Quantifying the Abundance of Alkane Moieties in Lignins with FTIR Spectroscopy and PLS Regression; Estimating Grafting Degree of Esterification

Abstract

As society is rapidly converting from fossil‐based materials to greener alternatives, the valorization of lignin through chemical modification has been given considerable attention. Characterizing this highly heterogeneous biopolymer is a constant challenge, and an emerging strategy for dealing with variations in material characteristics is combining traditional analytical techniques with chemometrics, such as Fourier‐transform infrared (FTIR) spectroscopy with partial least squares regression (PLSR). Here, a calibration data set was built based on FTIR spectra and the total carbon‐hydrogen bond (CHB) content of mixtures of technical lignins and alkanes, meant to emulate esterified samples. From this data, a PLSR model was built which predicted the CHB content of esterified lignin reaction products with an RMSE_CV=5.685 mmol/g and RMSE_Pred=5.827 mmol/g, and from which the weight percentage of ester‐to‐lignin was determined. When compared to wet‐chemical analysis, good agreement between the techniques was found with an obtained RMSE_Pred=8.3 % and a R² _Train=0.9752 for the degree of esterification. This indicates high model predictability and goodness of fit, and that the calibration data set successfully emulated esterified lignin samples.

By combining Fourier‐transform infrared (FTIR) spectroscopy and partial least squares regression (PLSR) a model was developed for quantifying the degree of esterification in lignin esterification reaction products. Quantification the degree of esterification (DoE) of esterified lignin samples gave RMSE_Pred=8.3 % and a R² _Train=0.9752 indicating that the model has good predictability.

Introduction

As part of the global effort to limit human influenced climate change, the substitution of traditional non‐biodegradable polymers and plastics with biopolymer‐based alternatives has been a central pursuit. ^[1] After cellulose, lignin is the second‐most abundant biopolymer on Earth making up 15 % to 35 % of lignocellulosic biomass, and despite its vast abundance it is still an under‐utilized natural resource. ^[2] Lignin has long been considered a low‐value waste product of the pulp and paper industry and has mostly been used as a low quality fuel for the generation of electricity and heat for the pulping process.[ ² , ³ ] It is a highly heterogeneous biopolymer comprised of three phenylpropane subunits, or monolignols, the hydoxyphenyl‐, guaiacyl‐ and syringyl units, and these are connected through a myriad of different chemical bonds, which add to the complexity of the biomolecule. ^[1] Many methods for extracting lignin from the lignocellulosic biomass exist, but three main processes are used industrially: the Kraft, soda and sulfite processes. Kraft and soda lignin are both produced from alkaline pulping, but Kraft lignin contains a considerable amount of sulfur that can impede further valorization. ^[4] Many alternative isolation techniques such as milled wood lignin (MWL) or ionic liquid extraction have been investigated towards the production of more native‐like lignins however, ^[5] these have been mostly limited to research activities so far. In general, the Kraft, soda and lignosulfonate technical lignins have more condensed structures than their native or native‐like counterparts.[ ⁶ , ⁷ , ⁸ ]

The esterification of lignins is a well‐reviewed topic with a myriad of potential applications.[ ⁹ , ¹⁰ , ¹¹ ] By taking advantage of lignins high abundance of hydroxyl groups, the material′s properties (e. g., hydrophobicity or compatibility with polyolefins) can be tailored by grafting various alkyl moieties. Esterified lignins have shown decreased Tg and improved solubility and melt characteristics in thermoplastics, ^[12] and increased solubility in lignin‐polymer plastics and resins. ^[13] They have also been investigated as biobased polyols for polyurethanes, ^[14] they have shown promise in biobased composites with polylactic acid,[ ¹⁵ , ¹⁶ ] and are being researched for uses in lignin‐based copolymers. ^[17] A constant challenge in lignin valorization research is connecting structural motifs to applications e. g., how successful a chemical grafting reaction really is in terms of the conversion of the reactive groups in question. The gold‐standard for quantification of reactive groups such as OH or COOH‐groups has long been nuclear magnetic resonance (NMR) spectroscopy, and in particular ³¹P and ¹³C experiments or the 2‐dimensional ¹H‐¹³C experiments HSQC and HMBC.[ ¹⁸ , ¹⁹ ] However, such analyses are costly and might not be easily accessible to all researchers, and more cost‐effective and available analyses such as ultraviolet‐visible‐ (UV‐vis) or Fourier‐transform infrared (FTIR) spectroscopy are more commonplace.[ ²⁰ , ²¹ ]

Partial least squares regression (PLSR) is a chemometric technique particularly well suited to analyze multicollinear data, such as spectroscopic data, and works well on datasets with a greater number of response variables than samples. ^[22] PLSR combined with FTIR has proven to be both a powerful and highly diverse analytical tool that has been used for the quantification of e. g., components in wood degradation and lignocellulosic biomass processing,[ ²³ , ²⁴ ] or the compositions of honey and wine samples.[ ²⁵ , ²⁶ ] Recently, this combinatory technique has been applied to lignin science for predicting thermal properties, ^[27] mapping the functional group compositions, ^[28] or tracking changes in molecular weight over time during depolymerization. ^[29]

In this work, a PLSR model was established for the quantification of the content of carbon‐hydrogen bonds (CHB) in lignin samples which were further used to calculate the degree of esterification in esterified lignin samples. Further, the samples used for modeling were mixtures of alkane compounds with technical lignins that simulated esterified samples, as a cost‐effective strategy that could be adapted by researchers for in‐house modeling. Although some data processing is required, basing the modeling on FTIR samples, rather on samples processed by time ineffective, destructive, and costly wet‐chemical methods, should result in a streamlined workflow with good estimates of grafting success. Additionally, a step‐by‐step guide is included for how to apply the developed model on new samples.

Results and Discussion

Data Preparation and Preprocessing

To build a representative dataset, three technical lignins were used; a softwood Kraft lignin (KL‐2), an Arkansas/straw lignin (ASL) and an organosolv lignin (OL) obtained from Norwegian spruce. Further, for the emulated ester counterparts three different alkane compounds were chosen with chain lengths in the C18–C21 range: stearic acid (C18), eicosane (C20) and NAFOL 1822 B (average: C21.28). From these components, mixtures based on weight percentage combinations in the 0 – 80 % range were prepared to emulate different degrees of esterification of the lignin material. The carbon‐hydrogen content (CHB) of the samples was chosen as the concentration metric since it can be applied to esterified samples of varying chain lengths. Further, instead of calculating merely the increase in CHB content, which required setting the CHB of the unmodified samples equal to zero, it was reasoned that including the small differences in composition between the technical lignins would likely better tune the quantification model. The CHB contents of the technical lignins were determined from elemental analyses (Table 1), combined with previous characterizations of the OH− and COOH− group contents of technical lignins. ^[30] For the alkanes, the CHB contents were determined from the number of C−H bonds per molecule divided by the molecular mass of the additive, see eq S1–S4 and table S1 in the supporting information for the calculations. Additionally for external model validation, eight esterified lignin samples were prepared, and their respective degrees of esterification determined by wet‐chemical methods (see experimental section). Lastly, the FTIR spectra of each sample was recorded as the average of two runs, from KBr pellets. This sample preparation technique has in a previous study given more accurate results for lignin samples as compared to ATR‐FTIR. ^[30]

Table 1.

Weight fractions for C, H, O and N of technical lignins and the associated calculated CHB concentrations used in the model development.

Abbrev.	Weight fraction (%)[a]				CHB[b] (mmol/g)
	C	H	O	N
ASL	61.0±1.3	5.9±0.0	26.8±0.1	3.7±0.9	50.6
HL	63.8±0.0	6.4±0.2	28.2±0.3	0.1±0.0	56.5
KL‐1	65.8±0.9	5.7±0.0	26.1±0.1	0.0±0.0	48.6
KL‐2	63.0±0.8	5.7±0.1	26.3±0.5	2.9±0.6	49.3
KL‐3	65.8±0.1	6.6±0.2	25.1±0.2	0.0±0.0	57.5
KL‐4	68.9±0.1	5.4±0.0	26.5±0.1	0.0±0.0	46.5
OL	62.6±0.9	5.6±0.0	26.6±0.4	1.9±0.0	49.8

[a] Obtained from elemental analyses. [b] Calculated from equations S1–S2.

The total number of samples used for the modelling was 72 alkane‐lignin blends, 3 unmodified technical lignins and 8 esterified samples. The data was arranged so that the predictor matrix consisted of the FTIR spectra (n×k), while the response vector was the calculated CHB concentrations (n×1). To increase the generalizability and robustness of the models, the samples of lignin‐alkane mixtures, and the unmodified technical lignins, were split 70 : 30 into training‐ (55 samples) and testing‐ (20 samples) sets. This data partitioning was done using a stratified random sampling approach, which addresses potential imbalances in the distribution of variables by splitting the data into subgroups based on percentiles before sampling within these groups to maintain a proportional distribution of variables in both the training and testing sets. ^[31] Further, the eight esterified lignin samples were included in both the testing set and in a separate prediction‐set. This was done to gauge the predictive ability of the models on real samples and to compare the predictions made on simulated‐ to real samples. Prior to initial model fitting, the data was baseline corrected with a simultaneous peak detection and baseline correction algorithm to remove baseline drift and systematic fluctuations. ^[32] The effectiveness of this preprocessing step was verified through both visual inspection and from the obtained model parameters.

Outliers in the training data set can disproportionately affect PLSR models, ^[33] so after initial model fitting the baseline corrected training set data was investigated for outliers via the reduced Hotelling′s T ² vs. Q‐residuals plot and from the residuals vs fitted response plot, see Figure 1. Hotelling′s T ² is a measure of the distance of an observation from center of the model, where scores=0, ^[34] for the chosen number of components (or latent variables). The $Q$ ‐Residuals, or the Sum of Squared Prediction Error (SPE), measures the residuals of the error matrix in relation to the projected value for the number of components used for the model. ^[34] See equation S5–S7 the supporting information for details on calculations. A total of 7 out of 55 samples fell outside the 95 % confidence limits and were flagged, and improvements in model performance was observed when removing these samples. However, as two of the flagged samples were the unmodified technical lignins, it seemed reasonable to retain these for increased robustness of the model on unseen samples.

Figure 1.

Potential outlier detection step, where the circled points represent samples outside the stapled 95 % confidence intervals. Left plot: Hotelling′s T ² statistic vs the Q‐residuals of the baseline corrected training set; Right plot: Residuals vs fitted(predicted) concentrations in outlier detection step.

The difference between retaining and removing these two samples were somewhat noticeable, with ${RMSE}_{CV}$ ‐values of 8.164 and 5.923 and $R_{CV}^2$ ‐values of 0.8241 and 0.9066, respectively. Retaining the technical lignins in the training set did however have a positive impact on model predictivity on the test and prediction sample sets. From the residuals vs fitted response plot some of the same samples were flagged as potential outliers. Additionally, no unusual patterns or trends were detected in this plot, indicating linearity in the data. ^[35]

Further preprocessing was done by first calculating the standard normal variate (SNV) then the 2^nd derivative using the Savitzky‐Golay (SG) function with a smoothing window width of 17, and a 3^rd order polynomial for the smoothing algorithm. ^[36] Variable selection was then performed, first based on domain knowledge, then secondly on Variable Importance in Projection (VIP) scores (equation S8 in the SI), see Figure 2 for illustrative examples of the preprocessing steps on the FTIR spectra. The manual feature selection included the ranges between 3044–2784 cm⁻¹ and 1636–952 cm⁻¹, but the prominent C=O stretching bands around 1600–1800 cm⁻¹ and the OH stretching band around 3700–3050 cm⁻¹ were purposely excluded, as the alkanes in the calibration data did not all contain carbonyl or OH functions. Important spectral features in the averaged sample spectra are shown in Table 2. Further, VIP feature selection was performed where a PLSR model was built and validated with 10‐fold cross‐validation. The optimum number of components was determined from the minimum value for the RMSE_CV and fed into the VIP algorithm. Although the empirical “rule” is to start with a VIP threshold of 1.0, the best results were obtained with a stricter VIP threshold of 1.2. Subsequently, the PLSR model was made with 10‐fold cross‐validation on the truncated data.

Figure 2.

Progressive preprocessing steps of the training set FTIR spectra colored by the type of technical lignin. a) Baseline corrected spectra; b) Standard Normal Variate (SNV) and 2^nd derivative preprocessing applied; c) Variable selection by domain knowledge; d) Variable selection by Variable Importance in Projection (VIP) score.

Table 2.

Averaged (normalized to 1508 cm⁻¹) FTIR spectra of the lignin‐alkane mixtures and esterified lignin samples, and important C−H signal regions with the associated type of vibrational modes.

Entry	Wavenumber (cm−1)[a]	Vibrational modes[a]
a – b	3042—2782	Alkane C−H Stretching: CH3, CH2, CH. Aromatic C−H Stretching
c	1605	Aromatic skeletal C−C stretching
d	1510	Aromatic skeletal C−C stretching
e	1465	C−H asymmetric deformation/bending
f	1426	Aromatic deformation +C−H in plane deformation
g	1367	Aliphatic C−H stretch in CH3, not OMe or Phen‐OH
h	1327	C−C stretching from S +G‐units condensation
i	1226	C−C +C−O +C=O stretching
j	1140	Aromatic C−H in plane deformation
k	1126	C−H deformation
l	1032	Arom. C−H in plane deformation

[a] Wavenumber values and the associated vibrational movements were obtained from O. Faix. ^[21]

PLSR Model Development

Even though several metrics have been developed in the debate on model validation,[ ³⁷ , ³⁸ , ³⁹ , ⁴⁰ , ⁴¹ , ⁴² ] a simple approach was chosen in this work. As proposed by Winkler, Tropsha and Alexander, validation based on the coefficient of multiple determination (R ² ) and the root mean square error (RMSE) should suffice, where R ² >0.6 and a low RMSE for the test set ensures good model fit and predictive ability. ^[43] Additionally, since RMSE penalizes large errors more than small ones, the mean absolute error (MAE) was also calculated, see equation S9–S11 in the SI. The linearity of MAE makes it more robust to extreme errors, ^[39] and verifying both the RMSE and MAE together can provide a fuller picture of a models performance. ^[44]

The generated models were investigated by both internal and external validation, see Figure 3. Good predictability was retained in both the prediction‐ and testing sets for the first 3 components, before the R ² _Test rapidly plummeted. A similar plateauing effect was seen in the RMSE values, which indicated the model was more robust toward the alkane‐lignin mixtures than esterified samples at higher component numbers, as expected. However, a low deviance between the calibration and cross‐validation R ² ‐values of 1.1 % was obtained at 3 components, ^[44] indicating a high goodness of fit and little overfitting of the model. This is also reflected in the measured vs predicted CHB content (Figure 3, bottom) and the validation metrics for the final PLSR model, see Table 3.

Figure 3.

Graphs over model development statistics for the training, testing and prediction data sets and for the cross‐validated training set as a function of the number of model components: RMSE of the CHB content (Top graph); coefficient of multiple determination (R2) statistics (middle); The measured vs predicted carbon‐hydrogen bond content of the Training and Testing sets (bottom).

Table 3.

Validation metrics of the best model generated with 3 components and 98.66 % and 93.83 % variance captured in X and Y, respectively.

	Dataset
	Train	CV	Test	Pred
R2	0.9383	0.9222	0.8032	0.6914
RMSE	5.062	5.685	7.250	5.827
MAE	3.928	4.494	5.608	4.978

The scores plots for the three components of the final model are shown in Figure 4. The top row of plots shows the group discrimination based on the origin of the technical lignin, which there is none of in components 1 and 2. Some separation occurs in the 3^rd component, but since this only accounts for ca 4 % of the X‐variance in the model it is largely indifferent to the lignin origin and indicates robustness in this regard. The plots in the bottom row are grouped based on the additive types. The samples are spread out in an increasing order based on their respective CHB content in the 1^st component and only the eicosane samples are discriminated in the 2^nd component. This is probably due to the inductive electron pull caused by the OH or COOH groups in NAFOL and Stearic Acid that decreases the electron density in the neighboring CH₂‐groups,[ ⁴⁵ , ⁴⁶ ] an effect which is not present in the eicosane samples.

Figure 4.

Scores plots of the three components with explained X‐variance contribution of each component, and where the points are colored according to CHB concentration indicated in the color scales. Top row: point shapes and shaded 95 % confidence ellipses correspond to technical lignin origin; Bottom row: point shapes and shaded 95 % confidence ellipses correspond to type of additive.

A similar discrepancy can also be seen in the loadings plot of the three components (Figure 5, top). The first two components weigh more heavily in the C−H stretching range at 3042–2782 cm⁻¹, while the third component weights the regions around 1500 cm⁻¹ which is associated with both aromatic ring stretching vibrations and aliphatic CH₃ antisymmetric deformations. ^[21] In the regression coefficients plot (Figure 5, bottom), the aliphatic range around 3042–2782 cm⁻¹ is weighted less, and the strongest positive and negative correlations are found in the band around 1500 cm⁻¹.

Figure 5.

Loadings plot (Top) and Regression Coefficients plot (bottom) for the three model components with the associated explained X‐variance contributions.

Calculations of Degree of Esterification

Although the CHB contents were used for quantification, the end goal was determining the degree of esterification of the lignin samples. This quantification was performed in two steps by first determining models predicted weight fraction of the alkane ($W_{Alkane}^m$ ) from each sample′s CHB using equation 1. This was further used to calculate the percentage degree of esterification of the ester set samples with equation 2, which were compared to the known values determined from wet‐chemical techniques.

$${\displaystyle W_{Alkane}^m=\frac{{CHB}_{Sample}-{CHB}_{Lignin}}{{CHB}_{Alkane}-{CHB}_{Lignin}}}$$ (1)

$${\displaystyle \%Esterification=100\%\times \frac{\frac{W_{Alkane}^m\times m_{Sample}}{{Mw}_{Additive}}}{c_{OH,Lignin}\times m_{Lignin}}}$$ (2)

The degree of esterification was also calculated for the lignin‐alkane blends, see Figure 6. Seeing as the samples were prepared based on weight% rather than molar%, the high‐wt % samples were though to help extrapolate towards real esterified samples of longer chain lengths. The obtained RMSE‐values for the Test‐ and Prediction sets of 12.790 % and 8.309 % could reflect that the model performs better at degree of esterification‐values where no extrapolation is done. This is also the case for the Training set samples, where lower values are more accurate than higher value samples. Arguably, higher accuracy might be obtained by removing the high‐wt % samples from the datasets, but this would be at the expense of the model′s extrapolative ability on longer chain length alkanes and was therefore not pursued further.

Figure 6.

The predicted and actual degree of esterification for the training‐, testing‐and prediction sets with the associated RMSE values. The calibration coefficient of determination (R2) for the testing set.

Conclusions

Traditionally, quantifying the degree of esterification from lignin esterification reactions is a tedious process requiring either time‐consuming wet chemical methods or costly analyses e. g., ¹³C‐ or ³¹P‐NMR. Often, a semi‐quantitative approach consisting of product mass yield and Fourier‐transform infrared spectroscopy (FTIR) is applied, where the qualitative increase in the C−H bands around 3042–2782 cm⁻¹ confirms successful grafting. In this work, based on a sample library consisting of mixtures of Kraft, soda or organosolv lignins with three different alkanes, chemometric techniques were combined with transmission FTIR spectroscopy to quantify the weight ratio of lignin‐to‐alkane in esterification reaction products. Additionally, by determining the content of carbon‐hydrogen bonds (CHB) in the analytical samples as a common measurement, the model is compatible with esterified lignins bearing fatty alkane chain of differing carbon lengths. However, for very short‐chain additives higher accuracies can be achieved by training a model on data including short chain samples, following the presented workflow. Good predictability was obtained for the CHB quantifications with RMSE_Test=7.250 mmol/g, MAE_Test=5.608 mmol/g, RMSE_Pred=5.827 mmol/g and MAE_Pred=4.978 mmol/g, and R₂‐values >0.6. Further, by finding the CHB content in a test set of esterified lignin reaction products and comparing to results from wet‐chemical analyses, the degree of esterification for each sample was calculated with an RMSE=8.3 %. Due to the samples consisting of two non‐bonded components, the model is not able to distinguish alkanes that are bonded to the lignin from those that are present as contaminants in impure reaction products. Thus, the results should be interpreted in tandem with other types of quantitative or semi‐quantitative analyses to provide a fuller picture of the quality of the material in question.

Experimental Methods

PLSR Modeling Software

The PLSR modeling was performed using RStudio (PBC, Boston, MA, USA, v2023.12.1 Build 402) and the pls package with the Orthogonal Scores algorithm (“oscorespls” method) with mean‐centering enabled for all PLSR model calculations. ^[47] Other relevant R packages used were the caret package for data partitioning, ^[31] the baseline package (with the “peakDetection” algorithm) for baseline correcions, ^[48] and the mdatools package for additional preprocessing steps (SNV and 2^nd derivative calculations). ^[36]

Materials, Chemicals and Solvents

The softwood kraft lignins BioPiva 100 (KL‐1) and BioPiva 395 (KL‐2) were obtained from UMP Biochemicals (Helsinki, Finland) while the Arkansas/Straw lignin Protobind 6000 (ASL) was purchased from PLT Innovations (Zürich, Switzerland). The Organosolv lignin (OL) was produced from Norwegian Spruce, as previously described by Ruwoldt & Tanase Opedal. ^[49] Additionally, a hydrolysis lignin (HL) of proprietary origin was used. The fatty alcohol NAFOL 1822 B was obtained from SASOL Germany GmbH (Hamburg, Germany), while all other reagents and all solvents were purchased from Sigma‐Aldrich (MERCK, Oslo, Norway) and were used as‐is without further purification.

FTIR ‐ Sample Preparation and Analyses

The analytical samples were prepared by mixing the appropriate additive (eicosane, NAFOL 1822B or stearic acid) and technical lignin (KL‐2, ASL or OL) in ratios of 10, 20, 30, 40, 50, 60, 70 or 80 wt %. The KBr discs were prepared by grinding the analyte mixture (1–3 mg) with dry KBr powder (350 mg) briefly in a mortar, before transferring the mixture to a pelleting press and applying a pressure of 10 tons for 2 minutes. After this the discs were immediately analyzed.

The FTIR analyses were performed on a PerkinElmer Spectrum 3 MIR spectrometer equipped with a Universal KBr Disc holder, and a MIR TGS (15,000–370) cm⁻¹ detector. The data was recorded with the PerkinElmer Spectrum IR (v 10.7.2.1630) software. Analyses were recorded between 4000 and 500 cm⁻¹, with 4 cm⁻¹ resolution and data point collection, and with 64 scans each. All analyses were run in duplicates, and after background measurements on a blank KBr disc with no analytes added.

Elemental Analysis

Elemental analysis was performed using a Flash 2000 CHNS/O Organic Elemental Analyzer (Thermo Fischer Scientific, Waltham, MA, USA) equipped with a MAS 200 R autosampler using He as the carrier gas. The composition of carbon, hydrogen and nitrogen was determined by combustion of samples pre‐weighed in a tin capsule in an oxygen‐rich environment at 1800 °C. The oxygen contents were determined by combustion at 1050 °C of samples pre‐weighed in silver capsules with no additional oxygen added. In both modes of analysis, the formed gases were transported onto a GC column for separation prior to reaching a thermal conductivity detector (TCD). All samples were compared to the calibration standard Atropine for quantification. From elemental analyses of each technical lignin preparation, the relative weight fractions ${\mathbf{w}}_{\mathbf{i}}$ of C, H, O and N were determined.

Hydrolysis and CG‐Quantification of Fatty Acids in Esterified Lignins

The esterified lignin samples (ca. 100 mg) were weighed into a round bottom flask and added a mixture of NaOH (aq., 1 M, 20 mL) in MeOH (20 mL). The solutions were heated to a gentle reflux for 1 hour before they were left to cool to ambient temperatures and were acidified with H₂SO₄ (10 %) to pH=2. Subsequently, the solutions were dried under N₂ at 55 °C, and the dry materials were extracted with known volume of a solution of cyclohexane/acetone (9 : 1). Samples of these solutions were then collected and added a solution of internal standards (heneicosanoic acid, betulinol, cholesteryl heptadecanoate and 1,3‐dipalmitoyl‐2‐oleoyl‐glycerol (not added for all analyses)) before the solvents were evaporated under N₂. To the dried mixtures, N,O‐bis(trimethylsilyl)trifluoroacetamide (BSTFA, 100 μL) and trimethylsilyl chloride (TMSCl, 50 μL) were added and the mixtures were heated to 70 °C for 30 min. Finally, the samples were centrifuged before the supernatants were analyzed by gas chromatography.

The GC analyses run on an Agilent 7890 A system equipped with an autoinjector and a J&W 125–1011 (70 m×530 μm×0.15 μm) column. Volumes of 0.2 μL were injected into a Cool on‐Column (CoC) injector kept at 80 °C. Helium (He 6.0) was used as the carrier gas at a constant flow of 21.6 mL/min. Sample detection was done with a flame ionization detector (FID) kept at 340 °C with He 6.0 as the makeup gas. The GC oven temperature program was: 90 °C (hold 1.5 min); 90 °C, 12 °C/min to 340 °C (hold for 5 min). The concentrations of the relevant fatty acids were then calculated using the response factors of the fatty acids and the internal standard, and the mass of the parent esterified lignin samples. Chromatograms of the analyses are shown in Figure S1.

Chemical Modification of Technical Lignins

THF Extracted Kraft Lignin (KL‐3) ^[50]

A mixture of BioPiva 395 Kraft Lignin and tetrahydrofuran (100 g/L) was added to a round bottom flask fitted with a condenser. The mixture was stirred at a gentle reflux for 5 hours, before it was filtered through a GF/A filter. The solvent was removed under reduced pressure, which yielded the extracted lignin.

Oxypropylated Kraft Lignin (KL‐4) ^[51]

BioPiva 395 Kraft Lignin (50 g) was dried by heating it to 55 °C under vacuum for 5 hours. Following this, propylene carbonate (5 equivalents per lignin OH group) and sodium carbonate (0.1 equivalent per lignin OH group) were added, and the mixture was stirred at 120 °C for 2.25 hours. The solution was let cool to ambient temperature before it was poured into HCl (4 L, 0.01 M) and stirred for 30 minutes. The mixture was added Celite 545 filtration aid (50 g), stirred for 10 minutes and filtered, and the solids were washed with distilled water (4×250 mL). The solids were then extracted with a solution of acetone and water (4 : 1), and the solvents were removed under reduced pressure to yield the oxypropylated Kraft lignin.

Esterified Lignins – General Procedure

The technical lignin (20 g) was first dried at 55 °C for 5 hours under vacuum. After drying, the lignin was added DMF (150 mL), and pyridine (10 mL) and the mixture was stirred for 30 minutes under a constant stream of N₂. A solution of fatty acid chloride in toluene (100 mL) was then added to the reaction mixture, and after stirring for 1 hour a second mixture of fatty acid chloride in toluene (100 mL) was added. Following this, the reaction was stirred for 20 hours at ambient temperature before it was quenched with ice–cold distilled water (500 mL). After formation of an emulsion in the top layer, the water phase was removed, and additional distilled water (500 mL) was added. The water phase was removed, and ethanol (0 °C, 500 mL) was added. The precipitated crude product was isolated and mixed with Celite filtration aid, before it was added isopropanol (200 mL), stirred for 30 minutes, and filtered. This was repeated two times, before the esterified lignin was released from the celite with 2‐Me‐THF. The solvent was removed under reduced pressure, which yielded the isolated esterified lignin product.

Author Contributions

Fredrik Heen Blindheim: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data Curation, Writing – Original Draft, Visualization. Jost Ruwoldt: Conceptualization, Methodology, Resources, Writing – Review & Editing, Project Administration, Funding acquisition.

Conflict of Interests

The authors declare no conflict of interest.

Supporting information

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer reviewed and may be re‐organized for online delivery, but are not copy‐edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors.

Supporting Information

Heen Blindheim F., Ruwoldt J., ChemSusChem 2025, 18, e202400938. 10.1002/cssc.202400938

Data Availability Statement

The data that support the findings of this study are available in the supplementary material of this article.