Identification of tropical wood species in paper: a new chemotaxonomic method based on extractives

2.1 Production of the reference pulps

To start with, the solid wood samples were manually cut into 3 x 3 cm wood chips with a self-built semi-automatic chipping machine. Afterwards, they were steamed with saturated steam at 1 bar prior to pulping for 30 min. The kraft pulping was carried out in a program-controlled 7-L M/K digester (M/K Systems INC., Williamstown, USA) with liquor circulation. Depending on the quantity of the raw material, between 550 and 1000 g dry matter wood chips were used per cooking. The liquor to wood ratio was 4:1 (v/w). Total NaOH chemical usage was 22–25 % with a sulfidity (Na₂S) of 35 %. The heating time (t) was 90 min followed by 120 min at a maximum temperature (T _max) of 165 °C. The obtained kraft pulps were slot screened to a maximum of 0.15 mm and subsequently bleached in five bleaching stages, adapted from the industrial bleaching practice of the Mercer Stendal pulp mill. The target brightness was 90 % ISO. The oxygen stage (O) was carried out with 12 % consistency (c), 2.5–4.0 % NaOH content and 0.4 % Mg SO₄ content for 120 min at 98 °C. The complexing agent stage (Q) was carried out with 3 % consistency and 0.2 % DTPA content for 30 min at 60 °C. The oxygen-enhanced peroxide stage (OP) was carried out with 12 % consistency, 2 % NaOH content, 0.1 % MgSO₄ content and 0.05 % DTPA content for 120 min at 95 °C. The chlorine dioxide stage (D) was carried out with 10 % consistency, 2–2.5 % ClO₂ content, 0.1 % DTPA for 180 min at 70 °C. Lastly, the peroxide stage (P) was carried out with 10 % consistency, 1.25–1.75 % NaOH, 0.1 % MgSO₄ content and 1–2% H₂O₂ content for 120 min at 80 °C.

The described kraft pulping process was based on the process steps and parameters commonly used in the industry. The fibers of the pulps produced are therefore similar to the fibers of industrially produced ECF pulps. A short discussion on the chosen pulping and bleaching parameters is included.

2.2 Grinding

The pulp was ground in a programmed and liquid nitrogen cooled ball mill (CryoMill, Retsch GmbH, Haan, Germany) with a frequency of 25 Hz using a 50 ml stainless steel grinding jar and one grinding ball of 25 mm diameter. Each pulp sample was ground five times for 1 min. Prior to the first grinding cycle, the grinding jar including the pulp was automatically cooled for 2 min. Cooling was applied for 0.5 min between all grinding cycles. The knife mill (M20 universal mill, IKA-Werke GmbH & CO. kg, Staufen, Germany) with a rotating vane blade was used with a speed of 20 000 1/min. The samples were ground three times for 30 s.

2.3 Extraction

The solvents were freshly distilled by rectification from both technical quality n-hexane and acetone (VWR International, LLC., Radnor, USA). The quality was analyzed via GC-MS. It is identical to GC quality n-hexane and PESTINORM acetone from VWR (VWR International, LLC., Radnor, USA). The extraction was performed by a Soxtherm SOX 6 from the company C. Gerhardt GmbH (Königswinter, Germany), run with the automatic MULTISTAT controlling system. The cellulose extraction thimbles and the viscose wadding used to plug the thimbles were pre-washed by a short Soxhlet extraction of 5–6 cycles for 60 min using a self-made azeotrope of n-hexane and acetone. The ground pulps of 5 g per extraction thimble were extracted with 140 ml of solvent with the automated self-programmed n-hexane extraction program: T-Class 200 °C; hotplate temperature 180 °C; setback interval 3.5 min; setback pulse 2 s; boiling phase 45 min; distilling off A, four intervals; extraction time 3 h; distilling off B, one interval; general distilling off interval time 5 min; total time 4 h 7 min. For acetone the extraction program was equal apart from the setback interval, which was 4.5 min. After extraction, the extracts were adjusted to a volume of 50 ml in volumetric flasks.

2.4 Sample preparation

The extracts were characterized by TD-GC-MS. Approximately 90 ± 5 µg of extract was applied into each sample cup by adding the extract solutions. Therefore, a defined volume of 50 µl of the solvents was gently evaporated by being placed under the fume cupboard at room temperature covered with a paper sheet to avoid dust pollution. Afterwards, the concentrations in the extracts were measured and sample cups filled up multiple times until the target mass was reached. After that, the sample cups were introduced into the pyrolysis system for TD-GC-MS analysis.

2.5 TD-GC-MS parameters

The TD-GC-MS analysis was performed using a micro furnace Double-Shot Pyrolyzer (Py-2020iD) equipped with an Auto-Shot Sampler (AS-1020E) both produced by Frontier Laboratories Ltd (Koriyama, Japan). The pyrolysis system was interfaced to a GC-MS (6890/5973N, Agilent Technologies Inc., Santa Clara, USA). The TD temperature for the dried and solid extractives was 325 °C and the interface was set to 330 °C. For pyrolysis of low-density polyethylene (LD-PE) retention index standards the pyrolyzer was set at 500 °C. The GC inlet and the GC-MS interface temperature were kept at 320 °C. A low polarity column (ZB-5HT, Phenomenex Inc., Torrance, USA) of 30 m × 0.25 mm i.d. and 0.25 µm film thickness was used with helium as carrier gas. The split ratio was set to 20:1. A flow rate of 1 ml/min (constant flow) was set for gas chromatographic separation. The signal data rate was 20 Hz, scan frequency was 2.22 scans/second and scan speed (u/s) was 1562 [N = 2].

The oven temperature of the GC started at 45 °C, which was held for 2 min and subsequently increased up to 340 °C with decreasing heating rates. This temperature was kept constant for 30 min. The exact configuration of the GC oven temperatures is listed in Table 1 and visualized in Figure 3. The total run time was 134.17 min. For mass spectral detection an 5973N MSD (Agilent Technologies Inc., Santa Clara, USA) was used with electron impact ionization energy of 70 eV. The scanning range for measurement in total ion current (TIC) mode was 29–700 m/z with a threshold of 100.

2.6 Data evaluation

Every dried pulp extract was analyzed at least twice. The multiple chromatograms per extract were visually compared in an overlay before checking its individual peaks. In case of unusually missing or additional peaks, these were screened and its reasons, such as a small air infiltration during the GC measurement for instance, were analyzed. Those samples were prepared again and measured twice to make sure the GC-MS measurement worked without any disturbances and that the chromatograms were equal. One of the representative chromatograms was further used. The data sets were pretreated and put together in a database of single-variety pulp extract reference chromatograms with the continuously developing database software OpenChrom^® (version: 1.4.0.202104211010, Lablicate GmbH, Hamburg, Germany) (Wenig and Odermatt 2010a,b) and its ChromIdent^® database tool. Compound identification of single peaks was performed using the NIST Mass Spectral Library (NIST20) (National Institute of Standards and Technology, USA).

Based on the fact that developing a new methodology is significant to this paper and therefore, its result, all key aspects of the methodology development are placed in Section 3 (Results and discussion). For instance, the outcome of comparing different mills for pulp milling and the discussion of suitable solvents for ground pulp extraction are part of the technique development process. Subsequently, the following chapter presents these alongside achieved bleached pulp brightnesses for different genera, their extract amount yield, the pulp extract chromatography data pretreatment and setting up/functioning of the reference chromatogram database.

3.1 Choice of wood species

The tropical forests of Southeast Asia are characterized by high species diversity among the tree species growing there. In cooperation with the NGOs Greenpeace and World Wide Fund For Nature (WWF), a list of 38 particularly relevant and interesting wood species was compiled (Table 2).

The pulp industry likes to use cheap and locally available woods. Therefore, the selection of relevant woods for the reference database was also based on the distribution and availability for the industry and correspondingly good procurement opportunities for the pulp industry. Some of the woods originate from plantations, for example Hevea brasiliensis or Paulownia tomentosa. They also find use as pulp as they are often co-products.

The list also includes Acacia mangium and Eucalyptus globulis, which are often legally used for pulp. These serve as a proof that the methods developed in this study can be applied to industrially produced pulps. In addition, these pulps are often found in papers and must be distinguished from tropical woods. The palms Cocos nucifera and oil palm Elaeis guineensis were included in this study because they can also be used for pulp production and it is anticipated that these palms, which are currently grown in plantations on former rainforest areas, will be increasingly cleared and utilized for pulp production (Onuorah E. O. et al. 2015; Wan Daud and Law 2011; Welling and Liese 2019). Table 2 also shows the achieved ISO brightness and extractive content of every samples bleached pulp.

According to the authors information, whole debarked logs are used in MTH pulp production and not sawmill residues. The authors therefore assume that heartwood always feeds into pulp production in large proportions and thus the use of heartwood as reference material was decided. Origins are not given in Table 2 since 2 kg of solid wood material were required for pulp production, which was not possible to obtain with a documented origin for most of the woods.

3.2 Grinding

For the selection of a suitable mill, the knife mill with rotating vane blade and the cryo ball mill were compared for pretreatment of MTH pulp before the extraction. Differences between the results may indicate that changes occur in the pulp during grinding. This could be caused not only by the different grinding techniques, but also by the temperature differences within the mills. The ball mill is cooled with liquid nitrogen (−196 °C) and produces a very fine powder, which provides a large surface area and thus good accessibility for the solvent (Figure 2c). It allows the solvent to flow uniformly through the sample and dissolve repeatable amounts of extractives, resulting in reproducible peak areas (Table 3). The knife mill, on the other hand, produces an inhomogeneous fluffy grind that contains both fibrous large particles (Figure 2b), but also very small powdery particles that are not visible in Figure 2b.

Figure 2:

Microscopic images of MTH pulp: (a) raw; (b) knife mill ground; (c) cryo ball mill ground. Staining was done with Alexander Herzberg stain.

The differently ground MTH pulps were extracted with petroleum ether (six samples per mill of 5 g each) and analyzed both quantitively, by gravimetric determination of the amount of extract, and qualitatively, by the detected number of peaks and peak areas in the TD-GC-MS chromatograms (Table 3). The quantitative results showed slightly higher extract amounts of 0.026 % (0.2 % CV) after the first extraction step for knife mill ground pulp in comparison to 0.020 % (2.3 % CV) for ball mill ground pulp. Furthermore, the number of peaks were higher within the knife mill treated pulp. These chromatograms had a mean number of 125.8 peaks (variance 17.1) whereas the ball mill ground pulps had a mean of 104.8 peaks (variance 6.2). The ball mill provided at least twice as good reproducibility of the peak areas from the extract chromatograms: the coefficient of variation (CV) of the areas of the 15 biggest peaks of the ball mill extracts chromatograms is 0.128 in contrast to 0.301 for the knife mill. The higher extract quantity and peak numbers of the knife mill spoke at first sight for this mill. But for the following analytical evaluation and the purpose of this paper the GC-MS measurement precision and the reproducibility of the results are crucial. The function of the fingerprint database relies on reproducible peak areas and proportions. Therefore, the extracts chromatograms of the pulps ground with the ball mill are more suitable for this study. The clearly better reproducibility of the peak areas when using the cryo ball mill could be due to the finer grinding under cold conditions.

3.3 Extraction

The successive extraction was also optimized on basis of the MTH industrial pulp milled with the cryo ball mill. The extract quantity was maximized through the selection of suitable solvents, extraction time and number of extraction cycles. The extract quantities of the individual extraction steps were determined. The extraction process used a combination of nonpolar and polar solvents, including three times petroleum ether (6 samples) followed by two times acetone and stand-alone n-hexane (5 samples) to obtain both fractions of the extracts as exhaustively as possible, as shown by Ponnuchamy et al. (2021) for kraft lignin and Krogell et al. (2012) for Norway spruce bark. Because the third extraction step obtained hardly any measurable extracts (Table 4), this method effectively extracts ground pulps exhaustively with two steps per solvent. Water was not used because it was assumed that water-soluble components were already discharged with the process water during pulping and bleaching. The results showed that most of the absolute extractive amounts were obtained in the first extraction step with either solvent. Table 4 also shows the relative extractives amount in percent and extract yields per solvent and extraction step.

Acetone was found to be quantitively more effective in extracting pulps than hexane, but high yield does not imply specificity of the extractives. Additionally, the acetone extracts were difficult to measure by GC-MS without derivatization due to their polarity. These polar components can be found in paper making effluents (Björklund Jansson 2005; Holmberg 1999; Örsa and Holmbom 1994; Valto et al. 2012; Willför et al. 2006), meaning they are affected at least in their quantity by pulping and bleaching. It was assumed that acetone mainly extracts polar extractives and hexane the more non-polar, more interesting extractives. The non-polar extractives are least affected by pulping and bleaching, as the process takes place in polar media, as water is used to dissolve the applied chemicals. Therefore, the hexane soluble extractives are more valuable for the database. The focus was on developing a practical method. That is why it was decided to leave the acetone extracts aside and concentrate on the hydrophobic fraction of the extractives. Another reason for the pre-fractionation is reduction of the complexity of the chromatograms. However, one problem with using the solvent petroleum ether is that it is not clearly defined in terms of composition, leading to a risk that extraction conditions may not be identical when changing batches or suppliers. Moreover, the ground pulps were not only cool extracted but boiled in the solvent in the first step. Monitoring the petroleum ethers (boiling range 40–60 °C) boiling temperature during the extraction process showed a temperature rise over time. This supported the assumption that the cool extraction was not carried out with the entire solvent mixture, but only with the less volatile part, because the composition of the petroleum ether presumably changed over time in favor of the higher-boiling fractions. So, petroleum ether is not perfectly suitable for extraction systems without pressure like Soxtherm or Soxhlet. In contrast, n-hexane is a well-defined solvent. The results of using n-hexane were found to be similar in terms of efficiency and reproducibility when compared to using petroleum ether (Table 4). Therefore, n-hexane was deemed the best choice for this purpose.

During the optimization process, sources of impurities for example in the extraction wadding, the extraction thimbles, the solvent or on glass surfaces and others were identified. They were largely eliminated by establishing a laboratory routine. These measures led to an improved quality of the pulp extracts with a significant reduction of impurities. A chromatogram of a blank extraction of empty and unwashed extraction thimbles and wadding with petroleum ether showing the impurities peaks (Supplementary Figure S5) and a list of identified compounds (Supplementary Table S40) were added to the supplementary material.

3.4 Optimization of TD-GC-MS chromatogram quality

Prior to development of data preprocessing systems, an optimization of the GC-MS analysis was a necessity. In this way, the quality of the chromatograms was increased by defining suitable GC-MS parameters such as the TD temperature of 325 °C and developing a fitting GC oven temperature program for the n-hexane pulp extracts. A better chromatographic separation of overlapping or sometimes called co-eluting peaks, which occasionally occur in one-dimensional gas chromatography (Blumberg 2012), was achieved especially of the mainly phytosterol peaks eluting around 280 °C in the range of RT 65–85 min (Figure 3). Their overlap was satisfactorily separated by the slow heating rates of 2.0 and 1.5 °C per minute. This was achieved by using polyethylene analyses as a neutral measure of the effect of the heating ramps. The heating ramps were adjusted so that the distances of the alkane and alkene peaks of the polyethylene were approximately equidistant over the relevant chromatogram range instead of increasingly narrow. Thus, a better capture of the information hidden behind coelution in this RT range of the chromatograms was achieved. In order to prevent artefacts of the high-boiling extract components in the following measurements, the final bake-out time needs to be 30 min long. Still the overall measurement time per extract of less than 140 min wasn’t unnecessarily extended. The optimized oven temperature program is shown in Figure 3.

Figure 3:

GC oven temperature program and chromatogram of Paulownia tomentosa (Thunb.) Steud. pulp extract before preprocessing. For the database references the chromatogram section of RT 7–88 min was analyzed.

3.5 Preprocessing of GC-MS data

The extractable constituents of the pulps were largely captured by the described conscientious sample preparation. For the following analysis they are available in the form of GC-MS chromatogram data sets. With the help of data preprocessing steps, it was possible to record the chromatographed analytes, i.e. their chemotaxonomic characteristics, and store them in a database. Appropriate pretreatment of the data eliminated or compensated artifacts. The quality of the chromatograms was thus improved and for each peak in the chromatogram, which in turn represents a component of the complex extractive mixture, characteristic properties such as percental peak area and retention index can be determined more reliably. The following nine most important preprocessing steps were found to be crucial for the pulp extract data matrix and the purpose of a subsequent fingerprint database. They need to be performed on the raw GC-MS data files for both references for the database and unknown pulp/paper samples for comparing against the database.

3.5.5 Smoothing chromatograms

For a well working peak detection and clean peak integration the whole chromatograms were smoothed. For this purpose, the chromatogram filter Savitzky-Golay smoothing with the settings order = 2 and width = 5 was applied. This filter algorithm was developed by Savitzky and Golay (1964). It is used to smoothen single ion signals or TIC signals and remove electric noise (Wenig 2011). The smoothing was done per ion, meaning every single m/z was smoothed separately, which is important for the following peak detection via deconvolution. Figure 4 shows how smoothing the raw m/z data also affects the peak shape in the TIC chromatogram. One can see how the blue raw chromatogram in Figure 4a looks unsmoothed in comparison to the smoothed brown TIC signal underneath (Figure 4b). For instance, the three mid-sized peaks in the center (RT 66.5, 67.7, 69.1 min) show rough detector signals and therefore tiny double peaks, which are removed in the smoothed data set below. Choppy signals are risky because they can be falsely detected as multiple peaks on both the ion level and the TIC level.

Figure 4:

Savitzky–Golay smoothing: TIC chromatogram sections from database reference sample Paulownia tomentosa (Thunb.) Steud., RT 66.10–70.50 min: (a) raw data versus (b) smoothed data.

3.5.6 Deconvoluting/detecting peaks

Multivariate curve resolution by alternate regression (MCR-AR) according to Gerber et al. (2012) was used to decompose the multivariate GC-MS data into individual pure component spectra and to detect the peaks. MCR-AR enables the detection of hidden peaks underneath the baseline noise or other peaks via their single ion traces. Thus, the underlying and overlapping peaks can be detected individually as shown in Figure 5. MCR involves a segmentation of chromatograms into non-overlapping minimum 50-scan windows using local minima/maxima. Mass channels are baseline-corrected within these windows by linear interpolation and area subtraction. Each window forms a data cube. This data cube is then unfolded into a matrix (N × K) × L, where N represents the number of samples, K represents scans, and L represents mass channels. The decomposition obtained from the initial MCR step is used as a starting point for AR. It alternates between deconvoluting chromatographic and mass spectral profiles until convergence. It starts by assuming a single distinct compound (rank = 1) and applies constraints like non-negativity to the solution. The pure component spectra are used as constraints in the regression process. The data is iteratively reprocessed (‘MCR-AR max iteration’: 50.0), by incrementally increasing the rank by one and calculating solutions with each iteration refining the decomposition. It continues until a predefined stopping criterion is met, indicating that the decomposition has reached a stable state. This means that the algorithm has found a solution that best represents the underlying pure components in the mixture (Gerber et al. 2012). This peak detection is very useful for the creation of the fingerprint database, as the complex wood extract samples often contain many similar substances that elute very close to each other in time. The overlaps cannot be completely separated chromatographically and sometimes appear as “mountains” in the chromatogram (Figure 3, RT 65–85 min). In the following some MCR-AR settings, that were used for best results, are given and explained. The ‘Local Maxima/Minima’ classifier uses an algorithm that follows the signal from left to right to find local maxima/minima based on the first derivative/slope of the signal. Although this classifier does not search for chromatographic peaks with start/maximum/stop, some kind of start point must still be found internally, to be able to evaluate the parameter ‘Signal difference threshold between start and extrema’, which was set to 1000.0. To set a starting point it is required that the slope exceeds a certain threshold (‘Local Maxima Scan Slope Threshold’: 0.001). Then the ‘Local Maxima/Minima Consecutive Scans’ parameter comes into play: this defines how many scans must remain above this threshold after it has been exceeded for a start point to be set. This helps to filter out the smallest local maxima in the signal noise range. It was set to 1.0. All detailed MCR-AR parameter adjustments including explanations can be found in Supplementary Table S39.

Figure 5:

Peak detection using deconvolution (MCR-AR): Section from smoothed database reference sample Paulownia tomentosa (Thunb.) Steud., RT 66.11–70.03 min.

3.5.6.1 Integrating peaks

To determine their peak areas the MCR-AR detected peaks were integrated with the trapezoid peak integrator. The deconvoluted TIC signal was integrated excluding the background. The peak areas were chosen to be put into the database as one of the comparison features rather than the intensity because the intensity and peak shape can vary whereas the peak areas and proportions to each other, which are used for fingerprinting, stay the same. The peak areas are normalized so that only a percentage value is used. Since this is not a quantitative but a qualitative identification method, the absolute content is not decisive here, thus only the ratios need to be repeatable.

3.5.6.2 High pass ions and peaks

The high pass filters keep all ions or peaks with the biggest intensity or area. For this database the 400 biggest peaks by area per reference chromatogram worked the best and were kept for further analysis. Of every of those peaks, the 150 ions with highest signal intensities were kept. Keeping more than 150 ion traces didn’t approve the database performance on random pulp mixtures because in the mixture the amount of every pulp is lower. Therefore, the extracts peaks are smaller and consequently due to the threshold settings the smallest single ions are not detected by the MS anyway.

3.5.6.3 Retention index (RI) calculation

The retention index (RI) is a measure used to characterize the retention time of a compound relative to the elution times of reference compounds. It is dimensionless and represents the relative affinity of a compound for the stationary phase compared to the reference compounds, meaning it converts retention times into system independent constants. An internal RI value is calculated for each peak maximum with LD-PE as standard reference, which is separated under the same conditions as the sample of interest. The RI are determined by relative distances of the compounds of interest to the LD-PE alkene peaks in the chromatogram: Retention Index RI = 100 × [n + (N − n) × (RT(unknown) – RT(n))/(RT(N) – RT(n))] where n is the carbon number of the n-alkene, N = n + 1 and RT = retention time. In every GC-MS sequences LD-PE RI standards are measured minimum after every 10th measurement. This is how a system-independent index value without time unit could be given, which is not influenced by any kind of RT shift due to column abrasion, cutbacks of the column or a changed column/measurement program. It is very important to calculate the RI for every peak because the RT shift can even on the same GC-MS system be substantial with longer times between measurements and therefore influences the matching accuracy of the database massively if no RI are used.

3.6 Functioning of the database

The pretreated chromatograms contain several pieces of information important for the functioning of the database: deconvoluted peaks consisting of the individual mass traces, their peak areas and RI. These are associated with the peaks. For each reference extract of a pulp produced from a single wood species, the above information results in an individual fingerprint of contained compounds. Although most of the substances are not wood species-specific marker substances, the complex combination of substances with their proportional ratios to each other is unique. This fingerprint information is fed into the database (Wenig and Odermatt 2010a,b), as shown in Figure 6.

Figure 6:

Schematic illustration of the database approach.

When adding a reference chromatogram to the database, each peak within a fixed RI window of ± 10 is compared with all other reference peaks in the same frame. In contrast to the NIST library peak comparison for compound identification by Stein and Scott (1994) for the fingerprint matching the cosine algorithm by Alfassi (2004) is used. The peaks are basically compared via their mass spectral match quality. Using retention indices is essential, otherwise the rate of false positive matches rises due to the similarity of wood extractive peaks, especially of alkanes due to their similar mass fragmentation pattern. Only if the peaks from different reference extracts match 80 % or better (Match Factor 80+), they are rated as the same compound and merged and added collectively to the database as one library peak. This combined library peak links to all the corresponding references where it originates from. If peaks at the same RT/RI match with a smaller percentage than 80, they are rated to be different peaks and put into the database separately as individual library peaks. The matching parameter was set to a high value of greater than MF 80 because the wood samples contain confusingly similar components.

When matching an unknown mixed species pulp extracts chromatogram for identification against the database, the first step again is to align the unknown chromatogram with the reference chromatograms in the database using an RI corridor of ±10. After that searching for matches in that RI corridor using the cosine algorithm starts. Also, peak areas and ratios (fingerprinting) are compared to find the best matches. The results, based on the similarity between the unknown and reference mass spectra, are ranked/sorted according to the scores, with higher scores indicating stronger matches. These MF scores are penalized by RI distance: the higher the RI (within the determined RI window) are lying apart from each other, the more the MF values are reduced. A minimum MF of 75 for every peak comparison of the unknown against the database reference peaks was chosen somewhat lower, since impure mass traces are expected due to overlaps within the mixed pulp extract sample. In addition, particular mass traces of wood species contained in small amounts in the mixed pulps may be below the mass spectrometers threshold intensity of 100. As soon as a peak of an unknown extract is identified against the database subset (RI ± 10, MF > 75), the information is available if this peak occurs only in one or more references. A peak, that occurs only in one reference sample or group is a direct hit/marker peak (Figure 9). Therefore, a relatively high number of marker peaks, compared to the number of total matched peaks assigned to a specific database reference, is an important indicator in the identification of unknown samples. All other peaks are ambiguous hits. The mathematics behind it is quite simple: principally it’s just counting the references. A direct hit counts 1. An ambiguous hit, which occurs in two references, counts with 0.5 and so on. The quantity of peaks and number of counting’s provide a hint of the unknown extract. When evaluating the comparison results of an unknown sample against the database the reverse similarity index (RSI) and reverse match factor are most important. The unknown pulps are rarely pure. They are mostly mixed samples from different wood species. Therefore, the match against the database is tested in reverse to see how well the fingerprint peaks of the single pure database references match the peaks of the unknown sample mixture, i.e., whether every pure reference can be found in the unknown mixed sample. Since the mixed sample consists of several species, matching the whole unknown mixture sample forwards against each reference of the database results in lower MF, because the entire sample mixture can never be found in one pure reference sample. As an example: a mixture of X, Y and Z is identified by a database consisting of the references X, Y and Z. Reverse matching results: X is 100 % included in the mix, Y is 100 % included in the mix and so is Z. Forward matching results: the whole mixture matches 33 % with each X, Y and Z. The results of a database matching are mean values of the matching of all individual peaks against each other. The individual peak matching can be represented as in the following Figure 7. Figure 7a shows the mass traces of an unknown deconvoluted peak of an unknown pulp extract mixture in the RT range of 67.50–68.00 min with its maximum at the RI of 3127 in comparison to the mass traces of the P. tomentosa database reference peak at the same RI in an inverted overlay format with the legend of all mass traces (m/z) on the bottom right. Figure 7b shows the comparison of the same P. tomentosa database reference peak in detail with all its mass traces not over an RT range but a single scan against the NIST20 library. The unknown extract substance is identified as (+)-Sesamin with an MF of 80.3 and a reverse MF of 87.2.

Figure 7:

Single peak database comparison: (a) comparison of an unknown deconvoluted peak of an unknown pulp mixture extract against the database library peak from the Paulownia tomentosa (Thunb.) Steud. reference pulp extract at RI 3127; (b) comparison of the database reference peak against the NIST20 library peak of the substance (+)-Sesamin.

In order to represent the database graphically and to make the dimension of the differences of the individual references visible, a principal component analysis (PCA) was calculated from the database entries with the above-mentioned information. In the score plot of the PCA (Figure 8) the PC 1 (principal component 1) with 10.98 % on the x-axis and the PC 2 on the y-axis with 7.96 % represent only about 19 % of the total database information. This is due to the fact that the constituents of the wood species are often similar and a huge data cloud of more than 6000 peaks times 38 species is difficult to reduce in its complexity to two dimensions. In the 3-dimensional interactive score plot (PC 1, PC 2, PC 3) (Supplementary Figure S3), a total of about 26 % of the database information is represented and the distances are clearer than in 2-D.

Figure 8:

(a) PCA-score plot of the pulp extract database, numbered after Table 2; (b) enlarged section of the Dipterocarpaceae family.

Figure 8b shows an enlarged section of the score plot exposing the Dipterocarpaceae family. As expected, the close relationship between the Dipterocarpaceae family members in particular the Shorea subgenera is evident in the chromatograms – they cluster together in the PCA representation. Thus, when identifying one of these species in an unknown sample, a closer look must be taken to ensure that there is no confusion. In order to refine the identification of Diptocarpaceae members, a separate database containing only these very similar species could be built in the future. Possibly, the matching parameters can then be chosen even more narrowly, and thus the distinctiveness of the matching results can be increased.

For each database reference chromatogram, there is a table of recorded library peaks including their RT, RI, peak area, NIST20 identified substances (MF > 80) and their 10 biggest ion traces as well as the other database references in which these peaks are also contained (Supplementary Tables S1–S38). The most decisive peaks are the specific marker peaks originating from only one reference. The whole database, including all original GC-MS data, can be found under Supplementary Figure S4. The reference chromatograms of A. mangium and P. tomentosa pulp extracts are also given as examples in Supplementary Figures S1 and S2.

The comparison of unknown paper extracts chromatograms against the database is supported by a visual comparison tool which shows the unknown chromatogram with unmatched, matched but ambiguous, meaning they are contained in more than one reference, and marker peaks (Figure 9).

Figure 9:

Comparison of an (a) unknown pulp mixture extract chromatogram with (b) the database reference chromatogram of Paulownia tomentosa (Thunb.) Steud.

When matching an unknown pulp extract chromatogram with the database, several comparison factors and statistics are crucial: the total RSI value, the number of marker peaks/ambiguous peaks and the total matching area which is calculated for every database reference compared to the unknown chromatogram. Ultimately, the observer must incorporate all of these comparison results/statistics into the decision-making process. An example of an actual species identification in an unknown pulp mixture extract is shown in Figure 10, a screenshot of the software interface. It shows the database query results from a chromatogram of a mixed pulp extract against the reference database. For the database reference of P. tomentosa an RSI of 88.7 % is given, 58 marker and 141 ambiguous peaks are matched. Additionally, 10.4 % of the peak area of the unknown chromatogram was identified as P. tomentosa. Analyzing these clear results, the authors correctly concluded that P. tomentosa is contained in the mixture. Tectona grandis (15 matched marker peaks) and Fagus sylvatica (22 matched marker peaks) are also contained one-third each in the mixed sample and correctly identified. As the sample was randomly mixed by a colleague, the composition was unknown to the authors until the analysis and identification decisions were made.

Figure 10:

Database query results from a chromatogram of a mixed pulp extract (containing Paulownia tomentosa) against the reference database: screenshot of the software interface.

3.7 Limitations

The authors acknowledge that limitations exist in the studies ability to comprehensively represent all possible variations within tree species. The researchers have taken these limitations into account.

Due to the requirement of 2 kg of material for pulp production, it was not possible to obtain samples with documented origin for most of the woods. As a result, there could be variations within tree species/genera that the current database is not capable of identifying, particularly in cases where trees are grown under extreme conditions.

In conclusion, the inability to obtain samples with a documented origin for most of the wood genera/species, along with the practical constraints on the number of samples analyzed, constitute limitations of the study. They highlight the future need for expanding the database to account for variations in tree species grown in diverse environments.