Absorbance and second derivative spectra
Figure 2a shows the average FTIR spectra in the fingerprint region (between 1800 and 800 cm−1) of the pine heartwood samples. The strong absorbance at 1690 and 1025 cm−1, and moderate to weak absorbance at 1590, 1510, 1460, 1380, 1265, 1155, 1105, 900 and 825 cm−1, are associated with the major biopolymers of wood, i.e. cellulose, hemicellulose and lignin (see Table 2). Furthermore, the band at 1690 cm−1 is specific to resin acid compounds (Nuopponen et al. 2003; Vahur et al. 2011).
In the second derivative spectra, many bands are identified more easily (Fig. 2b) confirming previous studies using the second derivative (Boeriu et al. 2004; Huang et al. 2008; Popescu et al. 2009; Zhang et al. 2016). The identified bands are listed with their most likely sources (based on literature) in Table 2. Most of the bands were assigned to lignin (825, 860, 960, 1225, 1265, 1425, 1465, 1510 and 1590 cm−1) and polysaccharides (805, 900, 985, 1005, 1025, 1055, 1105, 1155, 1185, 1315, 1335, 1360, 1385, 1635, 1655 and 1730 cm−1) (Faix 1991; Pandey 1999; Schwanninger et al. 2004). In general, the FTIR spectra of individual wood components (extractives, polysaccharides and lignin) support the band assignments given above. The FTIR spectrum of the extractives showed that it is composed almost exclusively of resinous materials (no bands from, e.g. tannin identified). Furthermore, the first three principal components from the PCA (accounting for 99% of the variation in the dataset) show striking resemblance with the spectra of the chemically isolated wood components (see Fig. 3 and also in Table 2). More specifically, PC1 is associated with polysaccharides (with higher scores for C–O bonds), PC2 with terpenoid constituents (high scores for carbonyl bonds) and PC3 with lignin (high scores for aromatic structural vibrations) (Fig. 3; Table 2). It is concluded that a combination of (1) comparison with band assignments from literature, (2) PCA of the whole dataset, and (3) chemical treatments for obtaining reference materials allowed for a very reliable identification of the primary structures responsible for the bands, including the minor ones, discussed in this study.
Discrimination based on lignin and polysaccharide absorption bands separately
The stepwise discriminant analysis applied to the bands that were assigned to lignin (DFL) presented a clear separation (Fig. 4a) between Pinus sylvestris (Ps-ART) and Pinus nigra samples (Pn-LIN and Pn-LSA) on the first discriminant function (DF1L), which showed highest loadings for bands at 860, 1265, 1425, 1465 and 1510 cm−1. However, the distinction between the two Pinus nigra locations was poor. Using these four bands, the discriminant analysis was performed again coding only for species (Pinus sylvestris vs. Pinus nigra) in order to confirm the discrimination between them. The results showed a clear separation between Ps-ART samples, with positive scores, and Pn-LIN and Pn-LSA samples, with negative scores (data not shown).
The discriminant analysis on the bands related to polysaccharides (DFP) provided a good separation (Fig. 4b) between species on the first discriminant function (DF1P) and the two locations of Pinus nigra (Pn-LIN and Pn-LSA) were separated by the second discriminant function (DF2P). The associated bands are 805, 900, 1105, 1155, 1187, 1315, 1335, 1383, 1635 and 1655 cm−1. To confirm the discrimination related to forest location, another analysis was performed using only samples from Pinus nigra (Pn-LIN and Pn-LSA). Again, the results showed good separation, Pn-LIN samples showing positive and Pn-LSA samples negative scores.
Stepwise discriminant analysis with all absorption bands
The DFT provided two discriminant functions that are based on twelve infrared bands (860, 900, 1105, 1225, 1315, 1335, 1385, 1405, 1425, 1610, 1635, 1655 cm−1). The first function (DF1T) explains 82% of the total variance while the second function (DF2T) accounts for the remaining 18%. The projection of the two functions (Fig. 5) shows that the samples plot in three main groups: Ps-ART, composed of samples of Pinus sylvestris trees from the Artikutza natural park; Pn-LIN, composed of samples of Pinus nigra trees from Linarejos, and Pn-LSA, composed of samples of Pinus nigra trees from La Sagra Mountain. Hence, DF1T (0.94 canonical correlation) showed a clear separation between the samples of the training set of Ps-ART, with positive scores, and the samples of the training sets of Pn-LIN and Pn-LSA, which have predominantly negative scores. The ANOVA test indicated that this difference is significant (P < 0.001) for DF1T scores (differences between pine species), as observed from boxplots (Fig. 6a). The DF2T (0.80 canonical correlation) showed a clear separation between samples of the training sets of Pn-LIN and Pn-LSA, with positive scores for Pn-LIN and negative scores for Pn-LSA, albeit that there is some overlap between the clusters of Pn-LIN and Pn-LSA. A highly significant difference (P < 0.001) was also found for DF2T scores (differences between locations for Pinus nigra trees) (Fig. 6b).
The validation set confirmed the accuracy of the discrimination between species and sampling locations. For the species differentiation (DF1T), Pinus sylvestris and Pinus nigra samples of the validation set were all correctly identified, with an average probability of 0.99 ± 0.01 (AVG ± SD); the same was found for each individual group of samples [Ps-ART-v (n = 6), Pn-LIN-v (n = 12) and Pn-LSA-v (n = 6)]. As for the growing location (DF2T), the samples of the validation sets of the two sites of Pinus nigra showed an average probability of correct site identification of 0.91 (SD 0.22). More specifically, samples from Pn-LIN-v were correctly assigned to their location with an average probability of 0.87 ± 0.26, whereas the probability for Pn-LSA-v was 0.99 ± 0.01. The lower probability for the Pn-LIN-v samples suggests that there is more variability in wood composition in the trees from this site. Even though it is known that the composition of arboreal wood depends on environmental factors of the growing location (Fritts 1976; Creber and Chaloner 1984), this variability can also be related to the sampling strategy (larger number of trees sampled). The relatively high chemical heterogeneity of the samples from Linarejos is also reflected by the relatively large distances of Pn-LIN samples to their group centroid in [DF1T-DF2T] space (Fig. 5). In fact, two samples of the training set and two of the validation set were grouped with Pn-LSA samples.
Interpretation of absorption bands that discriminate between species and locations
To elucidate the chemical features of the wood that allow the models to differentiate between species and sites, the interpretation was limited to the lignin and polysaccharide bands that were identified from the individual discriminant models (DFL and DFP) and by the global model (DFT), i.e. the bands at 860 and 1425 cm−1 for lignin and 900, 1105, 1315, 1335, 1385, 1635 and 1655 cm−1 for polysaccharides. The standardized canonical coefficients for the discriminant function are provided in Table 3. Bands with positive standardized canonical discriminant coefficients for DF1T are more intense in Ps-ART samples (positive discriminant scores) and bands with negative coefficients are more intense in Pn-LIN and Pn-LSA samples (negative discriminant scores), whereas bands with positive coefficients for DF2T show higher absorption in Pn-LIN samples (positive discriminant scores) and bands with negative coefficients show higher absorption in Pn-LSA samples (negative discriminant scores).
Lignin absorption bands
Firstly, the band at 860 cm−1 is associated with the CH out of plane bond vibrations in guaiacyl lignin (Evans 1991; Sills and Gossett 2012; Zhou et al. 2015). The positive standardized coefficient of this band implies that its absorbance is relatively strong in Pinus sylvestris wood. However, the band at 860 cm−1 can also be attributed to the propanoid side chain in lignin (Scheinmann 1970; Sharma 2004). Secondly, the band at 1425 cm−1 is assigned to the CH asymmetric deformation in methoxyl and aromatic skeletal vibrations in lignin (Faix 1992; Kubo and Kadla 2005). In this case, the standardized coefficients are negative, so that this band is associated with Pinus nigra samples. This suggests that aromatic methoxyl groups may be slightly more abundant in the structure of lignin in the Pinus nigra samples than the Pinus sylvestris samples.
These differences in lignin structure can be related to the lower shade tolerance of Pinus sylvestris than Pinus nigra (Trasobares et al. 2004). Lignin is the most receptive wood component to interactions with electromagnetic energy because of its macromolecular architecture (Chang et al. 1982). This higher sensitivity of Pinus sylvestris may be due to the lignin structure in the middle lamella, which is sensitive to sunlight exposure. Evans et al. (1992) showed that the CH out of plane bending vibration is the most influenced by factors that modify wood lignin composition. Thus, it is assumed that in Pinus sylvestris lignin structures with more CH out band bending vibration are produced in order to maintain phenylpropane derivatives, which are considered to protect meristem cells against light stress (Higuchi 1997).
The lignin signature was not useful for distinguishing different growing locations.
Polysaccharide absorption bands
For species differentiation, the band at 1635 cm−1 is relatively strong in Pinus nigra, whereas the band at 1655 cm−1 was associated with Pinus sylvestris samples. Both bands are related to OH bond vibrations including intermolecular hydrogen bonds between polysaccharide chains (Genest et al. 2013; Karunakaran et al. 2015). The intensity of hydrogen bonding exhibits strong influence on the rigidity of the cellulose chain and provides mechanical stability in fibres (Higuchi 1997; Klemm et al. 2004; Popescu et al. 2009). Karunakaran et al. (2015) stated that the band near 1635 cm−1 is characteristic of insoluble xylan. Insolubility could be due to the strong hydrogen bonds that lead to a large interaction between xylan chains (Poletto et al. 2014). Therefore, it is hypothesized that there are stronger interactions between xylan chains in Pinus nigra than in Pinus sylvestris. It had previously been found that xylan content could vary between wood species, even from the same genus (Timell 1964; Sjostrom 1981).
Regarding the differentiation of growing locations on the basis of polysaccharide FTIR signatures, the band near 900 cm−1 is stronger for pine samples from Linarejos (Pn-LIN) than for La Sagra Mountain (Pn-LSA). This band corresponds to the CH deformation of β-glycosidic linkages, which is related to the abundance of amorphous cellulose (Faix and Böttcher 1992; Evans et al. 1992). The bands at 1105 and 1385 cm−1 are assigned to the C–O stretching in cellulose and hemicellulose (Faix 1991; McCann et al. 1992; Zhang et al. 2010); however, the former, with a positive coefficient, seems to be associated with Pn-LIN samples while the latter, with a negative coefficient, seems associated with the Pn-LSA samples. There is no convincing explanation for this observation.
The bands at 1335 and 1315 cm−1 are assigned to amorphous and crystalline cellulose, respectively (Colom and Carrillo 2005; Popescu et al. 2007). Samples from the Cazorla Mountain (Pn-LIN) had higher intensities of the band corresponding to the amorphous cellulosic structure, while samples from La Sagra Mountain (Pn-LSA) were characterized by higher intensities of the crystalline structure of cellulosic compounds. Indeed, the ratio between crystalline and amorphous cellulose bands (1315/1335 cm−1) indicates a relatively high degree of crystallinity in samples from LSA (Fig. 7). This ratio has been used as an empirical crystallinity index to compare amorphous and crystalline cellulose between hardwood and softwood (Colom et al. 2003; Colom and Carrillo 2005). The ANOVA test indicated that cellulose crystallinity is significantly higher (P < 0.001) in samples from La Sagra Mountain (Pn-LSA) than in those from Linarejos (Pn-LIN).