1. Introduction

The material properties of wood are critical for the acoustics of string instruments. In violins and cellos, the soundboard (top) is made of specially selected spruce resonance tonewood (Picea abies), while the back is maple tonewood (Acer species) [1]. Curiously, leading violinists mostly prefer to play instruments made three centuries ago by old masters in Cremona, Italy, particularly Antonio Stradivari and Giuseppe Guarneri “del Gesù” [2,3]. It has long been speculated that Cremonese tonewood had special properties or hidden secrets [4,5], which continue to attract wide public interest and ongoing scientific examinations. Wood density measurements using computed tomography obtained similar results for Cremonese violins and modern instruments [6]. X-ray diffraction studies showed that Cremonese and modern maples still possess similar cellulose domain sizes [7]. The ¹³C{¹H} cross-polarization magic angle spinning (CPMAS) nuclear magnetic resonance (NMR) spectroscopy results gave the evidence of natural oxidation of lignin [7]. However, NMR and infrared spectroscopy revealed unusual chemical changes in the maples of Stradivari and Guarneri [8]. Elemental analyses suggest that Stradivari and Guarneri treated their maple wood with complex mineral recipes [7,9]. These treatments appear to have caused structural rearrangements in the holocellulose complex (cellulose plus hemicellulose), giving rise to an extra peak in the differential scanning calorimetry thermogram [7]. The rearrangement could have resulted from two factors: (i) hemicellulose degradation promoted by base-catalyzed hydrolysis (potash or lime solutions) in combination with spontaneous hydrolysis due to aging; (ii) molecular movement caused by long-term vibrations [10].

Thus far, maple wood in Cremonese violins have been subjected to more scientific examinations than spruce wood, but the latter is more important for instrument acoustics [3,11]. Investigations into Cremonese spruce have been impeded by the scarcity of wood samples removed during instrument repairs. In this study, we seek to develop a non-destructive optical method to investigate whether Cremonese spruce samples also exhibit unusual chemical and structural changes. Importantly, the analytical method would require only minimal samples, down to a few milligrams.

The wood cell wall is composed of three biopolymers: cellulose, hemicellulose, and lignin [12]. Only lignin shows strong absorbance and autofluorescence in the UV-visible region, which can be used as a label-free and non-destructive method for investigating the chemical modification of wood [13]. Lignin is a natural polymer abundant especially in xylem. Multiple fluorophore types within the lignin molecules lead to a broad excitation (UV and visible) and emission range [14,15]. Due to the complexity of lignin molecule, the fluorescence centers have yet to be identified [16]. Lignin fluorescence is sensitive to the molecular environment, such as pH variation, heat treatment, and chemical treatment [14,17,18]. It may also be affected by the physical structure of the polymer [19]. The presence of covalent linkages to hemicellulose may suppress the fluorescence, while removal of carbonyl groups tends to enhance the fluorescence [15,20]. Therefore, extraction of lignin from wood alters the fluorescence behavior inevitably. To study the lignin fluorescence in its natural state, imaging technique like confocal fluorescence microscopy has been applied [17,21]. Combined with extrinsic fluorescent labeling, the spatial distribution of lignin and how lignin associates with other biopolymers can be observed.

In this study, a two-photon hyperspectral imaging (TP-HSI) system [22] has been employed to the study of wood for the first time. Three spruce samples from old instruments and two modern spruce samples were observed. Each HSI stack provides both spatial and spectral information. The optical-section nature of two-photon excitation allows the examination of cellular structures of wood at different depths. In addition to two-photon fluorescence (TPF) of lignin, second harmonic generation (SHG) provides further information about the cellulose distribution [23]. This was the first time that TPF and SHG techniques were used to investigate the lignin and cellulose contents in the spruce woods of the historic instruments. Each spectrum in the HSI stack can be deconvolved into two fluorescence bases and a single SHG base via Gaussian fitting. Through linear unmixing method [24], each HSI stack can be divided into three spectral images, corresponding to three bases. More details about the spatial variation of lignin fluorescence and SHG can thus be obtained.

2. Methods

2.1 System setup

The TP-HSI system employed in this study is based on the non-de-scanned two-photon hyperspectral microscope developed in our laboratory [22]. To investigate the multiple fluorophore types induced by different excitations, two femtosecond lasers with central wavelength at 830 nm (Ti:sapphire laser, Avesta) and 1064 nm (FPL-03UFF0, Calmar Laser) were used. As the system configuration shown in Fig. 1(a), two laser beams were merged into a single light path by a dichroic beamsplitter (DB₁; ZT1064rdc-sp, Chroma). Passing through a galvo mirror (GM; GVSM002/M, Thorlabs), a pair of tube lenses (L₁ and L₂), and an objective (Obj; UPLSAPO 60XW, Olympus), the laser beam was focused onto the sample and scanned along the y-direction. The fluorescence signals arising from the line-shaped excitation area were collected by the same objective and reflected into the spectroscopic system by a dichroic beamsplitter (DB₂; FF801-Di02-25×36, Semrock). The signals were than dispersed along λ-direction by a grating (53067BK01-321R, Newport) and focused onto a CMOS camera (ORCA-Flash4.0 V2, Hamamatsu). For each y-scanning, the CMOS camera recorded a y-λ image, as shown in Fig. 1(b). To get a x-y-λ HSI stack, the sample was moved along the x-direction by a translation stage (TS) to achieve x-scanning. For each HSI stack, x-y images of different wavelength can be obtained (yellow dashed square in Fig. 1(b)) and the field-of-view is 100 µm and 120 µm in the x- and y-direction, respectively. The spatial resolution was measured from the x-y images as ∼504 nm and ∼645 nm under 830-nm and 1064-nm excitation, respectively. The spectral range was 346–792 nm and the spectral resolution was ∼0.7 nm. To make sure the fluorescence intensities of different samples can be compared to one another, the imaging depth was fixed around 60 µm beneath the sample surface.

 

Fig. 1. (a) The system configuration of the TP-HSI system and (b) the 3D HSI stack obtained by the system. Within yellow dashed square is a x-y image at certain wavelength. L: lens; M: mirror; DB: dichroic beamsplitter; G: grating; GM: galvo mirrors; Obj: objective; S: sample; TS: translation stage.

Download Full Size | PDF

2.2 Sample preparation

The historical spruce samples used in this study were removed from old instruments during repairs. There are five different samples (Fig. 2), including old spruce samples taken from a viola made by Amati in 1619 (AS), an English old violin made around 1790 (ES), and a cello made by Stradivari in 1720 (SS), and two control spruce samples taken from modern violin workshops (M3 and M5). The old samples appear darker than the modern ones, which may be caused by aging process. During observations, the samples were mounted with antifade mountant (S36939, Molecular Probes) to suppress photobleaching and preserve the fluorescence signals. Without any staining and sectioning, the samples were observed in their natural state.

 

Fig. 2. The photos of (a) AS, (b) ES, (c) SS, (d) M3, and (e) M5 samples. Scale bar: 3 mm.

Download Full Size | PDF

2.3 Frequency domain spectrum fitting

Due to the effects of natural broadening and Doppler broadening, each spectral line resulted from transition between two electronic states usually has a Gaussian line profile in frequency domain [25]. Based on this characteristic, the measured spectra were transformed into frequency domain first. Then, weighted sum of Gaussian functions was used to do the spectrum fitting. The number of the Gaussian functions was determined as the smallest number with which the R-square value of the fitting results can be higher than 0.995.

3. Results and discussion

3.1 Spatial and spectral information of spruce woods

Figure 3(a)–3(j) show the fluorescence images at 525 nm of spruce samples under 830-nm excitation; while Fig. 3(k)–3(t) are the images at 603 nm obtained under 1064-nm excitation. The wavelengths were chosen around the peak wavelengths of the fluorescence. The brightness of the images had been adjusted to reveal low-level details. The cellular structures can be clearly resolved in these optically-thick samples because of the optical section nature of TP excitation. The wood cell wall consists of multiple sublayers (from inside to outside): secondary cell wall (S1/S2/S3 layers), primary cell wall, and middle lamella (ML). ML is particularly enriched in lignin and serve as adhesives that hold neighboring cells together [12]. In spruce and other conifers, the most abundant cell type is the tracheid cell (TC), which grows longitudinally along the vertical tree axis. TCs have bordered pits (BP) on their side walls for lateral fluid transport [26], regulated by the pit membrane which is made of cellulose, hemicellulose, and pectin [27].

 

Fig. 3. The two-photon excited images of AS, ES, SS, M3, and M5 samples. (a)–(j) The 830-nm excited two-photon images of (a)–(e) the ML and (f)–(j) the TC cell walls and BPs with a wavelength of 525 nm; (k)–(t) the 1064-nm excited two-photon images of (k)–(o) the ML and (p)–(t) the TC cell walls and BPs with a wavelength of 603 nm. Yellow arrows: ML; red arrow: BP; blue star: TC cell wall. Scale bar: 10 µm.

Download Full Size | PDF

From Fig. 3(a)–3(e) and 3(k)–3(o), ML light up as linear structures between the TCs, due to its high concentration (∼70%) of lignin [28,29]. Adjusting the imaging depth by several micrometers, Fig. 3(f)–3(j) and 3(p)–3(t) show the structures of ML, TC cell walls, and BPs (Fig. 3(f) and 3(j) were acquired at different positions). The samples were found to be the tangential sections of the wood because BPs mostly exist at the tangential side of the cell walls. In each sample, ML shows the strong fluorescence intensity due to higher lignin contents [28,30]. The BP contour can be resolved in both old and modern samples. It appears that BP is more easily affected by aging than ML. Besides, it is worth noting the fluorescence intensity is generally higher in the old samples (0.015 ± 0.0033 a.u.) than the modern samples (0.003 ± 0.0009 a.u.) (Fig. 4(c) and 4(d)). Since aging process leads to hemicellulose decomposition and lignin oxidation [7,31], the increased fluorescence may result from the generation of fluorophores or reduced linkages between lignin and hemicellulose [15]. Moreover, chemical treatments which alter the molecular environment and physical structures of lignin could enhance fluorescence in old samples [13].

 

Fig. 4. The two-photon excited average spectra of AS, ES, SS, M3, and M5 samples. (a)(b) The average spectra obtained from ML and BP of 5 samples under (a) 830-nm excitation and (b) 1064-nm excitation; (c)(d) the original peak intensities (red stars) and the peaks and half-maximum bandwidths of the average spectra shown in (a) and (b), respectively. Blue arrows: SHG peaks.

Download Full Size | PDF

As shown in Fig. 3, the fluorescence signals arise mainly from ML and BP. To investigate the spectral variations between these two regions, the average spectra of ML and BPs in different samples were analyzed (Fig. 4). Figure 4(a) shows the normalized average spectra under 830-nm excitation. The old samples’ spectra have an average peak wavelength 530.4 ± 7.17 nm, which shows red shift in contrast to the modern samples’ average peak wavelength, 511.7 ± 3.5 nm. It implies that, in the old samples, lignin fluorescence at longer wavelengths has increased. The original peak intensities are shown in Fig. 4(c) (dark green). Interestingly, old samples (0.02 ± 0.014 a.u.) exhibit stronger peak intensities than modern samples (0.007 ± 0.006 a.u.), both in ML and BP regions. Applying two sample t-test, the p-value between the fluorescence intensities of old and modern samples under 830-nm excitation is p < 0.001, which shows significant difference. Due to the unavailability of samples, 30 individual measurements from old and modern samples (N = 30) were used for t-test. In addition to broad fluorescence peaks, a sharp peak (blue arrow) is observed around 415 nm. This is attributed to SHG signals because of its central wavelength (half of 830 nm) and bandwidth (∼4 nm).

Figure 4(b) shows the normalized average spectra under 1064-nm excitation. The original peak intensities (dark green) and the peak wavelength and bandwidth of each spectrum are shown in Fig. 4(d). Again, the fluorescence intensities of the modern samples are lower than those of the old samples. Applying two sample t-test, a p-value, p < 0.001, between the fluorescence intensities of old (0.014 ± 0.0093 a.u.) and modern (0.004 ± 0.0021 a.u.) samples under 1064-nm excitation shows significant difference. Due to the unavailability of samples, 30 individual measurements from old and modern samples (N = 30) were used for t-test. We noticed some differences compared to 830-nm excitation: 1) the spectra from the modern samples’ ML region behave just like those from the old ones (p = 0.06); 2) the spectra from the modern samples’ BP region have an average peak wavelength 556.8 ± 6.56 nm, much lower than that of the ML region, 598.9 ± 10.37 nm (p < 0.001); 3) the fluorescence intensity is always lower in BP than in ML. This reveals two points: 1) under 1064-nm excitation, the environment differences of lignin between ML and BP can be observed more clearly; 2) whatever the woods experienced, aging process or chemical treatment, the changes occurred in the BP region were more obvious than in the ML region. Besides the fluorescence, the SHG peak (blue arrow) can also be seen around 532 nm.

3.2 Spectral analysis

To obtain insights into the spectral variation, the spectra can be first deconvolved into several spectral bases. The spectrum fitting was carried out in the frequency domain, where both the fluorescence and SHG spectra have a general form of Gaussian function [25]. Performing Gaussian curve fitting on the spectra in Fig. 4(a), under 830-nm excitation, three spectral bases with central wavelengths at 561 nm (534759 Hz; base 1), 507 nm (591716 Hz; base 2), and 415 nm (722233 Hz; base 3) were found. As shown in Fig. 5(a), these three bases fit well with the fluorescence curve – all samples and regions have R-square values higher than 0.998. For the spectra shown in Fig. 4(b), under 1064-nm excitation, three spectral bases with central wavelengths at 606 nm (495050 Hz; base 4), 557 nm (538600 Hz; base 5), and 532 nm (563498 Hz; base 6). The fitting result is shown in Fig. 5(b) – all samples and regions have R-square values higher than 0.995.

 

Fig. 5. The curve fitting results of the average spectra under (a) 830-nm excitation and (b) 1064-nm excitation. The spectra were obtained from M3’s BP region. The spectra under 830-nm and 1064-nm excitation were fitted by bases 1–3 and bases 4–6, respectively.

Download Full Size | PDF

The next step is to deconvolve each average spectrum by these bases. For the spectra in Fig. 5(a) and 5(b), the normalized intensities of different bases are shown in Fig. 6. In all four cases, the two modern maples (M3 and M5) display similar spectra; while old samples clearly differ from modern sample except in the 1064-nm-excited spectra of ML. Referring to base 3 and base 6 (B3 and B6) which represent SHG, the average relative content of SHG (compared to fluorescence) is found to be 10 times higher under 1064-nm excitation (0.13 ± 0.102 a.u.) than 830-nm excitation (0.01 ± 0.009 a.u.), especially in the modern samples’ BP region (0.31 ± 0.175 a.u.). This is strongly related to the higher cellulose contents in the BP region than in the ML region, since SHG is only sensitive to the cellulose but not hemicellulose and lignin [23]. The fluorescence bases under 830-nm excitation (B1 and B2) show that, the older samples have lower B2/B1 ratios (1.02 ± 0.48) than modern samples (2.52 ± 0.82), suggesting a red shift in emission over time. The p-value of the B2/B1 ratios between the old and modern samples is 0.0049, which shows significantly difference. Due to the unavailability of samples, 30 individual measurements from old and modern samples (N = 30) were used for t-test. Especially, the ratio in the SS sample is much lower, 0.43 in ML region and 0.73 in BP region, drastically different from the AS and ES samples. As for 1064-nm excitation, the ratio of B5/B4 in all samples is lower than 1. In ML region, all samples have similar B5/B4 ratio (0.16 ± 0.11). In BP region, the B5/B4 ratio is higher in the modern samples (0.9 ± 0.006) than in the old samples (0.2 ± 0.17). Again, the SS sample has even lower ratio than the AS and ES samples. From the above observations, the fluorescence spectra show clear differences between the modern and old samples. This is most likely attributable to the natural aging process. Furthermore, there are obvious discrepancies between the SS sample (Stradivari cello) and two other old samples, which may suggest that Stradivari applied artificial treatments to the spruce wood.

 

Fig. 6. The relative contents of bases 1–3 in the average spectra of 5 samples under 830-nm excitation and the relative contents of bases 4–6 in the average spectra of 5 samples under 1064-nm excitation.

Download Full Size | PDF

3.2 Spectral images of spruce woods

Employing linear unmixing method to the HSI stacks [24], each image shown in Fig. 3 can be deconvolved into three images, corresponding to three spectral bases. The unmixed results of Fig. 3(f)–3(j) and Fig. 3(p)–3(t) are shown in Fig. 7 and Fig. 8, respectively. The Fluo.1, Fluo.2, and SHG in Fig. 7 refer to bases 1–3, and those in Fig. 8 refer to bases 4–6. The brightness and contrast of the images have been adjusted to reveal more details. However, to maintain the intensity relations between two fluorescence bases, the enhancement performed on all fluorescence (B1, B2, B4 and B5) images is the same. Since SHG intensity (0.09 ± 0.063 a.u.) is much lower than fluorescence (0.56 ± 0.295 a.u.) as shown in Fig. 6, the enhancement for SHG (B3 and B6) images is much stronger than fluorescence images.

 

Fig. 7. The images of (a)–(e) B1 (Fluo.1), (f)–(j) B2 (Fluo.2), and (k)–(o) B3 (SHG) of 5 samples under 830-nm excitation, obtained via linear unmixing. The brightness and contrast of the images were enhanced in order to reveal the details. The enhancement of the B1 and B2 images was the same, while the enhancement of the B3 images was much stronger than the other two image sets. (p)–(t) the added-up images of B1 and B2; (u)–(y) the added-up images of B2 and B3. Scale bar: 10 µm.

Download Full Size | PDF

 

Fig. 8. The images of (a)–(e) B4 (Fluo.1), (f)–(j) B5 (Fluo.2), and (k)–(o) B6 (SHG) of 5 samples under 1064-nm excitation, obtained via linear unmixing. The brightness and contrast of the images were enhanced in order to reveal the details. The enhancement of the B4 and B5 images was the same, while the enhancement of the B6 images was much stronger than the other two image sets. (p)–(t) the added-up images of B4 and B5; (u)–(y) the added-up images of B5 and B6. Scale bar: 10 µm.

Download Full Size | PDF

From the added-up fluorescence (B1+B2) images under 830-nm excitation (Fig. 7), the distribution of B2/B1 ratio can be easily revealed through the color variations. In the AS and ES samples, the ML regions show reddish to yellowish color, while the BP region show greenish color. These indicate the B2/B1 ratio of ML is lower than that of BP. The more reddish color in the SS samples shows B1 content is always higher than B2, whether in the ML or BP regions. Otherwise, the greenish color throughout whole M3 and M5 images shows a B2/B1 ratio higher than unity. From the B2+B3 images, SHG was found to distribute in the BP region and TC cell wall but little in the ML region. Surprisingly, the SS sample shows hardly any SHG signals. From the added-up fluorescence (B4+B5) under 1064-nm excitation (Fig. 8), all the images appear strong reddish color which indicates a B5/B4 ratio lower than unity, as the results shown in Fig. 6. Difference can be only found in the BP region, where the color appears more yellowish in the modern sample than in the old samples. Similar to those under 830-nm excitation, only in the BP region and TC cell wall can SHG signals be observed. Even though the SHG signals can also be found in the SS samples, the structures seem sparser and more fractured than in other samples. Because cellulose is quite stable in antique instruments [7], the reduction of SHG signals in SS cannot be explained for cellulose degradation. It is plausible that the chemical treatments have led to hemicellulose degradation and cleared some space for cellulose rearrangement, such as coalescing to form thicker fibrils [32]. Cellulose rearrangement has been previously implicated in the study of maple wood in Stradivari violins by differential scanning calorimetry [7].

4. Conclusion

In this study, TP-HSI system with both 830-nm and 1064-nm excitation has been applied to unsectioned (optically-thick) and unstained spruces samples for the first time. This system can provide two-photon hyperspectral images with a spatial resolution of ∼500 nm and spectral resolution of ∼0.7 nm. With the optical section nature of two-photon excitation, the tangential-section subcellular structures of wood, including the ML, TC cell wall, and BP, were easily observed at different depths. For each region, the spectral information arising from lignin fluorescence and cellulose SHG was obtained and analyzed. The spectral differences (peak wavelength and bandwidth) were found between the old and modern samples (due to aging) as well as between the ML and BP regions (due to composition). To reveal the sources of these variations, the spectral bases, including two fluorescence bases and SHG, were obtained via Gaussian curve fitting. Then, the relative contents of these bases in different samples’ spectra were analyzed. Furthermore, the spatial distribution of each base was obtained by employing the linear unmixing method. Simply through the color variations of the added-up images, the relative content of the spectral bases can be realized.

Among the old samples, AS (Amati viola) and ES (inexpensive English violin) exhibit similar properties, but the SS specimen (Stradivari cello) is clearly different. It appears that Antonio Stradivari had deliberately manipulated his spruce wood, but we did not observe that with his teacher, Nicolo Amati. Materials engineering of wood may have been one of the genius innovations of Stradivari. We observed a general trend of increased and red-shifted fluorescence emission in aged woods, likely attributed to the natural oxidation of lignin. The unusual reduction of SHG signals in the Stradivari cello may suggest cellulose rearrangement, possibly caused by special wood treatments that promoted hemicellulose decomposition. Cellulose microfibrils are the primary structural support elements in wood, and their physical rearrangement is expected to perturb mechanical properties. However, there is no previous report on wood with diminished SHG signal. Hence, its physical meaning and potential acoustic influences remain unclear and warrant further investigations.