Comparison of Two-Dimensional Fast Raman Imaging versus Point-by

Sep 20, 2012 - The evaluation was based on the comparison of the DS-Avg versus the point-by-point mapping mode in real usage conditions...
0 downloads 0 Views 2MB Size
Download PDF
Article pubs.acs.org/ac

Comparison of Two-Dimensional Fast Raman Imaging versus Pointby-Point Acquisition Mode for Human Bone Characterization Guillaume Falgayrac,*,†,‡ Bernard Cortet,†,‡,∥ Olivier Devos,†,§ Jacques Barbillat,†,§ Vittorio Pansini,†,‡,⊥ Anne Cotten,†,‡,⊥ Gilles Pasquier,†,‡,# Henri Migaud,†,‡,# and Guillaume Penel†,‡ †

Univ Lille Nord de France, F-59000 Lille, France UDSL, EA 4490, PMOI, F-59000 Lille, France § Laboratoire de Spectrochimie Infrarouge et Raman (LASIR CNRS UMR 8516), F-59655 Villeneuve d’Ascq ∥ Department of Rheumatology, ⊥Department of Musculoskeletal Radiology, and #Department of Orthopedic Surgery, Centre Hospitalier Régional Universitaire de Lille, F-59000 Lille, France ‡

S Supporting Information *

ABSTRACT: Recent technical developments gave rise to a new technology for two-dimensional fast Raman imaging: the DuoScan averaging mode (DS-Avg). This technology allows the acquisition of a Raman spectrum over a rastered macro spot. The aim of this study was to evaluate the interest of the DS-Avg applied on trabecular human bone. The evaluation was based on the comparison of the DS-Avg versus the point-by-point mapping mode in real usage conditions. The signal-to-noise ratio, the spectral difference, and the physicochemical parameters were estimated for comparison of the efficiency of both modes. Principal component analysis was performed to explore the capacity of both modes to detect compositional variations. Results showed that the DS-Avg spectrum was equivalent to the average spectrum of individual spectra acquired with the point-by-point mode for the same sample area. The physicochemical parameters can be also determined from DS-Avg acquisition. The DS-Avg combined with an objective ×50 allows a drastic decrease of the acquisition time, but the information about the micrometric composition is lost. The combination of the DS-Avg with an objective ×100 is a good compromise between acquisition time and resolution. The DS-Avg is a useful technology for imaging mineral and organic phases of bones and for assessing their spatial distribution on large samples. The point-by-point imaging mode is more appropriate to assess the heterogeneous composition of bone within the micrometer scale. For the first time, this study compares the DuoScan averaging mode to the point-by-point imaging mode on a trabecular human bone.

B

Combined with an imaging method, Raman microspectroscopy enables the determination of the spatial distribution of composition and physicochemical parameters.13 A variety of methods have been developed to image materials. They can be classified in two groups: as widefield source illumination approaches or scanning approaches.14 In widefield imaging, the entire sample field of view is illuminated and analyzed simultaneously by recording an image at discrete wavenumber increments through the imaging spectrometer.15 This approach is advantageous when a limited spectral range provides sufficient chemical and spatial information. Within the scanning approaches, there are three configurations existing: the point-by-point imaging, the line scanning, and the DuoScan averaging (DS-Avg) modes. In the point-bypoint scanning mode, a laser beam is focused perpendicularly onto the sample surface, and a spectrum from each spatial

one is a heterogeneous material in both composition and structure. The inorganic component is mainly constituted of carbonated apatite. It forms mineral crystals between collagen fibers and undergoes changes in its crystal size, perfection, and composition during maturation of bone.1,2 Type 1 collagen is the most important protein in bone and comprises 90% of the organic matrix.3 Cross-links are formed within and between the collagen fibers, and over time, they change from immature to mature cross-link spontaneously or through catalysis.4−6 These components are self-organized in a specific arrangement that varies as a function of the localization in bone and the in vivo mechanical stress.7 Raman spectroscopy contributes to efforts for a better understanding of bone composition and structure,8 and recent studies support the fact that Raman spectroscopy is becoming a popular tool to investigate properties of healthy and pathological bones.9,10 This technique indeed gives an indication of the bone quality through analyses of its physicochemical parameters (i.e., mineral/organic ratio, type B carbonatation, and crystallinity), since these parameters are strongly correlated with the mechanical properties of bone.11,12 © 2012 American Chemical Society

Received: June 25, 2012 Accepted: September 20, 2012 Published: September 20, 2012 9116

dx.doi.org/10.1021/ac301758y | Anal. Chem. 2012, 84, 9116−9123

Analytical Chemistry

Article

position is collected using a spectrograph. This method gives Raman images with a spatial resolution closed to 1 μm and a spectral resolution under 1 cm−1. The main drawback of the point-by-point scanning is that technology is time-consuming to acquire a Raman mapping. That time could be optimized by increasing the step-size and decreasing the acquisition time per spectrum (under-sampling). Hutchings et al. dealt with these parameters in order to evaluate the potential of rapid Raman imaging to analyze tissue sections of esophagus within a clinically practicable time scale.16 Krafft et al. used the undersampling in a comparative study between Raman and IR spectroscopy on intracranial tumors biopsies.17 In the line-scanning technology, a cylindrical lens is used to create a line-focused laser. As such, only the sample spatial dimension perpendicular to the line-focused laser is scanned by moving mechanically the sample. Data are swept synchronously across the detector as the le line moves across the sample and are read-out continuously. The line-scanning technology offers the advantage of a lateral resolution around 1 μm and a spectral resolution under 1 cm−1. All the applications have reported that the line-scanning technology is the fastest method for acquiring spectral information at a reasonable spatial resolution.18−21 The trade-off in improving acquisition time is in the decrease of both the signal-to-noise ratio (SNR) and quality of the reconstructing Raman image. Recently, a new method was developed for two-dimensional (2D) fast Raman imaging: the DuoScan technology (HORIBA, Jobin Yvon).22 This technology is based on a combination of scanning mirrors that scan the laser beam across a pattern chosen by the operator. The averaging mode consists in generating a rastered “macro laser spot” for large area mapping. Depending on the objective ×10, × 50, or ×100, the rastered macro spot varies from 300 to 3 μm, respectively. To our knowledge, there is only one study of Aina et al. that compared the DS-Avg with the transmission configuration when investigating polymorphic content of pharmaceutical formulations.23 Recently, the DS-Avg was also used to analyze the structure of carbonaceous materials and to identify phase transitions in molecular solids.24,25 The aim of this study is to evaluate the relevance of this new DuoScan averaging technology for fast Raman imaging applied to human bone samples in real usage conditions. The DS-Avg is compared to the point-by-point imaging mode, based on the SNR, the spectral difference, the physicochemical parameters, and the principal component analysis (PCA). Advantages and limitations of the DS-Avg are discussed in order to evaluate which relevant information can be extracted by this technique applied to human bones.

Figure 1. (a) MRI of the necrosed femoral head; the red shape shows the extent of the necrotic area. The white dots localize the sampling area: necrotic (N), inflammatory (I), and distant (D) biopsies. (b) Representation of point-by-point mapping mode: each red dot represents a spectrum acquired at the position (x, y) for a step of 1 μm. (c) Representation of DuoScan averaging mode: the area of 30 × 30 μm2 is sampled in one acquisition.

between the necrotic and distant biopsy. Biopsies were stored at −80 °C until sample preparation and analyses. The pathological (N) and healthy (I and D) biopsies are only used for the evaluation of the efficiency of the DS-Avg mode. Spectral features in relation with this pathology are not discussed in this work. Each biopsy had a diameter of 1 cm and a length of 3 cm. A transversal slice of ∼5 mm thickness was cut from each biopsy. The slice was fixed in an ethanol solution at 70% during 48 h and then set on a microscope slide with Araldite (Huntsman Advanced Materials Bâle, Switzerland). The sample in Araldite was polished using abrasive papers with a decreasing grain size (30, 3, and 0.3 μm). Raman Acquisitions. Raman spectra were acquired with a Raman microspectrometer LabRAM HR800 (Jobin-Yvon, France) provided with the DuoScan technology. The instrument is equipped with a XYZ motorized stage and a diode laser at 785 nm. The acquisitions were done on an objective ×50 or ×100 (numerical aperture 0.75 or 0.9, respectively). The acquisition time was set at 5 s. The 300 g/mm grating allows a spectral acquisition in the 300−1700 cm−1 range, and the spectral resolution was 4 cm−1. Raman spectra were processed using Labspec software (HORIBA, Jobin-Yvon, France). A Savitzky−Golay smoothing filter (filter width: 7; and polynomial order: 2) and polynomial baseline correction (degree 4) were applied to data set prior to the evaluation of SNRν, difference spectrum, physicochemical parameters, and the analysis by PCA. Imaging System. The instrument allows the acquisition of Raman mapping according to two modes. In the point-by-point imaging mode (Pt-Img), the laser beam is focused perpendicularly according to the sample surface. The laser beam is stepped in two dimensions (x and y) with a spectrum being recorded at each position (x, y) (red dots in Figure 1b). The PtImg mode generates x × y spectra. A scanning mirror is added on the laser pathway to switch from the Pt-Img to the DuoScan averaging mode (DS-Avg). The DS-Avg mode generates a rastered “macro laser spot” for large area mapping (Figure 1c). In this mode, the laser beam is focused perpendicularly only at the center of the surface analyzed. For the rest of the surface analyzed, the laser beam reaches the sample surface with different angles of incidence. The DS-Avg mode gives a single representative spectrum of the rastered macro spot. The size of the rastered macro spot depends on the objective. The maximum size for × 50 and ×100 objectives is a square of 30 × 30 μm2 and 3 × 3 μm2, respectively. The confocality of the system is preserved with this mode. The Raman signal (in DS-Avg or Pt-Img mode) is collected in confocal mode and

■

MATERIALS AND METHODS Bone Samples. Human bone samples were obtained after prosthetic joint replacement surgery of the hip. The surgery was undergone on a male of 53 years old in the orthopedic surgery unit of the Lille University Hospital. This protocol was done under the agreement of ethics committee. Three bone biopsies were harvested from the resected femoral head, immediately after the surgery. The localization of each biopsy was based on the analysis by magnetic resonance imaging (MRI) and the diagnostic of radiologists (Figure 1a). The necrotic biopsy (N) was sampled in the necrosis area (pathologic sample). The distant biopsy (D) was sampled near the femoral neck and considered as healthy according to radiological observations (healthy sample). The inflammatory biopsy (I) was sampled 9117

dx.doi.org/10.1021/ac301758y | Anal. Chem. 2012, 84, 9116−9123

Analytical Chemistry

Article

SNRν was calculated for each band of interest with a customized function under Matlab software (v7 R2010a). The bands of interest are the main bands of the mineral compound (427, 588, 960, and 1070 cm−1) and the main bands of the organic matrix (1244, 1270, 1450, and 1670 cm−1).9 The comparison of DS-Avg mode and Pt-Img mode is based on the mean SNRν calculated over the first 10 spectra (of the 450) and the 900 individual spectra, respectively. The use of the 450 spectra of DS-Avg could bias the results due to the high number of spectra, and it will not be representative of the standard usage. Only the first 10 DS-Avg spectra are selected to be close to the experimental parameters of the standard usage and take into account the reproducibility of the measure. Difference Spectrum. Raman spectra from point-by-point imaging (Pt-Img) and DuoScan averaging (DS-Avg) mode were compared by subtracting their respective spectrum. The mean spectrum was evaluated for both modes and each couple of data. In addition to data preprocessing mentioned in the Raman acquisition section, all mean spectra were normalized with the respect to the band at 960 cm−1. The difference spectrum was calculated by subtracting the mean spectrum of DS-Avg from the mean spectrum of Pt-Img in the 300−1700 cm−1 range. For the objective ×50, the Pt-Img mean spectrum was calculated on the average of 900 spectra. The DS-Avg mean spectrum was calculated on the first 10 spectra among 450, for the same reason mentioned in the Signal-to-Noise Ratio section. For the objective ×100, Pt-Img and DS-Avg mean spectra were calculated from the average of nine spectra. Physicochemical Parameters. The physicochemical parameters were evaluated and compared between DS-Avg and Pt-Img acquisition modes. The mineral/organic ratio was calculated by dividing the intensity of the phosphate band at 960 cm−1(ν1PO4) with vibrations resulting primarily from the CH2 side chains of collagen molecules (1450 cm−1). Similarly, type-B carbonate substitution was calculated between the respective intensities of the type-B carbonate band (1070 cm−1) and the phosphate band (ν1PO4). Crystallinity was calculated as the inverse of the full-width-at-half-maximum (fwhm) of the phosphate band (ν1PO4).27 The fwhm was evaluated by the Gaussian fitting function implanted in the PLS Toolbox (v6.7 eigenvector Research, Inc., USA). The physicochemical parameters were evaluated at 10 different positions of the biopsy with the objectives ×50 and ×100. For each position and parameter, the mean value in the DS-Avg mode was determined over 10 spectra, and the mean value in the Pt-Img mode was determined over 900 spectra. The physicochemical parameters were compared by statistical analysis. Statistical Analysis. A test of Mann−Withney U test (nonparametric) was used to evaluate if the physicochemical parameters from DS-Avg mode were different to Pt-Img mode. For each biopsy, 10 couples of data set were obtained. Each physicochemical parameter was tested among the 10 couples per biopsy. For example, the mineral/organic ratio from the DS-Avg and Pt-Img was compared over the 10 couples of data from the necrotic biopsy. The procedure was repeated for the type-B carbonate substitution and crystallinity for the necrotic biopsy. The same comparisons were also done for the inflammatory and distant biopsies. The difference was considered as significant at the level of p ≤ 0.05. Principal Component Analysis. In addition to preprocessing steps mentioned in the Raman Acquisitions section, each data set composed by DS-Avg and Pt-Img spectra (objectives ×50 or ×100) were mean centered in order to focus

focused on the same point of the entrance slit. Thus, the same region of the detector is always illuminated either in Pt-Img mode or DS-Avg mode. It is also assumed that the detector read noise is almost equal between the two modes because the Raman signal is read only once per spectrum either in Pt-Img mode or in DS-Avg mode (at equivalent laser power and acquisition time). The Pt-Img and DS-Avg modes were successively run to ensure analyses of an identical area. Spectrometer settings were kept identical across all measurements. First, the acquisition was carried out with the ×50 objective. The acquisition with the Pt-Img mode was run with a step of 1 μm and a surface delimited by a square of 30 × 30 μm2. Then, the DS-Avg mode was set to acquire one spectrum over a square of 30 × 30 μm2 with the same objective. The same spectrum is acquired at exactly the same position 450 times. This protocol generated one couple of data sets that were composed of 900 Raman spectra from the Pt-Img mode and 450 spectra from the DSAvg mode. The same protocol was repeated at 10 different positions of the bone surface of each biopsy in order to have 10 couple of data sets, and this was done for the 3 biopsies. The protocol described above was repeated with the objective ×100. The Pt-Img was run with a step of 1 μm and a surface delimited by a square of 3 × 3 μm2. Then, the DS-Avg was run to acquire one spectrum over a square of 3 × 3 μm2, at the same position. The same spectrum was acquired nine times. This procedure gave one couple of data sets composed of nine spectra from Pt-Img mode and nine spectra from DS-Avg mode. This procedure was also repeated at 10 different positions of the biopsy and for the 3 biopsies. The analyses generated 30 couples of data sets for the objective ×50 and 30 couples of data sets for the objective ×100. The experimental parameters are summarized in Table 1. Table 1. Description of One Couple of Data Sets for Objectives ×50 and ×100a objective × 50 size no. of spectra time/ spectrum

objective × 100

Pt-Img

DS-Avg

Pt-Img

DS-Avg

30 × 30 μm2, 1 μm step 900

square 30 × 30 μm2 450

3 × 3 μm2, 1 μm step 9

square 3 × 3 μm2 9

5s

5s

5s

5s

a

Comparison of the imaging settings between Pt-Img and DS-Avg for each objective. The number of spectra corresponds to the number of the individual spectrum.

Signal-to-Noise Ratio. The signal-to-noise ratio (SNR) is defined as the ratio between mean intensity to the standard deviation (SD) of the peak height, at a given wavenumber:26 SNR ν =

Imean(ν) SD(ν)

(1)

where Imean(ν) is the mean intensity and SD(ν) is the standard deviation at a given wavenumber (ν). A high SNRν ensure a good spectrum quality. This definition is applicable on a homogeneous sample. However, for a heterogeneous sample, the SD is not only representative of the signal variations from the experimental set up but also due to the heterogeneity of the chemical composition within the sample. Therefore the SNRν will be calculated to evaluate the both influences. 9118

dx.doi.org/10.1021/ac301758y | Anal. Chem. 2012, 84, 9116−9123

Analytical Chemistry

Article

Figure 2. SNRν as a function of bands of interest for the necrotic biopsy (objective ×50): comparison between (a) DS-Avg mode and (b) Pt-Img mode. SNRν as a function of bands of interest for the necrotic biopsy (objective ×100): comparison between (c) DS-Avg mode and (d) Pt-Img mode.

on spectral differences. For the objective ×50, a couple of data set was composed by 900 spectra (Pt-Img) and 450 spectra (DS-Avg). For the objective ×100, a couple of data was composed by 9 spectra (Pt-Img) and 9 spectra (DS-Avg). The PCA analysis allowed the evaluation of difference or similar spectral features between both methods of acquisitions. The PCA was carried out using PLS Toolbox (v6.7 eigenvector Research, Inc., USA).

range for the necrotic biopsy (Figure 3). The Pt-Img mean spectrum represents the average of 900 spectra. The DS-Avg

■

RESULTS Comparison of the SNRν Values for Objectives ×50 and ×100. The SNRν values were calculated for each couple of data according to eq 1. Spectra were acquired in DS-Avg mode with the objective ×50 and over an area of 30 × 30 μm2. Each point is the average of 10 DS-Avg spectra (Figure 2a). The SNRν values are in the range of 7−162 for the bands of interest. The acquisitions in Pt-Img mode were done with the objective ×50 and over an area of 30 × 30 μm2 (Figure 2b). Each point represents the mean value over 900 spectra. The SNRν values from the Pt-Img mode are between 8 and 22; maximum values are much lower than those obtained with the DS-Avg mode, even for the band at 960 cm−1. The same protocol was carried out with the objective ×100 on the necrotic biopsy at 10 different positions. Spectra were acquired in the DS-Avg mode over an area of 3 × 3 μm2. Each point is the mean of 9 DS-Avg spectra (Figure 2c). The SNRν values are in the range of 9−90. The acquisitions in the Pt-Img mode were done over an area of 3 × 3 μm2 (Figure 2d). The SNRν values from the Pt-Img mode show similar trends comparatively to DS-Avg. The band at 960 cm−1 has variations within the same range of 23−92. The bands at 427, 588, 1070, 1244, 1270, 1450, and 1670 cm−1 have SNRs values in the range 7−30. The same comparison is done on SNRν measurements from inflammatory and distant biopsies for the objectives ×50 and ×100 (Supporting Information, Figures S1 and S2). The same trends are observed for both objectives and biopsies. Difference Spectrum. The differences between both methods of acquisition were evaluated on the whole spectral

Figure 3. (a) Comparison of the mean spectra obtained on the necrotic biopsy in the DS-Avg and Pt-Img modes and the difference spectrum (objective ×50). (b) The difference spectrum (objective ×50). (c) Comparison of the mean spectra obtained on the necrotic biopsy in the DS-Avg and Pt-Img modes and difference spectrum (objective ×100). (d) The difference spectrum (objective ×100).

mean spectrum represents the average of 10 spectra. Mean spectra were normalized in respect to the intensity of the band at 960 cm−1. Mean spectra are quite similar (Figure 3a). This is confirmed by the difference spectrum that is low comparatively to the mean spectra (Figure 3b). The principal difference concerns the most intense band of the spectrum, at 960 cm−1, and represents around 4% of the total intensity. The difference values are much lower for the rest of the difference spectrum, which are under 1%. Spectral shifts in wavenumber are observed at 430, 586, 960, and 1067 cm−1. Those results are 9119

dx.doi.org/10.1021/ac301758y | Anal. Chem. 2012, 84, 9116−9123

Analytical Chemistry

Article

1303, 1441, 1635, 1656, and 1747 cm−1 (Supporting Information, Figures S13 and S14).28,29 The contribution of lipids are observed on 13 data sets among the 30 data sets analyzed by PCA. PC2 does not separate both groups. The PCA was carried out on 30 couples of data sets, built from acquisition of both methods and the objective ×50. The Figure 5 is representative of all PCA analysis carried out on each couple of data acquired with an objective ×50 (Supporting Information, Figures S12−S14). Among the 30 data sets composed of Raman spectra from Pt-Img and DS-Avg acquisitions with objective ×50, the PC1 score was representative of at least 75% of the variance. The PCA was also carried out on the 30 couples of data sets, built from acquisitions of both methods and the objective ×100. The PC1/PC2 score plots and loadings are similar to Figure 5c and d. The PC1/PC2 score plots show two distinct groups of points corresponding to DS-Avg and Pt-Img acquisitions (Supporting Information, Figures S15−S17).

representative of the 10 different positions of the necrotic biopsy. Similar results were observed for the acquisitions done with the objective ×100 on the necrotic biopsy (Figure 3c and d). The difference spectrum represents lower than 0.5% of the total intensity all over the spectral range. Identical acquisitions were carried out on the inflammatory and distant biopsies and gave the same results (Supporting Information, Figures S3− S8). Comparison of Physicochemical Parameters. The physicochemical parameters were evaluated at 10 different positions of the biopsy with the objectives ×50 (Figure 4). The

■

DISCUSSION The aim of this work was to evaluate a recent technology for 2D fast Raman imaging: the DuoScan averaging system (DSAvg). In the present study, that technology is applied to the analysis of human trabecular bone. The DS-Avg mode is compared with the point-by-point imaging (Pt-Img). This evaluation is based on the comparison of the following parameters: the SNRν, Raman spectra difference, physicochemical parameters, and PCA. The SNRν were calculated according to eq 1. The SNRν are higher with the objective ×50 in the DS-Avg mode, compared to the Pt-Img mode, especially for the band at 960 cm−1. In the Pt-Img mode, SD is related to the heterogeneous composition of bone, since 900 spectra correspond to 900 different points of analyses. In the DS-Avg mode, SD is related to the reproducibility of measurements, since the 10 spectra correspond to the same point. SD values of the Pt-Img mode with the objective ×50 are higher than those obtained in the DS-Avg mode (inset Figure 5a and b). Thus, the low SNRν in Pt-Img are due to the heterogeneous composition of bone. This result is observed for all measurements made with the objective ×50 for the three biopsies. As a consequence, the DS-Avg mode with objective ×50 is not representative of micrometric variations that are observed in the Pt-Img mode. With the objective ×100, SNRν varies within the same range. This result make sense because the surface analyzed in the DSAvg mode (3 × 3 μm2) is close to the surface analyzed by a single point in the Pt-Img mode (1 × 1 μm2). Thus, the DSAvg analysis with objective ×100 seems to be more close to the micrometric variations observed in Pt-Img with the same objective. The comparison of normalized mean Raman spectra by subtraction allows analyses on the entire spectral window (300−1700 cm−1). Concerning the objective ×50, the difference spectrum is near 4% for the band at 960 cm−1, around 1% for all other bands, and below 0.1% for the rest of the spectrum. This result is observed for all analyses performed on the three biopsies. Spectral shifts represented by the curve with “S shape” are observed at 430, 586, 960, and 1067 cm−1 on all difference spectra. These shifts are not due to spectral instabilities of the laser wavelength. Similar spectral features were observed on difference spectra obtained from acquisitions done with a stable laser He−Ne (data not shown). The origin of the shift may arise from instabilities of the spectrometer. Nevertheless, the

Figure 4. Mean values of the physicochemical parameters evaluated in 10 positions of the necrotic biopsy (objective ×50). The mineral/ organic ratio, type-B carbonate substitution, and crystallinity parameters are presented with their respective standard deviation (SD).

mineral/organic ratio acquired in the Pt-Img mode is close to that of the DS-Avg mode for each position. Similar observations are made for the type-B carbonate substitution and crystallinity parameters. These results are confirmed by statistical tests with a significant threshold set at p = 0.05. No statistical difference is obtained between the two methods of acquisition. This observation was also validated for spectra acquired with objective ×100 and the three biopsies (Supporting Information, Figures S9−S11). Principal Component Analysis. The comparison of both systems was made by PCA on each couple of data acquired with the objective ×50. The data set was composed of 450 and 900 spectra acquired in the DS-Avg and Pt-Img modes, respectively. Figure 5a and b represents their mean spectrum with their respective standard deviation (SD). The SD of the point-by-point imaging mode (Pt-Img) is higher than the SD of the DuoScan Averaging mode (DS-Avg). Figure 5c and d represents the score plot of PC1/PC2 for a data set from necrotic area and the loading plots of the two first principal components (PC1 and PC2). PC1 represents 92.52% of the variance of the data set. PC1 corresponds to the Raman spectrum of bone due to the presence of mineral bands (427, 588, 960, and 1070 cm−1) and organic bands (1244, 1270, 1450, and 1670 cm−1). PC1 separates the group DS-Avg from the group Pt-Img. Raman spectra acquired in Pt-Img mode show a high range of scores on PC1, compared to DS-Avg mode. PC2 represent 2.81% of the variance. Shifts in wavenumber are observed at 430, 586, 960, and 1067 cm−1 in PC2. These shifts are the same as in the difference spectra. PC2 has also contributions of residual lipids at 1064, 1261, 9120

dx.doi.org/10.1021/ac301758y | Anal. Chem. 2012, 84, 9116−9123

Analytical Chemistry

Article

Figure 5. Mean spectrum from (a) DS-Avg acquisition and (b) Pt-Img acquisition; PCA applied on data Pt-Img and DS-Avg from necrotic biopsy (objective ×50). (c) Score plot for PC1/PC2; (d) loading plot for PC1/PC2.

spectral features and correspond to the bone spectrum. PC2 shows shifts in the wavenumber that are similar from that obtained in the difference spectra. The score plot shows two distinct groups of points corresponding to DS-Avg and Pt-Img acquisitions. The group of Pt-Img is more scattered compared to DS-Avg. This behavior is inherent to the Pt-Img acquisition and the heterogeneous composition of bone, as we saw for the SNRν evaluation. The same trend was obtained by PCA on data sets built from acquisitions with the objective ×100. The results confirm that DS-Avg acquisitions are not representative of the heterogeneity at the micrometer scale, compared to Pt-Img. This feature has to be taken into account when the DS-Avg mode is used to analyze heterogeneous samples. In the literature, the comparison of fast Raman acquisition technologies was mainly done between the line-scanning and wide-field approaches, since the DuoScan technology is more recent. Several comparative studies were done on using various homogeneous samples. The performance of the line-scanning method was evaluated by comparison with the point-focus method.30 Four conventional spectrometers were compared. One of them was equipped with the line-scanning technology. The tested sample was a glassy carbon. Spectrometers have different collection efficiency and sampling optics. The Figure of Merit (FSNR) and SNR were determined for each spectrometer to aid the comparison. The conclusion was that a spectrometer capable of monitoring a larger sample area will yield higher SNR for a given power density. Another study compared the three methodologies (point-by-point, Linescan, and wide-field methods) using a single experimental configuration. The tested sample was composed of 5 μm squares of silicon. The comparison was based on the spectral SNR, spatial

spectral differences are very low. These results show that the DS-Avg spectrum corresponds to the mean of 900 individual spectra of the same area in spite of the bone heterogeneity and spectral shifts. It represents an advantageous gain of time. The DS-Avg spectrum is acquired in 5 s against 75 min for the PtImg mode. Concerning the objective ×100, the difference between the DS-Avg mode and the Pt-Img mode is under 1% of the total intensity. The gain in acquisition time remains interesting, because the same area is scanned in 5 s with the DS-Avg mode instead of 45 s with the Pt-Img mode. The difference spectrum for the objective ×50 is greater than the difference spectrum for the objective ×100. This difference is due to the analyzed surface. The surface analyzed with objective ×50 is 100 times greater than the surface analyzed with the objective ×100. The objective ×50 is more likely to probe mature and remodeling trabecular bone within the same area (30 × 30 μm2) compared to the objective ×100 (3 × 3 μm). The physicochemical parameters obtained with DS-Avg are statistically not different from that obtained with the Pt-Img acquisition. This result was checked on 10 different positions of the necrotic, inflammatory, and distant biopsies. The physicochemical parameters are more and more used, since it exists correlations with mechanical properties of bone.12 The DS-Avg mode would permit to screen rapidly large bone samples and identify a relevant zone that needs accurate measurements. The comparison of the two modes was also carried out by PCA on data sets built from acquisitions with the objective ×50. PC1 captures the majority of the variance and corresponds to the spectrum of calcified tissue. The results confirm that DS-Avg and Pt-Img spectra share the same 9121

dx.doi.org/10.1021/ac301758y | Anal. Chem. 2012, 84, 9116−9123

Analytical Chemistry

■

Article

CONCLUSIONS This study compares the 2D fast Raman imaging technology the DuoScan averaging mode to the point-by-point imaging. To the author knowledge, this comparison is carried out for the first time on human biopsies from the femoral head. Bone represents a complex and heterogeneous sample. The DS-Avg mode gives an average spectrum characteristic of bone that is equivalent to the average spectrum acquired by the point-bypoint mode, over the same area. In addition, the physicochemical parameters can be also evaluated by DuoScan. The DS-Avg mode offers very useful capabilities for imaging both mineral and organic phases of bone and for assessing their spatial distribution over large zones of interest. It allows a drastic decrease of the data acquisition time and has thus made it possible to collect Raman signal over millimeter-sized samples. The trade-off is the loss of the information of the composition within the micrometric scale. The DS-Avg mode is suitable to assess the composition of bone tissues within the osteon scale (50−100 μm). The point-y-point imaging is suitable to assess the composition of bone tissues within the lamellae scale (5−7 μm). This 2D fast Raman imaging technique is particularly suitable for the characterization of large and heterogeneous calcified samples.

resolution, and data acquisition time. The SNR of point-bypoint imaging and line-scanning method are similar. The two main conclusions were the Linescan method represents the fastest method for acquiring spectral information at a reasonable spatial resolution (∼1 μm) while yielding reconstructed Raman images of moderate quality. The widefield Raman imaging approach is the method of choice for obtaining submicrometer spatially resolved high-quality images from flat samples, within a reasonable time length. The widefield approach needs to know a priori the composition of the sample and the position of well-defined Raman bands.26 More recent comparative studies presented the evaluation of fast Raman imaging using heterogeneous samples. The linescanning method was evaluated on complex geological samples, containing mineral and organic phases.19,20 The line-scanning method allows a drastic decrease of acquisition time without losing either spatial or lateral resolution. Thus, it makes possible the analysis of millimeter-sized and heterogeneous samples. The line-scanning method was also evaluated to assess the homogeneity of drug in polymer formulation.18 Recently, the under-sampling method was compared with point-by-point method on esophageal biopsies.16 As expected, the undersampling method is faster than the point-by-point method. The DS-Avg was compared with transmission Raman spectroscopy (TRS) to study polymorphic content in a model pharmaceutical formulation.23 To our knowledge, this work represents the only comparison of the DS-Avg (objective ×50) with another system found in the literature. The assessment was done on the performances of partial least square modeling and model-free analysis. The conclusions were in favor of the TRS because it yields to a more robust model. The DS-Avg spectrum does not reflect the variability in the bulk composition of the sample compared to TRS. This issue is inherent to the backscattering configuration. In our study, the results showed that DS-Avg (objective ×50) did not reflect the variability of the human trabecular bone composition at the micrometer scale. This issue is inherent to the resolution of DSAvg with an objective ×50 and the heterogeneity of human bone at the micrometer scale. The lateral resolution of the Raman spectrometer used to be either the micrometer or millimeter scale. The IR techniques were able to analyze from large areas down to 30 × 30 μm2, but the micrometer resolution was not achievable. Nowadays, these differences between Raman and IR techniques do not stand anymore, in regard to lateral resolution. A synchrotron source permits achievement of micrometer resolution with the IR radiation. The technical development of Raman spectrometers gives the rise of techniques that allow spectral acquisitions of large area. The DS-Avg mode enables the acquisitions of area from 3 × 3 μm2 (with an objective ×100) to 300 × 300 μm2 (with an objective ×10). The main advantage observed from this study is the gain of acquisition time. The DS-Avg acquisition time is 900 times shorter than that of the Pt-Img, with objective ×50 and a step of 1 μm over an area of 30 × 30 μm2. With an objective ×100, the DS-Avg acquisition time is nine times shorter than that of the Pt-Img, over an area of 3 × 3 μm2. The trade-off is the loss of the spectral information relative to the compositional heterogeneity of human bone sample because the DS-Avg does not permit to retrieve individual spectra (as in Pt-Img mode). For the heterogeneous sample, since the surface analyzed is larger than Pt-Img, the DSAvg spectrum will be more likely complex (superposition of multiple spectral species) than the point spectrum.

■

ASSOCIATED CONTENT

S Supporting Information *

Additional information as noted in text. This material is available free of charge via the Internet at http://pubs.acs.org.

■

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS The authors want to thank the University of Lille 2, the “Institut de Médecine Prédictive et de Recherche Thérapeutique”, and the “Region Nord-Pas-de-Calais” for their financial supports.

■

REFERENCES

(1) Akkus, O.; Adar, F.; Schaffler, M. B. Bone 2004, 34, 443−453. (2) Boskey, A. Osteoporosis Int. 2003, 14, 16−21. (3) Weiner, S.; Traub, W.; Wagner, H. D. J. Struct. Biol. 1999, 126, 241−255. (4) Eyre, D. R.; Koob, T. J.; Van Ness, K. P. Anal. Biochem. 1984, 137, 380−388. (5) Paschalis, E. P.; Shane, E.; Lyritis, G.; Skarantavos, G.; Mendelsohn, R.; Boskey, A. L. J. Bone Miner. Res. 2004, 19, 2000− 2004. (6) Yamauchi, M.; Chandler, G. S.; Tanzawa, H.; Katz, E. P. Biochem. Biophys. Res. Commun. 1996, 219, 311−315. (7) Wagermaier, W.; Gupta, H. S.; Gourrier, A.; Burghammer, M.; Roschger, P.; Fratzl, P. Biointerphases 2006, 1, 1−5. (8) Falgayrac, G.; Facq, S.; Leroy, G.; Cortet, B.; Penel, G. Appl. Spectrosc. 2010, 64, 775−780. (9) Morris, M.; Mandair, G. Clin. Orthop. Relat. Res. 2010, 1−10. (10) Gamsjaeger, S.; Buchinger, B.; Zwettler, E.; Recker, R.; Black, D.; Gasser, J. A.; Eriksen, E. F.; Klaushofer, K.; Paschalis, E. P. J. Bone Miner. Res. 2011, 26, 12−18. (11) Donnelly, E.; Chen, D. X.; Boskey, A. L.; Baker, S. P.; van der Meulen, M. C. Calcif. Tissue Int. 2010, 87, 450−460. 9122

dx.doi.org/10.1021/ac301758y | Anal. Chem. 2012, 84, 9116−9123

Analytical Chemistry

Article

(12) Burket, J.; Gourion-Arsiquaud, S.; Havill, L. M.; Baker, S. P.; Boskey, A. L.; van der Meulen, M. C. H. J. Biomech. 2011, 44, 277− 284. (13) Donnelly, E.; Boskey, A. L.; Baker, S. P.; van der Meulen, M. C. J. Biomed. Mater. Res., Part A 2010, 92, 1048−1056. (14) Chalmers, J. M.; Griffiths, P. R. Handbook of Vibrational Spectroscopy; Wiley: Chichester, U.K., 2001. (15) Stewart, S., Panza, J., Drauch, A. In Vibrational Spectroscopic Imaging for Biomedical Applications; Srinivasan, G., Ed.; McGraw-Hill: Columbus, OH, 2010. (16) Hutchings, J.; Kendall, C.; Smith, B.; Shepherd, N.; Barr, H.; Stone, N. J. Biophotonics 2009, 2, 91−103. (17) Krafft, C.; Sobottka, S. B.; Schackert, G.; Salzer, R. J. Raman Spectrosc. 2006, 37, 367−375. (18) Brown, S. C.; Claybourn, M.; Sievwright, D.; Fearnside, V.; Ashman, C. Appl. Spectrosc. 2010, 64, 442−447. (19) Bernard, S.; Beyssac, O.; Benzerara, K. Appl. Spectrosc. 2008, 62, 1180−1188. (20) Carter, E.; Pasek, M.; Smith, T.; Kee, T.; Hines, P.; Edwards, H. Anal. Bioanal. Chem. 2010, 397, 2647−2658. (21) Hutchings, J.; Kendall, C.; Shepherd, N.; Barr, H.; Stone, N. J. Biomed. Opt. 2010, 15, 066015−066010. (22) Horiba Jobin-Yvon. Technical note: RA-TN04: DuoScan Imaging, 2012. (23) Aina, A.; Hargreaves, M. D.; Matousek, P.; Burley, J. C. Analyst 2010, 135, 2328−2333. (24) Schmid, J.; Grob, B.; Niessner, R.; Ivleva, N. P. Anal. Chem. 2011, 83, 1173−1179. (25) Alkhalil, A.; Nanubolu, J. B.; Burley, J. C. RSC Adv. 2012, 2, 209−216. (26) Schlücker, S.; Schaeberle, M. D.; Huffman, S. W.; Levin, I. W. Anal. Chem. 2003, 75, 4312−4318. (27) Turunen, M. J.; Saarakkala, S.; Rieppo, L.; Helminen, H. J.; Jurvelin, J. S.; Isaksson, H. Appl. Spectrosc. 2011, 65, 595−603. (28) Lakshmi, R. J.; Alexander, M.; Kurien, J.; Mahato, K. K.; Kartha, V. B. Appl. Spectrosc. 2003, 57, 1100−1116. (29) Penel, G.; Delfosse, C.; Descamps, M.; Leroy, G. Bone 2005, 36, 893−901. (30) Ramsey, J.; Ranganathan, S.; McCreery, R. L.; Zhao, J. Appl. Spectrosc. 2001, 55, 767−773.

9123

dx.doi.org/10.1021/ac301758y | Anal. Chem. 2012, 84, 9116−9123