Article pubs.acs.org/ac
Raman Spectroscopic Studies on Screening of Myopathies Rekha Gautam,† Sandeep Vanga,† Aditi Madan,‡ Narayanappa Gayathri,§ Upendra Nongthomba,*,‡ and Siva Umapathy*,†,∥ †
Department of Inorganic and Physical Chemistry, ‡Department of Molecular Reproduction, Development and Genetics, and Department of Instrumentation and Applied Physics, Indian Institute of Science, Bangalore, 560012, India § National Institute of Mental Health and Neuro Sciences (NIMHANS), Bangalore, 560029, India ∥
S Supporting Information *
ABSTRACT: Myopathies are among the major causes of mortality in the world. There is no complete cure for this heterogeneous group of diseases, but a sensitive, specific, and fast diagnostic tool may improve therapy effectiveness. In this study, Raman spectroscopy is applied to discriminate between muscle mutants in Drosophila on the basis of associated changes at the molecular level. Raman spectra were collected from indirect flight muscles of mutants, upheld1 (up1), heldup2 (hdp2), myosin heavy chain7 (Mhc7), actin88FKM88 (Act88FKM88), upheld101 (up101), and Canton-S (CS) control group, for both 2 and 12 days old flies. Difference spectra (mutant minus control) of all the mutants showed an increase in nucleic acid and β-sheet and/or random coil protein content along with a decrease in α-helix protein. Interestingly, the 12th day samples of up1 and Act88FKM88 showed significantly higher levels of glycogen and carotenoids than CS. A principal components based linear discriminant analysis classification model was developed based on multidimensional Raman spectra, which classified the mutants according to their pathophysiology and yielded an overall accuracy of 97% and 93% for 2 and 12 days old flies, respectively. The up1 and Act88FKM88 (nemaline-myopathy) mutants form a group that is clearly separated in a linear discriminant plane from up101 and hdp2 (cardiomyopathy) mutants. Notably, Raman spectra from a human sample with nemaline-myopathy formed a cluster with the corresponding Drosophila mutant (up1). In conclusion, this is the first demonstration in which myopathies, despite their heterogeneity, were screened on the basis of biochemical differences using Raman spectroscopy.
M
effectively.4,8 Although there are numerous studies on the basis of genetics, a complete understanding of the pathophysiology from a molecular point of view is still lacking, which hampers the development of a suitable therapy. Furthermore, since there is no effective and complete cure for any of the myopathies or dystrophies, a sensitive and early diagnosis of these disorders is a necessity.4 Therefore, a fast, reliable, sensitive, and specific screening technology for disease detection is needed. Raman spectroscopy, a label free, noninvasive, and multiplexing modality, is an emerging analytical technique in biomedical research.9−12 As an analytical tool, Raman spectroscopic methods provide molecular structure and conformationdependent spectral markers of the chemical components in a heterogeneous sample. This method is based on the principle of inelastic scattering of monochromatic (laser) light. Upon interaction with molecules, the laser light generates scattered radiation of different wavelengths that together provides a fingerprint spectrum of the molecular structure. The Raman
yopathies or muscle disorders may arise due to many factors, such as myofibrillar and internal cytoskeletal protein gene mutations, infection, nutrient deficiency, and so forth.1 These disorders can be broadly divided into two groups: (i) the neuromuscular diseases, which include Dystrophies, Spinalatrophies, Parkinson’s, and so forth; and (ii) the musculoskeletal diseases, which include disorders such as cardiomyopathies, nemaline-myopathies, and so forth.1−4 Globally, such muscle-related disorders are known to be one of the leading causes of fatality.5 Usually, diagnosis of myopathies involves numerous tests, including but not limited to muscle strength analyses using electromyography, ultrasound, and magnetic resonance imaging (MRI), which detect electrical activity of the muscle, muscle inflammation, abnormal muscle, and so forth.4−7 However, none of these techniques are able to provide chemical (molecule) specific information on the pathology to be studied. The traditional pathologist’s interpretations of muscle biopsies generally rely only on the visual morphological changes. Myopathies are a clinically and genetically heterogeneous group of diseases with a wide spectrum of phenotypes. Thus, they sometimes remain undiagnosed until they progress to a point at which the disorder can obviously be identified but has become difficult to treat © 2015 American Chemical Society
Received: September 23, 2014 Accepted: January 12, 2015 Published: January 12, 2015 2187
DOI: 10.1021/ac503647x Anal. Chem. 2015, 87, 2187−2194
Article
Analytical Chemistry spectrum provides information on the chemical bonds in the molecular structure, intermolecular interactions, and the environment of the molecule. Therefore, disease conditions, known to induce changes at the molecular level, can readily be probed by Raman spectroscopy. These relative differences in chemical composition at different disease states are reflected in the spectra, enabling not only the development of diagnostic tools based on Raman spectroscopy but also help in understanding the underlying mechanism behind the progression of diseases.13−17 Furthermore, over the past decade, Raman microscopy combined with multivariate analysis techniques has been shown to have great potential for cell and tissue characterization in model organisms and also pathological tissue classification in humans.18−22 In this paper, for the first time, applicability of Raman spectroscopy toward the study of muscle disorders in a well-known model system the fruit fly is presented. The fruit fly, Drosophila melanogaster, has been used extensively as a model system to study the genetic and molecular mechanisms of many diseases. Inherently, Drosophila is an ideal model system owing to its short life span, easy maintenance, cost effectiveness, high fecundity, well-sequenced genome, and high degree of conservation of developmental pathways. Importantly, 70% of human disease genes have counter parts in D. melanogaster.3,23,24 Many muscle mutations with known molecular lesions and heterogeneous phenotypes exist in Drosophila. The indirect flight muscles (IFMs) of Drosophila have been used as a model of choice due to their structural similarity with human muscles and availability of heterogeneous muscle mutants. In particular, IFMs provide a good genetic system to investigate muscle structure and function.25 These muscles are physiologically similar to cardiac muscles and structurally similar to skeletal muscles of humans.26 Muscles are made up of multinucleated cells called myofibers that are composed of several myofibrils that in turn are made up of myofilaments. Sarcomeres, the smallest contractile functional units of the skeletal muscle, are composed of thick and thin myofilaments. The thick filaments are made up of myosin and myosin-binding proteins, while the thin filaments are composed of actin, tropomyosin, troponin I (TnI), troponin C (TnC), and troponin T (TnT) protein subunits (Figure 1). The interaction of actin and myosin is required for muscle contraction. In resting muscles, this interaction is hindered by the tropomyosin− troponin complex, which prevents the force production. However, when TnC binds to Ca2+, the inhibitory actions of TnI and TnT are alleviated, and tropomyosin slides on the thin filament, which enables the cross bridge formation between actin and myosin to produce force.27 The IFM contraction is regulated by both stretch and Ca2+-induced thin filament (actin− tropomyosin−troponin complex) activation.28,29 Mutations in the major thin (actin, tropomyosin, TnI, TnC, TnT) and thick (myosin and myosin-binding protein flightin) filament proteins hamper this contraction mechanism and induce muscle degeneration. Mutations affecting the IFMs show two broad kinds of phenotypes. The hypercontraction muscle phenotype is one where muscles develop and assemble normally, but with the initiation of muscle contraction, they experience unregulated acto-myosin interactions, and then, the muscles pull apart leading to muscle degeneration. Many point mutations of the thin filament such as upheld101 (up101) and heldup2 (hdp2) (described below) are known to generate this phenotype.24 The pathophysiology of hypercontraction is grossly comparable to cardiomyopathies or muscle damage arising from excessive contraction in heart and in human muscle injuries.24,30 In the
Figure 1. (A) Polarized light image of a hemi thorax of Drosophila showing six dorsal longitudinal muscles and schematic of skeletal muscle organization showing from left to right: bundles of myofibers, myofibrils in a single myofiber, myofilaments in a single myofibril, and sarcomere, the smallest contractile functional units of the skeletal muscle. (B) Detailed drawing of sarcomere, showing thin and thick filaments.
case of upheld1 (up1) and actin88FKM88 (Act88FKM88) mutants (explained below), aggregates of Z discs termed nemaline bodies are formed as a result of the presence of truncated proteins or absence of structural proteins (Supporting Information, Figure S-1).31,32 These nemaline bodies are counterparts of similar structures seen in nemaline and other protein aggregate myopathies in humans.32 The muscle (IFM) mutants used in the current study are up1, hdp2, myosin heavy chain7 (Mhc7), Act88FKM88, up101, and Canton-S (CS). up1 is a TnT splice mutation, where an adult specific TnT isoform fails to accumulate in the IFMs; hdp2 is a TnI hypomorph; Mhc7 is an IFM-specific myosin null mutation that removes all the myosin from the IFMs; Act88FKM88 is an IFM-specific actin isoform null that removes all the actin from the IFMs; and up101 is a TnT hypomorph mutant.24 The objective of this study is to find specific markers using Raman spectroscopy that will be helpful for classification and better understanding of the biochemical pathways during progression of muscle abnormalities induced by mutations. To the best of our knowledge, this is the first demonstration of classification of myopathies on the basis of biochemical changes in a Drosophila model using Raman spectroscopy.
■
EXPERIMENTAL SECTION Tissue Preparation. All flies were maintained at 25 °C on cornmeal−agar medium. up1, hdp2, Mhc7, Act88FKM88, and up101 mutant flies were procured from John Sparrow, York University, U.K., and CS was used as the control in all experiments. The flies were anaesthetized with diethyl ether and mounted on a glass slide using glycerol with the ventral side up. These mounted flies were snap frozen in liquid nitrogen, bisected longitudinally into two halves using a razor blade, and kept in 70% ethanol overnight. IFM fibers were removed under the binocular microscope using a 4× objective from each half and mounted on an aluminum slide for Raman spectroscopic study. Details about the human sample are included in the Supporting Information (Section S.1). 2188
DOI: 10.1021/ac503647x Anal. Chem. 2015, 87, 2187−2194
Article
Analytical Chemistry
Figure 2. Raman difference spectra for all mutants compared to control flies, from bottom to top: (i) up1 minus CS, (ii) Act88FKM88 minus CS, (iii) Mhc7 minus CS, (iv) up101 minus CS, and (v) hdp2 minus CS of (A) 2 days old flies and (B) 12 days old flies. Crests and troughs in the difference spectrum indicate an increase or decrease of the corresponding biomolecule levels relative to the total Raman-active constituents in mutant muscles.
the corresponding bands are based on literature reports33−39 and are presented in Table S-1 (Supporting Information). For better visualization of the differences of each mutant with respect to the control, the difference spectra (mutant minus CS) were calculated. Interestingly, the difference spectra revealed various biochemical changes of prominent Raman bands occurring in mutant muscles (Figure 2A, B). Crests and troughs in the difference spectrum indicate whether there is an increase or decrease, respectively, in the relative levels of a certain type of biomolecules in comparison to the total Raman-active constituents in mutant tissue.21 Usually, our conclusions on relative increases or decreases of a certain biomolecule are based on several Raman bands showing the same trend. The mutants up1 and Act88FKM88 show similar patterns in the difference spectra (Figure 2A, B), as both are known to produce similar muscle phenotypes.31,32 The difference spectra of 2 and 12 days old flies demonstrate increased concentrations of nucleic acids at 782−788, 810, 1091, 1338, 1422, and 1576 cm−1 assigned to cytosine, thymine, uracil-ring breathing, and OPO stretching (DNA backbone), OPO stretching (RNA backbone), PO2− stretching (DNA backbone), nucleic acids (adenine-A and guanine-G)/protein CH2 deformation, and nucleic acids (A, G), respectively.33,39 The amide I, mainly due to the CO stretch (1640−1680 cm−1), and amide III, mostly due to NH in-plane bending and CN stretching (1220−1310 cm−1) regions of the Raman spectrum, are characteristic of protein secondary structure. Usually, proteins with a high α-helical content show an amide I band centered in the region of 1645− 1658 cm−1 and an amide III band centered around 1260−1310 cm−1. In contrast, proteins that are dominated by β-sheet and/or random coil structure have an amide I peak at 1660−1680 cm−1
Raman Spectroscopy. Details of Raman instrumentation, data preprocessing, and multivariate analysis steps are included in the Supporting Information (Section S.2). For an experimental set, Raman spectra were recorded from IFM fibers of approximately 10−12 flies of each type (up1, hdp2, Mhc7, Act88FKM88, up101, and CS), and the same experiment was repeated four times with new groups of flies. So, in total, muscles from 40−45 flies were analyzed for each type. For the mutants, the spectra were predominantly recorded from the degenerated sites. All spectra were acquired for 5 s, and for each spectrum, 30 scans were coadded, and a minimum of 140 spectra were obtained from each type of flies.
■
RESULTS Drosophila has a short life span of around 65−75 days. Therefore, it is an ideal model to understand the progression of diseases by analyzing different stages. We considered two different time points, 2 and 12 days, to investigate different states of muscle disorders. The 2nd day samples depict the early stage of muscle degeneration, and in the 12th day samples, muscles get almost completely degenerated. Raman spectra of IFMs from up1, hdp2, Mhc7, Act88FKM88, up101, and CS were recorded, and 130 spectra from each group were averaged for both 2 and 12 days old flies as shown in Figure S-2A and S-2B (Supporting Information), respectively. In order to show the variation within a group, average Raman spectra (700−1700 cm−1) ± 1 standard deviation of all mutants and CS for both 2 and 12 days old flies are shown in Figure S-3A and S-3B (Supporting Information), respectively. The Raman spectra (700−1700 cm−1) illustrate the numerous bands attributable to various chemical bonds of biochemical entities present in the muscles; the bond specific assignments for 2189
DOI: 10.1021/ac503647x Anal. Chem. 2015, 87, 2187−2194
Article
Analytical Chemistry
Figure 3. PC-LDA scores plot of (A) 2 days old flies and (B) 12 days old flies with (i) all five mutants (up1, Act88FKM88, Mhc7, up101, hdp2) and control (CS); (ii) four mutants (up1, Act88FKM88, Mhc7, hdp2) and CS; (iii) four mutants (up1, Act88FKM88, Mhc7, up101) and CS. The mutants up1, Act88FKM88 (phenotypes similar to nemaline-myopathy) and up101, hdp2 (phenotypes similar to cardiomyopathy) form two different clusters when projected on the LDF1−LDF2 plane. up1 and Act88FKM88 are also well separated from each other compared with up101 and hdp2. Mhc7 that shows mixed phenotypes is also well separated from these two groups.
and an amide III band at 1230−1260 cm−1.34−36 The difference spectra in the amide I and amide III regions show positive contribution at 1660−1680 cm−1 and 1230−1260 cm−1 and decrease at 1645−1658 cm−1 and 1260−1310 cm−1, which can be interpreted as an increase in β structure or random coil at the expense of α-helical content. These changes are further corroborated by a considerable increase in the intensity of the band at 901 cm−1 assigned to CC stretching vibrations, which is due to increased β structure.37 Although the 938 cm−1 band is assigned to an α helix, we have not considered it, as this band could also be due to glycogen.38 Other features which may alter with the change in structure of proteins are the tryptophan band at 757 cm−1 assigned to the symmetric ring breathing and a slight decrease in this band indicates the exposure of buried tryptophan residues.36 In addition, changes are also observed in protein side chains. A positive Raman band at 1402 cm−1, due to the COO− symmetric stretch of aspartic and glutamic acid residues, indicates that most acid side chains are deprotonated.36,37 Interestingly, the 2nd day difference spectrum of Act88FKM88 shows a decrease in band intensity of the 1004 cm−1 symmetric ring-breathing mode of phenylalanine and the 1449 cm−1 CH2 bending mode in proteins. Act88FKM88 is an actin null mutant and completely lacks other major thin filament proteins32 and therefore demonstrates a decrease in band intensity at 1004 and 1449 cm−1. The positive peaks, in 12th day difference spectra of up1 and Act88FKM88, at 855, 938, 1082, and 1125 cm−1 assigned to ring-breathing (tyrosine)/COC stretching (glycogen, polysaccharides), CC stretch (α helix/glycogen), CO/ CC stretch (carbohydrates), and CO stretch (glycogen)/ CN stretch (proteins) vibrations, respectively, suggest the glycogen accumulation in muscles (Figure 2B).38,39 Notably, two
characteristic bands of carotenoids at 1156 and 1524 cm−1 due to CC and conjugated CC bond stretches are clearly visible in the 12th day difference spectra of up1 and Act88FKM88 (Figure 2B), which provides evidence that the content of carotenoids is higher in mutant muscles than in CS.39 These glycogen and carotenoid bands are not seen in the case of the other group of mutants, that is, hdp2 and up101. The hdp2 and up101 are TnI and TnT mutants, respectively, and they produce hypercontraction of IFMs in which the IFMs normally degenerate. Therefore, the corresponding difference spectra (Figure 2A, B) indicate denaturation and an overall decay in protein levels. Further, the difference spectra for hdp2 and up101 of both 2 and 12 days old samples show increased concentrations of nucleic acids. The difference spectra for Mhc7 (Mhc7 minus CS) of 2 and 12 days old flies (Figure 2A, B) show resemblance to both groups: (i) up1, Act88FKM88 and (ii) up101, hdp2. Mhc7 is a Mhc null mutant, and the mutation prevents the accumulation of MHC protein and hence fails to assemble thick filaments in their IFMs.24 The remaining proteins, mainly actin, show aggregation in the absence of myosin. Mhc7 mutants also show aggregation of Z bands with other thin filament proteins that are seen in hypercontracted muscle mutants (where muscles pull apart after initial proper assembly). Therefore, difference spectra demonstrate features analogous to both Act88FKM88 and up101. Multivariate and Statistical Analysis. In the difference spectra, we have considered the average spectrum of each mutant and in particular the intense bands. However, to visualize small differences and to handle the large data set, a multivariate analysis method that utilizes the entire Raman spectrum instead of some particular peaks is required. Principal component analysis (PCA) is a widely used unsupervised data transformation procedure for 2190
DOI: 10.1021/ac503647x Anal. Chem. 2015, 87, 2187−2194
Article
Analytical Chemistry reduction of dimensionality of a complex data set while retaining the most diagnostically significant information for sample classification.13,21 It reduces high-dimensional data into lowdimensional data called PC scores that are the projections of the data onto the corresponding PCs. PCs are nothing but orthonormal directions of maximal variance in the data set. The first principal component (PC1) demonstrates the maximum variance in the data set, and the second principal component (PC2) illustrates the largest residual variance and so forth. PCA is capable of efficiently capturing the data patterns but has less discrimination power to classify the data into subgroups. Therefore, for further classification of data into subgroups, linear discriminant analysis (LDA), K-means clustering, partial least squares (PLS), and so forth40−42 can be performed using the most significant PC scores as features. LDA is a supervised learning algorithm that looks for projections, called linear discriminant functions (LDFs), which maximize the ratio of interclass variance to intraclass variance. The PC-LDA model improves the efficiency of classification as it automatically brings out the most diagnostically significant features.43−45 Here, we generated a PC-LDA model using a training set consisting of ∼90 spectra per class that were randomly selected from a gamut of valid spectra after outlier removal. PCs were sorted in the order of the highest to lowest eigenvalues of covariance in the data, and each PC represents distinct Raman features.40−45 Only the first 60 PC scores were used in the LDA model as the noise contribution increases in the remaining PCs. For the 2 days old samples, the first three PCs account for 29.5%, 12.6%, and 9.2% variance, respectively, and the first 60 PCs contribute to about 95.3% of the total variance. Similarly, for the 12 days old data set, the first 60 PCs contribute to about 96.1% of total variance with a contribution of 22.5%, 18.0%, and 12.2% variance from PC1, PC2, and PC3, respectively. Figure 3A(i), B(i) shows the classification results based on the PC-LDA technique for both the 2nd day and 12th day data sets. The three-dimensional scatter plot of scores of LDFs illustrates a good separation between the six classes. Interestingly, when projected onto a LDF1−LDF2 plane (i) up1 and Act88FKM88 and (ii) hdp2 and up101 were clustered together leading to the formation of two different groups pertaining to nemaline-myopathies and cardiomyopathies, respectively, as shown in Figure 3A(i), B(i). In the case of the 2nd day training set, sensitivities greater than 93% and specificities greater than 99% across the different classes and an overall accuracy (OA) of 97% were observed (Supporting Information, Table S-2). Likewise, in the case of the 12th day training set, sensitivities greater than 85% and specificities greater than 97% across the different classes and an OA of 93% were observed. Sensitivity represents the proportion of correctly classified spectra of mutants/control (true positive rate) and 1specificity means the false positive rate. A high value of sensitivity and specificity indicates a good classification performance of the proposed PC-LDA algorithm. From the remaining spectra, 30 spectra per class were randomly selected as a test set to validate the PC-LDA model. To ensure that the results are unbiased, training and testing sets were arbitrarily selected from the whole data set. The same process was repeated multiple times using new randomly chosen sets.41−43 Each time, similar values of sensitivities and specificities were observed (Table 1). In order to evaluate and compare the performance of the PCLDA algorithms, receiver operating characteristic (ROC) curves were generated for each mutant for both the 2nd and 12th day data sets (Supporting Information, Figure S-4A and S-4B). Further, for comparison of the different ROC curves, the line of
Table 1. Sensitivity and Specificity Table of the PC-LDA Model for the Test Set 2 days class up1 hdp2 Mhc7 Act88FKM88 up101 CS up1 hdp2 Mhc7 Act88FKM88 CS up1 Mhc7 Act88FKM88 up101 CS
sensitivity (%)
12 days specificity (%)
sensitivity (%)
For All Six Classes 95.7 99.7 80.4 62.3 98.5 69.0 84.5 98.3 83.1 98.1 96.7 93.4 70.8 91.4 65.8 92.3 96.3 89.7 For Five Classes (Excluding up101) 93.8 100.0 82.3 70.8 97.7 85.6 86.0 97.9 82.6 93.5 96.4 87.5 93.7 92.6 91.5 For Five Classes (Excluding hdp2) 92.4 99.6 85.4 82.4 98.5 82.5 100.0 95.7 93.6 81.0 96.8 89.1 94.3 96.8 95.9
specificity (%) 99.6 90.9 94.1 96.5 98.1 97.2 99.3 95.1 93.5 97.8 96.7 99.4 95.8 97.5 98.7 95.1
no discrimination (indicated by a solid black line) is also shown. The accuracy of the diagnostic model increases as the ROC curve moves toward the upper left corner. So, the area under the curve (AUC) is an important measure of the accuracy of a diagnostic test.22,43,46 The AUC values of up1, hdp2, Mhc7, Act88FKM88, and up101 for the 2nd day samples were 0.96, 0.90, 0.97, 0.97, and 0.87 (Supporting Information, Figure S-4A(i)), respectively, and for the 12th day samples, the AUC values were 0.92, 0.81, 0.89, 0.97, 0.87 (Supporting Information, Figure S-4B(i)), respectively. This indicates a high-quality performance of the PC-LDA algorithm (the AUC for an ideal classifier is 1.00, and for the line of no discrimination, the AUC is 0.5). The sensitivities and AUC values of up101 and hdp2 are slightly poorer compared with the other classes as most of the times either up101 is misclassified as hdp2 or vice versa. This misclassification occurs because up101 IFM hypercontraction is indistinguishable from that of hdp2, although the hdp2 phenotype is more severe than that of up101.23,47 To further confirm this, we have generated a five class PC-LDA model by excluding up101 and hdp2 one at a time for the 2nd day (Figure 3A(ii, iii)) and the 12th day (Figure 3B(ii, iii)) samples. In both cases, a significant increase in specificities and sensitivities of up101 and hdp2 was observed (Table 1), and the corresponding ROC curves (Supporting Information, Figure S4A(ii, iii) and S-4B(ii, iii)) move toward the upper left corner, indicating the excellent classifying capability of the PC-LDA model.22,41 On the other hand, the PC-LDA model seems to discriminate between up1 and Act88FKM88 in spite of similar features in the corresponding difference spectra. This is because the extent of the changes is different as seen in Figure 2A(i, ii); the y axis ranges from −0.5 to 0.5 for up1 and from 0.5 to 1.0 for Act88FKM88. Also, the PC-LDA model considers whole spectra (all the wavenumbers) for discrimination that is difficult to visualize. In addition, a negative control study was also performed by feeding randomized class labels (ground truth) to LDA as explained by Barman et al.48 Across the multiple iterations, positive predictive values, sensitivities of each class, and OA of the diagnostic model never exceeded 30%. This confirms that our 2191
DOI: 10.1021/ac503647x Anal. Chem. 2015, 87, 2187−2194
Article
Analytical Chemistry
cytoplasm.2,50 Furthermore, the increase in nucleic acid bands could also be due to a reduction in thick/thin filaments in the mutants that in turn leads to a relative increase in visibility of nuclei and mitochondria. In the amide I and III regions, the Raman difference spectra of all mutants collectively illustrate the transformation from αhelical content to β sheet and/or random coil. These changes indicate unfolding and/or denaturation of helical structures and formation of sheet structures through intermolecular interactions between exposed hydrophobic residues. These denatured proteins are not stable and therefore form aggregates as indicated by the increase of β-sheet bands at 901, 1402, and 1680 cm−1.34−36 Protein unfolding can also be monitored by the characteristic bands of aromatic amino acid side chains, as these bands change with the polarity of the microenvironment. During unfolding and/or modification in the structure of the protein, the tryptophan residue gets exposed to the polar environment from a buried hydrophobic microenvironment as indicated by the decrease in band intensity at 757 cm−1.36 According to Pezolet et al., this could be related to the random orientation in degraded muscles as opposed to the regular fiber structure in control.51 Remarkably, Act88FKM88 and up1 mutants, which are grouped together in the LDF1−LDF2 plane, show an increase in the characteristic bands of glycogen in the 12th day difference spectra. IFMs, being high energy demanding tissues, require an easily accessible and easily convertible energy source such as glycogen for flight. The mutant muscles are inactive and do not utilize glycogen efficiently, and it keeps accumulating, provided there is no feedback mechanism to stop its synthesis.37 Furthermore, in the mutated state, they follow an anaerobic metabolic pathway where glycogen is not required, and this leads to accumulation of glycogen.37 Many muscle myopathies are known to show this behavior of glycogen accumulation.1 Additionally, the bands of carotenoids at 1156 and 1523 cm−1 are significantly increased in the 12th day difference spectra of both Act88FKM88 and up1 mutants. Carotenoids are antioxidant molecules and are known to inhibit the malignant transformation and mutagenesis.52,53 In muscles, carotenoids usually bind nonspecifically to the hydrophobic pocket in the actomyosin protein complex by weak interactions.54 As they are closely related to muscle proteins, any alteration in muscle protein profiles may influence carotenoid deposition. Further studies need to be performed to corroborate this hypothesis. In the case of up101 and hdp2, which are grouped together in the LDF1−LDF2 plane, the hypercontraction process gets initiated during the late pupal states, and muscles are seen completely pulled to either sides of the Drosophila thorax by eclosion that leads to age-dependent degeneration of myofibers.24 As a result, Raman spectra showed an overall reduction of the proteins bands. Genes involved in muscle remodeling are known to be activated during hypercontraction,50 so up101 and hdp2 muscles showed increased nucleic acid accumulation, suggesting that Raman spectroscopy is a sensitive instrument to detect the biochemical changes reflecting the pathophysiology of the mutant phenotype.
algorithm is robust to potential confounding variables and correlations in the data set. Human Nemaline-Myopathy Shows a Phenotype Similar to up1. To corroborate a possible extension of the proposed PC-LDA based diagnostic algorithm for human samples, we trained our model with Drosophila samples up1, Act88FKM88 (nemaline-myopathy like phenotype), and CS and then tested with spectra recorded from human nemalinemyopathy (HNM) tissue (Figure 4). The average spectrum of 70 Raman spectra from the human sample is shown in Supporting Information, Figure S-5. Out of 70 HNM spectra, 63 formed a cluster with up1 (Table 2), which indicates that the proposed PC-LDA algorithm using Raman features based on the Drosophila model is also suitable for modeling human disorders. Although only a single sample was available, these first results are very promising, and we plan to extend this study to other human myopathies as well.
Figure 4. PC-LDA scores plot of up1, Act88FKM88, and CS of 2 days old flies and human nemaline-myopathy (HNM). Raman spectra from up1, Act88FKM88, and CS were used as the training set to build the classification model, and HNM is used as the test set. HNM is clearly classified as up1 (a nemaline-myopathy-like phenotype).
Table 2. Truth Table of Human Nemaline-Myopathy (HNM) Samples Classified Using the Drosophila Model class (2 days old flies)
HNM (70 spectra)
CS up1 Act88FKM88
5 63 2
■
DISCUSSION Our findings suggest that Raman spectroscopy is a potential tool for the detection and classification of myopathies on the basis of biochemical changes. Changes were primarily observed in nucleic acid content, protein conformations, and also in glycogen and carotenoids levels. The increased content of nucleic acids (DNA/RNA) in all mutants with respect to the control could be explained on the basis of the following reasons. A mutation in filament proteins shows a reduction not only in thin filament complex proteins but also in their transcript levels.29,31,49 Therefore, RNA is not translated, and mutant RNA accumulates. Second, in myopathic conditions, muscle biopsy is characterized by the presence of regenerative and necrotic myofibers. Muscle abnormalities also lead to an increase in the transcription of genes involved in remodeling of the fibers. So, regenerative fibers contain an enlarged nucleus and basophilic RNA-rich
■
CONCLUSIONS Classification of myopathies, a heterogeneous group of diseases, is an essential step to avoid misdiagnosis and improve therapy effectiveness and survival. Raman spectroscopy integrated with multivariate analysis not only classifies various mutants but also reveals the progressive changes in biochemical components associated with disease states. Raman spectroscopy could 2192
DOI: 10.1021/ac503647x Anal. Chem. 2015, 87, 2187−2194
Article
Analytical Chemistry discriminate nemaline-myopathies (Act88FKM88 and up1) and cardiomyopathies (hdp2 and up101) into separate groups on a molecular basis in the Drosophila model. Importantly, characteristic bands of glycogen and carotenoids increased significantly only in the case of Act88FKM88 and up1. Further studies are required to understand the physiological reasons for this observation. Remarkably, using the Drosophila PC-LDA model as a training set, we could classify HNM samples in the appropriate category as up1. This preliminary study indicates the possibility of using Raman spectroscopy for understanding myopathies in humans. Hence, we are currently investigating in more detail whether that Raman spectroscopy in the Drosophila model has the potential to study the mechanism of other human diseases such as neurodegenerative diseases, which is difficult to follow in other model organisms due to their long life span and is almost impossible in humans.
■
(6) Weissleder, R.; Pittet, M. J. Nature 2008, 452 (7187), 580−589. (7) Keren, S.; Zavaleta, C.; Cheng, Z.; de la Zerda, A.; Gheysens, O.; Ghambhir, S. S. Proc. Natl. Acad. Sci. U.S.A. 2008, 105 (15), 5844−5849. (8) Agrawal, P. B.; Strickland, C. D.; Midgett, C.; Morales, A.; Newburger, D. E.; Poulos, M. A.; Tomczak, K. K.; Ryan, M. M.; Iannaccone, S. T.; Crawford, T. O.; Laing, N. G.; Beggs, A. H. Ann. Neurol. 2004, 56 (1), 86−96. (9) Jain, R.; Calderon, D.; Kierski, P. R.; Schurr, M. J.; Czuprynski, C. J.; Murphy, C. J.; McAnulty, J. F.; Abbott, N. L. Anal. Chem. 2014, 86 (8), 3764−3772. (10) Ghita, A.; Pascut, F. C.; Mather, M.; Sottile, V.; Notingher, I. Anal. Chem. 2012, 84 (7), 3155−3162. (11) Krishna, C. M.; Sockalingum, G. D.; Kegelaer, G.; Rubin, S.; Kartha, V. B.; Manfait, M. Vib. Spectrosc. 2005, 30 (1−2), 95−100. (12) Kong, K.; Rowlands, C. J.; Varma, S.; Perkins, W.; Leach, I. H.; Koloydenko, A. A.; Williams, H. C.; Notingher, I. Proc. Natl. Acad. Sci. U.S.A. 2013, 110 (38), 15189−15194. (13) Mo, J.; Zheng, W.; Low, J. J.; Ng, J.; Ilancheran, A.; Huang, Z. Anal. Chem. 2009, 81 (21), 8908−8915. (14) Avetisyan, A.; Jensen, J. B.; Huser, T. Anal. Chem. 2013, 85 (15), 7264−7270. (15) Michalet, X.; Pinaud, F. F.; Bentolila, L. A.; Tsay, J. M.; Doose, S.; Li, J. J.; Sundaresan, G.; Wu, A. M.; Gambhir, S. S.; Weiss, S. Science 2005, 307 (5709), 538−544. (16) Hartsuiker, L.; Zeijen, N. J. L.; Terstappen, L. W. M. M.; Otto, C. Analyst 2010, 135 (12), 3220−3226. (17) Pully, V. V.; Lenferink, A.; van Manen, H.-J.; Subramaniam, V.; van Blitterswijk, C. A.; Otto, C. Anal. Chem. 2010, 82 (5), 1844−1850. (18) Singh, B.; Gautam, R.; Kumar, S.; Kumar, B. N. V.; Nongthomba, U.; Nandi, D.; Mukherjee, G.; Santosh, V.; Somasundaram, K.; Umapathy, S. Curr. Sci. 2012, 102 (2), 232−244. (19) Cheng, H. C.; Wei, W. C.; June, T. W.; Ta-Chau, C. J. Biomed. Opt. 2011, 16 (1), 016012−1−7. (20) Zuser, E.; Chernenko, T.; Newmark, J.; Miljkovic, M.; Diem, M. Analyst 2010, 135 (12), 3030−3033. (21) The, S. K.; Zheng, W.; Ho, K. Y.; Teh, M.; Yeoh, K. G.; Huang, Z. Br. J. Cancer 2008, 98 (2), 457−465. (22) Haka, A. S.; Shafer-Peltier, K. E.; Fitzmaurice, M.; Crowe, J.; Dasari, R. R.; Feld, M. S. Proc. Natl. Acad. Sci. U.S.A. 2005, 102 (35), 12371−12376. (23) Lawrence, T. R.; Lorraine, P.; Sam, C.; Michael, G.; Ethan, B. Genome Res. 2001, 11 (6), 1114−1125. (24) Nongthomba, U.; Cummins, M.; Clark, S.; Vigoreaux, J. O.; Sparrow, J. C. Genetics 2003, 164 (1), 209−222. (25) Cripps, R. M. Nature’s Versatile Engine: Insect Flight Muscle Inside and Out; Vigoreaux, J., Ed.; Springer: New York, 2006. (26) Taylor, J. L.; Todd, G.; Gandevia, S. C. Clin. Exp. Pharmacol. Physiol. 2006, 33 (4), 400−405. (27) Gordon, A. M.; Homsher, E.; Regnier, M. Physiol. Rev. 2000, 80 (2), 853−924. (28) Peckham, M.; Molloy, J. E.; Sparrow, J. C.; White, D. C. J. Muscle Res. Cell Motil. 1990, 11 (3), 203−215. (29) Gordon, S.; Dickinson, M. H. Proc. Natl. Acad. Sci. U.S.A. 2006, 103 (11), 4311−4315. (30) Cammarato, A.; Hatch, V.; Saide, J.; Craig, R.; Sparrow, J. C.; Tobacman, L. S.; Lehman, W. Biophys. J. 2004, 86 (3), 1618−1624. (31) Nongthomba, U.; Ansari, M.; Thimmaiya, D.; Stark, M.; Sparrow, J. C. Genetics 2007, 177 (1), 295−306. (32) Haigh, S. E.; Salvi, S. S.; Sevdali, M.; Stark, M.; Goulding, D.; Clayton, J. D.; Bullard, B.; Sparrow, J. C.; Nongthomba, U. Neuromuscular Disord. 2010, 20 (6), 363−374. (33) Matthews, Q.; Jirasek, A.; Lum, J. J.; Brolo, A. G. Phys. Med. Biol. 2011, 56 (21), 6839−6855. (34) Xu, X.-L.; Han, M.-Y.; Fei, Y.; Zhou, G.-H. Meat Sci. 2011, 87 (3), 159−164. (35) Careche, M.; Li-Chan, E. C. Y. J. Food Sci. 1997, 62 (4), 717−723. (36) Herrero, A. M. Crit. Rev. Food Sci. Nutr. 2008, 48 (6), 512−523. (37) Barret, T. W.; Peticolas, W. L.; Robson, R. C. Biophys. J. 1978, 23 (3), 349−358.
ASSOCIATED CONTENT
* Supporting Information S
Additional text describing details about the human sample and Raman instrumentation and spectral data analysis; Raman band assignments (Table S-1); confusion matrix of the PC-LDA model for the training set (Table S-2); the polarized light images of control (CS) and mutants up1 and up101 flies (Figure S-1); average Raman spectra of control (CS) and mutants for 2 and 12 days old flies (Figure S-2); average Raman spectra ± 1 standard deviation of control (CS) and mutants for 2 and 12 days old flies (Figure S-3); ROC curves for one versus all PC-LDA diagnostic models (Figure S-4); and average Raman spectrum of human sample HNM (Figure S-5). This material is available free of charge via the Internet at http://pubs.acs.org.
■
AUTHOR INFORMATION
Corresponding Authors
*(U.N.) Phone: +(91)-80-22933258/3262; e-mail: upendra@ mrdg.iisc.ernet.in. *(S.U.) Phone: +(91)-80-22932595; e-mail:
[email protected]. ernet.in. Notes
The authors declare no competing financial interests.
■
ACKNOWLEDGMENTS We would like to thank lab members Rishita Nag, Shreya Verma, Neha Ruhela, Esha Patnaik, Mamta Rai, Mohan Jayaram, and Sanjay Prasad. We thank Prof. Freek Ariese (VU University, Amsterdam) for making thoughtful suggestions on the early draft of the manuscript. This work was supported by financial assistance from Council for Scientific and Industrial Research (CSIR), Department of Biotechnology (DBT), and Department of Science and Technology (DST), Government of India.
■
REFERENCES
(1) Fredericks, C. M.; Saladin, L. K. Pathophysiology of the Motor Systems: Principles and Clinical Presentations; F. A. Davis Company: Philadelphia, PA, 1996. (2) McNally, E. M.; Pytel, P. Annu. Rev. Pathol.: Mech. Dis. 2007, 2, 87− 109. (3) Xu, Y.; Condell, M.; Plesken, H.; Edelman-Noyensky, I.; Ma, J.; Ren, M.; Schlame, M. Proc. Natl. Acad. Sci. U.S.A. 2006, 103 (31), 11584−11588. (4) Emery, A. E. H. Lancet 2002, 359 (9307), 687−695. (5) World Health Organization. World Health Organization Report, 2011; WHO: Geneva, Switzerland, 2011. 2193
DOI: 10.1021/ac503647x Anal. Chem. 2015, 87, 2187−2194
Article
Analytical Chemistry (38) Jong de, B. W. D.; Bakker Schut, T. C.; Coppens, J.; Wolffenbuttel, K. P.; Kok, D. J.; Puppels, G. J. Vib. Spectrosc. 2003, 32 (1), 57−65. (39) Stone, N.; Kendall, C.; Shepherd, N.; Crowand, P.; Barr, H. J. Raman Spectrosc. 2002, 33 (7), 564−573. (40) Mahadevan-Jansen, A.; Mitchell, M. F.; Ramanjamt, N.; Malpica, A.; Thomsen, S.; Utzinger, U.; Richards-Kortum, R. Photochem. Photobiol. 1998, 68 (1), 123−132. (41) Kendall, C.; Stone, N.; Shepherd, N.; Geboes, K.; Warren, B.; Bennett, R.; Barr, H. J. Pathol. 2003, 200 (5), 602−609. (42) Taleb, A.; Diamond, J.; McGarvey, J. J.; Beattie, J. R.; Toland, C.; Hamilton, P. W. J. Phys. Chem. B 2006, 110 (39), 19625−19631. (43) Harvey, T. J.; Hughes, C.; Ward, A. D.; Faria, E. C.; Henderson, A.; Clarke, N. W.; Brown, M. D.; Snook, R. D.; Gardner, P. J. Biophotonics 2009, 2 (1−2), 47−69. (44) Fletcher, J. S.; Henderson, A.; Jarvis, R. M.; Lockyer, N. P.; Vickerman, J. C.; Goodacre, R. Appl. Surf. Sci. 2006, 252 (19), 6869− 6874. (45) Baker, M. J.; Brown, M. D.; Gazi, E.; Clarke, N. W.; Vickerman, J. C.; Lockyer, N. P. Analyst 2008, 133 (2), 175−179. (46) Soaresa, J. C.; Barman, I.; Dingari, N. C.; Volynskaya, Z.; Liu, W.; Klein, N.; Plecha, D.; Dasari, R. R.; Fitzmaurice, M. Proc. Natl. Acad. Sci. U.S.A. 2013, 110 (2), 471−476. (47) Beall, J. C.; Fyrberg, E. J. Cell Biol. 1991, 114 (5), 941−951. (48) Barman, I.; Dingari, N. C.; Kang, J. W.; Horowitz, G. L.; Dasari, R. R.; Feld, M. S. Anal. Chem. 2012, 84 (5), 2474−2482. (49) Nongthomba, U.; Clark, S.; Cummins, M.; Ansari, M.; Stark, M.; Sparrow, J. C. J. Cell Sci. 2004, 117 (9), 1795−1805. (50) Montana, E. S.; Littleton, J. T. J. Biol. Chem. 2006, 281 (12), 8100−8109. (51) Pezolet, M.; Pigeon-Gosselin, M.; Nadeau, J.; Caille, J. P. Biophys. J. 1980, 31 (1), 1−8. (52) Bakker Schut, T. C.; Puppels, G. J.; Kraan, Y. M.; Greve, J.; van der Maas, L. L.; Figdor, C. G. Int. J. Cancer (Pred. Oncol.) 1997, 74, 20−25. (53) Hata, T. R.; Scholz, T. A.; Ermakov, I. V.; McClane, R. W.; Khachik, F.; Gellermann, W.; Pershing, L. K. J. Invest. Dermatol. 2000, 115, 441−448. (54) Ytrestoyl, T.; Coral-Hinostroza, G.; Hatlen, B.; Robb, D. H. F.; Bjerkeng, B. Comp. Biochem. Physiol., Part B: Biochem. Mol. Biol. 2004, 138, 29−40.
2194
DOI: 10.1021/ac503647x Anal. Chem. 2015, 87, 2187−2194
