Lasers Med Sci DOI 10.1007/s10103-013-1515-y
ORIGINAL ARTICLE
Monitoring of chemotherapy leukemia treatment using Raman spectroscopy and principal component analysis Jos´e Luis Gonz´alez-Sol´ıs · Juan Carlos Mart´ınez-Espinosa · Juan Manuel Salgado-Rom´an · Pascual Palomares-Anda
Received: 20 September 2013 / Accepted: 9 December 2013 © Springer-Verlag London 2014
Abstract In this research, we used the Raman spectroscopy to distinguish between normal and leukemia blood serum and identify the different types of leukemia based on serum biochemistry. In addition, monitoring of patients under chemotherapy leukemia treatment (CHLT) was studied. Blood samples were obtained from seven patients who were clinically diagnosed with three leukemia types and 21 healthy volunteers. In addition, other five leukemia patients were monitored during the CHLT, two patients were declared healthy, one patient suspended it; the health of the other two patients worsened, and no improvement was observed along CHLT. The serum samples were put under an Olympus microscope integrated to the Raman system, and several points were chosen for the Raman measurement. The Horiba Jobin Yvon LabRAM HR800 Raman system is equipped with a liquid nitrogen-cooled detector and a laser of 830 nm with a power irradiation of 17 mW. It is shown that the serum samples from patient with leukemia
J. L. Gonz´alez-Sol´ıs () · J. M. Salgado-Rom´an Biophysics and Biomedical Sciences Laboratory, Centro Universitario de los Lagos, Universidad de Guadalajara, Enrique D´ıaz de Le´on 1144, Paseo de la Monta˜na, CP 47460, Lagos de Moreno, Jalisco, M´exico e-mail: [email protected] J. C. Mart´ınez-Espinosa Mathematics and Biotechnology Academy, Instituto Politecnico Nacional-UPIIG, Silao de la Victoria CP 36275, M´exico P. Palomares-Anda Instituto Mexicano del Seguro Social, Av. P. los Insurgentes S/N, Col. los Para´ısos, CP 37320, Le´on, Guanajuato, M´exico
and from the control group can be discriminated when multivariate statistical methods of principal component analysis (PCA) and linear discriminant analysis (LDA) are applied to their Raman spectra obtaining two large clusters corresponding to the control and leukemia serum samples and three clusters inside the leukemia group associated with the three leukemia types. The major differences between leukemia and control spectra were at 1,338 (Trp, α-helix, phospholipids), 1,447 (lipids), 1,523 (β-carotene), 1,556 (Trp), 1,587 (protein, Tyr), 1,603 (Tyr, Phe), and 1,654 (proteins, amide I, α-helix, phospholipids) cm−1 , where these peaks were less intense in the leukemia spectrum. Minor differences occurred at 661 (glutathione), 890 (glutathione), 973 (glucosamine), 1,126 (protein, phospholipid C–C str), 1,160 (β-carotene), 1,174 (Trp, Phe), 1,208 (Trp), 1,246 (amide III), 1,380 (glucosamine), and 1,404 (glutathione) cm−1 . Leukemia spectrum showed a peak at 917 cm−1 associated with glutathione, but it was absent in the control spectrum. The results suggest that the Raman spectroscopy and PCA could be a technique with a strong potential of support for current techniques to detect and identify the different leukemia types by using a serum sample. Nevertheless, with the construction of a data library integrated with a large number of leukemia and control Raman spectra obtained from a wide range of healthy and leukemic population, the Raman-PCA technique could be converted into a new technique for minimally invasive real-time diagnosis of leukemia from serum samples. In addition, complementary results suggest that using these techniques is possible to monitor CHLT. Keywords Leukemia · Serum · Chemotherapy leukemia treatment · Raman spectroscopy · Principal component analysis
Lasers Med Sci
Introduction Leukemia is the 11th most common cancer worldwide with more than 250,000 new cases each year in the USA [1]. Principal leukemia types are identified on the basis of malignancy involving either lymphoid (B and T cells) or myeloid (granulocytic, erythroid, and megakaryocytic cells), and upon whether the disease is acute or chronic in onset [2]. For leukemia types, acute lymphoblastic leukemia (ALL), chronic lymphoblastic leukemia (CLL), acute myeloid leukemia (AML), and chronic myeloid leukemia (CML), generally, diagnosis may be suspected from examination of peripheral blood under a microscope, but it is necessary to be confirmed by bone marrow biopsy. Once the diagnosis is confirmed, the patient undergoes chemotherapy leukemia treatment (CHLT). Chemotherapy is a treatment to destroy cancer cells using medicine. There are more than 50 chemotherapy medicines available that can be used in a variety of ways according to the type of cancer, how advanced it is, and general health of the patient. In CHLT, high doses of medicines are commonly necessary, and the patient will be monitored regularly throughout the treatment by physical examination, X-rays, and blood and urine tests. All CHLT medicines work by attacking cells that are dividing rapidly, by affecting DNA in cancer cells for leukemia to disappear completely. However, CHLT also kills healthy cells, which can lead to side effects. At any one time, some cancer cells will be resting and may not be killed until a later round of chemotherapy. A CHLT is made up of several of these cycles, and a perfect monitoring of CHLT is reflected in the quality of life of the patient. The original type of medicine, called first line, may be changed if it is not controlling the cancer or if it causes serious side effects. Due to that, for many people, CHLT can have a huge impact on life; new control techniques of the cycle of chemotherapy, which alerts when a patient is completely healthy, are then necessary. Such techniques will need to be faster, less invasive, and cost-effective while retaining adequate sensitivity and specificity to perform as practical screening techniques. The Raman spectroscopy is a technique that is fast emerging as promising alternatives in biology and medicine, including cancer diagnosis and monitoring. The Raman spectroscopy is a vibrational spectroscopic technique that provides detailed information about the chemical and molecular composition of cell and tissues [3]. When incident laser light strikes a tissue sample, some of its photons undergo a change in energy, known as the Raman effect. Molecular excitations and interactions between the molecules of the sample cause this change and lead to scattering of photons. A spectrometer counts scattered photons and measures the intensity and energy change of the resultant light in units per centimeter. Because each molecule has
unique vibrations, the Raman spectrum of the tissue will consist of a series of bands or peaks, characteristic of the biochemical composition of that tissue. Pathologic changes in tissue lead to morphological and molecular alterations. Morphological changes are routinely applied for disease classification using pathologic examination. The detection of molecular abnormalities in diseased tissue by spectroscopy has been only recently recognized as a possible useful diagnostic method [3]. The clinical application of this technique includes rapid identification of microorganisms and cancer detection. An end goal of this research is to provide surgeons with a minimally invasive real-time diagnosis tool that can distinguish normal and diseased tissue. Such a device can guide for surgical resections as recently shown in an in vivo testing in adults [4], reduce the amount of time needed for pathologic examination by frozen section or routine histological, and help reduce operation time and cost. Many Raman spectroscopic studies performed on adults suggest that is possible to differentiate normal, benign, premalignant, and malignant lesions in the breast and gastrointestinal tract [4–6]. Some reported works about leukemia using the Raman spectroscopy and principal component analysis (PCA) were presented by our research group in 2008 [7] and by Frank et al. in a subsequent work using T cells and B cells [8]. It is the first report on results evaluating the usefulness of the Raman spectroscopy and PCA in the detection and identification of the different leukemia types as well as the monitoring of patients under CHLT. In the last two decades, multivariate analysis (MA) has been applied to the Raman spectroscopy to classify epithelial precancers and cancers [9]. In particular, PCA has been used to differentiate between epithelial precancers and cancers [10]. The Raman spectroscopy and PCA seem to be very promising tools to be used in biomedical research. PCA is used for dimensionality reduction by finding orthogonal dimension in descending order of variance. Other widely used techniques in exploratory data analysis are linear discriminant analysis (LDA) and hierarchical clustering (HC), which allow to identify in a natural way the classes present in data and which is depicted by a tree or dendrogram. It has been used in a large variety of engineering and scientific disciplines such as gene expression [11, 12], stock indices [13], and astrophysics [14]. In our leukemia research, LDA naturally allows to identify the control and leukemia patient groups, and within the leukemia patient group, the different leukemia types are identified. Timely knowledge on the leukemia types allows to establish the most affective CHLT for improving the survival of patient. The main contributions of the paper are identification of corresponding patterns of control and leukemia samples,
Lasers Med Sci
and identification of ALL, CLL, AML, and CML samples within the leukemia group. Finally, the present paper proposes a minimally invasive real-time diagnosis method to monitor the leukemia patient throughout CHLT by reducing subjectivity to human error. In the both studies, the principal component loading are plotted as a function of the Raman shifts by revealing which bands account for the greatest differences between the leukemia and control spectra. We report on the main biomolecules corresponding to the spectral differences.
Methodology Patients and protocol Fresh blood samples were obtained from seven patients who were clinically diagnosed with leukemia and 21 healthy volunteers. In addition, other five leukemia patients were monitored during the CHLT, two patients were declared healthy, one patient suspended it, and the health of the other two patients worsened, and no improvement was observed along CHLT. All patients were from the central region of Mexico and had similar ethnic and socioeconomic backgrounds. The age for the leukemia and control groups was between 8 and 50 years. Written consent was obtained from the subjects, and the study was conducted according to the Declaration of Helsinki. Table 1 presents the most relevant clinical information for each leukemia patient.
Table 1 Clinical diagnosis of leukemia patients Patient
Leukemia type
CHLT
Response to CHLT
1 2 3 4 5 6 7 8 9 10 11 12
ALL ALL ALL AML AML CML CML ALL ALL CML AML AML
No No No No No No No Yes Yes Yes Yes Yes
Pos Pos NCH Neg Neg
Yes, on treatment; No, no treatment CHLT chemotherapy leukemia treatment, Pos positive response, Neg negative response, NCH no change
Sample preparation and the Raman spectroscopy measurements Blood samples were obtained between 7:00 a.m. and 9:00 a.m. and were centrifuged to get the serum. A drop of serum was placed onto an aluminum substrate, which was examined by an Olympus microscope integrated to the Raman system. All spectra were measured on the same day when the samples were obtained. In order to ensure statistically sound sampling, we collected at least five spectra from different regions of each fresh serum sample using a Horiba Jobin Yvon LabRAM HR800 confocal Raman microscope with a laser of 830 nm and power of 17 mW. The laser beam was focused on the surface of the sample with a ×50 microscope objective and an exposure of 40 to 60 s. Analyzed region was from 400 to 1,800 cm−1 , with a spectral resolution of ∼0.6 cm−1 . The Raman system was calibrated with a silicon semiconductor using the Raman peak at 520 cm−1 .
Results and discussion A total of 144 spectra were collected with 102 spectra from 21 control patients and 42 spectra from 7 leukemia patients. For the case of patients under CHLT, at least five spectra were measured after each chemotherapy dose. Each acquired spectrum was normalized to the highest peak. Raw spectra were processed by carrying baseline correction, smoothing and normalization to remove noise, sample florescence, and shot noise from cosmic rays, through a filter based on the baseline correction with asymmetric least squares smoothing algorithm [15]. Unlike other algorithms, it is fast and simple, even for large signals, and asymmetric weighting applies everywhere. Once the spectra was processed, PCA was implemented, where the main information is described by the first principal components. By plotting the loading vectors as a function of the wave number, the position of relevant differences [16] between the control and leukemia groups could be determined. PCA and all the algorithms for data analysis were implemented in MatLab commercial software. Figure 1a shows an example of the recorded Raman spectra of blood serum samples of the control group, Fig. 1b shows the same data with smoothing, and Fig. 1c shows the same data with both smoothing and baseline corrections. After the initial processing, the mean spectrum of each group was calculated. The mean spectra were analyzed to obtain general biochemical information for each data group
Lasers Med Sci Fig. 1 a Raman spectra of raw data. b Raman spectra after smoothing. c Raman spectra after smoothing and baseline correction
[17, 18]. Sensitivity and specificity were used to judge diagnostic ability: sensitivity =
TP , TP + FN
specificity =
TN , TN + FP
where TP is true positive, FN is false negative, TN is true negative, and FP is false positive. The mean Raman spectra of control and leukemia patients are shown in Fig. 2. The mean spectra were analyzed to obtain general biochemical information for each data group [17]. These spectra are typical of biological tissues exhibiting a combination of broad Raman bands arising from the molecular vibration of proteins, nucleic acids, lipids, and other constituent molecules. The control spectrum showed the presence of higher amounts of carotenoids indicated by peaks at 1,002, 1,160, and 1,523 cm−1 [19]. Carotenoids have been shown to inhibit cancerous changes
Fig. 2 Mean Raman spectra of control and leukemia serum samples
in several organs including the skin, mammary gland, lung, liver, and colon [20]. In our study, some carotenoid-related peaks were absent from leukemia samples, suggesting that their absence may also play a role in the leukemia action mechanism. In addition, the mean spectra show that some peaks look similar to each other, but some peaks show differences in intensity. Some authors have reported that some peak positions in the Raman spectra of the tissue from malignant tumors are shifted when compared with the spectra of tissue from benign tumors [8, 21, 22]. For example, looking at 1,002 cm−1 , the peak intensity of the control is higher than the peak intensity of the leukemia group. This peak is attributed to the ring deformation mode associated with aromatic amino acids, phenylalanine, and proline, and is often observed in the Raman spectra of proteins [23]. The band of amide III at 1,246 cm−1 is wide and strong, which is assigned to the overlapping of helix, folding, and random coil. Other examples are peaks 1,338 (tryptophan, α-helix, and phospholipids), 1,447 (sheet and phospholipids), 1,523 (β-carotene), and 1,656 (proteins and phospholipids) cm−1 . The major differences between leukemia and control spectra were at 1,338 (Trp, α-helix, and phospholipids), 1,447 (lipids), 1,523 (β-carotene), 1,556 (Trp), 1,587 (protein, Tyr), 1,603 (Tyr, Phe), and 1,654 (proteins, amide I, α-helix, phospholipids) cm−1 , where these peaks were less intense in the leukemia spectrum. Minor differences occurred at 661 (glutathione), 890 (glutathione), 1,126 (protein, phospholipid C–C str), 1,160 (β-carotene), 1,174 (Trp, Phe), 1,208 (Trp), 1,246 (amide III), and 1,404 (glutathione) cm−1 . Leukemia spectrum showed a peak at 917 cm−1 , associated with glutathione, but it was absent in the control spectrum. Table 2 shows the main bands observed in the control and leukemia spectra, the corresponding assignment
Lasers Med Sci Table 2 Main bands observed in control and leukemia serum spectra, the corresponding assignment of biomolecules, and the comparison of the band intensities Band (cm−1 )
Biomolecules
Comparison of the band intensities
446 509 545 566 622 642 661 695 714 742 754 760 828 853 875 890 897 917 938 955 973 1,002 1,028 1,063 1,083
Glutathione Trp Trp
IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC
1,103 1,126 1,160 1,174 1,208 1,230–1,282 1,300–1,345 1,380 1,404 1,447 1,523 1,556 1,587 1,603 1,620 1,654
Phe Tyr Glutathione Polysaccharides Phospholipid Protein Trp Tyr Tyr Trp Glutathione C–O–C str Glutathione Skeletal str α CH2 rock Glucosamine Phe Phe Phe Phospholipids O–P–O and C–C Phe Protein, phospholipid C–C str β-carotene Trp, Phe Trp Amide III Trp, α-helix, phospholipids Glucosamine Glutathione Phospholipid, C–H scissor in CH2 β-carotene Trp Protein, Tyr Tyr, Phe Tyr, Trp C=C str Proteins, amide I, α-helix, phospholipids
< IL > IL ≈ IL < IL ≈ IL ≈ IL > IL ≈ IL ≈ IL ≈ IL ≈ IL ≈ IL < IL > IL > IL > IL ≈ IL < IL > IL > IL < IL > IL > IL > IL > IL
of biomolecules, and the comparison of the band intensities in the control and leukemia spectra. Alternatively, PCA and LDA techniques have been applied to discriminate between the control and leukemia spectra in an easier way. Leukemia detection and leukemia type identification PCA was carried out after removing the fluorescence contribution, smoothing, and applying the baseline correction. The main information obtained from the PCA is described by the first three principal component loadings, PC1, PC2, and PC3. Figure 3 corresponds to PCA of all the spectra. LDA was applied to the PCA result by dividing the new data according to natural classes present, obtaining two large clusters corresponding to the control and leukemia serum samples and three clusters inside the leukemia group associated with the three leukemia types (see Fig. 3). Clearly, PCA and LDA allowed to discriminate between the control and leukemia patients and between the three different leukemia types. The differences are due to some of the peaks analyzed above. The prediction of the pathological state of the samples using the Raman spectroscopy and PCA is compared with that of pathology shown in Table 1. In order to test the validity of our prediction classification, we have performed
IC > IL IC > IL IC IC IC IC IC
> IL > IL > IL > IL > IL
IC < IL IC > IL IC > IL IC IC IC IC IC IC
> IL > IL > IL > IL > IL > IL
IC and IL are the intensities of the same bands in the control and leukemia spectra, respectively
Fig. 3 PCA plot comparing control and leukemia patients. Clearly, the plot allows to discriminate between control and leukemia patients. In addition, PCA and LDA allow to discriminate between the three different leukemia types
Lasers Med Sci
a leave-one-out cross-validation. The results of the crossvalidation are compared with the classification results in Fig. 3. Since all 42 spectra from the leukemia case were correctly separated from the control spectra, the detection of leukemia in terms of sensitivity and specificity using the cross-validated data are 100 and 100 %, respectively. On the other hand, ALL, AML, and CML leukemia types were separated from one another with 100 % sensitivity and 100 % specificity. Rather than analyzing the first three loadings, we analyze the second, third, or fourth loadings, and the leukemia and control spectra continue to be separated with 97.5 % sensitivity and 98.1 % specificity. The higher values of sensitivity and specificity are encouraging, which may be guided by in-depth future studies. By plotting the loading vectors as a function of the wave number, the position of relevant differences [16] between the control and leukemia groups could be determined. Figure 4 corresponds to the plot of the first loading against the Raman shift. PC1 plot of Fig. 4 indicates peak positions where these differences occur. Unlike a Raman spectrum, loading spectrum contains positive and negative bands, and their corresponding frequencies can be correlated with some of the major variations in the molecular composition between both groups, which are seen for the more intense peaks. In Fig. 4, the control-leukemia plot is compared with control and leukemia samples, and controlcontrol plot is compared with control samples between themselves. By comparing the control sample with themselves, biochemical differences should exist, and they are indicated by the most intense peaks in the control-control plot. Consequently, whether we compare the control-control and control-leukemia plots, it should provide us only the significant biochemical differences between the control and leukemia patients by discounting the existing biochemical differences between the control patients. By discarding the most intense peaks matching between the control-control
and control-leukemia plots, we obtain real biochemical differences among the control and leukemia serum samples. These most significant differences appear at 460, 613, 774 (glutathione), 828 (Tyr), 853 (Tyr), 875 (Trp), 917 (glutathione), 955 (CH2 rock), 973 (glucosamine), 1,002 (Phe), 1,015 (glutathione), 1,028 (Phe), 1,126 (protein, phospholipid C–C str), 1,160 (β-carotene), 1,174 (Trp, Phe), 1,208 (Trp), 1,265 (amide III), 1,353 (Trp), 1,380 (glucosamine), 1,404 (glutathione), 1,425 (Trp), 1,523 (β-carotene), 1,603 (Tyr, Phe), 1,620 (Tyr, Trp C=C str), 1,654 (proteins, amide I, α-helix, phospholipids) per centimeter. These differences between control and leukemia samples obtained using both PC1 and PC2 plots [18] are consistent with those shown in Table 2. The rest of the bands show small differences in shape and position, and the nature of these differences probably comes from the intensity variation, shifting of the band positions, or a mixture of both features. By using either Table 2 or the plot of first loading, we showed that there were more significant differences between the control group and leukemia patients in glutathione, tryptophan, glucosamine, tyrosine, phenylalanine, and β-carotene. Monitoring of chemotherapy leukemia treatment In addition to the detection of the different types of leukemia, the present paper proposes the PCA as a method to monitor to leukemia patients under CHLT by offering a faster alternative technique by reducing subjectivity to human error. Monitoring is based on observing the biochemical changes throughout CHLT. Throughout the treatment, patient improvement is observed when the leukemia group is going to come near to the control group (see Fig. 3). We studied the serum of five patients under CHLT. Each patient was assigned a protocol of CHLT according to the type of leukemia, how advanced it is, and the general health
Fig. 4 First loading plots between control samples between themselves and between leukemia samples
Lasers Med Sci
of the patient. The most used chemotherapy drugs in the five leukemia patients studied were cyclophosphamide, doxorubicin, vincristine, cytarabine, and daunorubicin. A patient could receive a combination of different drugs throughout CHLT by changing one of them if no improvement is observed. Serum samples were taken after each chemotherapy dose and analyzed using the Raman spectroscopy and PCA until the patients finished their CHLT. The Raman measurements for the CHLT monitoring of the five patients supplemented the Raman measurements for leukemia detection and leukemia type identification. According to clinical records, the first of the two ALL patients under CHLT (see Table 1) received at least 12 chemotherapy doses. Due to the null response to CHLT, from the fifth day, a second line treatment or drug change was applied by observing immediately a positive response. Figure 5 shows the monitoring PCA of the first ALL patient under CHLT. In Fig. 5a, we observe that the cluster of the patient comes near to the cluster of the control patients
throughout CHLT. In the day number 33 of CHLT, the cluster is totally within the cluster of the control patients, meaning that the biochemical differences are minimal between the leukemia patient and control patients, and thus, the leukemia patient may be declared healthy. Figure 5b shows the behavior of the third loading throughout CHLT, observing that PC3 values remained almost constant until the fourth day. After this day, the Fig. 5b shows that PC3 values begin to vary, approaching to the PC3 values of control patients. Figure 5 shows clearly the health improvement of the patient and the precise moment of drug change, which is in total agreement with the clinical record. Therefore, the Raman spectroscopy and PCA seem to be excellent tools to monitor CHLT. Figure 6 shows the monitoring PCA of the first CML patient. According to the clinical record, this patient left the treatment and therefore never was declared clinically healthy. In Fig. 6a, we observe that in spite that the cluster of CML patient moves in the right direction, it never reaches
a
b
Fig. 5 a Monitoring PCA plot of the first patient under CHLT. b Plot of the third loading after each chemotherapy dose
Fig. 6 a Monitoring PCA plot of the CML patient under CHLT. b Plot of the first loading after each chemotherapy dose
Lasers Med Sci
the cluster of control patients. In addition, in Fig. 6b, PC1 values of the CML patient approximately never reach to PC1 values of control patients. On the other hand, the clinical records of two AML patients showed that the first AML patient received only one chemotherapy dose, and the second AML patient received three chemotherapy doses. The health of the two AML patients was deteriorating throughout CHLT. Figure 7 shows the monitoring PCA of one of the two AML patients. Clearly, we observe that clusters of leukemia patient move away from the control cluster after each chemotherapy dose, indicating that the health of the patient worsens, and no improvement is observed along CHLT. The result shown in Fig. 7 is in complete agreement with the clinical record of this AML patient. Monitoring can also be analyzed by using plots of loading against the Raman shift as it was shown in Fig. 4. Figure 8 corresponds to the third loading of control patients between themselves (black plot) and loading between control patients and the first LLA patient on days 0 (purple plot), 27 (red plot), and 33 (blue plot) of CHLT. These
Fig. 8 Third loading plots of the first LLA patient under CHLT. Shaded regions show the patient improvement after each chemotherapy dose
plots show the changes of the major biochemical differences between the control and leukemia groups after each chemotherapy dose in the 400–500 cm−1 region observing the shift of peaks. In one shaded region, we can observe that the peak at 460 cm−1 of day 0 (purple plot) shifted after each chemotherapy dose until it matched with the peak at 456 cm−1 of the third loading of the control patients (black plot) at the end of the treatment (day 33). The same behavior was observed for peaks 425, 434, 443, and 450 cm−1 and their corresponding shaded regions. At the end of the chemotherapy treatment, we observed that some peaks as 774, 917, 1,015, and 1,404 cm−1 corresponded to glutathione; 973 and 1,380 cm−1 associated to glucosamine; 875, 1,174, 1,208, 1,353, 1,425, and 1,620 cm−1 assigned to tryptophan; 828, 853, and 1,603 cm−1 corresponded to tyrosine; 1,002, 1,028, and 1,603 cm−1 associated to phenylalanine; and 1,160 and 1,523 cm−1 assigned to β-carotene, shifted to match with peaks of control samples. Our result is consistent with the publication where they claim that glucosamine could induce proliferation of leukemia [24–26].
Conclusion
Fig. 7 a Monitoring PCA plot of one of the two AML patients under CHLT. b Plot of the first loading after each chemotherapy dose
In this study, the Raman spectroscopy and PCA were used to highlight differences in the chemical composition of serum samples from patients with a clinical diagnosis of leukemia vs healthy control subjects. There exists a strong line at 1,002 cm−1 assigned to the phenylalanine in all samples, whose intensity is not easy to change and can often be a standard of the Raman lines in serum samples. By using PCA and LDA, it was possible to find the natural clusters present in the Raman spectra data allowing to identify two large groups corresponding to healthy and
Lasers Med Sci
leukemia patients, and within the leukemia group, it was possible to identify the patients with the different types of leukemia. Due to that, all spectra from leukemia cases were correctly separated from the control spectra; the detection of leukemia in terms of sensitivity and specificity using the cross-validated data were 100 and 100 %, respectively. On the other hand, ALL, AML, and CML leukemia types were separated from one another with 100 % sensitivity and 100 % specificity. We showed that by using the Raman spectroscopy and PCA techniques, it is possible to monitor patients throughout their CHLT, being sensitive to changes of chemotherapy medicines applied to the patient. Throughout a full CHLT, we monitored the shift of peaks in the loading plots allowing to identify the biomolecules that mark the differences between leukemia and control patients. The Raman measurements for the CHLT monitoring supplemented and validated the Raman measurements for leukemia detection and leukemia type identification. In our research, we showed that the main molecules involved in leukemia study were glutathione, glucosamine, tryptophan, tyrosine, phenylalanine, and β-carotene. Our results indicate that the Raman spectroscopy and PCA seem to be excellent tools with a strong potential of support for current techniques to detect and identify the different leukemia types based on probing changes at the molecular level. Nevertheless, with the construction of a data library integrated with a large number of leukemia and control Raman spectra obtained from a wide range of healthy and leukemia population, the Raman-PCA technique could be converted into a new technique for minimally invasive diagnosis of leukemia from serum samples and a real-time guide for the surgeon about the health improvement of patients under CHLT by reducing time and cost of treatment.
Acknowledgments The authors wish to thank CONACYT for financial support under grant number 45488 and Research Network Soft Condensed Matter. Also, we are thankful to Q. F. B. Olga Leticia Brizuela Gamio for preparation of the serum samples.
References 1. National Cancer Institute (2008) What you need to know about leukemia. NIH publication no. 08-3775. National Cancer Institute, Bethesda 2. Freireich EJ, Lemak NA (1991) Milestones in leukemia research and therapy. Johns Hopkins University Press, Baltimore 3. Choo-Smith LP, Edward MHG, Endtz HP et al (2002) Medical applications of Raman spectroscopy: from proof of principle to clinical implementation. Biopolymers 67:1–9 4. Haka AS, Volynskaya Z, Gardecki J et al (2006) In vivo margin assessment during partial mastectomy breast surgery using Raman spectroscopy. Cancer Res 66:3317–3322 5. Bohorfoush AG (2006) Tissue spectroscopy for gastrointestinal diseases. Endoscopy 28:372–380
6. Pichardo-Molina JL, Frausto-Reyes C, Barbosa-Garca O, HuertaFranco R, Gonz¨lez-Trujillo JL, Ramrez-Alvarado CA, GutirrezJu¨rez G, Medina-Gutirrez C (2006) Raman spectroscopy and multivariate analysis of serum simples from breast cancer patients. Laser Med Sci 10103:432–438 7. Gonz´alez-Sol´ıs JL, Mart´ınez-Espinosa JC, Frausto-Reyes C, Miranda-Beltr´an ML, Soria-Fregoso C, Medina-Valtierra J (2009) Detection of leukemia with blood samples using Raman spectroscopy and multivariate analysis. AIP Conf Proc 1142:99–103 8. Frank CJ, McCreery LR, Redd DCB (1995) Raman spectroscopy of normal and diseased human breast tissues. Anal Chem 67:777– 783 9. Mahadevan-Jansen A, Richards-Kortum RR (1996) Raman spectroscopy for the detection of cancers and precancers. J Biomed Opt 1(1):31–70 10. Stone N, Kendall C et al (2002) Near-infrared Raman spectroscopy for the classification of epithelial pre-cancers and cancers. J Raman Spectrosc 33:564–573 11. Qiong Z, Yingsha Z (2006) Hierarchical clustering of gene expression profiles with graphics hardware. J Pattern Recogn Lett 27:676–681 12. Dhaeseleer P (2005) How does gene expression clustering work Nat Biotechnol 23:1499–1501 13. Kullmann L, Kertesz J, Mantegna RN (2000) Identification of clusters of companies in stock indices via Potts superparamagnetic transitions. Phys A 287:412–419 14. Dekel A, West MJ (1985) On percolation as a cosmological test. Astrophys J 288:411 15. Boelens HF, Eiler PH, Hankemeier T (2005) Sing constrains improve the detection of differences between complex spectral data sets: LC-IR as an example. Anal Chem 77(24):7998– 8007 16. Nogueira VG, Silveira L (2005) Raman spectroscopy study of atherosclerosis in human carotid artery. J Biomed Opt 10(3):031117 17. Stone N, Kendall C, Smith J et al (2004) Raman spectroscopy for identification of epithelial cancers. Faraday Discuss 126:141– 157 18. De Gelder J, De Gussem K, Vandenabeele P, Moens L (2007) Reference database of Raman spectra of biological molecules. J. Raman Spectrosc 38:1133–1147 19. Schultz H, Baranska M, Baranski R (2005) Potential of NIR-FTRaman spectroscopy in natural carotenoid analysis. Biopolymers 77:212–221 20. Hata TR, Schlz TA, Ermakov IV et al (2000) Non-invasive Raman spectroscopic detection of carotenoids in human skin. J Invest Dermatol 115:441–448 21. Alfano RR, Liu CH et al (1991) Human breast tissue studied by IR Fourier transform Raman spectroscopy. Lasers Life Sci 4:23–28 22. Hanlon EB, Manoharan R et al (2000) Prospects for in vivo Raman spectroscopy. Phys Med Biol 45:R1–R59 23. Krafft C, Sobottka SB, Schackert G, Salzer R (2005) Near infrared Raman spectroscopy mapping of native brain tissue and intracranial tumors. Analyst 130:1070–1077 24. Wang Z, Qiao Y, Huang GS, Wang AQ, Zhang YQ, Feng JL, Yang GR, Guo Y, Liang R (2003) Glucosamine and glucosamine hydrochloride induced leukemia cell line K562 differentiation into macrophage. Chin Pharmacol Bull 19(3):290–293 25. Wang Z, Qiao Y, Huang GS, Wang AQ, Zhang YQ, Feng JL, Yang GR, Guo Y, Liang R (2003) Induction of macrophagic differentiation of leukemia cell line K562 by N-acetyl-D-glucosamine. J Fourth Mil Med Univ 24(1):46–48 26. Zhang L, Liu W, Peng Y, Wang D (2006) Antitumor activities of D-glucosamine and its derivatives. J Zhejiang Univ Sci B 7(8):608–614
ORIGINAL ARTICLE
Monitoring of chemotherapy leukemia treatment using Raman spectroscopy and principal component analysis Jos´e Luis Gonz´alez-Sol´ıs · Juan Carlos Mart´ınez-Espinosa · Juan Manuel Salgado-Rom´an · Pascual Palomares-Anda
Received: 20 September 2013 / Accepted: 9 December 2013 © Springer-Verlag London 2014
Abstract In this research, we used the Raman spectroscopy to distinguish between normal and leukemia blood serum and identify the different types of leukemia based on serum biochemistry. In addition, monitoring of patients under chemotherapy leukemia treatment (CHLT) was studied. Blood samples were obtained from seven patients who were clinically diagnosed with three leukemia types and 21 healthy volunteers. In addition, other five leukemia patients were monitored during the CHLT, two patients were declared healthy, one patient suspended it; the health of the other two patients worsened, and no improvement was observed along CHLT. The serum samples were put under an Olympus microscope integrated to the Raman system, and several points were chosen for the Raman measurement. The Horiba Jobin Yvon LabRAM HR800 Raman system is equipped with a liquid nitrogen-cooled detector and a laser of 830 nm with a power irradiation of 17 mW. It is shown that the serum samples from patient with leukemia
J. L. Gonz´alez-Sol´ıs () · J. M. Salgado-Rom´an Biophysics and Biomedical Sciences Laboratory, Centro Universitario de los Lagos, Universidad de Guadalajara, Enrique D´ıaz de Le´on 1144, Paseo de la Monta˜na, CP 47460, Lagos de Moreno, Jalisco, M´exico e-mail: [email protected] J. C. Mart´ınez-Espinosa Mathematics and Biotechnology Academy, Instituto Politecnico Nacional-UPIIG, Silao de la Victoria CP 36275, M´exico P. Palomares-Anda Instituto Mexicano del Seguro Social, Av. P. los Insurgentes S/N, Col. los Para´ısos, CP 37320, Le´on, Guanajuato, M´exico
and from the control group can be discriminated when multivariate statistical methods of principal component analysis (PCA) and linear discriminant analysis (LDA) are applied to their Raman spectra obtaining two large clusters corresponding to the control and leukemia serum samples and three clusters inside the leukemia group associated with the three leukemia types. The major differences between leukemia and control spectra were at 1,338 (Trp, α-helix, phospholipids), 1,447 (lipids), 1,523 (β-carotene), 1,556 (Trp), 1,587 (protein, Tyr), 1,603 (Tyr, Phe), and 1,654 (proteins, amide I, α-helix, phospholipids) cm−1 , where these peaks were less intense in the leukemia spectrum. Minor differences occurred at 661 (glutathione), 890 (glutathione), 973 (glucosamine), 1,126 (protein, phospholipid C–C str), 1,160 (β-carotene), 1,174 (Trp, Phe), 1,208 (Trp), 1,246 (amide III), 1,380 (glucosamine), and 1,404 (glutathione) cm−1 . Leukemia spectrum showed a peak at 917 cm−1 associated with glutathione, but it was absent in the control spectrum. The results suggest that the Raman spectroscopy and PCA could be a technique with a strong potential of support for current techniques to detect and identify the different leukemia types by using a serum sample. Nevertheless, with the construction of a data library integrated with a large number of leukemia and control Raman spectra obtained from a wide range of healthy and leukemic population, the Raman-PCA technique could be converted into a new technique for minimally invasive real-time diagnosis of leukemia from serum samples. In addition, complementary results suggest that using these techniques is possible to monitor CHLT. Keywords Leukemia · Serum · Chemotherapy leukemia treatment · Raman spectroscopy · Principal component analysis
Lasers Med Sci
Introduction Leukemia is the 11th most common cancer worldwide with more than 250,000 new cases each year in the USA [1]. Principal leukemia types are identified on the basis of malignancy involving either lymphoid (B and T cells) or myeloid (granulocytic, erythroid, and megakaryocytic cells), and upon whether the disease is acute or chronic in onset [2]. For leukemia types, acute lymphoblastic leukemia (ALL), chronic lymphoblastic leukemia (CLL), acute myeloid leukemia (AML), and chronic myeloid leukemia (CML), generally, diagnosis may be suspected from examination of peripheral blood under a microscope, but it is necessary to be confirmed by bone marrow biopsy. Once the diagnosis is confirmed, the patient undergoes chemotherapy leukemia treatment (CHLT). Chemotherapy is a treatment to destroy cancer cells using medicine. There are more than 50 chemotherapy medicines available that can be used in a variety of ways according to the type of cancer, how advanced it is, and general health of the patient. In CHLT, high doses of medicines are commonly necessary, and the patient will be monitored regularly throughout the treatment by physical examination, X-rays, and blood and urine tests. All CHLT medicines work by attacking cells that are dividing rapidly, by affecting DNA in cancer cells for leukemia to disappear completely. However, CHLT also kills healthy cells, which can lead to side effects. At any one time, some cancer cells will be resting and may not be killed until a later round of chemotherapy. A CHLT is made up of several of these cycles, and a perfect monitoring of CHLT is reflected in the quality of life of the patient. The original type of medicine, called first line, may be changed if it is not controlling the cancer or if it causes serious side effects. Due to that, for many people, CHLT can have a huge impact on life; new control techniques of the cycle of chemotherapy, which alerts when a patient is completely healthy, are then necessary. Such techniques will need to be faster, less invasive, and cost-effective while retaining adequate sensitivity and specificity to perform as practical screening techniques. The Raman spectroscopy is a technique that is fast emerging as promising alternatives in biology and medicine, including cancer diagnosis and monitoring. The Raman spectroscopy is a vibrational spectroscopic technique that provides detailed information about the chemical and molecular composition of cell and tissues [3]. When incident laser light strikes a tissue sample, some of its photons undergo a change in energy, known as the Raman effect. Molecular excitations and interactions between the molecules of the sample cause this change and lead to scattering of photons. A spectrometer counts scattered photons and measures the intensity and energy change of the resultant light in units per centimeter. Because each molecule has
unique vibrations, the Raman spectrum of the tissue will consist of a series of bands or peaks, characteristic of the biochemical composition of that tissue. Pathologic changes in tissue lead to morphological and molecular alterations. Morphological changes are routinely applied for disease classification using pathologic examination. The detection of molecular abnormalities in diseased tissue by spectroscopy has been only recently recognized as a possible useful diagnostic method [3]. The clinical application of this technique includes rapid identification of microorganisms and cancer detection. An end goal of this research is to provide surgeons with a minimally invasive real-time diagnosis tool that can distinguish normal and diseased tissue. Such a device can guide for surgical resections as recently shown in an in vivo testing in adults [4], reduce the amount of time needed for pathologic examination by frozen section or routine histological, and help reduce operation time and cost. Many Raman spectroscopic studies performed on adults suggest that is possible to differentiate normal, benign, premalignant, and malignant lesions in the breast and gastrointestinal tract [4–6]. Some reported works about leukemia using the Raman spectroscopy and principal component analysis (PCA) were presented by our research group in 2008 [7] and by Frank et al. in a subsequent work using T cells and B cells [8]. It is the first report on results evaluating the usefulness of the Raman spectroscopy and PCA in the detection and identification of the different leukemia types as well as the monitoring of patients under CHLT. In the last two decades, multivariate analysis (MA) has been applied to the Raman spectroscopy to classify epithelial precancers and cancers [9]. In particular, PCA has been used to differentiate between epithelial precancers and cancers [10]. The Raman spectroscopy and PCA seem to be very promising tools to be used in biomedical research. PCA is used for dimensionality reduction by finding orthogonal dimension in descending order of variance. Other widely used techniques in exploratory data analysis are linear discriminant analysis (LDA) and hierarchical clustering (HC), which allow to identify in a natural way the classes present in data and which is depicted by a tree or dendrogram. It has been used in a large variety of engineering and scientific disciplines such as gene expression [11, 12], stock indices [13], and astrophysics [14]. In our leukemia research, LDA naturally allows to identify the control and leukemia patient groups, and within the leukemia patient group, the different leukemia types are identified. Timely knowledge on the leukemia types allows to establish the most affective CHLT for improving the survival of patient. The main contributions of the paper are identification of corresponding patterns of control and leukemia samples,
Lasers Med Sci
and identification of ALL, CLL, AML, and CML samples within the leukemia group. Finally, the present paper proposes a minimally invasive real-time diagnosis method to monitor the leukemia patient throughout CHLT by reducing subjectivity to human error. In the both studies, the principal component loading are plotted as a function of the Raman shifts by revealing which bands account for the greatest differences between the leukemia and control spectra. We report on the main biomolecules corresponding to the spectral differences.
Methodology Patients and protocol Fresh blood samples were obtained from seven patients who were clinically diagnosed with leukemia and 21 healthy volunteers. In addition, other five leukemia patients were monitored during the CHLT, two patients were declared healthy, one patient suspended it, and the health of the other two patients worsened, and no improvement was observed along CHLT. All patients were from the central region of Mexico and had similar ethnic and socioeconomic backgrounds. The age for the leukemia and control groups was between 8 and 50 years. Written consent was obtained from the subjects, and the study was conducted according to the Declaration of Helsinki. Table 1 presents the most relevant clinical information for each leukemia patient.
Table 1 Clinical diagnosis of leukemia patients Patient
Leukemia type
CHLT
Response to CHLT
1 2 3 4 5 6 7 8 9 10 11 12
ALL ALL ALL AML AML CML CML ALL ALL CML AML AML
No No No No No No No Yes Yes Yes Yes Yes
Pos Pos NCH Neg Neg
Yes, on treatment; No, no treatment CHLT chemotherapy leukemia treatment, Pos positive response, Neg negative response, NCH no change
Sample preparation and the Raman spectroscopy measurements Blood samples were obtained between 7:00 a.m. and 9:00 a.m. and were centrifuged to get the serum. A drop of serum was placed onto an aluminum substrate, which was examined by an Olympus microscope integrated to the Raman system. All spectra were measured on the same day when the samples were obtained. In order to ensure statistically sound sampling, we collected at least five spectra from different regions of each fresh serum sample using a Horiba Jobin Yvon LabRAM HR800 confocal Raman microscope with a laser of 830 nm and power of 17 mW. The laser beam was focused on the surface of the sample with a ×50 microscope objective and an exposure of 40 to 60 s. Analyzed region was from 400 to 1,800 cm−1 , with a spectral resolution of ∼0.6 cm−1 . The Raman system was calibrated with a silicon semiconductor using the Raman peak at 520 cm−1 .
Results and discussion A total of 144 spectra were collected with 102 spectra from 21 control patients and 42 spectra from 7 leukemia patients. For the case of patients under CHLT, at least five spectra were measured after each chemotherapy dose. Each acquired spectrum was normalized to the highest peak. Raw spectra were processed by carrying baseline correction, smoothing and normalization to remove noise, sample florescence, and shot noise from cosmic rays, through a filter based on the baseline correction with asymmetric least squares smoothing algorithm [15]. Unlike other algorithms, it is fast and simple, even for large signals, and asymmetric weighting applies everywhere. Once the spectra was processed, PCA was implemented, where the main information is described by the first principal components. By plotting the loading vectors as a function of the wave number, the position of relevant differences [16] between the control and leukemia groups could be determined. PCA and all the algorithms for data analysis were implemented in MatLab commercial software. Figure 1a shows an example of the recorded Raman spectra of blood serum samples of the control group, Fig. 1b shows the same data with smoothing, and Fig. 1c shows the same data with both smoothing and baseline corrections. After the initial processing, the mean spectrum of each group was calculated. The mean spectra were analyzed to obtain general biochemical information for each data group
Lasers Med Sci Fig. 1 a Raman spectra of raw data. b Raman spectra after smoothing. c Raman spectra after smoothing and baseline correction
[17, 18]. Sensitivity and specificity were used to judge diagnostic ability: sensitivity =
TP , TP + FN
specificity =
TN , TN + FP
where TP is true positive, FN is false negative, TN is true negative, and FP is false positive. The mean Raman spectra of control and leukemia patients are shown in Fig. 2. The mean spectra were analyzed to obtain general biochemical information for each data group [17]. These spectra are typical of biological tissues exhibiting a combination of broad Raman bands arising from the molecular vibration of proteins, nucleic acids, lipids, and other constituent molecules. The control spectrum showed the presence of higher amounts of carotenoids indicated by peaks at 1,002, 1,160, and 1,523 cm−1 [19]. Carotenoids have been shown to inhibit cancerous changes
Fig. 2 Mean Raman spectra of control and leukemia serum samples
in several organs including the skin, mammary gland, lung, liver, and colon [20]. In our study, some carotenoid-related peaks were absent from leukemia samples, suggesting that their absence may also play a role in the leukemia action mechanism. In addition, the mean spectra show that some peaks look similar to each other, but some peaks show differences in intensity. Some authors have reported that some peak positions in the Raman spectra of the tissue from malignant tumors are shifted when compared with the spectra of tissue from benign tumors [8, 21, 22]. For example, looking at 1,002 cm−1 , the peak intensity of the control is higher than the peak intensity of the leukemia group. This peak is attributed to the ring deformation mode associated with aromatic amino acids, phenylalanine, and proline, and is often observed in the Raman spectra of proteins [23]. The band of amide III at 1,246 cm−1 is wide and strong, which is assigned to the overlapping of helix, folding, and random coil. Other examples are peaks 1,338 (tryptophan, α-helix, and phospholipids), 1,447 (sheet and phospholipids), 1,523 (β-carotene), and 1,656 (proteins and phospholipids) cm−1 . The major differences between leukemia and control spectra were at 1,338 (Trp, α-helix, and phospholipids), 1,447 (lipids), 1,523 (β-carotene), 1,556 (Trp), 1,587 (protein, Tyr), 1,603 (Tyr, Phe), and 1,654 (proteins, amide I, α-helix, phospholipids) cm−1 , where these peaks were less intense in the leukemia spectrum. Minor differences occurred at 661 (glutathione), 890 (glutathione), 1,126 (protein, phospholipid C–C str), 1,160 (β-carotene), 1,174 (Trp, Phe), 1,208 (Trp), 1,246 (amide III), and 1,404 (glutathione) cm−1 . Leukemia spectrum showed a peak at 917 cm−1 , associated with glutathione, but it was absent in the control spectrum. Table 2 shows the main bands observed in the control and leukemia spectra, the corresponding assignment
Lasers Med Sci Table 2 Main bands observed in control and leukemia serum spectra, the corresponding assignment of biomolecules, and the comparison of the band intensities Band (cm−1 )
Biomolecules
Comparison of the band intensities
446 509 545 566 622 642 661 695 714 742 754 760 828 853 875 890 897 917 938 955 973 1,002 1,028 1,063 1,083
Glutathione Trp Trp
IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC IC
1,103 1,126 1,160 1,174 1,208 1,230–1,282 1,300–1,345 1,380 1,404 1,447 1,523 1,556 1,587 1,603 1,620 1,654
Phe Tyr Glutathione Polysaccharides Phospholipid Protein Trp Tyr Tyr Trp Glutathione C–O–C str Glutathione Skeletal str α CH2 rock Glucosamine Phe Phe Phe Phospholipids O–P–O and C–C Phe Protein, phospholipid C–C str β-carotene Trp, Phe Trp Amide III Trp, α-helix, phospholipids Glucosamine Glutathione Phospholipid, C–H scissor in CH2 β-carotene Trp Protein, Tyr Tyr, Phe Tyr, Trp C=C str Proteins, amide I, α-helix, phospholipids
< IL > IL ≈ IL < IL ≈ IL ≈ IL > IL ≈ IL ≈ IL ≈ IL ≈ IL ≈ IL < IL > IL > IL > IL ≈ IL < IL > IL > IL < IL > IL > IL > IL > IL
of biomolecules, and the comparison of the band intensities in the control and leukemia spectra. Alternatively, PCA and LDA techniques have been applied to discriminate between the control and leukemia spectra in an easier way. Leukemia detection and leukemia type identification PCA was carried out after removing the fluorescence contribution, smoothing, and applying the baseline correction. The main information obtained from the PCA is described by the first three principal component loadings, PC1, PC2, and PC3. Figure 3 corresponds to PCA of all the spectra. LDA was applied to the PCA result by dividing the new data according to natural classes present, obtaining two large clusters corresponding to the control and leukemia serum samples and three clusters inside the leukemia group associated with the three leukemia types (see Fig. 3). Clearly, PCA and LDA allowed to discriminate between the control and leukemia patients and between the three different leukemia types. The differences are due to some of the peaks analyzed above. The prediction of the pathological state of the samples using the Raman spectroscopy and PCA is compared with that of pathology shown in Table 1. In order to test the validity of our prediction classification, we have performed
IC > IL IC > IL IC IC IC IC IC
> IL > IL > IL > IL > IL
IC < IL IC > IL IC > IL IC IC IC IC IC IC
> IL > IL > IL > IL > IL > IL
IC and IL are the intensities of the same bands in the control and leukemia spectra, respectively
Fig. 3 PCA plot comparing control and leukemia patients. Clearly, the plot allows to discriminate between control and leukemia patients. In addition, PCA and LDA allow to discriminate between the three different leukemia types
Lasers Med Sci
a leave-one-out cross-validation. The results of the crossvalidation are compared with the classification results in Fig. 3. Since all 42 spectra from the leukemia case were correctly separated from the control spectra, the detection of leukemia in terms of sensitivity and specificity using the cross-validated data are 100 and 100 %, respectively. On the other hand, ALL, AML, and CML leukemia types were separated from one another with 100 % sensitivity and 100 % specificity. Rather than analyzing the first three loadings, we analyze the second, third, or fourth loadings, and the leukemia and control spectra continue to be separated with 97.5 % sensitivity and 98.1 % specificity. The higher values of sensitivity and specificity are encouraging, which may be guided by in-depth future studies. By plotting the loading vectors as a function of the wave number, the position of relevant differences [16] between the control and leukemia groups could be determined. Figure 4 corresponds to the plot of the first loading against the Raman shift. PC1 plot of Fig. 4 indicates peak positions where these differences occur. Unlike a Raman spectrum, loading spectrum contains positive and negative bands, and their corresponding frequencies can be correlated with some of the major variations in the molecular composition between both groups, which are seen for the more intense peaks. In Fig. 4, the control-leukemia plot is compared with control and leukemia samples, and controlcontrol plot is compared with control samples between themselves. By comparing the control sample with themselves, biochemical differences should exist, and they are indicated by the most intense peaks in the control-control plot. Consequently, whether we compare the control-control and control-leukemia plots, it should provide us only the significant biochemical differences between the control and leukemia patients by discounting the existing biochemical differences between the control patients. By discarding the most intense peaks matching between the control-control
and control-leukemia plots, we obtain real biochemical differences among the control and leukemia serum samples. These most significant differences appear at 460, 613, 774 (glutathione), 828 (Tyr), 853 (Tyr), 875 (Trp), 917 (glutathione), 955 (CH2 rock), 973 (glucosamine), 1,002 (Phe), 1,015 (glutathione), 1,028 (Phe), 1,126 (protein, phospholipid C–C str), 1,160 (β-carotene), 1,174 (Trp, Phe), 1,208 (Trp), 1,265 (amide III), 1,353 (Trp), 1,380 (glucosamine), 1,404 (glutathione), 1,425 (Trp), 1,523 (β-carotene), 1,603 (Tyr, Phe), 1,620 (Tyr, Trp C=C str), 1,654 (proteins, amide I, α-helix, phospholipids) per centimeter. These differences between control and leukemia samples obtained using both PC1 and PC2 plots [18] are consistent with those shown in Table 2. The rest of the bands show small differences in shape and position, and the nature of these differences probably comes from the intensity variation, shifting of the band positions, or a mixture of both features. By using either Table 2 or the plot of first loading, we showed that there were more significant differences between the control group and leukemia patients in glutathione, tryptophan, glucosamine, tyrosine, phenylalanine, and β-carotene. Monitoring of chemotherapy leukemia treatment In addition to the detection of the different types of leukemia, the present paper proposes the PCA as a method to monitor to leukemia patients under CHLT by offering a faster alternative technique by reducing subjectivity to human error. Monitoring is based on observing the biochemical changes throughout CHLT. Throughout the treatment, patient improvement is observed when the leukemia group is going to come near to the control group (see Fig. 3). We studied the serum of five patients under CHLT. Each patient was assigned a protocol of CHLT according to the type of leukemia, how advanced it is, and the general health
Fig. 4 First loading plots between control samples between themselves and between leukemia samples
Lasers Med Sci
of the patient. The most used chemotherapy drugs in the five leukemia patients studied were cyclophosphamide, doxorubicin, vincristine, cytarabine, and daunorubicin. A patient could receive a combination of different drugs throughout CHLT by changing one of them if no improvement is observed. Serum samples were taken after each chemotherapy dose and analyzed using the Raman spectroscopy and PCA until the patients finished their CHLT. The Raman measurements for the CHLT monitoring of the five patients supplemented the Raman measurements for leukemia detection and leukemia type identification. According to clinical records, the first of the two ALL patients under CHLT (see Table 1) received at least 12 chemotherapy doses. Due to the null response to CHLT, from the fifth day, a second line treatment or drug change was applied by observing immediately a positive response. Figure 5 shows the monitoring PCA of the first ALL patient under CHLT. In Fig. 5a, we observe that the cluster of the patient comes near to the cluster of the control patients
throughout CHLT. In the day number 33 of CHLT, the cluster is totally within the cluster of the control patients, meaning that the biochemical differences are minimal between the leukemia patient and control patients, and thus, the leukemia patient may be declared healthy. Figure 5b shows the behavior of the third loading throughout CHLT, observing that PC3 values remained almost constant until the fourth day. After this day, the Fig. 5b shows that PC3 values begin to vary, approaching to the PC3 values of control patients. Figure 5 shows clearly the health improvement of the patient and the precise moment of drug change, which is in total agreement with the clinical record. Therefore, the Raman spectroscopy and PCA seem to be excellent tools to monitor CHLT. Figure 6 shows the monitoring PCA of the first CML patient. According to the clinical record, this patient left the treatment and therefore never was declared clinically healthy. In Fig. 6a, we observe that in spite that the cluster of CML patient moves in the right direction, it never reaches
a
b
Fig. 5 a Monitoring PCA plot of the first patient under CHLT. b Plot of the third loading after each chemotherapy dose
Fig. 6 a Monitoring PCA plot of the CML patient under CHLT. b Plot of the first loading after each chemotherapy dose
Lasers Med Sci
the cluster of control patients. In addition, in Fig. 6b, PC1 values of the CML patient approximately never reach to PC1 values of control patients. On the other hand, the clinical records of two AML patients showed that the first AML patient received only one chemotherapy dose, and the second AML patient received three chemotherapy doses. The health of the two AML patients was deteriorating throughout CHLT. Figure 7 shows the monitoring PCA of one of the two AML patients. Clearly, we observe that clusters of leukemia patient move away from the control cluster after each chemotherapy dose, indicating that the health of the patient worsens, and no improvement is observed along CHLT. The result shown in Fig. 7 is in complete agreement with the clinical record of this AML patient. Monitoring can also be analyzed by using plots of loading against the Raman shift as it was shown in Fig. 4. Figure 8 corresponds to the third loading of control patients between themselves (black plot) and loading between control patients and the first LLA patient on days 0 (purple plot), 27 (red plot), and 33 (blue plot) of CHLT. These
Fig. 8 Third loading plots of the first LLA patient under CHLT. Shaded regions show the patient improvement after each chemotherapy dose
plots show the changes of the major biochemical differences between the control and leukemia groups after each chemotherapy dose in the 400–500 cm−1 region observing the shift of peaks. In one shaded region, we can observe that the peak at 460 cm−1 of day 0 (purple plot) shifted after each chemotherapy dose until it matched with the peak at 456 cm−1 of the third loading of the control patients (black plot) at the end of the treatment (day 33). The same behavior was observed for peaks 425, 434, 443, and 450 cm−1 and their corresponding shaded regions. At the end of the chemotherapy treatment, we observed that some peaks as 774, 917, 1,015, and 1,404 cm−1 corresponded to glutathione; 973 and 1,380 cm−1 associated to glucosamine; 875, 1,174, 1,208, 1,353, 1,425, and 1,620 cm−1 assigned to tryptophan; 828, 853, and 1,603 cm−1 corresponded to tyrosine; 1,002, 1,028, and 1,603 cm−1 associated to phenylalanine; and 1,160 and 1,523 cm−1 assigned to β-carotene, shifted to match with peaks of control samples. Our result is consistent with the publication where they claim that glucosamine could induce proliferation of leukemia [24–26].
Conclusion
Fig. 7 a Monitoring PCA plot of one of the two AML patients under CHLT. b Plot of the first loading after each chemotherapy dose
In this study, the Raman spectroscopy and PCA were used to highlight differences in the chemical composition of serum samples from patients with a clinical diagnosis of leukemia vs healthy control subjects. There exists a strong line at 1,002 cm−1 assigned to the phenylalanine in all samples, whose intensity is not easy to change and can often be a standard of the Raman lines in serum samples. By using PCA and LDA, it was possible to find the natural clusters present in the Raman spectra data allowing to identify two large groups corresponding to healthy and
Lasers Med Sci
leukemia patients, and within the leukemia group, it was possible to identify the patients with the different types of leukemia. Due to that, all spectra from leukemia cases were correctly separated from the control spectra; the detection of leukemia in terms of sensitivity and specificity using the cross-validated data were 100 and 100 %, respectively. On the other hand, ALL, AML, and CML leukemia types were separated from one another with 100 % sensitivity and 100 % specificity. We showed that by using the Raman spectroscopy and PCA techniques, it is possible to monitor patients throughout their CHLT, being sensitive to changes of chemotherapy medicines applied to the patient. Throughout a full CHLT, we monitored the shift of peaks in the loading plots allowing to identify the biomolecules that mark the differences between leukemia and control patients. The Raman measurements for the CHLT monitoring supplemented and validated the Raman measurements for leukemia detection and leukemia type identification. In our research, we showed that the main molecules involved in leukemia study were glutathione, glucosamine, tryptophan, tyrosine, phenylalanine, and β-carotene. Our results indicate that the Raman spectroscopy and PCA seem to be excellent tools with a strong potential of support for current techniques to detect and identify the different leukemia types based on probing changes at the molecular level. Nevertheless, with the construction of a data library integrated with a large number of leukemia and control Raman spectra obtained from a wide range of healthy and leukemia population, the Raman-PCA technique could be converted into a new technique for minimally invasive diagnosis of leukemia from serum samples and a real-time guide for the surgeon about the health improvement of patients under CHLT by reducing time and cost of treatment.
Acknowledgments The authors wish to thank CONACYT for financial support under grant number 45488 and Research Network Soft Condensed Matter. Also, we are thankful to Q. F. B. Olga Leticia Brizuela Gamio for preparation of the serum samples.
References 1. National Cancer Institute (2008) What you need to know about leukemia. NIH publication no. 08-3775. National Cancer Institute, Bethesda 2. Freireich EJ, Lemak NA (1991) Milestones in leukemia research and therapy. Johns Hopkins University Press, Baltimore 3. Choo-Smith LP, Edward MHG, Endtz HP et al (2002) Medical applications of Raman spectroscopy: from proof of principle to clinical implementation. Biopolymers 67:1–9 4. Haka AS, Volynskaya Z, Gardecki J et al (2006) In vivo margin assessment during partial mastectomy breast surgery using Raman spectroscopy. Cancer Res 66:3317–3322 5. Bohorfoush AG (2006) Tissue spectroscopy for gastrointestinal diseases. Endoscopy 28:372–380
6. Pichardo-Molina JL, Frausto-Reyes C, Barbosa-Garca O, HuertaFranco R, Gonz¨lez-Trujillo JL, Ramrez-Alvarado CA, GutirrezJu¨rez G, Medina-Gutirrez C (2006) Raman spectroscopy and multivariate analysis of serum simples from breast cancer patients. Laser Med Sci 10103:432–438 7. Gonz´alez-Sol´ıs JL, Mart´ınez-Espinosa JC, Frausto-Reyes C, Miranda-Beltr´an ML, Soria-Fregoso C, Medina-Valtierra J (2009) Detection of leukemia with blood samples using Raman spectroscopy and multivariate analysis. AIP Conf Proc 1142:99–103 8. Frank CJ, McCreery LR, Redd DCB (1995) Raman spectroscopy of normal and diseased human breast tissues. Anal Chem 67:777– 783 9. Mahadevan-Jansen A, Richards-Kortum RR (1996) Raman spectroscopy for the detection of cancers and precancers. J Biomed Opt 1(1):31–70 10. Stone N, Kendall C et al (2002) Near-infrared Raman spectroscopy for the classification of epithelial pre-cancers and cancers. J Raman Spectrosc 33:564–573 11. Qiong Z, Yingsha Z (2006) Hierarchical clustering of gene expression profiles with graphics hardware. J Pattern Recogn Lett 27:676–681 12. Dhaeseleer P (2005) How does gene expression clustering work Nat Biotechnol 23:1499–1501 13. Kullmann L, Kertesz J, Mantegna RN (2000) Identification of clusters of companies in stock indices via Potts superparamagnetic transitions. Phys A 287:412–419 14. Dekel A, West MJ (1985) On percolation as a cosmological test. Astrophys J 288:411 15. Boelens HF, Eiler PH, Hankemeier T (2005) Sing constrains improve the detection of differences between complex spectral data sets: LC-IR as an example. Anal Chem 77(24):7998– 8007 16. Nogueira VG, Silveira L (2005) Raman spectroscopy study of atherosclerosis in human carotid artery. J Biomed Opt 10(3):031117 17. Stone N, Kendall C, Smith J et al (2004) Raman spectroscopy for identification of epithelial cancers. Faraday Discuss 126:141– 157 18. De Gelder J, De Gussem K, Vandenabeele P, Moens L (2007) Reference database of Raman spectra of biological molecules. J. Raman Spectrosc 38:1133–1147 19. Schultz H, Baranska M, Baranski R (2005) Potential of NIR-FTRaman spectroscopy in natural carotenoid analysis. Biopolymers 77:212–221 20. Hata TR, Schlz TA, Ermakov IV et al (2000) Non-invasive Raman spectroscopic detection of carotenoids in human skin. J Invest Dermatol 115:441–448 21. Alfano RR, Liu CH et al (1991) Human breast tissue studied by IR Fourier transform Raman spectroscopy. Lasers Life Sci 4:23–28 22. Hanlon EB, Manoharan R et al (2000) Prospects for in vivo Raman spectroscopy. Phys Med Biol 45:R1–R59 23. Krafft C, Sobottka SB, Schackert G, Salzer R (2005) Near infrared Raman spectroscopy mapping of native brain tissue and intracranial tumors. Analyst 130:1070–1077 24. Wang Z, Qiao Y, Huang GS, Wang AQ, Zhang YQ, Feng JL, Yang GR, Guo Y, Liang R (2003) Glucosamine and glucosamine hydrochloride induced leukemia cell line K562 differentiation into macrophage. Chin Pharmacol Bull 19(3):290–293 25. Wang Z, Qiao Y, Huang GS, Wang AQ, Zhang YQ, Feng JL, Yang GR, Guo Y, Liang R (2003) Induction of macrophagic differentiation of leukemia cell line K562 by N-acetyl-D-glucosamine. J Fourth Mil Med Univ 24(1):46–48 26. Zhang L, Liu W, Peng Y, Wang D (2006) Antitumor activities of D-glucosamine and its derivatives. J Zhejiang Univ Sci B 7(8):608–614
