The use of FTIR and Raman spectroscopy in combination with chemometrics for analysis of biomolecules in biomedical fluids: A review

5.1.Carbohydrate analysis using FTIR spectroscopy

Jessen et al. [33] have developed FTIR spectroscopy combined with PLSR for quantitative analysis of glucose in serum using combined 1/λ region of 988–1440 cm⁻¹ and 2620–3060 cm⁻¹ without any preparation step. FTIR spectra were subjected to the first derivative using the Savitzky–Golay algorithm with 9 smoothing points. The spectra were recorded in absorbance mode and subtracted by water blank spectrum. The actual values of glucose were measured using hexokinase method as described by [54]. The correlation between actual values (x-axis) and FTIR predicted values (y-axis) at concentration range of 2.7–23 mmol/L yielded the equation of y=0.89x+0.45 with R2 of 0.994. The standard error and limit of detection obtained were 0.29 mmol/L and 0.55 mmol/L, respectively. The authors concluded that this technique offered reliable and rapid analysis with the total time of whole analysis from sample measurement until data report is about 2–3 min.

5.2.Qualitative and quantitative analysis of protein

The changes in protein concentration in human blood can be used as an indicator for the nutritional status and certain illnesses state. ATR-FTIR spectroscopy has been applied to detect the minor difference of protein qualitatively and quantitatively in the biological samples [60]. The human serum samples spiked with proteins of human serum albumin (HSA) and immunoglobulin G (IgG) at seven different concentrations were directly analyzed on crystal ATR and their spectra were used for making prediction and validation models using multivariate calibration of PLSR. The wavenumber (1/λ) of 1640-1560 cm⁻¹, associated with stretching vibration of amide groups, were used for quantitative analysis. This band was increased with the increasing levels of proteins (HSA and IgG). In the sipked samples model, the R2 values for the prediction of HSA and IgG were 0.982 and 0.947, respectively in liquid patient samples. The cross validation model using leave one out technique, root mean square error of cross validation (RMSECV) obtained were relatively low, namely 3.065 mg/mL and 2.365 mg/mL respectively for HSA and IgG. This method was succesfully applied for analysis of patient samples with acceptable accuracy and precision. ATR-FTIR spectroscopy was also used for the quantitative analysis of IgA, IgG and IgM in myeloma affected blood samples. Quantitative analysis was performed by investigating the ratio (R) of absorbance values at several wavenumbers (cm⁻¹), namely R₁ (Abs_2874/2930), R₂ (Abs_1542/1654), R₃ (Abs_1315/1404), R₄ (Abs_1165/1244) and R₅ (Abs_953/1025). The results showed that these ratios are decreased in myeloma samples than those in normal serum samples [55].

The combination of ATR-FTIR spectra and multivariate calibration of PLSR was succesfully applied for quantification of IgA, IgG, IgM in the plasma samples. The spectra of dried human plasma samples were analyzed at mid IR region (4000 to 500 cm⁻¹) at resolution of 2 cm⁻¹ and number of scanning of 32. The R2 for the correlation between actual values of IgA, IgG and IgM and FTIR-predicted values were 0.98, 0.98, and 0.97, respectively. The error in validation model expressed by root mean square error of prediction (RMSEP) was relatively low, namely 0.05 g/L (IgA), 0.4 g/L (IgG), and 0.03 g/L (IgM). The relative standard deviation for reproducibility study was less than 2% for all analytes [9].

The complex nature of biological fluid samples is still the limiting factors for direct analysis using ATR-FTIR spectroscopy. For this reason, some innovations by combining sample preparation step and direct measurement using FTIR spectroscopy has been proposed. Elsohaby et al. [22] have coupled centrifugal ultrafiltration and ATR-FTIR spectroscopy for quantitative analysis of IgA in human serum. The actual values of IgA were determined using reference method of immunoturbidimetric assay. The whole serum samples and residue reconstituted with saline after centrifugal ultrafiltration were subjected to FTIR spectral measurement at 4000-650 cm⁻¹. PLSR calibration models using 5 factors (for whole serum) and 9 factors (for the reconstituted residue) was succesfully applied for relating the actual values of IgA and FTIR predicted values. The RMSEP value reported was in the range of 79-102 mg/dL.

ATR-FTIR at 1/λ of 4000-600 cm⁻¹ has been used for determination of proteins (albumin, immunoglobulin, total globulin, and albumin/globulin coefficients) in serum and blood samples. The actual values of proteins were determined using electrophoresis which involved 200 samples in calibration sets and 120 samples in validation sets [48]. Some 1/λ region and FTIR spectral processing (normal versus derivative) have been optimized to get the best results in terms of the highest R2 value and the lowest root mean square error of calibration (RMSEC) and RMSEP values. The regions related to amide I and amide II bands were used for quantitative analysis, namely albumin at 1/λ 1228-1024 cm⁻¹ using first derivative (1st der) spectra with 6 factors, γ-globulin at 1/λ 1392-1024 cm⁻¹ using 1st der spectra with 6 factors, total globulin at 1/λ 1440-1018 cm⁻¹ using 1st der spectra with 4 factors, and albumin/globulin coefficients using 1/λ 1531-1242 cm⁻¹ using 1st der spectra and standard normal variate (SNV) with 5 factors. The R2 values for the correlation between actual values and FTIR predicted values were 0.932, 0.850, 0.886 and 0.900 for albumin, γ-globulin, total globulin and albumin/globulin, respectively. The RMSEC values ranged 0.087-0.136 g/dL, RMSEP values of 0.093-0.191 g/dL. The cross validations resulted RMSECV value of 0.095-0.149 g/dL. This result indicated that FTIR spectroscopy combined with PLS regression could be used to predict proteins in serum samples.

5.3.Quantitative analysis of lipids

ATR-FTIR spectra have been applied for rapid analysis of very low density lipoprotein-triglyceride (VLDL-TG) serum in maternal serum without any sample preparation. The multivariate calibration of PLS was applied for prediction the levels of VLDL-TG. Using second derivative FTIR spectra at combined 1/λ region of 3100–2800 and 1800–1000 cm⁻¹ with 3 PLS factors, the correlation between actual values (x-axis) of VLDL-TG as determined by Triglyceride GPO-PAP spectrophotometric assay (Roche Diagnostics) and FTIR predicted values (y-axis) was y=0.985x+0.784 with R2 value of 0.985. The calibrated model was validated using independent samples to yield R2 value of 0.890 and RMSEP of 0.20 mol/L. This method has been successfully applied in serum and in ultracentrifugated VLDL fractions from 32 patients with acceptable accuracy and precision. Quantitative analysis of VLDL-TG by FTIR spectroscopy seems to be promising screening technique to identify women with increased risk of adverse pregnancy outcomes [47].

The presence of triglycerides (TG) in serum was analyzed using FTIR spectroscopy and multivariate calibration of PLSR. FTIR spectra were subjected to mean centering and first derivatization. The combined 1/λ regions of 1117–1200, 1429–1483, 1700–1785, 2780–3060 cm⁻¹ was used to predict the levels of TG without any preparation step. The correlation between actual values of TG as determined using enzymatic colorimetry (x-axis) according to Wahlefeld [65] and FTIR predicted values (y-axis) resulted an equation of y=0.73x+0.15 with R2 of 0.989, standard deviation of response of 0.035 mmol/L and limit of detection of below 0.34 mmol/L.

5.4.Analysis of nucleic acids in biomedical fluids

Quantification of DNA in eukaryotic cells has been performed using FTIR spectroscopy method in combination with chemometrics of PLS regression. DNA quantification was performed using standard consisting of DNA and hemoglobin standards. PLS model was built at the fingerprint region of FTIR spectra (1800-850 cm⁻¹). The model was evaluated using R2 values. The peaks of DNA were observed at 1225 cm⁻¹ (asymmetric phosphate stretching), 1087 cm⁻¹ (symmetric phosphate stretching), 1051 cm⁻¹ (C–O–C stretching), 970 cm⁻¹ (C–C stretching), and 1715 cm⁻¹ (C=O stretching). PLS was performed either in dehydrated or hydrated models. The R2 values were higher than 0.96 for all build models with lower RMSEP. It indicated that the PLS model has good accuracy, good predictivity, and good of fitness for DNA quantification [67].

FTIR spectroscopy has been used for analysis of nucleic acid in bladder cancer subject. FTIR measurement was performed at mid infrared region (4000-400 cm⁻¹) using resolution of 4 at 100 scans number. The peak of PO₂⁻ stretching vibration in nucleic acid appeared specific at 1086 cm⁻¹ measured using FTIR spectrophotometer. In the condition of bladder cancer, the peak shifted to around 1087 ± 0.87 cm⁻¹ compared to control. Some vibrations especially stretching vibration was increased significantly in cancer subject compared to control. The peak appeared at 967 cm⁻¹ could be considered as marker peak to detect tumors. FTIR measurement of sample from bladder cancer subjects showed slight shifting from 967.22 ± 0.33 cm⁻¹ to 967.40 ± 0.31 cm⁻¹. Moreover, the peak at 1/λ 930 cm⁻¹ in bladder cancer appeared at higher intensity than in control. The shifting of FTIR spectra is associated with considerable changes of nucleic acid (DNA) in the case of cancer because of uncontrolled cell proliferation, uncontrolled cell growth, and decreasing of nucleid acid stability. The use of chemometrics principal component analysis was also successfully differentiate between samples from normal and bladder cancer subjects [29].

6.1.Analysis of carbohydrate in serum using Raman spectroscopy

Application of Raman spectroscopy has been explored for glucose analysis in blood serum. Sugar concentration in human diabetic serum has been successfully analyzed using Raman Spectroscopy. Blood serum samples were obtained from normal and diabetic subjects. Raman measurement was performed in the range of 2000-800 cm⁻¹ at 532 nm. Raman spectra of glucose in diabetic mellitus patients appeared at 1168 cm⁻¹ which correspond to the carbohydrate peak. This peak appeared at very low intensity or almost lacking intensity in the normal serum (healthy subjects). The C=C (carbon double bond) could be observed from peak appeared at 1650 cm⁻¹. One specific peak at 1463 cm⁻¹ appeared only in blood serum from diabetic subjects that can be used as marker peak of diabetic subject. The glucose peaks at 1168 cm⁻¹ and 1531 cm⁻¹ were plotted against intensity to determine glucose concentration. Eventhough glucose from normal serum still could be detected, the intensity of glucose peaks from diabetic serum showed very high intensity than normal serum. Therefore, the increasing concentration of glucose could be detected using Raman spectroscopy to differentiate glucose level from serum of normal and diabetic subjects [26].

Quantitative analysis of glucose in human serum has been performed using Raman spectroscopy and chemometrics of PLS regression method. Raman measurement was carried out using diode laser at 830 nm, and each sample was measured in three replicates. The measurement was performed in the range from 1800-400 cm⁻¹. The bands of glucose were observed at 507 cm⁻¹ (skeletal modes of glucose), 1065 cm⁻¹ (C–C, C–O, C–H and C–OH modes), and 1128 cm⁻¹ (C–O stretching). Raman spectra were subjected to PLS regression to predict glucose concentration in human serum measured by Raman spectroscopy. The actual value of glucose was obtained from colorimetric method measurement. PLS model was evaluated using R2 and RMSECV values. Result of PLS model showed that Raman spectroscopy combined with PLS regression method is a good combination for quantitative analysis of glucose in human serum. PLS could be used to predict glucose concentration in the range of 0-200 mg/dL. Using 10 latent variables (LVs), PLS showed high R2 value (0.86) and lower RMSECV value (26.4 mg/dL) [58].

6.2.Qualitative and quantitative analysis of lipids using Raman spectroscopy

Conventional Raman spectroscopy (CRS) has been used for analysis of very low-density lipoprotein-triglyceride (VLDL-TG) serum in maternal serum without any sample preparation. The samples were excited using Nd:YAG 1064 nm laser at power of 300 mW. Raman signal was collected at 1/λ of 1000–3700 cm⁻¹ at resolution of 4 cm⁻¹ for 90 min and averaged over 2048 accumulations. The multivariate calibration of PLS was applied for prediction the levels of VLDL-TG. Using first derivative Raman spectra at combined 1/λ region of 3100–2800 and 1800–1000 cm⁻¹ with 3 factors, the correlation between actual values (x-axis) of VLDL-TG as determined by triglyceride spectrophotometric assay (Roche Diagnostics) and Raman predicted values (y-axis) was y=0.994x+0.281 with R2 value of 0.994. The prediction models using PLS revealed R2 of 0.800 and RMSEP of 0.27 mol/L. This method has been successfully applied in serum and in ultracentrifugated VLDL fractions from 32 patients with acceptable accuracy as indicated by high R2 and precision as indicated by low RMSEP [47].

Quantification of lipid components has been performed using Raman spectroscopy and chemometrics of PLS regression. The lipid components consist of cholesterol, triglycerides, high-density lipoprotein (HDL), and low-density lipoprotein (LDL). Raman spectral data were collected using a near-infrared dispersive Raman spectrometer. The diode laser is 830 nm with 25 mW. For excitation and collection of Raman spectra, a Raman probe was used and connected to spectrometer. PLS model was carried out using wavenumber of 1800-400 cm⁻¹. The actual concentration of cholesterol, triglycerides, HDL, and LDL were measured using colorimetric methods. Lipid bands could be observed at 1676 cm⁻¹ (C=C stretching of cholesterol), 1659 cm⁻¹ (C=C stretching of unsaturated lipid), 1452 cm⁻¹ (CH₃, CH₂ bending of lipids), 1269 cm⁻¹ (C=C stretching of saturated lipid), 1085 cm⁻¹ (C–O stretching or C–C skeletal of unsaturated lipid), and 700 cm⁻¹ (stretching of steroid ring from cholesterol). The R2 values obtained from PLS regression were 0.96; 0.98; 0.75; and 0.86, while the RMSEC values were 15.9; 35.4; 17.8; and 24.2 mg/dL for cholesterol, triglyceride, HDL, and LDL respectively. Whereas the latent variables (LVs) used were 9 for cholesterol and 8 for triglyceride, HDL, and LDL, respectively [58].

Raman spectroscopy has also been used for the quantitative analysis of some compounds in serum including cholesterol, HDL, LDL, triglycerides, and glucose. The concentration was compared to that obtained from enzymatic assay. Chemometrics of PLSR was used to build calibration models. PLSR showed good correlation between Raman and enzymatic assay as shown by the correlation coefficient value for all measured compound was higher than 0.940. The region used for building PLSR model was in the range of 4000-450 cm⁻¹ and the peaks were normalized to the peak at 1514 cm⁻¹ (peak with highest intensity). The use of Raman was also applied for serum differentiation from normal serum and gastric cancer serum samples. Raman spectra were subjected to chemometrics of partial least square-discriminant analysis (PLS-DA) for classification. Serum from normal sample could be clearly differentiated with serum from gastric cancer sample by using PLS-DA method [3].

6.3.Analysis of nucleic acid in biomedical fluids using Raman spectroscopy

Surface-enhanced Raman spectroscopy (SERS) and multivariate analysis of principal component analysis-linear discriminant analysis (PCA-LDA) have been used for analysis of ribonucleic acid (RNA) in serum from normal and colorectal cancer subjects. Mixture of colloidal silver solution and MgSO₄ were used as substrate for RNA measurement to enhance the Raman intensity of RNA. Raman measurement was performed using diode laser at 785 nm in the region of 1500-500 cm⁻¹. Raman spectra from normal subjects were similar with those from colorectal cancer subjects. SERS peaks of RNA appeared at 527 cm⁻¹ (uracil), 573 cm⁻¹ (cytosine, guanine), 678 cm⁻¹ (guanine), 714 and 736 cm⁻¹ (adenine), 859 cm⁻¹ (phosphate group), 1008 and 1033 cm⁻¹ (C–O and C–C stretching), 1240 cm⁻¹ (phosphate asymmetric stretching), 1277 cm⁻¹ (nucleic acid and phosphate), 1316 cm⁻¹ (guanine), and 1391 cm⁻¹ (C–H rocking). However, there were several differences located in several peaks, especially in the peak intensities of RNA between normal and cancer groups. In normal subjects, the peaks at 1391, 1240, 1008, 860, and 714 cm⁻¹ are higher than in cancer subjects, whereas the peak intensities at 1316, 1277, 1033, and 573 cm⁻¹ are higher in cancer subjects than in normal. Multivariate analysis of PCA-LDA successfully differentiated RNA serum from normal and cancer subjects. RNA serum samples from normal and cancer subjects were clearly separated using LDA analysis. LDA showed good accuracy (89.1%) for predicting RNA samples from cancer subjects and showed good accuracy (92.35%) for predicting RNA samples from normal subjects [13].

Raman spectroscopy in combination with chemometrics of PCA and PLS-DA was used for detection of nucleic acids and protein changes during breast cancer development [45]. Some Raman spectral markers were identified in relation to the specific biomolecule markers such as peak at Raman shift of 689 cm⁻¹ (nucleotide conformation), 788 cm⁻¹ (phosphodiester bands in DNA), 828 cm⁻¹ (tyrosine/protein), 848 and 1663 cm⁻¹ (DNA) have not previously been identified. Principal component analysis-factorial discriminant analysis (PCA-FDA) was successfully used for the classification of Raman spectra related to biomolecules changes in the breast cancer patients, according to different stages of the disease.

Surface-enhanced Raman Spectroscopy (SERS) has been used for analysis of nucleic acids in human infected Hepatitis C (HCV). The Raman spectra of healthy samples were different from those in infected patients with HCV. Several peaks from infected samples were used as markers for nucleic acid in the infected patients. The peaks appeared at 720 cm⁻¹ (adenine), 1077 cm⁻¹ (PO₂ RNA stretching), 1678 cm⁻¹ (C–O stretching of RNA), 178 cm⁻¹ (RNA), 1641 cm⁻¹ (amide), 1721 cm⁻¹ (protein stretching), and 1738 cm⁻¹ (C–O ester group in lipids). Principal component analysis has been successfully used for differentiation between normal serum and infected Hepatitis C serum. SERS spectra from normal subjects were clearly separated from SERS spectra obtained from cancer subjects as shown by PCA score plot [19].

6.4.Protein analysis in biomedical fluids using Raman spectroscopy

Surface-enhanced Raman spectroscopy (SERS) combined with multivariate analysis of PLSR has been successfully applied to differentiate saliva protein from healthy subject, patient with benign breast tumor, and patient with malignant breast cancer. There are some differences of SERS spectra resulted from normal and tumor group. Some intense peaks appeared at 1684 cm⁻¹, 1340 cm⁻¹, 1208 cm⁻¹, and 1004 cm⁻¹ in benign and malignant breast tumor group, whereas peaks at 1176 cm⁻¹, 1049 cm⁻¹, and 621 cm⁻¹ are higher in normal group. This indicates the changes in saliva proteins either in benign or malignant breast tumors. Spectra shifting in saliva protein of malignant breast cancer could also be monitored using SERS. The peak of saliva protein at 1310 cm⁻¹ which appeares in normal subject is shifted to 1318 cm⁻¹ in malignant breast cancer subject. The changes in SERS peak could be used as marker peak for the detection of benign and malignant breast cancer. More exploration was carried out to confirm the spectra from three different groups using multivariate analysis of PLS-DA. Model validation was carried out using leave one out cross validation technique and the score plot of PLS-DA showed clear classification among these three groups (normal, benign tumor, and malignant tumor). PLS-DA showed good diagnostic sensitivity (75.75%, 72.73%, and 74.19%) and good diagnostic specificity (87.63%, 78.35%, and 82.47%) for normal, benign, and malignant breast tumor subject, respectively [25].

Raman spectroscopy and multivariate analysis of PLS have been used for qualitative and quantitative analysis of albumin and immunoglobulin protein in human serum samples. Raman measurement was performed at 785 nm. Raman spectra of protein fraction could be observed at the band of 1650 cm⁻¹ (vibration of amide I), 1436-1460 cm⁻¹ (CH₂ deformation), 1300-1343 cm⁻¹ (CH deformation), 1004 cm⁻¹ (phenylalanine), 944 and 890 cm⁻¹ (C–C stretching), and 856 cm⁻¹ (tyrosine) [18,35,66]. Raman spectra of albumin, Immunoglobulin A (Ig A), and Immunoglobulin G (Ig G) are similar. However, the intensities of albumin spectra have the highest spectra intensity compared to Ig A and Ig G. Quantitative analysis for determination of albumin, Ig A, and Ig G was performed using PLS regression. PLS models were evaluated using R2, RMSEC, and RMSECV value. PLS model showed a good prediction model to determine albumin and immunoglobulin concentrations. The R2 values for calibration model were 0.95, 0.94, and 0.99 for albumin, Ig A, and Ig G, respectively. The high values of R2 showed good accuracy and predictivity of PLS models. The RMSEC values were 5.9, 0.3, and 0.4 for albumin, Ig A, and Ig G, respectively whereas the RMSECV values were 7.8, 0.5, and 1.2 for albumin, Ig A, and Ig G, respectively. Lower RMSEC and RMSECV values indicated lower error values in PLS models [1].

The combination of Raman spectroscopy and chemometrics of PLS has been used for protein analysis of human blood fractions. Raman measurement was carried out at 785 nm and the blood samples were divided into two categories; whole blood and blood plasma. Raman spectra of blood plasma showed several peaks corresponding to protein or protein component signals such as 820 cm⁻¹ (tyrosine), 950 cm⁻¹ (deformation of C-H group), 1002 and 1080 cm⁻¹ (phenylalanine), 1160 cm⁻¹ (deformation of C-C group), 1250 cm⁻¹ (α-helix of Amide III), 1330 cm⁻¹ (tryptophan), 1450 cm⁻¹ (CH₂ group), and 1650 cm⁻¹ (β-helix of Amide I). Raman spectra of whole blood were similar to blood plasma except in some peaks, namely 570 cm⁻¹ (deformation of FeO₂ group), 750 and 670 cm⁻¹ (pyrrole), 1120 cm⁻¹ (C–N group), 1227 cm⁻¹ (deformation of C-H group), and 1550 cm⁻¹ (phenylalanine). PLS method was performed to quantify protein either in blood plasma or whole blood. The variable in projection (VIP) analysis from PLS measurement was carried out to identify peaks which have strong contribution to build PLS model. Raman peaks at 570-700 cm⁻¹, 1120 cm⁻¹, 1550 cm⁻¹ are the most informative peaks in whole blood that correspond to hemoglobin and they are not found at blood plasma spectra. Peaks at 1002 cm⁻¹, 1227 cm⁻¹, 1440 cm⁻¹, and 1630 cm⁻¹ are the largest VIP value and they can be found both in blood plasma and whole blood that correspond to the peaks of albumin and globulin [41].

Raman spectroscopy has been employed for protein analysis in patient with asthma. Protein could be used as marker molecule to differentiate between normal and pathology condition. In pathology condition, certain protein could be different either in composition or structure from normal condition. Serum samples from asthma patients have been analyzed using Raman spectroscopy to investigate the changes in protein. Result showed that the proteins structures changed as the increasing severity of asthma [53]. Raman spectroscopy has been used for protein analysis in serum. Human blood serum in normal health was collected and then measured using Raman at wavelength 780 nm. Protein could be detected using Raman and the signals appeared at several regions. Most of protein signals appeared in the region of 1544 cm⁻¹ and 1648 cm⁻¹, which correspond to the signal of amide II and amide I. Other peak at 904 cm⁻¹ is associated with deformation of C-H bond in protein. Proteins have either single bond of C–O or double bond of C=O that appear at 1740 cm⁻¹ for single bond stretching vibration as well as 1780 cm⁻¹ and 1816 cm⁻¹ for double bond stretching vibration [17].