2.1.

Samples

This work was approved by the Human Research Ethics Committee at the Universidade Camilo Castelo Branco (UNICASTELO) (protocol no. 19690113.3.0000.5494) following Brazilian guidelines for the use of materials of human origin. Subjects were recruited from employees of the Laboratório Celso Matos (Santarém, Parà, Brazil). The samples consisted of urine from 54 volunteers with no renal complaints ($n=54$), the sample size being estimated using the Krejcie and Morgan estimation.³⁰ Inclusion criteria were as follows: individuals aged between 18 and 65 years, without history of pre-existing kidney disease or renal failure requiring hemodialysis. Hypertensive and diabetic individuals with signs of pyelonephritis were excluded from the study.

After explaining the goals of the study, each volunteer who agreed to participate read and signed the Consent Terms and answered a questionnaire on lifestyle habits and presence of previous disease and pathologies in close relatives. The urine of volunteers was collected in sterile vials provided by the researcher, after prior hygiene of the external genitalia, using the technique of midstream urine collection, whereby the initial flow is not collected (thus avoiding contamination from the urethra). Urination took place directly into the vial with immediate transport to the laboratory, where the colorimetric biochemical analysis took place to quantify the urea and creatinine in the samples.⁹ The colorimetric assay employs reagents to evaluate absorbance in the UV–visible range. Urea was evaluated using the UV kinetic-enzymatic method and creatinine using the modified Jaffé kinetic colorimetric method, using specific reagents (urea/BUN UV—urease/glutamate dehydrogenase, BioSystems S.A., Barcelona, Spain; and creatinine “Mod. Jaffe” 2 reagents, Eurotech Prod. Lab. Serv. Ltda., Goiânia, Brazil)^31,32 and an absorption spectrometer for medical diagnostics (Cobas Modular P, Roche Diagnóstica Brasil Ltda., São Paulo, Brazil). The repeatability and reproducibility for these colorimetric assays are, respectively, 3.3% and 4.3% for urea and 1.3% and 3.6% for creatinine.^31,32

After biochemical assay, about 10 mL of each sample was placed in a glass screw-cap vial and stored frozen ($-20°\mathrm{C}$), and at the end of a week of sample collection, samples were transported refrigerated (dry ice at $-78°\mathrm{C}$) to the RS laboratory for spectroscopic analysis, as described below. The storage of samples in a freezer in glass vials for 1 week and transport using dry ice seem to preserve the physicochemical characteristics of urine samples.³³ This study also used the Raman spectra reported by Bispo et al.,⁵ who assessed urine samples from normal subjects and diabetic/hypertensive patients without and with complications related to diabetes and HT by means of near-infrared RS and PCA with the aim of identifying potential biomarkers in the urine of diseased compared to healthy subjects. The procedures used to collect, store, and transport the urine samples and conduct Raman signal collection/analysis, including the Raman spectrometer parameters, were the same as those used in Bispo et al.

The spectra of reference samples of urea and creatinine (U5378 and C4255, Sigma-Aldrich Brasil Ltda., São Paulo, Brazil), diluted in 0.9% saline, were also obtained.

2.2.

Raman Spectroscopy

A dispersive Raman spectrometer (model Dimension P-1, Lambda Solutions, Inc., Massachusetts) was used. In this spectrometer, a diode laser (830 nm) was coupled to a fiberoptic cable (Raman probe), which was used to deliver radiation to the sample and collect the signal scattered by the sample. The probe was positioned at a distance of 10 mm from the urine sample holder, as shown in Fig. 1, and the laser power at the probe tip was adjusted to 250 mW. The light signal scattered by the sample was detected by a spectrometric CCD camera (back-thinned, deep-depleted, $1340\times 100$ pixels, Peltier-cooled, working temperature of $-75°\mathrm{C}$). The spectrometer used a holographic grating with $1200\text{lines}/\mathrm{mm}$, providing a spectral resolution of about $2{\mathrm{cm}}^{-1}$ in the useful spectral range of 400 to $1800{\mathrm{cm}}^{-1}$ (fingerprint region).

Fig. 1

Schematic diagram of the dispersive Raman system used to collect the spectral data from urine samples. Spectra were taken in triplicate. Laser wavelength: 830 nm, laser power at the probe tip: 250 mW, spectral resolution: $2{\mathrm{cm}}^{-1}$, spectral range: 400 to $1800{\mathrm{cm}}^{-1}$.

Urine samples were placed on a sample holder made of polished aluminum with holes of 4 mm diameter and about $100\mu \mathrm{L}$ volume in triplicate. The spectra of the urine samples were assessed via fiberoptics, with repeatability of excitation and signal collection geometry, thus it was possible to study the spectral differences related to differences in the biochemical constitution of the urine of different subjects and correlate spectral changes with differences in the concentration of the analytes of interest.

The acquisition and storage of the spectra were performed by a PC microcomputer using the software RamanSoft (version 1.7, Lambda Solutions, Inc., Massachusetts), which controls the exposure time of the detector and the number of acquisitions per sample via USB connection. The exposure time for collecting the spectra was 30 s (3 s and 10 accumulations), resulting in spectra with a signal-to-noise ratio as high as 100 for the Raman peak of urea ($1006{\mathrm{cm}}^{-1}$) and 20 for creatinine ($681{\mathrm{cm}}^{-1}$). Triplicate spectra of each sample were averaged for the spectral analysis.

The calibration of the spectrometer was checked before data collection. The positions (Raman shift) of the main bands of naphthalene²⁴ were verified, since the compound has characteristic intense and well-spaced bands in the fingerprint region 500 to $1700{\mathrm{cm}}^{-1}$. The calibration of the spectral response was performed by the equipment supplier and was checked by measuring the spectrum of a tungsten filament lamp with a National Institute of Standards and Technology traceable spectrum.³⁴

Prior to data analysis, spectra were preprocessed. First, background fluorescence was removed by fitting and then subtracting a fifth-order polynomial from the gross spectrum. Then, background-free spectra were subjected to smoothing using the Savitzky–Golay algorithm, decreasing the shot-noise contribution and thus increasing the spectral signal-to-noise ratio. Finally, normalization was performed using the area of the vibrational band of water at $1660{\mathrm{cm}}^{-1}$. These procedures (calibration, removal of fluorescence emission, smoothing, and normalization) allow the collection of trustworthy and high-quality Raman spectra.

2.3.

Model for Quantification of Urea and Creatinine Using Partial Least Squares

A multivariate regression model based on PLS was developed for predicting the concentrations of urea and creatinine using the spectral information contained in the Raman spectra and the concentrations obtained via the colorimetric method. The PLS routine employed the leave-one-out cross-validation approach, where a sample is left out of the model and the concentration of this sample is estimated by modeling using the remaining $n-1$ samples with a number of latent variables (LVs).^35,36 The process is repeated $n$ times for predicting the concentrations of all samples. The number of LVs to be included in the model was chosen from those with the lowest error of cross-validation in the left-out samples. By plotting the biochemical concentration versus the predicted one, the correlation coefficient ($r$) and the root mean square error of cross-validation (RMSEcv)³⁷ of the model for the dataset were calculated, which was used to estimate the assessing prediction error in the calibration model. Two quantitative models were developed using the software MATLAB 6.0 and the Chemoface toolbox,³⁸ using selected spectral regions for urea (500 to $560{\mathrm{cm}}^{-1}$, 960 to $1043{\mathrm{cm}}^{-1}$, and 1120 to $1192{\mathrm{cm}}^{-1}$) and for creatinine (650 to $940{\mathrm{cm}}^{-1}$) as predictor variables ($x$) and the concentrations of these biochemicals as predicted variables ($y$).

The developed quantitative models have been applied to a Raman dataset composed of spectra of urine from the study of Bispo et al.,⁵ aiming to estimate the concentration of urea and creatinine in those samples. In the study of Bispo et al.,⁵ spectra were measured in urine samples from healthy subjects ($n=18$) and from two groups of subjects with diabetes mellitus (DM) and arterial HT, one group ($n=20$) without complications associated with DM and HT (G1; and therefore at low risk of developing kidney disease), and another group ($n=16$) with complications such as cardiac disease, cerebrovascular disease, and peripheral vascular disease (G2; and therefore at high risk of developing renal disease), aiming at discrimination of groups using principal components analysis and discriminant analysis. Spectra were measured using the same Raman instrument and acquisition parameters presented here, and the Raman features of urea and creatinine in the same range were used to predict the concentrations from the developed model.

3.1.

Quantitative Model to Estimate Urea and Creatinine in Urine

The Raman spectra of a sample of urine from a patient without renal disease and the reference spectra of urea and creatinine diluted in 0.9% saline are presented in Fig. 2. The urine spectrum is dominated by peaks, especially those for urea and creatinine. The peaks at 527, 1006, and $1160{\mathrm{cm}}^{-1}$ indicate the presence of urea and the peaks at 681, 846, and $908{\mathrm{cm}}^{-1}$ creatinine, with attribution assigned to the symmetric stretching peak $\mathrm{C}─\mathrm{N}$ for the peak at $1006{\mathrm{cm}}^{-1}$ and $\mathrm{C}─{\mathrm{NH}}_2$ and $\mathrm{C}═\mathrm{O}$ stretching and vibrations of the aromatic ring for the peak at $681{\mathrm{cm}}^{-1}$,^39–41 confirming the peaks found in other studies of urine.^3,42,43

Fig. 2

Raman spectrum of a sample of urine from a healthy patient and the reference spectra of urea and creatinine diluted in 0.9% saline. Spectral regions inside the boxes were used in the PLS model (500 to $560{\mathrm{cm}}^{-1}$, 960 to $1043{\mathrm{cm}}^{-1}$, 1120 to $1192{\mathrm{cm}}^{-1}$ for urea and 650 to $940{\mathrm{cm}}^{-1}$ for creatinine). Wavelength: 830 nm, power: 250 mW, spectral resolution: $2{\mathrm{cm}}^{-1}$, exposure time: 30 s (3 s, 10 accumulations).

PLS models were constructed in order to estimate the concentrations of urea and creatinine using the spectral information from selected regions and the biochemical concentrations of urea and creatinine in midstream urine measured using the colorimetric method. These models with the predicted concentrations versus the colorimetric concentrations are presented in Fig. 3. In both cases, the PLS models presented higher correlation coefficients and higher RMSEcv when using five LVs. These values were $r=0.90$ and 0.91, and $\mathrm{RMSEcv}=312\mathrm{mg}/\mathrm{dL}$ and $25.2\mathrm{mg}/\mathrm{dL}$, for urea and creatinine, respectively.

Fig. 3

Correlation between concentration of (a) urea and (b) creatinine in midstream urine predicted by the PLS model versus the concentration using colorimetric method. The resulting $r$ and the RMSEcv error are presented for each biochemical.

McMurdy and Berger² found an RMSE of $4.9\mathrm{mg}/\mathrm{dL}$ for estimating the concentration of creatinine in unaltered human urine samples from a calibration dataset. For comparison, the chemical method used by them showed an error of $1.1\mathrm{mg}/\mathrm{dL}$, indicating that the model based on RS is capable of quantifying creatinine independently of clinical variations among patients.² Qi and Berger²² first demonstrated the use of liquid-core optical fiber RS and multivariate statistics in 61 clinical urine samples (excluding 12 outliers) and found that the error for assaying urea by the standard technique was $42.4\mathrm{mg}/\mathrm{dL}$, compared to $53.0\mathrm{mg}/\mathrm{dL}$ for the Raman assay. For creatinine, the errors were $5.1\mathrm{mg}/\mathrm{dL}$ for the standard assay and $6.8\mathrm{mg}/\mathrm{dL}$ for the Raman assay. These errors were ameliorated when the spectra were corrected by the absorption and scattering coefficients taken with a white light, and the model used the total integration time of 64 s, with errors of 40.4 and $4.3\mathrm{mg}/\mathrm{dL}$ for urea and creatinine, respectively.

In the study conducted by Dou et al.,⁴² the concentration of urea, creatinine, acetone, and glucose artificially added to urine samples was analyzed by Raman technique via the measurement of the intensities of the bands at 1013 and $692{\mathrm{cm}}^{-1}$ for urea and creatinine, respectively. The authors found correlation coefficients of $r=0.991$ and $r=0.998$ for urea and creatinine, respectively. The detection limits were 4.9 and $1.5\mathrm{mg}/\mathrm{mL}$, respectively, thus demonstrating the potential of near-infrared RS employed in quantitative analysis of glucose, acetone, urea, and creatinine in urine.

SERS has been proposed for the detection of metabolites in urine. Wang et al.⁶ used SERS to evaluate creatinine in samples of artificial and human urine, and found a correlation coefficient of $r=0.99$ in artificial urine over the range of 38.4 to $154\mathrm{mg}/\mathrm{dL}$ and $r=0.96$ for human urine over the range of 2.56 to $115\mathrm{mg}/\mathrm{dL}$. In a subsequent study that also evaluated these metabolites by SERS,⁷ the same authors observed a good linear correlation for creatinine concentrations over the range from 2.56 to $6.4\mathrm{g}/\mathrm{dL}$, with a coefficient of determination of $R^2=0.96$.

The correlation coefficients and the RMSEcv obtained in this work for urea and creatinine in midstream urine ($r=0.90$ and 0.91, and $\mathrm{RMSEcv}=312\mathrm{mg}/\mathrm{dL}$ and $25.2\mathrm{mg}/\mathrm{dL}$, respectively) may be considered suitable for screening analysis of clinical samples. In fact, the results achieved by other authors and presented in this work show the feasibility of performing urinalysis rapidly in single urine samples without sample preparation in an attempt to screen for renal disease. This is important as Raman analysis, which needs a low volume of sample, has no need for reagents and no residues to be discarded, and exhibits rapidity in obtaining the Raman spectrum, usefulness for screening samples, and subsequent adequacy of clinical treatment in a rapid way.

3.2.

Quantitative Analysis of Urine Samples from Diabetic and Hypertensive Patients Using Partial Least Squares

The PLS models for estimating the concentrations of urea and creatinine were applied to the spectra of urine from a group of healthy subjects and two groups of subjects with DM and HT (one group without complications associated with DM and HT—G1, and the other group with complications associated with DM and HT and risks for development of renal disease—G2), the dataset from the study of Bispo et al.⁵ These concentrations were estimated by the PLS models using the water-normalized intensities of the Raman peaks of the urine samples at the same spectral regions of urea (500 to $560{\mathrm{cm}}^{-1}$, 960 to $1043{\mathrm{cm}}^{-1}$, 1120 to $1192{\mathrm{cm}}^{-1}$), and creatinine (650 to $940{\mathrm{cm}}^{-1}$), and the mean concentrations are shown in Fig. 4. Large error bars in these groups may be attributed to the existence of subjects with considerable differences in pathological status (clinical samples), and the observance of negative estimated concentrations of creatinine and urea in some samples of G1 and G2 that might be due to noise.

Fig. 4

Mean ($\pm \mathrm{SD}$) concentrations of (a) urea and (b) creatinine estimated by the PLS model applied to the selected regions of the Raman spectra of urine from Bispo et al.⁵ in the groups of healthy subjects and subjects with DM and HT (without clinical complications—G1, and with clinical complications—G2). Symbol * indicates statistically significant differences between group G1 or G2 versus healthy group (ANOVA, $p<0.01$). Solid lines present the reference values¹² for urea and creatinine in isolated urine samples.

It was found that the group of healthy subjects that were asymptomatic for complications associated with DM and HT, such as renal disease, and had a urea concentration of $1626\pm 686\mathrm{mg}/\mathrm{dL}$ ($\text{mean}\pm \mathrm{SD}$) and a creatinine concentration of $136\pm 63\mathrm{mg}/\mathrm{dL}$, while the group of patients with DM and HT, but without clinical complications (G1), had lower concentrations of $808\pm 866\mathrm{mg}/\mathrm{dL}$ urea and $28\pm 81\mathrm{mg}/\mathrm{dL}$ creatinine. The group of patients with clinical complications (G2) had even lower concentrations of urea and creatinine, at $533\pm 561\mathrm{mg}/\mathrm{dL}$ and $20\pm 52\mathrm{mg}/\mathrm{dL}$, respectively. Since the reference values for these parameters in isolated urine are urea 0.9 to $3.0\mathrm{g}/\mathrm{dL}$ and creatinine 28 to $217\mathrm{mg}/\mathrm{dL}$ (female) and 39 to $259\mathrm{mg}/\mathrm{dL}$ (male),¹² the mean of G1 is at the limit of the reference value and the mean of G2 is outside the reference value. ANOVA indicated a significant difference between the healthy group and both $\mathrm{DM}+\mathrm{HA}$ groups ($p<0.01$). Thus, the results of Bispo et al.,⁵ who found differences in the spectral variables among healthy groups and subjects with DM and HT with and without complications, and the results presented in this study, in which the estimated concentrations of urea and creatinine in the DM and HT groups were lower than the concentrations in healthy ones, suggest that the differences in the concentrations of urea and creatinine in healthy patients and patients with DM and HT can be related to the severity of the complications that could lead to a decrease in renal function and culminate with a renal failure, and these differences can be identified using a quantitative RS model applied to single-collection midstream urine. It is important to notice that none of the patients in G2 presented a diagnosis of renal disease at the time of Raman analysis.

To assess the functional status of the kidneys, it is necessary to monitor reliable biomarkers. The use of RS for the quantitative analysis of creatinine concentration in the urine is described in the literature as an option that provides rapid measurement (within 10 s), high sensitivity (limit of detection of creatinine $<0.5\mu \mathrm{g}/\mathrm{mL}$), and reliability (variation $<5\%$) in urine analysis,⁶ together with the evaluation of multiple constituents in the urine.^13,22 In this work, the RMSE values of 312 and $25.2\mathrm{mg}/\mathrm{dL}$ for urea and creatinine in clinical urine samples indicate satisfactory results in terms of using the Raman technique for the screening of large populations using a single-collection sample.

RS has been described as a new and powerful diagnostic tool relative to routine biochemical tests.^{2,3,5,7,13,23} The Raman method would allow the simultaneous and rapid identification of biochemical components and their quantitative determination, including glucose, acetone, creatinine, urea, lipids, uric acid, and total blood protein.^22,44 The potential for the characterization of samples associated with the possibility of in situ analysis makes this analytical technique a promising tool for the evaluation of biochemical constituents. The results indicate that RS may prove to be a promising technique for the quantification of urea and creatinine in midstream urine, which could be used for the early diagnosis of renal changes at different stages of renal disease. Early diagnosis, with prompt routing to reduce or even stop the progression of renal problems, is among the key strategies to improve clinical outcomes and reduce costs associated with renal diseases.⁴³ The Raman technique could be an important tool for assessing the risk of developing renal injury in clinical diagnosis, thus providing a better quality of life for many patients due to the early detection of disabling diseases such as renal failure. The results of this study demonstrate that the use of RS is a promising technique in urinalysis for renal evaluation using a single spectrum.

Given its availability at low cost and easy implementation,⁴⁵ the technique may become an alternative for existing laboratory methods in the near future and could be applied in the presumptive identification of previously unknown renal lesions, not only for diagnosing renal failure itself but also for diagnosing the risk factors for the development of future renal disease with subsequent clinical referral.