2.1.

Domestic Cat Blood Serum Samples

A total of 59 cat blood serum samples were obtained from the Protection Animal Society (São Paulo, SP, Brazil) and used in the presented work. The samples were previously evaluated by the ELISA test, and following that, the Raman spectrum was taken for each sample.

2.2.

ELISA

An ELISA test was performed at the Department of Pathology of Universidade Estadual Paulista (UNESP) (Jaboticabal, SP, Brazil) aiming to detect the presence of anti-T. gondii antibodies (IgG) in the cat serum. The ELISA test was performed following the protocol described by Domingues ³⁷ Serological results were quantified and defined as a set of ELISA levels (between 0 and 9), corresponding to a range of the ELISA optical density (OD) of the samples,³⁸ according to Table 1 .

Table 1

OD by ELISA test, corresponding ELISA levels, and IgG serology.

2.3.

Raman Spectroscopy

A Raman spectrometer was used to collect the spectra; this instrument is described elsewhere.^{18, 24} Briefly, the excitation source was a Ti:sapphire laser (Spectra Physics, model 3900S) pumped by a $6\text{-}\mathrm{W}$ multiline argon laser (Spectra Physics, model 2709S), tuned to $830\mathrm{nm}$ . The Raman scattered signal from the sample was collected and coupled into the spectrometer entrance slit by a set of lenses. The Rayleigh scattering from the sample was strongly attenuated in the collecting system by an holographic notch filter (Kaiser Optical Systems, model HSPF 830AR2.0). The optical detection system was composed of a high coupling spectrograph (Kaiser Optical Systems, Inc., $f∕1.8\mathrm{i}$ ) and a liquid-nitrogen-cooled, deep depletion CCD camera and controller (Princeton Instruments, model EEV $1024\times 256$ and model ST130, respectively). To collect the Raman signal with an optimal resolution, the spectrometer entrance slit width was set³⁹ to $200\mu \mathrm{m}$ , with a spectral resolution of about $10{\mathrm{cm}}^{-1}$ . The CCD controller was connected to a personal computer that stored and processed the Raman spectra.

The quartz cuvette with the serum sample was positioned facing the spectrometer entrance slit. The pumping laser beam impinged on the cuvette, forming a right angle with respect to the signal-collecting optical axis. The serum samples were irradiated with $80\text{-}\mathrm{mW}$ laser power and an acquisition time of $100\mathrm{s}$ for each sample.

The optical detection system (spectrometer and CCD) was calibrated in wave numbers using an organic compound, indene $\left({\mathrm{C}}_9{\mathrm{H}}_8\right)$ , which has a well-established Raman spectrum that presents intense and well-distributed bands covering the whole spectral region between 800 and $1800{\mathrm{cm}}^{-1}$ .

2.4.

Preprocessing of NIR Raman Spectra

The contribution of the fluorescence emission to the Raman spectra was eliminated by fitting and subtracting a third-order polynomial from the gross spectrum. After calibration and background subtraction, the Raman spectra were filtered using the MATLAB^® software to reduce the high-frequency noise (associated with the shot and readout noise). Cosmic ray spikes were manually removed. Once filtered, spectra were normalized by the most intense band and submitted to a multivariate statistical analysis, composed of the PLS regression method.

2.5.

Multivariate Statistical Analysis—PLS method

Samples were classified according to their IgG serology into two groups for PLS analysis: negative and positive samples. Then the 59 preprocessed spectra were separated randomly in two groups: 49 were used as the calibration dataset and the remaining 10 as the validation (prediction) data set, including positive and negative samples in the calibration and validation groups. The PLS model was implemented by using the calibration set and the corresponding latent variables (LVs) were calculated from the first LV to the 10th LV.

To determine the number of LVs to be included in the PLS model that provides the minimum error predicted by the PLS compared to the ELISA OD values, we calculated the root mean square error of prediction (RMSEP) for each LV used in the validation data set as follows:

Likewise, the parameters sensitivity (Se), specificity (Sp), accuracy, positive predictive value (PPV), and negative predictive value (NPV) for the Raman/PLS model were calculated. Those parameters are defined as $\mathrm{Se}=\mathrm{TP}∕(\mathrm{TP}+\mathrm{FN})$ , $\mathrm{Sp}=\mathrm{TN}∕(\mathrm{TN}+\mathrm{FP})$ , $\text{accuracy}=(\mathrm{TP}+\mathrm{TN})∕(\mathrm{TP}+\mathrm{FP}+\mathrm{TN}+\mathrm{FN})$ , $\mathrm{PPV}=\mathrm{TP}∕(\mathrm{TP}+\mathrm{FP})$ , and $\mathrm{NPV}=\mathrm{TN}∕(\mathrm{FN}+\mathrm{TN})$ , where TP, FP, TN, and FN are the numbers of true positives, false positives, true negatives, and false negatives, respectively.

3.1.

ELISA

The 59 samples were classified as negative serum (N), medium positive (MP) and strong positive (SP) according to the following criterion (Table 1): N—ELISA level equal to 0, MP—ELISA levels ranging from 1 to 6, and SP—ELISA levels above 6. Based on this classification, it was obtained $13\mathrm{N}$ , $30\mathrm{MP}$ , and 16 SP blood sera.

3.2.

Raman Spectroscopy

Figure 1 shows the Raman spectra of three characteristic samples with N, MP, and SP ELISA serologies in the spectral range between 900 and $1800{\mathrm{cm}}^{-1}$ . All spectra have a similar profile, with prominent band positions at 944, 1003, 1058, 1169, 1242, 1331, 1451, and $1655{\mathrm{cm}}^{-1}$ for the three types of serology. An attempt to assign the bands presented in the serum, according to that reported by the literature, would be band at $944{\mathrm{cm}}^{-1}$ (proteins: $\mathrm{C}\mathrm{C}$ stretching mode of skeletal backbone α-helix conformation); at $1003{\mathrm{cm}}^{-1}$ (phenylalanine: symmetric ring breathing mode); at $1058{\mathrm{cm}}^{-1}$ (phenylalanine and lipids: skeletal $\mathrm{C}\mathrm{C}$ stretch); at $1169{\mathrm{cm}}^{-1}$ (amino acids and nucleic acids); at $1242{\mathrm{cm}}^{-1}$ (amide III from proteins: $\mathrm{N}\mathrm{H}$ bending, $\mathrm{C}\mathrm{N}$ stretch; $\mathrm{C}{\mathrm{H}}_2$ wag); at $1331{\mathrm{cm}}^{-1}$ (proteins: $\mathrm{C}{\mathrm{H}}_3$ and $\mathrm{C}{\mathrm{H}}_2$ wagging; polynucleotide chain; phospholipids); at $1451{\mathrm{cm}}^{-1}$ (lipids and proteins: $\mathrm{C}{\mathrm{H}}_2$ deformation and proteins: $\mathrm{C}{\mathrm{H}}_2$ bending); and at $1655{\mathrm{cm}}^{-1}$ (lipids: $\mathrm{C}\mathrm{C}$ stretching and amide I from proteins: $\mathrm{C}=\mathrm{O}$ stretching α-helix conformation).^{14, 16, 40, 41} Differences observed in the band location compared to literature are within the spectrometer resolution $(<10{\mathrm{cm}}^{-1})$ . The overall spectrum profiles are very similar for negative and positive sera; however, it was possible to observe Raman bands with higher intensities for the MP and SP samples when compared to the N samples, which could be explained by the presence of IgG antibodies. For instance, the bands at $1242{\mathrm{cm}}^{-1}$ (proteins) and $1058{\mathrm{cm}}^{-1}$ (lipids) have significant differences, increasing with disease gravity.

Fig. 1

Raman spectra of cat blood sera with the following serology to toxoplasmosis (IgG): negative (N), medium positive (MP), and strong positive (SP).

3.3.

PLS Model

The PLS regression model was used to develop a method to obtain the equivalent OD for the Raman spectra, thus obtaining an OD quantification and serology identification for the Raman/PLS model of cat blood sera. The data set was divided into two groups: 49 spectra for training the model and 10 spectra for validation (prediction), including negative and positive sera in both groups. First, we developed the calibration curve with the 49 spectra, in which the OD obtained by the ELISA test were used to determine the equivalent OD predicted by Raman/PLS. To determine how many LVs should be included in the PLS model to give the lowest prediction error, the 10 spectra from the validation data set were submitted to the PLS model using the number of LVs from 1 to 10. Then the RMSEP parameter was calculated for the OD values of the validation data set (Figure 2 ). We found that a lower RMSEP parameter is reached using seven LVs, so the OD values calculated using seven LVs for both training and validation groups versus the OD by ELISA are plotted in Fig. 3 , whereas that for the validation group alone, separated by serology type, are shown in Fig. 3. We observed that all points lie close to the diagonal, zero-error line. The prediction RMSEP had a value of about $\mathrm{OD}=0.18$ , which is satisfactory for a method of rapid serology assay considering that the upper cutoff for the negative serology was $\mathrm{OD}=0.221$ .

Fig. 2

Plot of the RMSEP values of the OD predicted by the Raman/PLS model compared to OD values by ELISA using the first 10 LVs of the validation Raman data set.

Fig. 3

(a) Plot of the OD values predicted by the Raman-PLS model using seven LVs applied to cat blood sera versus the OD values obtained by ELISA test. Training group: 49 spectra, validation group: 10 spectra. The diagonal line represents the zero-error prediction line. (b) Plot of the OD achieved by the Raman/PLS model versus the OD from the ELISA test on a scale of 0 to 1.8 of the validation data set. The horizontal line represents $\mathrm{OD}=0.221$ , which discriminates negative from positive IgG serologies.

3.4.

Diagnosis Analysis

A comparison between OD values obtained by the Raman/PLS regression model and those obtained by the ELISA test in the validation data set is shown in Table 2 . The table also shows the Se, Sp, accuracy, and PPV and NPV for the validation data set, considering positive the sera with $\mathrm{OD}>0.221$ . We did not observe any false negative diagnosis, although we found only one sample out of the 10 analyzed that was identified as negative by the ELISA and diagnosed as positive by the Raman/PLS method. With respect to this sample, it was verified that this misclassified sample presented an OD value that is very close to the upper cutoff value for negative serum, and, therefore, the prediction seems to be more uncertain.

Table 2

Classification capabilities of the Raman/PLS diagnosis correlated to the ELISA test (OD values) for the IgG from serum of domestic cats of the validation spectra.

Riley studied the diagnosis of the failure of transfer of passive immunity in horses using Fourier transform (FT) infrared spectroscopy employing PLS regression analysis.¹⁵ They found parameter values of about 92, 96, 95, 86, and 98%, for Se, Sp, accuracy, PPV, and NPV, respectively. To compare Riley’s work and the presented study, one should notice that the meaning of negative and positive is different in these works: FPT-positive in the Riley paper is equivalent to serum-negative for toxoplasmosis in this study and vice versa. Therefore, the meanings of Se, Sp, PPV, and NPV are apposite in these works. These values are comparable to the ones presented in this model for toxoplasmosis diagnosis using Raman/PLS regression analysis (Table 2).

Table 3 compares the capabilities of the dispersive Raman spectroscopy in conjunction with a multivariate statistical analysis in diagnostic models for different diseases as reported in the literature.^{42, 43, 44, 45} We can see that our results on toxoplasmosis identification agree well with those obtained for other diseases using the same spectroscopic technique.

Table 3

Comparison of the diagnostic capabilities of the Raman spectroscopy associated with multivariate statistical analysis in the diagnosis of several diseases.

The NIR Raman technique has been proven to be suitable for biotissue analysis (including tissues, blood, and serum) due to the intrinsic vibrational information that is carried out by the Raman signal without interference from water. By using Raman spectroscopy, usually no sample preparation is required, a small amount of sample is used, the biochemical alterations could be obtained directly in real time, and depending on the biochemical of interest, it results in a unique molecular fingerprint, which can be used to discriminate qualitatively and quantitatively the diseased tissue. Recent developments in Raman instrumentation include the use of low-cost diode lasers, CCD detectors with electronic cooling, fiber optic probes designed to reduce fiber background and increase SNR, and compact spectrometers with high throughput.⁴⁶