Preparation of parasite culture
Plasmodium falciparum parasites were cultured in a complete medium at 1% haematocrit at 37 °C in a 5% CO2/3% O2/balanced N2 gas mixture. The complete media, a mixture of healthy O+ human blood and RPMI 1640 culture media, were used as the control samples. Infected human red blood cells were washed three times (to remove any undesired components such as bacteria) and stored at between 4–8 °C. They were added to the complete media to initiate the culture process. Whole blood samples were taken from the consenting 20 malaria patients as well as 23 healthy volunteers at the Kenyatta National Referral Hospital, Kenya. To take part in the study, all of the participants were well briefed and agreed to provide informed consent. They all also agreed willingly and signed a written consent form that was part of the Ethics Approval forms. The study was approved by the Kenyatta National Hospital-University of Nairobi Ethics and Research Committee (ERC certificate number: P112 / 03 /2018). All the methods that were used to handle and prepare samples in this study were performed in accordance with the relevant guidelines provided in the Ethics Research Committee guidelines and regulations. The Haematocrit (initial parasitemia) was measured from a packed red blood cells volume that had been centrifuged in a swinging bucket rotor at 2000xg for 5 minutes at room temperature. Schizonts were isolated on Percoll cushions. For RS parasite analysis, aliquots were layered on top of 70% Percoll in 1.6 mL tubes and centrifuged at 4,000X g for 5 minutes. Red blood cell pellets were then washed once with RPMI 1640 medium and immediately frozen at -20 °C until they were used. The pellets were added to the complete media and kept in the CO2 incubator at 37 0C and thereafter harvested after every 24 hours for four days [52].
Staining and examination of thick and thin smears
Each time the media was harvested both thick (6 μL) and thin (2 μL) blood smears were prepared on a glass slides. The slides were immersed in absolute methanol for fixation then dried. Ten percent (1:9 mL) for 10 minute and 3% (3: 97 mL) for 45–60 minute fresh, working Giemsa stains were prepared. The stained blood smears were washed with water to remove excess stain, dried and observed under a microscope. The thick blood smear was used to microscopically confirm infection by plasmodium falciparum; while the thin smears were used to determine the parasite morphology. The smear was first screened at a low magnification (10X × 40X objective lens) to detect suitable fields for analysis then later examined using X100 oil immersion. Parasitemia and morphology were determined at each harvested stage. The percent of infected red blood cells was determined by enumerating the number of infected in relation to the number of uninfected cells. This method required the preliminary determination of the number of erythrocytes present in the average microscopic field. Although in this work sub-culturing was not done and therefore the media and pellets were not renewed hence the parasitemia should be higher, the parasite load, expressed as a percentage parasitemia was in agreement with other studies where plasmodium falciparum was cultured without change of media [53, 54].
LIBS instrumentation and sample analysis
The LIBS system used in this study is the Ocean Optics LIBS2500plus with a laser wavelength of 1064 nm. This is a broad-band, high resolution system which allows spectral analysis in the wavelength range 200 - 980 nm with a resolution of 0.06 nm. The system has seven spectrometers with a specified wavelength range for each channel. Each spectrometer has a 2048 pixel linear silicon charge coupled detector (CCD) array with an optical resolution of 0.065 nm. These spectrometers are synchronized to acquire data using the OOILIBS software and store it in a PC. The laser fitted in this system is a 50 mJ maximum energy CFR Nd:YAG from Big Sky Laser Technologies emitting laser pulses of 10 ns width and 10 Hz fixed pulse frequency. Analysis samples are placed on a manually controlled X-Y stage and a laser beam focused on them to produce plasma.
Trace biometal analysis using artificial neural networks (ANN) calibration models
Multivariate ANN calibration models that had been developed and successfully tested for accuracy using a standard reference material (SRM)-1598a (inorganic constituents in animal serum) [52] were applied to analyze the biometals in the plasmodium falciparum infected blood samples and the controls. SRM-1598 is a serum sample derived from a mixture of serum from healthy bovine and porcine animals and is used to evaluate the accuracy of analytical methods for selected elements in biological fluids similar to blood serum and plasma.
ANN is one of the computational ways of mapping non-linear input data to a target space. ANN is capable of performing different function such as curve fitting, pattern recognition as well as clustering. The most common of the network architectures is the multilayer feed-forward system in which the input data proceeds forward only (to the hidden layer and then to the output layer) and never makes loops as opposed to other architectures like the recurrent neural network system. The power of the network depends on the transfer function and the learning rule [42]. Prior to analysis the spectral data was pre-processed in order to eliminate data that was irrelevant to the study. The data was transformed in order to make the distribution of given variables and samples more suitable for a powerful analysis and for a more relevant analysis. Smoothing of the data was done by the moving average technique where the segment size was set to three. The degree of smoothing is usually determined by the width of the smoothing window i.e. the number of data points averaged. The preprocessed spectral databases were compressed into a smaller matrix for each element by obtaining data at different spectral regions for each element at emission wavelengths where LIBS has a spectral response. The output data (targets) were developed with the same number of columns as the input data. The datasets were imported to the neural network platform in MATLAB for analysis. The network was trained using the Levenberg-Marquardt training algorithm which randomly divides the input and output data set into three categories, that is, 60% training, 20% validation and 20% test set.
Table 1 shows the analytical accuracy of the developed ANN multivariate calibration models on a SRM-1598a. The obtained results demonstrate the good accuracy of the ANN calibration strategies that were developed. The Shapiro-Wilks normality test for the ANN model-determined concentrations in the blood indicated that the spectral data used for analysis was normally distributed; that is, Fe-0.8773, Cu-0.8412, Zn-0.8571 and Mg-0.9048. Therefore multivariate exploratory modeling for malaria diagnostics utilizing the determined biometal concentrations may be trusted.
Table 1 Comparison of Cu, Fe, Zn, Mg by ANN and SRM-1598a concentration (Source: [52])
|
Cu
|
Fe
|
Zn
|
Mg
|
NIST/SRM-1598a
[Actual value – (ppm)]
|
1.58±0.09
|
1.68±0.06
|
0.88±0.02
|
Not given
|
ANN model (ppm)
|
1.14±0.74
|
1.14±0.32
|
0.88±0.56
|
17.90±3.63
|
Spectral data modeling by principal components analysis (PCA)
PCA is a powerful tool for exploratory analysis as it performs a projection of the original data which allows for the visualization of the natural clustering of the data, evaluation of class similarities as well as reasons behind the observed classes or patterns. The technique is based on the evaluation of the total variance within data such that the greatest variance by any projection of the data lies on the first coordinate, the first principal component, the second greatest variance on the second coordinate, and so on.
The original data matrix is denoted as X, with n rows, termed ‘objects’, which correspond to the samples, and p columns, termed ‘variables’, which comprise the measurements made on the objects. PCA will provide an approximation of X in terms of the product of two small matrices T and Pas in Eqn. 1 which captures the essential data patterns of X such that
X=T.P+ E………………………………………………………………………………... (1)
where T represents the scores matrix, calculated as (n x A), and P, the loading matrix, obtained as (A x p). A is the intrinsic dimension; that is, the number of principal components necessary to describe all the information in the data set. The scores matrix expresses the relation among the samples and shows the sample coordinates in the new system of axes. The loading matrix P shows the relations among the variables where in this case are the intensities at different wavelengths. E represents a matrix of residuals. This procedure applied to spectra helps to visualize and extract information from the data and can also be applied to show clustering of similar groups [55]. The PCA modeling used both entire spectral region and spectral features.