https://orcid.org/0000-0002-4518-2608
Figures
Abstract
Arbovirus and malaria infections affect more than half of the world’s population causing major financial and physical burden. Current diagnostic tools such as microscopy, molecular and serological techniques are technically demanding, costly, or time consuming. Near-infrared spectroscopy has recently been demonstrated as a potential diagnostic tool for malaria and Dengue virus and as a screening tool for disease vectors. However, pathogen specific absorption peaks that allow detection of these infections are yet to be described. In this study, we identified unique visible and near-infrared peaks from existing laboratory strains of four major arboviruses including Barmah Forest virus, Dengue virus, Ross River virus, Sindbis virus and Plasmodium falciparum. Secondly, to determine the diagnostic ability of these peaks, we developed machine learning algorithms using artificial neural networks to differentiate arboviruses from media in which they were grown. Signature peaks for BFV were identified within the visible region at 410, 430, 562 and 588 nm and the near-infrared region at, 946, 958, 1130, 1154 and 1780 nm. DENV related peaks were seen at 410nm within the visible region and 1130 nm within the near-infrared region. Signature peaks for Ross River virus were observed within the visible region at 410 and 430 nm and within the near-infrared region at 1130 and 1780 nm, while Sindbis virus had a prominent peak at 410 nm within the visible region. Peaks at 514, 528, 547, 561, 582, and 595 nm and peaks at 1388, 1432, 1681, 1700, 1721, 1882, 1905, 2245, 2278, 2300 nm were unique for P. falciparum. Near-infrared spectroscopy predictive sensitivity defined as the ability to predict an arbovirus as an infection was 90% (n=20) for Barmah Forest virus, 100% (n=10) for Ross River virus and 97.5% (n=40) for Dengue virus, while infection specificity defined as the ability to predict media as not-infected was 100% (n=10). Our findings indicate that spectral signatures obtained by near-infrared spectroscopy are potential biomarkers for diagnosis of arboviruses and malaria.
Citation: Goh B, Soares Magalhães RJ, Ciocchetta S, Liu W, Sikulu-Lord MT (2025) Identification of visible and near-infrared signature peaks for arboviruses and Plasmodium falciparum. PLoS ONE 20(4): e0321362. https://doi.org/10.1371/journal.pone.0321362
Editor: Raquel Inocencio da Luz,, Institute of Tropical Medicine: Instituut voor Tropische Geneeskunde, BELGIUM
Received: September 23, 2024; Accepted: March 5, 2025; Published: April 17, 2025
Copyright: © 2025 Goh et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are within the paper and its Supporting Information files.
Funding: This research was funded by NHMRC (National Health and Medical Research Council) (APP1159384) awarded to M.T.S.-L., Advance Queensland Industry Research Fellowship (AQIRF2018019) awarded to M.T.S.-L. and the University of Queensland earmarked Ph.D. scholarship awarded to B.G. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Arboviruses persist in nature through a life cycle involving a vertebrate host, an organism that carries the virus and an infected arthropod, usually mosquitos or ticks [1]. Vector borne diseases have been on the rise due to increased geographical distribution and abundance of arthropod vectors mainly because of concurring factors such as climate change, migration and urbanization [2]. For instance, infection due to Dengue virus (DENV) is now considered the most common vector-borne infection globally as it affects more than half of the world’s population [3]. In 2023, over 6.5 million cases of DENV were reported, resulting in around 6,800 fatalities. The highest incidence was observed in areas of South America and Southeast Asia [4]. Malaria is a mosquito-borne disease caused by the Plasmodium parasite which is transmitted to humans through bites of infected female Anopheles mosquitoes. In 2023, an estimated 263 million malaria cases and 597,000 malaria related deaths were reported by the World Health Organisation highlighting malaria as a major public health concern [5]. Traditionally, malaria is diagnosed using microscopy and Giemsa-stained blood smears [6]. However, with a limit of detection of >50 parasites/μL of blood, it requires a well-trained microscopist [7]. Rapid diagnostic tests are also common for malaria diagnosis because they are very easy to use and do not require qualified personnel, but their sensitivity and specificity is low in detecting low parasitaemia [8,9].
Current diagnosis of arboviruses include molecular methods such as reverse transcriptase-polymerase chain reaction (RT-PCR) [10] and serological techniques such as Enzyme-linked immunosorbent assay (ELISA) [11,12]. Molecular methods are the most accurate and sensitive. For example, RT-PCR for DENV has detection limits that vary between 10–100 copies/reaction depending on the DENV serotype being tested. ELISA assay for detection of DENV-2 specific antibodies has been reported to be 90% (n=20) specific and 100% sensitive (n=20) in 40 human serum samples [13]. Despite high sensitivity, these methods are restricted to the laboratory settings, can be time consuming and costly for programmatic diagnosis and surveillance purposes. Molecular diagnosis techniques for malaria such as Polymerase Chain Reaction (PCR), quantitative-PCR, nested PCR and ELISA have been developed [14–16] but they are time consuming, costly and require trained personnel. Rapid diagnosis tests for malaria are based off the detection for HRP2 antigens as recommend by the World Health organisation [17]. In a recent study, rapid diagnosis test (CareStart™ Malaria HRP2 AccessBio kit) and microscopy both failed to detect more than 40% of infections identified by varATS qPCR [18]; indicating the necessity of novel tools to ensure accurate and prompt diagnosis of clinical malaria cases.
Near-infrared spectroscopy (NIRS) is a potential novel diagnostic/surveillance tool for arboviruses. It involves the interaction of near-infrared light with biological samples to produce a reflectance spectrum [19]. Based on chemical and structural differences between biological samples, unique spectra are produced. The spectra reflect the amount and type of biochemical composition of the sample and can be used to typify those samples. NIR spectra can provide insights into functional groups of samples, such as C–H, O–H, and N–H. Any molecule that contains hydrogen will exhibit a detectable NIR spectrum, making a wide variety of biological substances appropriate for NIR analysis [20]. NIRS can therefore be used as a biomarker for biological samples. Moreover, NIRS is rapid, non-invasive, and does not require skilled personnel to operate, enhancing its effectiveness in the field and in areas lacking scientific equipment.
To date, four studies have shown the ability of NIR technique to detect DENV, Chikungunya, Wolbachia and Zika in Ae. aegypti mosquitoes with accuracies above 90% [21,22]. However, only one study has demonstrated absorbance frequencies in the visible region for arboviruses. Firdous and colleagues reported that NIR peaks at 533 and 580 nm are indicative of the presence of the DNA mixture of DENV2 and DENV3 in human blood samples [23]. The application of NIRS for the detection of the plasmodium parasite has been reported in several studies [24–26]. These studies identified prominent peaks at 650 nm (mice whole blood) [24], 930–1660 nm (human skin) [25] and 1503–2306 nm (human whole blood) [26]. In this study, we identified NIR biomarkers for Barmah Forest virus (BFV), DENV, Ross River virus (RRV), Sindbis virus (SINV) and purified P. falciparum. These signature peaks provide a valuable foundation for diagnosing arboviruses and malaria using NIRS, serving as a basis for future analysis of human or vector samples.
Materials and methods
Dengue virus culture
DENV prototype strains DENV-1 Hawaii (1944), DENV-2 NGC (1944), DENV-3 H-87 (1956) and DENV-4 H241 (1956) were used in this study. DENV was propagated in C6/36 Aedes albopictus cells, maintained at 28°C in RPMI and supplemented with 10% FBS and 1% PSG. Following three passages in C6/36 cells, virus stocks were concentrated using Ultracel-100k filters (Amicon, Tullagreen, Cork Ireland) [27] and frozen at -80°C until further use.
Barmah Forest virus and Ross River virus culture
BFV QML and BFV WEN 1631 and RRV QML1 strain (GenBank No. GQ433354) were used in this study. The virus strains were passaged three times in Vero cells, maintained at 37°C in RPMI and supplemented with 10% FBS and 1% PSG. Following three passages in Vero cells, virus stocks were concentrated using Ultracel-100k filters (Amicon, Tullagreen, Cork Ireland) [27] and frozen at -80°C until further use. One vial of the viral stocks was thawed to determine virus titre using 50% tissue culture infectious dose (CCID50/ml) on Vero cells as described [28]. Briefly, virus stocks were 10-fold serially diluted and 100 µL of diluted virus was inoculated onto monolayers of Vero cells grown in 96 well plates in cell culture media and maintained at 37°C, 5% CO2. Ninety-six hours later, cells were fixed with 3.7% formaldehyde, stained with 1% crystal violet for 1 hour, washed in tap water and dried. The cell culture infectious dose 50% was determined from titration endpoints as described elsewhere [29] and expressed as the Vero cell CCID50/mL.
Sindbis virus culture
SINV strain (SINV 18953) was propagated in C6/36 Ae. albopictus cells, maintained at 28°C in RPMI and supplemented with 10% FBS and 1% PSG. Following three passages in C6/36 cells, virus stocks were concentrated using Ultracel-100k filters (Amicon, Tullagreen, Cork Ireland) and frozen at -80°C until further use. One vial of the viral stocks was thawed to determine virus titre by 50% tissue culture infectious dose (CCID50/ml) on Vero cells as described by Sudeep et al [30]. Briefly, virus stocks were serially diluted 10-fold and 100 µL of diluted virus was inoculated onto monolayers of Vero cells grown in 96 well plates in RPMI 1640 supplemented with L-glutamine, 5% FBS, 1% PSG and maintained at 30°C, 5% CO2. After 4 days of incubation, cells were fixed with 3.7% formaldehyde, stained with 1% crystal violet for 1 hour, washed in tap water, dried, and counted. The CCID50/ml were calculated according to published Reed-Muench method [31].
Virus free media controls
Virus free cell culture media used in this study consisted of sterilised Roswell Park Memorial Institute Medium (RPMI), 1640 Medium (Sigma Life Sciences, USA) with 10% heat-inactivated fetal bovine serum (FBS) (Thermo Fisher Scientific, USA) and 1% Penicillin-Streptomycin Glutamine solution (PSG) (Thermo Fisher Scientific, USA).
Arboviruses viral titer determination for NIRS spectral collection
All arboviruses used were passaged in 3 separate batches. Samples from each batch were used as a single biological replicate (Table 1). Virus stocks were titrated using a modification of the Enzyme-linked immunosorbent assay procedure of Broom et al. [32]. Briefly, virus stocks and samples were serially diluted 10-fold and inoculated onto monolayers of C6/36 cells grown in C6/36 cell culture media which consisted of RPMI, 1640 Medium (Sigma Life Sciences, USA) with 5% heat-inactivated FBS (Thermo Fisher Scientific, USA) and 1% PSG (Thermo Fisher Scientific) and maintained at 30°C, 5% CO2. After 7 days of incubation, cells were fixed in acetone: methanol (1:1) for 1 hour at 4°C. Plates were air-dried and antigen was detected using a cocktail of anti-flavivirus monoclonal antibody hybridoma supernatants; 4G2 [33] 6B-6C1:3H5 [34], at a ratio of 1:1:1, followed by horseradish peroxidase (HRP-) conjugated goat anti-mouse polyclonal antibody (DAKO, Carpinteria, CA, USA) (1:2000 in PBS-Tween 20). Antibodies bound to the cell mono-layers were detected by the addition of 3,3’,5,5’-tetramethylbenzidine liquid substrate system for membranes (Sigma-Aldrich). The CCID50 was determined from titration endpoints as described elsewhere [29] and expressed as C6/36 CCID50/mL. This experiment was repeated three times at three separate time points to create three independent biological replicates (Table 1).
Ethics
Five biological replicates of serum samples each consisting of 150 mL of pooled human serum samples were obtained from Australian Red Cross Lifeblood using human ethics protocol approved by The University of Queensland (Ethics approval number 2020001077, from 10 July 2020 to 31 August 2023). Participants were recruited beginning on 8 June 2023 to 22 June 2023. Following collection from donors, all samples were routinely tested for Hepatitis B and C, HTLV I/II, Syphilis HIV 1/2, and ABO/Rh antigens. Human serum was stored at -25°C for 3–4 days prior to running the described experiments to preserve proteomic profile integrity and was thawed fully at room temperature before use. All samples supplied by the Australian Red Cross Lifeblood were fully anonymized before handover.
P. falciparum cell culture and Plasmodium free media
Base media prepared for P. falciparum Welch was ATCC medium 2196 - Malaria medium (American Type Culture Collection, USA) which consist of a sterilized mixture of RPMI-1640 (Sigma R-0883), HEPES buffer (1 M), Gentamicin (50 mg/ml), L-glutamine (100 mM), Hypoxanthine (100 mM), Glucose (20%) and NaOH (1 N). The mixture was sterilised by filtering through 0.22 μM Millex® filter (Millipore, USA). Complete medium was made by adding the heat-inactivated (at 56°C for 1 hour) human serum to 10% (vol/vol) to the base medium and was used to culture parasites as previously described [35]. P. falciparum Welch strain FCR-3/FMG (American Type Culture Collection, USA) had an initial concentration of 255 parasites/mL. The base media alone was used as the control.
Spectra collection and analysis
LabSpec 4 near-infrared spectrometer (ASD Malvern Panalytical, Malvern, United Kingdom) was used to scan all samples. Details of the spectrometer used is published elsewhere [36]. RS3 software (Malvern ASD Panalytical) was used for NIRS spectra collection. Baseline calibration and optimization were done at the beginning of each experiment and every 30 minutes by scanning an empty space on the glass slide placed on a white Spectralon plate. Five µL of each arbovirus, P. falciparum, and respective media were aliquoted onto glass microscope slides to obtain a sample. A total of 10 technical replicates were scanned for each biological replicate of arbovirus, P. falciparium and media. Samples were scanned at approximately 2 mm from the light source by pointing the probe down to the centre of the sample for approximately 3–5 seconds. Full visible and NIR spectra was collected in the 350–2500 nm range of which 350–750 nm and 751–2500 nm belong to the visible light and NIR light regions, respectively.
Artificial neural network predictive analysis
Reflectance spectra were converted to absorbance using the formular Log . All spectral signatures were converted from txt to csv in ViewSpecPro software (Analytical Spectral Devices Inc, Boulder, CO, USA). To identify arbovirus and Plasmodium peaks of importance, raw spectra was converted into 2nd derivative using the Savitzky–Golay [37] with 2nd order smoothing by combining 10 neighbouring data points. 2nd derivative spectra graphs were visualised in GraphPad Prism 9 (GraphPad Software, Inc, California, USA). In the 2nd derivative graph, the difference between pathogen positive and negative samples is shown and unique absorption peaks for pathogens are identified. Model screening and data analysis were conducted in JMP Pro 16 software on raw data (SAS Institute Inc., Cary, NC, USA). Spectra of DENV1, DENV2, DENV3 and DENV4 were combined into a single identifier referred to as DENV. Likewise, spectra of BFV QML strain and BFV WEN 1631 strain were combined into a single identifier referred to as BFV. This is because no differences were observed between strains of these arboviruses (Table S1). SINV data was analysed separately due to spectral outliers that misclassified other arboviruses. Unsupervised learning using principal component analysis and discriminant analysis (PCA-DA) were conducted on all raw spectra as an initial differentiation step for visualisation.
The spectral data was first split into two groups: model training/validation (consisting of 16 biological replicates) and an independent test set (consisting of 8 biological replicates). The training (T), validation (V) and test sets (t) were separate biological replicates grown and analysed at different time points. Appropriate machine learning algorithm for each raw spectral data underwent model screening where the following model types were screened for preliminary accuracy: Bootstrap Forest [38], Naïve Bayes [39], Artificial Neural Network (ANN) [40] and Support vector machines [41] (Table S2). ANN produced the most accurate preliminary results and was therefore selected for further analysis. Spectral signatures from 410 to 2140 nm were used. This region was exclusive of spectral noise between 350–409 and 2141–2500 nm and NIR transition filters at 995–1100 and 1795–1810 nm. Spectra were used as model predictors whereas infection status (positive or negative) was used as the response factor. ANN model was developed using random K-Fold cross-validation (n=5 samples). The Neural Networks consisted of one layer with three TanH activation nodes boosted at a learning rate of 0.1 iteratively for 100 tours. Models were built to differentiate arboviruses from media. A summary of sample distribution between training, validation and independent test set is shown in Fig 1.
The flow of information from data collection to analysis including the number of samples used for training, validation and test sets. Training and validation sets were split based on biological replicates. ‘T’ represents training set, ‘V’ represents validation set and t represents the test set.
Results
2nd derivative spectra for arboviruses
Signature peaks for DENV are seen at 410 nm within the visible region and at 1130 nm within the NIR region. A DENV peak is observed to have lower absorbance than media at 410 nm but a higher absorbance value than media at 1130 nm (Fig 2A).
Peaks of importance are shown with black arrows.
A total of 10 prominent peaks were identified for BFV. These prominent peaks are within the visible region at 410, 430, 562 and 588 nm and within the NIR region at 946, 958, 1130, 1154, 1287–1331 and 1780 nm. Absorbance peaks within the visible region at 430 and 588 nm and within the NIR region at 946, 1130 nm have higher absorbance values than media while peaks at 410 and 562 nm in the visible region and 958, 1154 and 1780 nm in the NIR region have lower absorbance values than media (Fig 2B).
Prominent peaks for RRV were identified at 410 and 430 within the visible region and at 1130, 1154, 1447, 1464 and 1780 nm within the NIR region. Overall, RRV was observed to have 7 prominent peaks (Fig 2C).
The 2nd derivative NIR spectra of SINV 18953 showed prominent peaks at 410, 1447 and 1463 nm. SINV had the lowest number of specific prominent peaks (3 peaks) compared to the other arboviruses (Fig 2D).
Summary of DENV, BFV, SINV and RRV signature NIR peaks
Four pathogen related wavelengths fell within the visible light region. Two of which (410 and 430 nm) are within the blue visible region and the other two (562, 588 nm) are seen within the green visible light spectrum. BFV prominent peaks at 946 and 958 nm were observed in the 3rd overtone region. Three pathogen related prominent peaks (1130, 1154 and 1780 nm) were seen in the 2nd overtone region. Peaks at 410, 430, 1130 and 1780 nm were observed in BFV, DENV, RRV and SINV. Only one signature peak was identified for SINV which fell within the visible light region, the single signature peak was also identified in all the other arboviruses. Based off this result, we excluded SINV in the analysis. A summary of unique peaks identified for all arboviruses relative to published literature is shown in Table 2.
Differentiation of BFV, DENV, RRV, SINV and Vero media using unsupervised machine learning
For initial differentiation of BFV, DENV, RRV and Vero media, we attempted to use PCA-DA using the 350–2500 nm region. PCA-DA differentiated arboviruses and media with a misclassification rate of 15.1% and R square value of 0.676 (n= 430) (Fig 3). The canonical plot grouped DENV and RRV more closely with almost half the data points overlapping. DENV had slight overlaps with Vero media. BFV was distinct from all other tested categories.
Crosses in the plot represent the average while larger circles represent the standard deviation of data points.
Differentiation of BFV, DENV, RRV, SINV and Vero media using supervised machine learning
To identify if supervised machine learning could differentiate arboviruses from Vero media samples using raw spectra, a model using ANN was applied on the raw spectra between 410–2140 nm (Table 3). Overall, the ANN model differentiated BFV, DENV, RRV and Vero media from each other with an R square value of 1 for both training (n=128) and validation (n=32) set (Table 3). The independent test set consisting of 80 NIR spectra was predicted using the training model. Overall, positive predictive rate defined as the proportion of samples predicted as positive out of those that were truly positive was 100% indicating all infected samples were predicted as infected. The negative predictive rate defined as the proportion of samples predicted as negative out of all those that were truly negative was 76.9% (n=13) meaning some negative samples were predicted as postive. Specificity defined as the proportion of samples predicted as negative out of all negative samples was 100% (n=10) meaning all media samples were predicted as not infected. A summary of technical replicates for the training, validation and test set is shown in Table 3. And a summary of positive prediction rate, negative prediction rate, sensitivity, and specificity of BFV, RRV and DENV for the test set is shown in Table 4.
Raw spectra for P. falciparum
Unique NIR peaks of P. falciparum were observed in the 350–650nm visible region and 1450, 1960 nm NIR region (Fig 4A). Relative to media, a unique peak for P. falciparum was observed at 440 and 543 nm. P. falciparum also showed a higher absorbance than media at both 1450 and 1960 nm. To further investigate the spectra, we plotted the 2nd derivative of the average spectra of P. falciparum. Peaks related to pure P. falciparum were seen at 514, 528, 547, 561, 582 and 595 within the visible region 1388, 1432, 1681, 1700, 1721, 1882, 1905, 2245, 2278 and 2300 nm within the NIR region (Fig 4B). Some of the wavelengths observed to be unique to P. falciparum were found within the visible light region at 514, 528, 547, 561, 582 and 595 nm. The highest peaks for P. falciparum in this region were at 582 and 595 nm (Fig 4B). Comparatively, no peaks were observed for media used to grow P. falciparum in the visible region.
Peaks of importance are indicated with black arrow.
Summary of P. falciparum signature peaks
Six peaks observed for P. falciparum belong to the visible light region, two within the NIR 2nd overtone region, five in the 1st overtone region and 3 in the combination band region. The majority of the peaks identified belong to the visible light region and the 1st overtone region (Table 5).
Discussion
The aim of this study was to identify visible and NIR peaks that are unique to DENV, RRV, BFV SINV and P. falciparum which could serve as potential diagnostic biomarkers for these pathogens. A total of 9 peaks of interest were identified for these arboviruses (Table 2). Distinct signature peaks for BFV were seen at 562, 588, 946, 958 and 1154 nm. Peaks at 562 and 588 nm are within the visible region and could be useful for identification of BFV. Peaks at 946, 958 and 1154 nm represent lipid molecular structures. The lipids could be due to the presences of a bi-lipid membrane anchored on the surface of BFV by proteins E1 and E2 [49].
Besides BFV, the rest of the peaks observed were all shared among the three arboviruses and a peak at 410 nm was observed in all arboviruses. This peak at 410 nm is commonly used for assays that require fluorescent excitation such as ELISA and microscopy [50–53]. DENV and RRV have been identified using this peak previously using a spectrofluorometer [54] and ELISA [55], respectively. A peak at 430 nm was observed for BFV strains and RRV but not DENV and SINV indicating the likelihood that visible light could possibly be used to distinguish between these arboviruses. Peaks at 1130 and 1780 nm were present in BFV strains and RRV. The peak at 1130 nm represents CH3 functional group whereas the peak at 1780 nm represents C=O functional group. Both wavelengths indicate the presence of lipid biomolecules [42–47]. For BFV, this peak could be due to the presence of the bi-lipid membrane [49]. RRV uses lipid droplet biogenesis for viral replication [56] and lipid rafts for infection [57] thus these lipids could be the residue from these processes. The peak at 1130 nm was also observed for DENV. Lipids present in DENV could be a by-product from cellular passaging of DENV in C6/36 Ae. albopictus cells which use lipid metabolism for efficient replication [58–60]. In addition, we identified this peak in a previous study that detected DENV1 in human blood plasma [61].
To further evaluate if the NIRS spectra could be used to differentiate between arboviruses and media, machine learning algorithms were run on the visible-NIR spectral signatures collected. Using the ANN model, the sensitivity and specificity of ≥90% and 100%, respectively were achieved for the independent test set when samples were grouped as infected or not infected. Sensitivity for predicting arboviruses into their actual group was 100%, 80% and 70% for RRV, DENV and BFV, respectively. Seven out of 40 DENV samples were predicted as BFV. Similarly, 3 out of 20 BFV samples were predicted as DENV (Table 3). This indicates a slight confusion by the training model in differentiating DENV and BFV which could be due to the shared absorption peaks at 410 and 1130 nm.
A total of 16 peaks were identified for P. falciparum. Six of those peaks belong to the visible region. This is not surprising as the parasite can be detected via light microscopy. Three of the 6 peaks within the visible region (547, 561 and 582 nm), were also identified at 540, 560 and 579 nm in the ring stage of P. falciparum in whole blood as reported by Adegoke and colleagues [62]. Of the 10 remaining wavelengths, 3 in the NIR region (1388, 2245 and 2300 nm) represent C-H bond vibrations [42–47]. C-H bonds are basic chemical building blocks of life and could be responsible for numerous structures within the P. falciparum parasite. Four wavelengths of interest (1681, 1700, 1721, 1905 nm) represent lipids [42–48]. Lipids have been found in P. falciparum and have been shown to play several roles such as toxicity [63], gametocytogenesis [64], and parasite development [65–68]. In addition, peaks at 1388 and 1432 nm are within the same range as those recently identified (1377 and 1431 nm) in malaria infected patients non-invasively using a handheld spectrometer [25]. The remaining 3 wavelengths; 1432, 1882, 2278 nm represent aromatic amine, water and polysaccharides, respectively [42–48].
The peaks for arboviruses and P. falciparium identified in this study will be useful biomarkers for the surveillance/diagnosis of these pathogens in the real world either in humans hosts, animal hosts or mosquito vectors. However, a further assessment in the field using naturally infected blood samples is required to validate the biomarkers identified under this study. With this additional tool in hand, NIRS has the potential to rapidly identify infections to stop an outbreak by facilitating timely isolation and treatment of patients and rapid identification of infected mosquitoes.
Conclusion
We have identified several novel visible and NIR biomarkers for BFV, DENV, RRV, SINV and P. falciparum. To our knowledge, this is the first investigation to report NIR biomarkers for arboviruses. The findings of this study provide insights into the potential future application of these peaks as diagnostic biomarkers for these pathogens. Future work should evaluate the capability of NIRS spectrometers coupled with machine learning to detect these pathogens, using these biomarkers, in human, animal and mosquito species. This would facilitate rapid, non-invasive, and cost-effective diagnosis and surveillance of these pathogens particularly in large-scale setting that require programmatic surveillance to stop outbreaks.
Supporting information
Table S1. A summary of the validation set from ANN testing of different BFV strains, DENV serotypes and SINV.
Overall, a misclassification rate of 0.4778 (n=90) was obtained indicating a low overall accuracy. Numbers in the table represent the number of technical replicates allocated to the prediction count for each arbovirus and media.
https://doi.org/10.1371/journal.pone.0321362.s001
(DOCX)
Table S2. A summary of statistical models tested with training data.
Statistical models are ranked in order of most accurate to the least. Accuracy is determined by the model having values: lowest for misclassification rate, highest for entropy RSquare, highest for area under the curve, lowest for root average square error, and highest for generalized RSquare.
https://doi.org/10.1371/journal.pone.0321362.s002
(DOCX)
Fig S1. The average visible-NIR raw spectra of DENV media (A), BFV/media (B), RRV/media (C) and SINV/media (D).
Prominent NIRS peaks for DENV can be observed in the 1450 nm and 1950 nm regions and water peaks were identified at 1450 and 1950 nm. Generally, the absorbance value of DENV was lower than that of media (A). All spectra of BFV QML and BFV WEN1631 were averaged to identify prominent absorbance peaks for BFV. Generally, BFV was observed to have higher absorbance values than media (B). The absorbance value of RRV within the visible and NIR regions was generally higher than media except for the water peak around 1940 nm (C). SINV 18953 generally absorbed less light than media from the visible through to the NIR region between 350–2500 nm. Absorbance values for water molecules at 1450 and 1950 nm of SINV 18953 was lower than media (D).
https://doi.org/10.1371/journal.pone.0321362.s003
(DOCX)
Fig S2. The 2nd derivative of the averaged visible and NIR spectra of arboviruses and media.
The 2nd derivative of the average visible and NIR spectra for DENV/media (A), BFV/Media (B), RRV/Media (C) and SINV (D) from 350–2500 nm.
https://doi.org/10.1371/journal.pone.0321362.s004
(DOCX)
S1 Supporting Information Data. All data for this study is included in the file titled “Supporting information data”.
https://doi.org/10.1371/journal.pone.0321362.s005
(XLSX)
References
- 1. Braack L, Gouveia de Almeida AP, Cornel AJ, Swanepoel R, de Jager C. Mosquito-borne arboviruses of African origin: review of key viruses and vectors. Parasit Vectors. 2018;11(1):29. pmid:29316963
- 2. Tajudeen YA, Oladunjoye IO, Mustapha MO, Mustapha ST, Ajide-Bamigboye NT. Tackling the global health threat of arboviruses: An appraisal of the three holistic approaches to health. Health Promot Perspect. 2021;11(4):371–81. pmid:35079581
- 3. Bhatt S, Gething PW, Brady OJ, Messina JP, Farlow AW, Moyes CL, et al. The global distribution and burden of dengue. Nature. 2013;496(7446):504–7. pmid:23563266
- 4. Haider N, Hasan MN, Onyango J, Asaduzzaman M. Global landmark: 2023 marks the worst year for dengue cases with millions infected and thousands of deaths reported. IJID Reg. 2024;13:100459. pmid:39497753
- 5. W. H. Organisation. World malaria report 2024. 2024. [Online]. Available from: https://www.who.int/teams/global-malaria-programme/reports/world-malaria-report-2024
- 6. Tangpukdee N, Duangdee C, Wilairatana P, Krudsood S. Malaria diagnosis: a brief review. Korean J Parasitol. 2009;47(2):93–102. pmid:19488414
- 7. Amir A, Cheong F-W, De Silva JR, Lau Y-L. Diagnostic tools in childhood malaria. Parasit Vectors. 2018;11(1):53. pmid:29361963
- 8. Ogunfowokan O, Ogunfowokan BA, Nwajei AI. Sensitivity and specificity of malaria rapid diagnostic test (mRDT CareStatTM) compared with microscopy amongst under five children attending a primary care clinic in southern Nigeria. Afr J Prim Health Care Fam Med. 2020;12(1):e1–8. pmid:32634015
- 9. Shankar H, Singh MP, Phookan S, Singh K, Mishra N. Diagnostic performance of rapid diagnostic test, light microscopy and polymerase chain reaction during mass survey conducted in low and high malaria-endemic areas from two North-Eastern states of India. Parasitol Res. 2021;120(6):2251–61. pmid:33772349
- 10. Deepak S, Kottapalli K, Rakwal R, Oros G, Rangappa K, Iwahashi H, et al. Real-Time PCR: Revolutionizing Detection and Expression Analysis of Genes. Curr Genomics. 2007;8(4):234–51. pmid:18645596
- 11. Yeh J-Y, Lee J-H, Park J-Y, Seo H-J, Moon J-S, Cho I-S, et al. A diagnostic algorithm to serologically differentiate West Nile virus from Japanese encephalitis virus infections and its validation in field surveillance of poultry and horses. Vect Borne Zoonotic Dis. 2012;12(5):372–9. pmid:22217162
- 12. Choi K-S, Ko Y-J, Nah J-J, Kim Y-J, Kang S-Y, Yoon K-J, et al. Monoclonal antibody-based competitive enzyme-linked immunosorbent assay for detecting and quantifying West Nile virus-neutralizing antibodies in horse sera. Clin Vaccine Immunol. 2007;14(2):134–8. pmid:17135450
- 13. Narayan R, Raja S, Kumar S, Sambasivam M, Jagadeesan R, Arunagiri K, et al. A novel indirect ELISA for diagnosis of dengue fever. Indian J Med Res. 2016;144(1):128–33. pmid:27834337
- 14. Hendershot AL, Esayas E, Sutcliffe AC, Irish SR, Gadisa E, Tadesse FG, et al. A comparison of PCR and ELISA methods to detect different stages of Plasmodium vivax in Anopheles arabiensis. Parasit Vectors. 2021;14(1):473. pmid:34526109
- 15. Lazrek Y, Florimond C, Volney B, Discours M, Mosnier E, Houzé S, et al. Molecular detection of human Plasmodium species using a multiplex real time PCR. Sci Rep. 2023;13(1):11388. pmid:37452123
- 16. Wang B, Han S-S, Cho C, Han J-H, Cheng Y, Lee S-K, et al. Comparison of microscopy, nested-PCR, and Real-Time-PCR assays using high-throughput screening of pooled samples for diagnosis of malaria in asymptomatic carriers from areas of endemicity in Myanmar. J Clin Microbiol. 2014;52(6):1838–45. pmid:24648557
- 17. W. H. Organization. Guidance on control materials for antigen detecting malaria RDTs: tools for preparation and validation. [cited 26 Nov 2024]. Available from: https://www.who.int/publications/i/item/WHO-CDS-GMP-2019.08
- 18. Opoku Afriyie S, Addison TK, Gebre Y, Mutala A-H, Antwi KB, Abbas DA, et al. Accuracy of diagnosis among clinical malaria patients: comparing microscopy, RDT and a highly sensitive quantitative PCR looking at the implications for submicroscopic infections. Malar J. 2023;22(1):76. pmid:36870966
- 19. Norris KP. Infra-red spectroscopy and its application to microbiology. J Hyg (Lond). 1959;57(3):326–45. pmid:14427662
- 20. Manley M. Near-infrared spectroscopy and hyperspectral imaging: non-destructive analysis of biological materials. Chem Soc Rev. 2014;43(24):8200–14. pmid:25156745
- 21. Santos LMB, Mutsaers M, Garcia GA, David MR, Pavan MG, Petersen MT, et al. High throughput estimates of Wolbachia, Zika and chikungunya infection in Aedes aegypti by near-infrared spectroscopy to improve arbovirus surveillance. Commun Biol. 2021;4(1):67. pmid:33452445
- 22. Fernandes JN, Dos Santos LMB, Chouin-Carneiro T, Pavan MG, Garcia GA, David MR, et al. Rapid, non-invasive detection of Zika virus in Aedes aegypti mosquitoes by near-infrared spectroscopy. Sci Adv. 2018;4(5):eaat0496. pmid:29806030
- 23. Firdous S, Ahmed M, Rehman A, Nawaz M, Anwar S, Murtaza S. Transmission spectroscopy of dengue viral infection. Laser Phys Lett. 2012;9(4):317–21.
- 24. Theint HT, Walsh JE, Wong ST, Voon K, Shitan M. Development of an optical biosensor for the detection of Trypanosoma evansi and Plasmodium berghei. Spectrochim Acta A Mol Biomol Spectrosc. 2019;218:348–58. pmid:31026712
- 25. Garcia GA, Kariyawasam TN, Lord AR, da Costa CF, Chaves LB, Lima-Junior J da C, et al. Malaria absorption peaks acquired through the skin of patients with infrared light can detect patients with varying parasitemia. PNAS Nexus. 2022;1(5):pgac272. pmid:36712329
- 26. Adegoke JA, Kochan K, Heraud P, Wood BR. A Near-Infrared “Matchbox Size” Spectrometer to Detect and Quantify Malaria Parasitemia. Anal Chem. 2021;93(13):5451–8. pmid:33759513
- 27. Hugo LE, Stassen L, La J, Gosden E, Ekwudu O, Winterford C, et al. Vector competence of Australian Aedes aegypti and Aedes albopictus for an epidemic strain of Zika virus. PLoS Negl Trop Dis. 2019;13(4):e0007281. pmid:30946747
- 28. Pushpakom S, Iorio F, Eyers PA, Escott KJ, Hopper S, Wells A, et al. Drug repurposing: progress, challenges and recommendations. Nat Rev Drug Discov. 2019;18(1):41–58. pmid:30310233
- 29. Charretier C, Saulnier A, Benair L, Armanet C, Bassard I, Daulon S, et al. Robust real-time cell analysis method for determining viral infectious titers during development of a viral vaccine production process. J Virol Methods. 2018;252:57–64. pmid:29154792
- 30. Sudeep AB, Vyas PB, Parashar D, Shil P. Differential susceptibility & replication potential of Vero E6, BHK-21, RD, A-549, C6/36 cells & Aedes aegypti mosquitoes to three strains of chikungunya virus. Indian J Med Res. 2019;149(6):771–7. pmid:31496530
- 31. Reed LJ, Muench H. A simple method of estimating fifty per cent endpoints12. American Journal of Epidemiology. 1938;27(3):493–7.
- 32. Broom AK, Hall RA, Johansen CA, Oliveira N, Howard MA, Lindsay MD, et al. Identification of Australian arboviruses in inoculated cell cultures using monoclonal antibodies in ELISA. Pathology. 1998;30(3):286–8. pmid:9770194
- 33. Roehrig JT, Mathews JH. The neutralization site on the E2 glycoprotein of Venezuelan equine encephalomyelitis (TC-83) virus is composed of multiple conformationally stable epitopes. Virology. 1985;142(2):347–56. pmid:2414905
- 34. Henchal EA, Gentry MK, McCown JM, Brandt WE. Dengue virus-specific and flavivirus group determinants identified with monoclonal antibodies by indirect immunofluorescence. Am J Trop Med Hyg. 1982;31(4):830–6. pmid:6285749
- 35. van Schalkwyk DA, Burrow R, Henriques G, Gadalla NB, Beshir KB, Hasford C, et al. Culture-adapted Plasmodium falciparum isolates from UK travellers: in vitro drug sensitivity, clonality and drug resistance markers. Malar J. 2013;12:320. pmid:24028570
- 36. Mayagaya VS, Michel K, Benedict MQ, Killeen GF, Wirtz RA, Ferguson HM, et al. Non-destructive determination of age and species of Anopheles gambiae s.l. using near-infrared spectroscopy. Am J Trop Med Hyg. 2009;81(4):622–30. pmid:19815877
- 37. Savitzky Abraham, Golay MJE. Smoothing and Differentiation of Data by Simplified Least Squares Procedures. Anal Chem. 1964;36(8):1627–39.
- 38. Tin Kam Ho. Random decision forests. Proceedings of 3rd International Conference on Document Analysis and Recognition. 1995;1:278–82.
- 39.
Webb GI. Naïve Bayes. In: Sammut C, Webb GI, editors. Encyclopedia of Machine Learning. Boston (MA): Springer US; 2010, p. 713–4.
- 40. McCulloch WS, Pitts W. A logical calculus of the ideas immanent in nervous activity. Bulletin of Mathematical Biophysics. 1943;5(4):115–33.
- 41. Hearst MA, Dumais ST, Osuna E, Platt J, Scholkopf B. Support vector machines. IEEE Intell Syst Their Appl. 1998;13(4):18–28.
- 42.
Badr Eldin A. Near Infra Red Spectroscopy. 2011.
- 43. El-Bacha T, Midlej V, Pereira da Silva AP, Silva da Costa L, Benchimol M, Galina A, et al. Mitochondrial and bioenergetic dysfunction in human hepatic cells infected with dengue 2 virus. Biochim Biophys Acta. 2007;1772(10):1158–66. pmid:17964123
- 44.
Workman JJ, Weyer L. Practical Guide to Interpretive Near-Infrared Spectroscopy. CRC Press; 2008, p. 277–300.
- 45. Tigabu M, Fjellström J, Odén PC, Teketay D. Germination of Juniperus procera seeds in response to stratification and smoke treatments, and detection of insect-damaged seeds with VIS + NIR spectroscopy. New Forests. 2007;33(2):155–169.
- 46. Hourant P, Baeten V, Morales MT, Meurens M, Aparicio R. Oil and Fat Classification by Selected Bands of Near-Infrared Spectroscopy. Appl Spectrosc. 2000;54(8):1168–74.
- 47. Butkutė B, Slepetiene A. Near-infrared reflectance spectroscopy as a fast method for simultaneous prediction of several soil quality components. Chemija. 2004;15(1):12–20.
- 48. Salzer R, Workman J. Practical Guide to Interpretive Near‐Infrared Spectroscopy. Von Jerry Workman, Jr. und Lois Weyer. Angewandte Chemie. 2008;120(25):4706–7.
- 49. Kostyuchenko VA, Jakana J, Liu X, Haddow AD, Aung M, Weaver SC, et al. The structure of barmah forest virus as revealed by cryo-electron microscopy at a 6-angstrom resolution has detailed transmembrane protein architecture and interactions. J Virol. 2011;85(18):9327–33. pmid:21752915
- 50. Denis H, Marullo S, Hoebeke J, Strosberg AD. Enzyme-linked immunosorbent assay for amitriptyline and other antidepressants using a monoclonal antibody. Clin Chim Acta. 1986;159(3):257–67. pmid:3769214
- 51. Arya P, Ponmariappan S, Singh L, Kumar O. Development of ELISA based detection system for lethal toxin of Clostridium sordellii. Indian J Med Res. 2013;137(6):1180–7. pmid:23852299
- 52. Wei L, Roberts DW, Sanai N, Liu JTC. Visualization technologies for 5-ALA-based fluorescence-guided surgeries. J Neurooncol. 2019;141(3):495–505. pmid:30554344
- 53. Polishchuk RS, Polishchuk EV, Marra P, Alberti S, Buccione R, Luini A, et al. Correlative light-electron microscopy reveals the tubular-saccular ultrastructure of carriers operating between Golgi apparatus and plasma membrane. J Cell Biol. 2000;148(1):45–58. pmid:10629217
- 54. Bujalowski PJ, Bujalowski W, Choi KH. Interactions between the Dengue Virus Polymerase NS5 and Stem-Loop A. J Virol. 2017;91(11):e00047-17. pmid:28356528
- 55. Way SJR, Lidbury BA, Banyer JL. Persistent Ross River virus infection of murine macrophages: an in vitro model for the study of viral relapse and immune modulation during long-term infection. Virology. 2002;301(2):281–92. pmid:12359430
- 56. Criglar JM, Estes MK, Crawford SE. Rotavirus-Induced Lipid Droplet Biogenesis Is Critical for Virus Replication. Front Physiol. 2022;13:836870. pmid:35492603
- 57. Dou X, Li Y, Han J, Zarlenga DS, Zhu W, Ren X, et al. Cholesterol of lipid rafts is a key determinant for entry and post-entry control of porcine rotavirus infection. BMC Vet Res. 2018;14(1):45. pmid:29433482
- 58. Heaton NS, Randall G. Dengue virus-induced autophagy regulates lipid metabolism. Cell Host Microbe. 2010;8(5):422–32. pmid:21075353
- 59. Koh C, Islam MN, Ye YH, Chotiwan N, Graham B, Belisle JT, et al. Dengue virus dominates lipid metabolism modulations in Wolbachia-coinfected Aedes aegypti. Commun Biol. 2020;3(1):518. pmid:32948809
- 60. Tongluan N, Ramphan S, Wintachai P, Jaresitthikunchai J, Khongwichit S, Wikan N, et al. Involvement of fatty acid synthase in dengue virus infection. Virol J. 2017;14(1):28. pmid:28193229
- 61. Goh B, Visendi P, Lord AR, Ciocchetta S, Liu W, Sikulu-Lord MT. First Report of the Detection of DENV1 in Human Blood Plasma with Near-Infrared Spectroscopy. Viruses. 2022;14(10):2248. pmid:36298803
- 62. Adegoke JA, De Paoli A, Afara IO, Kochan K, Creek DJ, Heraud P, et al. Ultraviolet/Visible and Near-Infrared Dual Spectroscopic Method for Detection and Quantification of Low-Level Malaria Parasitemia in Whole Blood. Anal Chem. 2021;93(39):13302–10. pmid:34558904
- 63. Debierre-Grockiego F, Schofield L, Azzouz N, Schmidt J, Santos de Macedo C, Ferguson MAJ, et al. Fatty acids from Plasmodium falciparum down-regulate the toxic activity of malaria glycosylphosphatidylinositols. Infect Immun. 2006;74(10):5487–96. pmid:16988223
- 64. Tanaka TQ, Tokuoka SM, Nakatani D, Hamano F, Kawazu S-I, Wellems TE, et al. Polyunsaturated fatty acids promote Plasmodium falciparum gametocytogenesis. Biol Open. 2019;8(7):bio042259. pmid:31221627
- 65. Tran PN, Brown SHJ, Rug M, Ridgway MC, Mitchell TW, Maier AG. Changes in lipid composition during sexual development of the malaria parasite Plasmodium falciparum. Malar J. 2016;15:73. pmid:26852399
- 66. van Schaijk BCL, Kumar TRS, Vos MW, Richman A, van Gemert G-J, Li T, et al. Type II fatty acid biosynthesis is essential for Plasmodium falciparum sporozoite development in the midgut of Anopheles mosquitoes. Eukaryot Cell. 2014;13(5):550–9. pmid:24297444
- 67. Mi-Ichi F, Kita K, Mitamura T. Intraerythrocytic Plasmodium falciparum utilize a broad range of serum-derived fatty acids with limited modification for their growth. Parasitol. 2006;133(Pt 4):399–410. pmid:16780611
- 68. Shears MJ, MacRae JI, Mollard V, Goodman CD, Sturm A, Orchard LM, et al. Characterization of the Plasmodium falciparum and Plasmodium berghei glycerol 3-phosphate acyltransferase involved in FASII fatty acid utilization in the malaria parasite apicoplast. Cell Microbiol. 2017;19(1):e12633. pmid:27324409