Screening of Wolbachia Endosymbiont Infection in Aedes aegypti

Mar 23, 2017 - The prediction errors using partial least squares discriminant analysis (PLS-DA) discrimination models for laboratory studies on indepe...
0 downloads 5 Views 2MB Size
Subscriber access provided by Fudan University

Article

Screening of Wolbachia endosymbiont infection in Aedes aegypti mosquitoes using Attenuated Total Reflection mid-infrared spectroscopy Aazam Khoshmanesh, Dale Christensen, David Perez-Guaita, Iñaki IturbeOrmaetxe, Scott L. O'Neill, Don McNaughton, and Bayden R. Wood Anal. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.analchem.6b04827 • Publication Date (Web): 23 Mar 2017 Downloaded from http://pubs.acs.org on March 27, 2017

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Analytical Chemistry is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

Screening of Wolbachia endosymbiont infection in Aedes aegypti mosquitoes using Attenuated Total Reflection mid-infrared spectroscopy Aazam Khoshmanesh1, Dale Christensen1, David Perez-Guaita1, Inaki Iturbe-Ormaetxe2, Scott L. O’Neill2, Don McNaughton1 and Bayden R. Wood1* 1

Centre for Biospectroscopy, School of Chemistry, Monash University, Clayton, Victoria

3800, Australia. 2

Institute of Vector-Borne Disease, Monash University, Clayton,

Victoria 3800, Australia. *

Correspondence to: [email protected]

Abstract Dengue fever is the most common mosquito transmitted viral infection afflicting humans, estimated to generate around 390 million infections each year in over 100 countries. The introduction of the endosymbiotic bacterium Wolbachia into Aedes aegypti mosquitoes has the potential to greatly reduce the public health burden of the disease. This approach requires extensive PCR (Polymerase Chain Reaction) testing of the Wolbachia-infection status of mosquitoes in areas where Wolbachia-A. aegypti are released. Here we report the first example of small organism mid-infrared spectroscopy where we have applied Attenuated Total Reflection Fourier Transform Infrared (ATR-FTIR) spectroscopy and multivariate modelling methods to determine sex, age and the presence of Wolbachia (wMel strain) in laboratory mosquitoes and sex and age in field mosquitoes. The prediction errors using Partial Least Squares Discriminant Analysis (PLS-DA) discrimination models for laboratory studies on independent test sets ranged from 0 to 3% for age & sex grading, and 3 to 5% for Wolbachia infection diagnosis using dry mosquito abdomens while field study results using an Artificial neural network yielded a 10 % error. The application of FTIR analysis is inexpensive, easy to use, portable, and shows significant potential to replace the reliance on more expensive and laborious PCR assays.

ACS Paragon Plus Environment

Analytical Chemistry

Page 2 of 25

2 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

1. Introduction Dengue is a systemic viral infection, which causes a potentially life-threatening illness in many patients1. It takes the form of four serotypes (DENV1 – DENV4), each of which can cause the full range of symptoms2-4 . Aedes aegypti mosquitoes act as the primary vectors in the transmission of the virus between humans, resulting in an estimated 390 million infections per annum globally5. These infections are mostly limited to the tropical and subtropical regions, where A. aegypti and to a lesser extent Aedes albopictus occurs. Dengue fever is an expanding problem throughout the tropics of the world as current control methods are proving to be largely ineffective. Wolbachia pipientis are small intracellular Gram-negative bacteria found in up to 60% of insect species, where they can induce a variety of reproductive phenotypes and confer protection against a variety of pathogens. Despite their prevalence in nature Wolbachia does not naturally infect A. aegypti (the main dengue vector), however several Wolbachia strains have been artificially introduced6-8 to create stable Wolbachia-infected A. aegypti that contain high Wolbachia densities and have a reduced ability to transmit dengue and other pathogens including9-10 ZIKA. These strains are able to spread through insect populations by means of Cytoplasmic Incompatibility (CI), a form of reproductive manipulation that favours Wolbachia-infected female mosquitoes. The ability to induce CI, together with their pathogen-blocking abilities have put these bacteria at the forefront in the fight against dengue and other pathogens transmitted by mosquitoes11-14. As part of the Eliminate Dengue Program (www.eliminatedengue.com), Wolbachia-infected A. aegypti mosquitoes have been released in several international field sites with the aim of introducing Wolbachia into the wild mosquito population and in turn reducing dengue transmission15. The use of Wolbachia-infected A. aegypti for the control of dengue involves the regular monitoring of local mosquitoes in release areas during and after the release. These

ACS Paragon Plus Environment

Page 3 of 25

Analytical Chemistry

3 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

mosquitoes are collected in traps and assessed for their presence of Wolbachia using molecular methods. There are several different methods of screening mosquitoes for Wolbachia. To date these have been based mainly on standard polymerase chain reaction (PCR) assays16-17 that detect Wolbachia DNA and are very sensitive and reliable. The current best-practice screening method involves quantitative polymerase chain reaction (qPCR) assays15. This process involves the application of DNA primers, which target A. aegypti- and Wolbachia-specific genes, together with fluorescently labelled hydrolysis Taqman probes that allow the simultaneous detection of both species using real time fluorescence. These qPCR assays are able to screen a high number of samples quickly, while remaining robust and accurate18, but they can be expensive due to the cost of real-time PCR machines, primers, probes and PCR reagents, in particular the enzymes used in the reaction, and require specialised training. Alternative methods, such as Near Infrared Spectroscopy (NIRS), which spans the range between 10,000-4,000 cm-1 of the electromagnetic spectrum have been applied to detect Wolbachia infection, species identification, sex and age grading in flies19-22 as well as identification of two strains of Wolbachia in A. aegypti23 and malaria infection in mosquitoes20. However, the interpretation of NIR spectra is very difficult due to the broad nature of the bands resulting from the overlap of many overtone modes and the ubiquitous presence of water absorbance. These challenges make calibration procedures quite laborious and often not transferable from one instrument to the next24.

Due to intrinsic band

broadening the differences between NIR spectra of different compounds are often very subtle, resulting in a lower inherent molecular selectivity compared to the mid-IR. Consequently NIRS is a much less sensitive technique and therefore unsuited to the analysis of low-level components below a few percent25.

ACS Paragon Plus Environment

Analytical Chemistry

Page 4 of 25

4 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Fourier Transform Infrared (FTIR) spectroscopy has the potential to detect functional group vibrations in the mid-infrared region 4000-400 cm-1 of the electromagnetic spectrum providing a snapshot of the metabolic fingerprint of the biological sample under investigation. The changes in this fingerprint can be detected using multivariate models and these changes used as predictive markers for the diagnosis of disease and other conditions. FTIR-based diagnostics have a proven track record in detecting viral infections in cells26-28, however, the technique has never been applied to detecting infections in mosquitoes. Here we show for the first time that ATR-FTIR spectroscopy, in combination with multivariate modelling methods, has the required ease of sample preparation, sensitivity, and ability to become a fast and cheap method to replace costly and time-consuming qPCR for diagnosis of Wolbachia-infected mosquitoes as well as for discrimination of 2 day and 10 day old mosquitoes along with sex. It should be noted that mosquito sex can also be distinguished by eye (using the antennae flagellum anatomy, for example) and the ATR measurements to determine sex in this case were performed to understand the chemical differences between males and females. The system combines a standard bench-top FTIR spectrometer with a diamond crystal ATR accessory (or silicon cell designed for liquid samples) but portable units are also available. The technique utilizes the property of total internal reflection to generate an evanescent wave, which penetrates on the order of 3 µm into a sample, which is enough to penetrate the abdomens of mosquitoes crushed by the loading jig. The whole process of sample deposition and spectral recording takes less than 2 minutes using a single pass ATR crystal enabling approximately 3000 samples to be processed a week (12 hour day) with a single instrument. Our results show real potential that a viable spectroscopic alternative to current qPCR methods can be developed, with efficiencies made in time, cost, waste reduction, and training

ACS Paragon Plus Environment

Page 5 of 25

Analytical Chemistry

5 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

of personnel, and represents a totally new approach to assist in the elimination of dengue fever.

2. Methods Mosquito preparation Dried Mosquito Abdomens: About 400 lab grown mosquitoes including male & female, 2 and 10 days old, infected and uninfected with Wolbachia (wmMel) (50 mosquitos from each group) were fixed in 80 % ethanol and the abdomens dried in an incubator at 37 oC before recording spectra on the ATR. Fresh Mosquito Abdomens: About 200 lab grown whole (male & female) mosquitoes (nonblood fed &16 days old) from two strains taken freshly from the insectary, stored in a freezer at -5 oC for ~20 min before spectral acquisition. The legs were removed and the abdomen placed onto the ATR diamond window and pressure applied using the loading jig. Field mosquitoes: Whole (male & female) mosquitoes were collected in traps deployed in houses for 7 days at different suburbs in Cairns, Queensland, Australia. Mosquitoes were sent to Monash in 70% ethanol.

The mosquitoes were washed in water and dried before

dissecting the abdomen. The head and thorax were used for Polymerise Chain Reaction (PCR) assay. Mosquito extracts: About 100 homogenised mosquito extracts were prepared from infected & uninfected samples in 200 uL methanol. A mosquito abdomen and a glass bead were added per tube and placed in a tissue-lyser for extraction. Samples were homogenised by shaking before analysis.

PCR

1,2

was used as a reference method to determine the infected

mosquitoes.

ACS Paragon Plus Environment

Analytical Chemistry

Page 6 of 25

6 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Equipment & Spectral Data Acquisition (ATR-FTIR measurements): A Bruker model EQUINOX 55-(Bruker Optic, Ettingen, Germany) FTIR spectrometer fitted with a N2-cooled mercury–cadmium–tellurium (MCT) detector and a golden gate diamond ATR accessory (Specac limited, Orpington, Kent, UK) with loading jig was used for spectral acquisition of the mosquito abdomens, while a silicon sample holder (an ATR accessory) was used for liquid mosquito extracts. The Bruker system was controlled using an IBM-compatible PC running OPUS version 6.0 software. For each extract sample spectra from 5 to 10 uL of the mosquito extracts were placed on the silicon cell and air-dried. For the mosquito abdomen trials only separated abdomens were placed on the ATR diamond cell. Spectra were collected with a spectral resolution of 6 cm-1 and 32 co-added interferograms. The sample spectrum was ratioed against a clean diamond background or clean silicon background. For each sample deposit, 3-5 replicate spectra were recorded to assess the reproducibility of each sample spectrum. Data pre-processing & modelling software:

Pre-processing of the spectral data was

performed in OPUS-(Bruker Optic, Ettingen, Germany) and the Unscrambler X (Version 10.0.1, Camo, Norway) software packages. For optimal modelling, both 1st or 2nd second derivatives were calculated using the Savitzky–Golay algorithm with 9 smoothing points and then mean centred. Unscrambler X was used for PCA (principal component analysis), PLS-R (partial least square regression) & PLS-DA (partial least square discriminate analysis) modelling and Matlab R2014a (www.mathworks.com). The Artificial Neural Network (ANN) toolbox (nprtool and nntool) were used for ANN modelling and ROC curves, predictions and pattern recognition. The full cross validation option was used for PCA, and PLS-DA studies. In the ANN-toolbox feed forward backpropagation algorithm was selected and TRAINLM, LEARNGDM and TANZIG were used for training, adaption learning and

ACS Paragon Plus Environment

Page 7 of 25

Analytical Chemistry

7 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

transfer functions.

For Random Forest Feature Selection the RF_MexStandalone-v0.02

package was used in Matlab. Data Analysis steps: To determine whether ATR could determine the mosquito gender, age as well as Wolbachia infection a few steps were trailed using different sample treatments such as mosquito extracts along with fresh and dry mosquito abdomens. In the first step, average and individual mosquito spectra and their derivatives were compared. Statistical spectral plots such as quantile, mean & standard deviation were also investigated (Supplementary Note 1). In addition, RF-variable selection was applied to the spectra of mosquitoes (Supplementary Note 2). In the second step, PCA was performed to examine variation between the groups and identify outliers using different spectral windows for the PCA decomposition. In the third step, calibration models were developed for each study from the derivative spectra for selected spectral ranges using both linear and nonlinear modelling methods such as partial least square discriminate analysis (PLS-DA), artificial neural network (ANN) as well as ANN-pattern recognition (Supplementary Note 3). Finally to assess further the prediction accuracy of the calibration models independent test sets from mosquitoes not involved in the calibration dataset were used as blind samples.

3. Results & Discussion Initial studies focused on determining the ultimate sample preparation for ATR analysis of Wolbachia infection, sex recognition and age of the A. aegypti mosquitoes. A data set containing spectra of male and female, 2 & 10 days old, and Wolbachia-infected and uninfected mosquitoes was developed for mosquito extracts as well as fresh and dry mosquito abdomens. The laboratory-grown mosquitoes used in this study were collected from different generations and cohorts to increase variability in the sampling. Wolbachia-infected

ACS Paragon Plus Environment

Analytical Chemistry

Page 8 of 25

8 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

mosquitoes were not directly treated with the bacteria; rather, they were descendent from a line of mosquitoes (established in 2009) in which the bacteria are stably and maternally inherited. We used Partial Least Squares-Discriminant Analysis (PLS-DA) models for classifying the spectra according to their gender, age and Wolbachia presence. PLS-DA is one of the most common supervised classification methods in chemometrics, being a powerful classification tool, which needs to be externally tested in order to avoid overoptimistic results caused by overfitting31. In this study we ensured a reliable assessment of the prediction errors by splitting the database into calibration and independent test sets. A calibration set was employed for creating several PLS-DA models using different combinations of data preprocessing and spectral regions with cross validation employed for selection of the best of those models. The selected model was then tested using an independent test set. A detailed explanation of the data analysis is available in the Supplementary Note 1. For laboratorygrown mosquitoes the highest sensitivity and specificity (>95%) for Wolbachia detection was achieved using methanol-dried female mosquito abdomens. Nonetheless, the increased variability intrinsic in field collected Wolbachia infected mosquitoes made the classification through PLS-DA models unsuccessful, and an artificial neural network (ANN) approach (Supplementary Note 2) was employed as an alternative, which achieved a high sensitivity (92%) and specificity (84%) comparable to that for laboratory samples. The field mosquito study was performed on fresh abdomens to simulate field conditions. Detailed information on the preparation of mosquito samples, reference method for infection diagnosis, data preprocessing, data analysis steps, and modelling software is presented in the Online Methods. Figure 1 shows the overall concept of mosquito identification with ATR-FTIR, which simply involves placing the mosquito abdomen, lysate or whole mosquito on the crystal and recording the spectrum.

ACS Paragon Plus Environment

Page 9 of 25

Analytical Chemistry

9 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

3.1. Sex identification The first aim was to determine if ATR-FTIR had the required sensitivity to distinguish male from female A. aegypti mosquitoes using a variety of sample types (dried abdomens, fresh abdomens and mosquito lysates infected & uninfected with Wolbachia). The sex of these mosquitoes was determined using the antennae flagellum anatomy by trained entomologists. For sex recognition dried mosquito abdomens showed more consistent spectra with fewer outliers compared to fresh abdomens and mosquito lysates (results not shown). Figure 2a shows averaged raw and second derivative spectra recorded from methanol-dried abdomens from male and female uninfected A. aegypti lab mosquitoes. Large variations were observed in the C-O stretching region (1250-1000 cm-1) indicative of differences in carbohydrates, with the carbohydrate levels being greater in females than in males. In this case the dataset contained 221 samples, being 121 female and 100 male mosquitoes. Figure 2b shows the cross validation results for the optimum PLS-DA calibration model (154 samples) selected as well as the prediction plot (Figure 2c) for a test set (67 samples) for sex recognition of methanol-dried lab mosquitoes. The best model was based on the 1186-770, 1469-1355 and 1589-1784 cm-1 regions, using 8 latent variables and a pre-processing regime that entailed calculating the second derivatives, Standard Normal Variate (SNV), Pareto scaling (PS) and mean centring. Cross validation and independent testing indicated 2.6 and 0% classification error, respectively, which included the outliers proving the efficacy of this approach. Figure 2d shows the related receiver operation characteristic (ROC) plot32 based on prediction values from the cross validation and the prediction model (See Supplementary note 3 for more information)33. The area under the ROC curve for cross-validation and the independent prediction is close to 1 indicating the high performance of the prediction model.

ACS Paragon Plus Environment

Analytical Chemistry

Page 10 of 25

10 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Figure 2e shows the pre-processed average spectra coloured according to the regression vector, where the blue areas indicate spectral features associated with males (positive values of the regression vector) while the red are the spectral features of females (negative values of the regression vector). To interpret the regression vector it is important to note that it was constructed from second derivative spectra, and therefore the maxima in the raw spectra correspond to minima in the regression vector. The red minima bands in the 1150-1000 cm-1 region are assigned to the C-O stretching bond of sugars, which are associated with females and indicative of high amounts of carbohydrate compared to the males (blue). There are also some differences in the amide I mode associated with proteins. The females have a strong weighting for the 1643 cm-1 band, which may be indicative α-helical protein but this band could also have contributions from bound water and other molecules and could also be affected by light dispersion and scattering. Males have a strong weighting at 1624 cm-1 possibly from β-pleated sheets. These shifts maybe indicative of protein conformational differences between males and females related to the differences in sex-specific proteins, which are more alpha helical by nature in the females but other factors, as mentioned above, may also explain such shifts in the amide I mode. There are a number of other spectral features that distinguish males (blue) from females (red) shown in Figure 2e but the assignments of these bands are unknown, as they do not appear as major bands in the raw spectra, demonstrating the power of the multivariate approach in recognizing subtle chemical changes. Subsequently the methanol-dried abdomens of the female mosquitoes were mainly used in the ensuing analysis.

ACS Paragon Plus Environment

Page 11 of 25

Analytical Chemistry

11 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

3.2. Identification of 2 and 10 day Aedes aegypti mosquitoes Figure 3a compares average spectra from 118 two day old mosquitoes including male & females, infected & uninfected and 149 ten day old mosquitoes. The averaged spectra appear very similar except for some minor changes in the protein region (1700-1500 cm-1), which show on average a slightly higher concentration of proteins in the older mosquitoes. PLS-DA calibration models were developed using the second derivative spectra of dried mosquito abdomens in the selected lipid and finger print range (3000-1186 & 3140-2770 cm-1) using individual & combined sexes. Second derivative spectra of female mosquito abdomens (including both 2 & 10 days old) showed the lowest prediction errors (1.7-1.9%) using randomly selected test set. False predictions mainly belonged to 2 day old and small mosquitoes. Figure 3 (b-e) shows the selected calibration model developed using 2nd derivative spectra of 179 dried female mosquitoes in the 3000-1186 & 3140-2770 cm-1 ranges using 9 latent variables along with the ROC curve and prediction model, which included 118 samples in the test set. The area under the ROC curve (AUC) was 0.9977 and 0.9914 for the independent test and cross validation set, respectively, and demonstrates the high quality of the prediction model. The regression vector shows the important bands that discriminate 2 day from 10-day-old mosquitoes where “red” indicates strong bands associated with the old mosquitoes and the “blue” for younger mosquitoes. The “blue” highlighted band at 3008 cm-1 is assigned to the methyne stretching vibration from unsaturated fatty acids. It is evident that younger mosquitoes have higher concentrations of unsaturated fatty acids compared to their more mature counterparts. Changes in lipids are also evident in the CH2 and CH3 region (3100-2800 cm-1). Moreover, the ester carbonyl band at ~ 1743 cm-1 is a unique marker for fatty acids and appears blue shifted for younger mosquitoes compared with older mosquitoes. This shift can be seen in the regression plot as “blue” (2 days old) to the “right” of the 1743 cm-1 band and red (10 days old) to the left of this band.

Other important bands that have

ACS Paragon Plus Environment

Analytical Chemistry

Page 12 of 25

12 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

strong weightings for the older mosquitoes include those at 2881, 1238, 1458 and 1628 cm-1, while bands that have strong weightings for younger mosquitoes include those at 2962, 2927 and 1547 cm-1.

3.3. Detection of Wolbachia infection status in Aedes aegypti mosquitoes Female A. aegypti mosquitoes require a human blood meal to develop their eggs and as such are the main vector for transmitting the dengue virus between people. Female mosquitoes are generally larger in size than males and they have a higher Wolbachia load and greater total absorbance than their male counterparts. For this reason they provide a more robust prediction model for diagnosing Wolbachia infection compared to combining both male & female mosquitoes in the same model (results not shown). Figures 4a-e show the results for spectral comparison and modelling for the detection of Wolbachia infection in methanol-dried abdomens from female A. aegypti laboratory mosquitoes. The average spectra from ~60 Wolbachia-infected and ~60 uninfected female abdomens are shown in Figure 4a. The spectra are very similar and show only small changes in the amide I band at ~1650 cm-1 and amide II mode ~1544 cm-1 between Wolbachia infected and uninfected mosquitoes. The result implies that on average Wolbachia infected mosquitoes have higher protein content. Although the changes in the average spectra are small, the PLS-DA modelling (Figure 4b and 4c) can readily distinguish between the two types, showing only one miss-classification in the independent test set. The ROC curve (Figure 4d) supports these results showing excellent sensitivity and specificity for both cross validation and independent test for the model selected. An inspection of the regression vector in Figure 4e indicates that changes in the CH stretching region (3100-2800 cm-1) are strongly weighted in the model indicating different types of lipids are responsible for the

ACS Paragon Plus Environment

Page 13 of 25

Analytical Chemistry

13 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

classification. Bands at ~2927 cm-1 and 2854 cm-1 assigned to the asymmetric and symmetric CH2 bands, respectively, which are shifted when comparing infected mosquitoes with negative controls. The shift in these bands has been correlated with changes in the order and disorder of membrane lipids and the length of the lipid chains34. In general these bands are shifted towards lower wavenumber values as the order of the membranes increase. In our case, the presence of Wolbachia shifts CH2 bands to higher wavenumber values, indicating a decrease in the lipid membrane order. The involvement of lipids is further supported by the ester carbonyl band from fatty acids at 1743 cm-1, which is strongly weighted towards negative Wolbachia infection. In contrast the 2690 cm-1 band, assigned to CH3 asymmetric vibrational modes from lipids and amino acid side chains, is weighted towards positive Wolbachia infection. In summary, the classification model is mainly based on the fact that Wolbachia infection invokes several changes in lipids, including an increase of the CH3 vs C=O bonds and a disordering of the lipid membranes. We postulate that lipid production and membrane disorder is triggered by the presence of Wolbachia infection35-36. Molloy et al.35 investigated Wolbachia lipid metabolism in Aedes aegypti mosquitoes and reported major shifts in the cellular lipid profile in the presence of Wolbachia.

They also found that

cholesterol contributes to the mechanism of pathogen blocking by Wolbachia. The extremely high sensitivity and specificity shows the enormous potential of this technique as a rapid and easy method to screen for Wolbachia infection. At this stage it appears that the discrimination is based on detecting phenotypic changes in the mosquito induced by the bacterium and to date we have no direct evidence of bacterial contributions to the spectrum. Nonetheless the differences between infected and uninfected mosquitoes are consistent and inspecting the regression vector can assist in identifying the infrared markers responsible for the classification.

ACS Paragon Plus Environment

Analytical Chemistry

Page 14 of 25

14 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Summary of the parameters and results from models selected in the study are given in Supplementary Table S1.

3.4. Field mosquito studies For the field mosquito study fresh mosquito abdomens & mosquito lysates were used for sex & infection detection by applying different data processing tools including linear & nonlinear modelling techniques and different ranges of spectra (for details refer to Online Methods). These mosquitoes were collected from areas of North Queensland (Australia) where Wolbachia-infected mosquitoes (wMel strain) had been released in order to prevent the spread of dengue (www.eliminatedengue.com). The mosquitoes were collected as part of periodic weekly sampling. This sampling took place over several weeks, so the mosquitoes used come from different periods and cohorts. In general, more spectral variation is observed in field mosquitoes compared to lab mosquitoes and hence a nonlinear modelling method, namely an Artificial Neural Network (ANN) was employed, which gave a lower prediction error for female infection detection as well as sex recognition compared to PLS-DA. ANN models had 10 to 17% prediction error & NN-Pattern recognition indicated 78% accuracy on infection from prediction of an independent test set (Supplementary Note 2) using fresh mosquito abdomens. As an example, spectra of ~200 female fresh field mosquito abdomens were studied to detect the presence of Wolbachia. The minimum prediction error of 10% was achieved using 70 spectra as unknowns (not used for calibration model) by developing an ANN prediction model using PC scores as inputs into the ANN from normalised 2nd derivative spectra for the spectral ranges 3050-2800 & 1300-700 cm-1. The selectivity and specificity were 84 and 92% respectively. Spectra of very small mosquitoes and recently blood-fed females were removed as outliers. When squashed the blood directly contacts the

ACS Paragon Plus Environment

Page 15 of 25

Analytical Chemistry

15 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ATR crystal and dominates the spectrum reducing the sensitivity and specificity. Sex identification of field mosquitoes could also be readily achieved using both linear and nonlinear methods where greater than 95% specificity and sensitivity was obtained (Supplementary Note 2). The larger spectral variability observed from field-collected samples is not unexpected given that these mosquitoes, unlike those collected from laboratory colonies, had different ages and sizes and had remained in BG sentinel traps in the houses were traps were deployed for up to 7 days, after which they were identified and collected in 80% ethanol for processing at Monash University. Moreover, the nutritional status (including whether they recently had a blood meal or not) of field-collected mosquitoes is not controlled and this likely affects and increases variability in the biochemical composition of the samples.

4. Conclusions In summary; i) dried mosquitoes give more robust spectral models compared to fresh & mosquito lysates in the case of laboratory grown mosquitoes, ii) PLS-DA performed very well for testing sex, age and Wolbachia infection when using dried lab bred mosquitoes, iii) both the lipid and nucleic acid/carbohydrate regions were important in establishing the calibration models for prediction of 2 and 10 day old mosquitoes, sex & infection status, iv) for infection detection only female mosquitoes should be used for calibration models as these provide a better signal-to-noise ratio compared to males and the variation between males and females far outweighs the variation between Wolbachia infected and non-infected mosquitoes. v) Due to the inherent variability in field mosquitoes and for ease of sample preparation spectra of fresh hydrated mosquitoes were modelled using an ANN. Larger datasets, more targeted data acquisition, and the application of different algorithms will likely

ACS Paragon Plus Environment

Analytical Chemistry

Page 16 of 25

16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

lower the error rate and further improve the sensitivity and specificity of the technique. Here we demonstrate the potential of ATR-FTIR spectroscopy as an alternative fast and cheap method for diagnosing Wolbachia infection in both lab and field mosquitoes. Our results indicate that application of mid IR spectroscopy can achieve an accuracy of 95-97% for mosquito wMel infection differentiation in Aedes aegypti compared and accuracy of 95-97% for classifying 2 and 10 day mosquitoes and sex identification. These results show better predictive outcomes compared to a recent near-infrared spectroscopy (NIRS) study by Sikulu-Lord23 where two strains of Wolbachia were investigated namely wMelPop and wMel. For wMelPop from uninfected wild type samples NIRS gave an accuracy of 96% (N = 299) and 87.5% (N = 377) for males and females, respectively. Similarly, females and males infected with wMel were differentiated from uninfected wild type samples with accuracies of 92% (N = 352) and 89% (N = 444)20. NIRS could differentiate wMelPop and wMel transinfected females with an accuracy of 96.6% (N = 442) and males with an accuracy of 84.5% (N = 443). The study by Sikulu-Lord did not actually distinguish males from females nor predicted the age of the mosquitoes, however, it is relatively easy to distinguish females from males by eye and it should be noted we only tested 2 and 10 day mosquitoes and not their full life span. Mid-IR enables the identification of features and assignment of bands associated with infection status, age and sex. More accurate and meaningful feature selection assists with the development of more robust models based on related bands and actual chemical features for identification without involving unnecessary bands that may cause overfitting. With current technology it is possible to analyse 3000 mosquitoes per week but future technological developments such as an automated ATR multi-well sample holder would make it possible to obtain over 10,000 mosquito spectra per week. The ATR-FTIR in combination with ANN was found to be particularly effective at discriminating spectra of Wolbachia infected and uninfected mosquitoes in the field and hence shows enormous

ACS Paragon Plus Environment

Page 17 of 25

Analytical Chemistry

17 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

potential for a new technique at the front line to help monitor the spread of Wolbachia infection. It should be noted that PCR assays can detect different strains of Wolbachia and future studies will determine if the ATR technique has the required sensitivity to detect different strains of Wolbachia and other bacterial species for that matter.

Acknowledgements BRW is supported by an Australian Research Council (ARC) Future Fellowship grant FT120100926. We acknowledge Mr Finlay Shanks for instrumental support, the Eliminate Dengue team in Cairns for providing field mosquito samples, the Eliminate Dengue insectary team at Monash for providing colony samples and the Eliminate Dengue diagnostics team at Monash for validating the Wolbachiainfection status of the mosquitoes by PCR.

Methods Supporting Information Available: Methodological Notes. This material is available free of charge via the Internet at http://pubs.acs.org.

References: 1.

2. 3. 4. 5. 6. 7.

8.

Deen, J.; Lum, L.; Martinez, E.; Tan, L. H., In Dengue: guidelines for diagnosis, treatment, prevention and control, World Health Organization & Special Programme for Research and Training in Tropical Diseases: 2009; Geneva, pp 23-55. Rodenhuis-Zybert, I. A.; Wilschut, J.; Smit, J. M., Cell. Mol. Life Sci. 2010, 67, 2773-2786. Mustafa, M.; Rasotgi, V.; Jain, S.; Gupta, V., Med. J. Armed Forces India 2015, 71, 67-70. Screaton, G.; Mongkolsapaya, J.; Yacoub, S.; Roberts, C., Nat. Rev. Immunol. 2015, 15, 745759. Bhatt, S.; Gething, P. W.; Brady, O. J.; Messina, J. P.; Farlow, A. W.; Moyes, C. L.; Drake, J. M.; Brownstein, J. S.; Hoen, A. G.; Sankoh, O., Nature 2013, 496, 504-507. Axford, J. K.; Ross, P. A.; Yeap, H. L.; Callahan, A. G.; Hoffmann, A. A., Am. J. Trop. Med. Hyg. 2016, 94, 507-516. Joubert, D. A.; Walker, T.; Carrington, L. B.; De Bruyne, J. T.; Kien, D. H. T.; Hoang, N. L. T.; Chau, N. V. V.; Iturbe-Ormaetxe, I.; Simmons, C. P.; O’Neill, S. L., PLoS Pathog. 2016, 12, e1005434. Hancock, P. A.; White, V. L.; Callahan, A. G.; Godfray, C. H.; Hoffmann, A. A.; Ritchie, S. A., J. App. Ecol. 2016, 53, 785-793.

ACS Paragon Plus Environment

Analytical Chemistry

Page 18 of 25

18 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29.

30. 31. 32. 33. 34. 35. 36.

Aliota, M. T.; Caine, E. A.; Walker, E. C.; Larkin, K. E.; Camacho, E.; Osorio, J. E., PLoS Negl. Trop. Dis. 2016, 10, e0004682. Dutra, H. L. C.; Rocha, M. N.; Dias, F. B. S.; Mansur, S. B.; Caragata, E. P.; Moreira, L. A., Cell Host Microbe 2016, 19, 771-774. Moreira, L. A.; Iturbe-Ormaetxe, I.; Jeffery, J. A.; Lu, G.; Pyke, A. T.; Hedges, L. M.; Rocha, B. C.; Hall-Mendelin, S.; Day, A.; Riegler, M., Cell 2009, 139, 1268-1278. Hoffmann, A.; Montgomery, B.; Popovici, J.; Iturbe-Ormaetxe, I.; Johnson, P.; Muzzi, F.; Greenfield, M.; Durkan, M.; Leong, Y. S.; Dong, Y., Nature 2011, 476, 454-457. Walker, T.; Johnson, P.; Moreira, L.; Iturbe-Ormaetxe, I.; Frentiu, F.; McMeniman, C.; Leong, Y. S.; Dong, Y.; Axford, J.; Kriesner, P., Nature 2011, 476, 450-453. Lambrechts, L., Ann. Transl. Med. 2015, 3, 288. Lee, S. F.; White, V. L.; Weeks, A. R.; Hoffmann, A. A.; Endersby, N. M., Appl. Environ. Microbiol. 2012, 78, 4740-4743. Braig, H. R.; Zhou, W.; Dobson, S. L.; O’Neill, S. L., J. Bacteriol. 1998, 180, 2373-2378. Calvitti, M.; Marini, F.; Desiderio, A.; Puggioli, A.; Moretti, R., PLoS One 2015, 10, e0121813. de Oliveira, C.; Gonçalves, D. d. S.; Baton, L. A.; Shimabukuro, P. H. F.; Carvalho, F. D.; Moreira, L. A., B. Entomol. Res. 2015, 105, 305-315. Aw, W. C.; Dowell, F. E.; Ballard, J. W. O., G3 - Genes Genom. Genet. 2012, 2, 1057-1065. Sikulu, M.; Killeen, G. F.; Hugo, L. E.; Ryan, P. A.; Dowell, K. M.; Wirtz, R. A.; Moore, S. J.; Dowell, F. E., Parasite. Vector. 2010, 3, 1-7. Mayagaya, V. S.; Michel, K.; Benedict, M. Q.; Killeen, G. F.; Wirtz, R. A.; Ferguson, H. M.; Dowell, F. E., Am. J. Trop. Med. Hyg. 2009, 81, 622-630. Perez-Mendoza, J.; Dowell, F. E.; Broce, A. B.; Throne, J. E.; Wirtz, R. A.; Xie, F.; Fabrick, J. A.; Baker, J. E., J. Med. Entomol. 2002, 39, 499-508. Sikulu-Lord, M. T.; Maia, M. F.; Milali, M. P.; Henry, M.; Mkandawile, G.; Kho, E. A.; Wirtz, R. A.; Hugo, L. E.; Dowell, F. E.; Devine, G. J., PLoS Negl. Trop. Dis. 2016, 10, e0004759. Baranska, M.; Schütze, W.; Schulz, H., Anal. Chem. 2006, 78, 8456-8461. Wilks, P., Spectroscopy 2006, 21, 32-34. Erukhimovitch, V.; Talyshinsky, M.; Souprun, Y.; Huleihel, M., In DNA viruses: methods and protocols, Lieberman, P. M., Ed. Springer: Totowa NJ, 2005; pp 161-172. Erukhimovitch, V.; Huleihel, M.; Huleihel, M., J. Spectrosc. 2013, 2013, Article ID 317458. Lee-Montiel, F. T.; Reynolds, K. A.; Riley, M. R., J. Biol. Eng. 2011, 5, 1-12. Hoffmann, A. A.; Montgomery B. L.; Popovici J.; Iturbe-Ormaetxe I.; Johnson P. H.; Muzzi P.; Greenfield M.; Durkan M.; Leong Y. S.; Dong Y.; Cook H.; Axford J.; Callahan A. G.; Kenny N.; Omodei C.; McGraw E. A.; Ryan P. A.; Ritchie S. A.; Turelli M.; O’Neill S. L., Nature, 2011, 476, 454-457 Lee, S. F., White, V. L., Weeks, A. R., Hoffmann, A. A.; Endersby, N. M., Applied and environmental microbiology, 2012, 78, 4740-4743 Brereton, R. G.; Lloyd, G. R., J. Chemomet. 2014, 28, 213-225. Metz, C. E., Semin. Nucl. Med. 1978, 8, 283-298. Fawcett, T., Pattern Recogn. Lett. 2006, 27, 861-874. Derenne, A.; Claessens, T.; Conus, C.; Goormaghtigh, E., In Encyclopedia of Biophysics, Roberts, G. C. K., Ed. Springer: Berlin, Heidelberg, 2013; pp 1074-1081. Molloy, J. C.; Sommer, U.; Viant, M. R.; Sinkins, S. P., App. Environ. Microbiol. 2016, 82, 31093120. Caragata, E. P.; Rancès, E.; Hedges, L. M.; Gofton, A. W.; Johnson, K. N.; O'Neill, S. L.; McGraw, E. A., PLoS Pathog. 2013, 9, e1003459.

ACS Paragon Plus Environment

Page 19 of 25

Analytical Chemistry

19 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Figure captions

Figure 1. Schematic showing concept of ATR spectroscopic diagnosis of mosquitoes showing the spectral acquisition using the ATR accessory, spectral pre-processing, data analysis and the outcome, which is a diagnostic model.

Figure 2a. Average spectra recorded from 100 males (blue) and 121 females (red) showing large differences in the carbohydrate region (1200-1000 cm-1). 2b. PLS-DA calibration model performed on 2nd derivative spectra of methanol-dried lab mosquito abdomens in the 31002800 cm-1 & 1250-1000 cm-1 ranges using 8-factors. 2c. Independent test set containing x samples projected onto the calibration set showing all samples classified correctly for sex. 2d. Receiver operation characteristic (ROC) plot with calculated area under curve showing a value near 1 indicating the high performance of the prediction model. 2e. Regression vector for PLS-DA model where “blue” is male infected and “red” is female (arrows indicate the band assignments).

Figure 3a. Overlaid average spectra of a 2 (red) & a 10 (blue) days old methanol dried female mosquitoes in the 3600-900 cm-1 range.

3b. PLS-DA calibration model of 2nd

derivative spectra from ~200 female dried lab mosquito abdomens 2 & 10 days old. 3c. PLSDA test model where an independent test set was projected onto the calibration model. 3d. ROC curve showing the extremely high sensitivity and specificity achieved in the modelling. 3e. Regression vector showing the weighted bands for the young (red) and old (blue) model.

ACS Paragon Plus Environment

Analytical Chemistry

Page 20 of 25

20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Figure 4a. Average ATR-FTIR spectra of ~200 spectra from infected (blue) and uninfected (red) methanol dried female A. aegpti lab mosquito abdomina 4b. PLS-DA calibration model built from second derivative spectra using the combined spectral ranges of 3100-2800 cm-1 and 1800-700 cm-1 of infected (blue) & uninfected (red) from ~200 female mosquito abdomina. 4c. PLS-DA independent test set projected onto the calibration model. 4d. ROC plot showing the high performance of the prediction model. 4e. Regression vector for PLSDA model where blue is Wolbachia infected and red is uninfected.

ACS Paragon Plus Environment

Page 21 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

TOC graphic 269x320mm (113 x 98 DPI)

ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Figure 1 354x160mm (72 x 72 DPI)

ACS Paragon Plus Environment

Page 22 of 25

Page 23 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

Figure 2 181x193mm (300 x 300 DPI)

ACS Paragon Plus Environment

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Figure 3 180x188mm (300 x 300 DPI)

ACS Paragon Plus Environment

Page 24 of 25

Page 25 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

Figure 4 188x193mm (300 x 300 DPI)

ACS Paragon Plus Environment