Study sites
Experimental studies were conducted in the UK and Burkina Faso. Experiments in the UK were conducted in the LITE facilities of the Liverpool School of Tropical Medicine (LSTM) in temperature and humidity controlled insectaries (26 ± 2 °C and a relative humidity of 80 ± 10% under a L12:D12 h light: dark cycle with a 1-h dawn and dusk). In Burkina Faso, larvae were collected in August and September 2021 and September and October 2022 from four villages Sitiena (-4.80490°W, 10. 60366 °N), Tiefora (-4.54916 °W, 10. 62368 °N) and Tounmousséni (-4.98095°W, 10. 20376 °N ) (2022 only) in the Cascades region in the southwest of the country. Larvae were collected as 3rd or 4th instar and reared to adults at the Centre National de Recherche et de Formation sur le Paludisme (CNRFP) insectaries in Banfora. Semi Field Studies were performed at the Institut de Recherche en Sciences de la Santé (IRSS) in the north of Bobo-Dioulasso in Vallée du Kou village VK5 (4.4201°W, 11.3824°N).
Mosquito collection and rearing
Laboratory colonies
The Kisumu strain of An gambiae s.s. was used for experiments at LSTM. Approximately 600 pupae were added to six cages and allowed to emerge. Any pupae that had not emerged 24 h later were discarded. A subsample of 100 mosquitoes were collected 2 days later (age group 2 days); four of the cages were then offered a blood meal (day 3) using human blood, procured from the non-clinical blood product stock of a UK blood bank, and a haemotek feeder with the remaining two cages being maintained on sugar only. The blood fed cages were re-fed a further two times (day 10 and day 17); on each occasion, females that had not blood fed were removed from the cage and discarded. Oviposition pots were provided on days 5 and 12. Fifty females were sampled from the blood fed and sugar fed cages at 2 days, 4–5 days, 8–10 days, 11–12 days, 15–17 days, 18–19 days and 20–21 days post emergence. All samples were snap frozen before being transferred to individual Eppendorf tubes with silica gel and stored for at least 7 days at room temperature before processing. The experiment was repeated a second time, using the subsequent generation of Kisumu. Mosquitoes from both replicates were pooled for analysis; mosquitoes from two consecutive days were combined to form 6 age groups. The experiment was designed to ensure that the sampled females contained a mixture of sugar and blood fed and gravid and non-gravid females at different ages to determine whether physiological status reduced the accuracy of the REIMS age grading.
In a separate experiment to determine whether REIMS could be used to assess the number of gonotrophic cycles female mosquitoes had undertaken (ovarian age, a potential proxy for biological age) mosquitoes were transferred to individual oviposition tubes after blood feeding. Briefly, approximately 2000 Kisumu pupae were transferred into a cage and the emerging adults were split into two groups, a blood fed and non-blood fed group. Mosquitoes were offered a blood meal on day 4 post emergence. Non-blood fed females were removed the following day and on day 6 females were aspirated into individual oviposition tubes (50 ml falcon tubes containing a dampened filter paper in the base of the tube to provide an egg laying substrate) for 72 h. A subset of 50 mosquitoes that had laid eggs and 50 that did not lay eggs were snap frozen and retained on silica gel. The remaining females were transferred back to a cage and blood feed again on day 10. The oviposition process was repeated as above with 50 mosquitoes that had completed two gonotrophic cycles stored on silica gel. The full experiment was repeated a second time, using a subsequent generation of Kisumu. In total, of 417 females offered a blood meal across both experiments, 76% blood fed and 230 of these (73%) oviposited, 113 (49%) in round one only and 117 (51%) in round one and two.
Larval collections
Anopheles larvae were collected from multiple aquatic habitats in each of the three villages and transferred to the Banfora insectaries. Larvae were maintained in chlorine-free borehole water and fed with cat food (Friskies ®). Pupae were transferred to cages (separate cages for collections from each of the three villages) and males and females were kept together under insectary conditions (27 ± 2 °C temperature, 70–80% humidity with 12:12 h light and dark cycle) with glucose (5%) provided ad libitum. Approximately 50 females and 20 males were aspirated from the cages on days 1, 3, 5, 9, 13 and 15 post emergences. Legs were removed for species ID and the remaining mosquitoes stored individually in Eppendorf tubes with silica gel. The remaining mosquitoes were harvested at 20 days.
In a separate experiment, the larval collections were repeated the following year, transferred to the Banfora insectaries and raised to adults as above but in this experiment, the abdomen was separated from the head and thorax prior to storage. Abdomens were used for REIMS analysis and the remaining parts for the species ID.
Species identification
Mosquito legs were used for mosquitoes’ identification. DNA extraction consisted of placing the six legs in a 1.5 mL microcentrifuge tube with 30 µL of STE (100 mM NaCl; 10 mM Tris-Cl, pH 8.0; 1 mM EDTA). The samples were heated at 95 °C for 15 min to release DNA, then cooled to room temperature and centrifuged to pellet debris. The supernatant containing DNA was transferred to a fresh tube and stored at -20 °C for further PCR-based species identification following the method of15.
Plasmodium berghei infection and analysis
P. berghei ANKA 2.34 parasites were maintained in 6- to 8-week-old female CD1 mice (Harlan) by mechanical passage (up to a maximum of eight). If required, hyper-reticulosis was induced 3 days before infection by treating mice intraperitoneally (i.p.) with 200 µL phenylhydrazinium chloride (PH; 6 mg/mL in PBS; ProLabo UK). Mice were infected i.p. and infections were monitored using Giemsa-stained tail blood smears as described previously16.
To infect mosquitoes, two groups of five mice were housed in individually ventilated cages. One (blood-feed only control) cage was left uninfected, whereas the remaining five mice were PH-treated, and 3 days later infected i.p. with 106 P. berghei ANKA 2.34. Three days post-infection, animals were anesthetized, and > 500 female Anopheles stephensi mosquitoes allowed to blood feed on each mouse. Twenty-four hours later, unfed mosquitoes were removed. Mosquitoes were maintained on 8% (w/v) fructose, 0.05% (w/v) p-aminobenzoic acid at 19–22 °C, and 50–80% relative humidity. Day 12 post-feeding, 50 individual mosquito midguts were dissected and oocyst intensity and prevalence observed by standard phase microscopy and recorded to establish actual infection levels. The remaining ∼450 mosquitoes were examined by REIMS as described below.
Semi field experiments
Anopheles larvae were collected from different breeding sites located in Vallee de Kou VK7 village, in Bama between September and October 2021 and transferred to the field insectaries at IRSS and reared until pupae stage. Pupae were released into a newly refurbished semi-field station (SFS) located in the same village. The SFS is a fully screened house divided in several compartments separated by a corridor as described17. The climatic conditions including temperature, light, humidity, wind speed within the SFS were the same as outdoor conditions. The whole SFS is protected by a surrounding narrow water-filled channel to exclude ants and other predators. The release was done in one compartment refurbished for the experiment. In this compartment four claypots, containing cotton wool soaked in sugar water (5% glucose) were provided to serve as refuge of mosquitoes. Potential oviposition sites were emptied every 3 days to discard any larvae; in this way the age of the adults collected in the SFS could be ascertained with an accuracy of +/- 1 day from the date of addition of pupae. Approximately 50 adult females were collected from the SFS by aspiration on day 1, 3, 7, 9, 14 and 16 post 1st emergence. Mosquitoes were stored in individual Eppendorfs containing silica gel.
In order to mitigate against low recapture rates in the SFS, a separate subset of 150 pupae were released into a big cage each round ((2 × 2 × 2 m). Pupae were introduced and aspirated on the same day into both the big cage and SFS. Mosquitoes in the big cage did not have access to a blood meal.
Adult collections
Approximately 900 adult female mosquitoes were aspirated from inside homes in Tengrela using electric aspirators. Collections were performed between the hours of 6h00 AM and 9h00 AM in September and October 2021. Aspirated mosquitoes were stored in paper cups and then transported to the Banfora insectaries conserved in a coolbox. At the insectary mosquitoes from the same location were pooled and transferred into cages. Three hundred mosquitoes were snap frozen and then transferred to individual Eppendorfs with silica gel on the day of collection. The remaining mosquitoes were retained in cages with a source of sugar before being harvested at either day 4 or day 8 post collection.
REIMS analysis
Samples were analysed via a rapid evaporative source (REIMS, Waters, Wilmslow, UK) attached to a Synapt G2Si instrument ion mobility equipped quadrupole time of flight mass spectrometer (Waters, UK). The specimens were burned/evaporated using a monopolar electrosurgical pencil (Erbe Medical UK Ltd, Leeds, UK), which was connected to a VIO 50 C electrosurgical generator (Erbe Medical UK Ltd, Leeds, UK), providing electrical current, and to the source inlet via plastic tubing. A black conductive rubber mat, placed underneath the samples, acted as a counter electrode and facilitated the flow of electric current. To avoid inhalation of fumes during analysis, the burning process was performed within a fume box (Air Science, Lydiate, Merseyside, UK). Insects were analysed using a 40 W setting on the generator and the cutting option of the pencil. To increase conductivity and to protect the counter electrode during analysis, specimens were placed on a piece of glass microfibre paper (GFP, GE Healthcare Whatman) on top of a wet paper surface (moistened with MilliQ water).
While burning the entire biomass of single specimens, the aerosol was aspirated through the pencil and the attached 3 m long tubing into the REIMS source, using a nitrogen powered venturi valve on the source inlet. To increase the aerosol capture, a wide bore piece of plastic tubing was additionally placed over the tip of the electrosurgical pencil. A whistle incorporated into the Venturi tube guided the aerosol as well as a lock mass solution of leucine enkephalin (Waters, UK) in propan-2-ol (CHROMASOLV, Honeywell Riedel-de-Haën) into the source. This also filters the incoming aerosol to prevent larger particles from entering the inlet capillary. Inside the source, the ionised particles were declustered through contact with a heated impactor (Kanthal metal coil at 900 °C).
Mass spectra were acquired in negative ion mode at a rate of 1 scan per second between 50 m/z and 1200 m/z. The sample cone and heater bias were set to 60 V. Instrument calibration was performed daily in resolution mode using a 0.5 mM solution of sodium formate (flow rate 50 µl/min). The lock mass solution (0.4 µg/ml, leucine enkephalin in IPA) was continuously introduced during sample analysis at 30 µl/min. For each experiment, samples were analyzed in a formally randomised order on one or more days.
Data analysis
The raw mass spectra were imported into the model building software package Offline Model Builder (OMB-1.1.28; Waters Research Centre, Hungary), which allows separation of sample groups (classifications) based on principal component analysis (PCA) and linear discriminant analysis (LDA). Data were additionally analyzed using R (version 4.2.1)18 and the R Studio environment19, by PCA and LDA, as well as random forest analysis.
For Offline Model Builder, the burn events of the analyzed specimens were defined individually, summing all the MS scans within each chosen area. The option to create one spectrum per sample was selected. Other pre-processing parameters included the intensity threshold, which was set between 4e5 and 9e5 (depending on the background baseline), spectra correction using the lock mass (leucine enkephalin, 554.26 m/z) and background subtraction. To reduce the complexity of the mass spectral data, all acquired data points from 50 to 1200 m/z were combined into mass bins, each 0.1 m/z wide. Subsequent model calculation was based on principal component and linear discriminant analysis (PCA-LDA).
The models built by Offline Model Builder were cross-validated (leaving out 20% of data, for outliers the standard deviation multiplier was set to 5) to obtain the correct classification rate, as well as the number of failures and outliers and a matrix displaying the number of correctly and incorrectly identified samples of each classification. To additionally test discrimination, sample classifications were randomised and the data were re-analyzed, the expectation being a random distribution of samples and the inability to achieve separation.
For further analysis with R, the data matrix of each model was exported as a .csv file from Offline Model Builder, containing information about classification and the relative intensities for every mass bin. The matrices were used to perform random forest analysis in R using the package ‘randomForest’20. The data sets were randomly split into a training set (approx. 70% of the data) and a test set (approx. 30% of the data). Random forest results are displayed in form of confusion matrices. Trees were conducted 10 times for every model (using a different, randomly selected subset of samples for training and testing every time); the numbers of correctly identified and confused samples were turned into percentages and averaged. The optimal number of trees and mtry value were determined during the first analysis of each model and kept the same for each repeated analysis. A second R package, called ‘randomForestExplainer’21, was used to identify the most informative bins/ions that were driving class separation. PCA-LDA was also performed within the R environment, using the in-built package ‘stats’ and the package ‘MASS’22 and results visualized in form of kernel density plots and 2D- and 3D-scatter plots created using ‘ggplot2’23 and ‘scatterplot3d’24. PC-LDA based models were built with a quarter of the maximum number of principal components (PCs) unless stated otherwise. In OMB, models were built with a maximum of 100 PCs.
