Population data on potential host species

Data on the population of human and domestic animals from three villages, Wama Kusaye (8°58.695′N, 36°48.558′E; 1443 m above sea level), Baka-Boro (8°57.715′N, 36°52.058′E; 1522 m above sea level) and Machara (8°58.028′N, 36°42.994′E; 1514 m), in the East Wollega Zone of western Ethiopia was obtained from agricultural extension workers and the local administration office. The common practice in this region is for livestock and people to share their living quarters, and as such, the assumption was made that the availability of potential hosts is similar both indoors and outdoors.

Mosquito collection and blood meal analysis

Blood-fed mosquitoes were collected from the three villages on five separate days, using standard collection methods [32]. Indoor resting mosquitoes were collected in ten houses, in each village, from 06:00 to 08:00. Mosquito-knockdown collections were performed by spraying with Kilit™ (Miswa Chemicals Ltd, UK), a synthetic pyrethrum. Outdoor-resting mosquitoes were surveyed at five pit shelters dug for the purpose (1.5 × 1.0 × 2.0 m, with horizontal ‘pockets’ dug in the four walls of each) [32] in each village.

Anopheles mosquitoes were counted and then sorted by sex, abdominal condition (unfed, freshly fed, half gravid and gravid), and species using morphological keys [33]. The Anopheles mosquitoes that were provisionally identified as An. gambiae s.l., were screened using polymerase chain reaction (PCR) described by Scott et al. [34] and conclusively identified.

Freshly blood-fed mosquitoes were cut transversely between the thorax and the abdomen, and the posterior portions containing the blood meal were tested for source host blood by the direct enzyme-linked immunosorbent assay (ELISA) [35]. Commercially available anti-host (IgG) conjugates against human, cattle, goat, sheep and chicken (Kirkegard and Perry Laboratories, MD, USA) were used in the ELISA. Control samples consisted of blood drawn from a human (KTJ), and blood obtained from cow, sheep and goat (Addis Ababa Abattoirs enterprise), as well as chicken blood obtained from a local restaurant. Each mosquito was tested simultaneously for human, cattle, goat, sheep, and chicken antibodies. Significant differences in blood meals found in indoor- and outdoor-resting mosquitoes were determined using Chi squared (χ ²) analyses (Prism v. 5, GraphPad, CA, USA).

Forage ratio

The forage ratio was calculated as the proportion of host species present in blood meals of An. arabiensis divided by the proportion of host species available in the environment [36].

Volatile headspace collections

Headspace collections were obtained from cows, sheep, goats, and chicken. For this purpose, at least five individuals of each species were randomly selected from the Wama Kusaye village. The host hair, wool or feathers were cut with sterilized scissors, enclosed in separate polyacetete bags (Toppits, Melitta, Sweden) and immediately transported to the laboratory. The mixed hair, wool or feathers (20 ± 1 g) were placed in a glass wash bottle. A charcoal-filtered, continuous airstream (100 ml min⁻¹) was drawn by a diaphragm vacuum pump (KNF Neuberger, Freiburg, Germany) through the bottle onto an aeration column for 24 h. The aeration column consisted of a Teflon tube (4 mm diameter × 40 mm length) holding 30 mg Porapak Q (80/100 mesh, Alltech, Deerfield, IL, USA) between polypropylene wool plugs. Adsorbed volatiles were desorbed by eluting each column with 500 µl of re-distilled n-hexane (≥99.9 % purity, Merck KGaA, Darmstadt, Germany) and condensed under N₂ to approximately one-quarter of the volume. Samples were stored at −20 °C.

Mosquito rearing

Anopheles arabiensis (Dongola strain) were maintained at 27 ± 2 °C, 70 ± 2 % relative humidity and at a light:dark cycle of 12:12 h. Larvae were reared in plastic trays (20 × 18 × 7 cm) and fed Tetramin™ fish food (Tetra, Melle, Germany). Pupae were transferred to Bugdorm cages (30 × 30 × 30 cm, MegaView Science, Taiwan) for adults to emerge. Adults were provided 10 % sucrose solution ad libitum. For colony maintenance, female mosquitoes were provided with sheep blood (Håtunalab, Bro, Sweden) using an artificial feeder (Hemotek, Discovery Workshops, Accrington, UK). Electrophysiological analysis was conducted on four- to six-day post-emergence non-blood fed female mosquitoes.

Electrophysiology

Antennal responses to the headspace volatile collections were examined by combined gas chromatography (GC) and electroantennographic detection (EAD) analysis as well as electro-antennography (EAG) using an EAG system (IDAC-2; Syntech, Kirchgarten, Germany) and an Agilent 6890 N GC (Agilent Technologies, Santa Clara, CA, USA). For the GC-EAD analysis, the GC was equipped with a HP-5MS (Agilent Technologies) fused silica capillary column (30 m × 0.25 mm; df = 0.25 µm). Hydrogen was used as mobile phase (Q = 45 cm s⁻¹). Two µl of each sample were injected (splitless mode, 30 s, injector temperature 225 °C). The GC oven temperature gradient was programmed from 30 °C (4-min hold) at 8 °C min⁻¹ to 250 °C (5-min hold). To the GC effluent, 4 psi of nitrogen was added and split 1:1 in a Gerstel 3D/2 low dead volume four way-cross (Gerstel, Mülheim, Germany) between the flame ionization detector and the EAD. The GC effluent capillary for the EAD passed through a Gerstel olfactory detection port-2 transfer line, which mirrored the GC oven temperature, into a glass tube (8 mm diameter × 10 cm length), where it was mixed with charcoal-filtered, humidified air (1 l min⁻¹). The antenna was placed 0.5 cm from the outlet of this tube.

For EAG recordings, the excised head of a female An. arabiensis was used. After removing the distal tip of the first flagellomere of one antenna, it was inserted into a recording glass electrode filled with Beadle-Ephrussi ringer (140 mM NaCl, 4.7 mM KCl, 1.9 mM CaCl₂·2H₂O) and connected to a pre-amplifier (10×) probe connected to a high impedance DC amplifier interface box (IDAC-2; Syntech). The indifferent electrode was inserted into the occipital foramen. At least six GC-EAD runs were made for each headspace volatile collection on different preparations.

Chemical analysis

Volatile collections were analysed on a combined gas chromatography and mass spectrometer (GC–MS) (6890 GC and 5975 MS; Agilent Technologies) operated in the electron impact ionization mode at 70 eV. The GC was equipped with a similar column as for the GC-EAD analysis. Helium was used as the mobile phase (Q = 35 cm s⁻¹). The GC oven temperature was programmed as for the GC-EAD analysis above. Compounds were identified according to their Kovat’s indices and mass spectra in comparison with custom made and NIST-05 libraries, and confirmed by co-injection of authentic standards (Additional file 1).

Dose–response experiments

For further verification of the physiological activity of the chemicals identified through GC-EAD and GC-MS analyses, dose–response experiments were conducted by EAG recordings using synthetic standards (Additional file 1). Concentrations ranged in decadic steps from 0.001 to 10 % (volume/volume) for each synthetic compound. Dilutions of compounds were prepared in redistilled n-hexane (LabScan, Malmö, Sweden), except for furfuryl alcohol for which absolute ethanol was used (LabScan). Odour stimuli were produced by loading 10 µl of each diluted synthetic test compound onto a filter paper (1 × 1.5 cm, Munktell Filter AB, Sweden) inserted inside a glass Pasteur pipette. Pipettes with formulated filter papers were kept for 30 min in a fume hood prior to use to allow for solvent evaporation. The pipette was connected via a silicone tube to a stimulus generator (CS-55; Syntech) and the tip of the pipette was inserted into the glass tube with an air flow (1 l min⁻¹) directed towards the antenna. Stimuli were produced by puffing air (0.5 l min⁻¹) through the pipette during 0.5 s; each pipette was used only once. Hexane was used as a solvent blank, as the first and last stimulus for every replicate, except ethanol that was used as a solvent blank for furfuryl alcohol. Each set of odour stimuli was tested on one antenna (n = 6). The responses to each test stimulus were calculated by subtracting the averaged response amplitude of the solvent controls from the response amplitude of the stimulus.

Field evaluation of identified host and non-host volatiles

Field experiments were conducted in the Wama Kusaye village. In the village, 11 thatched houses were selected based on similarities in size, with houses separated approximately 200 m apart. The experimental design followed a Latin square, in which treatments were randomly assigned to houses on the first day and then rotated between houses to minimize location bias over the following days, for a total of 11 days. The experiments were conducted in November and December 2012, i.e., after the long rainy season, when host-seeking An. arabiensis were readily available. In each house, a single volunteer (27–36 years old) slept under an untreated bed net. A Centers for Disease Control and Prevention (CDC) mini-light trap (BioQuip Products, Inc, CA, USA), with the light bulb removed, was hung next to the foot of the bed net, approximately 1 m above ground level. Ethical clearance was obtained from the Ethical Committee of the Faculty of Science, Addis Ababa University conforming to the WMA Declaration of Helsinki.

Synthetic compounds of nine of the GC-EAD active compounds identified in the volatile headspace collections of the non-host (chicken) and hosts (cattle, goats, and sheep) of An. arabiensis were used in the study. Dispenser vials (PE# 733, Kartell, Italy), each containing 0.5 g of a synthetic compound released at a rate of 1 mg h⁻¹, were suspended approximately 10 cm beside and 20 cm below the trap using wire hooks (Fig. 1). The required release rate was achieved by varying the number of caps attached to each trap, and the size of the hole in the cap from which the chemical could volatilize. The number of caps and hole size required was determined: full caps were weighed and reweighed after 1, 2, 3, 4, 5, 6, 12, and 24 h of exposure to field conditions (25 ± 1 °C, 60 % RH). This procedure was repeated six times to calculate an average release rate for each compound. As a negative control, a similar trap, with solvent alone, was used. In addition, a caged chicken surrounded by a fine mesh screen, to prevent chicken-mosquito interactions, and suspended in a similar way as the dispensers, served as a control (Fig. 1). The traps were turned on at 18:00 and turned off the following morning at 06:00. Caught mosquitoes were enumerated and identified to species, as described above. The effect of compounds on the number of mosquitoes caught (distributed response variable) was subjected to a generalized linear mixed effect model procedure (GLMM, lmer) in the R statistical software version 3.1.1. (“house” and “day” were controlled for as random effects). The model used a Poisson distribution and log-link function for its construction, and AIC was used for model evaluation. For a comparative analysis among the different compounds, a posthoc test, adjusted for multiple comparisons, was performed on a linear mixed effects model (R, lme4, multcomp; Chi squared, χ ²; P < 0.05).

Mosquito species identification and composition

Four species of Anopheles mosquitoes, An. arabiensis, Anopheles funestus s. l., Anopheles nili and Anopheles coustani, were collected and identified in the study villages (n = 4844). Anopheles arabiensis, as determined by PCR analysis of 386 mosquitoes (more than 5 % of the mosquitoes caught), was the most abundant species, comprising more than 98.5 % of the total mosquitoes caught. A total of 4739 female An. arabiensis were collected from the study villages, using pyrethrum spray sheet collections (n = 1036, 758 and 503 for Wama Kussaye, Baka-Boro and Machara, respectively) and artificial pit shelters (n = 1264, 639 and 539 for Wama Kussaye, Baka-Boro and Machara, respectively). During the field evaluation of the non-host volatiles, two species of mosquitoes, An. arabiensis and An. coustani, were collected and identified. Anopheles arabiensis, as determined by PCR, was the most abundant species comprising more than 97 % of the total mosquitoes caught (n = 583).

Host-species abundance and feeding preference of Anopheles arabiensis

GC-EAD and GC-MS analyses of headspace volatile collections

In the headspace volatile collection from the non-host, chicken, 11 GC-EAD active compounds were detected (Table 2). Of these, limonene, β-myrcene, nonanal, sulcatone and cis-limonene oxide were also found within volatile collections of one or more of the non-human hosts. The remaining compounds, hexadecane, naphthalene, isobutyl butanoate and trans-limonene oxide, were specific to chicken. This study was unable to confirm the identity of two chicken-specific compounds using commercially available synthetic standards and are here referred to as unknown 1 and 2.

For further verification of the physiological activity of the compounds identified through GC-EAD and GC-MS analyses, dose–response experiments were conducted by EAG recordings using synthetic standards (Additional file 1). The EAG dose–response analysis of the GC-EAD active compounds demonstrated that An. arabiensis respond to all tested synthetics in a dose-dependent manner, and confirmed that the antennae were differentially sensitive to these compounds (Additional file 2).

Field evaluation of non-host and generic volatiles

Overall, the tested volatiles had a significant effect on trap catches when tested in the field using suction CDC traps (CDC traps without light; \(\varvec{\chi}_{10}^{2}\) = 226.76, P < 0.001; Figs. 1 and 2). Traps baited individually with the chicken-specific volatiles, isobutyl butanoate, naphthalene, hexadecane and trans-limonene oxide, and with the generic compounds, limonene, cis-limonene oxide and β-myrcene, caught significantly fewer An. arabiensis compared to the solvent baited negative control trap (Fig. 2). Similarly, a significantly lower number of mosquitoes were caught in a trap baited with a live, caged chicken (Fig. 2). In contrast, CDC traps baited with either of the generic compounds, sulcatone or nonanal, did not affect the number of An. arabiensis caught, compared to the solvent baited negative control trap (Fig. 2).