Pyrethroids

The 8 × 15 K Agilent microarray design chip (A-MEXP-2196) (Mitchell et al., 2012 (link)) was used to detect the set of genes differentially expressed between the resistant population of Vallée du Kou and a susceptible laboratory colony Ngoussou. Each array contains 60 mer probes designed from all 13,000 transcripts of the Ensembl P3.5 A. gambiae genome annotation, plus additional probes for the detoxification genes from a previous microarray design, the ‘detox chip’, used previously to explore metabolic resistance in A. gambiae (David et al., 2005 (link)).
RNA was extracted from three batches of ten females 3 day old A. gambiae s.s. from a F₁ sample from the VK6 population (nonexposed to insecticide but known to be resistant to multiple insecticides from bioassays results of VK) and from the Ngoussou strain which is fully susceptible to pyrethroids, DDT, carbamates and organophosphate with 100% mortality observed 24 h after 1 h exposure. RNA was isolated using the Picopure RNA isolation kit (Arcturus). The quantity and quality of extracted RNA were assessed using NanoDrop ND1000 spectrophotometer (Thermo Fisher) and Bioanalyzer (Agilent, Santa Clara, CA, USA) respectively. Complementary RNA (cRNA) of each sample was amplified using the Agilent Quick Amp labeling Kit (two-color) following the manufacturer's protocol. cRNA from the VK6 samples were labeled with cy3 dye while the cRNA from the susceptible strain Ngoussou was labeled with the cy5 dye. cRNA quantity and quality were checked before labeling using the NanoDrop and Bioanalyzer. Labeled cRNAs were hybridized to the arrays for 17 h at 65 °C according to the manufacturer's protocol. Five hybridizations between cRNA from VK and Ngoussou were carried out by swapping the biological replicates (Fig. S6).
Microarray data were analyzed using Genespring GX 11.0 software. In order to identify differentially expressed genes, a cut-off of 2-fold-change and a statistical significance of P < 0.05 and P < 0.01 were applied. The P values were generated from a t-test against zero using the data from the five hybridizations (Fig. S6) after a multiple testing correction using the Benjamin–Hochberg test. Enrichment analysis was carried out using the Blast2Go software (Conesa et al., 2005; Gotz et al., 2008 ) to detect the major Gene Ontology (GO) terms over-represented among the set of probes up or under-transcribed in the VK population in comparison to the entire microarray chip using a Fisher's test for statistical significance. The microarray data from this study were submitted to Array Express, accession number: E-MTAB-1083.

Magude district is a rural district located in the northwest of Maputo province, southern Mozambique (26° 02’ 00” south, 32° 17’ 00’ east), away from Mozambique’s main road. A baseline census conducted at the beginning of 2015 identified 48,448 residents living in 10,965 households. Malaria transmission is perennial with marked seasonality between November and April, coinciding with the rainy season. P. falciparum accounts for the majority of infections. Studies in nearby areas have identified Anopheles funestus s.s. as the most abundant vector responsible for the majority of transmission, followed by Anopheles arabiensis [10 (link),11 (link)]. High levels of pyrethroid resistance have also been described in A. funestus, but not in A. arabiensis [12 (link),13 (link)]. Artemether-lumefantrine, the national first-line antimalarial treatment, remains fully efficacious [14 (link)].The district has 27 community health workers (CHWs), nine rural health facilities (HFs), one referral health center with an inpatient ward in Magude Sede (main town), although the more severe cases are sometimes referred to the neighboring hospitals of Xinavane or Manhiça (Manhiça District). Magude’s baseline sociodemographic characteristics and health systems have been described elsewhere [15 (link)].

In the context of EHTs, a superiority trial could be used to demonstrate that a novel type of ITN kills significantly more mosquitoes than the current standard of care. Here we consider a trial designed to evaluate whether a next-generation ITN, designed to control pyrethroid-resistant mosquitoes, outperformed a standard pyrethroid-only net. If both unwashed and washed nets are included in the trial, one must decide which assessment(s) to make, e.g. a comparison of the unwashed nets, a comparison of both the unwashed nets and the washed nets, or an assessment of the average induced mortality across the unwashed and washed nets. In the Supplementary file 1 we show a worked example, using the output from the regression model to decide whether the null hypothesis should be rejected (i.e. if the next-generation ITN is deemed superior to the pyrethroid-only ITN). Based upon the trial design outlined above, a statistical test (implemented within the R package lme4) is used to determine whether one net is superior to another at a given significance level. As discussed in the Tutorial (Supplementary file 1), various statistical tests can be used to do this. As a default, the regression analysis carried out using the lme4 package uses a Wald test. As this is very quick to calculate, we shall favour this approach when carrying out time-consuming power analyses. In the Tutorial, we discuss the limitations of this approach, and how its accuracy can be verified.
In this work, we use PBO-pyrethroid ITNs as an example of a next-generation ITN. When comparing the mortalities of PBO-pyrethroid ITNs to pyrethroid-only ITNs, we will utilise a mathematical relationship between the two, fitted by Sherrard-Smith et al. (2022b) (link). The relationship, calibrated using data from a recent meta-analyses of experimental hut trial data (Nash et al., 2021 ), states that if the ITN-induced mortality of a pyrethroid-only ITN is $l$ then the mortality induced by a PBO-pyrethroid ITN is given by $1/(1+\mathit{exp}\left[-\left({\alpha }_1+{\alpha }_{2}l\right)\right]$ . Here we use the best-fit parameters obtained by Sherrard-Smith ( ${\alpha }_1=-1.43$ ; 95% CrI [−1.54, −1.33], ${\alpha }_2=5.60$ ; 95% CrI [5.29, 5.93]) (Sherrard-Smith et al., 2022b (link)). We use this result to generate parameters for simulated EHTs. The results are compared to mean mortality estimates recorded in the EHT studies in order to identify potential outliers.

The methodology used here for evaluating ITNs using a non-inferiority trial design is that proposed by the WHO (WHO, 2018 ). These trials should include an untreated net, a pyrethroid-only net (referred to here as ‘the standard comparator’), a first-in-class product that has already shown epidemiological impact (‘the active comparator’), and a novel product (‘the candidate net’). Both unwashed and washed ITNs should be included, which means that the trial should have at least 7 arms. The assessment of whether the candidate net is non-inferior to the active comparator is made by calculating the odds ratio (OR) of the mortality induced by these two products. As shown in the Supplementary file 1, we fit a regression model for which data for the unwashed and washed ITNs of the same type are grouped together. The OR is constructed from the parameters estimated from the regression modelling, along with their confidence intervals. For the candidate product to be judged non-inferior to the active comparator, the lower limit of the 95% confidence interval of the OR must be greater than the pre-selected non-inferiority margin (NIM). The NIM is a compromise between the measurement error of the assay and the minimum acceptable epidemiological impact of an inferior product. For mosquito mortality, the WHO have chosen an OR of 0.7 for the NIM. Rather than assessing non-inferiority for the unwashed and washed nets separately, we use the averaged mortality across the unwashed and washed arms to construct the OR. The non-inferiority assessment is illustrated in Fig. 2: the top panels show a scenario in which the candidate product is judged to be non-inferior to the active comparator, and the lower panels outline a scenario where this is not the case (here we would say that the candidate product is ‘not non-inferior’ to the active comparator). In the Tutorial (Supplementary file 1), we present a worked example of an assessment of non-inferiority, carried out for simulated EHT data.Fig. 2

Illustration of the non-inferiority assessment. Here we show estimates of mosquito mortality after 24 ​h in a simulated experimental hut trial. Points (with 68% and 95% confidence interval estimates on the horizontal lines) show unwashed ITN (red), washed ITN (yellow) and an untreated control net (grey). Overall estimate of the ITN mortality is shown in orange and is calculated by combining data from washed and unwashed nets. We show two scenarios, one in which a candidate product is deemed to be non-inferior to an ITN already evaluated in a randomised control trial (top panels: A, B), and another in which it is not non-inferior (lower panels: C, D). Both unwashed and washed nets are included in the simulated trials: the averaged mortality across unwashed and washed nets was used to calculate the odds ratio between the two products. The right-hand panels (B, D) show the odds ratios for the two scenarios, along with their confidence intervals. If the lower confidence interval is greater than the pre-selected non-inferiority margin (NIM), the candidate product is judged to be non-inferior to the pre-evaluated ITN. Here we use an NIM of 0.7, as recommended by the WHO (WHO, 2018 ). We note that the same underlying parameters were used to generate the two synthetic datasets, illustrating the variation present in the assay.

Fig. 2