The bulk of insecticide susceptibility testing is performed by testing phenotypic resistance using standardized bioassays. These insecticide susceptibility bioassays have been designed for ease of use by local mosquito surveillance authorities. The tests assess the proportion of mortality in local mosquito populations exposed to a certain diagnostic dose (predetermined using susceptible mosquito strains) under controlled laboratory conditions. Phenotypic assays can be performed on either adult mosquitoes (to test the susceptibility to adulticides) or aquatic immature mosquitoes (susceptibility to larvicides).

Genetic resistance surveillance is regularly used in antimalarial drug resistance surveillance to determine the prevalence or frequency of drug resistance in a given area, population, or even a single patient (Nsanzabana et al., 2018; World Health Organization, 2010). If genetics is used in insecticide-resistance surveillance, the objective is typically to identify the resistance mechanism once phenotypic resistance has been detected rather than as a monitoring tool (Ishak et al., 2015; Menze et al., 2016). This is because genetic assays to detect insecticide resistance still face several challenges (Vontas and Mavridis, 2019). While the genetic base for several resistance mechanisms has been identified, genetic markers are still missing for others, and their identification is a slow and complicated process (Donnelly et al., 2016). Additionally, where genetic markers have been identified, they may not fully explain the resistance phenotype, as shown for DDT-resistant Anopheles gambiae in West Africa (Mitchell et al., 2014). The mismatch between single-nucleotide polymorphisms (SNPs) and resistance phenotype could be a result of several factors. First, not every gene involved is likely to have been identified (Weetman and Donnelly, 2015). Furthermore, copy number variations (CNVs) additionally confer resistance (Assogba et al., 2015; Labbé et al., 2007), and these CNVs have been demonstrated to be common in the genome and widespread in the population (Lucas et al., 2019). Such CNVs are more difficult to detect than SNPs and are therefore largely understudied (Lucas et al., 2019). Moreover, CNVs are harder to correlate with phenotype due to the duplications of both susceptible and resistance alleles. Other factors, such as ecological and climatic variability (discussed in the next section) and interactions between resistance markers, may also lead to differences in insecticide susceptibility (Edi et al., 2014; Mitchell et al., 2012). Finally, as mosquitoes are diploid, heterozygosity at the resistance loci may confer different levels in insecticide susceptibility compared to susceptibility in homozygotes (South et al., 2020). While the advantage of genetic tools is that they result in reliable replicable data, they cannot accurately assess the prevalence or frequency of phenotypic insecticide resistance at this moment in time. However, the field is rapidly improving and genetic tools may become more widely used and informative in the near future (Vontas and Mavridis, 2019).

Standardized phenotypic insecticide-resistance bioassays are unlikely to reflect the efficacy of vector control intervention on mosquito control in the field for the aforementioned reasons. Indeed, this gap between bioassay data and mosquito control is frequently discussed in the literature (Churcher et al., 2016; Grossman et al., 2020; Hancock et al., 2018; Ranson and Lissenden, 2016; Sherrard-Smith et al., 2018; Venter et al., 2017; Vontas and Mavridis, 2019) and even acknowledged in the WHO test procedure guidelines (World Health Organization, 2016b). This disparity between the insecticide efficacy in a lab bioassay and in a field setting may explain the fact that the strong increase in pyrethroid resistance globally (as determined by laboratory bioassays) does not always translate well to mosquito control failure, nor to transmission control failure (as reviewed in Alout et al., 2017a; Alout et al., 2017b). Since the aim of vector control is not necessarily to reduce the mosquito burden, but to reduce vector-borne disease transmission, we should not just focus on mosquito control failure but on the impact of insecticide resistance on transmission control failure. Deltamethrin LLINs have in fact been demonstrated to provide protection in Malawi despite pyrethroid resistance being common in the major mosquito vector(s) (Lindblade et al., 2015). In Kenya, schoolchildren living in areas with a high prevalence of kdr mutations in vector mosquitoes were less likely to be infected with malaria if they reported using pyrethroid-based LLINs (Okoyo et al., 2015). Moreover, a cohort study carried out in five countries on two continents found no significant association between pyrethroid resistance in malaria vectors and malaria prevalence or incidence (Kleinschmidt et al., 2018). However, in contrast, some other studies do suggest a link between insecticide resistance and reduced malaria control. In South Africa, switching from pyrethroids to DDT in IRS campaigns following widespread pyrethroid resistance did lead to large reductions in malaria transmission (Maharaj et al., 2005). Furthermore, a large randomized control trial in Tanzania comparing pyrethroid-only LLINs with pyrethroid-piperonyl butoxide (PBO) LLINs (with the synergist piperonyl butoxide that can partially restore pyrethroid susceptibility), in combination with or without IRS with a non-pyrethroid, showed improved malaria control with IRS or the pyrethroid-PBO LLINs, suggesting that pyrethroid resistance decreased the efficacy of the standard LLIN alone (Protopopoff et al., 2018).

In order to predict the impact of insecticide resistance on disease transmission, we need to understand which mosquito traits play a role in disease transmission and how resistance interacts with these traits (Alout et al., 2016). Vectorial capacity is the ability of mosquito populations to transmit a specific pathogen between humans (Cohuet et al., 2010). Transmission dynamics models can be used to extrapolate resistance data to make predictions on the impact of insecticide resistance on disease transmission (Briët et al., 2013; Churcher et al., 2016; Enahoro et al., 2020; Okumu et al., 2013; Sherrard-Smith et al., 2018). However, many unknowns around the impact of insecticide resistance on important parameters influencing vectorial capacity persist, such as mosquito density, blood-meal frequency, life span, pathogen development time within the mosquito, and the mosquito’s vector competence (i.e., the capacity of a mosquito to acquire, maintain, and transmit a parasite). Therefore, there is an urgent need for detailed data of randomized control trials of vector control interventions combined with contemporaneous experimental hut data to support the predictive value of these models in transmission control (Sherrard-Smith et al., 2018). For example, insecticide resistance is expected to reduce insecticide-induced mosquito mortality and thus increase the transmission potential of diseases. However, delayed mortality due to insecticide exposure has frequently been observed across different mosquito species (see for instance, Agnew et al., 2004; Martins et al., 2012) and may result in a reduced disease transmission given malaria parasites’ relatively long incubation time within the mosquito host (ranging from 7 days to several weeks, depending on temperature [Mordecai et al., 2013; Paaijmans et al., 2009] and frequency of blood-meal intake [Habtewold et al., 2021; Shaw et al., 2020]). Alarmingly though, recent observations of highly pyrethroid-resistant mosquito populations from Burkina Faso show a lack of such delayed mortality in some populations (Hughes et al., 2020).

Insecticides and insecticide resistance can influence several other vector behaviors that affect disease transmission, such as host choice and biting frequency (Alout et al., 2017b; Rivero et al., 2010; Thomas and Read, 2016). As target-site resistance is mediated through mutations of key proteins of the neural network, this type of resistance can (a) impact mating competitiveness of resistant individuals, with a kdr heterozygote mating advantage observed in Anopheles coluzzii (Platt et al., 2015) and (b) reduce biting rates, as resistant individuals may be less responsive to olfactory cues (Siegert et al., 2009). Metabolic resistance has also been associated with behavioral changes: resistant Culex pipiens were more predated upon than their susceptible counterparts, hence reducing mosquito densities. This is likely a side effect of esterase overproduction inducing resource depletion (Berticat et al., 2004). Vector competence can also be impacted by insecticide resistance (Rivero et al., 2010). A striking example is Wucheria bancrofti, a parasitic worm causing lymphatic filariasis, that can develop in susceptible mosquitoes but is unable to fully develop in highly resistant Culex quinquefasciatus (McCarroll and Hemingway, 2002). Such processes could especially happen in the case of metabolic resistance, whose important energetic costs could generate trade-offs between resistance and immunity and/or between resistance and parasite development. These mechanisms have been proposed among possible explanations for the increased competence of ace-1-resistant Cx. quinquefasciatus for West Nile virus (Atyame et al., 2019) and of An. gambiae for P. falciparum infection (Collins et al., 2019). In addition, glutathione S-transferase metabolic resistance has also been reported to be associated with increased Plasmodium infections in Anopheles funestus (Ndo et al., 2019; Tchouakui et al., 2019). While kdr resistance has been observed to reduce vector competence in the lab setting (Alout et al., 2013), contrasting results have been observed in An. gambiae in the field, with no association (Collins et al., 2019) or a significant positive association (Kabula et al., 2016; Ndiath et al., 2014) observed between kdr resistance and P. falciparum development. However, the majority of these studies have the limitation that they are either performed with laboratory strains of both the mosquito and the parasite or use field observations, which are confounded by the impact of insecticides on longevity, meaning older mosquitoes are more likely to be resistant but are as such also more likely to be infectious due to the long Plasmodium incubation period (Minetti et al., 2020). Of note is that insecticides themselves (dissociated from insecticide resistance) may also impact disease transmission. Sub-lethal exposure to insecticides may reduce host-seeking or blood-feeding success. Complex interactions between insecticide exposure and biting behaviors, giving contrasting results (increase or decrease in the host-seeking or blood-feeding success of resistant mosquitoes following exposure depending on the insecticide used), have also been observed (Diop et al., 2020; Glunt et al., 2018; Liu et al., 1986; Thiévent et al., 2019). Failure in blood feeding can be of great importance regarding malaria transmission, given that blood-meal intake could be critical for malaria development (Habtewold et al., 2021; Shaw et al., 2020). Furthermore, the exposure to insecticides negatively impacted P. falciparum parasites’ development in the midgut of insecticide-resistant An. gambiae (Alout et al., 2014a; Kristan et al., 2016) and similarly reduced the infection rate of Plasmodium berghei in An. gambiae (Hauser et al., 2020). Similarly, exposure to residual levels of insecticide to Aedes albopictus may lower infection rates and body titers of dengue virus (Richards et al., 2017).

The complexities described above may explain some of the widely acknowledged discrepancies between laboratory resistance bioassay results and the observed nonfailure of vector control interventions (Churcher et al., 2016; Grossman et al., 2020; Ranson and Lissenden, 2016; Sherrard-Smith et al., 2018; Vontas and Mavridis, 2019). The WHO introduced intensity bioassays by incorporating 5x and 10x discriminating concentrations in their resistance testing guidelines, mentioning that “resistance phenotypes detected using the discriminating concentrations do not necessarily provide information in terms of efficacy failure of that insecticide in the field” (World Health Organization, 2016b). Similarly, the CDC bottle bioassay has a modified version, the Resistance Intensity Rapid Diagnostic Test (I-RDT) with higher discriminating dosages (Bagi et al., 2015; Brogdon and Chan, 2010). However, the higher discriminating concentrations, while adding some predictive benefits, still suffer from the same problems as described above (Venter et al., 2017). The missing link in this story is our understanding of the efficacy of actual vector control products in targeting local vector populations under local and natural conditions. WHO cone bioassays (World Health Organization, 2013a) with locally collected mosquitoes are an easy method to assess resistance under ambient conditions and are already frequently performed. However, these are most typically executed on the non-blood-fed F1 of field-collected mosquitoes or field-collected larvae reared to adults in the lab under room temperature or ambient lab conditions (e.g., Allossogbe et al., 2017; Bagi et al., 2015; Glunt et al., 2015; Hughes et al., 2020; Ibrahim et al., 2019; Riveron et al., 2019; Sahu et al., 2017). This approach likely fails to capture the impact of a vector control intervention on the local mosquito populations, as important factors such as mosquito age distribution, feeding status, as well as local environmental conditions that can affect insecticide toxicity, are ignored. Furthermore, the cone test does not replicate the natural contact of the mosquito with interventions, as highlighted above. There are some tests (such as WHO’s tunnel test, tent assay, cup assay, and the Mosquito Contamination Device (MCD) bottle assay) that allow for more natural mosquito behavioral responses (including transient contact and avoidance) to insecticide-treated nets, thereby assessing if the efficacy of those nets is underestimated compared to standard cone bioassays (Grossman et al., 2020; Sternberg et al., 2014; World Health Organization, 2013a). This is particularly relevant for insecticides that have a high excito-repellent effect, such as permethrin. The tunnel test uses a small experimental chamber (60×25×25 cm) between two mosquito cages. Mosquitoes are released on one side of a holed piece of net, while a suitable bait (e.g., a guinea-pig or rabbit) is placed on the other side and made available for mosquito feeding. Mortality and blood-feeding inhibition are the recorded outcomes. The MCD bottle assay consists of a 1 l bottle in which mosquitoes are released and are lured toward a piece of LLIN netting on the other side by placing a glass jar filled with hot tap water (approximately 35°C) that acts as an attractant. Immediate knockdown and mortality after 24 hr are the recorded outcomes. In addition, net efficacy tests can be performed with the Ifakara Ambient Chamber Test (I-ACT), which allows mosquitoes to interact with a human host sleeping under an actual net, mimicking the field situation (Massue et al., 2019). Its big advantage is that tests can be conducted under local natural conditions during the night when the nets are normally used and mosquitoes are active. Recorded outcomes are mortality and blood-feeding inhibition. Tunnel assays are also recommended to be conducted overnight, but other net-exposure assays are most likely performed during daytime (i.e., regular working hours), not taking the insect’s circadian rhythm into account (World Health Organization, 2013a). All the above tests are typically conducted with laboratory-reared mosquitoes. Experimental hut studies probably mimic natural conditions more closely when the efficacy of vector control interventions is evaluated, by allowing local vector populations to enter (or not) the huts (Briët et al., 2012; Massue et al., 2016), but those set-ups are costly, time-consuming, limited to few sites globally, and raise the question of how representative the artificial structures are compared to local natural houses. Meta-analyses on the predictive value of phenotypic bioassay data on the effectiveness of LLINs and IRS in terms of deterrence and blood-feeding success using hut trials may be a useful future tool for policy decision-making. However, the uncertainty in the estimates is large due to the high level of variability in the data (Churcher et al., 2016; Sherrard-Smith et al., 2018).

We, and others before us (Bagi et al., 2015; Grossman et al., 2020; Ranson and Lissenden, 2016), argue that assessments of local mosquito susceptibility against vector control products under local and natural conditions (in other words, practical resistance) are currently of utmost importance to adequately guide the decision-making process of insecticide-based interventions. Whilst technical (or standardized) resistance bioassays are valuable for tracking the emergence and spread (i.e., evolution) of insecticide resistance over space and time, they do not inform vector control programs whether the envisioned vector control product (with the same active ingredient) leads to operationally significant resistance. More worryingly, certain chemical classes may be removed from the national/local list with useful public health insecticides based on those resistance diagnostics, while actual products with the same active ingredient may still be effective (acknowledging this efficacy may differ between areas and between seasons for the variety of reasons mentioned before). With a limited insecticide arsenal available, it is important to fully understand which interventions still play a critical role in vector control.

To address the issue at hand, we urge policy makers and partners to develop more pragmatic and practical testing procedures to evaluate the efficacy of actual vector control tools (e.g., LLINs and IRS products). There will be no single set of guidelines to bridge the gap between technical and practical resistance, as different products have different properties. A few examples to illustrate this problem are (1) products that are known to repel mosquitoes in addition to killing them (N'Guessan et al., 2001), (2) products with slow-acting active ingredients that require delayed mortality observations (Kweka et al., 2018), and (3) non-uniform products, such as nets with deltamethrin-coated side panels but with a mixture of deltamethrin and PBO on the top panel (Kweka et al., 2017). As such, each product (or group of products with the same active ingredients) may need its own practical testing guidelines. What is clear, however, is that the body of literature highlights a need for tests to be performed under local environmental conditions (in local houses in areas where their implementation is envisioned; for IRS/LLINs: under relevant microclimatic conditions, for IRS: on relevant wall types) using local vector populations. If these mosquitoes are collected (with, e.g., non-lethal host-seeking traps [behavior relevant for LLINs] or by aspiration of mosquitoes off the walls [behavior relevant for IRS]) and used immediately in the tests, they reflect the actual local vector population we aim to kill and/or repel. One should conduct the tests at relevant times in terms of season (when tools are typically distributed, sprayed, and/or used) and time of day (when local mosquitoes are active and come in contact with the insecticide), as these times can affect their flight activity and detoxification processes (Balmert et al., 2014; Rund et al., 2016). Tests should be replicated and have an adequate number of control treatments (untreated nets or walls) due to the higher degree of variability in practical testing procedures. Two further arbitrary factors contributing to practical resistance are contact time and contact frequency (i.e., how long and how often a mosquito will contact the insecticide), which can be easily addressed by having a set of mosquitoes exposed shorter (e.g., 3 min like in the current LLIN tests) or longer (e.g., 30 min as in current residual efficacy IRS test, but note this will depend on the mode of action of the active ingredient), as well as exposing a cohort of mosquitoes multiple times during a single night to the product. However, these decisions are best guided by performing studies on actual contact times of local mosquito species with the intervention(s), similar to what has been suggested by WHO for assessing the effect of holes in LLINs on mosquito entry into a damaged net (World Health Organization, 2013b). Finally, when a product has excito-repellency properties, tests can be performed in larger cages (instead of the current WHO cones) placed against the nets, to create a refuge for mosquitoes. For larvicides, tests could be performed outside at ambient temperatures, with field-collected larvae of all stages, using original source water. Even better, but more labor intensive, is to treat several natural water bodies with larvicides following all national procedures and to add collected larvae of all stages randomly to floating meshed containers in those pools and several untreated breeding sites (serving as controls). This allows for larvae to be exposed to natural physiochemical water properties (e.g., temperature, salinity, conductivity, total dissolved solids) and food sources, but leaves them protected from predators. As larval survival is known to be dependent on larval densities (Epopa et al., 2018; Muriu et al., 2013), two or three different densities could be assessed, depending on local larval habitats. Such practical resistance data could be used to inform policy decisions, for instance, changing insecticide class when the insecticides under these natural conditions lead to a mortality below an established threshold.

While practical guidelines will inform vector control programs about the efficacy of actual vector control products, the remaining outstanding question will be how practical resistance outcomes translate to changes in disease transmission, a similar open question to technical resistance. This is especially important since practical resistance surveillance does not take the impact of resistance or insecticide exposure on mosquito transmission potential into account, which is beyond the capability of routine surveillance. We, therefore, argue for a need to collect practical resistance data to re-evaluate the association between insecticide resistance and vector-borne disease transmission, using practical resistance testing guidelines. With malaria control stagnating over the past 5 years, this new surveillance data will allow us to better understand the link between insecticide resistance, vector control efficacy, and disease epidemiology. In addition, these practical resistance data will provide vector control programs with relevant data on the efficacy of the vector control tools tailored to their local context and, therefore, the ability to make decisions on implementation strategies.