Evaluating the Evidence from Molecular Structure and Population Studies for Cross-Resistance to the Pyrethroids Used in Malaria Vector Control

The primary malaria control intervention in high burden countries is the deployment of long-lasting insecticide-treated nets (LLINs) treated with pyrethroids, alone or in combination with a second active ingredient or synergist. It is essential to understand whether the impact of pyrethroid resistance can be mitigated by switching between different pyrethroids or whether cross-resistance precludes this. Structural diversity within the pyrethroids could mean some compounds are better able to counteract the resistance mechanisms that have evolved in malaria vectors. Here we consider variation in vulnerability to the P450 enzymes that confer metabolic pyrethroid resistance in Anopheles gambiae s.l. and Anopheles funestus. We assess the relationships among pyrethroids in terms of their binding anity to key P450s and the percent dep letion by these P450s, in order to identify which pyrethroids diverge from the others. We then investigate whether these same pyrethroids also diverge from the others in terms of resistance in vector populations. We found that etofenprox, which lacks the common structural moiety of other pyrethroids, potentially diverges from the commonly deployed pyrethroids in terms of P450 binding anity and resistance in malaria vector populations, but not depletion by the P450s tested. These results are supplemented by an analysis of resistance to the same pyrethroids in Aedes aegypti populations, which also found etofenprox diverges from the other pyrethroids in terms of resistance in wild populations. In addition, we found that bifenthrin, which also lacks the common structural moiety of most pyrethroids, diverges from the commonly deployed pyrethroids in terms of P450 binding anity and depletion by P450s. However, resistance to bifenthrin in vector populations is largely untested. The prevalence of resistance to the pyrethroids α-cypermethrin, cyuthrin, deltamethrin, λ-cyhalothrin, and permethrin was correlated across malaria vector populations and switching between these compounds as a tool to mitigate against pyrethroid resistance is not advised without strong evidence supporting a true difference in resistance.


Introduction
The primary malaria control intervention in high burden countries is the deployment of long-lasting insecticide-treated nets (LLINs) treated with pyrethroids, alone or in combination with a second active ingredient or synergist [1,2]. Widespread and increasing resistance to pyrethroids is, therefore, a serious potential threat to malaria control [3,4]. Because the options for LLINs are limited, it is essential to understand whether the impact of resistance can be mitigated by switching between different pyrethroids or whether cross-resistance precludes switching. Pyrethroids listed by the World Health Organization (WHO) for malaria control are differentiated into two groups based on biological activity that is associated with the absence (Type I) or presence (Type II) of an alpha-cyano group (Fig. 1). Type II pyrethroids are more lethal to insects because of their higher potency to the voltage-gated sodium channel (VGSC) in nerve membranes, the primary target site of pyrethroids [5,6]. The higher potency of Type II pyrethroids such as deltamethrin and α-cypermethrin translates into much lower doses being required to treat vector control products compared with Type I pyrethroids such as permethrin. This has led to increased deployment of alpha-cyano pyrethroids, in particular α-cypermethrin, which is currently used in 28% of the prequali ed vector control products [2]. Generally, pyrethroids used in vector control possess the common structural motif of phenoxy benzyl alcohol coupled with a cyclopropane ring via an ester bond, except for bifenthrin and etofenprox (Fig. 1). This narrow spectrum of chemical variation among pyrethroids makes it likely that cross-resistance will occur in malaria vector populations.
The high burden countries where LLINs are deployed are concentrated in Africa where the most important vectors are Anopheles gambiae s.l. and Anopheles funestus [7]. Pyrethroid resistance in malaria vectors is primarily associated with target-site insensitivity due to mutations in the Vgsc gene known as knockdown resistance (kdr) and increased detoxi cation activity known as metabolic resistance.
Metabolic mechanisms of resistance are found in all African malaria vectors whereas kdr mutations are common in species of the Anopheles gambiae complex but not in the An. funestus subgroup [8][9][10][11][12][13][14]. There are multiple amino acid substitutions that cause target-site insensitivity resulting in pyrethroid resistance [15]. This includes a mutation, M918T, that produces a super-knockdown (s-kdr) phenotype in house ies. Structure modelling studies in M918T phenotypes indicate that the highest degree of resistance in s-kdr house ies depends on the chemical structure of the insecticide which is positively correlated with the presence of an α-cyano group coupled with a phenoxybenzyl moiety in the larger Type II pyrethroid molecules such as deltamethrin and fenvalerate [16]. By comparison, the most common Vgsc resistance allele in west African An. gambiae populations, L1014F, is not in uenced by pyrethroid chemical structure when expressed alone in house ies [17].
Although kdr mutations are common in An. gambiae s.l., they may have a relatively modest impact on resistance, and they are absent from highly pyrethroid-resistant An. funestus populations, suggesting that metabolic mechanisms have a greater impact in African malaria vectors [18][19][20]. Metabolic resistance is most commonly mediated by elevated levels of cytochrome P450 enzymes [21]. Transcriptome-wide studies of gene expression in resistant and susceptible mosquito strains have found upregulation of several P450 genes is associated with resistance to both a Type I pyrethroid (permethrin) and a Type II pyrethroid (deltamethrin). For example, upregulation of the CYP6P3 gene and its orthologues CYP6P9a and CYP6P9b, and of the CYP6AA1, CYP6Z1 and CYP6Z3 genes, is associated with resistance to both pyrethroids in An. gambiae / An. coluzzii and An. funestus [22][23][24][25][26][27][28][29][30][31][32]. In addition, upregulation of the CYP6Z2 gene in An. gambiae and An. coluzzii, and the CYP6M7 gene in An. funestus, is also associated with resistance to both pyrethroids [23-25, 27, 28, 30-32]. These ndings from studies of gene expression in resistant and susceptible strains provide evidence for P450-mediated cross-resistance in Anopheles populations, particularly to deltamethrin and permethrin, but associations with resistance to more than one pyrethroid have not always been found, a limited range of pyrethroids have been tested, and these studies don't give an indication of whether cross-resistance is stronger between some pyrethroids than others. Like the Anopheles vectors, target-site mutations and metabolic resistance are also thought to be the main resistance mechanisms in Aedes mosquitoes [33,34].
An assessment of the impact of individual structural variation within the pyrethroid class on resistance in the eld is required to inform the best use of different compounds. A previous study assessed resistance in malaria vector populations at over one thousand African sites and showed that when spatio-temporal trends were separated from noise in the susceptibility test data, strong associations among the resistance trends for three structurally similar pyrethroids (deltamethrin, -cyhalothrin and permethrin) were found [35]. The variance in the mean percent mortality values was 28 for the west Africa model and 23 for the east Africa model, re ecting the noisiness of the mortality data. This study also noted that the prevalence of resistance to permethrin was typically higher than that to deltamethrin, however, caution is needed when interpreting differences found using susceptibility test data because they may be due to real differences in the prevalence of resistance or differences in the calibration of the diagnostic dose or both. Diagnostic doses currently recommended for use were calculated by doubling the dose of a compound which kills 100% of a susceptible strain of a species, or doubling the LC 99 in this strain [36,37]. A robust recommendation should be based on data from multiple strains in different testing centres, but where this is not possible doses may not be well calibrated between compounds. It is clear that differences in resistance between individual pyrethroids cannot be generally assumed, but it remains unclear whether meaningful differences can occur, particularly when a wider range of pyrethroid chemistries is considered.
Here we take a new approach to assess variation in resistance among pyrethroids. We rst assess differences in pyrethroid chemistry that in uence inhibition of the key enzymes that confer metabolic resistance in African malaria vectors, and the rate of depletion of each pyrethroid by these enzymes [38]. Of the primary resistance genes, the P450 superfamily is most frequently associated with metabolic resistance to pyrethroids in malaria vectors. Therefore, we assessed the relative differences among six pyrethroids in terms of their molecular interactions with P450 enzymes from the major African malaria vectors by constructing a P450s structure-activity relationships model (P450s-SAR). We focus on αcypermethrin, deltamethrin and permethrin as most relevant for recommendations regarding the current LLIN options. However, for broader future consideration, we include bifenthrin, etofenprox, cy uthrin and λ-cyhalothrin, structurally varied pyrethroids that are also in the WHO's prequali ed list for malaria vector control ( Fig. 1) [2]. We then analyse resistance to these pyrethroids in multiple vector populations to determine whether the relative differences found by P450s-SAR studies translate into relative differences in resistance within wild populations. This is supplemented by an analysis of resistance in arbovirus vector populations. Finally, the resistance associations found across insecticide classes are also analysed in order to put the relationships found within the pyrethroids into the wider context of crossresistance generally and to further investigate whether cross-resistance predicted by laboratory studies can be detected as general trends in the eld data.

Material And Methods
In order to test whether relationships identi ed by SAR studies can be detected in the eld, we constructed dendrograms for the hierarchical relationships between pyrethroids found by a series of molecular and eld studies, and then compared the dendrograms obtained.
Relationships among pyrethroids in terms of functional activity data P450 inhibition assays using uorogenic probe substrates have become commonplace in drug discovery screening cascades and are a rapid method of screening for insecticide interactions with mosquito P450s to predict insecticide binding, metabolism, cross-resistance and synergism [38][39][40][41]. In this study, the half maximal inhibitory concentration (IC 50 ), which provides a value for inhibition of each P450 by each pyrethroid, and the percent depletion, which gives a value for metabolism of each pyrethroid by each P450, were both included to establish a P450s structural activity relationship model. This model was used to understand the chemistry of the pyrethroids, and the interaction with mosquito P450s that function as monooxygenases in metabolic resistance and to predict cross-resistance liabilities in vivo. Low IC 50 values indicate the pyrethroid is a potent inhibitor that may be able to counter resistance mediated by P450s. Low percent depletion indicates low metabolism of the pyrethroid, which means it may be less vulnerable to resistance mediated by P450s.
The IC 50 values for permethrin, etofenprox and bifenthrin (Type I) and deltamethrin, λ-cyhalothrin and αcypermethrin (Type II) pyrethroids that were exposed to recombinant P450s from the An. gambiae Kisumu strain (CYPs 6Z2, 6M2, 6P2, 6P3 and 9J5) and the An. funestus FUMOZ strain (CYP6P9a) were extracted from two studies [38,41]. In addition, inhibition activity data for these pyrethroids exposed to CYP6Z3 from the An. gambiae Kisumu strain were also generated (Additional File 1).
The values for percentage depletion (metabolism) of each pyrethroid by three of the enzymes, CYP6M2, CYP6P3 and CYP6P9a, which were expressed in a single plasmid construct, were also extracted from the same sources and used for the comparative analysis.
The two datasets were analysed using hierarchical clustering of rows (insecticide) and columns (P450) by Perseus v1.6.14.0 to produce two visual heat maps representing the clustered matrices for relative insecticide binding a nity and insecticide vulnerability to metabolic attack. The clustered matrices for functional activity data for these six pyrethroids against these seven P450s were then used to construct dendrograms for the hierarchical relationships among the pyrethroids.

Relationships among pyrethroids in terms of susceptibility test mortality in malaria vector populations
We accessed a published database of insecticide resistance in African malaria vectors [14] and identi ed all instances where a mosquito sample from the eld had been tested using two or more pyrethroids.
Pairs of results were extracted, rather than instances where a sample had been divided between tests of three or more pyrethroids, because there were insu cient data from studies testing > 2 pyrethroids against a single mosquito collection. This provided 3,153 pairs of WHO susceptibility test results from samples of the An. gambiae complex. Only data that detected resistance to at least one pyrethroid were included. That is, results from samples that had 100% mortality to all pyrethroids tested were excluded.
We conducted a series of correlation analyses to assess how closely associated each pair of pyrethroids is in terms of resistance. The mean value for the Pearson's correlation coe cient was calculated across 1,000 bootstrapped samples for each pyrethroid pair using SPSS Statistics v25. A Holm-Bonferroni correction was applied to identify signi cant correlations among the multiple tests conducted while avoiding false positives [42]. The mean correlation coe cients generated were ranked to identify the most and least closely correlated pyrethroids. These bootstrap mean correlation coe cients were used to construct a dendrogram of the hierarchical relationships among pyrethroids using the unweighted pairgroup method using the arithmetic mean (UPGMA [43]), where the highest correlation coe cient indicated the most closely related pair.
The analyses conducted using data from An. gambiae s.l. samples were repeated using data from An. funestus subgroup, An. arabiensis, An. coluzzii, An. funestus and An. gambiae samples. The same approach was also used for susceptibility test data from Aedes albopictus and Ae. aegypti to investigate whether the same relationships could be detected in these vectors of arboviruses, as detailed in Additional File 3. There were much lower data volumes for the individual Anopheles species, compared to An. gambiae s.l., and a limited selection of pyrethroid pairs could be tested so no dendrograms were constructed from these data. Finally, the correlations between resistance to deltamethrin and resistance to insecticides from other classes were calculated in order to put the relationships found within the pyrethroids into the broader context of cross-resistance.
The CYPs 6P3, 6M2 and 6P9a were selected for comparative metabolism analysis because they are commonly associated with pyrethroid resistance and amongst the earliest pyrethroid resistance markers to be functionally validated and most heavily used for in-vitro screening [29,41,45,46]. All of the pyrethroids apart from bifenthrin were strongly metabolised by 6P3 and its orthologues 6P9a expressed from An. gambiae and An. funestus respectively (Fig. 2B, Table S3). However, lower metabolism pro les were observed with 6M2 expressed from An. gambiae (Fig. 2B, Table S3). Notably, etofenprox was strongly metabolised by 6P3, 6M2 and 6P9a. Overall, the metabolism data presented in Fig. 2B and Table   S3 ranked etofenprox, deltamethrin and permethrin as the most vulnerable insecticides for metabolic attack by the three enzymes, followed by α-cypermethrin and -cyhalothrin, and bifenthrin demonstrated the lowest vulnerability.
The dendrograms indicate that permethrin and deltamethrin are closely related whereas bifenthrin diverges from these pyrethroids, in terms of inhibition of P450s and metabolism by P450s (Fig. 2).

Relationships among pyrethroids in terms of susceptibility test mortality in malaria vector populations
Each of the 15 pairs of values for pyrethroid resistance within An. gambiae s.l. was signi cantly correlated (Table 1). That is, populations with a higher prevalence of resistance to one pyrethroid tended to have higher prevalence of resistance to the others (Figs. 3 and S2). The pyrethroid pairs were ranked from the most closely correlated pair, deltamethrin vs λ-cyhalothrin, to the most divergent pair, etofenprox vs λ-cyhalothrin (Table 1, Fig. 4A). The correlation coe cients were used to construct a dendrogram of the hierarchical relationships among these pyrethroids (Fig. 4B). Deltamethrin, λ-cyhalothrin, permethrin, cy uthrin, and α-cypermethrin were closely related whereas etofenprox diverged from the other ve pyrethroids. Comparison of pyrethroid relationships seen in molecular and eld studies The three dendrograms using i) resistance in eld populations, ii) P450 inhibition and iii) depletion by P450s were re-constructed incorporating only the ve pyrethroids that were included in all three analyses (Fig. 5). The dendrograms for P450 inhibition and vector population resistance both show that deltamethrin, λ-cyhalothrin, permethrin are most closely related to each other, then α-cypermethrin, and etofenprox is most divergent (Fig. 5A and 5B). The dendrogram constructed using values for insecticide depletion metabolism by 6P3, 6M2 and 6P9a reveals different relationships among these pyrethroids, although permethrin and deltamethrin are still closely related (Fig. 5C).
Correlations in pyrethroid resistance within malaria vector species Across An. funestus subgroup communities, there were signi cant correlations between resistance to deltamethrin and λ -cyhalothrin, permethrin and λ-cyhalothrin, and deltamethrin and permethrin, and the same was true for the four species tested (Table 2, Figures S3 and S4). There were insu cient data to test the other pyrethroid combinations for the African malaria vector species. Across Ae. aegypti populations, resistance to cy uthrin, deltamethrin, λ-cyhalothrin and permethrin was signi cantly correlated whereas there were no signi cant correlations between these four pyrethroids and etofenprox (full results are given in Additional File 3).
In order to put the relationships found within the pyrethroids into the wider context of cross-resistance across the insecticide classes used for malaria vector control, the correlations between deltamethrin and six commonly used non-pyrethroid insecticides were also calculated. Signi cant correlations with the prevalence of resistance to DDT were found for species within the An. gambiae complex but not for An. funestus (Table S4 and Figure S5). No signi cant correlations were found between the prevalence of resistance to deltamethrin and that to bendiocarb or propoxur (carbamates), malathion, fenitrothion or pirimiphos-methyl (organophosphates) for species within the An. gambiae complex or An. funestus. Variation in pyrethroid resistance within populations of African malaria vector species The results presented above show signi cant correlations in resistance among the pyrethroids tested, but this result does not preclude the possibility that the prevalence of resistance is generally higher in one pyrethroid compared to the others across populations with differing levels of pyrethroid resistance. The insecticide depletion data presented above indicates that some pyrethroids are potentially more vulnerable to P450 attack, particularly etofenprox which was most depleted by the three P450s (Table  S3). This leads to the question of whether higher levels of resistance to this compound can be detected in wild mosquito populations. An analysis of the paired data from An. gambiae s.l. samples collected across Africa provides no evidence that the prevalence of resistance is consistently higher for etofenprox compared to the other pyrethroids in An. gambiae s.l. (Figure S6) but mortality was signi cantly lower after Ae. aegypti populations were exposed to etofenprox compared to mortality following exposure to deltamethrin, cy uthrin, λ-cyhalothrin and permethrin (Table S7).
To put the mortality differences found among pyrethroids ( Figure S6-S8) into the wider context of crossresistance, the prevalence of resistance to deltamethrin was compared to the prevalence of resistance to six non-pyrethroid insecticides in paired susceptibility tests ( Figure S9). A reversal in the differences between resistance to deltamethrin and to the organochlorine DDT was found, with An. gambiae s.l. species having signi cantly higher resistance to DDT whereas An. funestus had signi cantly higher resistance to deltamethrin. In all species tested, mortality was lower following deltamethrin exposure compared to bendiocarb and propoxur (carbamates), malathion, fenitrothion and pirimiphos-methyl (organophosphates) exposure.

Discussion
The results of this study highlight which of the pyrethroids used in malaria control are closely related in terms of inhibition of and depletion by P450s. Other studies of structurally diverse pyrethroids have also shown variation in P450 metabolism of pyrethroids with different structure. An in vivo study of the An. funestus strain, FUMOZ-R, which is characterised by upregulated P450 levels without any target site mutations, found that trans uthrin, which contains a poly uorobenzyl alcohol, was effective in the absence of the generic P450 inhibitor, piperonyl butoxide (PBO), whereas the other pyrethroids that contain common phenoxy benzyl moiety including cypermethrin, ß-cy uthrin, deltamethrin and permethrin were only effective when partnered with PBO [47]. This effect was associated with an inability of detoxifying enzymes to bind to the uncommon structure of trans uthrin. A similar observation was reported earlier from agriculture where an isogenic metabolic resistance strain isolated from a pyrethroidresistant eld population of Helicoverpa armigera showed signi cant cross-resistance between the pyrethroids characterised by having both the phenoxy benzyl and aromatic acid moieties whereas the substitution of the phenoxybenzyl group with a poly uorobenzyl group, as occurs in te uthrin, ben uthrin and trans uthrin, overcame most of this resistance [48]. These studies support the aim of identifying pyrethroids that are active against resistant populations when P450-mediated resistance plays a major role. In our study, bifenthrin diverged from the other pyrethroids in terms of both inhibition and depletion by P450s, but no susceptibility test data were available for resistance to bifenthrin in populations of African malaria vectors. Susceptibility test data were available for etofenprox and it was found to diverge from the more commonly deployed pyrethroids in terms of inhibition of An. gambiae and An. funestus P450s, and in terms of resistance in An. gambiae s.l. and Ae. aegypti populations.
The susceptibility test data from these populations show strong associations between resistance to the most commonly used pyrethroids (deltamethrin, λ-cyhalothrin, permethrin and α-cypermethrin), in agreement with the results for binding a nity and with earlier studies of spatio-temporal trends in An. gambiae s.l. [3,35]. The correlations in resistance among these pyrethroids, which were demonstrated in all the major African malaria vectors, suggest that if differences in resistance to these pyrethroids (as well as the less commonly deployed cy uthrin) are found using susceptibility tests conducted on a small number of eld samples of malaria vectors, further evidence should be obtained before any decision is made to switch between them.
Greater differentiation was found for resistance to bifenthrin in terms of both inhibition and depletion by P450s. The results for bifenthrin are interesting because they show that 1) this pyrethroid differs from the other pyrethroids in terms of P450 binding and metabolism, and 2) it may be less susceptible to common P450 enzymes. Bifenthrin is the active ingredient in one indoor residual spray (IRS), Bistar 10WP [2, 49], which is used in India. Bifenthrin IRS was trialled in Nigeria in 2006 and Zambia in 2011 [50][51][52] but has not been widely deployed in Africa where concerns about the duration of residual activity have been raised [52][53][54]. There are no eld data from susceptibility tests on African malaria vectors conducted using bifenthrin, presumably because this compound is rarely deployed and because there is no recommended diagnostic dose for use in a susceptibility test. One study of Anopheles sinensis in Korea collected blood-fed adults in the eld and exposed the F1 larvae to each of the pyrethroids considered by our study. They calculated resistance ratios using LC 50 values from a susceptible strain and found that the larvae were most susceptible to bifenthrin, cy uthrin and etofenprox, in that order, and least susceptible to permethrin [55]. Further evidence comes from studies of Aedes vectors, including three studies that tested bifenthrin [34]. One study in Mexico tested seven populations of Aedes aegypti with eight pyrethroids and compared the concentrations required for 50% knockdown (KC 50 ) and mortality (LC 50 ) to the same values obtained using a susceptible strain to give a resistance ratio (RR) [56]. Across the seven populations, resistance to deltamethrin, lambda-cyhalothrin, permethrin and α-cypermethrin were highly correlated (in terms of both RRKC 50 and RRLC 50 ), indicating the existence of strong crossresistance. However, the resistance values for bifenthrin were not correlated with any those for the other four compounds and the study concluded bifenthrin could be an alternative insecticide for Ae. aegypti in Mexico. Two independent studies in Thailand tested three Ae. aegypti and three Ae. albopictus populations, respectively, and calculated the diagnostic doses for each pyrethroid including bifenthrin using a susceptible strain [57,58]. In both instances, the population with the highest deltamethrin resistance also had the highest bifenthrin resistance, so no evidence for divergence in resistance was observed for these two species in Thailand. Given the known data noise in susceptibility test results, caution is needed when interpreting the results from a single study at a small number of sites. It is also worth noting that bifenthrin's relative immunity to depletion by CYP6M2, CYP6P3 and CYP6P9a described here was not found when tested previously [28]. Metabolism assays conducted by two earlier studies showed that CYP6M7, CYP6P9a and CYP6P9b from An. funestus metabolized bifenthrin (62%, 68% and 71% respectively) as well as permethrin, deltamethrin and λ-cyhalothrin (ranging from 46-81% depletion). Field tests for bifenthrin resistance in malaria vector populations are needed before we can reach a rm conclusion about whether bifenthrin should be recommended in situations where resistance to other pyrethroids has been found.
The analyses of binding a nity data and of eld data from malaria vector populations both show that resistance to etofenprox diverges, to a degree, from resistance to the more commonly deployed pyrethroids. This result is backed up by data from studies of resistance in Ae. aegypti. However, the depletion activity data suggests that etofenprox is more vulnerable to P450 metabolism and if resistance to this compound is found to be greater in malaria vector populations then a switch would not be advised. A trend for higher resistance to etofenprox was not seen in the data from malaria vector populations but was found in the data from Ae. aegypti populations, although caution is needed when interpreting differences found using susceptibility test data (particularly tests using diagnostic doses that have not been calibrated for Aedes species [34]). Etofenprox is the active ingredient in two WHO prequali ed products; a kit for insecticide-treated nets (Vectron 10EW) and an IRS formulation (Vectron20WP) [59]. The latter product is listed by the Global Fund, but etofenprox is not widely deployed in Africa and was last reported as the active ingredient used for IRS in 2012 in parts of Zambia [51,52].
We found some variation in the relationships among pyrethroids when different types of evidence were considered. In particular, the results for insecticide depletion were largely not repeated in the ndings for resistance in mosquito populations. The results for both insecticide inhibition and insecticide depletion depend on which enzymes are included in the activity tests. Seven P450s (three for the depletion analysis) were included here whereas at least 14 have been implicated in An. gambiae s.l. and An. funestus resistance so far [21, 22, 24-32, 46, 60-74] and many more in Aedes vectors [34]. It is also important to note that detoxi cation by P450s is not the only mechanism of resistance found in these vector species. Target site mutations are common in many of these species [9][10][11][12][13], upregulation of other detoxifying enzymes is also linked to pyrethroid resistance [75] and there is some evidence for cuticular thickening in resistant mosquitoes [76]. Upregulation of the GSTE2 gene is associated with resistance to both permethrin and deltamethrin, as well as DDT, in An. gambiae and An. coluzzii [71,73,77], An. funestus [29,72,75] and Ae. aegypti [78][79][80], and allele frequencies for target site mutations in the voltage-gated sodium channel gene, Vgsc, have been shown to be useful partial predictors of resistance in An. gambiae s.l. [35]. Thus, we would not expect the ndings from molecular studies of P450 activity alone to be exactly replicated in eld populations, except in instances where P450-mediated metabolic resistance dominates in a mosquito population.
The results for pyrethroid cross-resistance within individual species reported here match our knowledge of other mechanisms of resistance found in these species. Mutations in the Vgsc gene (kdr mutations) confer cross-resistance to pyrethroids and DDT, and are partial predictors of patterns of resistance to these compounds in the An. gambiae complex, but have not been found in An. funestus or other members of the An. funestus subgroup [3,[8][9][10][11][12][13][14]35]. In our study, correlations between pyrethroid and DDT resistance were found for members of the An. gambiae complex but not for the An. funestus subgroup or species. No correlations were found between pyrethroid resistance and resistance to the carbamates or organochlorines, underlining the nding that it is cross-resistance within the pyrethroids, as well as between the pyrethroids and DDT, that is most important. Some metabolic resistance mechanisms do confer cross-class resistance, e.g. between the pyrethroids and DDT and/or the carbamates [24,30,32,74], but the impact of these mechanisms within the array of resistance types that co-occur is more nuanced, and no cross-class resistance other than the aforementioned pyrethroid-DDT resistance in An. gambiae s.l. was detected here.
In conclusion, we have found that evidence for cross-resistance among pyrethroids predicted by SAR studies of metabolic resistance can be detected across African mosquito populations, as exempli ed by i) the close associations between the binding a nities of permethrin and deltamethrin to a range of anopheline P450s, ii) the close associations between depletion of permethrin and deltamethrin by these P450s, and iii) correlations in resistance to permethrin and deltamethrin in populations of An. arabiensis, An. coluzzii, An. gambiae and An. funestus. Importantly, populations with higher resistance to one of the pyrethroids studied here, which all contain the common structural motif of phenoxy benzyl alcohol coupled with a cyclopropane ring (the primary target for metabolic oxidation), are likely to have higher resistance to the others and these cross-resistance trends could be detected despite the noise in these susceptibility test data. It is unlikely that resistance to those pyrethroids most commonly deployed for malaria control diverges within vector populations and it would be unwise to switch between these compounds based on the results from a small number of susceptibility tests alone. There are, however, pyrethroids that are not commonly deployed that show greater potential for true divergence in resistance, such as bifenthrin and possibly etofenprox. It is worth noting that there are still signi cant correlations between resistance to etofenprox and resistance to the pyrethroids in common use, and that this is largely untested for bifenthrin. Systematic SAR analyses of these more structurally diverse pyrethroids are required to estimate the affect of structural diversity on pyrethroid resistance and these ndings need to be veri ed by studies of resistance in wild populations.

Declarations
Ethics approval and consent to participate Not applicable

Consent for publication
All authors provided their consent for publication.

Availability of data and material
All susceptibility test data analysed during this study are included in Additional File 4.

Competing interests
The authors con rm they have no competing interests.  Chemical structure of pyrethroid insecticides used for malaria vector control. The common scaffold of pyrethroids, boxed in red, was identi ed by searching 230 million compounds available in the ZINC database (https://zinc.docking.org).

Figure 1
Chemical structure of pyrethroid insecticides used for malaria vector control. The common scaffold of pyrethroids, boxed in red, was identi ed by searching 230 million compounds available in the ZINC database (https://zinc.docking.org).

Figure 2
Cluster analysis of functional activity data for six pyrethroids against P450s from African malaria vectors. (A) Inhibition data from the screening of six pyrethroids (scaffold structures indicated on the right of data panels) against a set of P450s are presented as a heat map. Target enzymes are arrayed along the x-axis, and each of the pyrethroids is arrayed along the y-axis. Colours indicate the inhibition potency of pyrethroids with an indicated variable scaffold for a designated target P450. Potent (hot) inhibitors are assigned a red colour, and weak or ineffective (cold) inhibitors are given a light green colour.
(B) Pyrethroid metabolism by the P450s most widely associated with resistance from An. gambiae (CYP6M2 and CYP6P3) and An. funestus (CYP6P9a) is clustered and presented as a heat map.
Pyrethroids susceptible to metabolism are assigned a red colour, and weak metabolism denoted light green. Dendrograms were obtained by hierarchical clustering. They indicate the degree of similarity as a function of the height of the lines connecting the pro les.

Figure 2
Cluster analysis of functional activity data for six pyrethroids against P450s from African malaria vectors. (A) Inhibition data from the screening of six pyrethroids (scaffold structures indicated on the right of data panels) against a set of P450s are presented as a heat map. Target enzymes are arrayed along the x-axis, and each of the pyrethroids is arrayed along the y-axis. Colours indicate the inhibition potency of pyrethroids with an indicated variable scaffold for a designated target P450. Potent (hot) inhibitors are assigned a red colour, and weak or ineffective (cold) inhibitors are given a light green colour.
(B) Pyrethroid metabolism by the P450s most widely associated with resistance from An. gambiae (CYP6M2 and CYP6P3) and An. funestus (CYP6P9a) is clustered and presented as a heat map.
Pyrethroids susceptible to metabolism are assigned a red colour, and weak metabolism denoted light green. Dendrograms were obtained by hierarchical clustering. They indicate the degree of similarity as a function of the height of the lines connecting the pro les.

Figure 3
The distributions of values for three example pyrethroid pairs (A) the most closely related pyrethroid pair in terms of resistance in wild mosquito populations (deltamethrin and λ-cyhalothrin), (B) a mid-ranked pair (permethrin and α-cypermethrin) and (C) the least closely related pair (λ-cyhalothrin and etofenprox).
Each point represents the results from a single An. gambiae s.l. sample that was subdivided between two susceptibility tests. The full results for every pair are shown in Figure S2, Additional File 2.

Figure 3
The distributions of values for three example pyrethroid pairs (A) the most closely related pyrethroid pair in terms of resistance in wild mosquito populations (deltamethrin and λ-cyhalothrin), (B) a mid-ranked pair (permethrin and α-cypermethrin) and (C) the least closely related pair (λ-cyhalothrin and etofenprox).
Each point represents the results from a single An. gambiae s.l. sample that was subdivided between two susceptibility tests. The full results for every pair are shown in Figure S2, Additional File 2.

Figure 4
Mean correlation coe cients for resistance to pairs of pyrethroids in An. gambiae s.l. 'alph.' is αcypermethrin, 'cy .' is cy uthrin, 'delt.' is deltamethrin, 'etof.' is etofenprox, 'lamb.' is λ-cyhalothrin and 'perm.' is permethrin. The bars represent the upper and lower 95% bootstap con dence intervals and the sample size for each pair is given below these bars.

Figure 4
Mean correlation coe cients for resistance to pairs of pyrethroids in An. gambiae s.l. 'alph.' is αcypermethrin, 'cy .' is cy uthrin, 'delt.' is deltamethrin, 'etof.' is etofenprox, 'lamb.' is λ-cyhalothrin and 'perm.' is permethrin. The bars represent the upper and lower 95% bootstap con dence intervals and the sample size for each pair is given below these bars.

Figure 5
Hierarchical relationships among pyrethroids measured using data on resistance in vectors and functional activity data. The dendrograms were constructed using (A) correlations in mortality across