The findings indicate variation in the preference for feeding upon humans rather than cattle between two populations of An. arabiensis, one in urban Dar es Salaam, and another in rural Kilombero. However, these observations become more interesting and seem to suggest an effect of urban environments on both An. arabiensis and An. gambiae, when compared with preceding studies that also measured host preference (rather than host choice) through carefully controlled experiments. Similar to all comparable previous studies of An. arabiensis in rural areas [21, 26], the rural Tanzanian An. arabiensis population studied here had no strong preference for humans or cattle (Figure 3). Indeed these results compared particularly well with those of Meza et al. 2019  (Figure 3), which also used juvenile cattle with relatively low biomass and therefore attractiveness . However, in urban Dar es Salaam, An. arabiensis appeared to exhibit a strong preference for cattle over humans that were significantly different from the same species in rural Kilombero over approximately the same period (Figures 2 and 3). Also unexpectedly, An. gambiae s.s. collected in Dar es Salaam, lacked its notoriously strong preference for humans when compared with equivalent indices derived from a previous study of the same species in rural Tanzania . It, therefore, appears that both siblings’ species have a stronger preference for non-human hosts in the city of Dar es Salaam than would reasonably be expected (Figure 3).
Position of Figure 3
The flexible feeding behaviour exhibited by the An. arabiensis in rural Kilombero is consistent with that reported by previous studies from the same setting  and beyond  that employed similarly direct, experimentally-controlled, host attack preference measurements using different capture methods. It is also reassuring that fitting host preference and availability models to historical blood meal host choice data for the same species across entire villages [21, 48] yields similar indirect estimates, indicating only a slight preference for cattle (Figure 3) even though such natural herds are dominated by larger adult cattle that may be reasonably expected to be more attractive . Indeed the Torr et al.  direct host preference experiments using electric grids similar to Meza et al , which also used larger adult cattle, yield almost identical estimates to these indirectly inferred from modelling analyses, confirming a slight preference of rural An. arabiensis for fully-grown cattle over humans. Such biologically and methodologically plausible triangulation of results from such different studies with such different methods reassuringly suggests that the experimental approach applied here, including the first use of ITT-C  for experimental host preference studies, provides reliable and readily comparable indices of host preference. It, therefore, appears reasonable to interpret the finding that An. arabiensis had a stronger preference for cattle in urban Dar es Salaam than in rural Kilombero or indeed any previous studies population of the same species (Figure 3) at face value.
It is also telling that a similar, and perhaps more surprising, the pattern was observed for the notoriously anthropophagic [6, 27, 49, 50] An. gambiae when compared with a previous study of the same species in a rural Tanzanian context (Figure 3). The lack of a clear preference for humans over cattle by An gambiae s.s. in this contemporary urban context contrasts starkly with historical records from Segera, only 258 km away from Dar es Salaam in rural Tanzania . This unusually flexible feeding behaviour for An. gambiae in Dar es Salaam may also contribute to the persistence of this species, even while it was virtually eliminated from other nearby ecological settings [11, 16], following widespread scale-up of LLIN use [11-13, 16]. The increasingly widespread use of LLINs  and high coverage of house window screening in urban Dar es salaam , which limit safe access of mosquitoes to human blood, may have forced this species to develop a strategy which enables them to evade personal target protective interventions for humans by exploiting animal blood whenever they can find it.
Urban Dar es Salaam has generally fewer cattle than Kilombero and probably most other rural settings, so it is certainly interesting that An. arabiensis now appears to have a stronger preference for feeding on cattle and perhaps other non-human hosts that were not assessed here. It will be important to investigate whether the two populations are genetically distinct or not [52-54]. This may be especially important following the recent surge of interest in genetic manipulation approaches to malaria vector control . Regardless of the underlying basis for this apparent trend towards greater zoophagy in both vector species in Dar es Salaam, on the one hand, it will limit the impacts of existing malaria vector control interventions like LLINs and mosquito-proofed window screening, while on the other it may provide opportunities for complementary approaches like veterinary insecticide treatments for livestock [6, 7, 56-59].
While this study was quite limited in terms of scale and sample size, it does raise some important questions that merit consideration beyond Dar es Salaam and Tanzania. Urbanization is known to influence host preferences in other mosquito taxa , and similar effects to those reported here might also occur in other African settings where An. gambiae and An. arabiensis continue to mediate malaria transmission, despite widespread use of LLINs . Indeed, it is notable just how few experimentally controlled host preference studies could be found to populate figure 3, despite the key role that this trait plays in malaria transmission and control. We, therefore, strongly encourage more widespread measurement of mosquito feeding preferences across a diversity of ecological settings, especially through routine programmatic surveillance , to help inform the selection and evaluation of complementary vector control interventions, ideally in an ecologically stratified manner.