Dramatic decreases of malaria transmission intensities in Ifakara, south-eastern Tanzania since early 2000s

Background Ongoing epidemiological transitions across Africa are particularly evident in fast-growing towns, such as Ifakara in the Kilombero valley, south-eastern Tanzania. This town and its environs (population ~ 70,000) historically experienced moderate to high malaria transmission, mediated mostly by Anopheles gambiae and Anopheles funestus. In early 2000s, malaria transmission [Plasmodium falciparum entomological inoculation rate (PfEIR)] was estimated at ~ 30 infectious bites/person/year (ib/p/yr). This study assessed the PfEIR after 15 years, during which there had been rapid urbanization and expanded use of insecticide-treated nets (ITNs). Methods Randomly-selected 110 households were sampled across Ifakara town and four adjacent wards. Mosquitoes were trapped nightly or monthly (June.2015–May.2016) using CDC-light-traps indoors, Suna® traps outdoors and human landing catches (HLC) indoors and outdoors. All Anopheles mosquitoes were morphologically identified and analysed by ELISA for Plasmodium circumsporozoite proteins. Mosquito blood meals were identified using ELISA, and sub-samples of An. gambiae and An. funestus examined by PCR to distinguish morphologically-similar siblings. Insecticide resistance was assessed using WHO-susceptibility assays, and some Anopheles were dissected to examine ovariole tracheoles for parity. Results After 3572 trap-nights, one Plasmodium-infected Anopheles was found (an An. funestus caught outdoors in Katindiuka-ward by HLC), resulting in overall PfEIR of 0.102 ib/p/yr. Nearly 80% of malaria vectors were from Katindiuka and Mlabani wards. Anopheles gambiae densities were higher outdoors (64%) than indoors (36%), but no such difference was observed for An. funestus. All An. funestus and 75% of An. gambiae dissected were parous. Anopheles gambiae complex consisted entirely of Anopheles arabiensis, while An. funestus included 84.2% An. funestus s.s., 4.5% Anopheles rivulorum, 1.4% Anopheles leesoni and 9.9% with unamplified-DNA. Anopheles gambiae were susceptible to bendiocarb and malathion, but resistant to pyrethroids, DDT and pirimiphos-methyl. Most houses had brick walls and/or iron roofs (> 90%), and 52% had screened windows. Conclusion Malaria transmission in Ifakara has decreased by > 99% since early-2000s, reaching levels nearly undetectable with current entomological methods. These declines are likely associated with ITNs use, urbanization and improved housing. Remaining risk is now mostly in peri-urban wards, but concerted efforts could further decrease local transmission. Parasitological surveys are required to assess actual prevalence, incidence and importation rates.


Background
Globally, human populations and settlements are undergoing rapid expansion, concurrent with speedy urbanization [1]. In the 1950s, less than one-third of global population lived in urban areas, but this had increased to 54% by 2014 and will likely exceed two-thirds by 2050 [1]. Most of these changes are expected to occur in developing countries in Africa and Asia [1,2]. In Tanzania, about 33% of people currently live in urban areas, compared to just 20% in 2000 [3,4].
Malaria burden has significantly declined worldwide over the past two decades, largely because of deliberate control efforts like scale-up of insecticide-treated nets (ITNs) and improved treatments [5] as well as due to socio-economic developments [6,7]. Urbanization is one of the major aspects of development that significantly impacts malaria transmission [7][8][9]. Processes of urbanization reduce malaria transmission primarily because urban environments lack suitable Anopheles breeding habitats [8][9][10]. Additionally, urban settings generally have better access of health services and improved housing as well as greater education and awareness than rural areas [7,9]. In a 2013 analysis, Tatem et al. concluded that ongoing urbanization in endemic regions would cause further declines in malaria transmission, and that these declines would be synergized by the deliberate scale-up of core malaria control tools [7].
Ifakara town, in the Kilombero river valley in the mostly agricultural Morogoro region, is among the fast growing towns in Tanzania. The town is surrounded by rain forests, wetlands and vast savannah lands, making it an attractive setting for both farmers and pastoralists. It is also a key trading hub in the region, mostly for farm and animal products. Towns such as Ifakara that appear in the middle of low-lying tropical floodplains present unique scenarios for assessing spatial and temporal variations of malaria transmission intensities [i.e. entomological inoculation rates (EIR), measured as number of infectious bites/person/year (ib/p/yr)]. Historical measures of EIR in Ifakara date back to 1965, when the town still had just one-fifth of its current population, and when Freyvogel and Kihaule conducted a small survey of Anopheles mosquitoes in the area [11]. Though parasite assessment techniques at the time were still fairly insensitive compared to options available today, Freyvogel and Kihaule reported monthly "apparent inoculation rates" for both Anopheles gambiae and Anopheles funestus, the two predominant malaria vectors at the time [11]. The "apparent inoculation rate" represented how frequently a single person was likely to get infected, and was estimated based on factors such as proportion of mosquitoes with sporozoites in their salivary glands upon dissection, gonotrophic cycles of the mosquitoes, and estimated daily biting rates (computed as ratio of total indoor mosquito catches and number of sleepers in collection room). In later years, there were multiple EIR assessments in the surrounding villages, including which reported 329 ib/p/yr from early 1990s in Namawala, a rural community 30 km away from the Ifakara town [12]. Because of the differences in assessment methods, it is difficult to directly compare these historical estimates against measures obtained after 1990s onwards.
In 2003, Drakeley et al. published findings of another entomological survey that assessed malaria transmission intensities in Ifakara, which at that time was still a periurban area surrounded by very high transmission villages in the Kilombero valley [13]. In the years preceding this report, the team had conducted fortnightly mosquito collections using CDC-light traps inside 32 randomly selected houses in Ifakara area. They used two different methods to estimate the PfEIR. First was the standard method, which involved calculations of the product of sporozoites rate (SR) and human-biting rate (BR), and the second method involved calculation of the ratio of sporozoites-positive mosquitoes to the number of nightly catches. Drakeley et al. estimated an EIR of 31 ib/p/yr using the standard method and 29 ib/p/yr using the alternative method [13].
This study aimed to assess malaria transmission intensities in Ifakara town and its adjacent wards, more than a decade after the assessment by Drakeley and colleagues. This new study employs similar mosquito sampling techniques, but with addition of indoor/outdoor comparisons of transmission risk, and an inclusion of an odour-bated trap to sample outdoor host-seeking vectors. Similar methods were also used for calculating and reporting PfEIR as used by Drakeley et al. [13]. The primary objective was to assess the extent to which malaria transmission in the study area had changed over the past 15 years, considering the rapid growth in population [2,14], urbanization trends [15] and the sustained use of effective malaria treatment as well as prevention measures such as ITNs [16][17][18][19].

Study site
This study was conducted in Ifakara town and its adjacent wards located in the Kilombero river valley, in south-eastern Tanzania (Fig. 1). The small but rapidlygrowing town lies at 8.1336° South and 36.855° East. It has an average altitude of 270 m above sea level, annual rainfall of 1200-1800 mm, annual temperature range between 20 and 32 °C, and relative humidity between 51 and 71%. The Ifakara town area is administratively divided into five wards, namely Ifakara Mjini (Ifakara town centre), Katindiuka, Lipangalala, Mlabani and Viwanja Sitini. In 2012, when the last national census was conducted, Ifakara town had a population of 55,956 people [20]. Given the estimated country population growth rate of 3.1%, the town currently has an estimated population of about 67,345 people as of 2018 [2,3]. In 2002 when Drakeley and colleagues had conducted the initial survey, Ifakara town had a population of 45,518 people [20], and between then and now the population has increased by nearly 50%. Ifakara Mjini and Viwanja Sitini, located on the western part of the town, are the most urban of the five wards, and the inhabitants are largely small business owners. These two wards make up of about 56% of the total population. On the other hand, Katindiuka and Mlabani, located on the eastern part of the town, are the most rural and least densely populated of the five wards and are surrounded by seasonal rice farms. This study was conducted in all the five wards ( Fig. 1) for a period of 12 months between June 2015 and May 2016.
Recent studies conducted in neighboring villages in the valley have shown that primary malaria vector species include Anopheles arabiensis and An. funestus, with the latter mediating > 80% of the ongoing transmission [21]. Other Anopheles mosquitoes found in the area are Anopheles coustani, Anopheles pharoensis, Anopheles squamosus, Anopheles ziemanni, Anopheles maculipalpis and Anopheles wellcomei. Malaria vector densities are highest during the rainy season, with peaks between March and May, though An. funestus densities may peak after the rains [22]. The major vector control intervention is long-lasting insecticide-treated nets (LLINs), mostly provided freely by the Government; the last such massdistribution of the nets having been in 2016.

Sampling procedures
Geo-referenced house-hold database obtained from Ifakara Health and Demographic Surveillance Systems (IHDSS) [23] covering the five wards was used to select the households for this study. A total of 22 households were randomly selected from each of the five wards, ensuring approximately equal representation from each Balozi (an administrative cluster consisting of 10-100 households). In each ward, two households were visited for 5 days every month for 12 months, and mosquito sampling was done both indoors and outdoors. The other 20 households were each visited twice every month, once for indoors and once for outdoors mosquito sampling. Altogether a total of 3600 trap-nights were planned in 110 households in the five wards for a period of 12 months between June 2015 and May 2016. The full sampling plan is shown in Table 1.

Mosquito collections indoors and outdoors
Indoor mosquito catches were done using standard Centres for Disease Control (CDC) light traps [24], set near human-occupied bed nets (Fig. 2). The light traps were suspended at the foot-end of the bed, approximately 1.5 m above ground [25]. The traps were set between 18:00 h and 06:00 h each sampling night in a room with at least one person sleeping under an intact bed net. Where no intact nets were available in the sampling room, one was provided freely to each consenting household. The traps were retrieved each morning starting 06:00 h; the field team taking approximately 30 min to collect all traps from all household stations. Outdoor mosquito collections were done using odourbaited Suna ® Traps [26] (Fig. 2). The Suna ® traps were suspended 30 cm above the ground, and 5-10 m from the selected house. The traps were baited using a combination of Ifakara synthetic lure [27] dispensed in pellets and CO 2 gas produced from yeast-molasses fermentation [28]. Like the CDC light traps, the Suna ® traps were set between 18:00 and 06:00 h and retrieved at 06:00 h each subsequent morning. To minimize any

Table 1 Illustration of complete sampling plan for the vector surveillance and estimation of malaria transmission in Ifakara town and its environs
The table shows the different trapping methods used, their frequency of use, position of use (indoors/outdoors), and whether the methods were used in sentinel or randomly selected households. The sampling plan does not include the additional collections done to assess parity direct competition between the indoor and outdoor traps, the sampling plan was such that the Suna ® and CDC light traps where never at the same house at any night, instead being placed in different nights. This specific dataset was, therefore, also not used for comparing indoor vs. outdoor biting rates. These CDC light trap and Suna ™ trap collections were done consistently in the 20 selected houses in each ward, totaling 100 households, each sampled monthly for indoors collections and monthly for outdoor collections.
To assess and compare indoor and outdoor biting preferences and nightly biting patterns, additional collections were done in 10 households using human landing catches method, working with consenting and trained adult male volunteers. In each of the 10 households, two adult male volunteers sat either indoors or outdoors and collected mosquitoes hourly, working for 45 min and resting for 15 min every hour. The collections were done between 18:00 and 06:00 h and each household was sampled for five nights every month for 12 months. In the 5-day period of HLC, the collections were done every other day, to allow volunteers a day of resting between the collections. All collected mosquitoes were kept in paper-cups and labeled to show date, hour of collection and household identification codes.
At the end of this survey, additional adult malaria vectors [An. gambiae sensu lato (s.l.) and An. funestus] were collected using CDC light traps, and used to assess parity rates and estimate physiological age of the mosquito populations in the area [29]. These additional collections were done for 1 month in 10 households per ward in three of the five study wards, i.e. Katindiuka, Mlabani, and Lipangalala.

Sorting and laboratory analysis of mosquito samples
Each morning, mosquitoes collected the previous night were killed using 70% ethanol and sorted by sex and taxa using morphological identification keys, primarily to separate Anopheles mosquitoes from other genera, and to identify major malaria vector groups [30,31]. Abdominal status of the female Anopheles mosquitoes was assessed and recorded as unfed, blood-fed or gravid. Anopheles mosquito samples were packed in labeled plastic microcentrifuge tubes (Eppendorf ® ) with silica gel as desiccant and sent to the Ifakara Health Institute laboratory for further analyses, including: (a) species identification using Polymerase Chain Reaction (PCR) [32,33], (b) detection of Plasmodium parasites in primary malaria vectors using Enzyme Linked Immunosorbent Assays (ELISA) [34], and (c) identification of mosquito blood meals using ELISA [35]. The specific methods used for the laboratory analyses were as follows:

PCR assays for species identification
DNA was first extracted from the mosquito hind legs. For An. gambiae s.l., PCR amplification was done for the species-specific nucleotide sequences of the ribosomal DNA (rDNA) intergenic spacer regions (IGS) in a 25 µl reaction volume of PCR mixture following methods first described by Scott et al. [33]. For An. funestus group mosquitoes, methods developed by Koekemoer et al. [32] were used to distinguish between up to five members of the An. funestus group, including the three that have been previously described in villages around the current study area, i.e. An. rivulorum, An. funestus sensu stricto (s.s.) and An. leesoni [21]. Here, amplification was done for the species-specific non-coding regions of the internally transcribed spacer 2 (ITS2) region on the rDNA. After amplification, the post-PCR amplicons for both the An. gambiae and An. funestus were analysed by electrophoresis in agarose gel stained with ethidium bromide. Visible DNA bands were photographed under ultraviolet light using Kodak Gel Logic 100 imaging system.

ELISA assays for parasite detection in mosquito salivary glands
To detect P. falciparum circumsporozoite proteins (CSP) in the salivary glands of the mosquitoes, ELISA assays were done on pooled samples separately for An. gambiae s.l. and An. funestus. Sample pooling was done by household ID, hour and date of collection and sampling method, and did not exceed 10 individual mosquitoes. Where fewer than 10 mosquitoes of one species complex were collected by a given method on a given hour or date at a given house, then the pool size was also less than 10. The optical density of post-ELISA lysate was measured at 405-414 nm after 45 min using ELISA plate readers. ELISA lysates for all postive samples were boiled at 100 °C for 15 min, to eliminate any false positives, usually associated with heat-labile protozoans in Anopheles mosquitoes, especially those with previous bovine blood-meals [36][37][38].

ELISA assays for identification of mosquito blood meal sources
Abdomens of all the blood-fed mosquitoes were screened to detect antigens for human, bovine, goat, dog and chicken blood antigens, as these are the most likely blood-meal sources in the area. Anti-IgG antibodies from each host were used to detect host antigen in the blood meal of the mosquito based on the method described by Beier et al. [35].

Assessment of insecticide susceptibility rates in malaria vectors
The World Health Organization (WHO) standard insecticide susceptibility assays [39] were performed on female adult mosquitoes raised from larvae sampled from two of the five study wards, i.e. Katindiuka and Viwanja Sitini. These assays were possible only with An. arabiensis, but not An. funestus because of the low densities and cryptic nature of An. funestus aquatic habitats in the area. Mosquito larvae were reared to adult stage under standard insectary conditions (temperature of 27 ± 2 °C and relative humidity of 70-90%) inside the Ifakara Health Institute mosquito laboratory, the Vect-orSphere. The larvae could feed on the mud and some algae brought together during collection but small quantities of Tetramin ® baby fish (Tetra, Melle, Germany) food was also added. Each morning pupae were collected and placed in plastic cups inside netting cages measuring 30 m × 30 m × 30 m. Upon emerging, adult mosquitoes were fed on 10% glucose solution prior to exposure to insecticides. Non-blood fed females of An. arabiensis (2-5 days old) were used in the assays, following the WHO test procedures [39]. The bioassays were done on 0.05% lambda cyhalothrin, 0.05% deltamethrin, 0.75% permethrin, 4% DDT, 0.1% bendiocarb, 0.25% pirimiphos-methyl and 5% malathion. Ifakara strain of An. gambiae s.s. was used as a reference susceptible strain. Four replicates of the exposure were done per insecticide, and each exposure used 25 mosquitoes. Mortality was recorded 24 h post-exposure.

Assessment of parity status
Additional adult malaria vectors were collected from inside the houses using miniature CDC light traps [24] to determine their physiological age [29]. These mosquitoes were collected from 10 households in three of the five wards: Katindiuka, Mlabani, and Lipangalala. Each morning, the mosquitoes were killed by refrigeration and immediately dissected under a stereo light microscope. The dissected ovaries were observed on 10× magnification under a compound microscope, to determine parity status by recording presence of stretched ovariole tracheoles (parous females) or coiled tracheolar skeins (nulliparous females), as detailed in the WHO Practical Manual for Entomology [29].

Household characterization
Additional information on the type of construction of each of the 110 houses was collected. This included type of construction material used for the roof and walls, materials used to cover the windows and doors, types of animals the houses kept and whether the houses had access to electricity.

Data processing, analyses and estimation of transmission intensities
Entomological data was recorded in standardized data forms used at Ifakara Health Institute. The data was quality-checked and double-entered into a Microsoft Excel database by two data entry clerks. The two data sets were compared to correct any errors, after which a single clean dataset was produced for further analysis.

Generalized linear mixed models for mosquito catch data
Data analysis was done using R statistical software version 3.3.2 [40]. Generalized linear mixed models (GLMM) [41] were used in analysis of mosquito densities by different study wards, mosquito trapping methods and position, i.e. indoors versus outdoors in cases where HLC had been used. In all analysis, study wards were treated as fixed factor while experimental day was treated as random term, controlling for seasonal variations. Initially, all the models were fitted using Poisson distribution models for count data with log-link functions, but due to poor-fitting and over dispersion, negative binomial distributions were fitted using glmmadmb (generalized linear mixed model automatic differentiation model builder) package [42], which allowed assessment of the zero inflations in the count data. Model testing was done using Akaike Information Criterion (AIC), by sequentially selecting models with lower AIC values. Graphs were produced using a ggplot2 package in R [43]. The mosquitoes caught using HLC were also analysed for nightly biting patterns and graphically presented to illustrate time of night when peak biting occurred.

Calculating PfEIR
Similar approaches as used by Drakeley et al. [13] were applied to calculate Plasmodium infection rates (sporozoites rates) and PfEIR. These analyses were initially done for each administrative ward, each month and each trap type, but was later pooled so that there was a single estimate of infection rate in all mosquitoes collected. Human-biting rates (HBR) were measured directly from human landing collections made indoors. The P. falciparum sporozoites rate (PfSR) was calculated by dividing the number of Plasmodium-positive mosquitoes by the total number of mosquitoes examined by ELISA, and this was expressed as a percentage. Plasmodium falciparum entomological inoculation rate (PfEIR) was then calculated considering the two methods previously used by Drakeley et al. [13] as follows: The first was the standard method of calculating PfEIR as a product of sporozoites rates and human-biting rates; i.e., C*(Number of sporozoites positive mosquitoes/ Number of mosquitoes tested) × (Number of mosquitoes collected/Number of trap nights) × 365 [44], where, C is the relative efficiency coefficient for the relationship between CDC-light trap catches and human landing catches. The second method involved calculation of the ratio of sporozoites positive mosquitoes and number of nightly catches; i.e., C*(no of sporozoites positive mosquitoes/number of trap nights) × 365 [13]. In this study all collected An. gambiae and An. funestus mosquitoes collected were analysed, therefore minimal variation in the two PfEIR estimates were expected. Although multiple trap types were used, PfEIR was only detectable using HLC, since the only sporozoite positive mosquito was collected using this method.
In the analyses by Drakeley et al., they had assumed that traps with extraordinarily large numbers of catches compared to the average, would be consisting mostly of young nulliparous mosquitoes, which are naturally sporozoites negative [45]. Such trap collections were therefore all considered "tested" but with negative CSP result. In this study however, mosquito samples were pooled in maximum pool size of 10 mosquitoes per assay, and all collected samples were analysed so that any biases associated with the said assumptions were effectively eliminated. In the Drakeley et al. analysis [13], a value of 1.605 (equivalent to 0.62 relative efficacy of CDC-light traps to HLC) was used based on previous studies by Charlwood et al. [45]. However, in recent estimates by Kaindoa et al. [21], values of 0.3 for An. funestus and 0.68 for An. gambiae s.l. were used based on 2004 estimates by Okumu et al. [46] for the relative efficiencies of CDClight trap catches relative to HLC. However, in a wideranging review of the estimates of CDC light trap and HLC catches across Africa, Briët et al. [47] concluded that there is no need to do these conversions, and that the CDC-Light trap estimate may be used as is. In this study therefore, PfEIR estimates were calculated without adjustments. The coefficient, C, in the PfEIR equations was fixed at 1, assuming all the trap types had similar efficiency across space and time. Nonetheless, any adjustments in human biting rates would be insignificant given no sporozoites positive mosquitoes were collected by any method other than HLC.

Assessing mosquito host preferences
Contributions of different blood meal sources in the malaria vectors were estimated as percentages of total mosquitoes assayed.

Analysis of spatial patterns in vector abundance
Mosquito densities and distribution across the study area was assessed for evidence of spatial associations and clustering. Spatial patterns of total nightly mosquito densities per house were examined using ArcGIS 10.4 spatial analyst tool (ESRI, USA). The Getis-Ord Gi* statistic [48,49] in Arc-GIS was used to identify households with significant clustering of high densities of the two main malaria vectors; An. arabiensis and An. funestus. Clusters depicting both the high vector density foci and low-vector density foci were identified and their statistical significance determined at Gi* P value ≤ 0.05, and Gi* Z score ≥ 1.96. Spatial relationships between households were assumed to be inversely related, such that houses far apart were more likely to differ in vector densities than households near one another. The analysis was done assuming Euclidean distances between neighboring mosquito sampling stations [48,49]. Indoor and outdoor densities of the mosquitoes were compared using the HLC method. Overall, An. gambiae densities were higher outdoors (64%) than indoors (36%), while only marginal differences were observed in An. funestus, 46% of which were indoors and 54% outdoors ( Table 2). Similar patterns were observed across the five wards (Fig. 3). The nightly biting patterns of mosquitoes collected by HLC in the ten sentinel houses in each ward are also shown in Fig. 4, which also shows that indoor collections were significantly lower than outdoor collections for An. arabiensis but not An. funestus throughout the night.

Spatial and temporal variations of malaria vector biting risk
Over the entire sampling period, 78.5% of all the primary malaria vectors were collected from two of the five wards, i.e. Katindiuka and Mlabani, which were the peri-urban parts. The remaining 21.5% were collected from the rest of the study area, i.e. Ifakara Mjini, Lipangalala and Viwanja Sitini, which were the more urbanized parts. In particular, Katindiuka ward, with houses constructed very close to mosquito breeding habitats (Fig. 5) had the higher densities of both An. arabiensis and An. funestus compared to the other wards (Table 3). Within Katindiuka ward itself, highest densities were in households closer to the adjacent rice fields on the eastern side of the ward, compared to households at the centre and on the western side ( Fig. 6). This distribution pattern was observed for both An. arabiensis and An. funestus, although An. funestus were caught at a much lower densities (Table 3). Densities of An. arabiensis were highest between January and May, peaking in February, while the An. funestus densities were highest between May and August, peaking in June (Fig. 4).

Sibling species of primary malaria vectors
A total of 7795 An. gambiae s.l. and 400 An. funestus s.l. mosquitoes were assayed in the laboratory by PCR for species ID, which represented all the mosquitoes of this species collected during the entire sampling season. The same mosquitoes were also assessed by ELISA (results shown elsewhere in this paper). All The unamplified DNA specimen may have been due to improper handling of the specimen prior to PCR, lack of appropriate primers in the PCR assays to detect actual species or incorrect morphological identification before the samples were analysed by PCR.

Mosquito blood meal sources
A total of 74 mosquitoes with blood meals in their abdomen were also assessed by ELISA to identify the source of their meals. These included 38 An. arabiensis and 36 An. funestus s.s. The blood-feeding proportions were relatively higher for An. funestus [i.e. 9% (36 out of 400) were blood-fed] compared to An. arabiensis for which 0.05% (38 out of 7795) were blood fed. Only human blood was identified in all 74 mosquitoes.

Insecticide susceptibility test results
The insecticide susceptibility test results are summarized in Table 4. Based on the WHO guidelines used here [39], the An. arabiensis mosquitoes raised from field-collected larvae were found to be susceptible to the carbamate, bendiocarb and to the organophosphate, malathion, but they were resistant to all the pyrethroids tested, i.e. permethrin, deltamethrin, and lambda-cyhalothrin, as well as the organochloride, DDT. Surprisingly, the mosquitoes were also resistant to the other tested organophosphate, i.e. pirimiphos-methyl. This compound has not been used for vector control in this region but is a common agricultural pesticide, and has been used for IRS in northern Tanzania over past 4 years. These insecticide resistance patterns were observed in both Katindiuka and Viwanja Sitini wards, where the tested mosquitoes were obtained. Full susceptibility (100% mortality) was observed in the susceptible Ifakara strain of An. gambiae s.s., used here as control. Knock down resistance (kdr) assays were not assessed, but a previous study done in villages in the neighboring Ulanga district indicated absence of kdr mutation genes [50].

Parity rates of An. arabiensis and An. funestus mosquitoes
A total of 109 An. arabiensis and 4 An. funestus were dissected to determine their parity status. All of these   (Table 5).

Plasmodium infectious rates and malaria transmission intensities
All 8613 Anopheles mosquitoes were analysed for Plasmodium circumsporozoite protein (CSP) using ELISA. These included 7795 An. gambiae s.l. (all of which were also determined by PCR to be An. arabiensis) and 400 An. funestus (a majority of which were determined by PCR to be An. funestus s.s.). In these ELISA assay, all An. arabiensis were uninfected and only one An. funestus s.s. mosquito was found to be infected with Plasmodium. This infected specimen had been collected outdoors between 03:00 and 04:00 h in Katindiuka ward in August 2015. Coincidentally, this was also the ward with highest malaria vector densities (Table 3). Since all Anopheles collected in this survey were analysed for Plasmodium infection, PfEIR estimations by the standard and alternative methods [13] returned similar results (Table 6). Sporozoite-infected mosquitoes were detectable only by human landing catches, but not any of the other traps (Table 6). Overall PfEIR was 0.102 ib/p/yr.  Table 4 Susceptibility of wild An. arabiensis mosquitoes collected in two wards in the study area, i.e. Katindiuka and Viwanja Sitini, (Fig. 1) The assays were conducted using WHO 2013 guidelines for testing insecticide resistance in malaria vectors [39] In tests conducted on the susceptible An. gambiae s.s. (Ifakara strain) used as an additional control, full susceptibility (100% mortality) was observed

Table 5 Plasmodium infectious status of primary malaria vectors collected indoors and outdoors in Ifakara town and estimates of the entomological inoculation rates, as contributed by the vector species
Since all Anopheles collected in this survey were analysed for Plasmodium infection, no differences were expected between overall PfEIR estimations obtained by either the standard or the alternative method as described by Drakeley et al. [13] a In this

Association between household characteristics and malaria vector densities
Of the 110 houses observed, 94.5% were constructed with brick walls and 5.5% had mud walls. Likewise, 97.3% of the houses had metal roofs and 2.7% had thatched roof. The houses had an average of 5.3 windows, ranging from 2 windows in small houses to 17 windows in the largest houses. The average number of outlet doors was 1.7, ranging from 1 to 7 doors. Half of the windows across the study area were covered with netting screens (51.8%), while the rest had wood or metal on the windows (17.3%), bricks (16.4%) or cardboard and clothes (13.6%). Only one house had completely uncovered windows (0.09%). The type of window cover had an effect on indoor densities of the primary malaria vectors, as lower indoor densities were observed in houses with netting screens on the windows compared to houses with other types of covers (Table 7). All (100%) of households had at least one insecticidetreated net (ITN). It was also observed that 51.8% of the houses had electricity while 48.2% did not. Electricity was mostly in the Ifakara Mjini, Lipangalala and Viwanja Sitini wards. All the sampled houses in Katindiuka ward had no electricity, and only 7 of the 22 households in Mlabani had electricity. Most of the houses in the periurban ward of Katindiuka were also surrounded by small rice fields and water ponds, especially during the wet season while houses in the urban settings of Ifakara Mjini and Viwanja Sitini wards were generally surrounded by small businesses such as small shops, vegetable stalls and bodabodas (motorcycle taxis) (Fig. 5). Lower mosquito densities were observed in houses with electricity compared to houses without electricity (Table 7). A great deal of brick-making activities was also observed in the dry season which resulted in a lot of pits that may have provided adequate breeding habitats for mosquitoes. Lastly, only one household had cattle, but 49 of the households (44.5%) had chickens or ducks.

Discussion
Africa is the world's most rapidly urbanizing continent [1,2]. Currently about 40% of the African population resides in urban settings, but that proportion is expected to reach 56% by 2050 [1,2]. Malaria, like most other infectious diseases, is negatively impacted by urbanization [7][8][9][10], since urban settings are characterized by better access to health care services, better housing and reduced availability of larval habitats [8][9][10]. In Tanzania, the urban population is increasing at 5.4% per annum. Tanzanian urban population was estimated at 5.7% in 1967 [15], but it had increased to 22.6% in 2002 [4], 29.1% in 2012 [51] and is estimated to be 33% in 2018 [3].
This current study focused on one of the fast-growing towns in Tanzania (i.e. Ifakara), and where demographic, socio-economic and epidemiological transitions are strongly evident. Ifakara town, in the Kilombero river valley is surrounded by rain forests, wetlands, savannah lands making it an attractive setting for both farmers and pastoralists. More importantly, as a town in the middle of an expansive low-lying flood-plain, only eight degrees south of the Equator, the area presents a unique epidemiological profile with very high malaria transmission in surrounding areas.
Though there have been several entomological surveys in the villages across the Kilombero valley over the past three decades [12,[52][53][54], there have been limited investigations within the Ifakara town centre itself, or in its closely adjacent wards. The most recent examination of malaria transmission in the town area was completed more than 15 years ago by Drakeley and colleagues [13]. Over the years, since the study by Drakeley et al. in early 2000s, the area has experienced significant increases in population concurrent with urbanization, due to inmigration by small business owners, farmers and pastoralists (Mayor of the Ifakara Town Council-Personal Communication). This current study presents the first reassessment of transmission in this community in nearly two decades.
A year-long entomological surveillance of malaria vectors was conducted in the five administrative wards making up the Ifakara town, an area with a total population estimated at 70,000 people. The magnitude and biting patterns of main malaria vector species, the levels of ongoing transmission, as well as the key factors driving the transmission in the town were identified. A general observation was that densities of malaria vectors within the area was found to be much lower compared the vector densities usually observed in nearby rural villages on both sides of the Kilombero river [21,22]. Overall PfEIR was estimated to be 0.102 ib/p/yr, which was entirely driven by An. funestus and only detectable by HLC. This was more than 99% decrease from the PfEIR reported by Drakeley et al. in 2003 from the same area. The PfEIR estimates are also much lower than those recently observed by Kaindoa et al. in the neighbouring rural villages of Ulanga district (PfEIR = 18.45), about 20 km south of Ifakara town [21]. Some of the factors that could be attributed to this dramatic decline could be urbanization [7,8,10], the universal bed net coverage campaigns that started in 2004 [17,55], improved housing [56] as well as improved diagnosis and treatment [16]. More than 90% of the houses in this study area now have brick walls and metal roofs, more than 50% had screened windows, and all (100%) had at least one bed net (Finda et al., unpublished). Unfortunately, this study was limited by the lack of aerial maps for the study area from 15 years ago, which would have enabled more detailed comparison of housing developments in the area. Nonetheless, these improvements, coupled with the ongoing largescale use of long-lasting insecticide-treated nets (LLINs) and other malaria control interventions, as well as the changes in potential mosquito breeding habitats, it can be expected that malaria transmission should plummet even further.
The ease of detecting Plasmodium circumsporozoite proteins in mosquitoes using simple ELISA assays and the simplicity of calculating entomological inoculation rates (PfEIR) has made this approach fairly popular for assessing malaria endemicity and transmission intensity [44,57]. It requires only simple measures of mosquito biting rates and sporozoites positivity rates, and is therefore a more direct method of measuring transmission intensity compared to other measures such as parasite incidence or prevalence and has been used widely for assessing impact of vector control programs in Africa [57,58]. This method estimates the number of bites by infectious mosquitoes per person per unit time. It is the product of the "human biting rate"-the number of bites per person per day by vector mosquitoes-and the fraction of vector mosquitoes that are infectious [44]. However, despite its widespread use, there have also been concerns that PfEIR estimates may be inconsistent, and that there can be large variabilities in PfEIR estimates from the same area [57,58]. Such variations are heightened by the differences in methods of measurement [57,58] and the lack of clear relationship between PfEIR estimates and malaria parasite prevalence of incidence rates in localities with low transmission intensities [59,60]. Besides, at very low PfEIR ranges, available methods are imprecise and there are very slim chances of getting mosquitoes at a precise time when innoculation takes place [60]. Therefore, the very low PfEIR estimates observed here do not necessarily mean absense of local transmission, but rather the lack of effective measurement methods.
In this study, only the human landing catches method was able to detect sporozoites positive mosquitoes. On the contrary, neither the CDC light traps nor the Suna ® traps caught any infected mosquito. This suggests the need for much more sensitive tools and approaches for measuring human exposure to malaria parasites in conditions such as this, where parasite densities in mosquitoes have become too low. Nevertheless, it has also been argued that if similar methods are used to consistently estimate transmission over time, such as this study has done, and if adequate stratification is performed, then interventions that drive malaria transmission to PfEIR < 1, could be effective in achieving local elimination [61]. In this survey, the overall PfEIR of 0.102 ib/p/yr was indeed far below the threshold of 1 ib/p/yr, beyond which sustained local efforts could lead to complete disruption of local malaria transmission.
Going forward, improved measurements are therefore a critical component of the malaria elimination agenda [62]. It will thus be essential to deploy improved measurement methods that can assess both the burden and the transmission of malaria pathogens in such situations so as to support further efforts for elimination. It would also be important to couple such an entomological survey with a malaria parasite prevalence study, using methods capable of detecting low-level parasitaemia so as to more comprehensively understand the residual malaria epidemiology in the area.
Within the Ifakara town, the highest density of malaria vectors was observed in Katindiuka and Mlabani (Fig. 6) wards, which were also the most rural of the five wards. Households in Katindiuka and Mlabani wards were surrounded largely by rice paddies and water ponds (Fig. 5), and nearly all the houses in these wards did not have electricity. Brick-making activities were fairly common in these areas in the dry season which resulted in a lot of pits with standing water in the rainy season, which may have provided adequate breeding habitats for mosquitoes. Although nearly all of the households surveyed were made with bricks and metal roof, majority of the households in Mlabani and Katindiuka lacked electricity, hence they were in the dark through most of the night, which may also have provided a suitable environment for host seeking mosquitoes [63].
In  [22,52,54,64]. While the most abundant malaria vector species was found to be An. arabiensis, it was An. funestus that was found to drive all of the transmission in the town. Only 400 An. funestus mosquitoes were collected compared to 7795 An. arabiensis, yet the only Plasmodium infected mosquito was an An. funestus. Though only a single infected An. funestus might be considered too few to conclude on the dominance of the species, evidence from neighboring villages suggest this is most likely the case. Similar patterns of dominance were indeed observed by Kaindoa et al. [21], who showed An. funestus to carry more than 80% of all the malaria transmission in neighboring villages in the Kilombero Valley. All blood-fed mosquitoes were found to contain human blood, and this can be explained by the absence of big livestock such as cows and goats in the urban settings, hence humans were the most available source of blood meals for the vectors. This could suggest that householdbased interventions could still be effective in targeting malaria vectors in the Ifakara town. However, the actual blood-feeding proportions were relatively higher for An. funestus (i.e. 9% (36 out of 400) were blood-fed) compared to An. arabiensis for which 0.05% (38 out of 7795) were blood fed. This further demonstrates the importance of An. funestus.
Relatively high parity rates were observed; over three quarters of all the mosquitoes dissected were found to be parous, a fact that also emphasizes the need for interventions that prevent man-vector contact to limit local transmission. On the basis of parity rates and human blood index, both An. funestus and An. arabiensis can be considered important vectors in the area. However, based on actual contribution to PfEIR estimates, An. funestus is likely mediating most of the on-going residual transmission in both the wider valley and in Ifakara itself. In this study, An. funestus densities were also observed to be higher in the dry season, which is the time of the year that people are most relaxed with regards to protection against mosquitoes.
There was evidence of phenotypic resistance in An. arabiensis to permethrin, deltamethrin, lambda cyhalothrin, pirimiphos methyl and DDT. However, the same mosquito populations from the study villages were found 100% susceptible to bendiocarb and malathion. The most surprising aspect of this specific finding was that the organophosphate, pirimiphos methyl has not previously been used in the area for vector control. It is however widely used in agriculture (Matowo et al., pers.comm.), which could be the source of this resistance pressure exhibited here. Nonetheless, the resistance profile of the mosquitoes collected here is worrying as it signals that there is a very narrow set of insecticide options now available for malaria vector control. One minor limitation with this aspect of the study was that the resistance assays were conducted only in two of the five wards, primarily due to availability of Anopheles larvae at the time of the tests. These findings can however still be considered fairly representative of the whole study area, and concur with other studies that have been done in surrounding villages in the valley [21,50].
Finally, the major decreases of malaria transmission seen are indeed impressive but should not be taken as a sign of impending local elimination. Instead, it should be interpreted as evidence that local elimination is possible given multisectoral approaches that combine house improvement to other technologies such as LLINs and effective case management, with proper diagnostics and medicines. The authors propose a parasitological survey to enable assessment of actual malaria cases prevalence and incidence rates, and also to examine in greater detail the actual rates of importation. Households should be encouraged to continue use of their long-lasting insecticide treated nets and to visit health facilities urgently whenever they experience any fevers. Given the significance of house improvement, these efforts too, should be encouraged to further reduce exposure to malaria and other mosquito-borne infections.

Conclusion
Malaria transmission intensity in Ifakara town and its surrounding environment has declined by over 99% over the past 15 years, reaching levels nearly undetectable with current entomological methods. Over a total of 3572 trap-nights, only one Plasmodium-infected Anopheles was found, an An. funestus mosquito caught outdoors in the peri-urban ward of Katindiuka. The overall PfEIR of 0.102 ib/p/yr is far below the threshold of 1 ib/p/yr, beyond which sustained local efforts could lead to complete disruption of local malaria transmission. This decline is likely associated with urbanization, improved housing, insecticide-treated nets and improved case management. House designs have vastly improved in the area and more than 90% of houses currently have brick walls and/or metal roofs, and more than half have screened windows. The remaining risk of Anopheles biting and malaria transmission in the valley is now mostly localized in the more rural ward adjacent to the town centre indicating that urbanization does indeed play a role in the control and possible elimination of malaria. On the basis of parity rates and human blood index, both An. funestus and An. arabiensis can be considered important vectors in the area. However, based on actual contribution to PfEIR estimates, An. funestus is likely mediating most of the on-going residual transmission in both the wider valley and in Ifakara itself, despite occurring in comparatively lower densities than An. arabiensis. Insecticide resistance has increased significantly, particularly against pyrethroids, organophosphate, pirimiphosmethyl and the organochloride, DDT, a situation which will clearly limit insecticidal options for malaria prevention. It is therefore possible that new cases reported in this area are now likely arising from infections outside the area than from within. It is also likely that concerted efforts could cause further decreases, possibly achieving complete disruption of local transmission in this area. Future surveys of malaria should deploy improved approaches for measuring transmission and parasite infection reservoirs in areas with such low transmission intensities. The authors specifically propose a parasitological survey to assess actual malaria prevalence and incidence rates, and estimate actual proportions of local and imported cases in the area.