Study area and selection of households
Data was collected in July and August 2019 in Nchelenge District, Zambia (Fig. 1). This region has three seasons: a rainy season from December to April, a cold and dry season from May to August and a hot and dry season from September to November. . Nchelenge district stretches along Lake Mweru and has several streams and tributaries that persist through the dry season, creating an expanse of swampy mosquito breeding grounds.
Twenty-four households were selected for this study: twelve located in 3 km from the lake (lakeside and inland, respectively). Sampled households had to have at least one animal enclosure outside the main bedding structure, and inland households had to be located within 0.6 km of a main watercourse.
During the study visits, household coordinates were uploaded into ArcGIS Pro (ESRI, Redlands, CA, USA). Flows were previously mapped and classified and the distance to the nearest main stream was calculated using the “near” tool in ArcGIS Pro (ESRI, Redlands, CA, USA).
Three lakeside and three inland households were sampled each day, and a Latin square method was used to rotate trap placement in each sampled household. . Mosquitoes were collected using CDC miniature light traps (John W. Hock Co., Gainesville) from 4:00 p.m. to 10:00 p.m. The head of household received a watch at the time of installation and was instructed to light the trap at 4:00 p.m. and tie the collection bag at 10:00 p.m. The traps were hung between 1.5 and 1.8m above the ground, either in a household’s indoor living room, an outdoor space where people gather in the evening, or outside next to a animal enclosure. Other than the glowing light from the trap and people or animals that spent time near the trap, no other lures were used. Household characteristics and human behavior surveys were conducted at the time of registration and additionally for each trap collection. The head of household answered all registration questions, and if he was not present for the follow-up questionnaire, the spouse or eldest answered.
Captured mosquitoes were killed by freezing, morphologically identified using Gillies and Coetzee , and stored individually in a microcentrifuge tube over silica gel to dry out in the field. The desiccated samples were returned to the Johns Hopkins Bloomberg School of Public Health (BSPH) in Baltimore, Maryland, USA, where each sample was divided into head/thorax and abdomen. Prior to fractionation, 17 samples identified morphologically as A.gibbinsi were sent to the University of the Witwatersrand to confirm morphological identity.
DNA from female Anopheles mosquito abdomens was crushed in lysis buffer using a Qiagen TissueLyser II (Qiagen, Hilden, Germany) and extracted using a DNA extraction method automated with a QIACube HT (Qiagen, Hilden, Germany) at Purdue University . After DNA extraction, each specimen underwent a PCR test to amplify a fragment of the internal transcribed spacer region 2 (ITS2) of the nuclear genome as previously described. [27,28,29]. A representative random subset of 5.4% of A.gibbinsi samples were selected for Sanger sequencing of the ITS2 target and the Barcode of Life fragment of the cytochrome oxidase I (COI) gene as previously described [27,28,29]. Sequencing of samples returned to BSPH was performed at the Johns Hopkins Medical Institutions (JHMI) Synthesis and Sequencing Facility. Forward and reverse sequences were imported into Geneious Prime (version 2021.2.2, Biomatters, Ltd, Auckland, New Zealand, https://www.geneious.com), cut to remove primer reads and sequences from poor quality, and aligned to create a consensus sequence for each sample. Consensus sequences were compared to the National Center for Biotechnology Information (NCBI) database and reference samples using BLASTn, and final identifications were confirmed if they had >99% identity with a NCBI sequence. Data were submitted to NCBI’s GenBank and accession numbers were acquired for the ITS2 (OM459737-OM459768) and COI (OM456780-OM456806) sequences.
Host detection scans
A host DNA detection test has been created to determine host preference. Primers from Kent & Norris Izadpanah et al. and Kumar et al.  were combined in a multiplexed PCR assay to detect individual and mixed blood meals from human, cow, pig, dog, chicken, and goat DNA, yielding differential product sizes for each host animal ( Supplementary File 1: Table S1). Each 25 μL PCR reaction consisted of 1 × buffer, 1.0 mM dNTPs, 0.625 units of Taq polymerase (New England Biolabs #M0273S), 50 pmol of each primer, and 1.0 µL of extracted abdomen DNA (Supplementary File 1: Table S1). Thermal cycler conditions consisted of an initial 5 min denaturation at 95°C, followed by 35 cycles at 95°C for 30 s, 58°C for 30 s, and 72°C for 45 s. The last extension step was at 72°C for 5 min. 12.0 μL of product was then run on an agarose gel for visualization and host determination.
Detection of sporozoites
The head/thoraxes were homogenized with BSPH in a buffer of boiled casein and Nonidet P-40 . Circumsporozoite protein (CSP) ELISA assays were performed using BEI Resources Malaria Research and Reference Reagent Resource Center (MR4) controls and protocols to detect the presence of CSP from P. falciparum sporozoites [18, 28]. Samples were tested in duplicate pools of five mosquito homogenates for the first ELISA, then individually tested in duplicate if the pool was positive, as previously described. [3, 29, 33]. Specimens were considered ELISA positive if the absorbance of the well for the individual mosquito was twice the absorbance of a negative insectary control mosquito.
Risk factor analysis
Ten of 216 collections were excluded from the analysis because the battery failed on the trap while it was operating (n=6), or the head of household forgot to attach the collection bag, potentially allowing to anophelines to escape (n = 4). A univariate negative binomial generalized linear mixed-effects model was performed with the number of A.gibbinsi by trap as the output for all variables of interest using the glmer.nb function of the MASS package in R. For each model, a random intercept term at the household level was included to account for repeat visits Household. Multivariate models were also performed using the glmer.nb function of MASS with a random intercept term at the household level. The best model was selected using Akaike’s Information Criterion (AIC).