Ecological Suitability and Spatial Distribution of Five Anopheles Species in Amazonian Brazil

Abstract

Seventy-six sites characterized in Amazonian Brazil revealed distinct habitat diversification by examining the environmental factors associated with the distribution and abundance of five anopheline species (Diptera: Culicidae) in the subgenus Nyssorhynchus. These included three members of the Albitarsis Complex, Anopheles oryzalimnetes, Anopheles marajoara, Anopheles janconnae; Anopheles triannulatus, and Anopheles goeldii. Anopheles janconnae abundance had a positive correlation to water flow and a negative relationship to sun exposure. Abundance of An. oryzalimentes was associated with water chemistry. Anopheles goeldii larvae were abundant in shaded, more saline waters. Anopheles marajoara and An. triannulatus were negatively associated with available resources, although An. marajoara also showed several local correlations. These analyses suggest An. triannulatus is a habitat generalist, An. oryzalimentes and An. janconnae are specialists, and An. marajoara and An. goeldii could not be easily classified either way. Correlations described herein provide testable hypotheses for future research and identifying habitats for vector control.

Introduction

The larval aquatic stage is an important part of the mosquito life cycle and habitat requirements may strongly influence the distribution and abundance of many species involved in pathogen transmission, including malaria vectors.^1–3 Many environmental variables can have a direct or indirect effect on mosquito oviposition^1,4 and on larval distribution, density, and development.^5,6 Knowledge of the influence of habitat factors on larval production is critical for understanding the spatial and temporal distribution patterns of mosquito vector species and for planning and implementing appropriate larval control strategies.^7,8

Despite five decades of intensive control efforts, Brazil continues to report the greatest proportion of malaria cases in the Americas.^9,10 Within Brazil, the incidence of malaria is almost exclusively (99.8% of the total number of cases) restricted to the Amazon Region, where a number of combined factors favor disease transmission and impair the use of standard control procedures.^11,12 Approximately 100 species of Anopheles occur in the Neotropical Region¹³ with 29 species in Latin America confirmed or potential human malaria vectors.¹⁴ Among these are the regionally important malaria vectors Anopheles triannulatus s.l.^15,16 and Anopheles goeldii,^16,17 and the Albitarsis Complex, which includes both local and regionally important vectors.^18,19

Anopheles triannulatus (Neiva & Pinto) was described as a polymorphic taxon²⁰; more recently it has been recognized as a complex of at least three species.^21–23 Originally considered zoophilic and exophilic, recent studies showed anthropophilic feeding and biting activity both indoors and out,^24,25 raising concerns about its potential as a vector in some areas of Latin America.^22,26 Anopheles triannulatus has the widest known distribution of the Triannulatus Complex, from Nicaragua to Argentina, and has been described from permanent pools, slow moving river margins, ditches, and swamps with full or partial shade.²⁷

Anopheles goeldii (Rozeboom & Gabaldón), often mistakenly identified as Anopheles nuneztovari,²⁸ has been reaffirmed as a separate species occurring in the Amazon Basin, whereas the distribution of An. nuneztovari s.s. is mainly across Colombia and Venezuela.²⁹ These species are hypothesized to be sympatric in some regions of northern South America.³⁰ The vector status of An. goeldii, originally described as zoophilic and exophilic,³¹ warrants further assessment because of its very close genetic association with An. nuneztovari, an important malaria vector particularly in Colombia and Venezuela.³² Previously, An. nuneztovari (likely An. goeldii) has been described as a species readily able to colonize and dominate in altered environments³³ as a result of its ability to survive in a variety of conditions, such as sunlit or shaded sites, fresh, clear, still, or flowing water³⁴ and occasionally in turbid and polluted sites.³⁵

The Albitarsis complex consists of at least eight species and a new lineage^36,37; only some of these have been incriminated as malaria vectors. The medical importance of Anopheles oryzalimnetes (Wilkerson and Motoki), previously Anopheles albitarsis B, is largely unknown, although this species has been attracted to human bait.³⁶ Larvae are common in some southern Brazilian rice fields,³⁶ often occurring in the flooded paddies associated with the vegetative and reproductive stages of the rice cycle. Anopheles marajoara has been implicated as a locally and regionally important vector in the lowland rainforest of eastern Amazonian Brazil.^16,18 The immature stages have been found in agricultural ponds and sunlit marshy areas, with increased abundance caused by forest clearance and water pollution.¹⁸ Anopheles janconnae (previously An. albitarsis E) has been implicated in malaria transmission in the savannah of northern Roraima state, Brazil.^18,19 There are no habitat data for Anopheles janconnae immature stages.

In contrast to extensive work on larval ecology in Africa, few studies of mosquito larval habitats have been conducted in Latin America^38,39 and the Brazilian Amazon in particular.^40–42 Statistical analysis of environmental variables using a principal component analysis or canonical correspondence analysis can reveal distinct habitat diversification among species that share a geographic distribution.⁴³ Habitat generalist species may be less affected by ecological variation than those that are dependent on a particular habitat type.^44,45 To date, no study has conducted this level of inquiry on these five species of Anopheles. Here, we examine the distribution and abundance of five mostly sympatric species from 76 larval sites to assess their habitat specificity and to identify environmental covariates associated with particular species.

Materials and Methods

Mosquito collection.

Anopheline larvae were collected from 76 breeding sites (randomly selected bodies of water—streams, pools, etc.; Supplemental Table 1) using standard dippers (500 mL) and preserved in 100% ethanol. Collection protocol included a minimum of 10 dips to determine presence or absence of anophelines and was approved by the Brazilian Instituto Evandro Chagas, Belém, Pará state Ethical Committee (Processo 010382/2009-7 “Biologia dos vectores de malaria no Brasil: Genetica e Ecologia”). If anophelines were present, additional dips were taken for a total of 2 hrs/breeding site in 2009, and a revised duration of one hr/breeding site in 2010/2011. To control for multiple collectors and duration, the relative abundance of larvae present was calculated as catch-per-unit of effort (total abundance/collectors present × time).⁴⁶ Collections in Roraima state occurred in eight localities between Amajari (N03°29.511′ W60°55.565′) and Ecuador (N00°07.115° W60°33.271′), along roughly 700 km of RR-174 in 2009 and 2011, and five localities in Pará state between Mojú (S01°51.684′ W48°45.611′) and Marabá (S05°20.052′ W49°05.079′), along 560 km of PA-475/263/150 in 2010 (Figure 1). Between two and nine habitats were sampled within each of the 13 localities. Localities were selected based on malaria incidence records in each municipality, in each state, for the collection year. Field collections were initially planned to be conducted twice in both Brazilian states, alternating years, however because of logistical constraints, duplicate sampling in Pará, and some localities in Roraima could not be done.

Figure 1.

Map of South America indicating the positions of the 13 localities (A–M) and species distributions. Pie charts depict the presence of a particular species in a given locality (town where 2–9 sites were sampled, see Supplemental Table 1), not the proportion of each species.

Initial morphological assignments of fourth instar larvae to An. triannulatus and An. nuneztovari were made using the key of Deane and collaborators⁴⁷ and confirmed by species-specific ITS2 RFLP.⁴⁸ Because morphological keys do not reflect the recent differences noted between the Colombian and Venezuelan An. nuneztovari and the Brazilian An. goeldii, samples were sequenced for COI and ITS2 and compared with those from Calado and collaborators.²⁹ Morphologically identified fourth instar Albitarsis Complex samples⁴⁷ were also confirmed by the Zapata⁴⁸ protocol and identified at the species level (data not shown) using species-specific primers from Li and Wilkerson.⁴⁹ For An. janconnae, all specimens initially identified as members of the Albitarsis Complex from 2009 and a subset from 2011 (randomly selected to ensure less than a 2.5% error rate for species delimitation) were sequenced by the Applied Genomics Technology Core (Wadsworth Center) for the barcode region and compared with samples from Ruiz and collaborators.³⁷ Total genomic DNA was extracted using the DNAeasy tissue kit (Qiagen, Valencia, CA) and preserved at the Wadsworth Center at −80°C.

Habitat characterization.

To achieve greater diversity of larval habitats, sampling was done once for each breeding site. Sites were randomly selected bodies of water identified at dawn within the city/town limits and habitat type (ditch, pool, pond, etc.) was classified based on definitions in Rejmankova and collaborators.⁵⁰ Multiple site measurements were taken, corresponding to the four cardinal points to account for different positions of the sun and surrounding vegetation, and the average recorded. Surface water⁵¹ of breeding sites was sampled and analyzed for the following: 1) concentration of nitrates, 2) concentration of nitrites,⁵¹ 3) alkalinity, and 4) pH were estimated with Eco-Check 5-in-1 test strips (Forestry Suppliers Inc., Jackson, MS). Variables 5) temperature,⁵² 6) conductivity, and 7) salinity¹ were measured using the handheld ExTech Exstik 100 meter (Extech Instruments, Waltham, MA). Aquatic variable 8) turbidity,^51,53 was estimated using a LaMotte (Chestertown, MD) individual test kit. Variables 9) water movement,⁵⁰ 10) type and density of algae, 11) density of surrounding vegetation,¹ and 12) relative shade provided by surrounding environment were estimated qualitatively on a scale of 0–3. Finally, 13) emergent light (measurement of light penetrating the canopy) estimated with Lutron LX-102 luxmeter (Taipei, Taiwan), and 14) canopy coverage estimated using a spherical densitometer (Forestry Suppliers Inc., Jackson, MS).⁵¹ All environmental covariates were measured at the time of mosquito larval sampling.

Data analysis.

Student's t tests were used to compare continuous environmental factors between positive larval sites and those negative for all anopheline species. Subsequent analyses examined differences between positive and negative sites for each species individually. To reduce the likelihood of false positives,⁵⁴ multivariate analysis of variance (MANOVA) was used to compare the mean abundance of anopheline larvae between different continuous environmental variables and species. Univariate F-tests for each of the eight variables were examined as a measure of variable contribution.

To examine species and environment effects from a multivariate perspective, a principal components analysis (PCA) was conducted to determine the number of meaningful factors, using the eigenvalue-minus-one criterion (JMP, Version 7; SAS software, Version 8; SAS Institute, Inc., Cary, NC). Among the 13 environmental variables considered, pH and alkalinity exhibited strong correlation, and the less variable of the pair (pH) was removed prior to analysis to ameliorate collinearity.⁵⁵ The concentration of nitrates and nitrites were excluded from further analyses based on consistent null readings. For each species and breeding site, we evaluated PCA scores as explanatory variables of relative abundance using linear and negative binomial regressions (R-2.9.2, with the MASS package, R Foundation for Statistical Computing, Vienna, Austria), and Pearson correlation analysis to determine significance (P < 0.05).⁵⁶ In addition, a canonical correspondence analysis (JMP, Version 7) was carried out as a secondary measure of the pattern of variation in community composition accounted for by the environmental variables.⁵⁷

All sites, regardless of species presence or absence, were included in the analysis. A species was considered a habitat generalist based on the following criteria: 1) large proportion of positive larval pools across both Brazilian states, and 2) no correlations with the environmental covariates.^45,58 To correct for the possibility of underestimation of distribution we examined the relationship of abundance and environment across all sites assuming both Brazilian states may possess adequate habitat and ecological conditions for each species.⁵⁶ To evaluate spatial differences and the possibility of local adaptation, PCA analysis was further stratified by locality. Only coefficients significant for one or more models are included and discussed, with coefficients greater than one excluded because of the questionable accuracy of the calculation.⁵⁹

Results

Distribution.

Anopheline species were found in 68 of the 76 sites examined (89.5%). Thirty-six (47.4%) sites had two or more species, and 18 (23.7%) had a single species (Table 1). Fourteen sites were positive for other anopheline species. Anopheles triannulatus and An. goeldii were found in the highest number of sites (39.5% and 36.8%, respectively), consistent with their wide distribution across Latin America.^20,30 Albitarsis Complex members had lower distributions overall, but had relatively high incidences (36–55%, Table 1) when evaluated among sites within their known distributions.

Table 1.

Proportion of positive larval habitats across all sampled sites and among sites within the known distributions of each species^*

	N	Number of habitats
		Across all sampled sites (%)	Across known range (%)	Shared among species	Unique to species
Total anopheline larvae	7,154	68 (89.50%)	–	–	–
An. janconnae	260	22 (28.95%)	55.00	17 (22.37%)	5 (6.58%)
An. triannulatus	203	29 (38.16%)	–	24 (31.58%)	6 (7.89%)
An. goeldii	230	30 (39.47%)	–	26 (34.21%)	4 (5.26%)
An. oryzalimnetes	64	13 (17.10%)	36.11	12 (15.79%)	1 (1.32%)
An. marajoara	238	14 (18.42%)	38.89	12 (15.79%)	2 (2.63%)

^*All cases except overall anopheline density counts are based on fourth instars only and therefore include the proportion positive for two or more identified species and those unique to an individual species across all samples sites. N = number of samples.

Variation in larval presence among species was compared by the Student t test and subsequently MANOVA. A number of associations were detected by the Student t test (Table 2), however MANOVA only supported the significant contribution of temperature and lux (Table 3). One-way MANOVA revealed no significant multivariate main effects for species (Table 3).

Table 2.

Two-tailed t test assuming unequal variances comparing mean variation between positive and negative larval pools per species^*

Species	Environ. variable	t Stat	P value
All anophelines	Vegetation	2.87	0.02
An. janconnae	pH	−2.21	0.03
	Alkalinity	−2.68	0.00
	Temperature	3.74	0.00
	Salinity	−2.20	0.03
An. oryzalimnetes	Temperature	−3.36	0.00
	Shade	−4.18	0.00
	Lux	2.52	0.02
An. marajoara	Temperature	−2.50	0.01
	Shade	−3.69	0.00
	Algal density	2.08	0.05

^*Only significant differences are shown of the 14 environmental variables examined. Positive and negative t-statistics indicate the direction of the association with presence of the species (e.g., the negative coefficient for pH indicates that species were more likely to be present in lower pH sites).

Table 3.

Abundance of species in different habitats and individual variable contribution^*

Species/variable	df	Wilks' λ	F
An. janconnae	63	0.929	0.602
Lux	1	–	3.120†
An. triannulatus	63	0.892	0.950
Temperature	1	–	2.783†
Lux	1	–	4.231‡
An. goeldii	63	0.978	0.179
An. oryzalimnetes	63	0.886	1.017
Lux	1	–	4.206‡
An. marajoara	63	0.922	0.670

^*Multivariate analysis of variance (MANOVA) is used to test the null hypothesis that the means vectors of all populations are equal against the alternative hypothesis that at least one mean vector is not equal to the others. Wilks' λ is a likelihood ratio test statistic commonly used in MANOVA. Significance:

^†P ≤ 0.05;

^‡P ≤ 0.01;

Five principal components based on 11 environmental variables were extracted using the eigenvalue-one criterion and supported by scree plot, explaining ∼73% of the total variation across all sites (Table 4). The first principal component (PC) described a strong gradient of water chemistry with high loadings on conductivity and salinity with a partial contribution of alkalinity. The second PC described sun exposure, with high loads of lux, temperature, turbidity, and a partial negative contribution of alkalinity. The third PC represents available resources with high loads on vegetation cover and algal density. The fourth PC characterized differences in protection from surrounding environment encompassing shade and canopy coverage. Orthogonal rotation was unable to resolve the various contributions of alkalinity and turbidity.⁵⁹ The smaller magnitude loading can be excluded so the PC is approximated by linear combination of the only remaining variable.⁶⁰ As this procedure is potentially misleading, both factor loadings were retained.

Table 4.

Principal components analysis—loadings of 11 environmental variables recorded in the field, using Eigenvalue-minus-one criterion^*

	Factor 1	Factor 2	Factor 3	Factor 4	Factor 5
Alkalinity	0.435	−0.540	−0.232	−0.277	−0.050
Temp	−0.297	0.597	−0.202	−0.264	−0.129
Conductivity	0.917	0.054	0.023	0.038	0.004
Salinity	0.865	−0.073	0.034	−0.126	0.013
Turbidity	0.303	0.469	−0.274	0.244	−0.547
Shade	−0.140	−0.148	−0.110	0.837	−0.123
Movement	0.083	0.079	−0.140	0.174	0.880
Algal Density	−0.147	−0.179	0.806	−0.120	0.030
Vegetation	0.186	0.045	0.856	0.106	−0.093
Lux	0.180	0.811	−0.067	−0.228	0.078
Canopy	0.026	−0.081	0.093	0.709	0.341

^*Items in bold depict significant loadings.

Analysis of all 76 sites consistently indicated that negative binomial regression analysis was a more suitable model for correlations between species compared with a linear regression model based on Akaike information criterion (AIC) scores. Negative binomial regression is based on discrete probabilities and assumes all points are isolated, indicating that individual habitats have no measure of bias toward the probability of species success in another habitat site.⁶¹ In contrast, linear regression examines the median of Y,⁶¹ suggesting it would be more appropriate in situations where the entire ecological range has been surveyed. Although species were present in a variety of water bodies, several exhibited slight specificity or preference based on varying principal components (Table 5). Negative binomial regression analysis showed moderate to high correlations of species abundance with four of the five principal components. Correlation coefficients indicated that An. janconnae was negatively associated with sun exposure (PCc = −0.75, P ≤ 0.01) and positively with water flow (PCc = 0.52, P ≤ 0.05). A single correlation was seen between species and abundance of An. triannulatus and An. goeldii, where the former was negatively associated with available resources (PCc = −0.48, P ≤ 0.01) and the latter positively correlated to water chemistry (PCc = 0.64, P ≤ 0.001). Abundance of An. oryzalimnetes is positively associated with water chemistry (PCc = 0.60, P ≤ 0.001) and negatively with available resources (PCc = −0.79, P ≤ 0.01). Abundance and available resources may be negative in An. marajoara (PCc = −0.68, P ≤ 0.05).

Table 5.

Comparison of linear and negative binomial regression analysis of larval abundance for all data points and significant principal components^*

Species	PC	Lin. regression		Neg. binomial
		Coefficient	AIC	Coefficient	AIC
An. janconnae	2	−0.99†	426.37	−0.75	164.67
	5	0.31	435.39	0.52‡	177.39
An. goeldii	1	0.43	417.83	0.64†	187.98
	4	0.33	419.33	0.48§	193.48
An. triannulatus	3	−0.29	330.53	−0.48§	187.19
An. oryzalimnetes	1	0.22§	256.93	0.60†	108.51
	3	−0.16	259.63	−0.79§	109.5
An. marajoara	3	−0.68‡	435.91	−1.63†	148.54

^*Significance:

^‡P ≤ 0.05;

^§P ≤ 0.01;

^†P ≤ 0.001. Akaike information criterion (AIC), a measure of the relative goodness of fit of a statistical model, is included, with the preferred model (bold) having the minimum AIC value.

Stratification by locality revealed additional spatial correlations, suggestive of local adaptation within individual species. Correlations at the locality level were just as likely to be explained by a linear regression model as they were by negative binomial (Table 6): excluding all correlations contraindicated by the AIC value, 18 relationships were evident at the locality level. Anopheles oryzalimnetes (PCc = 0.70, P ≤ 0.01), An. triannulatus (PCc = 0.45, P ≤ 0.05), and to a lesser extent An. goeldii (PCc = 0.18, P ≤ 0.05), exhibited positive correlations with water chemistry in Tucurui, whereas An. marajoara (PCc = 0.34, P ≤ 0.01) exhibited a similar trend in Jacunda. Sun exposure continued to exhibit a negative correlation among An. goeldii (PCc = −0.56, P ≤ 0.05), An. marajoara (PCc = −0.89, P ≤ 0.001), and An. oryzalimnetes (PCc = −0.24, P ≤ 0.05) in Alto Alegre, Jacunda, and Maraba, respectively. Interestingly, An. janconnae exhibited a positive correlation in Boa Vista (PCc = 0.71, P ≤ 0.05), with subsequent negative associations in Mucajai (PCc = −0.64, P ≤ 0.001) and Ecuador (PCc = −0.09, P ≤ 0.001), possibly suggesting some local adaptation to the grasslands that predominate in northern Roraima. Anopheles oryzalimnetes also exhibited a positive association with solar exposure in Moju (PCc = 0.18, P ≤ 0.01). Anopheles goeldii exhibited positive associations with available resources in Moju (PCc = 0.39, P ≤ 0.01) and Jacunda (PCc = 0.20, P ≤ 0.05). Environmental protection exhibited positive correlations with abundance of An. oryzalimnetes (PCc = 0.64, P ≤ 0.05), An. marajoara (PCc = 0.86, P ≤ 0.05), and An. triannulatus (PCc = 0.79, P ≤ 0.05) in Jacunda, but a negative association with An. janconnae (PCc = −0.55, P ≤ 0.05) in Petrolina do Norte. Finally, with respect to water flow, An. goeldii exhibited local adaptation, whereas in the metropolitan Boa Vista An. goeldii appeared to favor standing water sources (PCc = −0.76, P ≤ 0.05), and in underdeveloped Martins Pereira the trend suggested an inclination toward open environments (PCc = 0.40, P ≤ 0.001).

Table 6.

Comparison of linear and negative binomial regression analysis of larval abundance and significant principal components for data points stratified by locality^*

Site	PC	Species	Lin. regression		Neg. binomial
			Coefficient	AIC	Coefficient	AIC
B	2	An. goeldii	−0.56†	20.99	−1.10‡	12.93
C	2	An. janconnae	0.66†	27.66	0.71‡	24.61
	5	An. goeldii	−0.76†	24.36	−1.17†	21.89
E	2	An. janconnae	−2.47†	40.72	−0.64§	36.4
	5	An. janconnae	2.46	43.92	0.77†	39.29
G	5	An. goeldii	0.40§	−17.1	19.37	7.45
H	2	An. janconnae	−0.09‡	−14.3	−15.74	6.78
I	3	An. goeldii	−0.14	38.26	0.39‡	33.32
J	1	An. oryzalimnetes	0.28‡	−12.3	78.88	7.17
	2	An. oryzalimnetes	0.18‡	−14.2	−33.53	7.17
K	1	An. triannulatus	0.52	20.75	0.45†	19.95
	1	An. goeldii	0.18†	6.22	20.22	8
	1	An. oryzalimnetes	0.69	25.69	0.70‡	19.65
L	1	An. marajoara	2.18	58.21	0.34‡	86.41
	2	An. marajoara	−5.44	56.25	−0.89§	75.51
	3	An. goeldii	0.20†	6.30	39.24	8
	3	An. marajoara	−2.57	57.49	−0.44§	81.57
	4	An. triannulatus	0.72‡	23.47	0.79‡	22.6
	4	An. oryzalimnetes	0.73	30.90	0.64†	27.85
	4	An. marajoara	2	58.05	0.86†	50.78
M	2	An. oryzalimnetes	−0.24†	−0.47	−2.65	9.37

^*See Figure 1 for localities.

^†P ≤ 0.05.

^‡P ≤ 0.01.

^§P ≤ 0.001. Akaike information criterion (AIC), a measure of the relative goodness of fit of a statistical model, is included, with the preferred model (bold) having the minimum AIC value. PC = principal component.

A graphical representation of the contribution of each environmental variable to the species distribution is shown in Figure 2, where the first two canonical axes explain 89% of the total variance in species distribution. The environmental variable that contributed most to species distribution was temperature (F = 4.93, P = 0.001), followed by shade (F = 3.59, P = 0.009). The remaining environmental components had a lower impact on species variation. Anopheles triannulatus and, to a lesser extent An. goeldii, mapped to the central part of the first axis, which reflects their widespread distribution along this ecological gradient. The first axis, which explains 79.435% of total species variation, was related to environmental variables that reflected from left to right: An. marajoara and An. oryzalimnetes in the lowland rainforest environment prominent in Para state, to An. janconnae in the savannah region of Roraima. The second axis, which explained 9.65% of the total variance, was related to turbidity and conductivity. These associations are more difficult to interpret with respect to the various localities/states, because they do not represent any unambiguous environmental gradient.⁵⁷

Figure 2.

Ordination biplot diagram showing the dispersion of the five anopheline species and the 6 contributing variables on the first two canonical axes of a Canonical Correspondence Analysis. Black dots indicate the average centroid for each mosquito species: JAN, An. janconnae; TRI, An. triannulatus; GOE, An. goeldii; ORY, An. oryzalimetes; MAR, An. marajoara. The contribution of each canonical variable to total species variance is depicted in parentheses.

Discussion

Mosquitoes breed in a wide variety of terrestrial water accumulations, but individual species appear to choose particular types of habitats.^6,62 The larvae of some species seem able to tolerate only a narrow ecological range, whereas larvae of other widely distributed species may be more broadly adapted.^63,64 Of the five species of anophelines examined, An. triannulatus appears the most likely habitat generalist as it demonstrates both widespread distribution and little if any environmental constraints. On the other hand, high numbers of correlations coupled with restricted distribution records indicate An. oryzalimentes and An. janconnae are more specialized. The classifications of An. goeldii and An. marajoara are more complicated because the different analyses supported sometimes opposite trends relative to specialization and associated environmental factors.

The broad distribution of An. triannulatus, lack of correlation among t tests, MANOVA and canonical correlation analysis (CCA), and minimal associations at the microgeographic scale (locality level) all support habitat generalist status. The ability of dipterans to readily adapt to and use a broad variety of ecological niches has shaped the diversity of Neotropical anophelines⁶⁵ with the more generalized populations more ecologically heterogeneous.⁶⁶ Thus, the genetic heterogeneity seen in the Triannulatus Complex may be explained, in part, by environmental heterogeneity,⁶⁷ including local or microgeographic adaption, past and present ecological barriers, and/or demographic events.⁴⁴ Further studies that include larval ecology of populations of An. triannulatus across its distribution may help to elucidate the extent of population/lineage divergence and generalist status.

The consistent correlation between An. oryzalimnetes abundance and water chemistry may indicate a local adaptation to more saline waters around the Amazon and on the eastern perimeter of Brazil, which corresponds to its proposed distribution.³⁷ The ability to tolerate salinity varies among species, and both salinity and conductivity have been correlated with the presence or development quality (size, rate of emergence) of Anopheles larvae in Africa¹ and Latin America.^68,69 Bradley⁷⁰ pointed out that only 5% of all extant species within the family Culicidae are capable of surviving in salt water, with at least five independent origins of salt tolerance.⁷¹ Given the large degree of specialization, An. oryzalimnetes may be more susceptible to population control through careful drainage, soil management, compaction, and erosion that can minimize salinity increases.⁷²

Anopheles janconnae abundance was repeatedly correlated with sun exposure and water movement. Close association with moving water may contribute to geographic distribution by aiding gene flow. Even though An. janconnae is more specialized, it has been identified recently in Para state,³⁷ albeit in low numbers, perhaps indicating a potential to expand in certain habitat types. This could ultimately contribute to an increase in the regional incidence of malaria as An. janconnae is a vector of both Plasmodium vivax and Plasmodium falciparum.⁷³

Finding a consistent positive correlation to water chemistry across the stratified data sets supports the hypothesis by Scarpassa and Conn that, unlike sister taxa An. nuneztovari, An. goeldii distribution is restricted to the Amazon Basin.³⁰ The ability to colonize and even dominate in altered or temporary environments³³ is most evident among ditch habitats, given the strong negative correlation between abundance and PC5, where increasing tree cover and water movement suggest minimal landscape modification (irrigation, farming, etc.). Though a large number of correlations were revealed through PCA, the broad distribution of An. goeldii, in addition to having shown no correlations among t tests and CCA, suggest this species is less specialized. Though the vector status of An. goeldii is unknown, the ability to adapt to a broader habitat range coupled with earlier peaks in seasonal abundance⁷⁴ may increase the importance of this species at a local and regional level.

Inconsistencies between PCA and CCA results make it difficult to interpret the generalist/specialist status of An. marajoara, although larvae appear less specialized than those of sibling species An. janconnae and An. oryzalimnetes. The co-occurrence of An. marajoara and An. oryzalimnetes larvae at multiple sites suggests either strong divergent natural selection or some other mechanism to counteract gene flow.⁷⁵ Additional studies regarding the ecology and behavior could clarify whether the co-occurrence is truly sympatric or whether niche divergences have led to microallopatry.

Where there are no identifiable geographic barriers limiting the distribution of species, a variety of biotic (e.g., competition or predation) and abiotic (e.g., resource availability, physiological limits, demography, or climate) factors may constrain further range expansion.^76,77 A single sampling scheme is limited and does not take seasonal variation into consideration; thus may not be representative of the full spectrum of habitats or habitat conditions.⁷⁸ Although no generalizations can be made, single sampled ecology work can contribute to current literature, by laying the foundation for further studies. Conclusions made in this study provide testable hypotheses for future work.^79–82 For example, the recorded distributions of the five species indicate many of these are co-occurring, with multiple sampled sites positive for more than two species. Co-occurring populations can offer important insight into differences in behavior, seasonality, divergence, and niche partitioning. Correlations at a local level can indicate population differences in response to urban development, host availability, and distribution limits. Understanding the associations of habitat gradients with populations of mosquitoes may be important for understanding disease transmission in different kinds of habitats and may aid in planning more effective vector control strategies.⁸³