An. gambiae s.s. pupae were obtained from a colony maintained at the Ifakara Health Research & Development Centre (IHRDC). This colony was established in 1996 from individuals collected from Njage village in Kilombero District, Tanzania. In the insectary, pupae were collected and held individually in plastic tubes (4.9 × 2.9 cm) that were covered by netting. Pupae were left overnight for emergence, and the sex of the emerged adults identified visually the following day. Thereafter, adults of the same sex and age were pooled in groups of 50 and held in netted cages (20 × 20 × 15 cm). Mosquitoes were classified as being '0' days old on the day of their emergence. All females were at least two days old before being used in experiments described below. While in cages, mosquitoes were fed on a 10% glucose solution that was administered by placing a soaked cotton wool pad on top of the cage.

Virgin males isolated at emergence were left for periods of 1–20 days. On each day of age, the gonads of at least 10 males were dissected and observations made as described below (following Mahmood & Reisen 28), in order to assess how the morphology of their reproductive organs changed through time. In order to test whether morphological traits could predict whether mating had occurred, and whether any observed relationships between male age and morphology were altered by mating, experiments were conducted to provide a sample of males who had copulated before observation. On each day of mating experiments, 50 male An. gambiae of the same age were placed in front of a window prior to dusk (age groups ranged from 1–18 days old). Activity inside the cages was observed to begin approximately 45 minutes before dusk. Once males began to swarm, 20 virgin females (2–4 days old) were introduced into the cage. Two observers monitored activity in the cage with the assistance of a red light. When pairs of mosquitoes in copula were observed, they were siphoned from the cage and transferred together into a separate holding cup. The following morning, both males and females captured in copula were dissected. First, the spermatheca of the female was dissected to confirm whether she had been inseminated 27. Then the male partner of the inseminated female was dissected, and the morphological features of their reproductive system compared to those of male virgins of the same age. A further sample of males caught in copula were left for a number of days after mating (2–5) before being dissected in order to estimate the duration of any observable morphological changes associated with mating. Males were not observed to mate on the day of their emergence, so these experiments were restricted to males that were 1 day post-emergence or older.

After participating in experiments, live males were incapacitated by shaking them in a holding cup. Once knocked out, they were placed in a drop of phosphate buffered saline (PBS) and examined under a dissecting microscope (50×). Their reproductive system was removed by using needles to pull out the last segment of their abdomen. Slow excision removed the whole male reproductive system including testes, accessory glands, and ejaculatory duct. Subsequent observations were made under a compound microscope (100×) as this more clearly revealed the ultrastructure of male reproductive organs; including the testes and accessory glands. Each testis holds spermatocysts that store immature spermatozoa, and a sperm reservoir that holds mature spermatozoa. Spermatocysts are visible from the posterior tip of the testis (Figure 1). As spermatocysts mature, they break down and release spermatozoa into the sperm reservoir, which occupies the anterior section of the testis (Figures 1 and 2). During copulation, spermatozoa leave the sperm reservoir and travel via the vas efferentia to the seminal vesicle and then the ejaculatory duct before entering the female. Under a compound microscope, spermatocysts appear as circular cellular structures (Figure 2) whilst the sperm reservoir is characterized by non-packaged, thread-like striations.

Mahmood and Reisen 2829 demonstrated that it was possible to predict the age and mating status of An. stephensi and An. culicifacies from changes in their testicular and accessory gland morphology. This study concentrated on observation of the same traits, specifically: (1) the number of spermatocysts in the testes, (2) the proportion of testes filled by the sperm reservoir, and (3) the presence or absence of a translucent border (defined as a clear area) surrounding the accessory gland (Figure 3).

Before subjecting the data to statistical analysis, some general trends between male age and their morphology were recognized. These general trends were used to develop a qualitative model for age-grading based on the values of each of the three key morphological traits (Table 1). In developing this model, the aim was to test whether male age could be predicted by following a general 'rule-of-thumb', or whether more intensive quantitative statistical analyses were required for accuracy. A qualitative model for mating status could not be developed as no clear morphological associations were observed prior to statistical analysis.

General linear models were performed to investigate the relationship between the known age of males and: (1) the number of spermatocysts in individual testes, and (2) the proportion of the testes occupied by the sperm reservoir. Before analysis, data collected as proportions (% of testes occupied by the sperm reservoir) were logit transformed to improve their fit to a normal distribution. The additional explanatory variable of male mating status (virgin or mated the night before) was included in these models to examine whether it influenced any apparent relationship between male age and spermatocyst number, or sperm reservoir. The initial maximal statistical models for both spermatocyst number and proportional size of the sperm reservoir included the main effects of male age (days), mating status and their interaction (age × mating status). Additionally, logistic regression was used to investigate relationships between the presence of a clear area around the accessory glands and male age and mating history. The presence of a clear area was treated as a binary response variable ('0' if absent, '1' if present), with male age, mating status, and their interaction, being treated as independent explanatory variables. In all analyses, non-significant terms were sequentially eliminated to yield the minimum statistically significant model for each trait.

After these individual investigations of each trait, logistic regression was used to examine the ability of all three traits to predict male life-history when considered in combination. In this analysis, the outcome variable of 'age' was predicted as one of two age categories of 'young' (≤ 4 days) or 'old' (> 4 days). There are two reasons why male age was predicted as a category and not as discreet days. First, preliminary observations of the association between traits and age showed considerable overlap of values between days; suggesting a categorical rather than continuous approach could be more successful. Secondly, ideally a predictive model of male age could provide comparable demographic information as the morphology-based model available for female An. gambiae, that can distinguish them into groups of 'nulliparous' (assumed to be approximately ≤ 4 days old) and 'parous' (assumed to be approximately > 4 days old). By categorizing males into age groups similar to those obtained for females, this method can provide comparable estimates of survival. In this analysis, male age group was treated as a dependent variable, and spermatocyst number, proportion of testes occupied by the sperm reservoir (logit transformed) and the presence of a clear area were treated as independent variables. Binary logistic regression was used to test whether these three traits could predict whether a male was 'young' or 'old'. Four different model-fitting techniques were tested to examine which yielded the highest success in age prediction (Table 2). These models differed in whether they treated explanatory variables as continuous or categorical (spermatocyst number and relative size of the sperm reservoir), and the manner in which these explanatory variables were pooled into categories. In the first model, spermatocyst number and percent sperm reservoir were fit as continuous variables. In the following three models, values of spermatocyst number and the relative size of the sperm reservoir were pooled into categories and treated as fixed factors. In the second model, the median value of spermatocyst number (3) and the relative size of the sperm reservoir (60%) were used as cut-off points to divide data into two categories (median model). In the third, these explanatory variables were split into four categories; with the quartile values for each trait serving as cut-off points between groups (quartile model). In the final model, data for these two explanatory variables were split into four categories on the basis of visual observation for natural breakpoints in their relationships with age (Figures 4 &5, breakpoint model). In testing each of these models, all three main effects and their two-way interactions with the presence of a clear area were fit as explanatory variables. Parameter estimates obtained from the minimal statistical model for each model were used to obtain an equation for predicting age category on the basis of the observed values of each morphological feature.

Logistic regression was also applied to examine the association between mating status (treated as a dependent variable) and all three morphological traits. As with age, four different quantitative models were fit to explain variation in mating status (Table 2). Similar to age, all three morphological traits and their interaction were combined in a logistic regression model and used to predict the probability that a male was a virgin or had mated. This model predicted male mating status as a probability ranging between 0 and 1. Males whose mating status was predicted to be less than 0.5 were classified as virgins, and those assigned a value higher or equal to 0.5 were designated as having mated. Unless otherwise stated, error estimates accompanying means represent one standard error. All statistical analyses were performed using SAS version 8.2.

In a blind trial, laboratory-reared males whose age and mating status were known only by one insectary worker were given to another researcher to dissect. The observer recorded the morphological features of these unknown males, and entered them as independent variables into the age and mating status predictive models described above. The accuracy of predictions of both age and mating status were compared to the actual values (as revealed post hoc). The predictions of male age obtained from the statistical model were compared with those obtained from the qualitative 'rule of thumb' (Table 1).

A total of 454 An. gambiae males of known age were dissected. As male An. gambiae aged, the number of spermatocysts in their testes decreased steadily (Figure 4, F1,451 = 562.61, P < 0.01). At emergence, males had an average of 3.68 ± 0.45 (Range = 3 – 5) spermatocysts. By day 8, the mean number of spermatocysts in each testis fell below 1 (Day8 = 0.57, ± 0.57), and none were observed in unmated males in the 14 – 17 day old age group. Interestingly, a small number of spermatocysts reappeared in mated males of 18 days old (mean = 0.50, ± = 0.15). In addition to these age effects, spermatocyst number was also influenced by mating status (F 1,451= 12.09, P < 0.01), with virgin males having slightly fewer than males that had mated once (Figure 4). The rate at which spermatocysts declined with age, however, was not influenced by mating status (age × mating status: F1,450 = 2.77, P = 0.10). Conversely, the proportion of the testes occupied by the sperm reservoir increased with male age, and at a faster rate in virgins than in mated males (age × mating status: F1,450 = 45.81, P < 0.01, Figure 5). On the first day after emergence, an average of 26% (± 0.32) of the testis was occupied by the sperm reservoir. In virgin males, the sperm reservoir expanded to occupy all of the testes by day 9 of adult life, whereas once-mated males required approximately 14 days for full expansion of their sperm reservoir (Figure 5). The final morphological trait, the presence of a clear area surrounding the accessory glands, also changed with male age. The probability of having a clear area generally decreased with age, though at a substantially faster rate in virgins than in mated males (age × mating status: χ²₁ = 15.85, P < 0.01, Figure 6). The majority of males in the 0–1 day old age group had a clear area surrounding their accessory glands, regardless of whether they had mated or not. By day three of adult life and onwards, none of the virgin males had a clear area, whereas some mated males still exhibited this feature up until 13 days of age. Thus males who had mated the night before dissection were more likely to have a clear area than virgins. However, this mating-related morphological change was lost through time. Sequential analysis of a cohort of 47 males who mated at 5 days of age indicated that 85% (11/13) had a clear area on the first day after mating, but only 47%, 29% and finally 0.9% displayed this trait 2, 3 and 4 days after mating respectively.

To assess the ability of morphological features to predict male age grade ('young' or 'old'), all three morphological variables (logit-transformed sperm reservoir, spermatocyst number, and the presence of a clear area) and their interactions with the clear area were combined in a logistic regression model. All of the four statistical models that were fit explained a substantial proportion of the variation in male age group (r² = 0.73–0.79), and could classify the age-grade of males in the original data set with an overall accuracy of 88.1–90.5% (Table 3). Of these four, the breakpoints model had the highest predictive success when applied to the independent blind trial data, and thus was selected as the optimum model. The remainder of this section focuses on the performance of this model. Using breakpoints, the minimal statistical model of male age grade included only the main effects of the relative size of the sperm reservoir (χ²₁ = 115.96, P < 0.01) and the presence of a clear area (χ²₁= 33.72, P < 0.01). When fit to the original data set, this model was able to correctly categorize 89.8 % of male mosquitoes that were ≤ 4 days old, and 90.8% of mosquitoes that were > 4 days old. The accuracy of this model decreased slightly when applied to the blind trial data, although it still correctly identified 81% of males in the young age group, and 95% of those in the old group (Table 3). This quantitative model was substantially more accurate in classifying male age than was qualitative model based on generalities deduced from observation (Table 4).

The accuracy with which statistical models based on morphological features (Table 2) could correctly predict the mating status of mosquitoes in the original and blind trial was also examined. In contrast to the analysis of age, these statistical models could explain only a small proportion of the variation in male mating history (r² = 0.13–0.17, P < 0.05 in all cases, Table 5). All models suffered from the same failing; an inability to correctly identify virgin males (81–100% failure rate). Indeed, the median, quartile and breakpoints models were unable to correctly identify any virgin males, and the best performing model of mating history (continuous model) could only identify 18% of them. In this model, mating status was significantly related to the presence of a clear area (χ²₁ = 21.36, P = 0.01), the proportion of the testes occupied by the sperm reservoir (χ²₁ = 8.08, P < 0.04), and the interaction between the presence of a clear area and the relative size of the sperm reservoir (χ²₁ = 9.81, P < 0.01). Despite the statistical significance of these terms, their ability of predict male mating success was very weak.