Diversity and complexity of auditory representation in the hearing systems of Aedes aegypti mosquitoes

Authors: Takuro S. Ohashi, Yifeng Y. J. Xu, Shunsuke Shigaki, Yukiko Nakamura, Tai-Ting Lee, YuMin M. Loh, Emi Mishiro-Sato, Daniel F. Eberl, Matthew P. Su, Azusa Kamikouchi

Abstract

Although low-frequency sounds have been reported to stimulate dispersal responses in male and female mosquitoes, only males show attraction to sound. Male attraction to female flight tones is important during courtship; however, groups of males show diverse responses to acoustic stimuli, suggesting that auditory processing can vary drastically between the sexes and individual males. To investigate diversity in auditory representation within and between the sexes, we used molecular and functional analyses to explore Aedes aegypti mosquito auditory processing. We identified shared and dimorphic neurons connecting mosquito ears to the brains’ primary auditory processing center. Calcium imaging from this brain region facilitated definition of multiple neuronal clusters based on auditory stimulation responses. More clusters with greater complexity were identified in males than females, with these clusters highly differentiated among males. Transcriptomic and proteomic analyses found enrichment of ciliary-related factors in male ears compared to females, potentially underlying sexual dimorphisms in hearing systems.

INTRODUCTION

Male attraction to the sounds of flying conspecific females (phonotaxis) plays an important role in the mating behaviors of the yellow fever mosquito Aedes aegypti (A. aegypti), a major vector of various diseases (1, 2). A. aegypti mating can occur in a male-dominated swarm that facilitates encounters between males and females, promoting copulation success (3). Males locate females by detecting their sexually dimorphic wing beat frequencies (WBFs), with males producing WBFs several hundred hertz higher than females (male, ~600 to 900 Hz and female, ~400 to 650 Hz across a temperature range of ~20° to 30°C) (4).

Female mosquitoes, on the other hand, do not appear to show attraction to male flight sounds (5, 6), although they have been reported to show dispersal behaviors in response to tones of 100 to 300 Hz, potentially mimicking predator avoidance (7, 8). Male mosquitoes have also been reported to show similar repulsion behaviors when exposed to these tones, suggesting potential conservation of some auditory behaviors between the sexes (9). Males may therefore be able to process a greater range of auditory stimulation than females to facilitate both attractive and repulsive phonotactic behaviors.

Substantial diversity of responses to a range of sounds has been observed between individual male mosquitoes; although most males in a group are attracted to sounds mimicking average female WBFs under specific environmental conditions, individual males can be attracted to sounds up to ±200 Hz of these tones (10). This diversity has been suggested to be mediated by the wide range of possible mosquito flight tones as female detection relies on the interaction of male and females flight tones in the male ear (5). Auditory processing may therefore be highly diverse not only between males and females but also between different males.

Mosquito hearing behaviors are mediated via their sensitive hearing systems. The hearing organ is located in the antennae (11) and comprises two (i) the flagellum, a sound sail that vibrates with mechanical stimulation, and (ii) the Johnston’s organ (JO), the site of auditory transduction. The male flagellum is densely covered with fibrillae, whereas females have only sparse fibrillae distributions (12). The mechanosensory neurons in JO (denoted JO neurons) are bipolar neurons, each having a single distal dendrite connected to the base of flagellum and an axon projecting to the brain (3, 13, 14). Male A. aegypti JOs contain >15,000 JO neurons, making them the largest chordotonal organs reported in insects, whereas female JOs contain half the male number of neurons (13, 15, 16).

Functional sexual dimorphisms also exist, with frequency matching between the female flight tone and the male peak flagellar vibration (“mechanical tuning”), both 400 to 500 Hz, facilitating male detection of the female’s flight tone (17). The peak oscillation frequency of the female flagellum, on the other hand is ~225 Hz (17–19), potentially linked to predator detection rather than localization of conspecific males. Furthermore, male (but not female) mosquitoes can demonstrate large amplitude, seemingly monofrequent flagellar oscillations referred to as self-sustained oscillations (SSOs). SSOs entrain only to pure tones close to female WBFs (17), suggesting that SSOs in males play a key role in the acoustic processing of female flight tones. SSOs rely, at least in part, on the dynein-tubulin motor machinery, which underlies active hearing (14).

The peak sensitivity of electrical responses recorded from the axon bundle of male JO neurons (“electrical tuning”) is distinct to the aforementioned mechanical tuning of the flagellum (10, 20). Females have not yet been reported to show differences in mechanical and electrical tuning (21). This mismatch in males has led researchers to suggest that male JO neurons respond most sensitively not to female flight tones but to linear interactions between male and female WBFs derived from nonlinear modules in the male ear, referred to as distortion products (DPs) (5, 14, 17). Several of these DPs, including the quadratic distortion product (QDP; ~350 to 400 Hz) (5, 10), are similar in frequency to the peak of the male electrical response (5, 22), supporting the hypothesis of a DP-based acoustic communication system (5, 17). This DP-based system potentially explains the wide array of responses observed between individual males as the sizable variation in male and female WBFs, up to ±50 Hz of mean values (10), likely leads to substantial differences in the resulting DPs that can be produced and perceived.

Male hearing systems must thus be highly sensitive and complex to process the diverse array of potential stimuli, including both repulsive and attractive sounds. However, investigations into this complexity have largely been restricted to peripheral anatomical and functional assays, with the molecular bases of male auditory processing rarely being tested (23). Furthermore, tests of central differences in mosquito auditory processing have not yet been conducted. In mosquitoes, JO neurons have been reported to project to a part of the antennal lobe (AL) denoted as the JO-Center (JO-C) (24); however, a recent study found that A. aegypti JO neurons mainly project to the antennal mechanosensory and motor center (AMMC), as for Drosophila melanogaster (16). In several mosquito species, including A. aegypti, the size of the male AMMC is approximately double that of the female (25), suggesting that the acoustic processing properties of the AMMC may also vary between sexes and the potential for substantially greater complexity in male auditory representation exists.

In this study, we sought to investigate the diversity and complexity of central processing of auditory information in A. aegypti mosquitoes. We first compared the projection patterns of JO neuron axons in the brains of male and female A. aegypti and found that male JO neurons had more branches compared to females. Next, we compared the auditory response properties of the AMMC using calcium imaging, identifying sex-specific frequency properties and male-specific functional diversification. Clustering analyses identified more male-specific (four) than female-specific clusters (two), with only one cluster shared, suggesting a higher degree of complexity in males. We also show that the low-frequency tones that elicit peak responses from female AMMCs can elicit negative responses in specific male neurons. Development of a quadratic state-space model suggested that males exhibiting SSOs can exhibit antiresonance, dampening male flagellar vibrations at specific frequencies similar to those inhibiting specific male neurons.

As SSOs have been suggested to be powered by dynein-tubulin motor machinery, we observed the ciliary ultrastructure of JO neurons using electron microscopy but found no obvious differences in dynein arm structures between the sexes. Integration of transcriptomic and proteomic analyses of male and female pedicels identified different sets of cilium-related genes expressed between the sexes with potentially distinct kinetic properties, as well as potential upstream modulatory factors. These molecular differences may underlie the sexual dimorphic nature of mosquito ear vibrations.

RESULTS

Projection patterns of JO neurons in the brain of A. aegypti

Investigation of central processing of sound first required confirmation how JO neurons projected into the central brain. The morphology of JO neurons in mosquitoes has been studied largely via fluorescent tracer application and head sectioning (16, 17, 26). However, examining the comprehensive projection pattern of JO neuron axons requires whole-brain imaging (27, 28). We therefore conducted whole-brain imaging using neural tracing (four male and three female replicates) from one of the pedicels which house JO neurons to compare the panoptic morphology of JO neurons between sexes (see Materials and Methods; fig. S1A). We also traced flagellar neurons [which potentially include olfactory, thermosensitive, and hygrosensitive neurons (29, 30)] to facilitate distinguishing projection patterns between different neuronal groups—in this case, flagellar and JO neurons.

In both sexes, most of the fluorescent tracers from the flagellum and a subset of tracers from the pedicel labeled the AL, known to be a primary olfactory processing center, matching a previous study (fig. S1, A and B) (24). The tracer applied via the pedicel further labeled axon bundles innervating the AMMC (Fig. 1, A to E) (26). These axon bundles are larger and broader in males than females, paralleling a previous study that reported the sex differences in size of the AMMC (25). The tracer applied via the pedicel also labeled several neuronal clusters whose cell bodies were located in the brain. These neurons, which are presumably efferent neurons that innervate the antenna, also exhibited sexual Males have at least six putative efferent neuron clusters (observed in three of four brains), whereas we found five such clusters in a female (observed in one of three brains, with the other brains having fewer clusters; Fig. 1, B and C).

To annotate these clusters according to their location, we first identified different brain structures following nomenclature established in a previous study; this included the AMMC, AL, gnathal ganglion (GNG; aka subesophageal zone), clamp, and optic lobe (OL) (Fig. 1A) (31). We next annotated clusters by their mPME-I/C (mosquito posterior medial efferent neurons-ipsilateral/contralateral) located in the clamp, mLALE-I/C (mosquito lateral antennal lobe efferent neurons) lateral to the AL, and mVME-I/C (mosquito ventral medial efferent neurons) in the GNG, which project bilaterally to both AMMCs (Fig. 1, B, C, and F, and table S1).

Axons projecting from the pedicel to the AMMC, classified as JO neuron axons, bifurcated into at least six branches in males and five branches in females (Fig. 1, D and E, and fig. S1, C and D; three of four male brains and two of three female brains). Only a subset of all branches identified was labeled in individual brains, potentially due to technical limitations associated with the use of dextran. Among the six branches, five branches (mosquito JO ipsilateral neurons 1 to 5; mJO-I1 to mJO-I5) projected to the ipsilateral brain, and one bundle (mosquito JO contralateral neurons 1; mJO-C1) innervated the contralateral brain. After the pedicel-derived axon bundle enters the frontal side of the brain and starts projecting toward the posterior side, mJO-I1 branched first and arced to innervate the dorsal posterior AMMC. Two medial branches, mJO-I2 (dorsal) and mJO-I3 (ventral), then branched out, further innervating the posterior. The most lateral branch, JO-I4, appeared only in males (fig. S1, C and D). A lateral axon, mJO-I5, projected to the medial brain by passing through the gap between mJO-I2 and mJO-I3. Female mJO-I5 extended further to the dorsal brain compared to male equivalents.

Together, JO neurons in both sexes projected to both ipsilateral and contralateral AMMCs, although male JO neurons appeared to have a more diverse branching pattern.

Sexual dimorphisms of sound representation in the brain

We next reconfirmed previous reports of peripheral hearing function (10–12, 17–19) and compared electrical tuning between male and female ears using a combined laser Doppler vibrometry (LDV)/electrophysiology paradigm. We simultaneously measured mechanical vibrations of the flagellum and antennal nerve compound action potential signaling in response to 1-s long sweep stimuli provided via electrostatic actuation covering either 1 to 1000 or 1000 to 1 Hz (Fig. 2A; see fig. S2A for analysis paradigm). We found mean peak mechanical tunings of 431 and 253 Hz for males and females, respectively, with mean peak electrical tuning estimates of 346 and 220 Hz (Fig. 2B and table S2). Males had significantly higher peak mechanical and electrical tuning values than females [Aligned Rank Transform analysis of variance (ART ANOVA) with Tukey method; P < 0.0001 for both tuning types], and both sexes have higher peak mechanical than peak electrical tunings (ART ANOVA with Tukey method; P < 0.0001 for males and P = 0.0222 for females).

Analysis of WBF recordings of free-flying mosquitoes gave mean values of 499 and 819 Hz for females and male, respectively (fig. S2B and table S3). Although male WBFs were significantly higher than female WBFs (Welch’s t test; P = 2.2 × 10^−16^), we observed variations within a sex of up to ±50 Hz of mean values (fig. S2B). The estimated QDP (equal to the difference between male and female WBFs) under these conditions is around 320 Hz, similar to the estimated peak electrical tuning of males of 346 Hz. In the electrical tuning, we found another peak at around 155 Hz in 3 of 11 males, which was smaller in amplitude than the higher frequency peak (fig. S2C). Although this lower frequency peak was more similar in frequency to the peak female electrical tuning frequency than the higher male peak, it appeared sharper in profile and may not be related to female tuning (fig. S2C). However, electrophysiological recordings from the entire antennal nerve bundle precluded the separation of potentially different populations.

We thus next used calcium imaging to investigate responses in the brains’ primary auditory processing center. Given that JO neurons mainly project into the AMMC in both sexes, we investigated the frequency response properties of the AMMC (Fig. 2, C and D, and fig. S3A to C), including both JO neurons and their downstream neurons. We used the Q/QF system to drive fluorescent calcium reporter GCaMP6s expression pan-neuronally using a brp-QF2w driver line (32) and observed sound-evoked calcium responses in the brain. We first delivered pure tones with a wide frequency range (50 to 850 Hz) and observed the area that corresponds to the AMMC (fig. S3D; see Materials and Methods for details regarding identification of the AMMC). In general, male AMMCs responded to a far broader range of frequencies than females, with males showing strong responses to tones between 150 and 500 Hz whereas females demonstrated strong responses to pure tones of around 100 to 200 Hz and almost no response to tones above 350 Hz (fig. S3E).

To analyze detailed response properties for each stimulus frequency, we focused on the frequency range to which responses were identified while providing sufficient time between stimuli to allow the fluorescence level to decrease to baseline. As it proved challenging to deliver a wide range of tone frequencies with sufficient intervals between stimuli while also maintaining mosquito health, we reduced the frequency range tested for subsequent experiments. We thus next delivered individual pure tones focusing on a frequency range of 100 and 500 Hz to confirm the peak frequency, which was detected at 362.5 Hz in males and 175 Hz in females (ART ANOVA; P = 0.0001; Fig. 2, F and G; fig. S4, A and B; and table S4). These estimates fit closely with peak frequency estimates from electrophysiology measurements (Fig. 2B).

Mosquito AMMC has functionally distinct auditory regions

To distinguish functionally different neuronal populations, we analyzed the response properties of each pixel of our AMMC images (Fig. 3A). We first normalized time traces of fluorescent changes in each pixel with maximum values of each pixel to enable comparison of response patterns. To extract only reliably responsive pixels, we discarded pixels without detectable responses for any stimulus type (see Materials and Methods; fig. S4C and table S4 for pixel counts used for the analysis), yielding an 11,646 x 1700 matrix (pixels x frames). We then mapped the frequency to which each pixel showed the largest response to visualize peak frequency responses in the AMMC. It should be noted that this analysis focused only on the frequency peak and not on other response features (e.g., the sharpness of the response frequency spectrum), which were analyzed later (see Fig. 4).

Female AMMCs showed maximal response peaks in the range of 100 to 200 Hz, whereas male responses were mainly restricted over 250 Hz (Fig. 3B). Unlike fruit flies (33–36), the spatial organization of best frequencies appeared varied across individuals, with male AMMCs showing a far greater range of peak frequencies than females (Fig. 3C). This agreed with the multiple peaks covering a broad range of frequencies observed in the electrical responses of some male JOs (fig. S2C).

Next, to more fully capture the range of response pattern properties beyond best frequencies, we mapped representations of whole response features (Fig. 4). By categorizing all time traces of calcium responses in each pixel that cover the AMMC, the response types were split into distinct clusters based on Euclidian distance, which allowed us to investigate neuronal groups with similar overall response properties, rather than focusing only on the maximum response properties. The number of clusters depends on the threshold of At large threshold distances (>200), the response types were divided into two, male- and female-type responses, possibly due to their significant difference in frequency range. Further decreasing the threshold distance increased the cluster number, with a distance of 85 (red line in Fig. 4A) yielding seven clusters (Fig. 4A).

At this distance, male and female response properties were still almost entirely Four clusters (#1 to #4) were exclusively found in male AMMCs, two clusters (#6 and #7) were observed only in females, and one cluster (#5) was present in both sexes (Fig. 4A). Most male clusters tended to show strong responses to a wide range of frequencies. In contrast, female AMMCs responded to more specific frequencies (Fig. 4, A and B). Notably, pixels in each cluster showed some extent of diversity in their best frequencies, reflecting diversity of the best frequency of each cluster (Fig. 4C). Male clusters include more diverse best frequencies, suggesting that the diversity of best frequencies in male AMMCs was in part due to these wider response properties (Fig. 4C). Pixels that belonged to the same cluster were also located in close spatial proximity, suggesting that functional organization is, to some extent, related to anatomical organization.

Although we tried to align the angle and depth of the AMMC as precisely as possible, we could not find any common pattern among individuals (Fig. 4B). Although female AMMCs show specific response patterns, male AMMCs appeared to have more functionally diverse pixels. Because the entropy of clusters across pixels was higher in males throughout the distance, we conclude that male AMMCs are more functionally diversified than females (ART ANOVA; P = 0.0289; Fig. 4, D and E, and table S4).

SSO may produce antiresonance of a flagellum

Unexpectedly, cluster #4 (identified only in males) showed a strong negative calcium response to 150 Hz (Figs. 4A and 5A), with at least two males tested displaying a drop in calcium concentration in response to stimulation of this frequency. One possible reason is that the flagella of these males were vibrating in distinct states to other males, i.e., demonstrating SSOs (17). We hypothesized that playback of a 150-Hz tone minimized the male flagellar oscillation while in the SSO state, causing a reduction of calcium level.

To test this hypothesis, we investigated the frequency components of the male flagellar response to white noise (1 to 2000 Hz; Fig. 5B). First, we injected male mosquitoes with dimethyl sulfoxide (DMSO), which has been shown to induce SSOs (Fig. 5C) (17). Using LDV, we then monitored the response movement of the SSO-induced flagellum when exposed to the white noise (Fig. 5B). Measuring the relationship between the vibration stimulus of white noise (input) and the flagellar behavior (output) is equivalent to performing system identification, which mathematically describes the dynamic system of the flagellum. Here, the behavior of the flagellum is identified in the state-space model (37), which has the advantage of being able to describe not only input-output relationships but also internal states. The continuous changes observed in the envelope of time traces of flagellar vibrations encouraged us to use a second-order state-space model for system identification (Fig. 5, D to F), which enabled simulation of the frequency characteristics of the flagellum.

Three of 12 male SSO flagella demonstrated a negative peak of power gain (antiresonance) around 150 Hz (Fig. 5G and fig. S5A). In the quiescent state (i.e., non-SSO state), on the other hand, no antiresonance was found in any individuals (fig. S5B). This suggests that ~150-Hz tones can minimize flagellar vibrations in some cases, bringing about a drop in calcium level in specific individuals.

Cilium-related genes are highly enriched in male pedicels

The cilium’s location at the distal dendrite of JO neurons has encouraged some researchers to hypothesize that it is a source of energy for SSOs (Fig. 6A) (3, 13, 14). We thus next examined the cilium structure of JO neurons in males and females. Analysis using transmission electron microscopy (TEM) found that both sexes shared a 9x2+0 structure of axonemal cilium as previously described (13, 15). We also found inner and outer dynein arms associated with microtubule doublets in both sexes, suggesting that both sexes share the basic structure of axonemal cilium in their JO neurons despite the male specific phenomenon of SSOs (Fig. 6A).

Although gross anatomical structures were conserved between the sexes, specific ciliary-related genes may still be differentially expressed, underlying differences in hearing function. We therefore compared the gene expression profiles of male and female pedicels, which house the JO and include cell bodies of JO neurons as well as supporting cells (Fig. 6B), as well as male and female heads (lacking mouthparts and antennae) (38). We first verified the quality of RNA sequencing (RNA-seq) with principal components analysis (PCA) and found that most of the observed variance was attributable to tissue (pedicel and head) (43%) and sex (20.7%), indicating that biological replication effects were relatively negligible (fig. S6A). Differentially expressed gene (DEG) analysis using DESeq2 (39) identified 820 transcripts enriched in male compared to female pedicels, whereas 1377 transcripts were found to be more abundant in female compared to male pedicels [P < 0.1, log2 fold change (FC) > 1; Fig. 6C and fig. S6B]. Gene Ontology (GO) enrichment analysis identified GO terms annotating up-regulated and down-regulated genes in male pedicels compared to female pedicels, female heads or male heads. In these GO terms, gene overlaps were frequently observed across tissues (tables S5 and S6).

To visualize GO terms that label the same genes, we measured the distance among terms by the proportion of shared genes and constructed a GO network. When we focused on the analysis of male and female pedicels (Fig. 6D), all GO terms related to up-regulated genes in the male pedicel were clustered into four groups, one of which was related to the construction of cilium. Female pedicel-enriched genes comprised 13 clusters, including regulation of blood coagulation and biotic stimulus detection. Similar comparisons between male pedicels and female or male heads also identified ciliary construction terms as being up-regulated in male pedicels (fig. S6, C and D).

Although transcriptomics can provide a comprehensive overview of gene expression, protein abundance does not necessarily strongly correlate with RNA level (40, 41). To test whether the up-regulation of cilium-related genes was also detected at a protein level and correlated with RNA expression levels, we next conducted liquid chromatography–tandem mass spectrometry (LC/MS-MS) proteomic analyses of mosquito pedicels and heads. PCA showed that over 50% of the variance could be ascribed to tissue (37.13%) and sex (13.06%), validating that the replication effects were also negligible in proteome analysis (fig. S6E). Linear model analysis using DESeq2 identified 1301 proteins enriched in male pedicel compared to females (P < 0.1, log2 FC > 1; Fig. 6E and fig. S6F). GO network analysis confirmed that cilium-related proteins were also up-regulated in male pedicels in comparison to female pedicels, as well as compared to male and female heads (Fig. 6F; fig. S6, G and H; and tables S5 to S7).

We found a significant correlation between log2 FC between male and female pedicels at RNA and protein levels, suggesting that the enrichment of most genes is comparatively similar across levels (Pearson’s correlation coefficient; R = 0.43, P < 0.001; Fig. 6G). Genes related to cilia were typically up-regulated in both RNA and protein (blue points in Fig. 6G). By focusing on genes overexpressed in male pedicels compared to female pedicels only at the protein level, we identified 1235 genes significantly up-regulated in males (P < 0.1 and log2 FC > 1 at protein level only; green points in Fig. 6G). GO enrichment analyses of these genes identified terms related to splicing and metabolic processes (Fig. 6H).

Although, in A. aegypti, dyneins are not explicitly annotated as “dynein arms,” such annotated genes exist in Anopheles gambiae (An. gambiae). We therefore first identified genes in An. gambiae annotated with GO terms related to inner and outer dynein arms and then identified their orthologs in A. aegypti (see Materials and Methods). We found 17 putative dynein genes at the transcriptomic level, with 12 of these also picked up at the protein level (Table 1; red points in Fig. 6G). Although most dyneins were present in both sexes, most were significantly up-regulated in males compared to females. Only three genes were not significantly dimorphic at the RNA level (AAEL002792, AAEL002819, and AAEL010197), and only three other genes were not dimorphic at the protein level (AAEL001340, AAEL008036, and AAEL022882). In general, therefore, dynein expression appears higher in males than females.

To identify transcription factors regulating these dyneins, as well as other ciliary-related factors [genes of interest (GOIs); Fig. 7A], we ran weighted gene correlation network analysis (WGCNA) with comparatively up-regulated genes in male pedicels compared to female pedicels (logFC of male pedicels versus female pedicels > 0; Fig. 7B). To screen potential transcription factors of GOIs, the directly connected genes to GOIs (neighboring genes) were visualized at either RNA (Fig. 7C and table S8) or protein (Fig. 7D and table S9) levels. Non-GOI genes with the highest levels of connections with GOIs (five to seven connections for transcriptomic data and two connections for proteomic data) were determined as the top neighboring hub genes (Fig. 7, C and D). Analyses at RNA and protein levels found six and five genes, respectively (Table 2). This included identification of AAEL009489 (ortholog of Drosophila fd3f, a known transcription factor regulating dyneins) and AAEL010682 (ortholog of Drosophila gudu, an important regulator of spermatogenesis) from the transcriptomic data analysis, and AAEL009677 (ortholog of Drosophila HEM, which plays an essential role in cytoskeletal organization) from the proteomic data analysis.

DISCUSSION

Here, we identified the sexual dimorphic architecture of the primary auditory center, AMMC, in the mosquito brain. Subsequently, by comparing the auditory response properties of the brain between males and females, we identified sex-specific spatial diversity in the auditory representation of the AMMC. We also investigated molecular differences in the hearing systems of male and female mosquitoes using a combination of gene expression profiles of hearing organs and electron microscopy.

Fluorescent tracer application into the pedicel allowed us to trace whole neurons connecting to the pedicel, labeling both JO neurons and putative efferent neurons. The width of the JO neuron axon bundle appeared larger in males than females, aligning with a previous report of the AMMC being significantly larger in males and males having approximately double the number of JO neurons compared to females (25). The neurites of JO neurons overlap with each other in the AMMC, precluding us from describing the detailed bifurcation pattern of JO neurons. However, the fluorescent tracer labeled at least five axon bundles of male JO neurons and four branches in females. A branch of JO neurons innervates the contralateral AMMC, in contrast to fruit flies whose JO neurons do not project to the contralateral AMMC (27, 28). As for optic nerve projection through the optic chiasm in vertebrate visual pathways, signals detected by one ear are presumably partially transmitted to the contralateral AMMC in mosquitoes (42).

One possible function of this binaural convergence is to represent the signals from both ears for encoding acoustic field representation, equivalent to visual field representation in the vertebrate brain to enable stereopsis. It should be noted, however, that the JO not only detects sound but also gravity and wind in Drosophila (34, 36). Further investigation is needed to elucidate whether and how the mosquito AMMC is divided into functional subregions and how the information delivered from both ears is interactively processed in the AMMC.

In this study, tracers applied to the gap created by removing the flagellum alone, or the hole made by removing the pedicel, labeled the AL. A previous study reported that mosquito JO neurons project to the subset of the AL known as the JO-C, where auditory and olfactory information may be integrated (24). Although it is possible that we labeled olfactory sensory neurons, tracer application to the hole made by removing the pedicel in Drosophila predominantly labeled JO neurons (28). Considering this, our neuro-tracing experiments lend support to this previous report. Calcium imaging of the AL would help test whether auditory information is also represented in this brain region.

Previous studies in humans have reported sexual dimorphisms in the cochlea, an equivalent hearing organ to the mosquito JO, in terms of size and stiffness gradient (43–46). However, large sex differences in frequency properties are rarely observed in the early stages of auditory processing across a wide array of animals, including humans [but see (44)] (47–49). In the present study, we found that the frequency characteristics of the AMMC, the first stage of auditory processing in the mosquito brain, are significantly distinct between the sexes, suggesting that strong sexual selection pressure on mosquitoes has brought about these substantial dimorphisms.

Here, we expanded on previous reports demonstrating differences in peripheral hearing function between the sexes, as well as differences between males in their responses to auditory stimulation, by studying central auditory representation (Fig. 8). Male AMMCs showed peak responses to 362.5-Hz tones, similar to the estimated peak electrical tuning of JO neurons estimated via electrophysiology (~346 Hz) and the best frequency of the QDP (320 Hz). Our study thus lends support to the DP-based communication hypothesis of mosquito acoustic communication (10, 17, 50) and opens potential avenues for further understanding the neural basis of their acoustic communication. Future work on calcium imaging of tethered mosquitoes could integrate male WBF monitoring during flight with simultaneous calcium imaging recordings from their AMMCs while they are stimulated with a range of female flight sounds. Given previous reports of the significant differences in attractive frequencies previously reported between individual males, simultaneous measurements of the sounds produced by males, auditory representation in their AMMCs, and their behavioral responses could facilitate deeper understanding of the bases of these differences between males.

Our profiling of auditory representation in mosquito AMMCs was limited to responses to pure tones of different frequencies. Similar profiling in fruit flies used other stimulation types to characterize AMMC clusters (33)—future equivalent profiling in mosquitoes could thus also use different stimulus types, such as static deflections to identify regions sensitive to large displacements of the flagellum (potentially marking these regions as containing wind/gravity neuronal subpopulations). Although our bin size of 50 Hz allowed us to present three repeats of pure tone stimuli covering a large range of frequencies, it also reduced the granularity of our analyses, particularly for females, whose range of responses was far smaller than males. Future experiments could use a smaller bin size covering a narrower frequency range to create more detailed neuronal profiles.

Furthermore, mosquito acoustic communication is highly sensitive to the environmental temperature, with both WBF and hearing function thermosensitive. As such, changes in background temperature across or within experimental assays could potentially influence the results of our functional/behavioral assays. Although the magnitude of differences we observe between males and females (>100 Hz for electrophysiological measurements, >300 Hz for WBF, and >100 Hz for calcium imaging experiments) suggests that this variability may not confound our analyses of sexual dimorphisms in hearing properties, future work could explore changes in mosquito JO neuron responses and auditory representation in the AMMC to stimulation across a temperature gradient.

Considering the distinct behaviors observed in male and female mosquitoes, sexual dimorphisms in the information processing pathways of various sensory modalities have likely evolved to support their sex-specific behaviors, such as phonotaxis, blood-feeding, and host-seeking. This is in part reflected in changes in brain morphology, with the AMMC significantly larger in males than females, and the AL significantly larger in females than males (25). The analytical tools we developed in this study should help identify the sex-biased landscape of sensory processing in the mosquito brain.

We also found a conserved response cluster shared between males and females (i.e., cluster #5). A previous study has shown that male Aedes diantaeus display repulsive behaviors to sounds between 140 and 200 Hz (51). Furthermore, female A. aegypti showed increase in flight speed when exposed to sounds of around 200 Hz (7). These responses, coupled with prior reports of auditory sensitivity to acoustic stimulation in this frequency range (8, 9), have been linked to predator avoidance. Numerous species from the order Odonata, such as dragonflies, are known predators of mosquitoes and have WBFs of ~30 to 50 Hz (52, 53). Although pure tones of these frequencies may be difficult to perceive due to background environmental noise, harmonics of these WBFs should thus fall directly into the range of audible sounds. This partial conservation of auditory representation between the sexes connected to shared auditory behaviors may explain gross anatomical similarities in inner/outer dynein arm structure we observed via TEM; female mosquitoes still use active hearing mechanics to some extent, whereas males require a more diverse auditory processing system to show both negative and positive phonotaxis.

Our state-space modeling suggested that SSOs can interfere with the flagellum response at 150-Hz tone, bringing about the negative calcium response of the AMMC detected in some males (Fig. 8). We measured the vibrations of SSO flagella induced by DMSO injection, which causes nonspecific physiological impairments. Antiresonance of the flagellum appeared only in 3 of 12 SSO-induced individuals, raising two questions. First, why was antiresonance not found in all SSO-induced mosquitoes? One possible reason is that DMSO damaged not only the suppression system of SSOs but also the key mechanism of active hearing in the other mosquitoes. Second, what is the biological role of antiresonance in SSOs? SSOs appeared so rarely that no physiologically specific induction of SSOs has been reported in A. aegypti (5, 17). SSOs entrain only to pure tones similar to female WBFs, suggesting that the antiresonance produced by SSOs is a by-product of maximizing sensitivity to female flight tones. Calcium imaging of auditory responses of SSO-induced brains would support our understanding of how SSOs alter auditory processing in the brain.

Our transcriptome and proteome analyses revealed enrichment of GO terms related to the regulation of blood coagulation and responses to biotic stimulus in females, potentially relevant for blood aspiration. On the other hand, microtubules, dynein, and related genes (i.e., cilium-related genes) were found to be highly enriched in the male pedicel. However, electron microscopy images suggest that both males and females share the basic cilium structure of the JO neuron’s dendrite. These similarities may be linked to shared properties of the JO across the sexes, such as facilitating escape responses. Shared and distinct sets of cilium-related genes expressed may therefore have different kinetic properties, enabling sex- and situation-specific flagellar oscillations, including SSOs.

Our WGCNA output found fd3f, a known transcription factor regulating dynein expression (as well as other auditory-related genes) in Drosophila melanogaster (54), to be a key neighboring hub gene, suggesting that further research into the role of fd3f in mosquitoes could provide insight into sexual dimorphisms in hearing properties. Further investigation of our datasets could help identify other, potentially mosquito-specific hearing transcription factors. Calcium imaging of males whose male-specific cilium-related genes are knocked down may help our understanding of the role of the male-specific gene set in determining the male-specific response properties in the AMMC.

Limitations of this study

Here, we used fluorescent tracers to visualize the anatomy of JO neurons and flagellar neurons in the brain. For JO neurons, we used 10,000–molecular weight (MW) dextran, which does not penetrate gap junctions, to distinguish efferent neurons innervating the antenna and neurons downstream of JO neurons. We also used 3000-MW dextran, which diffuses at a faster rate, to label the full structure of neurons. This methodology can result in variations in staining levels among individuals. Furthermore, we did not observe strong staining of the AL after application of dextran following removal of the entire pedicel. Studies in Drosophila have found that JO axons in the antennal nerve are much larger than the axons of olfactory sensory neurons (55, 56). If this is also the case in mosquitoes, the much smaller axon diameter of olfactory sensory neurons may make it considerably more difficult to take up sufficient dextran for labeling. Future anatomical analyses of JO neurons using a genetic driver [e.g., trpVa^QF2^; (57)] should help overcome some of these limitations.

The pan-neuronal driver used in this study labeled not only JO neurons but also other auditory and nonauditory responsive neurons. Therefore, although the calcium responses we observed during sound playback appeared localized to the AMMC (fig. S3C), our calcium imaging data may still potentially include responses of downstream neurons and noise unrelated to auditory responses. To elucidate more precise spatial arrangement of frequency properties, genetic drivers that label specific neurons should be generated [as has previously been done in Drosophila; (35, 36, 58)].

Although multiomic analyses can lend power to molecular studies, they can also cause complications given differences in sensitivity between transcriptomics and proteomics, as well as difficulties in comparing transcript and protein abundance at a single time point given the time lag between transcription and translation, in addition to posttranslational protein modifications. For example, although our transcriptomic analyses identified 2441 transcripts up-regulated in male pedicels compared to male heads (as compared to 1542 down-regulated in male pedicels compared to heads), our proteomic analyses identified far fewer up-regulated proteins (433). Further work conducting these comparisons across multiple time points is necessary to fully capture the complexity of tissue-specific gene expression.

SSOs have been modeled using a harmonic oscillator in the past (17). When modeling self-excited vibrations, such models are highly suitable; here, however, we focused on vibrations when acoustic stimuli were applied, making it necessary to account for the application of external forces. Our model allows for the switch between a harmonic oscillator and a forced vibration system depending on the presence or absence of an external force, such as acoustic stimulation.

The results of our frequency response analysis (using Bode diagrams) showed different responses across mosquitoes (fig. S5). This could simply be the result of mosquitoes exhibiting different behaviors, but it is also possible that our one-dimensional (1D) LDV measurements of the 3D movement of the flagellum were insufficient. Given this lack of dimensionality in our dataset, we modeled the flagellar movement obtained by 1D measurements as forced vibrations. Further work is therefore required to collect 3D LDV data to model the 3D flagellar movement, as well as correspondingly increase the complexity of the model.

MATERIALS AND METHODS

Mosquito rearing

All A. aegypti mosquitoes were reared using a 12-hour:12-hour dark cycle at 28°C and 60 to 70% relative humidity. Adults were fed a 10% glucose solution. For calcium imaging, we used brp-QF2w (32) as a driver line and QUAS-GCaMP6s (32) as a reporter line. The wild-type Liverpool strain was used for all other experiments.

Tracer application

Mosquitoes 3 to 10 days after eclosion were anesthetized on ice, and one of their flagella was clipped from the pedicel. To visualize the nerve from the flagellum, a drop of fluorescent tracer (dextran, tetramethylrhodamine, and Biotin 555/580, 3000 MW, lysine fixable, micro-ruby; D7162, Invitrogen) dissolved in phosphate-buffered saline (PBS; #T9181, Takara) was placed onto the gap created by removing the flagellum. Mosquitoes were then kept at 4°C for an hour. To visualize the nerve from the pedicel, the pedicel whose flagellum was clipped was then removed on ice and a drop of Alexa Fluor 488 dextran (10,000 MW, anionic; Thermo Fisher Scientific), which does not cross gap junctions, was placed onto the hole. The mosquitoes were then kept at 4°C for 3 hours (fig. S1A).

Mosquito heads were fixed in 4% paraformaldehyde (PFA) in PBS at 4°C for 3 hours. Brains were dissected from the head in PBS, kept at 4°C in PBS containing 0.5% Triton X-100 (PBT) overnight, and incubated with primary (Mouse anti-Bruchpilot nc82; DSHB, RRID: AB2314866) and secondary (Alexa Fluor 647–conjugated anti-mouse IgG) antibodies for 3 days, respectively, at 4°C. After rinses with PBT and PBS, brains were incubated in 50% glycerol in PBS for an hour and 80% glycerol in deionized water for 30 min. Four male and three female brains were imaged on glass slides.

Confocal microscopy and image processing

Confocal images of brains were obtained at 0.84-μm intervals with an FV 1200 laser scanning confocal microscope (Olympus, Tokyo, Japan) equipped with a silicone-oil immersion 30x Plan-Apochromat objective lens [numerical aperture = 1.05].

Combined vibrometry and electrophysiology Recordings

Combined vibrometric and electrophysiological recordings used 2-day entrained mosquitoes at Zeitgeber time (ZT) 11 to ZT 13, in a temperature-controlled room at 25° ± 1.5°C. Mosquitoes were mounted on rods using glue such that only their right flagella were free to move. The rod was then placed in a micromanipulator on a vibration isolation table.

A reference electrode was inserted into the mosquito thorax to facilitate charging to −30 V relative to ground. Electrostatic actuators were placed around the freely moving flagellum so that electrostatic stimulation could be provided. The insertion of a recording electrode into the base of the pedicel enabled electrical recordings from the antennal nerve. A laser Doppler vibrometer (Vibroflex, Polytec) was focused on the tip of the flagellum to record flagellar displacements to stimuli.

Next, a calibrated force-step stimulus was provided to calibrate the displacement of the flagellum to ~±3.5 μm; the stimulus was provided to the electrostatic actuators using a data acquisition unit CED Micro1401 (Cambridge Electronic Design). Following this calibration, sweep stimuli consisting of 1.0-s long chirps of linearly increasing or decreasing frequency (i.e., 1 to 1000 or 1000 to 1 Hz), were provided. Each loop consisted of 10 sets of sweeps, each containing four different forward phasic, forward antiphasic, backward phasic, and backward antiphasic. Between phasic and antiphasic sweeps 0.1 s of silence was provided, whereas 0.4 s of silence was provided between forward and backward sweeps. Flagellum displacement and nerve recording data were recorded using the Spike2 software (Cambridge Electronic Design).

Combined vibrometry and electrophysiology Data analysis

For mechanical tuning analyses to calculate the flagellar tuning frequency, only data for which the laser quality data were recorded as greater than zero was included. A dc remove with time constant of 0.001 s was first applied to the laser data, before rectification and a smooth function with time constant of 0.0005 s (fig. S2A). A slope function with 0.1-s time constant was used to find when the channel gradient was equal to zero; this represents the time point of maximum flagellar vibration. Calculation of this time allowed for estimation of the corresponding stimulus frequency, referred to as flagellar ear mechanical tuning frequency.

For electrical tuning analyses, nerve responses from sequential phasic and antiphasic stimuli were averaged to cancel artifacts recorded in the nerve channel resulting from electrostatic actuation (fig. S2A). This processed nerve signal was analyzed as described above for the laser signal. dc remove (time constant = 0.001 s) and rectification and smooth (time constant = 0.0005 s) functions were applied, followed by a slope function (time constant = 0.1 s). Estimating the time at which the slope equaled zero enabled calculation of the corresponding stimulus frequency, denoted as the ear electrical tuning frequency.

In total, 11 male and 10 female electrophysiology recordings were collected and analyzed.

WBF measurements

Groups of 50 virgin male or female mosquitoes were entrained in incubators for 2 days at 25° ± 1.5°C. During dusk on the third day, a microphone (EK series, Knowles) was inserted into the cage and flight tones recorded during the period immediately preceding dusk using a Picoscope 2408B and the Picoscope 6.14 software (Pico Technology) at a 50-kHz sampling rate. Data were analyzed using the SimbaR package in R with only flight recordings longer than 150 ms included in the final analysis. The median of all estimated WBFs for a single cage was then calculated. These median values were found to be normally distributed for both females and males (Shapiro-Wilk test, P > 0.05 for both groups), so the mean of these medians was calculated.

In total, WBF recordings were collected and analyzed from 11 male and 12 female cages.

Calcium Data acquisition

Mosquitoes were collected within 24 hours after eclosion, and fewer than eight mosquitoes of each sex were separately kept in individual vials for 5 to 9 days after eclosion. Mosquitoes collected at ZT 10 to ZT 13 were anesthetized on ice and stabilized ventral side up onto an imaging plate using ultraviolet-cured adhesive (Norland Optical Adhesive 81, Norland). Head angles were adjusted to prevent the flagellum from dipping into Ringer’s solution, and the dorsal head and thorax were glued with the adhesive to fix the flagellum direction. The posterior head and thorax were also glued to avoid destroying the cap structure during the following steps. The ocelli of the medial head and the mouthparts were removed using an incision scalpel (MICRO FEATHER, Feather) and forceps to open a window to monitor GCaMP fluorescence from the AMMC in the brain, which was determined by location (the region between the OL and GNG when observed from the ventral side). A drop of saline solution with the following components was added to prevent 108 mM NaCl, 5 mM KCl, 2 mM CaCl2, 8.2 mM MgCl2, 4 mM NaHCO3, 1 mM NaH2PO4, 5 mM trehalose, 10 mM sucrose, and 5 mM Hepes, pH 7.5, 265 mosmol. A fluorescent microscope (Axio Imager.A2, Carl Zeiss) equipped with a water-immersion 20× objective lens (W Achroplan/W N-Achroplan, numerical aperture = 0.5; Carl Zeiss), a spinning disk confocal head CSU-W1 (Yokogawa Electric Corporation), and an OBIS 488 LS laser (Coherent) for excitation at 488 nm was used.

To provide sound stimuli, a loudspeaker (FF225WK, Fostex) was positioned ~11 cm from the antenna of the mosquito. To monitor calcium responses to pure tones, a series of pure tones with different frequencies (50 to 850 Hz by 50 Hz for fig. S3E and 100 to 500 Hz by 50 Hz for Figs. 2 to 4) were delivered. Each tone was delivered for 2 s with a 2-s pause for 50- to 850-Hz stimulation and a 15-s pause for 100- to 500-Hz stimulation paradigms. The peak-to-peak amplitude of the tone was ~5 mm/s for each pure tone.

Calcium Analysis

Calcium imaging data were analyzed using MATLAB (MathWorks), Fiji, and R software. Motion artifacts due to animal movement were canceled using the NormCorre algorithm by running the CaImAn toolbox with MATLAB. For fig. S3E, the region of interest (ROI) was set at the bulging area in the lateral anterior brain, where the AMMC is located. The mean fluorescence intensity in the ROI for each frame was calculated using Fiji. For Figs. 2 to 4 and fig. S4, the 50 x 50 pixels around the AMMC were selected, and responsive pixels were extracted using the following process.

The relative fluorescence change (dF/F) was calculated to indicate the calcium response, which was given bydF/F=(Fn−FBase)/FBase(1)where Fn is the corrected fluorescence intensity at n seconds from the sound onset. FBase is the average of the corrected fluorescence intensity during the 5 s before the stimulus onset. Noise (>3 kHz) for dF/F was cut with a Butterworth filter using the “signal” package via R for fig. S3E and “designfilt” function via MATLAB for Figs. 2 to 4 and fig. S4A.

The filtered value was normalized by the maximum value of each trial (normalized dF/F), allowing us to compare the response properties with the following process. To ensure the quality of analysis in Figs. 2 to 4 and fig. S4, we set two criteria for pixels to be (i) The maximum response is located between the onset of the stimulus and 3 s after the offset. (ii) The SDs of dF/F at the same time point in three trials were less than 0.5. Pixel counts in male and female brains are shown in fig. S4C and table S4. For Fig. 2, time traces of calcium responses (dF/F) in each individual were averaged across pixels and the best frequency of each individual was determined as the frequency of the tone at which the maximum value of the time trace was identified. In Figs. 3 and 4, the average time traces of normalized dF/F of all replicates in each pixel were analyzed. The best frequency in each pixel was identified based on the timing of the maximum fluorescent value.

Hierarchical clustering was conducted based on Euclidian distance, and the cutoff criterion was set at 85. Entropy (H) in Fig. 4 (D and E) was calculated to indicate the diversity of clusters across pixels, which was given byH=−∑E∈UP(E)logP(E)where P(E) is the proportion of cluster E in all pixels in each individual. U is a total set of cluster E.

LDV with DMSO injection

Male mosquitoes were glued to rods so that no body part apart from their right flagella was able to move. After gluing, individual rods were held in a micromanipulator on a vibration isolation table at 25° ± 1.5°C. A laser Doppler vibrometer was focused on the mosquito’s right flagellum, and 50% DMSO in Ringer’s solution was injected into the base of the left pedicel. Flagellar displacement amplitudes increase by several orders of magnitude while exhibiting SSOs (Fig. 5C) (17). Flagellar movements without sound were measured over 60 s. Individuals that showed oscillations of at least 100 nm/s throughout the recording time were used for experiments. White noise playback (1 to 2000 Hz at a particle velocity level of 1.58 × 10^−4^ ms^−1^, equivalent to a sound pressure level of 70 dB) was provided using a loudspeaker (FF85WK, Fostex). Flagellar velocity in response to white noise stimulation was measured for the next 3 min at a sample rate of 12 kHz using the VibSoft 6.0 software (Polytec). Data were analyzed from 12 males injected with DMSO and 9 males not injected.

Modeling of antennal vibrations using the state-space method

Second-order or higher differential equations can be used to model vibrations, the most basic of which is the mass-spring-damper system. In this study, the antenna is described as a two-degree-of-freedom mass-spring-damper system (Fig. 5D). Two-degree-of-freedom mass-spring-damper systems are generally known as coupled vibrations (forced vibration systems). When considering the antenna as coupled vibrations with two degrees of freedom, we assume one end to be a fixed end and the other to be a free end as the base of the flagellum is fixed and the tip can move freely. Hence, they can be expressed by the simultaneous differential equations shown in Eq. 2m1x¨1+c1x˙1+c2(x˙1−x˙2)+k1x1+k2(x1−x2)=0m2x¨2+c2(x˙2−x˙1)+k2(x2−x1)=f(2)where, m, c, and k represent the coefficient of the mass, damper, and spring that make up the coupled vibration system, respectively, and the subscripts 1 and 2 denote the number of components. The minimum coupled vibration system has two degrees of freedom and consists of two masses, springs, and dampers. The equation of motion is derived with subscript 1 as the fixed end and subscript 2 as the free end. Here, f is an external force and is expressed as Acos(ωt).

To find the unknown parameters m, c, and k in Eq. 2, we used the state-space conversion shown in Eq. 3, which is the most fundamental in modern control theory (56, 57). We measured the behavior of the antennae of an actual mosquito during stimulation with 1- to 2000-Hz white noise [equivalent to changing ω in f = Acos(ωt)] and identified the unknown parameters in Eq. 3. Identification of the variables was performed using the System Identification toolbox of MATLABz=Az+b′f=[OI−M−1K−M−1C]z+[O−M−1b]f(3)where the elements of each matrix are defined as followsz=[x1x2x˙1x˙2]T,M=[m100m2],C=[c1+c2−c2−c2c2]K=[k1+k2−k2−k2k2],b=01

This method is known as gray-box modeling. We found that, when the same white noise input was given to the model, it showed a response very similar to the behavior of a mosquito’s antennae, as shown in Fig. 5 (E and F).

Electron microscopy

Mosquitoes were decapitated, and the proboscis and flagellum clipped from the head in PBS. Heads were fixed overnight at 4°C in a fixative containing 2% EM grade PFA, 2.5% glutaraldehyde, and 0.08 M cacodylate. Heads were washed three times in the fixative and postfixed in 2% OsO4 for 3 hours. After washing in the fixative, heads were dehydrated by immersion in increasing concentrations of ethanol solutions, up to and including pure anhydrous ethanol. To infiltrate with resin, heads were passed through pure propylene oxide (PO) for 10 min, 1 PO/resin for 4 hours and pure resin for a day. Samples were then cured in silicon molds at 60°C for 48 hours. Pedicels from fixed heads were sectioned using an ultramicrotome (UC7k, Leica) with diamond knives (ultra 35° and histo, DiATOME). Thin sections were collected, stained with 2% uranyl acetate and lead stain solution (Sigma-Aldrich), and imaged using a transmission electron microscope (JEM-1400PLUS, JEOL).

Transcriptomics: Sample preparation

Groups of virgin male or non–blood-fed female mosquitoes were flash frozen in liquid nitrogen at ZT12 after entrainment in an incubator for 2 days. Tissues were dissected on ice in RNAiso Plus (#9109, Takara Bio Inc.). Five repeats of pedicels and heads were collected and RNA extracted from homogenized samples. RNAiso Plus was added to lysed samples for a final volume of 1 ml. Samples were incubated at room temperature for 5 min before 200 μl of chloroform (Kanto Chemical Co. Inc.) was added prior to mixing. Following a further 15-min incubation at room temperature, samples were centrifuged at 12,000g for 15 min at 4°C. A 0.5 ml volume of 2-propanol (Sigma-Aldrich) was added, and samples were stored for 30 min at −20°C.

The supernatant was discarded following centrifugation at 12,000g for 10 min at 4°C, and the remaining RNA pellet washed with 1 ml of 75% ethanol (Sigma-Aldrich) solution. Samples were centrifuged for 5 min at 7500g at 4°C, with this ethanol wash cycle performed twice before a further wash with pure ethanol (Sigma-Aldrich). The RNA pellet was next dissolved in Nuclease Free Water (Invitrogen), and RNA quality assessed using a Nanodrop (Thermo Fisher Scientific). RNA was then submitted to the Center for Gene Research at Nagoya University.

Transcriptomics: Analysis

Next-generation sequencing of submitted RNA was performed using NextSeq550 (Illumina Inc.). The quality of sequenced reads was checked using FastQC, and no sequence flagged as poor quality was detected. The RNA sequence length was 81. Single-end reads were aligned using HISAT2, with more than 90% of reads aligned and used for the following analysis. The number of reads mapped to each gene of a gene transfer format (GTF) file of the AaegL5.0 genome from vectorbase (https://vectorbase.org/vectorbase/app) was counted with featureCounts (59). Analysis of the read count data was conducted via “iDEP 2.0” (60).

The library size across samples was normalized using DESeq2’s median of ratios method. For PCA, the count data were transformed with EdgeR and Pseudo count c = 4 was selected (fig. S6A). The median value of all genes within a sample was assigned to missing values for that sample. DEGs were analyzed using DESeq2 with independent filtering of lower counts. The output csv file was downloaded and visualized using the “ggplot2” package in R (Fig. 6, C and E, and fig. S6, B and F).

Proteomics: Sample preparation

Groups of virgin male and female mosquitoes were collected as for the RNA-seq experiments described above. Tissues were then dissected in PBS and transferred to microfuge tubes containing PBS at room temperature. PBS was replaced by sample buffer [100 mM tris-HCl (pH 8.0), 1% SDS, and 20 mM NaCl], and tissues were homogenized on ice before being centrifuged at 16,100g for 10 min at 4°C to remove debris. The extracted supernatants were transferred to new microfuge tubes, with 2 μl of the supernatant per sample used to test protein concentration (Pierce 660 nm Protein Assay, Invitrogen, catalog no. 22662). Samples were stored in a −20°C freezer prior to LC/MS-MS. Extracts were reduced with 25 mM dithiothreitol for 30 min at 25°C, followed by alkylation with 50 mM iodoacetamide for 30 min at 25°C in the dark. Five repeats of pedicels and heads were collected.

Protein purification and digestion were performed using the sample preparation SP3 method (61). The beads were then resuspended in 100 μl of 50 mM tris-HCl (pH 8.0) with 1 μg of Lys-C (Fujifilm-Wako) at 37°C for 4 hours followed by digestion with 0.2 μg of trypsin (Promega) and gentle mixing at 37°C overnight. The digested samples were acidified with 5 μl of 20% trifluoroacetic acid (TFA). Digested peptides were desalted using GL-Tip SDB (GL Sciences, Tokyo, Japan), evaporated in a SpeedVac concentrator, and redissolved in 0.1% TFA and 2% acetonitrile.

Data-independent LC-MS/MS analysis

LC-MS/MS analysis of the resultant peptides was performed on an UHPLC connected to a Q Exactive Orbitrap mass spectrometer (Thermo Fisher Scientific) through a nanoelectrospray ion source (AMR Inc., Tokyo, Japan). The peptides were separated on a 125-mm C18 reversed-phase column with an inner diameter of 100 μm (Nikkyo Technos, Tokyo, Japan) with a linear 5 to 40% acetonitrile gradient for 0 to 100 min, followed by an increase to 95% acetonitrile for 5 min. The mass spectrometer was operated in data-independent acquisition (DIA) mode. MS1 spectra were collected in the range of mass/charge ratio (m/z) 500 to 740 (isolation window width, 10 Da) at 70,000 resolution to set an automatic grain control (AGC) target of 3 × 10^6^. MS2 spectra were collected at 200 to 1800 m/z at 70,000 resolution to set an AGC of 3 × 10^6^, a maximum injection time of “auto,” and stepped normalized collision energies of 22%.

Protein identification and quantitative analysis from MS data

DIA-MS files (raw files) were centroided, and noise was removed using ProteoWizard’s default centroiding algorithm, with parameters optimized for DIA datasets. The files were searched against an in silico mosquito spectral library using DIA-NN (version: 1.8.1). First, a spectral library was generated from the VectorBase-62_AaegyptiLVP_AGWG_AnnotatedProteins database using DIA-NN (62). Parameters for generating the spectral library were as digestion enzyme, trypsin; missed cleavages, 2; peptide length range, 7 to 45; precursor charge range, 1 to 4; precursor m/z range, 495 to 745; and fragment ion m/z range, 200 to 1800. “FASTA digest for library-free search/library generation,” “deep learning-based spectra, RTs, and IMs prediction,” “N-term M excision,” and “C carbamidomethylation” were enabled. DIA-NN search parameters were as mass accuracy, 10 parts per million (ppm); MS1 accuracy, 10 ppm; protein inference, genes; neural network classifies, single-pass mode; quantification strategy, robust LC (high precision); and cross-run analysis without normalization, RT dependent. “Use isotopologues,” “heuristic protein inference,” and “no shared spectra” were enabled.

Matching between runs was turned on for quantitative analysis. Protein identification threshold was set at 1% or less for both precursor and protein false discovery rates (FDRs). The abundance values of quantified peptides and proteins were output as a tsv file. To check the quality of the analysis, tryptic digest of human cell lysate was analyzed before and after sample analysis, with the amount of identified protein found not to decrease between analyses.

Proteomics: Analysis

Most analysis steps were conducted with “Bioconductor” packages (63) in R. The protein abundance data were imported as an MSnSet file, and PCA was conducted using the “plot_pca” function in the “MSnSet.utils” package. DEG analysis was performed with count data (64) using the “limma_contrasts” function with trend = FALSE (prior variance was constant) and robust = TRUE [robust empirical Bayes procedure of (65) used]. These results were visualized using the “ggplot2” package in R.

GO analysis

To conduct GO enrichment network analysis, we used the background R script of “iDEP 2.0” (60). GO terms for GO enrichment analysis were obtained from Ensembl Metazoa [LVP AGWG (aalvpagwg_eg_gene) dataset] using the “RSQLite” package in R. Biological_process GO terms containing less than 10,001 genes were used for all GO analyses. Clusters in Fig. 6 (D, F, and H) were labeled by manually summarizing GO terms based on the predominant types of terms included in each cluster.

Although, in A. aegypti, dyneins were not explicitly annotated as “dynein arms,” relevant annotations related to inner and outer dynein arms exist in An. gambiae (66, 67). Dyneins were selected based on the identification of orthologs of An. gambiae genes annotated with the GO terms, “inner dynein arm assembly (GO: 0036159),” “outer dynein arm assembly (GO: 0036158),” “inner dynein arm (GO: 0036156),” or “outer dynein arm (GO: 0036157).”

Enrichment GO networks were visualized using the “Plotly” package in R.

Weighted correlation network analysis

WGCNA focused on all genes enriched in male pedicels (logFC > 0) at transcriptomic or proteomic levels and used the “WGCNA” package (68) in R.

To construct a network with high scale-free topology, the soft-thresholding power was determined as the lowest value for which the scale-free topology fit R^2^ exceeded 0.8 (set as 0.7 for transcriptomic data and 0.4 for proteomic data; see fig. S7, A and B). Connections with higher adjacency for which the scale-free topology fit R^2^ exceeded 0.9 (threshold: 0.3 to 0.9 for transcriptomics and 0.32 to 0.57 for proteomics) were then screened (fig. S7, C and D).

After constructing the network with all genes enriched in male pedicels, a subset of the network was visualized in the following steps. As GOIs, ciliary-related genes appearing in both transcriptomic and proteomic GO network analyses were screened from the set of significantly up-regulated genes in male pedicels (logFC > 0 and FDR < 0.1). Dyneins were selected based on the identification of orthologs of An. gambiae genes annotated with GO terms related to inner or outer dynein arms. Neighboring hub genes were identified and screened based on the number of network connections to GOIs (tables S8 and S9).

WGCNA networks for either transcriptomic or proteomic data including GOI and neighboring genes directly connected to GOI were visualized using the “visnetwork” package in R.