Microbial interactions in the mosquito gut determine Serratia colonization and blood feeding propensity

How microbe-microbe interactions dictate microbial complexity in the mosquito gut is unclear. Previously we found that Serratia, a gut symbiont that alters vector competence and is being considered for vector control, poorly colonized Aedes aegypti yet was abundant in Culex quinquefasciatus reared under identical conditions. To investigate the incompatibility between Serratia and Ae. aegypti, we characterized two distinct strains of Serratia marcescens from Cx. quinquefasciatus and examined their ability to infect Ae. aegypti. Both Serratia strains poorly infected Ae. aegypti, but when microbiome homeostasis was disrupted, the prevalence and titers of Serratia were similar to the infection in its native host. Examination of multiple genetically diverse Ae. aegypti lines found microbial interference to S. marcescens was commonplace, however one line of Ae. aegypti was susceptible to infection. Microbiome analysis of resistant and susceptible lines indicated an inverse correlation between Enterobacteriaceae bacteria and Serratia, and experimental co-infections in a gnotobiotic system recapitulated the interference phenotype. Furthermore, we observed an effect on host behaviour; Serratia exposure to Ae. aegypti disrupted their feeding behaviour, and this phenotype was also reliant on interactions with their native microbiota. Our work highlights the complexity of host-microbe interactions and provides evidence that microbial interactions influence mosquito behaviour.


Abstract. 32
How microbe-microbe interactions dictate microbial complexity in the mosquito gut is 33 unclear. Previously we found that Serratia, a gut symbiont that alters vector competence 34 and is being considered for vector control, poorly colonized Aedes aegypti yet was Enterobacteriaceae bacteria and Serratia, and experimental co-infections in a 45 gnotobiotic system recapitulated the interference phenotype. Furthermore, we observed 46 an effect on host behaviour; Serratia exposure to Ae. aegypti disrupted their feeding 47 behaviour, and this phenotype was also reliant on interactions with their native 48 microbiota. Our work highlights the complexity of host-microbe interactions and provides 49 evidence that microbial interactions influence mosquito behaviour. 50 51 52

Introduction. 53
Mosquitoes harbour a variety of diverse microbes that profoundly alter host 54 phenotypes 1-3 . In general, the bacterial microbiome can vary considerably between 55 mosquito species and individuals, but within an individual, it is comprised of relatively 56 few bacterial taxa 4,5 . It is becoming more apparent that a variety of factors contribute to 57 this variation, but we have a lack of understanding regarding why some taxa are present 58 in a host, yet others are absent. In mosquitoes and other insects, much effort has been 59 undertaken to characterize the infection status of species and populations for specific 60 endosymbiotic bacteria such as Wolbachia 6-9 , yet few studies have examined the 61 infection prevalence of specific gut-associated bacteria in mosquito vectors. It is evident 62 that several gut-associated bacterial taxa are common between phylogenetically diverse 63 mosquito species 4,5 , but less attention has been paid to identifying incompatible host-64 microbe associations and the mechanism(s) behind this incompatibility. 65 66 Microbiome assembly in mosquitoes is influenced by the environment, host and 67 bacterial genetics, and stochastic processes. While the host is instrumental in 68 maintaining microbiome homeostasis 10-14 , evidence is emerging that bacterial genetics 69 and microbe-microbe interactions also dictate the prevalence and abundance of 70 microbiota 15-18 . It is clear that the microbiome can influence several important 71 phenotypes in mosquito vectors 3,19,20 , including the ability to transmit pathogens. 72 Therefore, a greater appreciation of factors that influence colonization of the mosquito 73 gut could assist our understanding of mosquito phenotypes important for vectorial 74

Selection of S. marcescens antibiotic resistant mutants 140
S. marcescens antibiotic resistant mutants were created as described 50   nucleotides below the quality threshold of 0.05 (using the modified Richard Mott 187 algorithm) and those with two or more unknown nucleotides or sequencing adapters 188 were removed. Reference based OTU picking was performed using the SILVA SSU 189 v128 97% database 54 . Sequences present in more than one copy but not clustered to 190 the database were placed into de novo OTUs (99% similarity) and aligned against the 191 reference database with 80% similarity threshold to assign the "closest" taxonomical 192 name where possible. Chimeras were removed from the dataset if the absolute 193 crossover cost was three using a k-mer size of six. Alpha diversity was measured using 194 Shannon entropy (OTU level), rarefaction sampling without replacement, and with 195 100,000 replicates at each point. Beta diversity was calculated and nMDS plots were  Fluorescence readings were taken at 60°C after each cycle before deriving a melting 208 curve (60-95°C) to confirm the identity of the PCR product. The PCR was carried out on 209 the ABI StepOnePlus Real-Time PCR System. Relative abundance was calculated by 210 comparing the Serratia load to the single copy mosquito gene. 211 212 Life history assays. 213 To determine blood feeding success, mosquitoes were offered a sheep blood meal 214 using a hemotek feeding system. Cups of 25 female mosquitoes were starved for 24 215 hours prior to blood feeding. Mosquitoes were given the opportunity to feed, and then 216 the number of blood fed mosquitoes were counted. For a subset of mosquitoes, the 217 prevalence of S. marcescens in blood-fed and non-blood fed mosquitoes was 218 determined by plating on selective media. To examine the reproductive output, we 219 measured the number of eggs produced by a blood feed female. Individual blood fed 220 females were placed into a vial with an oviposition site. After 4 days, the number of 221 eggs were counted. Females that did not lay were excluded from the analysis. For most 222 assays, the mortality of mosquitoes was quantified daily by counting and removing dead 223 mosquitoes in cups. In addition, bacterial strains were submitted for short-read Illumina sequencing to 30X 241 coverage using the Standard Whole Genome Service from the MicrobesNG service 242 (www.microbesng.com, Birmingham, UK). Assemblies were performed using 243 unicycler 57 , generating a hybrid assembly using both long-and short-read sequences as 244 input for each strain, respectively, for the assembly process. FastANI (average 245 nucleotide identity) was used on a set of Serratia reference genomes retrieved from 246 NCBI ( Table S3) to confirm the species allocation. ANI analysis shows that CxSm1 and 247 CxSm2 are highly similar to Serratia sp. Y25, which likely forms a subspecies of S. 248 marcescens with an average ANI distance of 0.054 (Table S4, Fig S1) file of all variant sites. The option to call genotypes at variant sites was passed to the 262 bcftools call. All bases were filtered to remove those with uncertainty in the base call. 263 The bcftools variant quality score was required to be greater than 50 (quality < 50) and 264 mapping quality greater than 30 (map_quality < 30). If the same base call was not 265 produced by all reads, the allele frequency, as calculated by bcftools, was required to 266 be either 0 for bases called the same as the reference, or 1 for bases called as a SNP 267 (af1 < 0.95). The majority base call was required to be present in at least 75% of 268 reads mapping at the base (ratio < 0.75), and the minimum mapping depth required 269 was 4 reads, at least two of which had to map to each strand (depth < 4, depth_strand 270 < 2). Finally, strand_bias was required to be less than 0.001, map_bias less than 271 0.001, and tail_bias less than 0.001. If any of these filters were not met, the base was 272

Serratia strain characterization. 277
Two strains of Serratia were isolated from Cx. quinquefasciatus by conventional 278 microbiology procedures. 16S rRNA sequencing indicated these strains were S. 279 marcescens, and each produced a red pigmentation when grown in a culture which is 280 indicative of this species (Fig S2A). Although the 16S rRNA sequence was identical 281 between strains, we saw phenotypic differences in their swimming motility, oxidase 282 activity, and capacity to form biofilms, suggesting they were phenotypically divergent 283 (Fig S2A-B). Swimming motility has been implicated in host gut colonization of several 284 hosts 58,59 , and these traits can influence pathogen infection in mosquitoes 60 . To further 285 characterize these strains (named CxSm1 and CxSm2), we sequenced their genomes 286 using nanopore and Illumina technologies. Comparative genome analysis indicated high 287 similarity between the two strains, and both showed 94.7% average nucleotide identity 288 (ANI) similarity to S. marcescens when comparing with a set of Serratia reference 289 genomes, indicating that these might represent a subspecies of S. marcescens (Fig S1, 290 Table S3 and S4). Recent work has indicated a population structure in S. marcescens 291 with at least two different clades 61 , which might be an indication for several subspecies 292 or indeed a species complex, as is for example seen for Klebsiella pneumoniae or 293 Enterobacter cloacae 62,63 .To aid our recovery of each of these S. marcescens strains on 294 media, we selected for rifampicin and streptomycin spontaneous antibiotic resistant 295 isolates for CxSm1 and CxSm2 respectively (antibiotic resistant strains named 296 CxSm1 RifR and CxSm2 SmR ). 297 298

Serratia colonization of mosquitoes 299
We investigated the ability of Serratia to colonize the novel Ae. aegypti host by 300 reinfecting bacteria into mosquitoes in a 10% sucrose meal and monitored infection 301 dynamics in the mosquito over time. The Serratia infection was completely lost from Ae. 302 aegypti by 12 dpi, whereas the bacterial prevalence in the native host, Cx. 303 quinquefasciatus, remained constantly high with infection levels ranging from 100% 304 infection for CxSm1 RifR to 80% infection for CxSm2 SmR at 12dpi (Fig 1A). Of the 305 mosquitoes that were infected, both S. marcescens strains infected Ae. aegypti 306 (Galveston) at significantly lower densities compared to their native host, Cx. 307 quinquefasciatus (Fig 1A). For example, at 3 dpi, we saw approximately 1000 times 308 less Serratia in Ae. aegypti compared to Cx. quinquefasciatus (Fig 1A). We also 309 examined other culturable microbiota by plating mosquito homogenates on non-310 selective LB plates, and in general, we saw few changes in the number of CFUs 311 between groups in either Ae. aegypti or Cx. quinquefasciatus (Fig S3), suggesting 312 Serratia infection had minimal effect on the total bacterial load of culturable microbiota 313 in mosquitoes. The inability of Serratia to persistently infect Ae. aegypti, which was not 314 observed for other bacteria (Fig S3), suggests that barriers, either of bacterial or host 315 origin, were promoting the maladaption between these Serratia strains and this line of 316

Microbial intereaction in the mosquito gut 319
To gain insights into the mechanism promoting the incompatibility between Serratia and 320 Ae. aegypti, we repeated infections in antibiotic treated mosquitoes as we speculated 321 that the native microbiota of mosquitoes might interfere with the colonization of the host 322 (Fig 1B). We formulated this hypothesis as we have previously seen evidence of 323 bacterial exclusion of symbiotic microbes in mosquitoes 4,18 . Strikingly, both CxSm1 RifR 324 and CxSm2 SmR colonized mosquitoes at significantly higher titers when mosquitoes 325 were treated with antibiotics compared to mosquitoes reared conventionally without 326 antibiotics (Mann Whitney test; CxSm1 RifR ; day 3 p < 0.002, day 9 p < 0.01; day 12 p < 327 0.0001, CxSm2 SmR ; day 3 p < 0.03, day 9 p < 0.01; day 12 p < 0.01) (Fig 1B). 328 Furthermore, for both Serratia strains, significantly more mosquitoes were infected at 329 day 12 in antibiotic treated mosquitoes compared to untreated (Fisher's exact test; 330 CxSm1 RifR p = 0.01, CxSm2 SmR p = 0.0007). The levels of Serratia in Ae. aegypti after 331 microbiome homeostasis was disrupted by antibiotics were comparable to infections in 332 the native host Cx. quinquesfasciatus (Fig 1A). These data indicated that the Ae. 333 aegypti (Galveston) line had the capacity to harbor Serratia, and that the incompatibility 334 in mosquitoes with an intact microbiome (Fig 1A, 4 ) was due to members of the native 335 microbiota inhibiting Serratia, as opposed to intrinsic host factors or genetic factors of 336 the S. marcescens strains. 337

338
To determine how widespread these microbial interactions were in Ae. aegypti 339 mosquitoes, we investigated eight diverse lines for native Serratia infections and their 340 capacity to be infected with CxSm1 RifR . When examining the native Serratia load by 341 qPCR, seven of the eight Ae. aegypti lines had significantly lower titers compared to C. 342 quinquefasciatus ( Fig. 2A). Intriguingly, an Ae. aegypti line from Thailand had a high 343 Serratia load that was comparable to the infection in the native Culex host. We also 344 quantified Serratia levels in two other Cx. quinquefasciatus lines and found similar or 345 higher loads of Serratia in these other lines (Fig S4), indicating the robust infection of 346 Serratia in Cx. quinquefasciatus was commonplace. We then infected the CxSm1 RifR 347 Serratia strain into these eight diverse Ae. aegypti lines. For these infections we 348 focused our attention on CxSm1 RifR , as overall, it appears this strain had a greater 349 capacity to infect Ae. aegypti compared to CxSm2 SmR . We therefore posited that this 350 strain would be more likely to infect non-native hosts. Similar to our previous 351 experiments, Serratia poorly infected the Galveston line and was eliminated by 12 dpi. 352 In the other lines, we saw some variation in the time it took for Serratia to be eliminated, constituted 99.5% of the relative abundance indicating that there was negligible 371 contamination in our sequencing (Fig S5). Across all mosquito lines, we identify a total 372 of 1,163 bacterial OTUs, but only 55 were present in mosquitoes at an infection 373 frequency above 1% (Table S5). 374

375
When examining taxa within the microbiome, the majority of sequences were from the 376 Proteobacteria, while others were associated with Verrucomicrobia and Bacteroidetes. 377 Within the Proteobacteria, the most abundant OTUs were in with the families 378 Enterobacteriaceae, Acetobacteriaceae, and Pseudomonasaceae, while the Thailand 379 line harboured a considerable amount of Verrucomicrobiaceae compared to the other 380 three lines (Fig 3A). Confirming our qPCR data, we saw minimal or no Serratia infection 381 in the Galveston, Iquitos, or Juchitan lines, but this bacterium comprised approximately 382 4% of the relative abundance of the Thailand line (Fig 3B). It was also noticeable that 383 the Thailand line possessed many more OTUs compared to the other lines (Fig S6; 384 Table S5). This was corroborated by alpha diversity measures, which indicated the 385 Thailand line had a significantly elevated Shannon's diversity index compared to the 386 other three lines (Fig 3C). To examine the community structure of the microbiome in 387 each line, we undertook non-metric multidimensional scaling (NMDS) analysis based on 388 Bray-Curtis dissimilarity. Strikingly, the microbiomes of each line were significantly 389 different from one other (Fig 3D, p < 0.05), however it was evident from the pattern of 390 clustering of the microbiota, that the Thailand line was considerably divergent compared 391 to the other three lines. 392

393
To examine specific taxa that may be the cause of microbial incompatibility, we 394 undertook pairwise comparisons to identify bacteria that were differentially abundant 395 between lines. We examined differences at the family level using ANCOM, which is 396 specifically designed to handle variable microbiome data 55 . While the abundance of 397 several families was significantly different between lines, the Enterobacteriaceae were 398 the only family that was consistently reduced in the Thailand line compared to the other 399 three lines (Fig 3E). In addition to amplicon sequencing, we used qPCR to determine 400 the total microbial load of mosquitoes and found each possessed a similar density of 401 bacteria (Fig 3F), indicating the increase in taxa in the Thailand line were not simply

Co-infections in gnotobiotic infection model 408
To functionally demonstrate that Enterobacteriaceae interfered with Serratia 409 colonization, we undertook a series of co-infection experiments in antibiotic treated 410 mosquito lines. Prior to infection of the CxSm1 RifR Serratia strain, we infected 411 mosquitoes with Cedecea RifR , an Enterobacteriaceae that commonly infects mosquitoes, 412 or other Acetobacteraceae and Pseuduomonadaceae bacteria as controls (Fig 4A). We 413 chose Cedecea as we have previously documented that this bacterium infects Ae. 414 aegypti effectively 4 . The infection prevalence of Serratia in the co-infected Ae. aegypti 415 Galveston line was significantly reduced in all time points (Fig 4B, p < 0.05, Fisher's 416 exact test). In the few mosquitoes that did harbour a Serratia infection, the density was 417 significantly lower compared to the single infection (Fig 4B, t-test p < 0.05). These data 418 indicated Serratia colonization was inhibited by the presence of Cedecea, and the 419 phenotype we observed previously in conventionally reared mosquitoes could be 420 recapitulated in a gnotobiotic setting. Similarly, we also found the prevalence of Serratia 421 was reduced by co-infection in the Ae. aegypti Thailand line (p = 0.05, Fisher's exact 422 test), although this effect was more subtle, and no significant difference was observed 423 at 12 dpi (Fig 4C). In contrast to co-infection with Cedecea, we found no effect in 424 Serratia prevalence or titers when co-infected with Asaia or Pseudomonas (Fig 4E and  425 F), which are members of the Acetobacteraceae and Pseduomonadaceae families, 426 respectively. Interestingly, there was evidence that Serratia interferes with Asaia 427 infections in Ae. aegypti, as there was an initial reduction in the prevalence of Asaia in 428 the co-infected group compared to the single infection (Fig 4F,  to 50% over the course of the experiment (Fig 5). Despite a lower level of infection, 438 Cedecea infection prior to Serratia reduced the infection of the latter. At 15 and 18dpi, 439 the prevalence of Serratia in the co-infection was 50% compared to 100% in the single 440 infection (Fig 5A, p = 0.03, Fisher's exact test). We also examined the effect of 441 Cedecea on an established Serratia infection by reversing the order each bacterium 442 was administered to the mosquito. In this case, the prevalence of Serratia in the co-443 infection was significantly reduced only at the 18 dpi time point (Fig 5B,

Effect of Serratia exposure on blood feeding behaviour. 448
Anautogenous mosquitoes require a blood meal to acquire nutrition for egg 449 development. Ingested blood alters the gut microbiota composition and abundance, 450 often increasing total bacterial load but decreasing species richness 64,65 . In other 451 mosquito species, Serratia has been seen to rapidly increase in titer after a blood 452 meal 66-68 and, in some cases, can be lethal to the host 36 . As such, we investigated the 453 influence of blood feeding on Serratia infected Ae. aegypti (Fig 6A). We measured 454 bacterial load in the mosquito (Fig 6B) as well as a range of life history traits. For these 455 experiments, we focused our attention on CxSm1 RifR . In contrast to a previous study 36 , 456 we observed no fitness costs to infection in terms of mosquito survival pre or post-blood 457 meal (Fig S7). After a blood meal, Serratia density precipitously increased around 100-458 fold. The increase in the antibiotic treated mosquitoes was more subtle, likely because 459 the bacterial load was initially greater, suggesting there is an upper limit to infections. 460 After blood feeding, Serratia infections were comparable to densities and infection 461 frequencies seen in sugar fed mosquitoes (Fig 1 & 4), with levels in antibiotic treated 462 mosquitoes being maintained at around 1x10 6 bacteria/mosquito. In conventionally 463 reared mosquitoes, Serratia was eliminated, albeit over a longer time period, likely due 464 to the increased density of the bacterium after stimulation from the blood meal. Post 465 blood feeding, Serratia densities equilibrated to levels around 10 6 , which were 466 comparable to infection densities seen in non-blood fed mosquitoes (Fig 6B). While we 467 saw no differences in egg number (Fig S8), in the process of conducting these 468 experiments, we observed that CxSm1 RifR infected mosquitoes were less inclined to 469 take a blood meal when reared on a convention sugar diet. 470

471
We, therefore, investigated whether Serratia infection altered mosquito blood feeding 472 behaviour. After providing mosquitoes with the opportunity to feed, we saw significantly 473 fewer females had imbibed a blood meal compared to uninfected or antibiotic treated 474 CxSm1 RifR infected mosquitoes (Fig 6C, ANOVA p < 0.001). Blood-feeding rates in 475 Serratia infected Ae. aegypti were restored when mosquitoes were fed antibiotics, 476 indicating these behavioural changes were mediated by the interplay between 477 CxSm1 RifR and other bacterial constituents of the microbiome susceptible to antibiotics. 478 Given this intriguing finding, we repeated these experiments with the CxSm2 SmR isolate. 479 Similar to findings with its close relative, the CxSm2 SmR Serratia strain altered the blood 480 feeding rates in mosquitoes (Fig 6D, ANOVA p < 0.001). Given the heterogeneity in the 481 prevalence of CxSm1 RifR and CxSm2 SmR in conventionally reared mosquitoes, we 482 examined individuals that did or did not blood feed for Serratia infection. For both 483 CxSm1 RifR (Fig 6E) and CxSm2 SmR (Fig 6F), the Serratia infection rate was significantly 484 higher in non-blood fed mosquitoes compared to blood fed (CxSm1 RifR p < 0.005; 485 CxSm2 SmR p < 0.005), indicating that mosquitoes that took a blood meal were less likely 486 to be infected with Serratia. When considering this, it is likely the reductions we 487 observed at the population level (Fig 6C & D) are conservative, and the effect of 488 Serratia infection on blood feeding behaviour is more pronounced. 489 490

Discussion. 491
The interplay between the host and microbes can dictate insect microbiome 492 homeostasis, but little is known regarding how microbe-microbe interactions within the 493 gut influence microbial composition and abundance. Previously we identified a Serratia 494 infection gradient in the arboviral vectors, Ae. aegypti, Ae. albopcitus, and Cx. 495 quinquefasciatus, with high loads in the latter and an absence of infection in the former 4 . 496 Here we show that Serratia poorly infects many Ae. aegypti strains and that the 497 mechanism mediating this incompatibility is competitive exclusion from other members 498 of the Enterobacteriaceae, which are close relatives of Serratia. Given that Serratia can 499 influence vector competence in mosquitoes and has been proposed as a microbe for 500 paratransgenic control 66,67 , it is imperative we enhance our understanding regarding the 501 factors that influence Serratia acquisition in the mosquito gut. Galveston line, but these factors were deficient in the Thailand line, which resulted in 531 the more subtle phenotype. In Galleria mellonella, the greater wax moth, both host and 532 bacterial factors synergize to control microbiome composition. When host immunity is 533 suppressed, and mutant symbionts that lack the capacity to produce bacteriocins 534 (proteins that inhibit closely related bacterial strains) are administered to the moth, 535 Serratia proliferates within the microbiome 74 . In Anopheles gambiae, mosquitoes 536 regulate Serratia infections by the complement pathway, and silencing of CLIP genes 537 increases Serratia load, which subsequently induced mortality 75 . Similar to the Oriental 538 fruit fly, in mosquitoes, Duox maintains redox homeostasis which in turn regulates 539 microbiota 13,76 . Taken together, these studies suggest host and microbial factors 540 together maintain microbiome homeostasis in mosquitoes, and when this is disrupted, 541 other taxa that would normally be excluded can proliferate within the microbiome. 542 543 Serratia has pathogenic effects in several insect species 74,77-79 , but this bacterium is 544 also a common taxa within the insect microbiome 23-28 . In hematophagous arthropods, 545 there is variation in Serratia's pathogenicity, ranging from inducing mortality or severe 546 fitness costs on the host under certain conditions to having no observed effect 36,78,80 . 547 We saw little evidence for Serratia affecting mortality or reproduction, similar that 548 observed in Culex mosquitoes 81 . However, intriguingly, our data suggests that Serratia 549 can altered the propensity of mosquitoes to take a blood meal. 550 551 While there are several examples that a broad range of microbes can influence feeding 552 behavior in insects, relatively little is known regarding how gut-associated microbes 553 contribute to these phenotypes. In flies, microbial communities affect affect feeding 554 preference and egg laying behaviour 82,83 , and in mosquitoes, pathogens can alter 555 feeding behaviour. For example, the fungus Metarhizium reduces blood feeding rates in 556 An. gambaie 84 , while arbovirus infections in Aedes mosquitoes can alter feeding 557 phenotypes 85,86 , potentially by altering expression of odorant binding proteins 87 . 558 Alternatively, in several insect systems microbe-mediated alteration in immunity affects 559 feeding behaviour [88][89][90] . Plasmodium infection in Anopheles alters host-seeking 560 response, but similar phenotypes are also induced by microinjection of heat killed 561 Escherichia coli, indicating immune challenge may mediate these behavioural 562 phenotypes 91 . In An. gambiae there is an interplay between S. marcescens and 563 gustatory receptors and odorant binding proteins 92 , and in flies, these gustatory 564 receptors have been implicated in influencing behaviour 93,94 . Our data indicate that 565 Serratia acts in concert with other microbes to reduce blood feeding. There is a complex 566 immune interplay between gut microbes and the host 95-97 , and it is possible that 567 disruption of microbiome homeostatis by Serratia infection may alter basal immunity 568 which subsequently affects feeding behaviour. Alternatively, these mosquitoes may be 569 suffering the effects of infection or microbiome dysbiosis resulting in a lack of interest in 570 feeding. 571 572 From a vector control standpoint, reducing blood-feeding rates will greatly influence 573 pathogen transmission. However, this phenotype is mediated by an interaction between 574 Serratia and other native microbes of the Ae. aegpyti. Given the inherent variability in 575 the microbiome of mosquitoes, further investigations are warranted to determine how 576 universal this phenotype is, and in general how microbiome dysbiosis alters mosquito 577 behaviour that can impact vectorial capacity. In the laboratory setting, reduced feeding 578 rates would act as a distinct mechanism to eliminate Serratia infections from the 579 microbiome of Ae. aegypti.

Acknowledgements. 606
We would like to thank the UTMB insectary core for providing mosquitoes and Alvaro 607 Acosta-Serrano for commenting on a previous draft. GLH was supported by the BBSRC 608      (Table S3)