Dynamic remodeling of lipids coincides with dengue virus replication in the midgut of Aedes aegypti mosquitoes

We describe the first comprehensive analysis of the midgut metabolome of Aedes aegypti, the primary mosquito vector for arboviruses such as dengue, Zika, chikungunya and yellow fever viruses. Transmission of these viruses depends on their ability to infect, replicate and disseminate from several tissues in the mosquito vector. The metabolic environments within these tissues play crucial roles in these processes. Since these viruses are enveloped, viral replication, assembly and release occur on cellular membranes primed through the manipulation of host metabolism. Interference with this virus infection-induced metabolic environment is detrimental to viral replication in human and mosquito cell culture models. Here we present the first insight into the metabolic environment induced during arbovirus replication in Aedes aegypti. Using high-resolution mass spectrometry, we have analyzed the temporal metabolic perturbations that occur following dengue virus infection of the midgut tissue. This is the primary site of infection and replication, preceding systemic viral dissemination and transmission. We identified metabolites that exhibited a dynamic-profile across early-, mid- and late-infection time points. We observed a marked increase in the lipid content. An increase in glycerophospholipids, sphingolipids and fatty acyls was coincident with the kinetics of viral replication. Elevation of glycerolipid levels suggested a diversion of resources during infection from energy storage to synthetic pathways. Elevated levels of acyl-carnitines were observed, signaling disruptions in mitochondrial function and possible diversion of energy production. A central hub in the sphingolipid pathway that influenced dihydroceramide to ceramide ratios was identified as critical for the virus life cycle. This study also resulted in the first reconstruction of the sphingolipid pathway in Aedes aegypti. Given conservation in the replication mechanisms of several flaviviruses transmitted by this vector, our results highlight biochemical choke points that could be targeted to disrupt transmission of multiple pathogens by these mosquitoes.


Introduction
The transmission cycle of dengue viruses (DENV) require a human host and mosquito vector. Mosquitoes acquire DENV via feeding on the blood of an infected human. The blood meal is deposited in the midgut of the mosquito and infection is first established in the midgut epithelium [1]. While digestion of the blood meal is complete within 48 hours [2], viral replication in the midgut tissue reaches its peak only at 7-8 days post-blood meal (pbm) ingestion [1]. Subsequently, the virus disseminates from the midgut and infects other tissues including the fat body and salivary glands. Approximately 10-14 days pbm the salivary glands become infected and the virus can be transmitted in the saliva to a human when the mosquito acquires another blood meal. Since the successful transmission of this virus depends greatly upon its ability to replicate efficiently in several mosquito tissues, the local biochemical and physical environment of each tissue plays a critical role in virus propagation.
In both human and mosquito cells, lipids play an integral role in the life cycle of DENV [3][4][5][6][7]. Host-derived membranes are incorporated into a lipid envelope that surrounds the capsid protein and genomic RNA of DENV particles [8]. This membranous structure facilitates virus release from infected cells by budding into the endoplasmic reticulum and re-entry into new cells through fusion of virus-host membranes [7,9,10]. Additionally, electron tomography has revealed that significant rearrangements of host cell membrane architecture occur upon DENV infection [4,7,[11][12][13]. Virus-induced membrane structures are required as platforms for virus replication and assembly and protect replicating genomes from antiviral defense mechanisms of the host. They have been identified in DENV-infected human and mosquito cells and the midgut epithelium and salivary glands of Culex mosquitoes infected with West Nile virus [14]. This intracellular membrane reorganization imposes a significant metabolic cost to the host cell. It requires activation of biosynthetic processes, and the trafficking and degradation of lipids and other related molecules. Our previous studies investigated the perturbation of lipid homeostasis in C6/36 mosquito cells following infection with DENV [5]. We identified lipids involved in maintaining the stability, permeability and curvature of membranes, as well as bioactive lipid molecules that were significantly changed upon infection. Specifically, a burst of glycerophospholipids was observed coincident with viral replication kinetics. This burst of lipids was attributed to the activity of fatty acid synthase (FAS), a key enzyme in glycerophospholipid biosynthesis. Inhibition of this enzyme was detrimental for DENV replication in both human and mosquito cells [5,6]. Collectively, these data demonstrate that lipids play critical roles in DENV infection in cell culture models.
Although lipid biochemistry has been studied in mosquitoes for several years, very little is known about the relationship between virus and mosquito host and the alteration of intracellular lipids that underpins infection capacity. In this study, we used high-resolution mass spectrometry to explore metabolic changes in the midgut of Ae. aegypti exposed to DENVcontaining blood meals as this tissue represents the crucial site of initial viral replication. Using a time course study, we compared metabolic profiles of infected and uninfected midguts at early-, mid-(peak viral replication) and late time points post-infection. Our results demonstrate significant fluctuations in molecules that function as membrane building blocks, bioactive messengers, energy storage molecules and intermediates in lipid biosynthesis and lipolysis pathways. Presumably these changes represent both manipulation of cellular resources for viral replication as well as the cellular response to infection. They may result from either de novo biosynthesis or the consumption/conversion of metabolites. Additionally, import/export may contribute to the perturbation of metabolite pools. Import pathways are specifically important in the mosquito for molecules such as cholesterol and its derivatives since they cannot be synthesized de novo [15,16]. Many unidentified metabolites (not found in currently available databases that include animal, plant, fungal and bacterial metabolites and synthetic compounds) were also significantly perturbed during virus infection and may represent mosquito-specific metabolites that have not yet been annotated. The flaviviruses transmitted by this vector share similar replication mechanisms; identifying metabolic bottlenecks that condition vector competence and transmission of these pathogens could be exploited for novel transmission-blocking interventions for control of globally important diseases.

DENV type 2 infection of the Ae. aegypti mosquito vector
To profile the alterations in the metabolic environment of the mosquito midgut over time post-blood feeding and during the course of DENV infection in the mosquito vector, Ae. aegypti Chetumal strain mosquitoes were fed an infectious blood meal containing DENV type 2 strain Jamaica-1409 (JAM-1409; blood titer 1x10 7 PFU/mL) or exposed to a noninfectious blood meal as a negative control (S1 Fig). Midguts were dissected from the mosquitoes on days 3, 7 and 11 pbm representing early viral infection in the midgut (day 3), high replication activity in the midgut with early dissemination to the salivary glands (day 7), and high replication activity in salivary glands and other tissues with virus being cleared from the midgut (day 11). Immediately after dissection, individual midguts were tested for DENV RNA by qRT-PCR ( Fig 1A). To ensure that the metabolic profile of the midgut tissues accurately represented the infected environment, every individual midgut was evaluated by qRT-PCR for DENV RNA and only infected tissues were pooled. On day 3 pbm, viral genomes were detected in 55% of the dissected midguts, and on days 7 and 11 pbm, detection increased to 73% and 76% respectively. Plaque assays of whole mosquito homogenates harvested on day 11 pbm (n = 30) indicated that 60% of mosquitoes showed detectable infectious virus averaging 8.5x10 3 PFU/ mL ( Fig 1B). These were similar infection rates to those reported previously by Salazar et al, 2007 [1]. Two independent pools of~100 tissues were included for each treatment group and time point representing a total of 200 biological samples. The metabolic profiles observed in the infected tissues represent true biological deviations corresponding to the infected state. It should be noted however, that estimation of variance and statistical power are compromised by the small sample sizes since the maximum limit of samples acquirable in order to maintain metabolite integrity at a given time point is limited.

LC-MS data analysis of mosquito midguts following DENV infection
Mosquito metabolites in both polar and non-polar phase extracts were analyzed by LC-MS in both positive and negative ionization modes to obtain the most comprehensive coverage possible. Experimental data from each extract in each mode were analyzed separately by the workflow described in materials and methods and shown in S1 Fig. Only features (molecular identities defined by a unique mass to charge ratio, m/z and retention time) detected in both pools (representing 200 biological replicates) were considered 'present' in the samples. For the midguts, a total of 6,103 molecular features from all modes, treatments and time-points pbm were detected. These features were subjected to statistical analysis to compare the differences in metabolite abundance between DENV-infected and uninfected midguts at each time point pbm. The features that showed statistically significant differences in abundance (|log 2 fold change|) ! 1 and p-value < 0.05; total of 936 features) were identified using LIPID MAPS, HMDB, and Metlin databases. Approximately 39% of the significantly different features (363 molecules) were identifiable to an accuracy of < 6 parts per million (ppm) ( Table 1 and S1 Table). About 93% of identified and significantly different features (341 molecules) have recognizable biological relevance. The metabolites detected in the non-polar phase from both negative and positive ionization modes comprised glycerophospholipids (GPs), sphingolipids (SPs), glycerolipids (GLs), sterols, fatty acyls, acyl-carnitines, prenols, polyketides, amino acids and peptides and other organic compounds. The metabolites detected in polar phases from both detection modes also contained the above molecular species with the exception of GLs and contained nucleosides instead. All nomenclatures and categories for lipids are based on the LIPID MAPS comprehensive classification system [17].

Specific lipids levels are elevated during DENV replication in the midgut of Ae. aegypti
To explore the metabolic environment in the mosquito midgut during DENV infection, we compared the metabolite repertoire of DENV-infected midguts to uninfected controls. A summary of the data is shown in Fig 2 and S1 Table. Of 6,103 features detected, 936 features (15%) have differential abundance in DENV-infected midguts compared to uninfected controls in at least one time point (Fig 2A and 2B). The Venn diagram shows the numbers of molecular features with differential abundance in the midgut (|log 2 fold change| ! 1 and adjusted pvalue < 0.05) upon DENV infection on days 3, 7, and 11 pbm. The profiles of 5,167 features were unaltered at any day post-infection (Fig 2A and 2B). Among the features with altered levels, 274, 283 and 114 features showed differential abundance only on day 3, 7 or 11 pbm, respectively, whereas 98 features showed differential abundance in all 3 days (Fig 2B).
Identified metabolites: About 39% of features (363 out of 936) with differential abundance were identifiable through use of three metabolite databases and the putative identifications were categorized into different metabolite classes. The metabolic relationships between the lipid classes are mapped in Fig 2C. Fatty acyl-CoA can be synthesized de novo. It is a precursor that can be modified and incorporated into more complex lipid molecules such as glycerophospholipids, sphingolipids, glycerolipids and sterols. On the other hand, fatty acyl-CoA can be degraded through β-oxidation when cells require energy. The overall trend of lipid molecular levels that changed upon DENV infection is summarized in Fig 2D. A majority of the metabolites were unchanged (had intensity changes that are less than 2-fold upon infection or p ! 0.05), while a majority of those that showed greater than 2-fold changes in intensity had higher abundances upon DENV infection ( Fig 2D and S1 Table, identified tab). Only 61 out of 363 features (~17%) decreased in abundance during DENV infection.
Unidentified metabolites: Interestingly, a large number (~61%) of metabolites that were differentially altered in abundance during DENV infection were unidentified (S1 Table, unidentified tab). A majority of these unidentified metabolites had elevated abundance in the DENVinfected samples compared to uninfected controls at all three time points post-infection. These metabolites represent a potential resource since they highlight molecules that are not found in currently available databases that include animal, plant, bacterial and fungal metabolites as well as synthetic compounds. Therefore, they could be compounds unique to Diptera, mosquitoes or Aedes species that are yet to be annotated. Given their importance to DENV infection, they should be exploited in the future as possible transmission-blocking control points.

Glycerophospholipids (GPs)
GPs are major components of cellular membranes. GPs are composed of a polar head group, a phosphate group and two fatty acyl chains. The specific composition of GPs in a membrane influences membrane fluidity, leakiness, nutrient exchange, assembly and function of signaling protein complexes and vesicular traffic [18][19][20][21]. In insects, the GP composition of membranes also affects tolerance to changing environmental conditions [22]. Pathogenic or endosymbiont infections also drive changes in the intracellular environment and could influence the regulation of GP content in membranes [5,[23][24][25][26]. As a point of reference, the metabolic pathways of the major classes of GPs are shown in Fig 3A. Our study identified 565 species of GPs in Ae. aegypti midguts with 87 species that were altered upon DENV-infection. The alteration landscape and distribution of the different GPs are shown in Fig 3B and 3C.
Minimal enrichment of GP species and levels can be observed on day 3 pbm, during the initial establishment of viral infection in the midgut (Fig 3B). This enrichment was significantly enhanced on days 7 and 11 pbm. These observations were coincident with known viral infection dynamics (peak viral replication in the midgut on day 7 and viral dissemination to other tissues and clearance from the midgut by day 11) [1]. The signaling GPs, PIs, were the most diverse ( Fig 3C) closely followed by PGs, important as surfactants and precursors of cardiolipins in mitochondrial membranes. PCs, PEs and PSs, which are primary components of most cellular membranes were also increased during infection especially at the peak of viral replication (day 7 pbm). Interestingly, levels of most of the lyso-or short chain GPs were decreased during infection.

Glycerolipids (GLs)
GLs, including mono-, di-and triacylglycerols (MAG, DAG and TAG), are critical effectors of energy metabolism in insects and mammals. Fatty acids absorbed from a digested blood meal or synthesized de novo from a sugar meal can be converted into DAG and TAG that can be transported to the fat body for energy storage or the ovaries for vitellogenesis [27-31] (Fig 4). DAG is also an important intermediate in GP synthesis [18, 32] and a critical second messenger regulating cell proliferation, survival, mitochondrial physiology, gene expression and apoptosis [33,34]. Seventy species of GLs were identified in our study, of which eight GLs were significantly changed in abundance upon infection. Most of the MAG, DAG and TAG levels   were higher during early time points (day 3 and 7 pbm) in DENV-infected midguts (Fig 4). This might be a result of transportation of these lipids from storage tissues to support the demand for lipids during infection [35,36]. Interestingly, the level of cytidinediphosphate (CDP-DAG (37:1)) were elevated only during the later phases of infection (days 7 and 11 pbm). CDP-DAG is a precursor for GP synthesis ( Fig 3A). As a result, increased CDP-DAG (37:1) might support the high demand for GP synthesis on days 7 and 11 pbm during virus infection.

Sphingolipids (SPs)
SPs are bioactive molecules that play important roles in the structural composition of cellular membranes and numerous cell-signaling pathways and are critical in microbial pathogenesis [22,[37][38][39][40]. De novo synthesis of SPs occurs in the endoplasmic reticulum through the condensation of L-serine and palmitoyl-CoA to form ceramide (Cer) via several intermediates [41,42]. Cer is a precursor for the biosynthesis of several complex SPs such as sphingomyelin (SM) and glycosylceramides or the production of fatty acids and sphingosines through its hydrolysis [43][44][45]. The SP pathway has been highlighted in many studies in mammalian cells [37] and in mosquito cells [5] as a lipid metabolic pathway altered by flavivirus infection. In this study, using the identified SPs from both infected and uninfected mosquito tissues we reconstructed a significant proportion of the SP metabolic pathway for Ae. aegypti (Fig 5). Several members of the pathway were identified at all three time points in both biological replicates of the infected and uninfected tissues. We observed accumulations of several SPs such as sphinganine, sphinganine-1-PC, sphingosine, Cer and hexosylceramide that were elevated in DENV-infected midguts on days 3, 7 and/or 11 pbm. GP-Cers, especially PE-Cers, are sphingomyelin analogues. They function as the principal membrane sphingolipids in Drosophila, which lacks sphingomyelin synthase [46]. In mosquito midguts, we found decreased abundance of PE-Cer (d32:2(2OH)) upon DENV infection on day 3 pbm but an increase in the levels of PE-Cer(d33:2(2OH)), PE-Cer(d36:1), PE-Cer(d36:3(2OH)) and PE-Cer(d36:2(2OH)) on day 7 and/or 11 pbm. Interestingly, sphingomyelin was found in Ae. aegypti mosquito midguts and in cells in culture (Fig 5 and S2-S4 Figs); however, the levels of sphingomyelin were not changed upon DENV infection at any of the time points tested. Glycosylated ceramides (hexosylceramide and hexosylceramidesulfate) play critical roles in modulating cellular signaling and gene expression resulting in changes in processes such as proliferation, apoptosis, autophagy and endocytosis [47]. The abundance of the neutral glycosphingolipid, FMC-6(d40:1 (2-OH)), was increased on all days and HexCer(d40:1(2OH) sulfate was increased on day 3 following DENV infection ( Fig 5).

Changes in the Cer-DHCer balance impair DENV infection
In this study, we identified a large number of SPs with significantly altered abundance following DENV infection of mosquito midguts. Important biological roles of the SP pathway are determined by conversion from DHCer to Cer, the central precursor of diverse SPs. The balance between Cer and DHCer concentrations plays a critical role in physiologically regulating properties such as membrane architecture, fluidity and function [48]. Therefore, we investigated if alteration of Cer or DHCer levels and the Cer/DHCer ratio had an effect on DENV infection, using Ae. aegypti-derived (Aag2) cultured cells, which are readily amenable to loss of function studies. Cells were treated with 4-hydroxyphenyl retinamide (4HPR) or long double stranded RNA (dsRNA) to inhibit activity of or knock down expression of sphingolipid Δ-4 desaturase (DEGS, VectorBase: AAEL013047, NCBI:LOC5577178), the enzyme that catalyzes conversion of DHCer to Cer, the final step in the de novo biosynthesis of Cer (Fig 5), and thus should greatly influence the Cer/DHCer ratio [49]. 4HPR is a synthetic retinoid that activates the Cer de novo synthesis pathway prior to DHCer synthesis, but inhibits the activity of DEGS We also did not see changes in sphingosine (d18:1), sphingosine-1-P (d18:1-P), sphinganine (d18:0) or SM levels following 4HPR treatment which is expected as the enzymes that produces these molecules are not the primary targets of 4HPR (S2F and S2G Fig). The effect of 4HPR on DENV replication therefore might not be due to a direct effect on the Cer/DHCer ratios, but due to off target effects such as ROS, changes in bcl-2, p53 mRNA expression and apoptosis [51]. A similar observation has been made in mammalian cells [50]. To validate that the reduction of DENV replication was caused by the loss of SP metabolites or Cer/DHCer imbalance and not due to off-target effects of the inhibitor, the expression of DEGS in Aag2 cells was transiently knocked down by RNA interference, using long dsRNA (Fig 6B-6E). In DEGS knockdown (DEGS-KD) cells, DENV titer and genome replication were significantly reduced compared to the GFP dsRNA (GFP-KD) negative control, but were similar to the DENV dsRNA (DENV-KD) positive control (Fig 6B and 6C). SP metabolic profiling of the DEGS-KD cells compared to GFP-KD cells revealed striking reduction of Cer(d18:1/ 18:0), DHCer(d18:0/20:0) and DHCer(d18:0/22:0) after DEGS-KD (Fig 6F and 6G, lower panel). Interestingly, DEGS-KD also greatly altered the Cer/DHCer ratio for molecules with 18-and 16-carbon long chain sphingoid bases, especially those with very long fatty acyl chain lengths (20:0-24:0) ( Fig 6F and 6G, upper panel). These results indicated that perturbation of the Cer-DHCer balance reduced both DENV genome replication and infectious virus formation. As expected we did not see changes in sphingosine (d18:1), sphingosine-1-P (d18:1-P), sphinganine (d18:0) or SM levels during dsRNA treatment (S3 Fig). When DENV-infected and uninfected Aag2 cells were compared for alterations in the SP pathway, we observed a significant increase in the abundance of Cer(d18:1/20:0) and Cer(d18:1/24:1) and DHCer(d18:0/ 22:0) in infected cells, but a decrease in Cer(d16:1/16:1) in infected cells (S4A and S4B Fig,  lower). These changes did not, however, affect the Cer/DHCer ratio for very long fatty acyl chain containing molecules (S4A and S4B Fig, upper). We did not see changes in sphingosine (d18:1), sphingosine-1-P (d18:1-P), sphinganine (d18:0) or SM levels during dsRNA treatment (S4C and S4D Fig).

Fatty acids and derivatives (Fatty acyls)
Fatty acyls can be acquired from the midgut digestion and absorption of a blood meal or can be de novo synthesized from central carbon metabolism precursors [27,52]. Fatty acids that are covalently linked to coenzyme A, fatty acyl-CoAs, can be incorporated into complex lipids such as GPs, SPs and GLs that have structural roles in membranes. In addition, free fatty acids or fatty acyls that are linked to other functional groups can have critical roles in signaling, energy homeostasis and the immune response in insects and mammals (e.g.: fatty amides and eicosanoids) [53][54][55]. In this study, at least nine subclasses of fatty acyls were detected and many of these had higher abundances in DENV-infected midguts compared to uninfected controls in at least one time point pbm (S5 Fig and S1 Table). The molecules observed were fatty amides, hydroxy fatty acids, free fatty acids, eicosanoids and leukotriene (i.e. immunomodulators), fatty-amines, glycosides, dicarboxylic acids, and keto-fatty acids. Fig 7 shows molecules with significantly increased or decreased levels at least at one time point following DENV infection and have known functions in immunomodulation ( Fig 7A) and N-acylamides and other fatty acyls whose accumulation is a marker for malfunction of fatty acid oxidation ( Fig 7B). Prostaglandins, leukotrienes, thromboxanes and lipoxins are eicosanoids that are synthesized in various tissues. Eicosanoids are inflammatory mediator molecules have been found to play several important roles in insect immunity [55]. In this study, three prostaglandins (prostaglandin a2, prostaglandin d2, and PGD2-dihydroxypropanylamine) and one thromboxane (dehydrodinor-TXB2) showed increased levels during DENV infection and one leukotriene (leukotriene e4) decreased during DENV infection. Anti-inflammatory molecules (resolvin d5 and epoxy-DHA) increased during DENV infection [56, 57]. Metabolites of ceramide; HexCer, hexosylceramide; FMC-6, acetyl-sphingosine-tetraacetyl-GalCer(d40:1(2OH)); PE-Cer, phosphatidylethanolamine-ceramide; Phytosphinganine-1-P, phytosphinganine-1-phosphate; PI-Cer, phosphatidylinositol-ceramide; Sphingosine-1-P, sphingosine-1-phosphate; Sphingosince-1-PC, sphingosine-1-phosphatidylcholine and SM, sphingomyelin. https://doi.org/10.1371/journal.ppat.1006853.g005 Dengue virus alters lipid homeostasis in Aedes aegypti mosquitoes linoleic acid that may regulate prostaglandin synthesis (HEPE, HpOTrE and TriHOME) also increased upon DENV infection [58]. N-acylamides play important roles in endocannabinoid signaling systems in mammals and have been detected and reported as potentially important signaling molecules in Drosophila [59]. Elevated levels of these molecules are also observed in the urine and blood of human patients with fatty acid oxidation disorders [60]. In our study, we found that five N-acylamide molecules (e.g. N-arachidonoylglycine, myristoylglycine, Nstearoylarginine, N-decanoylglycine, and N-undecanoylglycine) had increased levels and Nheptanoylglycine displayed a decreased level in midguts during DENV infection (Fig 7B).

Acyl-carnitines
One of the most prominent observations in this study is the change in abundance of acyl-carnitines in midgut tissue after DENV infection (Fig 8). These molecules are intermediates that shuttle fatty acyl-CoA from the cytoplasm into the mitochondria for β-oxidation and subsequent energy (ATP) production (Fig 8). A total of 54 acyl-carnitine molecules were detected in the midgut metabolome. Following DENV infection, 26 acyl-carnitines had a significant increase in abundance and only one had decreased abundance ( Fig 8A). Interestingly, of those that increased in abundance, 25 out of 26 molecules contained medium chain-length fatty acyls of 4-12 carbons. The accumulation of acyl-carnitines with medium-length carbon chains was reported to be a result of incomplete β-oxidation [61].

Sterol lipids
Sterol lipids are a group of cyclic organic compounds composed of 17 carbon atoms arranged in a four-ring structure. In insects, cholesterol is a component of cellular membranes and a precursor of the ecdysone hormone [15, [62][63][64][65]. However, insects cannot synthesize cholesterol de novo but must absorb it from dietary sources and/or microflora [15,16,66]. During infection of mammalian cells, cholesterol in the flavivirus envelope seems to be critical for viral entry and fusion [67,68]. It has also been shown that flaviviruses can manipulate cellular cholesterol homeostasis to facilitate genome replication [69]. However, the role of cholesterol in flavivirus replication in the mosquito vector has not been determined. In this study, we detected 111 sterol lipid molecules. Many of these have no reported function in insects. Since sterols in the mosquito are acquired from dietary sources (in this case, raisins, sucrose, sheep's blood and cell culture medium), we did not exclude sterol lipids identified as plant or other animal from our analyses. Of the 111 molecules, 25 showed different levels of abundance during DENV infection compared to controls (S1 Table). Twenty-one molecules increased during infection while only four molecules decreased during infection. A majority of the changes occurred on day 3 (10 molecules) and day 7 (14 molecules) pbm. Only one molecule showed significant changes (decreased) on day 14 pbm.

Discussion
The dynamic metabolic environment of an organism is an indicator of physiological changes that occur following exposure to varying environmental conditions. When arboviruses such as DENV infect a mosquito vector, they must strike a delicate balance between metabolic

DEGS-KD (red) and GFP-KD cells (blue) using LC-MS/MS Multiple Reaction Monitoring (N = 3). (F) and (G) shows Cer and DHCer molecules with
18-and 16-carbon long chain sphingoid bases, respectively. Lower panels show the abundance of individual species of Cer(d18:1/xx:x or d16:1/xx:x) or DHCer(d18:0/xx:x or d16:0/xx:x) where xx:x refers to the fatty acyl chain attached to each sphingoid head group. The upper panels show the ratio of Cer/DHCer species with the same fatty acyl chain lengths. These ratios highlight the Cer-DHCer imbalance caused by manipulation of DEGS gene expression. Student's t-test was applied for statistical analysis. Ã p < 0.05, ÃÃ p < 0.01, ÃÃÃ p < 0.005, ÃÃÃÃ p < 0.001.
https://doi.org/10.1371/journal.ppat.1006853.g006  Dengue virus alters lipid homeostasis in Aedes aegypti mosquitoes commensalism and competition to achieve persistent replication in mosquito tissues to allow transmission to a new host. Here, we have explored the metabolic landscape of Ae. aegypti, the primary vector of DENV, ZIKV, YFV and CHIKV, during infection with DENV. Our study focused on the midgut, the site of initial viral replication prior to viral dissemination to other tissues and transmission to the vertebrate host [1]. The observed biochemical changes highlight specific metabolites that may be markers of infection. They may be required for viral replication or produced as part of the defensive response of the vector to the invading pathogen. We have consistently observed a link between DENV infection and an increased abundance of GPs; our previous work on DENV-infected human cells showed that fatty acid synthase (FAS), a key enzyme required for fatty acid synthesis (essential pre-cursors of GP synthesis), is activated by the expression of virus-encoded nonstructural protein NS3, and relocates to sites of viral RNA replication through interaction with NS3 [6]. Inhibition of FAS activity using C75 caused a reduction in viral replication and we observed this in both cultured human and mosquito cells [5,6]. We profiled the metabolic changes upon C75 inhibition of FAS in mosquito cells and demonstrated that the GP repertoire was directly reduced coincident with the reduction in viral replication. Therefore, in the cell culture models an abundance of GPs is required for viral replication. The current study supports these observations and reveals that DENV infection also significantly alters the abundance of GPs and other lipid-related molecules in the midgut of the mosquito vector during infection (summarized in Fig 9). While our previous studies showed that de novo synthesis of lipids plays an important role in DENV replication, other processes such as lipolysis, lipid conversion/consumption or import/export may also critically shape the metabolite pool observed in these infected tissues [70,71].

Glycerophospholipids (GPs)
One of our most striking observation was the marked elevation of GP levels coincident with the kinetics of viral replication in the midgut [1]. Digestion of the blood meal, is completed within~48 hours post-ingestion, and provides a major source of amino acids and fatty acids used for the biosynthesis of macromolecules [74,75]. These macromolecules are required for vitellogenesis (yolk formation) and energy storage [29, [76][77][78]. In DENV-infected mosquitoes, the rapid increase in GP content is observed at day 7 pbm when viral replication reaches its peak in the midgut. While it is currently unknown if biosynthesis alone contributes to this burst in GP content the increase is consistent with requirements for DENV replication within human and mosquito cells in culture [5,6]. Specifically, DENV infection induces a massive expansion of intracellular membranes required for the assembly and function of viral replication complexes and assembly and release of enveloped progeny virions [4,7]. This membrane expansion is coincident with increased GP biosynthesis through the co-opted functions of FAS, a key enzyme in the pathway. Electron microscopic studies have revealed similar membrane expansions associated with the midgut of WNV-infected Culex mosquitoes but equivalent studies have not been performed in Ae. aegypti infected with DENV [14].

Sphingolipids (SPs)
Reconstruction of the SP pathway in Ae. aegypti revealed significant conservation with that of Homo sapiens, allowing us to identify a majority of orthologous metabolites in the pathway. Importantly, most metabolites were identified in both pools of 100 midguts, at all time points and in both infected and uninfected tissues confirming a significant presence in the midgut metabolome. Additionally, we were also able to identify orthologous effector enzymes for a majority of the SP metabolic pathways in our study using annotations in the AaegEL5 assembly of the Ae. aegypti genome on VectorBase [79]. SPs are important components of lipid rafts in membranes that in mammalian systems, together with cholesterol support the assembly and function of various protein signaling complexes [80]. Metabolites of SPs are well-studied signaling molecules [41]. These lipids have been studied in Drosophila [23, 81,82] and it has been demonstrated that these lipids are critical for development, reproduction and maintenance of tissue integrity in the fruit fly [83]. SPs have also been identified in Anopheles stephensi by spatial mapping [84] and in Aedes [5] and Culex cells in culture [85]. However, the specific This study revealed that the abundance of lipids in six main lipid metabolic pathways [72] was altered following DENV infection of the mosquito midgut (names of lipids that were altered in abundance during infection are in black (bold) and lipids that remained unchanged are in grey). All of these pathways require fatty acyl-CoA (FA-CoA), activated fatty acids, as a precursor for the synthesis of complex lipids (red). Names of enzymes involved in FA-CoA homeostasis are italicized (FAS, ACOT and ACSL). Previous studies by Heaton et al, 2010 [6] and Perera et al, 2012 [5] have shown that DENV infection requires and can enhance the activity of FAS to synthesize in human and mosquito cells. Ye et al, 2013 [73] have shown that Wolbachia , the insect endosymbiont used to control DENV transmission by Ae. aegypti mosquitoes can increase FA-CoA catabolism by increasing the expression of ACOT enzymes. Therefore, FA-CoA metabolism represents a 'hub' that may control lipid metabolic competition in the midgut that mediates the success of DENV replication. Abbreviations: ACC, acetyl-CoA carboxylase; ACSL, long-chain acyl-CoA synthetase; ACOT, acyl-CoA thioesterase; CDP-DAG, cytidine diphosphate diacylglycerol; Cer, ceramide;; CL, cardiolipin; DAG, diacylglycerol; DHCer, dihydroceramide; G-3-P, glycerol -3-phosphate; GP-Cer, glycerophospholipid-ceramide; FAS, fatty acid synthase; HexCer, hexosyl ceramide; LPA, lysophosphatidic acid; MAG, monoacylglycerol; PA, phosphatidic acid; PC, phosphatidylcholine; PE, phosphatidylethanolamine; PI, phosphatidylinositol; PG, phosphatidylglycerol; PS, phosphatidylserine; SM, sphingomyelin; sphinganine-1-P, sphinganine-1-phosphate; sphinganine-1-PC, sphinganine-1-phosphatidylcholine; sphingosine-1-P, sphingosine-1-phosphate; sphingosine-1-PC, sphingosine-1-phosphatidylcholine; TAG, triacylglycerol and Wol, Wolbachia.
https://doi.org/10.1371/journal.ppat.1006853.g009 functionality of these lipids has not been extensively studied either in vitro or in vivo in mosquitoes. All viruses and most bacteria are incapable of SP synthesis and rely on the host to provide these lipids. The specific species of SPs that are increased in abundance during infection could influence the balance between commensalism, competition and/or host damage [37,86].
In our data, the most striking observation is that several SP metabolites that are precursors or derivatives of Cer showed increased abundance in DENV-infected midguts compared to controls. In Ae. aegypti-derived cells we observed a significant increase in the abundance of select Cer and DHCer species and a reduction in other Cer species during infection. Importantly, altering the ratio of Cer to DHCer (through inhibition of DEGS) significantly reduced viral genome replication and infectious virus release, highlighting the importance of this hub for infection. In our previous studies in C6/36 Ae. albopictus mosquito cells, we also observed elevated levels of Cer and SM during DENV infection [5]. An important observation from the three systems studied thus far (Ae. aegypti mosquito midgut tissue, Ae. aegypti-derived Aag2 cells and Ae. albopictus-derived C6/36 cells) is that the SP synthetic pathway is significantly activated during infection, with the Cer concentration being a focal point. In contrast, work by Aktepe et. al., on mammalian cells showed that DENV replication was enhanced upon the inhibition of Cer synthesis [86]. Therefore, while a critical focal point during infection, the SP synthetic pathway requires further investigation to determine how the Cer hub and other intermediates in the SP metabolic pathway might support or limit infection.

Mitochondrial β-oxidation and metabolite trapping
During mitochondrial β-oxidation activated fatty acids (FA-CoAs) in the cytoplasm are esterified with carnitine to produce acyl-carnitines that are then shuttled to the mitochondrial matrix for energy production [87][88][89]. In this study, we observed a temporal accumulation of medium-length carbon chain-containing acyl-carnitines in DENV-infected midguts compared to uninfected midguts. The accumulation of medium-length carbon chain acyl-carnitines was reported to be indicative of incomplete β-oxidation [90]. Given that DENV replication requires FA-CoAs, for lipid biosynthesis and subsequent membrane expansion, it is possible that acyl-carnitine accumulation results from a manipulation of β-oxidation (an increase or decrease) in order to provide for viral replicative needs [5,71,86].
There are two possible hypotheses to explain the accumulation of medium-chain length acyl-carnitines during DENV infection (Fig 10). In hypothesis I, accumulation of acyl-carnitines might be caused by inhibition or blockage of β-oxidation in the mitochondria reducing the production of ATP (Fig 10I). This phenomenon has been observed in cells exposed to hypoxic conditions [91][92][93]. Similar effects were observed during DENV infection of human cells; in infected HepG2 cells, mitochondrial membrane damage appeared to lead to a reduction in ATP levels [94]. To maintain energy homeostasis, energy consumption is diverted towards increasing glucose uptake and glycolysis [91]. This response has also been observed in DENVinfected primary human cells [95]. Due to the stalling of acyl-carnitine transport and a blockage of β-oxidation, FA-CoA lipid partitioning may occur to divert FA-CoAs towards a synthesis of complex lipids and membrane expansion at the expense of fatty acid oxidation [92,95,96].
Alternatively, in hypothesis II, an accumulation of medium-length carbon chain acyl-carnitines could occur because of mitochondrial overload (Fig 10II). In this scenario, due to increased energetic demands during infection, a large proportion of FA-CoAs enter mitochondria but are only partially broken down by β-oxidation, which serves as a bottle neck for this reaction. This phenomenon was observed in human skeletal muscle insulin resistance [61]. The ATP that is produced by β-oxidation is sufficient to supply the energetic needs for producing more virions, and acetyl-CoA (2 carbon product) can be recycled for de novo synthesis of new longer-chain FA-CoA molecules directly at the site of virus replication [6]. This hypothesis is supported by Heaton et al. during DENV infection of Huh7.5 human liver cells, where they observed an increase in autophagosome processing of cellular lipid droplets and TAGs to produce free fatty acids and a stimulation of mitochondrial β-oxidation [71]. The molecular mechanisms of how mitochondrial energetics are diverted to benefit DENV replication in the mosquito midgut are still unclear and the observations in this study present the first insight into the possibilities.

Metabolic competition versus commensalism
FA-CoAs (activated fatty acids) form a primary hub that integrates multiple lipid metabolic pathways, some of which were found to be perturbed in this study during DENV infection (Fig 9). These molecules are key to the synthesis of more complex lipids, which are important for DENV replication in both human and mosquito cells [4][5][6][7]. FA-CoAs are also important for activating several signaling pathways in the cell [97,98]. Interestingly, there is evidence that Wolbachia, the endosymbiotic bacterium used to control DENV infection in Ae. aegypti, might drive this metabolic balance in the opposite direction (FA-CoA to free fatty acid) suggesting opposing metabolic driving forces resulting in a competition between virus and endosymbiont [73]. Recent studies in Wolbachia-infected, Ae. albopictus (Aa23) cells have supported this hypothesis and indicated that several lipid groups such as SP, PC, and DAG previously shown to be elevated in DENV-infected C6/36 cells were depleted upon Wolbachia infection of the cells [5,23,24].
The hypothesis that metabolic challenge rather than immune pathways form the basis for pathogen-blocking capabilities of Wolbachia is further supported by observations by Caragata et al, 2013 [99] who showed that cholesterol might also be a limiting factor that drives metabolic competition between virus and endosymbiont. Specifically, increasing cholesterol availability via an enriched diet increases virus replication in the insect and reduces the impact of Wolbachia-mediated pathogen suppression. Mosquitoes are incapable of de novo synthesis of  100], and must sequester these molecules from the blood meal or from microbiota that are capable of synthesizing these lipids. In the presence of Wolbachia, it is possible that cholesterol "stealing" from the vertebrate blood meal may make cholesterol unavailable for DENV infection/replication. In mammalian systems, it has been shown that cholesterol is critical for DENV replication [67,68], and therefore, it is interesting to contemplate what might happen in the mosquito where these lipids are limited. Alternately, other species of lipids could substitute for the function(s) of cholesterol during DENV replication in the mosquitoes. In this study, we detected over hundred cholesterol molecules with one fourth of the molecules showing alterations during early phases of DENV replication in the midgut. The specific role these molecules might play in the virus-vector interactions yet remain to be explored.

Comparison to cell culture models
These in vivo studies are the obvious next step to previous studies that have evaluated metabolic changes induced by DENV infection in cell culture models. Interestingly, in both human [6] and mosquito cell cultures models [5] we observed a significant increase in GP expression during DENV infection. We determined that this burst of GP abundance is due to the activity of FAS that seems to be stimulated and re-localized to sites of viral RNA replication by DENV nonstructural protein 3 (NS3) [5]. We anticipated that this might also be true in vivo. Accordingly, we saw a significant increase in GPs in Ae. aegypti midguts that directly coincides with the peak of viral replication (similar to that observed in both human hepatoma and C6/36 cells) indicating that it could be a key modulator of DENV infection. In mosquitoes, in vitro (C6/36 cells) we did not see significant PG/PI involvement the during infection. However, we do see activity in mosquito midguts following infection. The concentrations of PG lipids are reportedly lower in eukaryotic membranes but higher in bacterial membranes, and therefore could be elevated in the midguts by contributions from the microbiome [101]. However, to accurately compare the specific GP landscape with respect to true concentrations of head groups, extensive targeted analyses will need to be carried out in the future. Lysophospholipids (LysoGPs) also seemed more heavily perturbed in C6/36 cells compared to mosquito midguts with evidence that phospholipase A2 may have increased activity during DENV infection [5]. A striking observation in this study is that among the GPs that were significantly altered in abundance during infection, lysoGPs were prominently decreased in abundance in infected midguts while they were increased in abundance in infected C6/36 cells. These lipids are known to be stress-related as well as to function as precursors and signaling molecules that also play important roles in influencing membrane architecture [102,103]. These different response patterns in vitro and in vivo indicate a more complex situation in the whole animal and is evidence that in vitro observations are not always translatable to the in vivo environment.
Sphingolipids (SPs) also showed differences between the cell culture and in vivo systems. While in C6/36 cells some of the major SPs like SM and Cer were perturbed during infection, in the midgut tissues, we saw both precursors and downstream products of Cer metabolism as well as complex SPs being perturbed. Interestingly, while several SMs were detected, we did not see any significant changes in SM abundance during infection. The observation of SPs in Aag2 cells was rather uninformative, with only a few Cer and DHCer molecules showing significant differences in abundance during infection. However, alteration of the Cer to DHCer ratio in these cells by RNAi knockdown significantly reduced DENV genome replication and infectious virus release. These observations suggested that there may be a battle between the virus and the host to regulate the Cer hub that results in keeping most of the SPs in balance. Actively changing this hub can significantly impact the viral life cycle. A caveat is that when comparing metabolic responses of these different in vivo and in vitro models systems to DENV infection, it is important to take into account the inherent differences between the systems that might impact this response. For example, C6/36 cells, are a cell line derived from Ae. albopictus larvae (a different species to Ae. aegypti), and are defective in the RNAi pathway [104]. The Aag2 cell line, while derived from Ae. aegypti embryos are persistently infected with cell fusing agent virus (CFAV) [105]. Our in vivo studies were carried out in adult female Ae. aegypti midgut tissue, representing a much more complex and biologically relevant cellular environment.

Summary
We present the first comprehensive analysis of the metabolome of Ae. aegypti, the primary mosquito vector of dengue, Zika, chikungunya and yellow fever viruses, during DENV infection. Alterations in the metabolome are quantitative indicators of the outcome of arbovirus-host interactions that facilitate virus replication or may result from host responses to infection. Our study provides evidence that DENV infection specifically alters the lipid repertoire at the initial site of viral replication, the midgut of the mosquito. Since this study was carried out using a highly susceptible mosquito strain with a laboratory-adapted virus, it highlights metabolic pathways that may be critical for achieving optimal levels of replication in the midgut required for successful dissemination and ultimately transmission. Given that vector competence for infection and transmission is dependent on the specific virus-vector pairing, it will be critical to evaluate the metabolic environment under conditions where the virus has a less robust infection rate and/or the vector presents infection and transmission barriers. This is particularly important in field settings where vector competence is also influenced by environmental conditions, co-infecting pathogens and the microbiome of the vector. In addition, there is a growing body of evidence that metabolic competition between arboviruses and endosymbionts such as Wolbachia may form the basis of endosymbiont-mediated control of viral transmission. Therefore, exploring vector metabolism is a powerful and integrative means to identify metabolic control points that could be exploited to interfere with pathogen transmission or to implement vector control.

Virus
DENV serotype-2 strain Jamaica 1409 (JAM-1409) was obtained from the Centers for Disease Controls and Prevention (CDC), Fort Collins, CO, USA [106]. The virus was passaged in C6/ 36 cells cultured in L15 medium supplemented with 3% fetal bovine serum (FBS), 50 μg/mL penicillin-streptomycin, and 2 mM L-glutamine by infecting the cells at a multiplicity of infection (MOI) of 0.01. Fresh medium was replaced at 7 days post-infection. At 12-14 days postinfection, virus-containing supernatant and infected cells were harvested to prepare the infectious blood meal.

Mosquito rearing
Ae. aegypti strain Chetumal was originally collected from Yucatan Peninsula, Mexico [107]. Adult mosquitoes were fed on raisins and water, with uninfected blood meals to stimulate oogenesis, and were maintained at 28˚C, 80% relative humidity with 12-12 h light-dark periods. Male mosquitoes (20-25 males) were placed in one-pint cartons with 200-250 female mosquitoes to maintain the colony.

Mosquito infection with DENV
Mosquitoes were orally exposed to a DENV-infectious blood meal using an artificial membrane feeder as described previously [108] but raisins and water were only removed at 24 and 4 hours prior to blood feeding. For the uninfected control group, we mixed uninfected C6/36 cells suspended in cell culture medium with blood meal to maintain as similar a metabolic input to the DENV-infected group as possible. Mosquitoes were allowed to feed for 45-60 mins. Blood-engorged mosquitoes were reared up to 11 days and fed on sucrose and water.

Sample preparation for metabolomics analysis
Mosquito dissection. On days 3, 7 and 11 pbm, midguts were dissected from 200 uninfected and 300 DENV-infected mosquitoes. Individual midguts were ground in 100 μl phosphate buffered saline (PBS). Two microliters of each individual DENV-exposed midgut (n = 799) homogenates were collected in a separate tube for qRT-PCR analysis. The rest of the homogenate was immediately frozen on dry ice and stored at -80˚C for metabolite extraction.
Quantitative RT-PCR detection of infected mosquitoes. To determine the presence of DENV RNA in the midgut of mosquitoes following an infectious blood meal, 2μL of midgut homogenate was heated at 95˚C for 5 mins and chilled on ice prior to performing a one-step quantitative RT-PCR (qRT-PCR) reaction without RNA extraction. Presence of viral RNA was quantified using Brilliant III Ultra-Fast SYBR qRT-PCR Master Mix (Aligent Technologies, CA) and with the primer pairs for DENV2 RNA as described previously [109]. Quantitative RT-PCR was performed using a iQ5 real-time PCR detection system using the following cycles: RT at 50˚C, 20 min; 95˚C, 5 min for RT inactivation and 40 cycles of 95˚C, 5 s and 60˚C, 1 min. The results were analyzed with the iQ5 optical system software version 2.0 (Bio-Rad, CA).
Metabolite extraction. Two equal pools of 75-100 midgut homogenates for DENVinfected and mock infected treatment groups were subjected to metabolic extraction. Internal standards (10 μg Alanine-Leucine-Alanine-Leucine and 10 μg Ceramide d17:1, Avanti, AL) were added to the pools for normalization of the variation introduced during tissue extraction. To extract metabolites, one volume of methanol and chloroform and 0.01% butylated hydroxyl toluene were added to an equal volume of homogenous tissue in PBS [110]. Samples were vortexed thoroughly, prior to separation of the aqueous phase from the organic phase by centrifugation at 1,250 x g for 10 minutes. To prevent degradation of metabolites, samples were processed on ice or at 4˚C throughout the process. Metabolites from aqueous and organic phases were dried separately by speed-vacuum centrifugation and were stored at -80˚C for further metabolomics analysis.

Liquid chromatography-Mass spectrometry (LC-MS) profiling analysis of metabolites
Two technical replicates of each pool of midguts were analyzed by a LTQ Orbitrap XL instrument (Thermo Scientific, Waltham, MA). It was coupled to an Agilent 1100 series LC (Agilent Technologies, Santa Clara, CA) equipped with a refrigerated well plate auto sampler and binary pumping device.
For polar metabolite analysis, reverse phase liquid chromatography was used to analyze the samples. An Atlantis T3 column (Waters Corp., Milford, MA) with 1.0 x 150 mm, 5.0 μm dimensions was used for the separation. Solvent A consisted of water + 0.1% formic acid. Solvent B consisted of acetonitrile + 0.1% formic acid. The flow rate was 140 μL/minute. A volume of 5 μL was loaded onto the column. The gradient was as follows: time 0 minutes, 0% B; time 1 minutes, 0% B; time 41 minutes, 95% B; time 46 minutes, 95% B; time 50 minutes, 0% B; time 60 minutes 0% B. The LC-MS analysis was run twice, using positive and negative polarity electrospray ionization (ESI). Data were acquired using data dependent scanning mode. FTMS resolution of 60,000 with a mass range of 50-1100 was used for full scan analysis and the FTMS was used for MS/MS data acquisition with a resolution of 7500 and collision induced dissociation (CID) mode.
For non-polar metabolite analysis, reverse phase liquid chromatography was also used to analyze the samples. An Xterra C18 column (Waters Corp., Milford, MA) with 2.1 x 150 mm, 5.0 μm dimensions was used for the separation. Solvent A consisted of water + 10mM ammonium acetate + 0.1% formic acid. Solvent B consisted of acetonitrile/isopropyl alcohol (50/ 50 v/v) + 10mM ammonium acetate + 0.1% formic acid. The flow rate was 300 μL/minute. A volume of 10 μL was loaded onto the column. The gradient was as follows: time 0 minutes, 35% B; time 10 minutes, 80% B; time 20 minutes, 100% B; time 32 minutes, 100% B; time 35 minutes, 35% B; time 40 minutes 35% B. The LC-MS analysis was run twice, using positive and negative polarity in ESI. For MS/MS experiments, data were acquired using a data dependent scanning mode. FTMS resolution of 60,000 with a mass range of 100-1200 was used for full scan analysis and the FTMS was used for MS/MS data acquisition with a resolution of 7500 and higherenergy collisional dissociation (HCD) mode. The top five most intense ions were acquired with a minimum signal of 500, isolation width of 2, normalized collision energy of 35 eV, default charge state of 1, activation Q of 0.250, and an activation time of 30.0. The acquired data were evaluated with Thermo XCalibur software (version 2.1.0).

MS data processing and statistical methods
Raw data were converted to.mzXML in centroid form using msConvert [111]. Downstream analyses were conducted in the open source R program [112]. The xcms package was used with the centWave [112][113][114] algorithm and a Gaussian fit for peak-picking, and the OBI--Warp method for retention time correction and alignment [115]. Parameters used are given in the supplement material. Features that had retention times outside of limits determined by the gradient of solutions used were not included in further analysis (2-33 minutes for nonpolar modes and 0-47 minutes for polar modes). Intensities for technical replicates were averaged to provide a single value for each biological replicate and then normalized using median ratio scaling described by Wang, et al. [116]. One was added to intensity values prior to transformation to log 2 because of the presence of zeroes. Statistical analysis was conducted in the R package limma [117], which fits linear models to the data using an empirical Bayes approach, allowing for information across all features to be used to develop inference about individual features [118]. P-values were based on a moderated t-statistic with a Bayesian adjusted denominator and adjusted for false discovery rate [119]. Each mode (negative / positive + polar / nonpolar) was processed and analyzed separately. Features were designated as significantly different when the absolute log 2 fold change was greater than or equal to one, and the adjusted p-value was less than 0.05. Where results from all modes are included in one table or graph, the significance is based on the analysis in single modes alone. Putative molecular names of features were identified by searching mass per charge ratio (m/z) against the Human Metabo-  [121] and manually inspected using NIST MS Search (v2.0) [122].

Generating long double-stranded RNA for mosquito gene expression knockdown by RNAi
Long double-stranded RNA (dsRNA) was reversed transcribed and amplified from Ae aegypti mosquito total RNA. Briefly, total RNA was extracted from a whole mosquito homogenized in TRIzol reagent (Ambion). Primers were designed to amplify a region of~500 bp of the gene of interest and a T7 promotor sequence was added to the 5' end of both forward and reverse primers (S2 Table). Reverse transcription reaction (RT) and polymerase chain reaction (PCR) were performed in two-steps using SuperScript III Reverse Transcriptase (Invitrogen) and Taq polymerase (NEB) respectively. PCR products were purified using the GeneJET PCR Purification kit (Thermo Scientific) and subjected to in vitro transcription using the MEGAscript T7 kit (Invitrogen) according to the manufacturer's protocol. The reactions were incubated at 37˚C for 12 h. The transcribed products were then heated to 75˚C for 5 minutes and cooled down at room temperature for 4 h to allow dsRNA to anneal. The products were then treated with DNase to get rid of the DNA template and purified by phenol-chloroform followed by ethanol precipitation. The purified dsRNAs were stored at -80˚C for future use.

Double stranded RNA knockdown of genes in Aag2 cells
Aedes aegypti (Aag2) cells were cultured in Schneider's insect medium (Sigma-Aldrich) supplemented with 2 mM L-glutamine, 1% non-essential amino acids, and 10% FBS. To perform dsRNA knockdown in Aag2 cells, the cells were seeded in a 48-well plate at 50,000 cells/well. Twenty four hours later, the cells were transfected with 260 ng of dsRNA mixed with TransIT-2020 Reagent (Mirus) following the manufacturer's protocol. New medium with 2% FBS was replaced at 6 hours post transfection. Cell viability assays were performed at 2 days post transfection using the CellTiter-Glo Luminescent Cell Viability Assay (Promega).

DENV infection of the RNAi-knockdown cells
Cells were infected with DENV at 48 h post dsRNA transfection. To prepare virus inoculum, DENV-containing cell culture supernatant was mixed with 1x PBS supplemented with 0.5mM MgCl 2 , 1.2mM CaCl 2 and 1% FBS to make a final concentration of virus at a MOI of 0.3. Medium from Aag2 cells was removed and replaced with 0.3 mL of DENV inoculum per well. The cells were allowed to absorb viruses at room temperature for 1 h. Then, the inoculum was removed and replaced with fresh Minimum Essential Medium (MEM; Gibco) supplemented with 1% non-essential amino acids, 2mM L-glutamine and 2% FBS. Quantification of infectious DENV was carried out by plaque assay on BHK cells [6]. Absolute quantification of intracellular DENV genome replication was performed by qRT-PCR as previously described in 'quantitative RT-PCR detection of infected mosquitoes' section above. DENV RNA standard was produced from in vitro transcribed DENV genome as previously described in [123].

Quantifying mosquito gene expression upon dsRNA knockdown
Cells were collected in 200 μl TRIzol reagent at two days post dsRNA transfection. RNA extraction was performed following the manufacture's protocol. Total RNA (500 ng) was subjected to RT using random primers (Invitrogen). The cDNA was quantified by quantitative PCR (qPCR) using PowerUp SYBR Green Master Mix (Applied biosystems). Primer sequences designed to quantify DEGS and actin (internal control) genes are shown in S3 Table. ΔΔCt method was used for calculating percent gene expression of the target gene compared to the GFP knockdown control.

4HPR inhibitor studies in Aag2 cells
Aag2 cells pre-treated with 3.75 μM of N-(4-hydroxyphenyl) retinamide (4HPR; Sigma-Aldrich) dissolved in DMSO or DMSO only as a vehicle control. The final concentration of DMSO was at 1% in Schneider media supplemented with 2% FBS. At 24 h post treatment, cells were infected with DENV at MOI of 0.3. Following absorption, cells were overlaid with MEM containing 2% FBS and fresh 4HPR or DMSO. Virus-containing supernatant was harvested at 24 hpi and titrated by plaque assay. Cell viability assays were performed with 4HPR or DMSO treatment without DENV infection using the CellTiter-Glo Luminescent Cell Viability Assay (Promega).

Quantification of SPs in Aag2 cells by multiple reaction monitoring (MRM)
SPs were extracted from Aag2 cells following the protocol published by [124]. Briefly, cells were trypsinized and washed twice in PBS. Equal cell numbers between treated and control samples were collected. Cer/Sph mixture I Internal standard (Avanti Polar Lipids) was premixed with chloroform at a final concentration of 2 nM. Metabolic extraction was performed as described. Metabolites were dried using nitrogen gas.
An Agilent 1200 Rapid Resolution liquid chromatography (LC) system coupled to an Agilent 6460 series QQQ mass spectrometer (MS) was used to analyze SPs in each sample according to Merrill et al., 2005 with some modifications [124]. A Waters Xbridge C18 2.1mm x 100mm, 3. Average abundance of fatty acyl molecule in DENV infected midguts was compared with uninfected midguts and represented as log 2 fold change. Each row shows a different fatty acyl molecule, grouped based on the classification of molecular structure. Columns represent 3, 7, and 11 day pbm. Log 2 fold changes that are zero represent the changes that were not significantly different in DENV infected versus uninfected tissues. Log 2 fold changes shown in dark red or dark blue represent log 2 fold changes that are greater than 5 or lower than -5. (TIF) S1 Table. Select metabolites from mosquito midguts that show differential abundance following DENV infection. Abundance of metabolites detected in DENV-infected and uninfected midguts was compared. Frist tab lists the molecules that were putatively identifiable and second tab lists the molecules were unidentifiable. The following information is provided for each feature: m/z Avg , average mass to charge ratio detected (averaged between the two biological replicate pools); RT Avg , average retention time detected for each metabolite (averaged between the two biological replicate pools); Detection mode, both polar and nonpolar extracts were detected in both negative and positive ionization modes; Name, identification for each lipid; Lipid classes, classification of each lipid according to LIPID