The SUMOylation pathway suppresses arbovirus replication in Aedes aegypti cells

Mosquitoes are responsible for the transmission of many clinically important arboviruses that cause significant levels of annual mortality and socioeconomic health burden worldwide. Deciphering the mechanisms by which mosquitoes modulate arbovirus infection is crucial to understand how viral-host interactions promote vector transmission and human disease. SUMOylation is a post-translational modification that leads to the covalent attachment of the Small Ubiquitin-like MOdifier (SUMO) protein to host factors, which in turn can modulate their stability, interaction networks, sub-cellular localisation, and biochemical function. While the SUMOylation pathway is known to play a key role in the regulation of host immune defences to virus infection in humans, the importance of this pathway during arbovirus infection in mosquito vectors, such as Aedes aegypti (Ae. aegypti), remains unknown. Here we characterise the sequence, structure, biochemical properties, and tissue-specific expression profiles of component proteins of the Ae. aegypti SUMOylation pathway. We demonstrate significant biochemical differences between Ae. aegypti and Homo sapiens SUMOylation pathways and identify cell-type specific patterns of SUMO expression in Ae. aegypti tissues known to support arbovirus replication. Importantly, depletion of core SUMOylation effector proteins (SUMO, Ubc9 and PIAS) in Ae. aegypti cells led to enhanced levels of arbovirus replication from three different families; Zika (Flaviviridae), Semliki Forest (Togaviridae), and Bunyamwera (Bunyaviridae) viruses. Our findings identify an important role for mosquito SUMOylation in the cellular restriction of arboviruses that may directly influence vector competence and transmission of clinically important arboviruses.


Introduction
Many clinically important arthropod-borne viruses (arboviruses), such as dengue (DENV), Zika (ZIKV) (Flaviviruses; Flaviviridae), chikungunya (CHIKV) (Alphavirus; Togaviridae) and Rift Valley Fever (Phlebovirus; Bunyaviridae) viruses, are transmitted by the mosquito vector Aedes aegypti (Ae. aegypti, Aa). Aedes-borne pathogens represent a substantial worldwide public health burden due to an ever expanding geographical vector range and associated threat of viral emergence and epidemic disease [1][2][3][4]. For example, approximately half of the world's population is now estimated to be at risk from DENV with around 390 million new infections worldwide every year [5]. In addition, arboviral diseases transmitted by Aedes have an important socioeconomic impact, with the 2015-2016 Zika outbreak in Latin America and the Caribbean estimated to cost 18 billion USD [6]. Without efficient treatments or vaccines against the majority of arbovirus infections, vector control is the most widely utilized strategy to limit viral transmission [7]. Insecticides have been commonly used as the first line of defence, but increased mosquito resistance hampers their efficacy [8,9]. As such, there is a need to develop new and effective vector control measures [10]. In this context, it is crucial to improve our understanding of mosquito biology and mosquito-arbovirus interactions to identify targets that influence arbovirus infection directly in mosquitoes.
The Small Ubiquitin-like MOdifier (SUMO) protein covalently modifies target proteins (in a process termed SUMOylation) in a reversible manner. This post-translational modification regulates a wide variety of cellular processes, including the cell cycle, transcription, DNA repair, protein stability and sub-cellular localization [11]. In Homo sapiens (H. sapiens), the process of SUMOylation is conducted in a 3-step enzymatic cascade, involving the E1 SUMO activating enzyme complex SAE1/2, the E2 conjugating enzyme Ubc9 (UBE2I), and E3 SUMO ligases, including the Protein Inhibitor of Activated STAT (PIAS) family of proteins [11]. H. sapiens possess 5 SUMO homologues, of which only SUMO1, 2, and 3 are constitutively expressed and conjugated to target proteins [12]. SUMO modification typically occurs on lysine (Lys/K) residues within a SUMO Consensus Motif (SCM) C-Lys-X-[Asp/Glu] of a target protein, where C represents a hydrophobic amino acid and X represents any amino acid [13]. SUMO2 and SUMO3 both contain an internal SCM that readily supports poly-SUMO2/ 3 chain formation [14,15]. SUMO chains have many important cellular functions, including roles in meiosis, mitosis, DNA repair, and proteasomal degradation [16]. In contrast, SUMO1 does not possess an internal SCM, and is therefore generally considered to be conjugated as a monomer onto target proteins or incorporated into poly-SUMO chains as a chain terminator [14].
SUMOylation plays an important function, both positively and negatively, in the replication of a wide range of DNA and RNA viruses in mammalian cells, either directly through the SUMOylation of viral proteins or indirectly by regulation of immune defences [17][18][19]. To date, studies to determine the role of SUMOylation during arbovirus infection have focused on flaviviruses in mammalian cells [20][21][22][23]. While two of these studies demonstrated depletion of Ubc9 could restrict DENV replication [20,21], a third study showed that SUMOylation of the DENV non-structural 5 protein (NS5) decreased ubiquitin-mediated degradation of NS5 to promote infection [23]. This proviral role for host SUMOylation in mammalian cells was supported by recent findings from Zhu et al. [22], which demonstrated the SUMOylation inhibitor 2-D08 to decrease flavivirus (ZIKV, DENV, Yellow fever virus, West Nile virus, and Japenese Encephalitis virus) RNA levels during infection. However, it remains to be determined what role (proviral or antiviral) SUMOylation plays during arbovirus infection in mosquito cells.
Genes of the SUMOylation pathway are conserved in various arthropods, including insects [24,25]. Most of the knowledge on SUMO function in insects has been gained from studies in the fruit fly Drosophila melanogaster (D. melanogaster). In this species, SUMO modification is known to be involved in embryogenesis, metamorphosis, gut regeneration, and innate immunity [26]. However, the functional conservation of the different effectors involved in the SUMOylation pathway, along with its role during infection with alphaviruses, flaviviruses, and bunyaviruses in arthropods has yet to be determined.
Here we characterise the sequence homology and functional conservation of the SUMOylation pathway between Ae. aegypti and H. sapiens. Our analysis reveals that AaSUMO is functionally more comparable to HsSUMO1 than HsSUMO3, the closest related orthologue, due to the lack of an internal SCM. Consistent with SUMOylation in H. sapiens, we show that the sole Ae. aegypti PIAS (AaPIAS) SUMO E3 ligase stimulates AaSUMO poly-SUMO chain formation in a SIZ/PIAS (SP) RING-domain dependent manner. We identify differential patterns of SUMO expression in Ae. aegypti cells and tissues known to support arbovirus replication. Finally, we present compelling evidence for a broad antiviral effect of Ae. aegypti host SUMOylation against arboviruses. Depletion of AaSUMO, AaUbc9, or AaPIAS resulted in increased arbovirus replication from three independent families; bunyamwera virus (BUNV), Semliki Forest virus (SFV) and ZIKV. To our knowledge, this research is the first study to investigate the biochemical properties and biological significance of the SUMOylation pathway on arbovirus replication in Ae. aegypti cells. Our findings reveal an important antiviral role for AaSU-MOylation in the regulation of antiviral defences to arbovirus infection in mosquito cells.
Alignment of SUMOylation effector proteins (SUMO, SAE1/2, Ubc9, and PIAS) between Ae. aegypti and H. sapiens demonstrated a high degree of protein conservation. AaSAE1, AaSAE2, and AaUbc9 shared 42%, 50%, and 85% amino acid identity with their respective H. sapiens orthologues (Fig 1A). Homology between AaSUMO and HsSUMO1-3 proteins ranged from 47 to 71% identity, with AaSUMO showing the highest degree of similarity to that of HsSUMO3 ( Fig 1A). Phylogenetic alignment of SUMO proteins from different species Hs, Homo sapiens (P55854); Mm, Mus musculus (Q9Z172); Gg, Gallus gallus (Q5ZHQ1); Xl, Xenopus laevis (Q6DK72); Is, Ixodes scapularis (B7PKS5); Cq, Culex quinquefasciatus (B0WVZ6); Aa, Aedes aegypti (Q16EQ3); Aal, Aedes albopictus (A0A182H3N0); Ag, Anopheles gambiae (Q7PNJ2 and Q7PY16 + ); Dm, Drosophila melanogaster (O97102). Branch lengths are proportional to sequence divergence with bootstrap values shown. Conserved residues are shown in orange; residues with similarity in green. SUMO consensus motifs (SCM; C-Lys(K)-X-Asp(D)/Glu(E), where C represents a branched hydrophobic amino acid and X represents any amino acid) are higlighted in grey. (C) Amino acid alignment of full length AaSUMO plotted against HsSUMO3. (D) Predicted structure of AaSUMO (blue) plotted against the resolved structure of HsSUMO3 (red; PDB ID: 2MP2; [85] indicated that SUMO orthologues from Ae. aegypti and other mosquito species lack an N-terminal SCM, contrary to the chelicerate Ixodes scapularis and vertebrate SUMO orthologues (Fig 1B and 1C; [25]). Homology modelling confirmed AaSUMO to contain a classical ubiquitin-related modifier tertiary structure with flexible N-and C-terminal regions (Fig 1D; [25,28]). In H. sapiens, PIAS SUMO E3 ligases are known to increase the rate of SUMO modification and poly-SUMO chain formation [29,30]. In contrast to the five PIAS proteins expressed by H. sapiens, only one PIAS protein has been identified in Ae. aegypti [24]. The amino acid identity of AaPIAS ranged from 32 to 37% between HsPIAS orthologues, increasing up to 70% in the catalytic SP-RING domain (Fig 1A). Sequence alignment confirmed the presence of Zn 2+ -coordinating residues within the SP-RING required for SUMO ligase activity ( Fig 1E; grey bars) [31]. These data indicate that the function of the SP-RING domain is likely to be conserved between H. sapiens and Ae. aegypti. Overall, our analysis revealed a high degree of protein conservation between Ae. aegypti and H. sapiens SUMOylation pathways. These data suggest a functional conservation of this pathway between mosquitoes and humans, with the notable exception of an absent N-terminal SCM within AaSUMO.

AaPIAS stimulates the efficient formation of poly-AaSUMO chains
As AaSUMO lacks an N-terminal SCM, this protein may not readily support poly-SUMO chain formation, unlike its closest H. sapiens orthologue HsSUMO3 [14]. We tested this hypothesis by incubating together recombinant proteins (S1 Fig) of either the AaSUMOylation pathway (AaSUMO, AaUbc9, and AaSAE1/2) or the HsSUMOylation pathway (HsSUMO3, HsUbc9, and HsSAE1/2) at either 28 or 37˚C, the optimal temperatures for mosquito and mammalian cells, respectively. We detected a build-up of AaSUMO-dimers along with high molecular weight (HMW) SUMO conjugates indicative of SAE1/2 auto-SUMOylation ( Fig  2A; [32,33]). The formation of HMW AaSUMO conjugates was more comparable to HsSUMO1 than HsSUMO3 at either 28 or 37˚C (Fig 2B and 2C). These data indicate that AaSUMO is more biochemically similar to that of HsSUMO1 than HsSUMO3 with respect to supporting the formation of HWM SUMO protein conjugates, consistent with the lack of an internal SCM (Fig 1B; [25]). In order to determine the effect of AaPIAS on poly-AaSUMO chain formation, biochemical assays were repeated in the presence of wild-type (WT) or catalytically inactive (C371A; Fig 1E) AaPIAS. Addition of WT AaPIAS led to a rapid build-up of HMW SUMO conjugates relative to AaPIAS C371A (Fig 2D and 2E). Collectively, these data indicate that AaPIAS stimulates the efficient formation of HMW SUMO conjugates in an SP-RING dependent manner. We next sought to investigate why AaSUMO could not efficiently form chains in the absence of AaPIAS. We hypothesized that this was due to the absence of an SCM (Fig 1B; [25]). We generated a chimeric SUMO protein, in which the Nterminus of AaSUMO (amino acids 1-13) was replaced by the one of HsSUMO3 (amino acids [1][2][3][4][5][6][7][8][9][10][11][12][13][14] to incorporate its SCM. This chimeric SUMO protein was found to support the rapid build-up of HMW SUMO conjugates in the absence of AaPIAS that was dependent upon the Lys acceptor residue (K11) within the SCM (Fig 2F). These data demonstrate that while AaSUMO shares the highest degree of amino acid identity to HsSUMO3, its biochemical activity as a ubiquitin-like protein is more functionally related to that of HsSUMO1 due to the absence of a functional SCM.
AaSUMO contains eight Lys residues ( Fig 1C) which could act as potential acceptor sites for SUMO modification in the presence of AaPIAS. We next investigated which Lys residues within AaSUMO were preferentially modified by AaPIAS. Mass spectrometry analysis was conducted on SUMO-dimers and HMW poly-SUMO chains generated in the presence or absence of either WT or C371A AaPIAS (S2 Fig). Spectral analysis demonstrated that AaSUMO was preferentially conjugated to surface exposed Lys (K) residues K5, K6, K9, K40, and to a lesser extent, K33 (Figs 3A and S3 and S2 Files). Addition of AaPIAS increased the proportion of detectible branched SUMO peptides, consistent with enhanced levels of poly-SUMO chain formation in a SP-RING dependent manner (Fig 2D and 2E) without strong preference for any specific solvent-exposed Lys residue (Figs 3B and 3C and S3). Collectively, these data demonstrate that AaPIAS plays an important role in the SCM-independent formation of HMW AaSUMO conjugates.

AaSUMO is differentially expressed in Ae. aegypti tissues
To determine if AaSUMO was expressed in tissues relevant to arbovirus infection, we determined the relative expression levels of SUMO transcripts in multiple mosquito tissues from female mosquitoes; including salivary glands, digestive tract, ovaries, carcass (remaining abdominal tissue), and perfused haemocytes (mosquito immune cells). RNA was extracted and transcript expression levels analysed by RT-qPCR. SUMO transcripts were detected in all

PLOS PATHOGENS
SUMOylation pathway is antiviral against arboviruses in mosquitoes tissues and haemocytes, with significantly higher levels of SUMO expression observed in haemocytes and ovaries relative to the carcass (Figs 4A and S4).
We next examined the expression pattern of AaSUMO protein in Ae. aegypti. Immunofluorescent confocal microscopy analysis was conducted using polyclonal antibodies raised to HsSUMO2/3, which detect both unconjugated and conjugated AaSUMO protein (Fig 2A and  2D). Microscopy assays were conducted on cultured Ae. aegypti-derived AF5 cells, perfused haemocytes, salivary glands, digestive tract, and ovaries from Ae. aegypti NBF females (Fig 4). Controls without primary antibody were also analysed under the same conditions to ensure signal specificity to AaSUMO (S5 Fig). In AF5 cells, AaSUMO was predominantly detected as nuclear punctate structures (Fig 4B and 4C). Consistent with our RT-qPCR results, AaSUMO expression was detected in perfused haemocytes and in every Ae. aegypti tissue analysed. In haemocytes, AaSUMO expression was restricted to the nucleus in prohaemocytes (Fig 4D), which have a high nuclear/cytoplasmic ratio [34], and differentiated granulocytes (Fig 4E), which are bigger cells with filopodia and a low nuclear/cytoplasmic ratio [34]. Very low levels of AaSUMO expression were detected in the lobes of salivary glands ( Fig 4F), with higher levels of AaSUMO detected in surrounding fat body cells ( Fig 4G). In the midgut, AaSUMO was expressed in the nucleus of enterocytes (recognizable by their large nuclei) and in smaller enteroendocrine and/or stem cells (Fig 4H). In the ovaries, AaSUMO was found to be highly expressed in the nuclei of the epithelial cells of the oviduct (Fig 4I). Similar staining was found in the ovarian sheath surrounding the ovarioles (Fig 4J). In the ovarioles, AaSUMO was expressed within the primary previtellogenic follicle and in the germarium (Fig 4K and 4L). In the primary follicle, AaSUMO was detected in the cytoplasm of the follicular cells surrounding the follicle. A strong circular signal of AaSUMO protein was detected close to the nucleus of the oocyte (Fig 4K and 4L; white arrows) and at perinuclear regions of nurse cells (Fig 4K; yellow arrows). In the germarium, AaSUMO was detected in the nucleus and cytoplasm of germline stem cells and follicle stem cells, in the cystoblasts as well as in the germline and follicular cells of the developing secondary follicle. Overall, our findings shows that AaSUMO is differentially expressed in multiple tissues and cells known to support arbovirus replication, including the gut, ovaries, and haemocytes.

PLOS PATHOGENS
SUMOylation pathway is antiviral against arboviruses in mosquitoes independently of the effector protein targeted for depletion (Fig 5B, 5E and 5H, respectively). These data indicate that multiple component proteins of the AaSUMOylation pathway contribute to the suppression of arbovirus replication; with BUNV and SFV being the most susceptible to AaSUMOylation-mediated restriction. Significant increases in BUNV, SFV, and ZIKV intracellular RNA levels could also be observed in cells depleted for AaSUMOylation proteins (Fig 5C, 5F and 5I, respectively); with SFV demonstrating the greatest increase in viral RNA (vRNA) levels (up to 10-fold relative to dsLacZ-treated control cells). It remains to be determined why depletion of AaUbc9 or AaPIAS had no significant effect on the intracellular levels of BUNV RNA. However, this may reflect the relative level of effector protein depletion between replicate experiments (Fig 5A; Ubc9). In the majority of cases, disruption of the AaSUMOylation pathway led to increases in both arbovirus replication (Fig 5B, 5E and 5H) and intracellular vRNA levels (Fig 5C, 5F and 5I). Collectively, these data indicate that the AaSUMOylation pathway has a broad antiviral action against arboviruses from multiple families in Ae. aegypti cells.

Discussion
The antiviral defences of mosquitoes are key determinants of vector competence, i.e. the ability of mosquitoes to get infected by an arbovirus following an infectious blood meal and subsequently transmit it to a new vertebrate host [44]. In mosquitoes, several pathways have been reported to play a role in the control of arbovirus infection. These include the RNAi and evolutionarily conserved innate immune pathways, such as the NF-κB signalling pathways (Toll and Imd) and the Janus kinase-signal transduction and activators of transcription (JAK/STAT) pathway [45,46]. Here, for the first time, we identify the Ae. aegypti SUMOylation pathway as being another evolutionarily conserved pathway that supresses arbovirus replication in mosquito cells.
To date, the majority of SUMO studies in arthropods has been conducted in D. melanogaster, which is known to possess a single SUMO gene (smt3; [47]). In contrast, two homologous SUMO genes have been previously identified in the Ae. aegypti genome [24,25]. Our analysis using the most recent annotated sequence [27] could only identify one SUMO gene (AAEL015064), a finding supported by our mass spectrometry analysis which identified unique peptides to all core effector proteins of the Ae. aegypti SUMOylation pathway (AaSUMO, AaSAE1/2, AaUbc9, and AaPIAS). We demonstrate that AaSUMO is highly related to HsSUMO3, but lacks an N-terminal SCM restricting its ability to support poly-SUMO chain formation. Our analysis supports previous studies, which have suggested this motif to be lost in insects, but to be present in vertebrates and ticks [24,25]. Addition of AaPIAS stimulated the accumulation of HMW AaSUMO conjugates in an SP-RING dependent manner, demonstrating AaPIAS to be a functionally conserved SUMO E3 ligase in mosquito cells. Notably, AaPIAS promoted AaSUMO conjugation onto multiple solvent exposed Lys residues, indicating that AaPIAS may stimulate the formation of multiple or branched poly-SUMO chain types. This contrasts with poly-SUMO2/3 chain formation in vertebrate cells, which can occur independently of SUMO E3 ligase activity in a SCM K11-dependent manner [14]. This highlights an additional level of regulation required for the formation of poly-SUMO chains in Ae. aegypti, which are known to play many important biological roles in mammalian cells [16]. It remains to be determined if AaPIAS, or other functionally conserved SUMO ligases, significantly contribute to poly-AaSUMO chain formation in vivo. Our mass spectrometry and western blot analysis revealed that AaSAE2 is SUMO modified in vitro, the levels of which are significantly enhanced in the presence of AaPIAS in a SP-RING dependent manner. In H. sapiens, SUMO modification of SAE2 leads to altered protein sub-cellular

PLOS PATHOGENS
SUMOylation pathway is antiviral against arboviruses in mosquitoes localisation and reduced affinity for Ubc9 [32,33]. Thus, SAE2 auto-SUMOylation is likely to be a functionally conserved property of the SAE1/2 heterodimeric complex, although further studies are required to confirm the biological importance of this phenotype in Ae. aegypti cells.
The analysis of SUMO expression in NBF female Ae. aegypti reveals that SUMO is expressed in haemocytes (mosquito immune cells) and in various tissues, including the digestive tract, the ovaries, and the fat body. Our analysis demonstrated SUMO transcript to be ubiquitously expressed throughout Ae. aegypti, with the highest levels of expression in the ovaries and haemocytes. Of note, previous microarrays and RNA-seq studies have shown that SUMO expression levels in whole females do not change upon blood feeding and ZIKV infection [48,49]. Similar ubiquitous tissue expression of SUMO (smt3) in D. melanogaster has been found by RNA-seq, with an enrichment in ovaries (http://flyatlas.gla.ac.uk/FlyAtlas2/index. html). Although the expression of SUMO has yet to be reported in the haemocytes of adult flies, Ubc9 (lesswright, Lwr) is specifically required in larval haemocytes for haematopoiesis [50], suggesting that SUMO is expressed in Drosophila haemocytes. SUMO and Ubc9 are also ubiquitously expressed in the shrimp species Fenneropenaeus chinensis [51], with high levels of expression in haemocytes and ovaries. Thus, the tissue expression profile of SUMO in arthropods seems to be conserved across evolution. Indeed, Ubc9 is highly expressed in the reproductive organs and the blood cells in the fish Cynoglossus semilaevis [52] and in immune cells, endocrine tissues, reproductive tissues, and adipose tissues in H. sapiens (Human Protein Atlas, https://www.proteinatlas.org/) [53,54].
In many of the Ae. aegypti tissues and cells studied here, SUMO is predominantly expressed in punctate structures in the nuclei. This pattern of expression in mammalian cells is due to the extensive SUMO modification of Promyelocytic leukaemia nuclear bodies (PML-NBs), a process required for PML-NB formation and associated protein-protein interactions [55]. PML-NBs are involved in many cellular processes, including the cell cycle, genome stability, and antiviral defence, through the SUMO-dependent sequestration, modification, and degradation of associated host factors [56]. The PML gene is not evolutionarily conserved among eukaryotes, being absent in lower eukaryotes such as insects [57]. However, as many other RING containing proteins (including other TRIM family members like PML; TRIM19) can form nuclear structures, including in insects [58], it is possible that SUMO modification of other RING proteins is involved in the formation of nuclear complexes in mosquito cells with similar regulatory functions.
SUMO modification is known to regulate the biological function of hundreds of cellular proteins [11]. Previous studies have clearly established a role of SUMOylation, both positive and negative, in the replication of DNA and RNA viruses in mammalian cells [17][18][19]. The SUMOylation pathway is also known to influence arbovirus replication in H. sapiens cells, although its pro-or antiviral action is still a matter of debate and warrants additional study [20][21][22][23]. We have now shown that the SUMOylation pathway influences replication of a range of arboviruses in Ae. aegypti cells. Indeed, depletion of components of the SUMOylation pathway led to a consistent increase in viral replication and intracellular RNA levels of BUNV, SFV, and ZIKV; albeit to varying extents. Such differences may reflect virus-specific interfaces with the AaSUMOylation pathway that requires further study. Importantly, depletion of SUMO effector proteins led to an increase in viral replication of BUNV and SFV equivalent to that of Ago2 depletion, a key mediator of the Ae. aegypti antiviral RNAi pathway [46]. Thus, we identify the AaSUMOylation pathway to contribute to the antiviral immune response to arbovirus infection in mosquito cells. However, we cannot discount a positive (pro-viral) role for AaSUMOylation, as there may well be virus-specific requirements for host SUMOylation in certain mosquito cell types or tissues. Notably, AaPIAS depletion had an overall weaker effect on SFV replication in mosquito cells relative to that of Ubc9 depletion. Such differences

PLOS PATHOGENS
SUMOylation pathway is antiviral against arboviruses in mosquitoes may be linked to virus-specific requirements for component proteins of the AaSUMOylation pathway to restrict arbovirus replication. Previous studies suggest that members of the H. sapiens PIAS family of E3 ligases, which are primarily known for their role as negative regulators of innate immunity, also possess intrinsic immune functions to suppress viral replication in mammalian cells [59][60][61]. In Ae. aegypti, transient depletion of AaPIAS decreased DENV and ZIKV infection intensity in the midgut [62,63], consistent with PIAS proteins being negative regulators of the JAK/STAT pathway. By way of contrast, activation of the JAK/STAT pathway in the mosquito fat body decreased DENV infection while it had no effect on ZIKV and CHIKV [64]. These data suggest virus-or tissue-specific roles for the JAK/STAT pathway in the immune regulation of arbovirus infection in mosquito cells, which may account for why we did not observe an overall net decrease in arbovirus replication following AaPIAS depletion in our study.
Collectively, our results demonstrate that the sequence, function, and tissue distribution of the Ae. aegypti SUMOylation pathway effector proteins are highly conserved. We have shown the SUMO pathway to suppress arbovirus replication from three independent viral families. Thus, we identify SUMOylation as an additional conserved pathway involved in antiviral defences in cells derived from Ae. aegypti. Our data helps to fill many long-standing gaps in the literature about the conservation and biochemical properties of the SUMOylation pathway in this transmitting vector species, and the overall net functional effect of Ae. aegypti SUMO modification during arbovirus replication. Areas for future research include determining in which tissues and cell types, where arboviruses can replicate (e.g. gut, salivary glands, and haemocytes; [65][66][67]), and how, mosquito SUMOylation controls arbovirus replication. Further analyses using tissue-specific genetic tools and vector-specific antibodies will be required to assess the precise location and mechanisms of restriction in Ae. aegypti.
SUMOylation can influence the function of a wide variety of proteins and cellular processes. Nevertheless, some hypotheses explaining the antiviral action of SUMOylation in mosquito cells can be proposed. In D. melanogaster, SUMOylation occurs on diverse immune proteins that can modulate innate immune function, including JAK/STAT and Imd patways [26,68,69]. In arbovirus (BUNV, SFV, and ZIKV) infected mosquito cells silenced for Ubc9, we failed to observe a significant decrease of vir-1 (JAK-STAT effector, [70]) or cecropin D (Imd effector, [71] expression (S6 Fig), which suggests that AaSUMOylation does not directly influence the transcription of these effector proteins. However, we cannot exclude that AaSU-MOylation could enhance the expression, stability, or function of other innate immune regulators to mediate an antiviral effect in mosquito cells. Alternatively, arboviral proteins may be targeted directly for SUMOylation to modulate their function or stability, as observed for the DENV NS5 protein in mammalian cells [23]. Further studies are warranted to determine the precise mechanism(s) associated with the antiviral role(s) of host SUMOylation in Ae. aegypti cells that suppress arbovirus replication.

PLOS PATHOGENS
SUMOylation pathway is antiviral against arboviruses in mosquitoes inhibitor cocktail (Roche, pH 7.5). Samples were then lysed by digital sonicator (7 repeats at 50% intensity for 30 sec, at 4˚C) and centrifuged (14,000 x g for 10 min). Supernatants containing recombinant proteins were filtered through a 0.45 μm filter, bound to 250 μl (w/v) of equilibrated Ni-NTA (Qiagen) or Strep-tactin Superflow agarose beads (Novagen) in the corresponding purification buffers. Samples were tumbled end-over-end for 90 min at room temperature (RT), beads sedimented centrifugation (1500 x g for 5 min), washed twice in 10 ml of purification buffer and three times in 1 ml of purification buffer. 6xHis tag proteins were eluted in 750 μl of 6xHis Purification Buffer containing 350 mM (final concentration) imidazole at pH 7.5 prior to dialysis in Dialysis Buffer (50 mM Tris (pH 7.5), 150 mM NaCl, 5% glycerol, 2 mM MgCl 2 and 5 mM β-mercaptoethanol, 0.1% Triton-X-100) for 4 hours. Strep.II-AaSAE2/1 was eluted in 600 μl Strep Purification Buffer with 2.5 mM D-desthiobiotin (pH 8.5) prior to dialysis in Dialysis Buffer for 4 hours. Proteins were visualized by SDS-PAGE (4-12% gradient PAGE; Invitrogen) alongside purified commercial BSA standards (BioRad) by Coomassie brilliant blue staining. Protein concentration was determined on the Li-Cor Odyssey Imaging System.

Biochemical assays
Master mixes of Ae. aegypti or H. Sapiens recombinant SUMO pathway component proteins were aliquoted into individual tubes containing SUMO, Ubc9, and SAE1/2 (50 ng each/reaction), and if required, WT or mutant (C371A) PIAS (10 ng/reaction). The reaction was initiated by addition of SAE1/2 and incubated at 37 or 28˚C for the desired length of time. The reaction was terminated by the addition of 1.5xBM (Laemmli buffer supplemented with 2.6 M Urea and 50 mM Dithiothretol). This sample mixture was boiled at 95˚C for 10 min prior to SDS-PAGE and western blotting. Biochemical assays for Mass spectrometry analysis were performed with 5 μg SUMO, 500 ng Ubc9, 500 ng SAE1/2, and 50 ng WT or C371A PIAS, and incubated at 28˚C for 1 hour. Reactions were terminated by the addition of NuPAGE Sample Reducing Agent (Invitrogen).

Western blots
Samples were resolved on a NuPAGE-Novex 4-12% Bis-Tris Gradient gel (Invitrogen) with 1X NuPage MES SDS running buffer (Invitrogen). PAGE gels were transferred onto an Amersham protran 0.2 μm nitrocellulose blotting membrane (GE Healthcare) in transfer buffer (1x NuPage transfer buffer and 10% Methanol) at 30 V for 60 min. The membranes were blocked in blocking buffer (5% FBS in PBS) for 1 hour. The mouse anti-SUMO2/3 primary antibody (Cat. No. Ab81371; Abcam) diluted 1:1,000 in blocking buffer was incubated on the blotting membrane for 1 hour or O/N at 4˚C. The membrane was washed 3 times for 5 min in PBST (0.1% Tween-20). A DyLight800 fluorescent goat anti-mouse secondary antibody (SA535521; Thermo Fisher Scientific) diluted 1:10,000 was incubated with the membrane for 1 hour. All steps were conducted at room temperature unless stated otherwise. Membranes were imaged and quantified using a LiCor Odyssey Imaging System.

dsRNA production
Total RNA was extracted from Ae.aegypti-derived A20 cells using an RNAeasy Plus Kit (Qiagen), following manufacturer's instructions. cDNA was synthesised with TaqMan Reverse Transcription Reagents kit (Life Technologies) with Random Hexamer primers according to the manufacturer's instructions. To synthesise dsRNA, cDNA was amplified with gene-specific primers (S1 Table) incorporating the T7 RNA polymerase promoter sequence at the 5' end. PCR was conducted using Phusion polymerase (New England Biolabs), PCR products were

PLOS PATHOGENS
SUMOylation pathway is antiviral against arboviruses in mosquitoes excised from the gel using QIAquick Gel Extraction Kit (Qiagen), and sequenced to confirm identity by Sanger sequencing (Source Bioscience). dsRNA were produced using the MEGAscript RNAi kit (Ambion) according to the manufacturer's instructions. The dsRNA were quantified using a Nanodrop spectrophotometer.

dsRNA transfection and infection
AF5 cells (2.5x10 5 / well) were seeded in 24-well plates and left in the incubator overnight (O/ N) at 28˚C prior to the transfection of 300 ng dsRNA with 2 μl DharmaFECT2 (Dharmacon) per well, following manufacturer's instructions. Cells were incubated for 72 hours prior to infection (MOI of 0.05 PFU per cell). After 48 hours, cells were harvested for RT-qPCR (viral load and depletion efficiency) or luciferase analysis.

Luciferase assay
Luciferase expression was determined using the Luciferase Assay System Kit (Promega) for SFV-Ffluc and the Nano-Glo Luciferase Assay System Kit (Promega) for BUNV-NLuc and ZIKV-NLuc. Luciferase activities were determined on a GloMax 20/20 Luminometer following the manufacturer's instructions. A minimum of 5 independent biological repeats were conducted in triplicate for each experimental condition.

Mosquito rearing
Ae. aegypti mosquitoes (Liverpool strain, gift from Pr. Eileen Devaney, University of Glasgow) were reared at 28˚C and 80% humidity with a 12:12 light photoperiod. Larvae were reared in water with some dry pieces of cat food pieces from larvae hatching to pupal stage. Emerging adult mosquitoes were transferred in cages with unlimited access to 10% w/v sucrose solution. For rearing, female mosquitoes were fed on rabbit blood supplemented with heparin (Orygen Antibodies Ltd) using a Hemotek (Hemotek Ltd., UK) at 37˚C.

Dissection of mosquito tissues and perfusion of haemocytes
Tissues were dissected from 5 day-old females in RNase free 0.05% PBS-Tween 20 (PBST) (v/ v). For RNA extraction, tissues were stored at -80˚C. For immunostainings, tissues were stored in PBS-T 0.05% on ice before fixation. Haemocytes were collected by perfusion. The last segment of the abdomen was cut and mosquitoes were then injected in the thorax with 0.01% PBS-T. For RT-qPCR experiments, PBS-T diluted haemocytes were collected in tubes on ice, further centrifuge at 2,000 RPM for 15 min at 4˚C. The supernatant was removed before the addition of TRIzol reagent (Thermo Fisher Scientific). For immunostainings, haemocytes were perfused on slides (ibidi) which were incubated for 30 min at 28˚C before removal of PBS-T for fixation and staining.

RNA extraction and reverse transcription (RT)-quantitative PCR
Total RNA was isolated from cells using the RNAeasy Plus Kit (Qiagen), following manufacturer's instructions. cDNA was synthesised with TaqMan Reverse Transcription Reagents kit (Life Technologies) with Random Hexamer primers, as per manufacturer's instructions. Samples were analysed in triplicate using SYBR Green Master mix (Applied Biosystems) with gene-specific primers (S1 Table). Quantitative PCR (qPCR) was conducted on an Applied Biosystems 7500 Fast-Real-Time PCR system using MicroAmp plates and Optical Adhesive covers (Applied Biosystems). Relative mRNA expression levels were determined using the 2 −ΔΔCt (cycle threshold) method [77], normalised to the S7 ribosomal gene and presented as relative

PLOS PATHOGENS
SUMOylation pathway is antiviral against arboviruses in mosquitoes values to that of the dsLacZ control. Data presented for the knockdown efficiency and the viral loads is the mean of 5 independent biological repeats. RQ means and standard deviations are presented.
Dissected mosquito tissues (pools of 25 digestive tracts, ovaries or carcasses, pools of 60 salivary glands, and pools of perfused haemocytes from 70 females per independent replicate) were homogenised in TRIzol (Thermo Fisher Scientific) with glass beads using the Precellys 24 homogeniser (Bertin Instruments). RNA from tissues and haemocytes were further extracted using the Trizol method according to the manufacturer's instructions except that 1-Bromo-3-chloropropane (Sigma) was used instead of chloroform and DNase (TURBO DNase, Ambion) treatment was performed. Reverse-transcriptions (RT) and RT negative controls were performed using the MMLV retro-transcriptase (Promega) from 25 ng/μl of RNA according to the manufacturer's instructions. cDNAs were aliquoted and stored at -20˚C until qPCR. qPCR was performed using Fast SYBR Green master mix (Applied Biosystems) following the manufacturer's instructions using the 7500 Fast machine (Applied Biosystems). Primers used are listed in S1 Table. Results were analysed with the 7500 Software v2.0.6. Data were analyzed as described by Taylor et al. [78] to obtain normalized expression values, relative to ribosomal S7 transcript expression, with a geometric mean (geomean) of 1 for the carcass condition. Log2-normalized expression values were used for statistical analyses.

Immunofluorescence assays
Salivary glands and ovaries (n�10 of each per replicate) were fixed at room temperature for 20 min in 4% (w/v) paraformaldehyde (PFA) diluted in PBS-T 0.05%. AF5 cells (3x10 4 cells/well, incubated O/N at 28˚C on ibidi slides), perfused haemocytes (from a pool of 5 females per replicate) and guts (n�10 per replicate) were fixed the same way with the exception that the PFA was diluted using PBS. Fixed haemocytes/tissues (except ovaries) were washed (three times, 15 min each at 4˚C) in PBS-T 0.05%, before being blocked for a minimum of 30 min in blocking solution (0.25% PBS-T 5% FBS (v/v), 5% BSA (w/v), 0.25% Triton X-100 (v/v) at 4˚C. Unlike other tissues, ovaries were dilacerated and washed in 1% PBS-T before being blocked in blocking solution containing 0.5% Tween and 0.5% Triton. AF5 cells, haemocytes, and tissues were incubated at 4˚C overnight with a mouse anti-SUMO2/3 antibody (Cat. No. Ab81371; Abcam) diluted 1:1000 in blocking solution. For each experiment, a negative control without primary antibody was performed. Samples were washed (five times, 15 min each at 4˚C) in 0.05% PBS-T (0.5% PBS-T for ovaries) and incubated with an Alexa Fluor 568 goat anti-mouse IgG (H+L) diluted 1:1,000, DAPI 1X (405 nm) and Phalloidin 1X (488 nm) in blocking solution for 2 hours at RT. Four washes in 0.05% PBS-T (0.5% PBS-T for ovaries) were done and tissues were mounted between a slide and coverslip (24 mm x 24 mm) with an imaging spacer (1 well, diam. x thickness 13 mm x 0.12 mm Grace Bio-Labs SecureSeal imaging spacer, Sigma-Aldrich) using ibidi Mounting Medium (ibidi). Mounting media was used to replace the PBS-T in haemocytes. Images were acquired on a Zeiss LSM 710 inverted confocal microscope, equipped with a 40X or 63X oil-immersion objective and processed with Fiji/ImageJ and Adobe Photoshop. Two independent replicates were performed.

Shotgun proteomic analysis to identify expressed SUMO pathway proteins
AF5 cell extracts were harvested in 1.5XBM and proteins fractionated by 4-12% Bis-tris SDS-PAGE gel running in MES buffer. Each lane was excised into 12 slices and tryptic peptides extracted as described previously [79]. All 132 peptide samples were analyzed by LC-MS/ MS on a Q Exactive mass spectrometer (Thermo Scientific) coupled to an EASY-nLC 1000 liquid chromatography system via an EASY-Spray ion source (Thermo Scientific) running a

PLOS PATHOGENS
SUMOylation pathway is antiviral against arboviruses in mosquitoes 75 μm x 500 mm EASY-Spray column at 45˚C. Peptides were fractionated over a 90 minute gradient and data were acquired in the data-dependent mode. Full scan spectra (m/z 300-1800) were acquired with resolution R = 70,000 at m/z 200 (after accumulation to a target value of 1,000,000 with maximum injection time of 20 ms). The 10 most intense ions were fragmented by HCD and measured with a resolution of R = 17,500 at m/z 200 (target value of 500,000, maximum injection time of 60 ms) and intensity threshold of 2.1x10 4 . Peptide match was set to 'preferred' and a 40 second dynamic exclusion list was applied. The resultant 132 raw MS data files were processed using MaxQuant (v 1.6.1.0) with the built-in Andromeda peptide search engine [80]. The Ae. aegypti uniprot protome (downloaded August 2017) was searched. Enzyme specificity was set to trypsin-P. Cysteine carbamidomethylation was selected as a fixed modification with methionine oxidation and protein N-terminal acetylation as variable modifications. Initial maximum allowed mass deviation was set to 20 parts per million (ppm) for peptide masses and 0.5 Da for MS/MS peaks. The minimum peptide length was set to 7 amino acids and maximum size 4600 Da. A maximum of two missed cleavages were considered. A false discovery rate (FDR) of 1% was required at both the protein and peptide levels. The 'match between runs' option was applied with a time window of two minutes matching between samples from the same gel slice position across different lanes, or a slice above or below. 5462 protein groups were identified after removal of decoy proteins, putative contaminants and those only identified by modified peptide. Peptide level data for SUMO, Ubc9, SAE1, SAE2 and PIAS proteins are shown in S1 File.

Shotgun proteomic analysis of in vitro SUMO conjugation reactions to identify sites of polymerisation
Three in vitro SUMO conjugation reactions were prepared in triplicate: without any E3, with WT PIAS, or with C371A mutant PIAS (see above). Proteins were fractionated by 10% Bis-tris SDS-PAGE gel running in MOPS buffer. Each lane was excised into upper and lower slices to gain information on SUMO-SUMO dimers, and high molecular weight polymers. Tryptic peptides were extracted as described above. Half of each tryptic peptide sample was further subjected to in solution digestion with GluC to shorten the C-terminal SUMO adduct attached to substrate peptides, hence increasing likelihood of detection of as many SUMO conjugation sites as possible. Two mass spectrometry runs were performed for tryptic peptides and one for the doubly digested Trypsin-GluC peptides. Instrument setting were essentially as described above except for the second trypic peptide run and the only Try-GluC run for which a top1 method was applied with a 1000ms fill time. These settings improve MS/MS spectra for large low abundance peptides. To automatically detect branched peptides a concatenated database approach was taken, similar to previously described [81]. Briefly, a sequence file in fasta format was generated whereby the C-terminus of SUMO was concatenated to all conceivable peptides yielded by protease digestion of SUMO, Ubc9, SAE1 and SAE2. This concatenated sequence database contains individual 'proteins' that mimic the SUMO-substrate branched peptides allowing MaxQuant to automatically search for them in the MS data. MaxQuant was set up as described above except the concatenated peptide fasta file was used along with a file containing all proteins present in the in vitro reactions, GluC and Try/P (6 missed cleavages) or just Try/P alone (4 missed cleavages) were defined as proteases, and max peptide mass was set to 10000. Representative spectra can be found in S7-S14 Figs and data summary can be found in S2 File.

Bioinformatics and statistical analysis
Amino acid alignment was conducted using the online software T-COFFEE (http://tcoffee.crg. cat/) [82]. These results were then plotted and exported with BOXSHADE (https://embnet.

PLOS PATHOGENS
SUMOylation pathway is antiviral against arboviruses in mosquitoes vital-it.ch/software/BOX_form.html). Models of AaSUMO were conducted by obtaining the annotated amino acid sequence from UniProt, and predicting the tertiary structure with Phyre2 and plotting the structure in Chimera 1.10.1 [83,84]. Statistical analysis and graphs were produced using GraphPad Prism 7.02. Statistical tests are indicated in figure legends. Immunofluorescence assay without primary antibody on perfused haemocytes (A), salivary glands (B), midgut (C) and ovaries (D). The signal was revealed by an Alexa Fluor 568 goat anti-mouse IgG (H+L) diluted 1:1000 (red). Nuclei are stained by DAPI (blue) and Factin is stained by Phalloidin 488 (green). The images were acquired on a Zeiss LSM 710 inverted confocal microscope with 40X or 63X oil-immersion objective and using the same parameters as those used for samples incubated with anti-SUMO primary antibody (Fig 5). Scale bar is 20 μm. Expression of vir1 and cecropin D (cecD) was analysed in dsUbc9-and dsLacZ-treated cells (as described in Fig 5) by RT-qPCR. N = 5 independent biological replicates. Values normalized to ribosomal S7 and expressed relative to dsLacZ-treated control samples set to 1; RQ mean and SD plotted. Statistical analysis, one sample (two-tailed) t test to a hypothetical mean of 1 (dsLacZ control), significant probability