**4. Discussion**

In humans, ZIKV infection was correlated with congenital microcephaly in newborns and with other neurological conditions in adults [4–7,10–12]. Still, little is known about the molecular mechanism of ZIKV infection and how it relates to neurological disorders. Here, we present a human host protein–ZIKV (Brazilian genotype) NS protein interaction network. This network was obtained by a combination of yeas<sup>t</sup> two-hybrid (Y2H) screens and tandem affinity purification coupled to mass spectrometry (TAP-MS). Y2H screens primarily reveal direct pair-wise interactions and are capable of detecting transient interactions, while TAP-MS will reveal proteins engaged in stable complexes, which will eventually result in the identification of indirect interactions [44–47]. The use of both methods results in a comprehensive panorama of ZIKV protein–protein interactions.

The merged network combining two complementary methods (Y2H and TAP-MS) contains 157 nodes and 189 interactions with a limited overlap between the two methods, consistent with other previously determined PIN [23,46,47]. Further, the subset of Y2H interactions validated in human cells displayed a false positive rate of ~24% (9/38) averaged across all seven ZIKV NS proteins, as judged by confirmation interaction experiments using SBP pulldown assays. This is in line with other published Y2H screens [23,48–51]. These results sugges<sup>t</sup> that this PIN contains high-confidence interactions.

Twenty-nine human proteins interacted with more than one ZIKV protein (19.3% of all hits). Similar relatively high levels of promiscuity of human proteins in relation to their viral interactors were also found in previous studies. Scaturro et al. [19] and Coyaud et al. [20] had 10.5% and 36% of all hits interacting to more than one ZIKV protein, respectively. Although Shah et al. [21] had a much lower (0.3% of all hits) rate of human proteins interacting to more than one ZIKV protein, a high level of promiscuity of human proteins is also apparent across studies, where the same human proteins are often found interacting with distinct ZIKV proteins. For example, all human proteins shared between Shah et al. [21] and Scaturro et al. [19], or between Shah et al. and this study, were found to interact with di fferent ZIKV NS proteins. In addition, comparisons across other studies showed consistently high levels of discordance in bait interactions (Figure 4B). These data sugges<sup>t</sup> that di fferent ZIKV NS proteins have common targets in the human proteome. However, it is unclear why di fferent studies detected exclusive interactions with di fferent baits. It is conceivable that several ZIKV NS proteins interact with large protein complexes, such as the 26S subunit of the proteasome complex, via di fferent targets; furthermore, di fferences in the biology of the cells providing the proteome (i.e., levels of protein expression and formation of specific protein complexes), the biochemical methods, or the filtering criteria for significant interactions may also determine which interactions are robust enough to result in detection.

Several aspects could account for the low level of overlapping between studies. Although every study used a di fferent cell line, three of them used derivatives of HEK293 cells (HEK 293T, HEK 293FT, HEK 293 T-rex) and one used SK-N-BE2 neuroblastoma cells [19]. Yet, low overlap was also encountered in pair-wise comparison between studies with HEK cells (Shah et al. and Golubeva et al.; Shah et al. and Coyaud et al.). Conversely, a slightly higher overlap can be found between studies with HEK cells (Golubeva et al. and Coyaud et al.), as well as with studies with di fferent cell lines (Coyaud et al. and Scaturro et al.). These observations sugges<sup>t</sup> that di fferences in cell lines are unlikely to explain the low overlap.

In addition to a ffinity purification followed by mass spectrometry used in every study, complementary methods were also used such as BioID (a proximity-dependent labeling approach) [20] and the yeas<sup>t</sup> two-hybrid (this study) that could partly explain the di fferences between studies. Other small di fferences in methods (expression vectors, a ffinity tags) and preys (all viral proteins versus only non-structural proteins; amino acid sequence di fferences between virus strains used) could also contribute to the di fferences across studies. Alternatively, some could represent spurious interactions detected as a result of the overexpression of the baits; however, all four studies used stringent cut-o ff measures and validation experiments, ensuring that the number of false positive results is likely to be low. Furthermore, the limited number of proteins in our dataset with high CRAPome [27] scores indicating consistent recovery in a ffinity proteomics as non-specific background also suggests that the di fferences across studies are unlikely to be explained by a large number of false positives. We propose that the low overlap among these studies suggests that they have not reached saturation and other ZIKV protein-interacting host proteins are still to be discovered.

Further, we identified multiple components of CCT (chaperonin containing TCP1 or TriC-TCP-1 ring complex) complex as targets. This complex plays a role in tra fficking of telomerase and small Cajal body (CB) RNAs through the proper folding of the telomerase cofactor, TCAB1 [52]. CBs are transcription-dependent nuclear compartments and play a critical role in neuron biology through snRNP and snoRNP assembly [53]. Interestingly, Coyaud et al. [20] demonstrated that ZIKV NS5 expression leads to an increase in the absolute number of CBs per cell, but to a reduction of the volume of these CBs, suggesting that NS5 expression could lead to CB fragmentation. Our data point to the interaction of NS1 with multiple components of the CCT complex, suggesting that NS1 could also play a role in CB stability and in neural disorders. Additionally, it has already been shown that the Dengue virus (DENV) infection occurs in an NS1/CCT-dependent manner [54].

Centrosomal abnormalities lead to impaired mitosis, which is a hallmark of MCPH. In fact, our data set presents multiple proteins related to phenotypes associated with impaired mitosis (Figure 3B,C). Furthermore, our PIN shares 24 (16% of all unique hits) known interaction partners of 14 (out of 18) MCPH loci plus CEP63 (Table S14).

In that context, CEP192 (identified as an NS3 interaction partner by Y2H) plays a central role in the initial steps of centriole duplications through the interaction and recruitment of CEP152 (MCPH9) and PLK4, respectively, which is necessary for the proper recruitment of SAS6 (MCPH14), STIL (MCPH7), and CENPJ (MCPH6) [55–60]. Our data sugges<sup>t</sup> that NS3 could interfere with centriole duplication and, consequently, could be important for the ZIKV-associated microcephaly phenotype. Furthermore, GO enrichment analysis and Mitocheck phenoclusters sugges<sup>t</sup> that NS1, NS2A, and NS3 target host factors are implicated in mitotic phenotypes.

In humans, viral infection activates the type-I interferon (IFN-I) signaling leading to STAT1/2 activation. The ZIKV5 protein acts as an antagonist of the IFN-I pathway by stimulating STAT2 (but not STAT1) degradation [61]. STAT1 activity is modulated by PIAS1, which has been implicated in herpes simplex viral replication [62]. We identified PIAS1 as an interacting partner of NS5 and showed that overexpression of PIAS1 results in a shorter NS5 protein half-life. Our data sugges<sup>t</sup> that PIAS1 can modulate the levels of ZIKV NS5, but it is unclear the extent to which this modulation may a ffect ZIKV replication. Interestingly, a CRISPR/Cas9 screening revealed that PIAS1-depleted cells are more sensitive to ZIKV infection-dependent lethality [38]. Collectively, these data sugges<sup>t</sup> that PIAS1 might play an important role in ZIKV biology by modulating NS5 protein levels.

In summary, the data presented here together with three previously published studies [19–21] provide a valuable resource to dissect the mechanistic underpinnings of central nervous system perturbations caused by ZIKV infection and to identify potential pharmacological targets. A small number of overlapping hits across di fferent studies sugges<sup>t</sup> that the screens are still far from reaching saturation.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2073-4409/9/1/153/s1, Figure S1: GO enrichment of the merged network; Figure S2: Phenoclusters of individual preys; Figure S3: Y2H bait expression, control transformations, and matings; Table S1. Yeast two-hybrid screening data; Table S2. Yeast two-hybrid hits; Table S3. TAP-MS hits - APOSTL output; Table S4. Merged PIN (Y2H + TAP/MS); Table S5. GO (Cellular component) enrichment membership; Table S6. GO (Biological Process) enrichment membership; Table S7. Bait-specific GO (Biological Process) enrichment ratios; Table S8. Bait-specific GO (Cellular component) enrichment ratios; Table S9. Phenoclusters of bait sets; Table S10. Phenoclusters of individual preys; Table S11. Integrated ZIKV PIN; Table S12. Flavivirus replication factors (functional screens) intersection with Merged ZIKV PIN; Table S13. Microcephaly-associated genes; Table S14. Merged ZIKV PIN and MCPH subnetwork.

**Author Contributions:** V.A.G., T.C.N., R.D.M., G.S.-K., M.A.C. and A.N.A.M. data curation; V.A.G. and T.C.N. formal analysis; V.A.G., T.C.N., G.S.-K., M.A.C. and A.N.A.M. supervision; V.A.G., T.C.N. and X.L. validation; V.A.G., T.C.N., G.d.G., S.D., X.L., P.P.G., V.I. and J.K. investigation; V.A.G., T.C.N., G.d.G., X.L., P.P.G., V.I. and J.K. methodology; V.A.G. and T.C.N. writing—original draft; M.A.C. and A.N.A.M. conceptualization; M.A.C. and A.N.A.M. writing—review and editing; M.A.C. and A.N.A.M. resources; M.A.C. and A.N.A.M. funding acquisition. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by Florida Department of Health Zika Research Grant Initiative pilot award 7ZK29 and in part by the Proteomics and Metabolomics Core at the Mo ffitt Cancer Center through its NCI CCSG gran<sup>t</sup> (P30-CA76292), and by Mo ffitt's Center for Immunization and Infection Research in Cancer (CIIRC). T.C.N. is a Fulbright Scholar.

**Acknowledgments:** We thank the individuals and families who have generously donated their time, samples, and information to facilitate research on ZIKV. We also thank Brent Kuenzi for help with proteomic data analysis.

**Conflicts of Interest:** The authors declare that they have no conflicts of interest related to the contents of this article.
