Reduced dengue virus pre-membrane protein-HMGB1 interaction activates host cell transcription to attenuate infection

Abstract

To successfully complete an infection lifecycle, viruses have to avoid activating host antiviral and transcriptional responses to avoid disrupting cellular processes and cause premature host cell death. How the host transcriptional response to dengue virus (DENV) infection and its impact on pathogenesis, however, remain unclear. Herein, we examined a pair of DENV-2, where the wild-type strain (16681) elicited few host transcriptional responses following infection, in contrast to its clinically tested attenuated derivative (PDK53 strain). Site-directed mutagenesis studies revealed a single D29V substitution on the pre-membrane (prM) protein, present not in 16681 but in PDK53, that altered transcriptional response to infection. Mechanistically, prM D29V substitution impaired interaction of prM protein with host high mobility group box 1 (HMGB1) protein; wild-type prM-HMGB1 interaction prevented HMGB1 cytoplasmic-nuclear translocation to support transcriptional response to infection. Remarkably, HMGB1 knockdown not only reduced transcriptional response but also slowed virus replication rate, suggesting that HMGB1 also controlled the expression of host factors necessary for DENV replication. Indeed, introducing 29 V to prM increased virus replication rate and transcription of antiviral response genes in both wild-type DENV-3 as well as chimeric DENV-3/-2, the DENV-3 component of TAK-003. Our findings suggest exploiting this interaction to improve immunogenicity and potentially efficacy of current vaccines.

Introduction

Dengue is an acute, Aedes mosquito-transmitted disease. It is caused by any one of the four genetically distinct dengue viruses (DENV-1 to -4). Despite an estimated 100 million cases occurring each year¹, the scale of the problem is set to worsen in the coming years due to urbanization and globalisation, both compounded by climate change². The World Health Organization has thus named dengue as one of the top ten threats to global health³. Despite the scale of the global public health burden of dengue⁴, many knowledge gaps on pathogenesis remain undefined, making dengue vaccine development an imprecise science. Amongst the gaps in knowledge are how DENVs avoid activation of host immune responses before DENV infection disseminates from the site of inoculation to target organs. Systemic dissemination is also particularly important for perpetuation of DENVs as their life cycle depends on transmission from infected humans to blood-feeding Aedes mosquitoes, and then back to susceptible humans^5,6. Stealth from immune and other responses that limit infection is hence an important trait of DENVs⁷. Crippling this stealth mechanism could attenuate wild-type DENVs, making them possibly suitable for use as live vaccines.

An aspect of stealth is control of host transcription⁸. Such control would not only avoid activating innate immune responses, including the antiviral type-I interferon (IFN) response, it would also avoid disrupting cellular processes that lead to premature death of host cells. Indeed, the wild-type DENV-2 (16681 strain) and its attenuated derivative (PDK53 strain)⁹, which is a key component of a licensed live attenuated dengue vaccine (TAK-003) show such differences in control of host transcriptional response to infection^10–12. We have shown reproducibly that DENV-2 16681 replicates more slowly and induces fewer genes to be expressed than its attenuated PDK53 strain^13–15. Similar trends in both virus replication rate and transcriptional response to infection have also been found when wild-type yellow fever virus (Asibi strain) was compared with its attenuated derivative (YF17D)¹⁶. We have also previously shown that host transcriptional response was associated with attenuated PDK53 infection, in part through a glycine-to-aspartate substitution in position 53 (G53D) of the non-structural 1 (NS1) protein of 16681¹⁵. NS1 G53D showed reduced interaction with ribophorin I (RPN1), necessary to ensure NS1 glycosylation and proper folding in the endoplasmic reticulum (ER). The resultant ER stress response activates host cell antiviral responses that constrain PDK53 infection dissemination. However, a single NS1 G53D substitution on the 16681 genomic backbone did not accelerate the replication rate of the mutant virus to levels observed with PDK53, suggesting that another mutation in PDK53 also contributes to the compromised stealth and hence attenuated infection¹⁵.

Herein, using infectious clone construction with site-directed mutagenesis, we identified the aspartate residue on position 29 of the pre-membrane (prM) protein of DENV-2 16681 strain as critical in achieving stealth. Wild-type prM interacted with a prominent factor of transcription regulation, the high mobility group box 1 (HMGB1) protein¹⁷. Substitution of the aspartate residue to valine on position 29 (D29V) of the prM protein compromised interaction with HMGB1, resulting in cytoplasmic-nuclear translocation of HMGB1 for increased transcription of genes, especially those involved in type-I IFN response to infection. Remarkably, HMGB1 induction of the host transcriptome also accelerated PDK53 replication, which, besides activating cytoplasmic innate immune sensors, may also be important for vaccine immunogenicity^13,18. Our findings not only reveal a stealth mechanism of DENVs, but they also suggest a path forward to increase the potency of a licensed dengue vaccine without re-invention.

Results

DENV-2 PDK53 differs genetically from its parental wild-type DENV-2 16681 by nine nucleotide substitutions¹⁹. Eight of these substitutions are in the coding region of the genome, three of which are synonymous substitutions (Fig. 1A). The remaining substitution is in the 5’ untranslated region (5’UTR) (Fig. 1A). We have previously rescued infectious clone-derived viruses, including 16681, PDK53 and 6 mutants of 16681, each bearing either the 5’UTR substitution or one of the non-synonymous substitutions found in PDK53¹⁵.

Fig. 1. prM D29V substitution restricted the plaque size of wild-type 16681 in type-I IFN dependent manner.

A Schematic diagram of the genetic differences between attenuated DENV-2 PDK53 and its wild-type DENV-2 16681 parent. Nucleotides and amino acids of wild-type 16681 and attenuated PDK53 represented in blue and red letters, respectively. Figure adapted from Guzman et al., 2010⁵⁹. B, C Plaque phenotype of 16681, prM D29V mutant and PDK53 on BHK-21 cells. B Representative pictures of plaques. C Plaque sizes in mm²; n ≥ 12. n represents the number of plaques. mean ± SEM. D DENV RNA levels relative to GAPDH mRNA levels in BHK-21 cells infected with either 16681, prM D29V mutant or PDK53 at MOI of 1 at 24 hpi; n = 3. n represents number of biological replicates. mean ± SD. E–H Plaque phenotype of 16681, prM D29V mutant and PDK53 on BHK-21 cells transfected with either siRNA targeting IRF3 (siIRF3) or scramble non-targeting siRNA as control (siControl). E Western blot analysis showing knockdown efficiency of IRF3. F Representative pictures of plaques. G Plaque sizes in mm²; n ≥ 15. n represents number of plaques. mean ± SEM. H Plaque counts per well of 24-well plate; n ≥ 5. n represents number of wells of 24-well plate. mean ± SD. I–L DENV RNA levels relative to PPIA mRNA levels (I), and IFNβ (J), IFIT2 (K) and CXCL10 (L) mRNA levels relative to PPIA mRNA levels in primary moDCs infected with either 16681, prM D29V mutant or PDK53 at MOI of 1 at 48 hpi; n = 3. n represents number of biological replicates. mean ± SD. *p < 0.05; **p ≤ 0.01; ***p ≤ 0.001; ****p ≤ 0.0001 in unpaired Student’s t test.

We have previously reported that NS1 G53D substitution attenuated 16681 infection in both Aedes aegypti and human cells¹⁵. We have now found another substitution—prM D29V—that also reduced the plaque size of 16681 (Fig. 1B, C). However, this substitution did not alter Aedes aegypti infection¹⁵. It was originally dismissed as an attenuating mutation although the study did not provide evidence for its conclusion²⁰. No reduction in plaque diameter was found with NS2A L181F or NS4A G75A substitution on the 16681 genomic backbone (Supplementary Fig. 1a, b).

prM D29V substitution restricted the plaque size of wild-type 16681 in type-I IFN-dependent manner similar to PDK53

We next investigated the basis of the small plaque phenotype of the 16681 infectious clone with the prM D29V substitution, hereon referred to as prM D29V mutant. Replication of prM D29V mutant genomic RNA in BHK-21 cells on which plaque assay was performed to visualize plaque morphology, as measured at 24 h post-infection (hpi), was faster compared to 16681 (Fig. 1D), indicating that the plaque diameter of prM D29V mutant, like that of PDK53¹³, was not due to a slow virus replication rate. Instead, type-I IFN response from infected cells reduced the susceptibility of neighboring uninfected cells to infection. Plaque assay conducted on BHK-21 cells after knockdown of IRF3 (Fig. 1E and Supplementary Fig. 1c), a transcription factor of type-I IFN response that has been shown to play a crucial role in controlling PDK53 plaque formation¹³, increased prM D29V mutant plaque diameter compared to negative controls transfected with scrambled siRNA (Fig. 1F, G); prM D29V mutant plaque count, however, remained unchanged (Fig. 1H). Plaque number but not diameter of PDK53 was significantly increased (Fig. 1F–H), similar to previously reported observations^13,15. In contrast, IRF3 knockdown had no appreciable effects on both plaque size and count of 16681 (Fig. 1F–H). Consistent with our findings in BHK-21 cells (Fig. 1D), intracellular viral RNA levels in diploid human fibroblast cell strain, MRC-5 and primary monocyte-derived dendritic cells (moDCs) were higher with prM D29V mutant compared to 16681 infection, as measured at 24 (Supplementary Fig. 1d) and 48 hpi (Fig. 1I), respectively. Increased IFNβ and IFN-stimulated genes (ISGs) induction with prM D29V mutant and PDK53 compared to 16681 infection were also observed (Fig. 1J–L and Supplementary Fig. 1e–g).

Reversion of prM 29V back to wild-type prM 29D partially restored wild-type phenotype

To validate prM D29V substitution as an attenuating mutation, we used site-directed mutagenesis to reverse the valine residue to aspartate on the PDK53 infectious clone. Compared to PDK53, prM V29D substitution resulted in a mutant that formed larger plaques (Fig. 2A, B). The reverse mutant also replicated more slowly than PDK53 (Fig. 2C), with significantly reduced expression of IFNβ and ISGs (Fig. 2D–F). Collectively, these findings indicate that the prM D29V substitution contributes to the attenuated phenotype of PDK53.

Fig. 2. Reversion of prM 29V back to wild-type prM 29D partially restored wild-type phenotype.

A, B Plaque phenotype of 16681, PDK53 and prM V29D reverse mutant on BHK-21 cells. A Representative pictures of plaques. B Plaque sizes in mm²; n ≥ 24. n represents number of plaques. mean ± SEM. C–F DENV RNA levels relative to PPIA mRNA levels (C), and IFNβ (D), IFIT2 (E) and CXCL10 (F) mRNA levels relative to PPIA mRNA levels in BHK-21 cells infected with either 16681, prM D29V mutant, PDK53 or prM V29D reverse mutant at MOI of 1 at 24 hpi; n = 3. n represents number of biological replicates. mean ± SD. *p < 0.05; **p ≤ 0.01; ***p ≤ 0.001; ****p ≤ 0.0001 in unpaired Student’s t test.

prM D29V substitution compromised the ability of PDK53 to control host transcriptional response, especially immune responses

To determine if the increased type-I IFN response to the prM D29V mutant could be a part of a larger transcriptional response from reduced stealth, we examined the changes in gene expression in primary moDCs following infection with different DENV-2s including 16681, prM D29V mutant and PDK53 (Supplementary Data 1). We also included in this analysis the previously identified NS1 G53D mutant (Supplementary Data 1), which was attenuated due to loss of interaction between viral NS1 protein and host oligosaccharyltransferase (OST) complex on ER membrane¹⁵. Similar to previously reported findings with PDK53¹³, prM D29V mutant infection elicited more differentially expressed genes (DEGs) than wild-type 16681 infection, as compared to mock-infected cells (Supplementary Fig. 2a, left), with substantial amounts of DEGs exclusively elicited by attenuated DENV-2s (Supplementary Fig. 2a, right). To further characterize the transcriptional signature of prM D29V mutant infection, we identified the DEGs elicited by prM D29V mutant infection compared to wild-type 16681 infection. Remarkably, prM D29V mutant infection elicited the largest number of DEGs compared to wild-type 16681 infection (Fig. 3A, left). More importantly, the DEGs triggered by prM D29V mutant infection not only overlapped with almost all of the DEGs induced by NS1 G53D mutant infection, it also overlapped with 71% of the DEGs from PDK53 compared to 16681 infection (Fig. 3A, right).

Fig. 3. prM D29V substitution compromised the ability of PDK53 to control host transcriptional response.

Whole genome microarray was performed on primary moDCs infected with either 16681, prM D29V mutant, NS1 G53D mutant or PDK53 at MOI of 1 at 48 hpi; n = 3. n represents number of biological replicates. A Table (left) and proportional Venn diagram (right), showing the number of genes differentially expressed during attenuated DENV-2s infection against wild-type 16681 infection. Differentially expressed genes (DEGs) were defined as genes with fold change more than 2 and adjusted p-value less than 0.05. Proportional Venn diagram created according to the diagram made by using DeepVenn (https://www.deepvenn.com/)⁶⁰. B–E Pathway enrichment analysis on the DEGs shared among three attenuated DENV-2s (B, C), and between prM D29V mutant and PDK53 only (D, E). B, D Pie charts of the percentages of the up- and down-regulated DEGs. C, E Pathway enrichment analysis conducted by using the EnrichR tool against the Gene Ontology Biological Process (GOBP) database^61,62. Significantly enriched pathways were defined as pathways with adjusted p-value less than 0.05 and comprising at least 5 genes. The left Y-axis and X-axis indicate enriched pathways and the number of DEGs enriched in each pathway, respectively. Significantly enriched pathways are further classified into biological functions and annotated on the right Y-axis.

Pathway enrichment analysis revealed that most of the DEGs induced by prM D29V mutant infection were involved in immune response to infection. Among the 102 DEGs shared by prM D29V mutant, NS1 G53D mutant and PDK53 (Fig. 3A, right), 76 upregulated DEGs mapped onto 46 pathways, where significantly enriched pathways were those with adjusted p-value less than 0.05 and comprised at least 5 genes (Fig. 3B, C). The majority of the positively enriched pathways were involved with innate and adaptive immune responses (Fig. 3C); the downregulated DEGs could not be mapped onto any pathway. Of the 187 DEGs shared by both PDK53 and prM D29V mutant (Fig. 3A, right), 117 upregulated DEGs mapped onto 15 significantly enriched pathways, which were mostly involved in immune response to infection (Fig. 3D, E). Again, no significantly enriched pathways were identified from the downregulated DEGs. Finally, of the 319 DEGs found uniquely with prM D29V mutant infection (Fig. 3A, right), 177 upregulated DEGs mapped onto significantly enriched pathways involved in immune response, post-translational modification, cell growth/proliferation and cell death (Supplementary Fig. 2b, c); the 141 downregulated DEGs mapped onto pathways involved in cellular metabolism and cytokine as well as chemokine responses (Supplementary Fig. 2b, d). These findings collectively suggest that the prM D29V substitution impaired the ability of 16681 to control host transcriptional response, particularly those with immune functions to infection.

To demonstrate reproducibility of our findings, we conducted a separate experiment by infecting primary moDCs with 16681, PDK53, prM D29V or NS1 G53D mutants. This time, we quantified RNA transcripts using NanoString analysis with the nCounter Human Immunology panel (Supplementary Data 2). Compared to mock infection negative control, a large number of immune genes were induced in cells following infection with PDK53 and the 2 mutants; few differentially expressed genes were found in cells following 16681 infection compared to negative control (Supplementary Fig. 2e).

Wild-type prM protein interacted with host HMGB1 to prevent nuclear translocation

Following translation, the prM protein is translocated across the ER membrane into lumen²¹, where it associates with the DENV envelope (E) protein to form heterodimers on the glycoprotein shell of the virion²². Structural reconstruction of the prM-E heterodimers suggests that the amino acid residue on position 29 of the prM protein is solvent-exposed and surface-accessible, and could interact with other proteins in the ER (Fig. 4A and Supplementary Fig. 3a). To identify the host proteins prM interacts with, we over-expressed flag-tagged wild-type prM 29D and flag-tagged mutant prM 29 V separately in HEK293T cells that are highly susceptible to transfection and support high protein expression. Total proteins from the cells were then extracted and subjected to co-immunoprecipitation using anti-flag antibody. Immunoprecipitated extracts were then examined using mass spectrometry (Supplementary Data 3). Each condition was done in triplicates and only proteins that were identified in all three replicates were further analyzed to identify proteins that uniquely interacted with wild-type and mutant prM (Supplementary Fig. 3b). Sorting of data according to average abundance and average Q-score revealed the high mobility group box 1 (HMGB1) protein and the Ras homolog family member A (RhoA) protein as the top and unique hits of wild-type prM 29D (Fig. 4B) and mutant prM 29 V (Supplementary Fig. 3c), respectively. While RhoA is known as a prominent factor in cytoskeletal regulation²³, HMGB1 is well known as a master regulator of transcription²⁴, including those of immune genes²⁵. Interaction of wild-type but not mutant prM with HMGB1 was reproducibly observed in another independent flag co-immunoprecipitation experiment but this time, HMGB1 was probed using western blot (Fig. 4C and Supplementary Fig. 3d). However, we were unable to co-immunoprecipitate flag-tagged HMGB1 with prM from 16681 infection, likely in part due to the slow replication of 16681 and hence low abundance of prM protein (Supplementary Fig. 3e). Nonetheless, given the central role HMGB1 plays in controlling gene expression, which is the difference in host response to infection with our different DENV strains, we thus focused subsequent investigations to understand DENV stealth through its interaction with HMGB1.

Fig. 4. Wild-type prM protein interacted with host HMGB1 to prevent nuclear translocation.

A The 3D structure of DENV prM-E trimer at neutral pH to represent the environment in the endoplasmic reticulum (ER) lumen (PDB: 4B03⁶³) visualized using Chimera (https://www.cgl.ucsf.edu/chimera/)^64,65. PrM and E proteins represented in purple and blue ribbons, respectively, and the amino acid residue 29 represented as green spheres. B, C Co-immunoprecipitation using anti-flag antibody with whole cell lysates from HEK293T cells expressing flag-tagged wild-type prM 29D followed by mass spectrometry (B) and western blot analysis (C). B Mass spectrometry revealed a group of host proteins interacting exclusively with wild-type prM 29D. Identified host proteins were sorted according to average abundance and average Q-score to identify the top hits. C Western blot analysis confirmed the loss of interaction of mutant prM 29V, but not wild-type prM 29D, to host HMGB1 protein, which is the first top hit among host proteins interacting uniquely with wild-type prM 29D. The blot showing HMGB1 and flag-tagged prM expression levels before (Input) and after (anti-flag IP) co-immunoprecipitation. D Immunofluorescence performed on primary monocytes infected with different DENVs at MOI of 10 at 24 hpi to locate HMGB1 during different virus infection. HMGB1 and prM proteins were stained with their respective primary antibody followed by anti-rabbit IgG Alexa Fluor 488 (green)- and anti-mouse IgG Alexa Fluor 568 (red)-conjugated antibodies as secondary antibody, respectively, while cell nuclei were stained blue with DAPI. White arrows indicate nuclear HMGB1 in infected cells. Representative pictures were shown. n ≥ 4. n represents number of frames taken. E, F HMGB1 protein expression levels in cytoplasm (E) and nucleus (F) of A549 cells infected with different DENVs at MOI of 1 at 24 hpi. Following infection, all A549 cells were harvested and subjected to compartmentalization into cytoplasmic and nuclear fractions. Western blot analysis was performed to detect HMGB1 along with α-tubulin and histone as cytoplasmic and nuclear markers, respectively. (Left) Representative blots were shown. (Right) Bar graphs showing HMGB1 expression levels normalized against its respective marker, and expressed as relative units against wild-type 16681; n = 4. n represents number of independent experiments. mean ± SD. *p < 0.05 in unpaired Student’s t test.

HMGB1 is a ubiquitous nuclear protein that regulates transcription by acting as a DNA chaperone, modulating the structure of DNA to facilitate protein binding, including those of transcription factors¹⁷. Since prM is translocated into the ER lumen after translation, we asked if wild-type prM interacted with HMGB1 in the ER after its expression to sequester it from translocating to the nucleus. Immunofluorescence staining for prM and HMGB1 in primary monocytes under physiological conditions showed that under mock infection condition, HMGB1 was found mainly in the nucleus (Fig. 4D). In 16681-infected cells, however, HMGB1 largely remained in the cytoplasm (Fig. 4D). In contrast, HMGB1 was more abundant in the nucleus in prM D29V mutant- and PDK53-infected cells (Fig. 4D).

To validate our immunofluorescence microscopy findings, we fractionated a DENV-susceptible human lung adenocarcinoma cell line, A549. We have chosen to use A549 instead of primary monocytes to obtain a sufficient number of cells for fractionating into nuclear and cytoplasmic fractions, followed by a western blot for HMGB1. We found that the levels of HMGB1 in the cytoplasmic fractions of cells following 16681 infection were significantly higher compared to cells with prM D29V mutant and PDK53 infection (Fig. 4E and Supplementary Fig. 3f). Conversely, HMGB1 levels in the nuclear fractions of cells following 16681 infection were lower than cells with prM D29V mutant and PDK53 infection, although the differences were not significant, likely due to cytoplasmic contamination as indicated by the presence of the α-tubulin marker (Fig. 4F and Supplementary Fig. 3g).

Besides translocating to the nucleus to increase gene expression, HMGB1 can also be secreted from the cell to mediate inflammatory and immune responses²⁶. To explore the possibility that cytoplasmic localization of HMGB1 in cells following 16681 infection was poised for secretion, we measured the levels of HMGB1 in the cell culture supernatant of A549 cells following infection using ELISA. Compared to prM D29V mutant and PDK53 infection, the amount of secreted HMGB1 was significantly lower with 16681 infection (Supplementary Fig. 3h), suggesting that the enrichment of HMGB1 in the cytoplasm of cells following 16681 infection did not result in an increased amount of secreted HMGB1 that could act as a cytokine²⁷. We also ruled out the possibility that cell-cultured prM D29V mutant and PDK53, which we used for our experiments, could have induced immune gene expression and thus confounded our results: Levels of HMGB1 in the supernatant of prM D29V mutant and PDK53 virus stocks were not higher than those in 16681 virus stock (Supplementary Fig. 3i). Collectively, these findings suggest that the wild-type prM interacted with HMGB1 to impair transcriptional response to infection, an interaction which was lost with D29V substitution.

HMGB1 knockdown reduced transcriptional response to prM D29V mutant and PDK53 infection

To determine the effects HMGB1 has on DENV infection and the host response to infection, we pre-treated A549 cells with siRNA against HMGB1 and compared these to scrambled siRNA control (Fig. 5A and Supplementary Fig. 4a, b). Absolute mRNA transcript measurement using the nCounter Human Immunology panel (Supplementary Data 4) showed, as expected, that HMGB1 knockdown reduced the number of upregulated immune genes compared to mock infection controls (Fig. 5B–D and Supplementary Fig. 4c). Unexpectedly, HMGB1 knockdown also reduced both prM D29V mutant and PDK53 but not 16681 replication (Fig. 5E). This finding suggests that beyond immune genes, HMGB1 could also induce expression of additional host factors required by attenuated but not wild-type DENVs to complete the infection cycle²⁸.

Fig. 5. HMGB1 knockdown reduced transcriptional response to prM D29V mutant and PDK53 infection.

A–E A549 cells transfected with either siRNA targeting HMGB1 (siHMGB1) or scramble non-targeting siRNA as control (siControl) were infected with either 16681, prM D29V mutant or PDK53 at MOI of 1 at 24 hpi; n = 3. n represents number of biological replicates. A Western blot analysis showing knockdown efficiency of HMGB1. B–D NanoString nCounter analysis was performed on infected siRNA-treated A549 cells using the nCounter Human Immunology panel. Volcano plots of differential immune gene expressions in siHMGB1-treated A549 cells versus siControl-treated A549 cells during 16681 infection (B), prM D29V mutant infection (C), and PDK53 infection (D). Each point represents the average value of each immune gene from three biological replicates. The expression difference is considered significant for log2 fold change more than 1.3 (outer vertical dashed black lines) and p-value less than 0.05 (above horizontal solid red line). Points are colored according to their average expressions. Blue represents significantly downregulated genes. Red represents significantly upregulated genes. Gray represents not significantly differentially expressed genes. E DENV RNA levels relative to GAPDH mRNA levels in siRNA-treated A549 cells following infection. mean ± SD. F, G Plaque phenotype of DENV-3 wild type and DENV-3 prM S29V mutant on BHK-21 cells. F Representative pictures of plaques. G Plaque sizes in mm2; n ≥ 18. n represents number of plaques. mean ± SEM. H–K DENV RNA levels relative to GAPDH mRNA levels (H), and IFNβ (I), IFIT2 (J) and CXCL10 (K) mRNA levels relative to GAPDH mRNA levels in Huh-7 cells infected with either DENV-3 wild type or DENV-3 prM S29V mutant at MOI of 1 at 48 hpi; n = 3. n represents number of biological replicates. mean ± SD. L DENV RNA levels relative to GAPDH mRNA levels in primary moDCs infected with either TAK-003 DENV-3 or its prM S29V mutant at MOI of 1 over the course of 48 hours after infection; n = 3. n represents number of biological replicates. mean ± SD. *p < 0.05; **p ≤ 0.01; ***p ≤ 0.001 in unpaired Student’s t test.

PrM S29V substitution increased replication rate and innate immune response to DENV-3 and chimeric DENV-3/-2

If indeed, besides inducing immune genes, HMGB1 also induces expression of pro-viral factors that support replication of attenuated DENVs, the absence of this substitution in the chimeric DENV-1, −3 and −4 components of TAK-003 could have compromised their replication. These chimeric viruses were constructed using the PDK53 genomic backbone but spliced with prM and E genes from wild-type DENV-1, -3 and -4. Although the amino acid residue on position 29 of the prM protein is different among the four DENVs, it is conserved within each DENV (Supplementary Fig. 4d). Moreover, residues 23 to 27 and 30 are highly conserved across the four DENVs (Supplementary Fig. 4d). Since TAK-003 has not shown efficacy in preventing symptomatic DENV-3 infection in children without prior DENV infection at the point of vaccination, we asked if substituting the serine residue on position 29 of the prM protein with valine in the chimeric construct DENV-3/-2 could increase replication; vaccine viremia has been shown to correlate with magnitude of antibody response to vaccination^29,30.

We first tested if S29V substitution on DENV-3 prM could reproduce the features observed with prM D29V mutant. For this purpose, we used a DENV-3 isolated from the blood of a local dengue patient³¹ to determine if prM S29V substitution would reduce the plaque diameter; chimeric DENV-3/-2 would not be suitable for assessing plaque size as this virus, even with wild-type prM, is already attenuated as evidenced from clinical trials of TAK-003^10,11,32. As expected, we found that prM S29V substitution reduced plaque size of wild-type DENV-3 (Fig. 5F, G). It also increased replication in a DENV-susceptible human hepatoma cell line, Huh-7 (Fig. 5H), and correspondingly the type-I IFN response to infection (Fig. 5I–K). Similarly, prM S29V substitution increased the replication of chimeric DENV-3/-2 virus, especially in the first 12 hpi, in both Huh-7 cells (Supplementary Fig. 4e) and primary moDCs (Fig. 5L) compared to unaltered DENV-3/-2 chimeric virus. These findings collectively suggest substituting wild-type amino acid residue with valine on position 29 of the prM protein to reduce HMGB1 interaction could, in addition to inducing immune gene expression, also increase the rate of DENV replication.

Discussion

Avoiding innate immune sensors and controlling the transcriptional response is vital for DENV infection to proceed until newly formed viral progenies reach levels sufficient for systemic dissemination and transmission. Uncontrolled infection burden, whilst beneficial to DENVs, harms their human hosts; patients with high viremia levels are at increased risk of severe dengue^33,34. Our findings reveal that, at least in part, DENV evades early immune detection by using its prM protein to sequester the master transcriptional regulator, HMGB1^17,26 in the cytoplasm. Such sequestration would likely also prevent it from undergoing post-translational modification necessary for secretion into the extracellular space³⁵, in which HMGB1 could act as a pro-inflammatory cytokine²⁷.

Early antiviral responses are activated by innate immune sensors³⁶. Pattern recognition receptors (PRRs) such as Toll-like receptors (TLRs) and RIG-I-like receptors (RLRs) can detect DENV RNA, both the positive-sense genomic RNA as well as the double-stranded RNA formed upon synthesis of the negative-sense RNA intermediate^37,38. Furthermore, innate immune response can also be triggered by the mitochondria due to changes in metabolism, resulting from energy consumption for DENV replication. Release of mitochondrial DNA into the cytoplasm can trigger cytosolic DNA sensors—cGAS—to induce STING and downstream type-I IFN response^28,39. To avoid such premature antiviral responses, DENV replicates within modified ER membranes—also referred to as replication organelles—that shield viral RNA from cytoplasmic RNA sensors^40–42. It also interacts with host factors to avoid disrupting mitochondrial function^43,44 and activate host antiviral responses^45–47. Indeed, avoidance of premature induction of host antiviral response has likely shaped DENV evolution⁴⁸. Furthermore, viral genetic investigations into epidemiological observations have shown repeatedly that DENVs associated with epidemics activated fewer innate immune and antiviral responses than DENV strains transmitted during endemic periods⁴⁹. Crippling the ability of DENVs to evade such early antiviral response would thus attenuate DENVs clinically.

Attenuation of DENVs through introducing mutations that activate host antiviral response early, however, has to be balanced by increased replication rate of the mutant virus. Amongst the four live attenuated DENV strains developed by Mahidol University in the 1980s, one of which was PDK53, was DENV-1 PDK13⁵⁰. Despite over expression of genes associated with antiviral response within 24 hours of infection, PDK13 replication was not faster than that of its wild-type parent DENV-1 (16007)⁵¹; thus whilst safe, PDK13 showed lacklustre immunogenicity⁵². That induction of immune genes and increased virus replication rate both resulted from reduced prM interaction with HMGB1 thus suggest that prM D29V is an important attenuating mutation of PDK53. Future studies will be needed to dissect the transcriptional response to determine which component increased DENV replication.

Our finding that induction of genes through the nuclear activity of HMGB1, besides constraining infection dissemination, also supported faster replication of PDK53 and prM D29V mutant compared to 16681 within infected cells, may explain the DENV-type-specific efficacy profile of TAK-003. Regardless of pre-vaccination baseline serostatus, the efficacy against symptomatic DENV-2 infection following TAK-003 vaccination is well sustained at over 80%^10,12,53. While the efficacies for preventing symptomatic DENV-1, -3 and -4 following TAK-003 vaccination were high in those seropositive at baseline, the efficacies in those seronegative at baseline were lower; there was also no efficacy against DENV-3 in those seronegative at baseline¹². As the DENV-1, -3 and -4 components were constructed by splicing the prM and E genes from these DENVs into the genomic backbone of PDK53, the chimeric DENVs retained the NS1 G53D substitution but without the additional substitution in the prM gene of PDK53; the amino acid residue on position 29 of the prM protein of all three chimeric components of TAK-003 is thus that of the wild-type DENV. Substitution of the valine residue into the DENV-3/DENV-2 chimeric virus accelerated replication. Such accelerated replication may be necessary for sufficient viral antigens to be expressed to stimulate an adaptive immune response before attenuated virus replication is controlled. Enhancing DENV-type-specific vaccine efficacy of TAK-003, even if to a limited extent, may thus just be a single nucleotide substitution away.

There are several limitations in this study. We have not been able to test the extent to which an additional substitution in position 29 of the prM protein of chimeric DENV-3/-2 would enhance neutralizing antibody titers to DENV-3. Testing immunogenicity in mouse models may be misleading since mouse infection is predicated on suppressing type-I IFN response. As type-I IFN response constrains PDK53 infection and hence those of the respective chimeric viruses, a suppressed type-I IFN response could exaggerate attenuated DENV infection and hence immunogenicity. The intricacies involved in HMGB1 and wild-type prM interactions also remain to be fully investigated. As HMGB1 is not transported through the conventional ER-Golgi pathway^54,55, prM-HMGB1 interaction is thus unlikely to occur at the ER-Golgi interface. Moreover, the pre-peptide (pr) region of prM protein would have been cleaved off in the Golgi apparatus in the process of virus maturation. It is thus likely that prM-HMGB1 interaction occurs in the ER lumen. Future studies will be required to pinpoint the exact subcellular location of these interacting proteins. In addition, which of the 3 domains where HMGB1 interacts with prM remains to be fully elucidated. Lastly, position 29 of the prM protein is located in the pr domain, which is proximal to the furin cleavage site. It is thus possible that the prM D29V substitution, besides impairing interaction with HMGB1, could also impair virus maturation. We have not been able to assess the proportion of mature to immature DENVs clearly in this study, as conventional plaque-forming units underestimates the true number of infectious PDK53 particles due to its attenuated state13. Finally, the interaction of prM D29V to RhoA may yet have an additional impact on virus attenuation that will need to be examined in future studies.

In conclusion, our findings demonstrate that DENV avoids disrupting transcriptional response to infection by interacting with HMGB1, through its prM protein, and sequestering it in the cytoplasm. Substituting the amino acid residue at position 29 of the prM protein with valine reduced HMGB1 interacting activity, the translocation of which into the nucleus not only induced innate immune gene expression but also factors that accelerated DENV replication, potentially necessary for vaccine immunogenicity.

Methods

Infectious clone-derived viruses

Infectious clones of viruses were constructed based on the published genome sequences using the Gibson assembly method⁵⁶. Briefly, viral genomic RNA was extracted from the laboratory stocks of DENV-2 16681 and DENV-2 PDK53 using the QIAmp Viral RNA Mini Kit. cDNA was synthesized from the RNA using the SuperScript III First-Strand Synthesis Kit with a primer targeting DENV-2 3’UTR (Supplementary Table 1). PCR fragments were amplified from the cDNA using primer pairs listed in Supplementary Table 1 with the Q5 Hot-Start High-Fidelity 2x Master Mix. The PCR fragments were gel-purified with the MinElute Gel Extraction Kit and TA cloning was performed into the pGEM-T Easy Vector. Single-nucleotide mutations were introduced by site-directed mutagenesis using primers listed in Supplementary Table 1 with the Q5 Site-Directed Mutagenesis Kit. After confirmation of the mutations by Sanger sequencing, all fragments were amplified from the vectors using the Q5 Hot-Start High-Fidelity 2x Master Mix. pUC19 vector was also amplified using primer pairs provided in Supplementary Table 1. The amplified fragments were gel-purified and assembled using the NEBuilder HiFi DNA Assembly Master Mix to generate infectious clones. The infectious clones were transfected into HEK293T cells using Lipofectamine 2000 reagent. Supernatants containing viruses were harvested 2-3 days after transfection. The viruses were further expanded in C6/36 or Vero cells. Full genome sequencing was performed using either Sanger sequencing or MinION sequencing to confirm that the viruses were composed of genetic sequences identical to the published data with the accession numbers listed in Supplementary Table 2 and the mutant viruses carrying the desired mutations previously published¹⁹. Viral titres were determined by plaque assay as described below.

Cell lines

Cell lines used were purchased from the American Type Culture Collection (ATCC). BHK-21 and A549 cells were cultured in RPMI growth media (RPMI media with HEPES supplemented with 10% fetal bovine serum (FBS)). Huh-7 and HEK293T cells were cultured in DMEM growth media (DMEM media with HEPES supplemented with 10% FBS). MRC-5 cell strain was cultured in EMEM growth media (EMEM media with HEPES supplemented with 10% FBS). During the course of infection, the cell lines were cultured in maintenance media, consisting of 2% FBS, 100 U/mL penicillin, and 100 μg/mL streptomycin in the respective base media of each cell line.

Primary monocytes

Peripheral blood mononuclear cells (PBMCs) were isolated from venous blood collected from a flavivirus-naïve healthy donor with the approval of the National University of Singapore Institutional Review Board (B-15-227) as previously described⁵⁷. The cells were incubated at 37 °C for 2.5 h to allow monocytes to adhere to the tissue culture flask surface. The adhered cells were washed 5 times with PBS to remove the non-adherent lymphocytes, and replenished with monocyte growth media (RPMI media with HEPES supplemented with 10% FBS, NEAA, sodium pyruvate, 100 U/mL penicillin, and 100 µg/mL streptomycin). These monocytes were incubated overnight at 37 °C before use in experiments.

Primary monocyte-derived dendritic cells (moDCs)

Monocytes were isolated from PBMCs obtained as described above using CD14 MicroBeads. To differentiate CD14+ monocytes into dendritic cells, the isolated cells were cultured in the monocyte growth media supplemented with 100ng/mL of IL-4 and 50ng/mL of GM-CSF for 6-8 days, with replacement of fresh cytokine-containing monocyte growth media on the third/fourth day.

Virus infection

Cells were seeded at 1 × 10⁵ cells per well in a 24-well tissue culture plate or 5 × 10⁴ cells per well in a 96-well tissue culture plate 1 day prior to infection. In the 24-well tissue culture plate, the cells were infected with DENV by adsorbing 100 μL of viral inoculum for 1 h at 37 °C with rocking at 15 min intervals prior to replacement with the respective maintenance media of each cell line. In the 96-well tissue culture plate, the cells were infected with DENV by adsorbing 100 μL of viral inoculum for 1 h (cell lines) or 2–6 h (primary cells) at 37 °C prior to replacement with the respective maintenance media of each cell line. At indicated time points after infection, cells were harvested by washing once with PBS before the addition of RLT buffer from the RNeasy Mini Kit and frozen at −80 °C until RNA extraction.

Plaque assay

Plaque assay was performed on BHK-21 cells as previously described⁵⁷. Briefly, each sample was diluted in serial 10-fold dilutions and incubated on confluent BHK-21 cell monolayer in a 24-well tissue culture plate for 1 h at 37 °C with rocking at 15 min intervals before replacement with 0.8% carboxymethylcellulose in RPMI maintenance media. After 5-6 days of incubation at 37 °C, the cultures were fixed with 20% formaldehyde solution at room temperature for at least 30 min. The fixed cultures were washed off with water and stained with 0.5% crystal violet solution. The staining solution was washed off with water and the plate was dried to visualize plaques. The plaques were imaged using the ImmunuoSpot S6 Versa 580 (Cellular Technology Ltd) or the ChemiDoc Touch Imaging System (Bio-Rad). Plaque sizes were determined using the BioSpot software or the Fiji software. Parameters in determining plaque sizes were adjusted to optimize settings depending on plaque morphology with plaques on the edges of each well excluded from the analyses. Viral titres are calculated as plaque-forming units per milliliter (PFU/mL).

siRNA knockdown of IRF3

BHK-21 cells were transfected with either siRNA targeting IRF3 or non-targeting control siRNA using Lipofectamine RNAiMax reagent. Forty-eight hours post-transfection, a plaque assay was performed as described above. Transfection efficiency was determined by western blot as described below using anti-IRF3 (1:1000) and anti-β-actin (1:5000) antibodies.

Quantitative real-time PCR

Total RNA was extracted using the RNeasy Mini Kit and cDNA synthesis was performed using the qScript cDNA Synthesis Kit or LunaScript RT SuperMix. Quantitative real-time PCR was performed and analyzed with the LightCycler 480 system (Roche). The primer pairs used are listed in Supplementary Table 1.

Microarray analysis

Total RNA was extracted using the RNeasy Mini Kit. The quality of the RNA samples was assessed by Bioanalyzer and the microarray was performed at the Duke-NUS Genome Biology Facility using the Affymetrix GeneChip Human Gene 2.0 ST Array. Partek Genomics Suite analysis software was used for data normalization. The differentially expressed genes (DEGs) were defined as transcripts with a fold change of more than 2.0 and an adjusted p-value less than 0.05 after Benjamini-Hochberg Step-Up FDR. Pathway analysis was performed against the Gene Ontology Biological Process (GOBP) database using the EnrichR tool. Significantly enriched pathways were defined as pathways with an adjusted p-value less than 0.05 and comprising at least 5 genes.

NanoString analysis

Total RNA was extracted using the RNeasy Mini Kit. NanoString profiling of host immune response was performed using the nCounter Human Immunology v2 Panel. Briefly, the RNA extracts were hybridized to capture and reporter probes set. The hybridized samples were aligned and immobilized on the nCounter Sprint Cartridge, and scanning was performed to collect data using the nCounter Digital Analyzer (NanoString Technologies). The data was analyzed using the nSolver analysis software. Counts of mRNA transcripts obtained for each gene were normalized against counts of mRNA transcripts of a set of housekeeping genes. A heatmap of the unsupervised hierarchical clustering of immune genes that were identified as DEGs in the microarray analysis was conducted using the R package ComplexHeatmap⁵⁸.

Construction of flag-tagged prM plasmids and mass spectrometry analysis

PCR fragments containing flag-tagged full-length prM sequence with and without prM mutation were amplified using primer pairs listed in Supplementary Table 1 with the Q5 Hot-Start High-Fidelity 2x Master Mix. pUC19 vector was also amplified using primer pairs shown in Supplementary Table 1. The amplified fragments were gel-purified using the MinElute Gel Extraction Kit and assembled using the NEBuilder HiFi DNA Assembly Master Mix. The assembled products bearing either wild-type prM or mutant prM sequence were transfected into HEK293T cells using Lipofectamine 2000 reagent. Each condition was done in biological triplicate. Forty-eight hours post-transfection, the cells were washed once with PBS and lysed in NP40 lysis buffer (1% Nonidet P-40, 150 mM NaCl, 50 mM Tris, pH 8.0) supplemented with the protease inhibitor cocktail. After clarification of the lysates, the supernatants were collected and subjected to co-immunoprecipitation using anti-FLAG M2 magnetic beads. The eluates were sent to BGI Genomics (Hong Kong) for mass spectrometry analysis to identify proteins interacting with wild-type prM and mutant prM. The proteins that were identified in all biological triplicates, and showed average Q-score and average number of peptide more than 2 were used to identify a set of proteins that uniquely interacted with wild-type prM and mutant prM. To identify the top hits, the proteins were sorted according to average relative abundance and average Q-score.

Western blot

Cells were washed once with PBS and either lysed in the RIPA buffer in the presence of protease inhibitor cocktail for whole cell protein determination or subjected to the Nuclear/Cytosol Fractionation Kit for protein determination in cytoplasmic and nuclear compartments. After clarification of the lysates, the supernatants were harvested and diluted with 4x Laemmli sample buffer in the presence of 2-Mercaptoethanol followed by heat treatment at 95 °C for 5 min. Proteins were separated by SDS-PAGE using the 4–20% Mini-PROTEAN TGX stain-free gel with the Tris/Glycine/SDS buffer. The proteins on the gel were transferred to a PVDF membrane using the Trans-Blot Turbo Transfer System Blotting instrument (Bio-Rad). The blot was blocked with the EveryBlot blocking buffer before incubation with a desired primary antibody, followed by its respective HRP-conjugated antibody (1:5000). To visualize chemiluminescent signals, the blot was developed with ECL Prime western blotting detection reagent and imaged with the ChemiDoc Touch Imaging System. Quantification of protein expression levels was performed with the Image Lab software.

Immunofluorescence assay

Following infection, primary monocytes were harvested and fixed with 3% paraformaldehyde. After that, the cells were permeabilized and blocked with 0.1% saponin in PBS containing 5% BSA and 1 μg/mL Fc block for 30 min. The cells were then stained for prM (undiluted hybridoma supernatant) and HMGB1 (1:250) using their respective primary antibody, followed by anti-rabbit IgG Alexa Fluor 488- and anti-mouse IgG Alexa Fluor 594-conjugated antibodies, respectively, as secondary antibody (1:100), while cell nuclei were stained with DAPI. The samples were then viewed with the LSM710 confocal microscope (Carl Zeiss).

HMGB1 ELISA

Levels of HMGB1 in cell culture supernatants and virus stocks were quantified using HMGB1 express ELISA Kit. The standards, controls and samples were assessed in technical duplicates. The absorbance at 450 nm was measured using the Infinite M200 microplate reader (Tecan). Quantitative analysis was performed using the standards provided.

siRNA knockdown of HMGB1

A549 cells were transfected with either siRNA targeting HMGB1 or non-targeting control siRNA twice at 24 h interval using Lipofectamine RNAiMax reagent. Twenty-four hours post the second transfection, a plaque assay was performed as described above. Transfection efficiency was determined by western blot as described above using anti-HMGB1 (1:10000) and anti-β-actin (1:5000) antibodies. Cell viabilities at the time of infection were assessed using CellTiter 96 AQueous One Solution Cell Proliferation Assay (MTS).

Overexpression of flag-tagged HMGB1

HEK293T cells were transfected with either the flag expression vector or the flag-tagged HMGB1 expression plasmid using Lipofectamine 2000 reagent. Twenty-four hours post-transfection, cells were infected with DENV-2s as described above. Forty-eight hours post-infection, cells were harvested and subjected to flag co-immunoprecipitation as described above. The inputs and eluates were subjected to western blot analysis, probing for flag-tagged protein and prM protein as described above.

Statistical analysis

GraphPad Prism software was used for all statistical analysis. Findings are reported as mean ± SD or mean ± SEM as indicated in the figure legends. The number of biological replicates or the number of independent experiments (n) is stated in the figure legends. All analysis was conducted with unpaired Student’s t test as indicated in the figure legends; p < 0.05 was considered significant.

Author contributions

E.E.O. conceived the study. T.S., W.C.N., and M.M.C. conducted the experiments. J.S.G.O. conducted the analysis. H.C.T., D.H.L.N., Y.X., and Z.Q. provided resources. W.C.H., S.Z., J.X.Y., Y.T.T., and E.Z.O. provided expertise. T.S., J.S.G.O., K.R.C., M.M.C., and E.E.O. analyzed and interpreted the data. T.S. and E.E.O. wrote the manuscript. T.S., J.S.G.O., K.R.C., and E.E.O. revised the manuscript.

Data availability

The datasets generated or/and analyzed during this study are available in the Array Express repository with the following accession numbers and URLs, E-MTAB-16201 (https://www.ebi.ac.uk/biostudies/ArrayExpress/studies/E-MTAB-16201?key=fc791244-9f73-4bf2-8fae-9a265255354e), E-MTAB-16202 (https://www.ebi.ac.uk/biostudies/ArrayExpress/studies/E-MTAB-16202?key=35e11804-255f-4ad6-8520-08bbb8c09473), and E-MTAB-16204 (https://www.ebi.ac.uk/biostudies/ArrayExpress/studies/E-MTAB-16204?key=7b94054e-f13e-42ab-a308-09030745d070).

Competing interests

The authors declare no competing financial interest but the following competing non-financial interests: E.E.O. has served as a dengue vaccine advisory board member for Takeda Pharmaceuticals, which has used DENV-2 PDK53 to develop TAK-003.

Supplementary information

The online version contains supplementary material available at 10.1038/s41541-026-01401-3.