Transcriptional Profiling Confirms the Therapeutic Effects of Mast Cell Stabilization in a Dengue Disease Model

To identify the host responses that were induced by DENV2 infection, we compared liver and spleen gene expression levels on days 1 and 3 postinfection in DENV2-infected mice versus mock-infected animals. These time points were chosen since they coincide with the peak time points of virus replication and DENV symptoms such as vascular leakage and thrombocytopenia in this mouse model (24). Active DENV2 replication in target organs was confirmed in this study by the quantitation of total DENV genome copy numbers (see Fig. S1A and S1B in the supplemental material) and negative-strand viral RNA PCR (Fig. S1C). We found that DENV2 infection resulted in the differential expression of 320 genes in the liver on day 1 but that the response in this organ was mostly resolved by day 3, when only 9 differentially expressed (DE) genes were detected (Fig. 1A and B). The liver response was enriched for various metabolic pathways and interferon (IFN) signaling on day 1 (Table 1). Carbohydrate metabolism genes such as SLC2A4 (solute carrier family 2 member 4), PFKFB3 (6-phosphofructo-2-kinase/fructose-2,6-biphosphatase 3), and RCAN1 (regulator of calcineurin 1) were upregulated, as were lipid metabolism genes such as NAMPT (nicotinamide phosphoribosyltransferase), SREBF1 (sterol regulatory element binding transcription factor 1), and LPIN1 (lipin 1) (Fig. 1C). DENV2 infection also induced the upregulation of interferon-stimulated genes (ISGs) such as IFIT3 (IFN-induced protein with tetratricopeptide repeats 3), ISG15, and OAS (2′-5′-oligoadenylate synthetase 1) (Fig. 1C). These ISGs have been shown to promote antiviral activity against DENV using in vitro model systems (33 – 35). On day 3, only two pathways, granzyme A signaling and protein kinase A signaling, were enriched (Table 1), suggesting a transition from an innate immune response toward a robust T cell response for clearing virus infection. The small number of upregulated genes, including GZMA (granzyme A), SPIC (Spi-C transcription factor), and Ly6a (lymphocyte antigen 6 complex, locus A), were associated with immune functions.

The kinetics of the host response in the spleen were similar to those in the liver, with the greatest number of gene expression changes occurring on day 1. Two hundred twenty DE genes were detected in the spleen on day 1, while 61 DE genes were detected on day 3 (Fig. 1A and B). The top most enriched pathways in the spleen were associated with pathogen recognition, IFN and cytokine responses, and immune cell activation on day 1 (Table 1). Cytokine genes such as CCL2, CCL4, IL-15 (interleukin 15), and CXCL10 were upregulated in the spleen (Fig. 1D). These pathways, along with immune cell signaling pathways, were the most perturbed in the spleen on day 3 (Table 1). For example, natural killer cell signaling genes such as KLRA6 (killer cell lectin-like receptor subfamily A6), KLRC1 (killer cell lectin-like receptor C1), and NCR1 (natural cytotoxicity-triggering receptor 1) were highly expressed in the spleen on day 3 (Fig. 1E). Overall, these data suggest active virus replication and systemic infection followed by the clearance of DENV in this wild-type (WT) mouse model, which are consistent with the viral kinetics reported previously for DENV2 in WT mice (24). Pathways enriched in both the spleens and livers of DENV-infected mice were consistent with those identified in human in vivo and in vitro studies (10 – 12, 14, 36, 37).

To investigate the effect of MC stabilization on DENV-induced host responses, we transcriptionally profiled the spleens and livers of DENV2-infected mice that had been treated with ketotifen. Animals received intraperitoneal injections of ketotifen or saline daily. The first dose was administered 1 h after infection, with daily drug injections thereafter at 24-h intervals. Although ketotifen treatment did not significantly affect DENV2 replication (see Fig. S1 in the supplemental material), it had a striking effect on the host response to DENV2 infection (Fig. 2). Ketotifen-treated, DENV2-infected mice had a more robust transcriptional response than did untreated, DENV2-infected mice (Fig. 2A and B). Approximately 12% of the liver genes and 18% of the spleen genes that were DE in ketotifen-treated, DENV2-infected mice were also DE in untreated, DENV2-infected mice (Fig. 2C to F).

To probe the responses that were specific to DENV2 infection, we focused on the genes that were DE in untreated, DENV2-infected mice and asked whether their expression was also affected in ketotifen-treated, DENV2-infected mice. We found that ketotifen treatment dampened the DENV-induced gene expression changes by at least 10% for approximately 55% of the genes on day 1 and 67% of the genes on day 3 in the liver (Fig. S2). In the spleen, ketotifen treatment reversed the DENV2-induced expression changes by at least 10% for approximately 45% of the genes on day 1 and 72% of the genes on day 3 (Fig. S3).

We next investigated the host pathways that were induced or inhibited by ketotifen. The responses that were most perturbed in the livers of ketotifen-treated, DENV2-infected mice were cholesterol and lipid metabolism pathways on day 1 (Table 2 and Fig. 3). Cholesterol biosynthesis genes such as DHCR24 (24-dehydrocholesterol reductase), CYP51A1 (cytochrome P450 family 51), SQLE (squalene epoxidase), and FDPS (farnesyl diphosphate synthase) were downregulated on day 1 relative to untreated, DENV2-infected animals (Fig. 3C). We also observed decreases in the expression levels of genes such as COL1A1 (collagen type 1 alpha 1), COL1A2, F11 (coagulation factor XI), and KNG1 (kininogen 1) that are involved in intrinsic prothrombin activation (Fig. 3C).

Perturbation of cholesterol biosynthesis, complement signaling, and liver X receptor (LXR)/retinoid X receptor (RXR) pathways dominated the splenic responses in ketotifen-treated, DENV2-infected mice on day 1 (Table 2 and Fig. 4). LXR is activated by oxysterol ligands and forms a heterodimer with RXR to initiate the transcription of target genes. LXR/RXR signaling is important for the regulation of lipid metabolism, inflammation, and cholesterol catabolism (38). Cholesterol biosynthesis genes such as DHCR24, CYP51A1, and SQLE were downregulated, but other lipid metabolism genes such as ABCG1 (ATP binding cassette subfamily G member 1) and ABCA1 were upregulated (Fig. 4C). Ketotifen also promoted the expression of complement genes, such as C2, C6, and C1QA, as well as genes that are required for cross talk between innate and adaptive immune cells and the detection of pathogens by lymphocytes, such as CD69, CD86, HLA-A (major histocompatibility complex, class I, A), IL-18, and KLRD1 (Fig. 4C).

Transcriptional profiling revealed that both DENV infection and ketotifen treatment perturbed innate and adaptive immune cell signaling pathways in the liver and spleen, suggesting that immune cell population changes were occurring in these organs. MCs are known to recruit NKT cells in the skin to help clear peripheral DENV infection (39). However, the recruitment of these cells to systemic sites like the spleen may lead to inflammation and associated pathology. To characterize the effect of ketotifen treatment on immune cell recruitment and inflammation, we employed a tissue deconvolution algorithm that uses whole-organ gene expression data as its input. The digital cell quantification (DCQ) algorithm predicts the quantities of immune populations in an organ by pairing mouse immune cell transcriptional profiles curated by ImmGen (Immunological Genome Project) with whole-organ transcriptional profiles (40). ImmGen is a public resource that contains a compendium of microarray data derived from mouse immune cell populations isolated under standardized conditions. Many of the cell types profiled in ImmGen were isolated from the spleen under steady-state conditions or after perturbations such as bacterial or viral infection. Dengue disease is characterized by acute increases in the numbers of cytotoxic NK cells, NKT cells, and CD8⁺ T cells (41 – 44). DCQ predicted increases in the numbers of splenic NKT, NK, and CD8⁺ T cells after DENV infection (Fig. 5). DENV infection was predicted to increase the numbers of NK cells of the Ly49H^+/− and Ly49C/I^+/− subsets, and ketotifen was predicted to reduce this increase (Fig. 5A). These splenic NK cells were predicted to be activated, as they are transcriptionally similar to NK cells isolated after viral infection (Fig. 5A).

Numbers of splenic CD4^+/− NKT cells were also predicted to be increased after DENV infection. Although ketotifen treatment was predicted to reduce the DENV-induced increase in the number of CD4⁺ NKT cells, it was projected to increase CD4⁻ NKT cell quantities. CD44⁺ and CD44⁻ NKT cells were also differentially affected by ketotifen (Fig. 5B). Ketotifen treatment was projected to reverse the DENV-mediated reduction in the quantities of CD44⁻ NK1.1⁻ NKT cells while increasing the quantities of CD44⁺ NK1.1⁺ NKT cells (Fig. 5B). The numbers of effector CD8⁺ T cells were predicted to increase after DENV2 infection, and ketotifen treatment was predicted to reduce the numbers of these cells with effector phenotypes (Fig. 5C). Thus, ketotifen was predicted to increase the number of memory cells in both the NKT and CD8⁺ T cell populations in infected animals relative to vehicle treatment of infection.

To further assess the influence of ketotifen on splenic inflammation and validate whether the DCQ algorithm reflected cellular changes in tissue, we performed flow cytometry at the same time points as those that were used for RNA isolation. When we examined the flow cytometry data for NKT cells, as expected, we observed overall reduced numbers of NKT cells in the spleen with ketotifen treatment (Fig. 6A and B), but consistent with the expression data, increased numbers of these NKT cells were CD44⁺ (Fig. 6C and D). This suggests that although ketotifen reduced inflammation in the spleen, it did not prevent NKT cells from becoming activated. There was little change in the numbers of NK cells predicted by the DCQ algorithm on day 1, but by day 3, this analysis suggested that there were many fewer NK cells in ketotifen-treated mice (Fig. 5A). Flow cytometry analysis revealed that numbers of NK cells were significantly reduced on both day 1 (data not shown) and day 3 (Fig. 6A) after infection in ketotifen-treated mice compared to vehicle-treated mice. Similarly, there was a reduction in the number of NK cells expressing CD4 (an activation marker for NK cells) after ketotifen treatment (Fig. 6E and F). This result is consistent with our previous observation that MC activation results in the recruitment of NK and NKT cells (39), since here, blocking of MC activation by using ketotifen decreased the number of NK cells in the spleen. DCQ analysis also revealed perturbations in T cell subsets, consistent with the transition to adaptive immune responses that we expected to occur between days 1 and 3 of infection. In comparison to vehicle-treated DENV-infected mice, there were overall reduced total numbers of total CD8⁺ T cells in the spleens of ketotifen-treated mice (Fig. 6G and H). Thus, the results of flow cytometry showed generally consistent relative changes in cell populations compared to the expression data, but DCQ analysis sensitively detected changes in genes associated with certain reference populations. Overall, ketotifen-treated mice have reduced inflammation in the spleen but do not appear to have a reduced capacity for the activation of key innate immune cell types that are responsible for clearing infection, such as NK cells and NKT cells.

C57B/6NTac mice were purchased from InVivos, Singapore, and housed in the Duke-National University of Singapore (Duke-NUS) vivarium. Mice were infected with 1 × 10⁶ PFU of DENV2 strain EDEN2 by i.p. injection in 100 μl of phosphate-buffered saline (PBS). Drug-treated animals were given 0.6 mg per mouse per day of ketotifen (Sigma) or an equivalent volume of saline by i.p. injection, as previously optimized (24). The first dose was administered 1 h after infection, with daily drug injections thereafter at 24-h intervals. DENV was propagated in c6/36 cells and titrated by using standard methods, as previously described (39).

RNA extraction from tissues of virus- and mock-infected C57B/6NTac mice was performed in triplicate. Probe labeling and microarray slide hybridization for each biological replicate were performed by using the Mouse Whole Genome Microarray 4x44K kit (Agilent Technologies) according to the manufacturer's instructions. Slides were scanned on an Agilent DNA microarray scanner (model G2505B) using the XDR setting, and raw images were analyzed by using Agilent Feature Extraction software (version 9.5.3.1). Extracted raw data were partitioned into liver and spleen samples and separately background corrected by using the “norm-exp” method with an offset of 1 and quantile normalized by using the limma package in the R environment (78). Probes were filtered for low intensity, requiring at least two samples with intensity above a threshold set at the 5% quantile for intensity. This resulted in 32,986 probes retained for the liver samples and 29,608 probes retained for the spleen samples. Probes were mean-summarized by gene.

RNA was isolated by using the TRIzol/RNeasy hybrid RNA isolation protocol. For the detection of the negative strand, cDNA was synthesized by using an iScript Select cDNA synthesis kit using a sense primer, C14A (AATATGCTGAAACGCGAGAGAAACCGCG), followed by PCR with primer pair C14A and C69B (5′-CCCATCTCITCAIIATCCCTGCTGTTGG-3′), to amplify a 170-bp region from the capsid-PrM region of the DENV genome (79, 80). For the quantification of total DENV genome copy numbers, cDNA was synthesized by using primer C69B and an iScript cDNA synthesis kit, and quantitative real-time PCR was performed by using primer pair C14A and C69B and probe VICD2C38B (AGC ATT CCA AGT GAG AAT CTC TTT GTC AGC TGT) (80).

Differential expression was determined by comparing DENV-infected replicates to mock-infected samples based on a linear model for each gene by using limma. Criteria for differential expression were an absolute log₂ fold change (FC) of 0.58 and an adjusted P value of 0.05, calculated by using a moderated t test with subsequent Benjamini-Hochberg correction. Between the DENV2- and mock-infected liver samples, 192 upregulated DE genes and 128 downregulated DE genes were identified at day 1, and 8 upregulated DE genes and 1 downregulated DE gene were identified at day 3. Between the DENV2-infected, ketotifen-treated and mock-infected liver samples, 846 upregulated DE genes and 804 downregulated DE genes were identified at day 1, and 7 upregulated DE genes and 1 downregulated DE gene were identified at day 3. Between the DENV2-infected and mock-infected spleen samples, 213 upregulated DE genes and 7 downregulated DE genes were identified at day 1, and 55 upregulated DE genes and 6 downregulated DE genes were identified at day 3. Between the DENV2-infected, ketotifen-treated and mock-infected spleen samples, 667 upregulated DE genes and 374 downregulated DE genes were identified at day 1, and 52 upregulated DE genes and 11 downregulated DE genes were identified at day 3.

Functional analysis of statistically significant gene expression changes was performed by using the Ingenuity Pathway Analysis Knowledge Base (IPA; Ingenuity Systems). For all gene set enrichment analyses, right-tailed Fisher's exact test was used to calculate the probability that the enrichment of each biological function was due to chance alone. All enrichment scores were calculated by IPA, using the probes that passed our quality control (QC) filter as the background data set.

To infer the immune cell quantities for each DENV2- and mock-infected sample, we used the decomposition-based digital cell quantification (DCQ) algorithm (81). The algorithm utilizes an immune cell compendium of transcriptional profiles for 207 isolated immune cell subsets and the signatures of 61 predefined gene surface markers discriminating these cell types. Given the relative expression across each gene from the microarray samples, DCQ models differential expression as the sum of changes in quantities of the 207 immune cell types by using an “elastic net” regression technique and the signatures of the gene markers. The DCQ output reflects relative immune cell quantities for each immune cell subtype. Ten output models were generated, and relative cell quantities were taken from their average. A lambda minimum, a parameter of L1 and L2 regularization, of 0.2 was used.

Spleens were harvested, dissociated with collagenase (Sigma), and processed to single-cell suspensions by using a 70-μm cell strainer (BD Biosciences). Total cell numbers were quantified by using a hemocytometer. Cells were then stained with the following antibodies: anti-CD45-BUV395, anti-CD3e-peridinin chlorophyll protein (PerCP)-Cy5.5, anti-CD4-BV650, anti-CD8a-Alexa 700, anti-CD62L-phycoerythrin (PE)-Cy7, anti-CD44-BV510 (all from BD Biosciences), and anti-NK1.1-PE (BioLegend). Flow cytometry data were acquired with an LSRFortessa cell analyzer (BD Biosciences) and analyzed by using FlowJo software (FlowJo, LLC).

The SingHealth Institutional Animal Care and Use Committee approved animal protocols (2012/SHS/774) in accordance with the National Advisory Committee for Laboratory Animal Research (NACLAR).

All primary expression microarray data have been deposited in the NCBI Gene Expression Omnibus (accession number GSE100196 ).