Research ResourceHost-Microbe Interactions

The cytotoxic type 3 secretion system 1 of Vibrio rewires host gene expression to subvert cell death and activate cell survival pathways

See allHide authors and affiliations

Sci. Signal.  16 May 2017:
Vol. 10, Issue 479, eaal4501
DOI: 10.1126/scisignal.aal4501

Vibrio survives by rewiring host cell signaling

Vibrio parahaemolyticus (V. para) is a major cause of food poisoning caused by eating raw or undercooked shellfish. Like other types of bacteria, V. para delivers effectors into host cells through type 3 secretion systems (T3SSs). Those delivered through T3SS1 cause nonapoptotic death in various cultured cells. De Nisco et al. analyzed the transcriptional response of primary human fibroblasts to infection with a strain of V. para that has an intact T3SS1, which enables the bacterium to survive in its natural habitat, but lacks T3SS2, which mediates gastroenteritis in humans. V. para T3SS1 induced a rapid, dynamic transcriptional response that initially suppressed cell death and stimulated cell survival networks. Later in infection, effectors delivered through T3SS1 inhibited prosurvival signaling posttranslationally and induced cell death.

Abstract

Bacterial effectors potently manipulate host signaling pathways. The marine bacterium Vibrio parahaemolyticus (V. para) delivers effectors into host cells through two type 3 secretion systems (T3SSs). T3SS1 is vital for V. para survival in the environment, whereas T3SS2 causes acute gastroenteritis in human hosts. Although the natural host is undefined, T3SS1 effectors attack highly conserved cellular processes and pathways to orchestrate nonapoptotic cell death. To understand how the concerted action of T3SS1 effectors globally affects host cell signaling, we compared gene expression changes over time in primary fibroblasts infected with V. para that have a functional T3SS1 (T3SS1+) to those in cells infected with V. para lacking T3SS1 (T3SS1). Overall, the host transcriptional response to both T3SS1+ and T3SS1 V. para was rapid, robust, and temporally dynamic. T3SS1 rewired host gene expression by specifically altering the expression of 398 genes. Although T3SS1 effectors targeted host cells at the posttranslational level to cause cytotoxicity, V. para T3SS1 also precipitated a host transcriptional response that initially activated cell survival and repressed cell death networks. The increased expression of several key prosurvival transcripts mediated by T3SS1 depended on a host signaling pathway that is silenced posttranslationally later in infection. Together, our analysis reveals a complex interplay between the roles of T3SS1 as both a transcriptional and posttranslational manipulator of host cell signaling.

INTRODUCTION

Viral and bacterial toxins have proven to be vital tools for studying the cell biology of eukaryotic hosts because they have evolved to become potent and specific inhibitors and activators of host signaling pathways that control cellular proliferation, cytoskeleton and protein trafficking, intra- and extracellular signaling pathways, and protein homeostasis (1). An important toxin-delivery system used by Gram-negative bacteria is the type 3 secretion system (T3SS) (2). Derived from the flagellar apparatus, the T3SS is a needle-like structure that transfers specialized toxins (effectors) directly from the bacterial cytosol into the host cell cytoplasm (3). Within the host cell, effectors target and manipulate specific proteins to alter host cell processes for the benefit of the bacterium (1). Vibrio parahaemolyticus (V. para), a Gram-negative bacterium found in warm marine and estuarine environments, uses T3SSs for its survival both in the environment and within human hosts (4, 5). With the effects of global warming expanding its ecological niche, V. para has become the leading cause of acute seafood-borne gastroenteritis worldwide (4, 6). In addition to causing a self-limiting gastroenteritis, V. para can infect wounds that have been exposed to contaminated seawater (7, 8),

The V. para genome encodes several virulence factors, including thermostable hemolysins (TDH and TRH), polar and lateral flagella, adhesins, and two T3SSs (9). T3SS1 is found in all sequenced clinical and environmental isolates of V. para, whereas T3SS2 is found only in clinical isolates (9, 10). Because T3SS2 mediates enterotoxicity and invasion of host cells, it is the principal virulence factor responsible for gastroenteritis in both humans and animal models (5, 11, 12). The focus of this study is T3SS1, which is believed to be necessary for V. para’s survival in the environment but does not cause clinical gastroenteritis and does not contribute to gastroenteritis in animal models (9, 11). T3SS1 orchestrates an efficient, nonapoptotic cell death by targeting cell processes and pathways that are highly conserved from yeast to humans (13, 14). Environmental V. para is found in various marine organisms, including finfish, codfish, bivalve shellfish, and seaweed and within marine sediments, but the specific cell type that T3SS1 has evolved to target remains undefined (15, 16). However, several human and mouse cell lines have been key in uncovering the conserved biochemical functions of the T3SS1 effectors (14, 17).

T3SS1-mediated cell death occurs in distinct and highly reproducible stages due to the temporal action of the effectors it delivers (Fig. 1A) (14). As early as 15 min after infection, the effector VopQ (VP1680) forms an outward-rectifying channel that neutralizes the lysosome and inhibits autophagic flux, resulting in massive autophagosome accumulation within the host cell (18, 19). Within the first hour after infection, the secreted phosphatase VPA0450 hydrolyses the 5′ phosphate on phosphatidylinositol 4,5-bisphosphate (PIP2) at the plasma membrane, which causes blebbing by dislodging the membrane-associated actin cytoskeleton and reducing membrane stability (20). After 1 hour of infection, the Fic [filamentation induced by cyclic adenosine monophosphate (cAMP)] domain–containing protein VopS (VP1686) covalently attaches an AMP to the hydroxyl group of a threonine residue in the switch 1 region of Rho guanosine triphosphatases (GTPases). This posttranslational modification, termed AMPylation, inactivates Rho GTPases and precipitates cytoskeletal collapse and cell rounding (21, 22). At least one additional effector, VopR (VP1683), operates through a yet-uncharacterized mechanism (9, 23). At 2 to 3 hours after infection, the host cell ruptures and releases its cellular contents (Fig. 1A) (14).

Fig. 1 V. para T3SS1 is cytotoxic to primary fibroblasts.

(A) Schematic of T3SS1-induced cell death as previously defined in HeLa cells (14). The cell death is divided into stages defined by the action of known T3SS1 effectors (yellow, VopQ; cyan, VPA02450; magenta, VopS). (B) Representative confocal micrographs showing primary human fibroblasts infected with V. para POR3:T3SS1+. DNA is shown in blue (Hoechst), and actin is in green (phalloidin–Alexa Fluor 488). Scale bar, 20 μm. n = 3 independent experiments. (C) Lactate dehydrogenase (LDH) released by primary fibroblasts and HeLa cells infected with POR4:T3SS1 and POR3:T3SS1+. Percent cytotoxicity is relative to uninfected (UN) (negative) and Triton X-100–treated (positive) controls. Data represent the means of nine replicates [n = 3 (biological replicates) and 3 (technical replicates)] ± SD.

Although T3SS effectors are most often studied in isolation, they work in concert to manipulate host cell signaling to benefit the bacterium. Previous work has demonstrated that disruption of autophagic flux by VopQ antagonizes phagocytosis of V. para and prevents cell death by apoptosis (24). VopQ has also been implicated as a modulator of c-Fos and c-Jun protein abundance and interleukin-8 (IL-8) secretion in V. para–infected Caco-2 cells (25). In addition, VopS, through AMPylation of its target proteins Rho, Rac, and Cdc42, inactivates nuclear factor κB and mitogen-activated protein kinase (MAPK) signaling pathways, which are critical pathways through which the cell responds to stimuli such as bacterial infection (22, 26). Both VopQ and VopS interfere with the activation of the host NLRC4 (NOD-like receptor CARD domain–containing 4) inflammasome during V. para infection (27).

Because a growing body of evidence suggests that the V. para T3SS1 is not only cytotoxic to the host cell but also actively subverts host cellular immune responses, we postulated that the T3SS1 may alter host gene expression as a critical part of its subversive mechanism. Here, using genome-wide RNA sequencing (RNA-seq) methods, we demonstrate that V. para lacking T3SS1 induced the expected host cell response to pathogen-associated molecular patterns (PAMPs), whereas V. para with intact T3SS1 rewired the temporal expression patterns of hundreds of host genes. Furthermore, our systems-level analysis revealed that the host transcriptional response manipulated by T3SS1 altered signaling through several vital host pathways, which resulted in the activation of cell survival and repression of cell death networks. Our data suggest that altering host gene expression by the posttranslational targeting of critical cell processes, such as autophagy and Rho GTPase signaling, is an important mechanism by which T3SS1 subverts host immune responses to a Gram-negative bacterial infection.

RESULTS

V. para T3SS1 orchestrates death of primary human dermal fibroblasts

Although the identity of the organism or cell type in which V. para T3SS1 evolved to act is unknown, T3SS1-orchestrated cell death has primarily been characterized in HeLa cells and in RAW267.4 macrophages (14, 24). However, both cell lines are transformed and their high genetic heterogeneity may influence their transcriptional response to infection. Consequently, we sought to characterize V. para infection in primary human dermal fibroblasts [American Type Culture Collection (ATCC) PCS-201-01]. To study the effects of T3SS1 alone, we used a V. para strain, POR3, which is a derivative of the clinical strain RIMD2210633. This POR3:T3SS1+ strain (tdhASvcrD2) does not produce functional hemolysins or a functional T3SS2 but maintains an active T3SS1 as its primary virulence factor (5, 28). As a control for host cell response to PAMPs such as lipopolysaccharide (LPS), we also infected primary fibroblasts with the noncytotoxic strain POR4:T3SS1. This POR4:T3SS1 strain was derived from POR3 (tdhASvcrD1vcrD2) and does not produce a functional T3SS1 (28).

As previously reported in HeLa and RAW264.7 cells, POR3:T3SS1+ induced both cell rounding and rupture in primary fibroblasts (Fig. 1B), whereas POR4:T3SS1 induced no toxic effects (fig. S1) (14). We observed a lag in cytotoxicity in primary fibroblasts, with maximal cell death occurring at 240 min after infection compared to 150 min after infection in HeLa cells (Fig. 1C). Correspondingly, we have previously observed that maximal lysis occurs only 120 min after POR3 infection in RAW264.7 macrophages, supporting the idea that cell type–specific factors govern the timeline of cell lysis in response to T3SS1 (14).

V. para infection provokes a rapid and robust host transcriptional response

After characterizing the cellular response to a POR3:T3SS1+ infection in primary fibroblasts, we used genome-wide transcriptional profiling methods (RNA-seq) to measure changes in gene expression over time in response to infection with POR4:T3SS1 or POR3:T3SS1+. Time points were collected 45, 60, 75, and 90 min after infection. In total, libraries from 27 samples (triplicates of uninfected cells and of cells infected with POR3 and POR4 at t = 45, 60, 75, and 90 min) were sequenced and mapped to the human genome. The sequencing data passed all statistical quality control tests (fastqc and fastq_screen), and principal components analysis revealed tight clustering of technical replicates (fig. S2). We performed differential expression analysis across the time points for the two comparisons: uninfected versus POR4:T3SS1 and uninfected versus POR3:T3SS1+. Applying statistical cutoffs of false discovery rate (FDR) ≤ 0.01 and log2 counts per million (log2CPM) ≥ 0 and fold change (FC) cutoffs of −1.5 ≥ FC ≥ 1.5, we found that 1205 genes were differentially regulated between uninfected versus POR4:T3SS1-infected fibroblasts and 1385 genes between uninfected versus POR3:T3SS1+-infected fibroblasts (tables S1 and S2).

Hierarchical clustering of the differentially expressed genes revealed a distinct temporal pattern in gene expression changes in both comparisons (Fig. 2). By grouping genes with similar temporal differential expression patterns, we pinpointed groups of transcripts that were dynamically regulated during specific phases of the infection. Two major clusters resulting from both the uninfected versus POR4:T3SS1 and POR3:T3SS1+ differential expression analyses were the “early response” (Fig. 2A, clusters 3 to 5) and “late response” (Fig. 2A, cluster 6) clusters. Transcript abundance of genes in early response clusters was maximal at 45 min after infection, whereas late response genes were maximally expressed at 90 min after infection. For example, the abundance of the transcripts of many cytokines, including IL6, CXCL1, CXCL2, CXCL3, CXCL6, and IL8, increased (log2FC > 2) in response to infection with both POR4:T3SS1 and POR3:T3SS1+ (tables S1 and S2). All of these cytokines followed the late response expression pattern and were maximally expressed at 90 min after infection [Fig. 2, A (cluster 4) and B (cluster 6)].

Fig. 2 V. para infection induces rapid changes in host gene expression.

(A) Heat map of normalized differential expression between uninfected and V. para POR4:T3SS1-infected primary human fibroblasts over time. The transcripts depicted here met cutoffs of −1.5 ≥ FC ≥ 1.5, FDR ≤ 0.01, and log2CPM ≥ 0. Yellow indicates transcripts with higher abundance in POR4-infected fibroblasts compared to uninfected fibroblasts, and blue indicates transcripts with lower abundance in infected cells compared to uninfected cells. Clusters (colored bars on the left) were defined through hierarchical clustering of the differential expression data and represent groups of genes with similar temporal expression patterns. (B) Heat map of normalized differential expression between uninfected and POR3:T3SS1+-infected fibroblasts. Cutoffs and color coding are the same as in (A), and clusters were also assigned by hierarchical clustering.

The T3SS1-induced expression of several cytokines is cell type–specific

It was previously reported that V. para–induced expression and secretion of IL-8 depends on the T3SS1 effector VopQ in Caco-2 cells, but we did not observe a significant difference in IL8 expression between POR4:T3SS1- and POR3:T3SS1+-infected fibroblasts (tables S1 and S2) (25). To resolve this discrepancy, we used quantitative reverse transcription polymerase chain reaction (qRT-PCR) to measure IL8 mRNA in POR4:T3SS1- and POR3:T3SS1+-infected HeLa cells at 75 min after infection. We observed that the increase in IL8 expression in HeLa cells was T3SS1-dependent (fig. S3A). Together, these data suggest that the wiring of signaling networks regulating IL8 expression differs per cell type. Another important inflammatory cytokine, IL11, fell into the late response clusters [Fig. 2, A (cluster 6) and B (cluster 4)]. However, its expression was repressed by both POR4:T3SS1 and POR3:T3SS1+ early in infection, and although the repression was gradually alleviated in later time points, the IL11 transcript was never significantly more abundant in infected fibroblasts than in uninfected fibroblasts (tables S1 and S2). Because this repression of IL11 expression was unexpected, we measured IL11 mRNA in HeLa cells and RAW264.7 macrophages by qRT-PCR at 75 and 60 min after infection, respectively. We found that IL11 expression was specifically induced by POR3:T3SS1+ infection in HeLa cells and RAW264.7 macrophages (fig. S3B). Consistent with the IL8 data, the expression of IL11 in response to POR4:T3SS1 or POR3:T3SS1+ was cell type–specific, which may be due to the differences in Toll-like receptor expression between species and cell types (2931).

Expression of a family of zinc transcription factors is induced during the early response to V. para

For both fibroblasts infected with POR4:T3SS1 and fibroblasts infected with POR3:T3SS1+, the early response genes were dominated by genes encoding zinc finger transcription factors, for example, ZNF554, ZNF681, ZNF665, and ZKSCAN3, that are largely uncharacterized (Fig. 2A, clusters 3 to 5). The expression of over 20 ZNF and ZKSCAN genes was induced with log2FC > 1.1 at 45 min after infection compared to that of uninfected cells (table S1). At later infection time points, the expression of these genes resembled that in uninfected cells. The conservation of the expression patterns of these genes in both infection conditions suggests that expression of these transcripts is activated by V. para PAMPs and not by the specific action of the T3SS1 [Fig. 2B (clusters 6 and 7) and table S2].

A total of 398 host genes are differentially expressed in response to V. para T3SS1

To understand specifically how T3SS1 affects host cell gene expression, we performed differential expression analysis on RNA-seq data sets from POR4:T3SS1– versus POR3:T3SS1+-infected fibroblasts. Applying FC cutoffs of −1.5 ≥ FC ≥ 01.5 and statistical cutoffs of FDR ≤ 0.01 and log2CPM ≥ 0, we found that 398 genes were differentially expressed between POR4:T3SS1– and POR3:T3SS1+-infected fibroblasts (table S3). We used hierarchical clustering to uncover the major temporal patterns of T3SS1-specific gene expression and found that the 398 genes were grouped into six distinct clusters (Fig. 3A).

Fig. 3 T3SS1 specifically induces the differential expression of 398 genes.

(A) Heat map of normalized differential expression between primary human fibroblasts infected with V. para POR4:T3SS1 and those infected with V. para POR3:T3SS1+. The transcripts depicted here met cutoffs of −1.5 ≥ FC ≥ 1.5, FDR ≤ 0.01, and log2CPM ≥ 0 and were assigned into clusters by hierarchical clustering. Yellow denotes transcripts with increased abundance in POR3- compared to POR4-infected cells, and blue denotes a decreased abundance. (B) Heat map of log2FC values of the most significantly differentially expressed genes in (A). The log2FC between uninfected and either POR4- or POR3-infected fibroblasts is shown. Yellow indicates transcripts with increased abundance in infected cells compared to uninfected cells, and blue represents decreased abundance in infected cells compared to uninfected cells.

Cluster 4, which contains genes that are strongly activated later in infection, was the largest cluster and contained the most significantly differentially expressed genes (Fig. 3A). Notably, the transcripts of FOS, RGS2, KLF6, and TSC22D3 followed the early response pattern in POR4:T3SS1-infected fibroblasts, where expression of these genes increased early (45 min) but then returned to baseline (Fig. 3B). In contrast, transcripts of these genes in POR3:T3SS1+-infected cells followed a similar pattern early but then maintained a steady or late maximal induction over the course of the infection (Fig. 3B). For example, the transcript abundance of TSC223D3, which encodes an anti-inflammatory leucine zipper protein, increased in both POR4:T3SS1- and POR3:T3SS1+-infected cells early in infection (2.3-fold) but, after 90 min of infection, was 11.9-fold higher in POR3:T3SS1+-infected fibroblasts than in POR4:T3SS1-infected fibroblasts (tables S1 to S3).

Cluster 4 also contained the gene encoding the GTPase RhoB (RHOB), which was only differentially expressed during T3SS1-mediated infection. The abundance of RHOB mRNA was 8.5-fold higher at 75 min after infection and 18-fold higher at 90 min after infection in POR3:T3SS1+- versus POR4:T3SS1-infected fibroblasts (Fig. 3B and table S3). GEM, which encodes an inner membrane GTP-binding protein of unknown function, also fell into cluster 4. GEM expression was very low in fibroblasts throughout POR4:T3SS1 infection but steadily increased over the course of POR3:T3SS1+ infection (Fig. 3B).

Several genes encoding transcription factors displayed increased expression in POR3:T3SS1+-infected fibroblasts compared to in POR4:T3SS1-infected fibroblasts. These transcription factors primarily fell into clusters 4 and 6. The most significant T3SS1-induced transcription factors in cluster 4 include the Krüppel-like factors (KLFs) KLF2, KLF6, and KLF4, the cAMP-dependent transcription factor ATF3, and the activator protein 1 (AP-1) complex members FOS and JUND [Fig. 3, A (cluster 4) and B]. The AP-1 complex member JUN, however, fell into cluster 6 [Fig. 3, A (cluster 6) and B]. The expression of most genes in cluster 6, including JUN, increased earlier in infection than the genes in cluster 4.

In addition to JUN, cluster 6 contained the inducible prostaglandin-endoperoxide synthase 2 (PTGS2; also known as COX2) [Fig. 3A (cluster 6) and table S3]. Starting at 60 min after infection, PTGS2 expression increased steadily in POR3:T3SS1+-infected fibroblasts and, at 90 min, was 39-fold higher than in uninfected and 9-fold higher than in POR4:T3SS1-infected fibroblasts (Fig. 3B and tables S2 and S3). Cluster 6 also contained the early growth response transcription factor EGR1. We found that POR4:T3SS1 also significantly induced EGR1 expression in fibroblasts early during the infection time course, but this increase was not sustained through later stages of infection. However, the POR3:T3SS1+ induction of EGR1 expression was sustained throughout the infection and was higher than the expression induced by POR4:T3SS1 (Fig. 3B).

T3SS1-induced expression of several genes is conserved in different cell types

To test the conservation of the host transcriptional response to T3SS1 in other cell types, we measured the mRNA abundance of several of the genes whose transcription was strongly activated by T3SS1 in primary fibroblasts in infected HeLa cells and RAW264.7 macrophages by qRT-PCR. RHOB and KLF2 expression was greatly increased during POR3:T3SS1+ infection, but not during POR4:T3SS1 infection, in both HeLa cells and RAW264.7 macrophages (fig. S4, A and B). KLF6 and FOS also showed similar expression patterns in HeLa cells and RAW264.7 macrophages as compared to primary fibroblasts because the abundance of both transcripts was significantly increased upon POR3:T3SS1+ infection but not upon POR4:T3SS1 infection (fig. S4, C and D).

The expression of several genes induced by POR4:T3SS1 infection differed between HeLa cells and RAW264.7 macrophages. In both cell types, we measured significantly increased transcript abundance for JUN, EGR1, ATF3, PTGS2, and GEM upon POR3:T3SS1+ infection (fig. S4, D to I). However, the relative expression of all these genes was also somewhat increased in POR4:T3SS1-infected RAW264.7 macrophages but not in POR4:T3SS1-infected HeLa cells (fig. S4, D to I). This ability of POR4:T3SS1 to induce significant expression of these transcripts in RAW264.7 macrophages, but not in HeLa cells, highlights the differential ability of these cell types to respond to environmental cues like bacterial PAMPs.

T3SS1 mediates differential activation or repression of core pathways over time

To understand how the observed differential gene expression in response to V. para T3SS1 changed host cell signaling events, we used Ingenuity Pathway Analysis (IPA; Qiagen) of our two sets of temporal differential expression data [uninfected versus POR4:T3SS1 and uninfected versus POR3:T3SS1+ (Figs. 2 and 3)]. To maximize the expression data available to the IPA software, we loosened the inclusion criteria by applying filters of −0.3 ≥ log2FC ≥ 0.3, log2CPM ≥ 0, and FDR ≤ 0.05. IPA canonical pathway analysis, which compares well-established signaling pathways, predicted that several host signaling pathways were significantly activated or repressed in each comparison (tables S4 and S5). Two pathways that were predicted to be strongly affected in POR4:T3SS1-infected, but not in POR3:T3SS1+-infected, primary fibroblasts were signaling through triggering receptor expressed on myeloid cells 1 (TREM1) and GαS [heterotrimeric GTP-binding protein (G protein) subunit α S] (Fig. 4). In response to the stimulus of POR4:T3SS1, the expression of genes encoding cytokines that are transcriptional targets of TREM1 signaling and the costimulatory proteins ICAM1 and CD83 was activated in all time points. Conversely, GαS signaling was predicted to be strongly repressed by POR4:T3SS1 due to the decreased expression of genes encoding components of GαS signaling and increased expression of the gene encoding RGS2, a negative regulator of GαS signaling, in POR4:T3SS1-infected fibroblasts.

Fig. 4 V. para differentially affects core pathways based on the activity of T3SS1.

Heat map of predicted activation (yellow) and repression (blue) Z scores calculated by Qiagen’s IPA using the temporal uninfected versus POR4:T3SS1 and uninfected versus POR3:T3SS1+ differential expression data. The color key correlates the displayed heat map color and calculated Z scores. Gray denotes that a pathway was unaffected (P < 0.05) in the infected condition relative to uninfected.

Several pathways were predicted to be activated or repressed specifically in POR3:T3SS1+-infected fibroblasts but not in POR4:T3SS1-infected fibroblasts. On the basis of the expression of genes encoding components of these pathways, signaling by Rho family GTPases, including RhoA and Rac, was predicted to be activated in response to POR3:T3SS1+ starting at 60 min after infection (Fig. 4). Correspondingly, RhoGDI (Rho guanosine diphosphate dissociation inhibitor) signaling, which inhibits Rho GTPases, was repressed by POR3:T3SS1+ starting at 60 min after infection (Fig. 4) (32). MAPK pathways were also differentially affected by POR4:T3SS1 and POR3:T3SS1+ infection. The IPA software predicted extracellular signal–regulated kinase (ERK) and MAPK signaling to be moderately activated at 60 min after infection during POR4:T3SS1 infection but not during POR3:T3SS1+ infection (Fig. 4 and tables S4 and S5). ERK and MAPK signaling were repressed in POR3:T3SS1+-infected fibroblasts 90 min after infection (Fig. 4). In addition, p38 MAPK signaling was differentially affected by POR3:T3SS1+ and POR4:T3SS1. The action of T3SS1 greatly repressed p38 MAPK signaling from 45 to 75 min after infection, but at 90 min after infection, the pathway was suddenly activated (Fig. 4). Given the role of p38 MAPK signaling in cytokine-induced mRNA stability, it is an intriguing hypothesis that the activation of this pathway contributes to the observed increase in mRNA abundance of some genes later in POR3:T3SS1+ infection (Fig. 3A, cluster 4) (33).

T3SS1 manipulates host gene expression through the MAPK pathway

In all cell lines tested, RHOB expression was specifically increased by T3SS1. Because many of the pathways affected by T3SS1 in our IPA canonical pathway analysis center on MAPK signaling, we tested whether the T3SS1-specific activation of RHOB expression was induced by MAPK signaling. We treated primary fibroblasts and RAW264.7 macrophages with either the MAPK kinase 1 and 2 (MEK1/2) inhibitor U0126 or dimethyl sulfoxide (DMSO) (vehicle), infected with POR3:T3SS1+, and then monitored RHOB expression by qRT-PCR. After U0126 pretreatment, POR3:T3SS1+ infection did not increase RHOB expression in fibroblasts or macrophages, and RHOB transcript abundance was similar to quantities observed in DMSO-pretreated POR4:T3SS1-infected cells (Fig. 5, A and B, and fig. S5, A and B). These data indicate that the T3SS1-induced expression of RHOB depends on MEK1/2 activity in these cell types. We next tested whether other transcripts induced by T3SS1 also depended on MEK1/2 activity for their T3SS1-specific increased expression. As reported previously in Caco-2 cells (25), the T3SS1-specific induction of FOS expression in primary fibroblasts was inhibited by U0126 treatment (Fig. 5C). However, in contrast to previous observations in Caco-2 cells (25), we found that the T3SS1-triggered increase in JUN expression was not significantly inhibited by U0126 treatment in primary fibroblasts (Fig. 5C). Whereas U0126 treatment attenuated the T3SS1-induced expression of KLF2, PTGS2, and EGR1 in primary fibroblasts, it did not significantly alter ATF3 expression (Fig. 5, D and E).

Fig. 5 T3SS1-specific induction of several transcripts requires MEK1/2 activity.

(A) Immunoblot showing phosphorylated ERK1 and ERK2 (p-ERK1/2) and total ERK1/2 in POR3:T3SS1+-infected primary human fibroblasts treated with DMSO (vehicle) or U0126 to demonstrate the activity of U0126 as an MEK inhibitor in primary fibroblasts. n = 3 independent experiments. (B to E) qRT-PCR showing the expression of RHOB, JUN, FOS, PTGS2, EGR1, KLF2, and ATF3 in primary human fibroblasts that were pretreated with the MEK1/2 inhibitor U0126 or DMSO before being infected with POR3:T3SS1+ or POR4:T3SS1 compared to uninfected cells. Expression was normalized to the housekeeping gene IPO8. Data are 2−ΔΔCt ± SD. n = 3 experiments. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, one-way analysis of variance (ANOVA) analysis and Tukey’s multiple comparisons test. ns, not significant.

Network analysis predicts that cell survival pathways are activated and cell death pathways are repressed during T3SS1-induced cell death

Next, we sought to understand how the observed changes in gene expression and pathway activation in response to POR3:T3SS1+ infection affected broader cell signaling networks and cellular functions by performing network analysis using IPA (table S6). Surprisingly, two of the most highly activated biological functions at 90 min after POR3:T3SS1+ infection were organismal cell survival and cell viability (Figs. 6 and 7A). Most of the differentially expressed genes in fibroblasts infected with POR3:T3SS1+ promote cell survival. These genes encode cytokines, transcription factors, receptors, kinases, peptidases, enzymes, and phosphatases and include genes focal to this study, such as KLFs, FOS, JUN, EGR1, and PTGS2 (Fig. 6). Although the expression state of some genes (ATF3 and TSC22D3) is predicted to repress cell survival, the expression state of most genes promotes cell survival. Collectively, this expression pattern activates the cell survival network. Conversely, cell death networks were predicted to be strongly repressed during POR3:T3SS1+ infection (Fig. 7A). These data suggest that by posttranslational targeting of key cellular functions, the T3SS1 effectors alter host gene expression to enhance prosurvival signals within the host, thereby masking their cytotoxic effects and perhaps dampening the host immune response (Fig. 7B).

Fig. 6 Key differentially expressed genes activate cell survival in POR3-infected fibroblasts.

Illustration of the action key POR3 differentially expressed genes in the cell survival network generated using IPA. The subcellular localization of the protein products of each gene and differential expression status (yellow, increased; blue, decreased) are depicted. Dashed lines denote indirect interactions. Orange lines indicate a gene expression state consistent with activation of cell survival; yellow lines indicate an expression state inconsistent with cell survival activation.

Fig. 7 V. para T3SS1 activates cell survival networks and represses cell death networks.

(A) Heat map of IPA Z score prediction of activation (yellow) or repression (blue) of biological networks in response to POR3:T3SS1+ infection over time using the temporal uninfected versus POR3:T3SS1+ differential expression data from Fig. 2B. (B) Predicted cell survival and organismal death network activation (>0) and repression (<0) by POR3:T3SS1+ over the time course of T3SS1-mediated cell death. Plotted values correspond to the IPA Z scores from cell survival and organismal death networks in (A) and table S6.

Many additional interesting disease and biological function networks were predicted to be activated or repressed during POR3:T3SS1+ infection (table S6), especially the expression and transactivation of RNA (Fig. 7A). These transcription-centered networks are both maximally activated later in infection, which is consistent with the high number of genes that we found to be activated by T3SS1 later in infection (Fig. 3A, clusters 4 and 6). The activation of these networks suggests that even as the T3SS1 toxins perturb vital cellular processes, T3SS1 does not shut down host transcription but activates it (Fig. 7B).

DISCUSSION

V. para–induced transcripts reveal new players in bacteria-induced innate immunity

Herein, we have discovered previously unrecognized genes whose expression increases in response to a Gram-negative bacterial pathogen. By measuring the temporal gene expression changes in host fibroblasts infected with a noncytotoxic strain of V. para (POR4:T3SS1) lacking the three major virulence factors (tdhASvcrD1vcrD2), we identified genes whose expression was primarily modulated by general bacterial PAMPs and not the specific action of V. para T3SS1 effectors. We identified a class of uncharacterized zinc finger transcription factors that are transcriptionally induced within 45 min after infection by either POR4:T3SS1 or POR3:T3SS1+ [Fig. 2, A (clusters 3 to 5) and B (clusters 6 and 7)]. This pattern of regulation suggests that these uncharacterized transcription factors may be involved in the early innate immune response to Gram-negative bacteria. The cellular function of most of these transcription factors is unknown, but ZKSCAN3 has been shown to repress autophagy (34). Perhaps, POR3:T3SS1+ counteracts this repression by injecting the effector VopQ, which acts early during infection to interrupt autophagic flux and induce autophagosome accumulation by a posttranslational mechanism (18, 19, 24).

The expression of CXCL1, CXCL2, and CXCL3, which encode chemokines that promote antimicrobial innate immune responses, was induced in fibroblasts infected by both POR4:T3SS1 and POR3:T3SS1+ starting at 45 min after infection (log2FC > 1) and rose steadily as the infection proceeded [Fig. 2, A (cluster 6) and B (cluster 4)] (35). This kinetic profile indicates that primary fibroblasts can quickly increase the expression of these transcripts in response to bacterial PAMPs and sustain them even through the cytotoxic action of T3SS1 effectors. Although Gram-negative bacteria, like V. para, are especially susceptible to the defense responses elicited by these chemokines, the V. para T3SS1 does not dampen their expression (35). It is possible, however, that these secreted chemokines might be subverted on the posttranscriptional level by a yet-to-be-discovered T3SS1 effector.

T3SS1 rewires host gene expression during infection

The primary aim of this study was to determine the extent to which the action of T3SS1 specifically alters host gene expression. Through differential expression analysis of POR4:T3SS1- versus POR3:T3SS1+-infected primary fibroblasts, we found that the expression of 398 genes was specifically affected by T3SS1 (Fig. 3A). Most of the affected genes had not been previously linked to V. para infection or T3SS1-mediated cytotoxicity. For example, expression of EGR1, encoding the transcription factor early growth response 1, is transiently activated in epithelial cells and macrophages by bacterial pathogens (Helicobacter pylori) and PAMPs (LPS and soluble peptidoglycan), but we observed that T3SS1 enhanced and extended EGR1 expression throughout V. para infection (36, 37). Our data suggest that EGR1 expression is not only activated by generic Gram-negative PAMPs but also by the specific action of the T3SS1. We observed that signaling through MEK1/2 is required for the increase in EGR1 expression in response to V. para T3SS1, which has been previously noted in response to Gram-negative PAMPs during H. pylori infection (Fig. 5D) (36).

The expression of PTGS2 (also known as COX2) is induced by both viral and bacterial PAMPs (38). We show that PTGS2 expression is more rapidly and robustly induced by POR3:T3SS1+ than by POR4:T3SS1 (Fig. 3, A and B) and that PTGS2 expression is induced by POR3:T3SS1+ through MEK1/2. Another study demonstrated that PTGS2-dependent inflammation allows uropathogenic Escherichia coli to circumvent host resistance to colonization and cause severe recurrent urinary tract infections (39). Clearly, PTGS2 plays an important role in Gram-negative infections, but overexpression of PTGS2 is also associated with resistance to apoptosis and promotion of cancer cell growth and survival (40). Deciphering the exact mechanism of MEK1/2-dependent increase in PTGS2 transcript abundance during POR3:T3SS1+ will provide important insight into how this gene is regulated in other disease states.

T3SS1-induced RHOB expression requires activation of a pathway that is silenced later in infection

The abundance of RHOB protein has previously been shown to increase in macrophages during POR3:T3SS1+ infection (41), but the kinetics of RHOB expression had not been analyzed. Our data clearly demonstrate that T3SS1 quickly and robustly increases RHOB transcript abundance in primary fibroblasts and that T3SS1-induced expression is conserved in both RAW264.7 macrophages and HeLa cells (Fig. 3B and fig. S4A). Previously, we identified Rho GTPases as targets of AMPylation by VopS and found that the modified GTPases could no longer activate downstream signaling molecules (21). However, VopS does not AMPylate Rho GTPases until 1 hour after infection when host cells start rounding (21, 22). In addition, AMPylation of Rho GTPases inhibits signaling by MAPK. RHOB transcription increased substantially before cell rounding in an MAPK-dependent manner, which leads us to propose that the mechanism and timing of RHOB transcript induction and posttranslational silencing by T3SS1 is likely more complex than predicted (22, 41, 42). Further work to understand the upstream mechanism of T3SS1-mediated induction of RHOB expression, as well as the downstream function of RhoB in infected cells, will be crucial to understanding the complex interplay between T3SS1 effectors and these host signaling outputs during infection.

T3SS1 specifically induces expression of genes encoding Krüppel-like transcription factors

A key observation in our analysis was the T3SS1-specific activation of transcripts encoding the Krüppel-like zinc finger transcription factors KLF2, KLF6, and KLF4. The T3SS1-induced expression of KLF2 and KLF6 was conserved in HeLa cells and RAW264.7 macrophages (fig. S3, B and C). KLF2 is an immunosuppressive transcription factor that induces quiescence in various T cell lines, and KLF6 controls transforming growth factor β (TGF-β) expression and represses cell proliferation by inducing the expression of the gene encoding the cell cycle inhibitor p21 and by interacting with Jun and cyclin D1 (43, 44). Our work demonstrates that the T3SS1-induced expression of KLF2 is dependent on signaling through the ERK-MAPK pathway, specifically on the activity of MEK1/2 (Fig. 5D). Rho-inactivating toxins, like YopT from Yersinia spp., ExoS and ExoY from Pseudomonas aeruginosa, and C3 exotoxin from Clostridium botulinum, induce KLF2 and KLF6 expression in several cell types, such as macrophages, endothelial cells, and epithelial cells (4547). Given that the T3SS1 effector VopS inactivates Rho GTPases, including RhoA, during T3SS1-mediated cell death, it is an attractive hypothesis that VopS is responsible for inducing the expression of KLF2, KLF6, and perhaps even KLF4 (22).

Pathway analysis reveals new signaling pathways targeted by T3SS1 effectors

By analyzing host gene expression over the course of an infection, we discovered new host signaling pathways targeted by T3SS1 effectors. Specifically, T3SS1 subverted the predicted activation of TREM1 signaling and repression of GαS signaling induced by POR4:T3SS1 (Fig. 4). It has been demonstrated that TREM1 can be activated by LPS, which may explain the predicted activation of the pathway in POR4-infected fibroblasts (48). TREM1 activation initiates signaling through Janus kinase 2 (JAK2), protein kinase B (PKB; also known as Akt), ERK1, and ERK2 that results in increased expression of genes involved in the inflammatory response (49). Our data suggest that, through an unknown mechanism, T3SS1 subverts TREM1 activation by LPS. It is not clear how GαS signaling is so strongly repressed in POR4:T3SS1-infected fibroblasts but not in POR3:T3SS1+-infected fibroblasts. GαS signaling centers on signal transduction between G protein–coupled receptors and GαS to activate downstream effectors, such as protein kinase A, and downstream signaling outputs affect cell proliferation, cytoskeletal remodeling, and cell cycle regulation (50).

Conversely, signaling by Rho GTPases and RhoGDI was specifically altered by POR3:T3SS1+ and not by POR4:T3SS1 (Fig. 4). Signaling by Rho GTPases was activated, whereas RhoGDI signaling was repressed. Previous studies have shown that depletion of RhoGDI constitutively activates Rho GTPases and increases PTGS2 (COX2) expression, which mirrors our observations of increased expression of PTGS2 and activation of Rho GTPase signaling in primary fibroblasts (Figs. 3B and 4) (51). Perhaps the increase of PTGS2 expression and predicted activation of Rho GTPase signaling in POR3:T3SS1+-infected fibroblasts are due to a specific suppression of RhoGDI signaling by T3SS1 effectors.

The Gram-negative T3SS controls host signaling at multiple regulatory levels

Network analysis of the differential expression data provided one of the most unanticipated insights of this study: Although the T3SS1 mediates a methodical, orchestrated posttranslational death through the action of an outward-rectifying ion channel (VopQ) to block autophagic flux, a PIP2-5-phosphatase (VPA0450) to compromise the integrity of the cell membrane, and a Rho GTPase AMPylator (VopS) to inhibit cellular cytoskeletal signaling, the T3SS1 also simultaneously manipulates the host response to infection by altering the expression of hundreds of host genes to activate cell survival signaling pathways (Fig. 7) (14, 18, 20, 21). Conceptually, this study provides unique insight into the classical view of the activity of a T3SS and its effectors. Beyond their roles as posttranslational mediators of cell invasion or cytotoxicity, these secretion systems profoundly rewire host cell signaling at the transcriptional level, thereby altering the host environment to promote the proliferation and survival of the bacterial pathogen.

MATERIALS AND METHODS

Bacterial strains and culture conditions

The V. para POR3 (POR1ΔvcrD2) and POR4 (POR1ΔvcrD1ΔvcrD2) strains were provided by T. Iida and T. Honda of Osaka University. The POR3 and POR4 strains were cultured at 30°C in MLB (LB broth + 3% NaCl).

Mammalian cell culture

Primary adult dermal fibroblasts were purchased from ATCC and revived and maintained at 5% CO2 and 37°C in a low-serum primary fibroblast medium (ATCC) according to ATCC instructions. HeLa cells (ATCC) and RAW264.7 macrophages were maintained at 5% CO2 and 37°C in high-glucose Dulbecco’s modified Eagle’s medium (DMEM; Gibco) supplemented with 10% (v/v) fetal bovine serum (Sigma-Aldrich), 1% (v/v) penicillin-streptomycin-glutamine, and 1% (v/v) sodium pyruvate.

Infection of primary and transformed cell lines

For RNA-seq, primary human fibroblasts were seeded onto six-well plates at a density of 1 × 105 cells/ml and grown for 18 to 20 hours to ~80% confluency. Overnight, V. para cultures were normalized to an optical density at 600 nm (OD600) of 0.2 and subcultured to an OD600 of 0.6. Bacteria were pelleted, resuspended in unsupplemented DMEM, and grown at 37°C for 30 min to induce T3SS1 expression (20). Primary fibroblasts were washed with unsupplemented DMEM and then infected with preinduced POR3 or POR4 strains at a multiplicity of infection (MOI) of 10. Plates were centrifuged at 1000g for 5 min to synchronize infection and incubated at 37°C and 5% CO2. At 45, 60, 75, and 90 min after infection, RNAprotect Cell Reagent (Qiagen) was added to halt the infection and preserve the RNA. Cells were harvested by scraping and subsequent cell pellets were resuspended in Buffer RLT Plus (Qiagen) and stored at −80°C. The same infection protocol was followed for qRT-PCR experiments with primary fibroblasts, HeLa cells, and RAW264.7 macrophages, except macrophages were seeded at a density of 5 × 104 cells/ml. For MEK1/2 inhibitor experiments, cells were starved in plain DMEM for 30 min and then incubated with 10 μM U0126 (Cell Signaling) or 10 μM DMSO (vehicle) for 1 hour before infection.

RNA isolation and gene expression RNA-seq

RNA isolation was performed the same for all cell types: primary fibroblasts, HeLa cells, and RAW264.7 macrophages. Cells were lysed with 27-gauge half-inch needles before homogenization with QIAshredder columns (Qiagen). Total RNA was then purified using the RNeasy Plus Kit (Qiagen). Purified total RNA samples were run on the Agilent 2100 Bioanalyzer to determine RNA quality. Only samples with an RNA integrity number score of 9 or higher were used. RNA concentration was determined with a Qubit fluorometer before library preparation. Four micrograms of total deoxyribonuclease-treated RNA was prepared with the TruSeq Stranded Total RNA Low-Throughput Sample Prep Kit from Illumina. In brief, polyadenylated RNA was purified and fragmented before strand-specific complementary DNA (cDNA) synthesis. cDNA was A-tailed, and indexed adapters were ligated. After adapter ligation, samples were PCR-amplified and purified with AMPure XP beads and then validated again on the Agilent 2100 Bioanalyzer. Samples were quantified with the Qubit fluorometer before being normalized, pooled, and then sequenced on the Illumina HiSeq 2500 using SBS v3 reagents. Read lengths were 50-nucleotide single-end, and the sequencing depth was at least 25 million reads per sample.

RNA-seq analysis methods

First, FASTQ files were analyzed using FastQC v0.11.2 (52) and FastQ Screen v0.4.4 (53), and reads were quality-trimmed using fastq-mcf (ea-utils/1.1.2-806) (54). The trimmed reads were mapped to the hg19 assembly of the human genome (the University of California, Santa Cruz, version from igenomes) using TopHat (55). Duplicated reads were marked using Picard tools (v1.127; https://broadinstitute.github.io/picard/), read counts were generated using featureCounts (56), and differential expression analysis was performed using edgeR (57). For differential expression analysis, statistical cutoffs of FDR ≤ 0.01 and log2CPM ≥ 0 and FC cutoffs of −1.5 ≥ FC ≥ 1.5 were used to identify statistically significant and possibly biologically relevant differentially regulated transcripts. Pathway and network analysis were conducted using Qiagen’s IPA tool (www.qiagen.com/ingenuity). Differentially expressed gene heat maps were clustered by hierarchical clustering using R (http://www.R-project.org).

Quantitative RT-PCR

Primers were designed to amplify 100 to 150 base pairs of each target gene (table S7) and tested for efficiency. RNA was isolated from infected primary fibroblasts and HeLa cells 75 and 90 min after infection and from infected RAW264.7 macrophages 60 min after infection by the methods described above. RNA concentration was measured via NanoDrop, and cDNA was generated using the iScript cDNA Synthesis Kit (Bio-Rad). Transcripts were quantified on a CFX384 Touch Real-Time PCR Detection System using the iTaq Universal SYBR Green Supermix (Bio-Rad) and 500 nM primers. Relative gene expression for each target gene was calculated by the −ΔΔCt method with respect to transcript levels in uninfected cells using the reference gene IPO8 for normalization (58).

Confocal microscopy

HeLa cells and primary fibroblasts were seeded at 1 × 105cells per well onto ultraviolet-sterilized glass coverslips in six-well plates. Following the infection protocol described above, samples were first washed with 1× phosphate-buffered saline (PBS) and then fixed with 3.2% (v/v) paraformaldehyde for 10 min at room temperature. Cells were permeabilized with 0.5% (w/v) saponin for 4 min at room temperature and then washed in 0.1% saponin. F-actin was stained with Alexa Fluor 488–phalloidin (2 U/ml; Thermo Fisher Scientific), and DNA was stained with Hoechst A33342 (1 μg/ml; Invitrogen). Coverslips were mounted on glass slides with ProLong Gold Antifade mounting medium (Thermo Fisher Scientific) and imaged on a Zeiss LSM 800 confocal microscope. Images were converted using ZEN 2 software (Zeiss).

LDH release assay

HeLa cells and primary fibroblasts were seeded in 24-well plates. Before infection, mammalian cells were washed in unsupplemented DMEM (phenol red–free). Infection was performed at an MOI of 10 with preinduced POR3 or POR4, as described above. At each time point, triplicate medium samples were collected in 96-well plates, and the activity of LDH was measured using a Cytotoxicity Detection kit (Takara). Percent LDH release was calculated compared to cells lysed in 1% Triton X-100.

Immunoblotting

Primary fibroblasts and RAW264.7 macrophages were infected as described above and, for each time point, were washed with 1× ice-cold PBS. Fibroblasts were treated with 0.25% trypsin-EDTA (Thermo Fisher Scientific) before collection by scraping to aid in sample collection, and macrophages were collected by scraping alone. Collected cells were pelleted (1000g) and washed twice in ice-cold 1× PBS + protease inhibitors and lysed in 50 mM tris (pH 8), 150 mM NaCl, 5 mM EDTA, 1% NP-40, 0.5% sodium deoxycholate, and 0.1% SDS with protease and phosphatase inhibitors (Roche Applied Science) for 20 min on ice before gel electrophoresis and immunoblotting. Total ERK1/2 and phospho-ERK1/2 were detected with p44/42 MAPK (137F5) and phospho-p44/42 MAPK T202/Y204 (197G2) primary antibodies, respectively, and donkey secondary antibodies against rabbit (GE Healthcare).

SUPPLEMENTARY MATERIALS

www.sciencesignaling.org/cgi/content/full/10/479/eaal4501/DC1

Fig. S1. Confocal microscopy of POR4:T3SS1-infected fibroblasts.

Fig. S2. Principal components analysis of RNA-seq data.

Fig. S3. qRT-PCR analysis of cytokines in HeLa cells and RAW264.7 macrophages.

Fig. S4. qRT-PCR analysis of top fibroblast POR4 versus POR3 differentially expressed genes in HeLa cells and RAW264.7 macrophages.

Fig. S5. U0126 treatment inhibits POR3-induced Rhob expression in RAW264.7 macrophages.

Table S1. Significantly differentially expressed genes in uninfected versus POR4:T3SS1-infected fibroblasts.

Table S2. Significantly differentially expressed genes in uninfected versus POR3:T3SS1+-infected fibroblasts.

Table S3. Significantly differentially expressed genes in POR4:T3SS1- versus POR3:T3SS1+-infected fibroblasts.

Table S4. IPA canonical pathways uninfected versus POR4:T3SS1.

Table S5. IPA canonical pathways uninfected versus POR3:T3SS1+.

Table S6. IPA network analysis (disease and biological functions) of uninfected versus POR3:T3SS1+.

Table S7. Primer list.

REFERENCES AND NOTES

Acknowledgments: We thank R. Kittler and the Next Generation Sequencing Core in the Eugene McDermott Center for Human Growth and Development for technical assistance. We also thank P. Douglas and the members of the Orth Lab for discussion and technical assistance. Funding: This work was funded by the NIH (grant R01-AI056404), the Welch Foundation (grant I-1561), and Once Upon A Time Foundation. N.J.D.N. is funded by the Infectious Disease Training (grant T32AI070116-10). J.F. is funded by the Microbiology Training (grant T32 AI007520-17). C.X. was partially supported by the NIH (grant UL1TR001105). K.O. is a Burroughs Wellcome Investigators in Pathogenesis of Infectious Disease, a Beckman Young Investigator, and a W.W. Caruth Jr. Biomedical Scholar and has an Earl A. Forsythe Chair in Biomedical Science. Author contributions: Conceptualization: N.J.D.N. and K.O. Methodology: N.J.D.N., M.K., C.X., and K.O. Formal analysis: N.J.D.N. and M.K. Investigation: N.J.D.N., P.L., and J.F. Resources: K.O. and C.X. Writing (original draft): N.J.D.N. and K.O. Writing (review and editing): N.J.D.N., K.O., J.F., P.L., and M.K.; and Funding acquisition: K.O. and C.X. Competing interests: The authors declare that they have no competing interests. Data and materials availability: The RNA-seq data are deposited in the Gene Expression Omnibus under accession no. GSE93783.
View Abstract

Navigate This Article