SARS-CoV Virus-Host Interactions and Comparative Etiologies of Acute Respiratory Distress Syndrome as Determined by Transcriptional and Cytokine Profiling of Formalin-Fixed Paraffin-Embedded Tissues

Abstract

These studies attempt to understand more fully the host response and pathogenesis associated with severe acute respiratory syndrome (SARS) coronavirus (SARS-CoV) by monitoring gene expression using formalin-fixed paraffin-embedded (FFPE) pulmonary autopsy tissues. These tissues were from patients in different hospitals in Singapore who were diagnosed with various microbial infections, including SARS-CoV, that caused acute respiratory distress syndrome (ARDS). Global expression patterns showed limited correlation between end-stage ARDS and the initiating pathogen, but when focusing on a subset of genes implicated in pulmonary pathogenesis, molecular signatures of pulmonary disease were obtained and appeared to be influenced by preexisting pulmonary complications and also bacterial components of infection. Many factors detected during pulmonary damage and repair, such as extracellular matrix (ECM) components, transforming growth factor (TGF) enhancers, acute-phase proteins, and antioxidants, were included in the molecular profiles of these ARDS lung tissues. In addition, differential expression of cytokines within these pulmonary tissues were observed, including notable genes involved in the interferon (IFN) pathway, such as Stat1, IFN regulatory factor-1 (IRF-1), interleukin-6 (IL-6), IL-8, and IL-18, that are often characterized as elevated in ARDS patients.

INTRODUCTION

The clinical course of severe acute respiratory syndrome coronavirus (SARS-CoV) infection is variable, with fatal cases progressing to acute respiratory distress syndrome (ARDS) and the predominant pathology being diffuse alveolar damage (DAD).^1–4 SARS has captured the attention of the world as a new disease, but it has been suggested in clinical terms that SARS is a variant of ARDS attributed to SARSCoV.⁵ SARS-CoV does not produce unique cytopathic changes and although suspicious for direct viral effect, the changes seen in formalin-fixed paraffin-embedded (FFPE) tissues are within the spectrum of those seen in end-stage ARDS-DAD cases initiated by various pathogens.² These present studies attempt to provide insight into the host response and ensuing pathogenesis associated with SARS-CoV by using FFPE pulmonary autopsy tissues from Singapore patients diagnosed with assorted microbial infections, including SARS-CoV. Because of the necessarily limited sample size of our study, we chose to focus our efforts on a subset of genes whose roles have been implicated in clinical ARDS and experimental pulmonary stress studies rather than trying to use microarray analysis as a hypothesis- generating discovery tool. Although it is feasible that host response to a pulmonary infection could be traced to the original etiologic agent, current clinical techniques cannot always establish this correlation. Therefore, we sought to determine molecular signatures of ARDS lung disease and if it was possible to distinguish ARDS cases initiated by SARS-CoV

MATERIALS AND METHODS

Autopsy samples were sent in consultation to the Armed Forces Institute of Pathology (AFIP) from Tan Tock Seng Hospital and Health Science Authority Singapore. A brief overview of the clinical and pathologic features of the FFPE lung tissues is shown in Table 1, as previously reported.^3,4 RNA was extracted from FFPE tissues at the AFIP as previously described.⁶ Total RNA isolated was double amplified using the RiboAmp RNA Amplification kit (Arcturus, Mountain View, CA), and quality of RNA was evaluated using the Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA). Microarray format and protocols are described on our website: expression.microslu.washington.edu. The experimental design for microarray analyses involved experimental samples being cohybridized with a reference sample to a human cDNA array (13,026 unique cDNA clones). Experimental samples were individual patient samples representative of various ARDS cases. The reference sample was created by pooling equal quantities of amplified RNA from the pulmonary experimental samples.

Table 1.

Clinical Information and Pathologic Features of FFPE Tissues^a

Case	Age/sex	Respiratory illness duration	Clinical diagnosis	Autopsy findings	Microbiology results	Details of DAD
−ARDS-1	1.3/F		Viral pneumonitis, status asthmaticus	Severe necrotizing tracheobronchitis	Adenovirus	Necrotizing
−ARDS-2	87/M		Nosocomial pneumonia (acquired after aortofemoral bypass surgery), sepsis, heart failure, ischemic heart disease	Severe necrotizing acute bronchopneumonia, iliac artery aneurysm, hypertensive nephrosclerosis	Pseudomonas aeruginosa, Candida tropicalis, Candida glabrata	Lobar pneumonia rather than DAD
−ARDS-3	47/M		Multiorgan failure, septic shock with disseminated intravascular coagulation	Acute necrotizing bronchopneumonia, fibrin thrombi within glomeruli	Streptococcus pneumoniae	Focal acute
+ARDS-4	67/F	4 days	Fever, brought in dead	Airspace edema and bronchiolar fibrin	SARS-CoV, alpha-hemolytic Streptococcus sp. and Klebsiella sp.	Acute-phase moderate with moderate parenchymal involvement
+ARDS-5	63/M	6 days	Multiorgan failure, ischemic heart disease, congestive heart failure, hypertension	Marked hyperplasia, mild acute bronchopneumonia, presence of multinucleated giant cells, coronary vessels: 10% occlusion, and nephrosclerosis	SARS-CoV, Mucor sp. and Aspergillus sp.	Marked acute-phase and mild organizing with high parenchymal involvement

^aPathologic features of pulmonary tissues were examined by histopathology as an initial attempt to classify these features in accordance with patient clinical information and possibly etiologic agent. The numbers of macrophages and neutrophils seen within alveolar spaces of ARDS initiated by SARS-CoV samples did not appear to be significantly different from that typically seen with other ARDS samples with organizing DAD. Briefly, histologic features of ARDS-4 were similar to those of other SARS cases of shorter duration (10 or fewer days), whereas ARDS-5 was also of shorter duration, but histologic features were similar to those of other SARS cases of longer duration (more than 10 days). Preinfection, ARDS-5 had congestive heart failure (CHF), which may account for the observed histologic features more typical of cases of longer duration.

Each single microarray experiment incorporated reverse dyelabeling techniques and resulted in four measurements for each gene, allowing calculation of a mean ratio among expression levels, standard deviations (SD), and p values within the Rosetta Resolver System (Rosetta Biosoftware, Seattle, WA). Genes were selected as differentially expressed based on two criteria, p ≤ 0.05 and a fold change ≥2. Furthermore, these parameters needed to hold true for two of the five experiments. Two was selected because we believed it would allow for similarity between the two SARS-positive cases or similarity between the two preexisting pulmonary complications (asthma or congestive heart failure [CHF]) to be studied. Two one-way analyses of variance (ANOVA) were performed. In the first ANOVA analysis, ARDS initiated by SARS-CoV was defined as one population, and ARDS initiated by other pathogens was defined as another population. In an alternative ANOVA analysis, ARDS with preexisting pulmonary complications was defined as one population, and ARDS without preexisting pulmonary complications was defined as another population. A stringent ANOVA p value cutoff of <0.0001 was used to select for genes that best discern each of the two groups.

Data normalization, error models, and all other calculations for our slide format are described at expression.microslu.washington.edu. This website is also used to publish all primary data in accordance with the proposed standards.⁷ Biosets were made through mining Protein Lounge pathways (www.proteinlounge.com) in combination with genes of interest from various literature sources.^8–10

RESULTS AND DISCUSSION

Although it is feasible that systemic host responses to pulmonary infection could be traced to the original etiologic agent, current clinical techniques and biologic assays cannot always establish this correlation, especially later in the course of infection if end-stage ARDS-DAD results, as a final common pathway to lung injury. Therefore, transcription profiling of pulmonary tissues was used to determine signatures of lung disease progression and whether SARS-CoV-induced ARDS (+ARDS-4 and +ARDS-5) could be distinguished from other ARDS cases (−ARDS-1, −ARDS-2, and −ARDS-3). FFPE tissues are amenable to preserving diagnostic qualities, yet the fixation process has previously limited their use in molecular biology applications because of modifications of RNA by formalin. Although these fixed tissues are not considered optimal for obtaining pristine RNA, methods are available for recovery of amplifiable RNA from these sources.⁶ Characterization of extracted total RNA and processed amplified RNA from these pulmonary clinical samples, as well as comparable control tissue sources (nonhuman primate; data not shown), showed a range of total RNA fragmentation sizes (100–2000 bp) but similar average amplified RNA fragmentation sizes (~750 bp), suggesting that sufficient concentrations of intact total RNA were present to attempt gene expression studies (Fig. 1).

FIG. 1.

The quality of total RNA extracted from tissues and amplified RNA was characterized using capillary electrophoresis. The ladder is observed as six distinct bands with sizes of 0.2, 0.5, 1.0, 2.0, 4.0, and 6.0 kb. (A) Comparison of total RNA extracted from FFPE pulmonary autopsy tissues (ARDS-1–5) and a typical fresh sample. (B) Comparison of amplified RNA generated using total RNA from those in A. The generation of more abundant shorter fragments of aRNA (as seen by the shoulder in the curve of the aRNA plot) did not directly correlate with the use of total RNA, with more abundance of shorter fragments.

As a control, in order to investigate the variability of gene expression due to chemical processing, a direct microarray comparison was performed using FFPE tissue obtained from one nonhuman primate as the experimental sample with matched fresh tissue obtained from the same nonhuman primate as the reference sample. It should be noted that these matched tissues came from a pulmonary sample that had been halved and then fixed in two different manners. A portion was placed in RNAlater (fresh) and a portion was placed in formalin (fixed). The formalin tissue was then paraffin embedded using standard techniques. The idea of doing a direct microarray comparison is grounded in the idea that if the mRNA populations and gene expression were identical in both the tissue samples, regardless of tissue fixation method used, zero genes would be detected by microarray that would be identified as differentially expressed. Contrarily, any gene that would be identified as differentially expressed by microarray could be further defined as variable between tissue samples due to tissue fixation methods (that chemically modify RNA) or tissue slice position in original stock tissue sample. This direct microarray comparison of matched primate FFPE tissues and fresh tissues showed ~3% of the genes were differentially expressed (Fig. 2). This demonstrates that even though there is some variability in the transcription profile elucidated from tissues that have undergone extensive chemical processing, ~97% of the information gained from the FFPE tissues concurs with that obtained from fresh tissues. As a reference, direct microarray comparisons of adjacent slices of FFPE tissue, as well as comparisons of adjacent slices of fresh tissue, showed ~0.1% of the genes to be differentially expressed, indicating that by gene expression characterization, these tissues really can be defined as same.

FIG. 2.

Genes that were differentially expressed when comparing matched FFPE with fresh tissue. Macaque tissues were excised and preserved in RNA-later or FFPE and then processed to extract total RNA. Each of the total RNA samples could be amplified and used to generate fluorescently labeled probes for microarray hybridization. The microarray experiment was a direct comparison of the FFPE tissue with the fresh tissue, incorporating reverse dye-labeling techniques and resulting in four measurements for each gene, as described in Materials and Methods. Doing a direct microarray comparison is grounded in the idea that if the mRNA populations and gene expression were identical in both the tissue samples, regardless of tissue fixation method used, zero genes would be detected by microarray that would be identified as differentially expressed. Selecting for a fold change >2 and p ≤ 0.05, 64 genes were shown to be differentially expressed from a total of 2171 unique genes to give ~3% differentially expressed genes. Signature genes (64 genes) are defined as those genes being upregulated (19 genes) and those genes being downregulated (45 genes). Downregulated refers to genes showing an increase in gene expression compared with the reference sample, and upregulated refers to genes showing a decrease in gene expression compared with the reference sample. This demonstrates that even though there is some variability in the transcription profile elucidated from tissues that have undergone extensive chemical processing, ~97% of the information gained from the FFPE tissues concurs with that obtained from fresh tissues.

Subsequent microarray analyses of the clinical samples, direct comparisons of each experimental sample to the reference sample, showed that each of the ARDS experimental samples exhibited similar numbers of differentially expressed genes, suggesting that the manner in which the ARDS lung tissues had been chemically processed could be considered grossly similar to one another (Color Plate 1A). This experimental design would, thus, highlight differences in pathogenesis among the various infectious agents and host responses, stratifying disease states, rather than differences between disease and health. This was to alleviate any concern that the use of autopsy samples at end-stage of disease would focus on attributes of autopsy tissue. In addition, although a reference source of a non-ARDS donor would have been more conventional and allowed us to investigate all aspects of the host response during ARDS, the use of a pooled ARDS reference allowed us to investigate whether pathogen-specific host responses could be elucidated, more specifically a unique molecular signature of SARS-induced ARDS. The microarray data are plotted as a heat map, a grid of colored points where each matrix entry represents a gene expression value, with red corresponding to high expression, green corresponding to low expression, and black corresponding to an intermediate level of expression. Hierarchical clustering methods are used to order rows (genes) and columns (samples) and to allow one to identify groups of samples that have similar expression level patterns and genes that are similar across samples.^11,12 Global analysis showed that −ARDS-1 and +ARDS-5 clustered together (Color Plate 1B). Because adenovirus was detected in −ARDS-1 and SARS-CoV and Mucor sp. and Aspergillus sp. were detected in +ARDS-5, clustering was not due to the initiating infectious agent. Curiously, however, it can be noted that −ARDS-2, −ARDS-3, and +ARDS-4 each tested positive for a bacterial infection, suggesting a potential gene expression pattern associated with infections containing a bacterial component.

COLOR PLATE 1.

Global expression profile of all genes. Each microarray experiment featured one of the five individual samples cohybridized with a portion of the pooled reference sample. This experimental design would, thus, highlight differences in pathogenesis among the various infectious agents and host responses, stratifying disease states, rather than differences between disease and health. This was to alleviate any concern that the use of autopsy samples at end-stage of disease, would focus on attributes of autopsy tissue. (A) Numerical summary of microarray data. The number of genes upregulated or downregulated with fold-change ≥2 and p ≤ 0.05 is indicated by n1(n2), with n1 being % of signature genes differentially expressed and with n2 being the % of total number of genes on the microarray differentially expressed. (B) Comparison of differentially expressed genes from five separate experiments using hierarchical clustering of an agglomerative algorithm, complete link heuristic criteria, and euclidean distance metric. Again, we used fold change ≥2 and p ≤ 0.05 parameters in two of five experiments, as discussed in Materials and Methods. The causative agent is listed, and CHF represents congestive heart failure. Genes shown in red were upregulated, and genes shown in green were downregulated in each individual patient sample compared with a pooled reference sample.

Because our sample set contains multiple etiologies related to ARDS, with only a single specimen from each etiology combination, we chose to focus our efforts on a subset of genes whose roles have been implicated in clinical ARDS and experimental pulmonary stress studies rather than trying to use the microarrays in this analysis as a hypothesis-generating discovery tool. First, we focused on cytokines, as cytokine dysregulation has been correlated with a hyperinflammatory response and disease severity.^8,13,14 Expression data used to generate Color Plate 1 were filtered by using an interferon (IFN) and cytokine bioset, and a heat map was generated using this subset of genes to determine similarities. When analyzing expression patterns associated with cytokine signaling,¹⁰ −ARDS-1 and +ARDS-5 lung tissues showed similar transcription patterns within notable genes included in the IFN pathway, such as Stat1, IFN regulatory factor-1 (IRF-1), interleukin- 6 (IL-6), and IL-18 (Color Plate 3).

COLOR PLATE 3.

Expression profile of genes involved in IFN and cytokine signaling. Comparison of differentially expressed genes from five separate experiments using hierarchical clustering (agglomerative algorithm, complete link heuristic criteria, and euclidean distance metric) was further focused by using an IFN and cytokine bioset to define a subset of genes. Again, we used fold change ≥2 and p ≤ 0.05 parameters in two of five experiments, as discussed in Materials and Methods. To determine additional nuances, ANOVA was used to compare two groups defined a priori. In the first ANOVA analysis, ARDS initiated by SARS-CoV was defined as one population, whereas ARDS initiated by other pathogens was defined as another population. In an alternative ANOVA analysis, ARDS with preexisting pulmonary complications was defined as one population, and ARDS without preexisting pulmonary complications was defined as another population. Using this scenario, this factorial-based analysis selected genes that are different between the two groups. Using these criteria, there were no genes attributed to SARS (absent yellow bar), but there were a number of genes attributed to preexisting pulmonary complications (blue bar).

Stat1 was shown to have increased gene expression in both −ARDS-1 and +ARDS-5. IL-6 and IL-18 both showed a decrease in gene expression in ARDS 2–4, −ARDS-1 showed an increase in regulation, and +ARDS-5 showed a very slight trend of increased regulation. Decreased levels of expression of IRF-1 were also demonstrated in ARDS-2–4, whereas +ARDS-5 showed increased levels of IRF-1 expression. Although −ARDS-1 showed a slight trend of decreased regulation, IRF regulation was still considered more similar to +ARDS-5 than to ARDS 2–4. IL-6 and IL-8, both proinflammatory cytokines, as well as IL-18 and activating factors (ATF) have been observed to be elevated in the plasma of SARS patients^14–17 as well as previously described ARDS patients.^18,19 Increased levels of expression of IL-8 and ATF3 were observed in both −ARDS-1 and +ARDS-5, with decreased levels observed in the other ARDS cases. Whereas previous studies have focused on detection of cytokines in plasma, it is interesting to note that patients with preexisting pulmonary complications (−ARDS-1, asthma and +ARDS-5, CHF) show the same trend at the transcriptional level in lung tissue. Another hallmark of proinflammatory cytokines and fibrotic mediator, IL-1α, is suppressed in this group but is elevated in −ARDS-2 and +ARDS-4 and somewhat elevated in −ARDS-3.

Two ANOVA analyses were next performed with groups defined either as ARDS initiated by SARS-CoV or ARDS initiated by other pathogens, in an attempt to tease out genes attributed to a direct SARS-CoV effect, or ARDS with preexisting pulmonary complications or ARDS without preexisting pulmonary complications, in an attempt to further define gene expression affected by an underlying condition. Using these analyses, many of the cytokines that make up the heat map in Color Plate 3 (36 of 61) could also be defined as genes that discern between ± preexisting pulmonary conditions (asthma or CHF); this is indicated by the blue bar. This suggests that gene expression of cytokines is affected by preexisting pulmonary conditions, such as asthma or CHF, before respiratory infection. Although ANOVA was used in an attempt to classify a set a genes attributed to a direct SARS effect, none were shown to have enough similarity between +ARDS-4 and +ARDS-5 to do so.

When singling out genes differentially expressed during pulmonary damage and repair, reportedly indicative of prognosis,⁸ transcription patterns of lung tissues from −ARDS-1 and +ARDS-5 again showed similar molecular profiles (Color Plate 2). ANOVA analyses were performed as described. Using these analyses, many of the genes that make up the pulmonary damage and repair heat map in Color Plate 2 (24 of 51) could also be defined as genes that discern between ± preexisting pulmonary conditions, as indicated by the blue bar. Again, this demonstrates that gene expression is affected by preexisting pulmonary conditions, such as asthma or CHF, before respiratory infection. Higher levels of extracellular matrix (ECM) components induced during lung injury and fibrosis, such as collagen (COL3A1 and COL1A1), fibrillin (FBN1), and keratin (KRT5), were observed. Enzymes responsible for remodeling of matrices were shown to be variable, including plasminogen activator (PLAU), responsible for fibrin binding and degradation of ECMs, and matrix metallopeptidase-2 (MMP- 2), both a gelatinase and a collagenase. Integrin β (ITGβ1), a fibronectin receptor responsible for fixating cells to the ECM and associated with activation of transforming growth factor (TGF), was shown to be differentially regulated, and TGF-β2 was included in downregulation as well. In addition, genes related to pulmonary stress were induced, including acute-phase proteins (FTH1), airway inflammation (DAP3), and antioxidants (AOC3, SOD1, and SOD2).

COLOR PLATE 2.

Expression profile of genes involved in pulmonary stresses. A subset of genes was investigated using a pulmonary stress bioset and selecting for fold change ≥2 and p ≤ 0.05 parameters in two of five experiments, as discussed in Materials and Methods. Again, to determine additional nuances, ANOVA was used as Color Plate 3. There were a number of genes attributed to ARDS cases having preexisting pulmonary complications (blue bar) and a few genes attributed to a SARS effect (yellow bar).

Again, most studies to date investigating components in pulmonary matrix remodeling survey bronchial lavage (BAL) fluids to detect soluble^8,20 mediators, so it is interesting to compare and contrast our results using lung tissues from clinical ARDS patients. Key events taking place at sites of matrix remodeling may not necessarily be reflected by soluble markers in BAL fluid, and our results from lung tissues prove to be more relevant. In addition, a few of the genes that make up the pulmonary damage and repair heat map in Color Plate 2 (6 of 51) were selected as genes that discern between ± SARS effect, as indicated by the yellow bar, most prominently MMP-8 and selectin P ligand (SELPLG), which were observed to have a decrease in gene expression in +ARDS-4 and +ARDS-5 compared with the other ARDS samples. As illustrated in Color Plate 2, MMP-8 and SELPLG tend to discern the two ANOVA groups (± SARS), whereas phospholipase A2, group IIA (PLA2G2A), immunoglobulin kappa constant (IGKC), immunoglobulin lambda constant 2 (IGLK2), and insulin-like growth factor 2 receptor (IGF2R) do not appear exclusive to one group. This is due to the nature of ANOVA analysis, and these represent genes that can be followed up with larger samples sizes to better understand how they segregate. With the biasing of our selection criteria, a few unique markers of lung damage and repair potentially caused by a direct SARS-CoV effect could be considered however overall SARS resembled ARDS.

Collectively, these studies show the feasibility of using FFPE samples for obtaining transcriptional profiling. Microarray molecular signatures obtained from pulmonary autopsy FFPE tissue samples correlated to what is currently known from studying soluble protein markers in either BAL fluids or sera samples from both ARDS and SARS patients. Even though more clinical samples in our study would have contributed to increased statistical significance, these small studies are not unprecedented.^21,22 The recurring clustering of −ARDS-1 and +ARDS5 hinted that pathogenesis may potentially be influenced by preexisting pulmonary complications. Although neither asthma (−ARDS-1) nor CHF (+ARDS-5) contribute directly to DAD, each disease has associated pulmonary components that could be considered a factor during subsequent infectious challenge and ensuing pathogenesis.^20,23–25 This parallels previous suggestions that patient outcome is related to the magnitude and duration of the host inflammatory response and independent of the precipitating cause of ARDS.¹³ Another point of interest is that +ARDS-4, having rapid disease progression (~4 days), showed a number of strongly upregulated cytokine signaling genes. These results suggest that if tissues were investigated earlier in disease progression, gene expression might reveal more insights into pathogenesis based on original pulmonary insult. This is similar to findings that early death (≤3 days) is related to the condition precipitating ARDS.¹³ In addition, −ARDS-2, −ARDS-3, and +ARDS-4 samples contained various pathogens, yet each had a bacterial component to their infection. Although it is outside the scope of this discussion to attempt to detail the host response to individual pathogen components or to use microarrays as a hypothesis-generating discovery tool (considering our sample size), it is believed that as samples are added to create a respiratory pathogen compendium, a more comprehensive view of ARDS progression at the genetic level would be able to supersede current characterization of ARDS at the clinical level.