A human in vitro model system for investigating genome-wide host responses to SARS coronavirus infection

Cell culture and virus infection

PBMCs were obtained by Ficoll-Hypaque separation of whole blood. 2 × 10⁵ PBMCs were seeded into each well of a 24-well culture plate, 0.5 ml of complete RPMI-1640 (Life Technologies-Invitrogen, USA) added to each and cultured overnight at 37°C (5% CO₂). A seed stock of SARS-CoV (strain SIN 2774) passaged in Vero E6 cells was used for infection. Vero E6 culture supernatants were added to each well in 50 μl volume at a concentration of 0.1 or 0.01 MOI (based on plaque forming units) and a control plate (media only). Each culture was set up in duplicate. After 4 hours of incubation, one set of the duplicate wells of the control plate and a 4 hours incubation plate were harvested while the rest received 0.5 ml media top-up and incubated for a further 2, 4, 6 and 8 days.

Cell harvest and RNA isolation

Harvesting was performed by gently flushing the wells with a Pasteur pipette to removed non-adherent cells, followed by a rinse of 1 ml RPMI. The rinsed fraction was pooled with the first harvest aliquot and spun at 1500 rpm. The cell pellet was washed twice with 2 ml RPMI to remove virus in the supernatant. 1.5 ml Trizol (Invitrogen, USA) was then added to the adherent cell fraction as well as the non-adherent cell fraction to lyse cells and stabilize the RNA. Extraction of total RNA was then performed following manufacturer's protocol and the resultant RNA dissolved in 40 μl water.

Real-time quantitative polymerase chain reaction (PCR)

The amount of SARS-CoV in each cell fraction was measured by real-time quantitative PCR assay. 2 μl of RNA was reverse transcribed and amplified in 20 μl using 0.9 μM each of forward (5'-GGTTGGGAT TATCCAAAATGTGA-3') and reverse (5'-AGAACAAGAGAGGCCATTATCCTAAG-3') primers, and 0.25 μM of TaqMan^® MGB probe: 5'-(6-FAM)AGAGCCATGCCTAACAT(NFQ)-3') in a one step PCR using master mix from Applied Biosystems (USA) according to manufacturers' recommendations. Reactions were performed using an ABI PRISM 7900 sequence detection system (48°C for 30 min, followed by 95°C for 10 min and 40 cycles of 95°C for 15 sec and 60°C for 1 min) and quantitation achieved using standard curves generated from in vitro transcribed RNA.

High density oligonucleotide array hybridization

At each time point (4 hours, 8 hours, 12 hours, 24 hours), 5 × 10⁷ cells of mock-infected and infected cells were harvested and lysed using Trizol (Invitrogen, USA). Total RNA was isolated according to the manufacturer's recommendation. Quality of the total RNA was judged from the ratio between 28S and 18S RNA after agarose gel electrophoresis. 20 μg of total RNA was labeled with Cy-3 or Cy-5 using the Superscript II reverse-transcription kit (Invitrogen, USA) and hybridization was carried out overnight (16 hours) at 42°C on high-density oligonucleotide arrays (~19,200 gene features, Compugen) using universal human reference (Stratagene, USA) as a reference. Hybridized arrays were scanned at 5 μm resolution on a GenePix 4000A scanner (Axon Instruments) with variable photo-multiplier tube voltage to obtain maximal signal intensities, and the resulting images were analyzed via GenePix Pro v4.0 (Axon Instruments) as described in the manual.

Microarray data analysis

Raw data were analyzed on GenePix analysis software version 4.0 (Axon Instruments) and uploaded to a relational database. The logarithmic expression ratio for a spot on each array was normalized by subtracting the median logarithmic ratio for the same array. Data were filtered to exclude spots with a size of less than 25 μm and any poor quality or missing spots. Since the correlation of the overall data from reciprocal labeling was good, values obtained from reciprocal labeling experiments were averaged. In addition, the data were distilled to the set of gene features that were present at all 4 time points in both the viral infected samples and the negative controls. The results were represented as the logarithmic ratio of gene expression between the viral infected samples and their corresponding negative controls at the various time points. Application of these filters resulted in the inclusion of ~12,900 of the total ~19,200 gene features in subsequent analyses.

To discover patterns of gene expression, the values associated with each gene feature f were translated so that their means were zero. Similar genes, whose translated gene-features exhibited same induction-repression pattern, were grouped together. Genes g _i , g _j were said to be similar if they satisfied the following condition:

, where U(x) = 1 if x > 0, U (x) = 0 otherwise,

where g _it and g _jt denote the translated values of gene features g _i , g _j at time t respectively; and N is the number of time points for which the expression of a gene was observed. Similar genes, based on the above criteria, are grouped together. Within each group, the genes were ordered in the descending order of their expression range (defined as the difference between the maximum and minimum ratios of gene expression). This algorithm is a special case of the Friendly Neighbor algorithm currently under development. The final plots were generated using the original expression ratios while preserving the clustering and ordering discovered by the above algorithm. To determine whether a gene observed to be responsive could appear merely by chance, 100,000 expression profiles were generated by randomly sampling the expression ratios from the entire dataset with replacement. The P value of a gene is the fraction of the random profiles whose logarithmic expression range is as good as, or better than that of the selected gene.

Kinetics of SARS-CoV replication

We obtained PBMCs from 6 healthy volunteers by Ficoll-Hypaque separation of whole blood. Of the 6 donor PBMCs tested, all were able to support SARS-CoV replication when infected with multiplicity of infection (MOI) of 0.1. The first sampling, taken from cells infected for 4 hours, showed an average copy number of 32 × 10³ (Fig. 1A) and represents the initial inoculum level. Over the course of the next 8 days, there was a steady rise in viral load, reaching as high as 480 × 10³ copies per well in one donor, which could only be explained by active replication of the SARS-CoV intracellularly. This work is supported by recent in vivo evidence suggesting that SARS-CoV may have infected and replicated within PBMCs of SARS patients [15] and cells from humans and animals [16]. An indication of the PBMCs lineage involvement was provided by repeating the experiment using the monocyte-macrophage cell line THP-1 [17], in which viral replication was similar to the primary cell culture over the first 4 days (Fig. 1B). In the primary cultures, the non-adherent cell fraction which comprises mainly lymphocytes and granulocytes showed dramatically less viral replication in our assay as did all cells infected at MOI of 0.01 (Fig. 1A).

The kinetics of viral replication was variable among the 6 donors (Fig. 1C). There was a lag phase of 2 days in the case of donors a, c and e; and 4 days for donors b, d and f before any significant increase could be detected. The viral replication generally peaked at either day 4 or day 6. The exception was donor b, in which the virus seemed to replicate at a much slower pace compared to the other 5 donor samples. Equally interesting was the different levels of virus attained. Donor d seems to stand out from the rest, reaching a peak of 480 × 10³ copies per well which is 4 times more than that attained by donor e, with 120 × 10³ copies per well. Such variation strongly suggests that there is an underlying host-pathogen interaction influencing the kinetics of SARS-CoV replication efficiency. These in vitro observations may reflect the wide range of patient outcomes after SARS-CoV infection [18].

Antibody blocking experiments were also performed in which SARS-CoV was pre-incubated with convalescent patient sera for 30 minutes before introduction to the PBMCs and after a 4 day incubation period, the adherent cell fraction was harvested and assayed for SARS-CoV viral titer. Results clearly showed that even at high dilution, convalescent sera inhibited SARS viral replication (data not shown), presumably by blocking viral entry. This supports other reports indicating that SARS-CoV is not endocytosed through antibody mediated mechanisms and confirms a protective role for antibodies elicited either by the infection or through immunization [19, 20].

SARS-CoV specific gene expression changes

To further elucidate the molecular processes of SARS-CoV infection, PBMCs from 3 healthy individuals were infected separately in vitro with SARS-CoV (0.1 MOI) and harvested at 4 hours, 8 hours, 12 hours and 24 hours time intervals post-infection. As controls, uninfected aliquots of the same PBMCs were also harvested at the corresponding time points. Total RNA extracted from the PBMCs of the 3 individuals were pooled, labeled and hybridized to human oligonucleotide arrays consisting of ~19,200 gene features. Reciprocal dye swap replicate hybridizations were performed to minimize technical noise. Analysis of variance in expression levels for each gene across all the time points indicated the ~1200 genes which showed the largest variability (Fig. 2A and 2B).

A number of CC chemokines (CCL4, CCL20, CCL22, CCL25, CCL27) and their receptors (CCR4 and CCR7) were highly expressed in response to the infection (Fig. 2C and Table 1), indicating a rapid mobilization and increased trafficking, in particular of the monocyte-macrophage lineage very early on in the infection [26]. CXC chemokines (CXCL9, CXCL12) were also highly expressed suggesting significant increase in neutrophil homing as well. These are likely to be lung directed as IL8 and IL17 were also highly expressed [27–32]. Specific trafficking of these cells to the lung may account for the localized nature of the response [33].

Surprisingly, a number of blood coagulation genes were highly expressed early during our in vitro infection (Fig. 2C and Table 1), in particular TBXAS, which has been implicated in vasoconstriction, platelet aggregation, membrane lysis and increased permeability [34, 35]; fibrin (FGB and FGG) and the coagulation pathway directly (SERPINs D1 and A3 together with Factors 10, 3 and 2). This gives a pro-coagulation profile, which mimics the clinical-pathological observations: at autopsy, many SARS patients have unusually disseminated small vessel thromboses in the lungs without evidence of disseminated intravascular coagulation [1, 36]. Again, these expression profiles provide an experimental framework to explore an important aspect of SARS pathobiology and treatment.

It is interesting to note that the TLR9 was highly expressed in comparison to other TLR receptors, implying some degree of TLR specificity for the virus (Fig. 3A). TLR9 is known to respond to CpG signaling motifs (GTCGTT) [37–39] and one possibility is that the virus is activating directly through this mechanism. In support of this, we found that the SARS-CoV viral sequence contains the highest number (7 copies) of such specific signaling motifs compared to other coronaviruses and significantly more than several other viruses involved in respiratory diseases (Fig. 3B). It is conceivable that TLR9 may be aiding host recognition of the virus via the CpG groups and contributing to the initiation of the innate host inflammatory response. An alternative explanation is that TLR9 is being stimulated by mechanisms unrelated to CpG recognition.

The emerging picture from this study implicates a central role for the immune response in the pathobiology of a SARS infection. While detailed in vivo studies of the host response are now required, the in vitro model described here will allow responses to specific modulators (such as therapeutics) to be investigated. In future developments of the model, it will be interesting to compare the host response to different SARS-CoV isolates with inactivated preparations of the virus. In other diseases, in vitro models have revealed a number of key processes relevant to the clinical diseases [9, 12, 13] and it is likely that the responses identified here will prove to be equally important. Although some clinical parameters have now been used as prognostic markers [40–42], further study of the regulatory mechanisms for chemokine-cytokine production will likely improve their accuracy and perhaps allow development of new treatment protocols.