Utility of DNA Microarrays for Detection of Viruses in Acute Respiratory Tract Infections in Children

Study Design

This study is a prospective case series of all consecutive samples sent for viral DFA testing from pediatric patients seen at UCSF between December 2003 and June 2004. All patient samples were collected according to protocols approved by UCSF's Institutional Review Board.

Sample Processing

Consecutive nasopharyngeal aspirate (NPA) samples sent to the UCSF clinical laboratory for DFA were analyzed with the Light Diagnostics Respiratory DFA Viral Screening and Identification Kit (Chemicon International, Temecula, CA). This kit detects 7 common respiratory viruses: respiratory syncytial virus (RSV); influenza A and B (FluA and FluB); human parainfluenza virus types 1, 2, and 3 (HPIV-1, 2, and 3); and adenovirus (AdV). Remaining sample material was transferred into a sterile 14-mL conical tube, frozen, and stored at −80°C.

For blinded Virochip analysis, frozen NPA samples were thawed, and 200-μL aliquots were used to extract RNA using the RNeasy Mini Kit (Qiagen Inc, Valencia, California) according to the manufacturer's protocol, including on-column DNase digestion. Microarrays used in this study were identical to those described previously (National Center for Biotechnology Information GEO platform GPL3429).¹⁶ RNA samples were amplified and labeled using a round A/B random PCR method and hybridized to the Virochip as reported previously.¹³ Microarrays were scanned with an Axon 4000B scanner and analyzed using Axon GenePix 6 software (Axon Instruments, Union City, California). All microarray data have been submitted to the National Center for Biotechnology Information GEO database (accession number GSE10294).

Virochip Data Analysis

Virochip data analysis was carried out in 2 stages. First, all microarrays were analyzed with the E-Predict algorithm,¹⁷ using a significance value of P < .01 to identify microarrays with statistically significant viral hybridization patterns (176 of 278). To generate an optimized set of oligonucleotide intensity weights for the third-generation Virochip platform, we took the remaining 102 presumably negative microarrays and manually fit a set of functions with the general equation

where w is a weight (value from 0 to 1) for a given oligonucleotide, i is the median of sum-normalized intensities of that oligonucleotide across the 102 negative microarrays, and lower boundary a is the median of medians of the sum-normalized intensities of all oligonucleotides. The upper boundary condition b was expressed as $b=a+c\sigma$, where c is a constant and σ is the standard deviation of sum-normalized intensities of the oligonucleotide across the 102 negative microarrays. A total of 40 weight sets with c and p as fitting measurements were evaluated using E-Predict profile separation statistics on 6 rhinovirus (RV)-positive, 6 RSV-positive, 4 HPIV-3–positive, and 2 FluA-positive microarrays. The optimal performance was achieved with weights corresponding to c = 0.15 and p = 1.5. These weights were used to generate negative null distributions based on the set of 102 negative microarrays mentioned above, and final microarray virus determinations were made by E-Predict. A microarray was considered E-predict–positive for a given virus if the corresponding energy profile attained a significance value of P < .05.

PCR Analysis

Blinded reverse-transcriptase (RT)-PCR assays for RSV, FluA, RV, and enterovirus (EV) were performed on RNA extracted from the frozen NPA samples at the Viral and Rickettsial Disease Laboratory, California Department of Health Services. For detection of RSV, an RT-PCR assay for RSV targeting the fusion (F) gene was performed using the 1-step Access RT-PCR system as described previously.² For detection of FluA, primers and fluorescent probes targeting the highly conserved matrix (M) gene of FluA were used with the Roche LightCycler Real-Time PCR System (Roche Diagnostics, Indianapolis, Indiana) as described previously.¹⁸ For detection of RV/EV, we first ran an in-house RT-PCR multiplex assay on randomly primed template cDNA using primers targeting the highly conserved 5′ untranslated region of RV¹⁵ and EV.¹⁹ Positive samples were identified by the presence of amplified PCR bands of the expected size on agarose gel electrophoresis. Follow-up real-time PCR assays for RV/EV detection on discrepant samples between the Virochip and the in-house multiplex RT-PCR were then carried out using the Roche LightCycler Real-Time PCR System as described previously.²⁰

Sequence Confirmation of Virochip-Positive/PCR-Negative Samples

Confirmation of 2 Virochip-positive/PCR-negative samples was done by PCR using alternative primers based on high-intensity microarray oligonucleotides and direct sequencing. The sequence from 1 case of RSV was 97% identical to a 155-bp fragment from the matrix (M) gene for RSV subgroup B strain 9320, and the sequence from 1 case of RV was 96% identical to a 302-bp fragment from the VP4/VP2 region for RV strain QPID03-0003. Another Virochip-positive/PCR-negative case (an EV) was confirmed by a repeat run of the real-time PCR assay.²⁰

Clinical Data Collection

After assay results for the NPA samples were obtained, subjects were identified, and a retrospective review of the medical record was performed. Data were systematically collected using a standard form that documented the following information: age, sex, date and location of sample collection, clinical presentation, presence of immunocompromised state, presence of acute respiratory failure, and DFA result. Location of sample collection was classified as outpatient (clinic or emergency department) or hospital admission. Presenting illness was defined as an upper RTI (ie, cough and/or congestion with or without fever or a clinician's diagnosis), lower RTI (ie, clinician's diagnosis of asthma exacerbation, bronchiolitis, croup, or pneumonia), or no respiratory illness (eg, febrile illness, seizures, or DFA collected on a routine basis, such as before transplantation surgery). Immunocompromised patients were defined as children with solid organ or bone marrow transplants, congenital or acquired immune deficiencies (including those on chemotherapy), or human immunodeficiency virus. Patients with respiratory failure were defined as children who developed acute respiratory decompensation necessitating mechanical ventilation as a result of their respiratory illness. For patients with more than 1 sample, a first-time sample was defined as the earliest sample collected during a single hospitalization or illness.

Demographic and Clinical Data

We analyzed a total of 278 NPA samples collected from 222 children for the presence of viruses using DFA and Virochip. Demographic and clinical data for the 222 patients are given in Table I. Most patients (73%) were hospitalized. Approximately 71% of the enrolled patients (n = 157) had an acute RTI, and the remaining patients (n = 65) had DFA sent for other reasons, most often a nonrespiratory febrile illness (n = 46). Eighteen percent of the patients (n = 39) were immunocompromised, and 8% (n = 17) developed respiratory failure necessitating mechanical ventilation as a result of their illness. Most of the subjects with RSV were younger (mean age, 1.5 years) and required hospitalization (76%), whereas subjects with influenza tended to be older (mean age, 4.1 years) and were treated mainly as outpatients (64%). All cases of human metapneumovirus (HMPV), coronavirus (CoV), and AdV were in hospitalized patients. The majority of viral infections in immunocompromised subjects (60%) were due to picornaviruses.

Detection of Viruses by DFA and the Virochip

Figure 1 shows the spectrum and frequency of detection of different viruses by the 2 methods. In patients with RTI, DFA detected a virus in only 36% of the samples, whereas the Virochip detected a virus in 64%, for a 75% improvement in the rate of detection. Detection of viruses not included in the DFA panel (non-DFA viruses) accounted for approximately 1/3 of all Virochip-positive identifications. Among these non-DFA viruses, picornaviruses was the largest group (74% of samples positive for a non-DFA virus, or 16% of all samples from patients with RTIs). These picornaviruses included 16 RVs, 14 EVs, and 1 parechovirus. Besides the picornaviruses, the Virochip identified 4 HMPV and 2 CoV infections from patients with RTIs. Human cytomegalovirus–like sequences were detected in 1 sample from 1 febrile patient with pneumonia. In 7 samples, the Virochip detected viruses not typically associated with respiratory disease and of doubtful clinical significance; these viruses included polyomaviruses (SV40 and JC viruses), plant viruses commonly found in the gastrointestinal tract (nanoviruses and geminiviruses),²¹ and bovine leukemia virus, a virus ubiquitous in cow's milk.²² A virus was detected in a higher proportion of patients with upper RTIs (68%) and bronchiolitis (77%) than in those with pneumonia (54%) or asthma exacerbation (41%). In patients with nonrespiratory illnesses (typically febrile illnesses of uncertain origin), the Virochip detected a virus in 30% of the samples, with the majority accounted for by picornaviruses, whereas DFA detected a virus in only 3%. Overall, at least 1 of the 7 DFA viruses was identified by Virochip in 31% of the RTI samples (n = 86), compared with 25% identified by DFA (n = 72), corresponding to a 19% overall increase in detection by the Virochip compared with DFA (P < .01 by χ² analysis).

Virochip Detection of Double and Critical Viral Infections

The Virochip detected 13 cases of simultaneous infection by 2 viruses (9% of Virochip-positive samples, 5% of all samples; Table II; available at www.jpeds.com). In contrast, DFA detected no cases of double infection. The Virochip also detected a viral pathogen in 12 of 17 critically ill patients who developed respiratory failure necessitating mechanical ventilation, whereas a virus was detected by DFA in only 5 such patients. Two double infections (1 case of FluA/AdV and 1 case of RSV/RV) were identified among these life-threatening cases.

Comparison of the Virochip with PCR

To assess Virochip results that were discordant with DFA and overall sensitivity and specificity, we carried out RT-PCR assays for RSV, FluA, and RV/EV, the 3 major groups of viruses detected by the Virochip (Figure 2). For detection of RSV and FluA, the overall sensitivity of the Virochip (86%) relative to PCR was significantly better than that of DFA (71%; P < .05 by χ² analysis). The corresponding specificities were ≥ 99%. There was also a greater overlap in positives between Virochip and PCR than between DFA and PCR for these 2 viruses; 16 of of 18 (89%) of Virochip-positive/DFA-negative RSV and FluA samples were confirmed to be true positives by PCR (Figure 2; Venn diagram).

For detection of RV/EV, we first used an in-house RT-PCR multiplex assay based on methods reported previously.15, 19 Using this assay, 31 NPA samples were positive by both Virochip and PCR, 18 NPA samples were Virochip-positive/PCR-negative, and 5 NPA samples were Virochip-negative/PCR-positive. Because these results were inconsistent with the rates of detection using clinically validated PCR assays for the other viruses (RSV and FluA), we further analyzed the 23 discrepant samples in a blinded fashion using a more sensitive clinically validated real-time PCR assay for RV/EV.²⁰ Results obtained by combining these 2 assays show that the Virochip had overall high sensitivity (90%) and specificity (99%) for detection of RV/EV (Figure 2C).

Virochip Detection of PCR-Negative Cases of Respiratory Virus Infection

Six cases (3 picornavirus, 2 RSV, and 1 FluA) detected by the Virochip tested negative in the corresponding PCR assays (Figure 2). We hypothesized that most of these cases were viral strains that had failed detection with standard PCR primers. To investigate this possibility, we recovered viral sequences from 2 Virochip-positive/PCR-negative cases (1 RV and 1 RSV) and separately confirmed a case of EV that was Virochip-positive and previously PCR-negative by repeating the real-time PCR assay. Similar attempts to confirm the remaining 3 Virochip-positive/PCR-negative cases with the limited amount of sample available were unsuccessful. Thus, at least 3 of the 6 Virochip-positive/PCR-negative represent PCR false-negatives, and, consequently, the reported sensitivity of the Virochip may be underestimated.