A previous version of the story incorrectly stated that the researchers received a patent for their classifier in 2017. They filed an application in 2017 but have not yet received a patent. We regret the error.
NEW YORK (GenomeWeb) – A group led by Mount Sinai researchers has used nasal brush collection and RNA sequencing (RNA-seq) to identify a genetic classifier for asthma. With further validation studies, the team believes its tool could lead to the development of a minimally invasive biomarker test that would cost only around $30.
In order to detect cases of asthma, clinicians usually perform pulmonary function tests (PFTs) using equipment to measure airflow and lung size, producing results within an hour. Patients also self-report past symptoms of asthma, whether mild or severe, during an office visit. However, PFTs are not always available in primary care settings, and they can produce unreliable results if measured poorly by physicians.
In a study published earlier this week in Scientific Reports, the Mt. Sinai-led team used RNA-seq to profile gene expression from nasal epithelial cells collected from 190 subjects with mild to moderate asthma, as well as 140 controls. Using a machine learning-based method, they developed an asthma classifier that was able to differentiate patients with and without mild-moderate asthma.
Senior author and associate professor at the Icahn School of Medicine Supinda Bunyavanich said that she and her colleagues used a nasal brush to swab patient airways containing nasal epithelial cells because of their ease of access and connection to the lungs.
"Given that the united airway and the nasal passages are directly connected to the lower airway ... it'd be great if we could take a nasal brush and figure out if someone has asthma based on [those samples]," Bunyavanich said.
After sample collection, the researchers randomly assigned 150 subjects with asthma as the development set — to be used for asthma classifier development — and used the remaining 40 subjects to evaluate the method's classification performance on additional datasets of independent subjects with asthma and other respiratory conditions.
Isolating RNA from the groups, the team then performed RNA-seq on the samples to find potential biomarkers. As part of a machine learning "pipeline," the team developed two algorithms aimed at narrowing down a set of genes that could be used in a classifier, resulting in a set of 90 genes. The researchers also developed a classifier algorithm and calculated the threshold needed to interpret the data.
After developing the asthma classifier, the team tested the tool on eight different patient groups: a dataset of people with or without asthma profiled by RNA-seq; two datasets profiled by microarrays and recruited by outside research groups; and five datasets comparing asthma to other respiratory conditions.
In order to assess the asthma classifier in other populations and using different profiling platforms, the team applied it to nasal gene expression data from groups of asthmatics and controls — Asthma 1 and Asthma 2 — profiled by microarrays. Despite discordances in study designs, sample collections, and gene expression profiling platforms, the researchers found that the asthma classifier performed "relatively well" and better than permutation-based random models.
The nasal brush-based classifier had a clinical sensitivity of 92 percent and specificity of 100 percent when her team performed the test on the asthma dataset profiled by RNA-seq, according to Bunyavanich.
In order to further evaluate the classifier's specificity, the team examined its performance on five additional cohorts of respiratory conditions with symptoms that mimic asthma symptoms.
In three cohorts — allergic rhinitis, cystic fibrosis, and smoking — the classifier properly identified the samples as "not asthma." However, the classifier slightly deviated in the two upper respiratory infection cohorts — observed at day two and day six — which the researchers believe was the result of common inflammatory pathways that caused viral inflammation and asthma. Overall, the team noted that the classifier's results indicate that it could accurately differentiate patients with asthma from those with other respiratory issues.
"We were pleasantly surprised that [the classifier] had a low-to-zero misclassification rate for other conditions to asthma," Bunyavanich explained. "We had expected overlapping symptoms between the different conditions, that there might be more classification than we wanted."
Bunyavanich highlighted that the classifier demonstrated "excellent" performance based on interpreting all 90 genes. While researchers could profile a smaller number of genes, she argued that the performance of a smaller gene set could be compromised. The 90-gene classifier, she said, balances performance with a "reasonable number of genes."
"Ninety genes is a well-circumscribed number of genes to profile in an automated fashion, and doesn't add to the burden of what needs to be done at the point of care," Bunyavanich said. "While we used sequencing to identify the 90 genes and [develop] the algorithm, we can use targeted profiling with another platform to perform [the process] faster."
Bunyavanich explained that the time to result will depend on the third-party platform used by researchers. Sample collection and the algorithm steps can be performed in "a matter of seconds" she said.
There were limitations to the study, however. The team acknowledged that the RNA-seq development set was performed at a single laboratory, and that some characteristics of the patients, such as race, do not cover all populations.
In addition, the group's asthma classifier did not perform as well in the microarray-based dataset versus RNAseq-based asthma datasets. However, Bunyavanich argued that her team expected those results because of differences in the study design and technological factors between RNA-seq and microarray profiling.
First, the team noted, the subjects differed in physical characteristics and condition phenotypes. Subjects in the RNAseq test set contained adults who were classified as mild and moderate asthmatics or healthy controls. In contrast, patients in the Asthma 1 microarray test set consisted entirely of children with nasal pathology issues.
The team also believes that the difference in performance may be caused by the difference in profiling approaches. Bunyavanich noted that gene mappings do not perfectly agree between RNA-seq and microarray techniques because of discrepancies between array annotations and RNA-seq gene models.
According to Bunyavanich, her team has developed one of the largest nasal RNA-seq dataset for asthma and is one of the first groups to identify a nasal brush-based genetic classifier for asthma. The authors highlighted that the asthma classifier selects signals from differential expression, in addition to genes below traditional significance thresholds that might play a role in identifying asthma.
Bunyavanich and her team filed a patent for the classifier set of 90 genes, as well as the classifier algorithm, in 2017. According to Bunyavanich, her team is currently "in the early stages" of commercializing the classifier tool for clinical use. She aims to set a target price of around $30 dollars per test, depending on the platform used by the clinician. She also noted that the classifier pipeline could eventually be used to develop classifiers for other upper respiratory tract issues.
Bunyavanich envisions the test being used "at the clinical frontline," whether in the office or urgent care settings, where a clinician can perform nasal brushings. Depending on the third-party platform used, she noted that the sample could be sent to an external lab or examined at the doctor's office.
In the future, Bunyavanich's team will recruit additional cohorts to increase nasal gene expression profiling and the test's validation data. The researchers believe that the first major step after validation will be to accurately identify patients afflicted with asthma. In addition, the following phase of research will be to develop a nasal biomarker to predict condition subtypes and monitor treatment response for asthma.