NEW YORK — Researchers have generated a gene signature that detects host response to COVID-19, which they say could improve diagnostic tools and lead to earlier treatments.
In a pair of papers appearing Wednesday in the journal Cell Systems, researchers from Yale University and the Icahn School of Medicine at Mount Sinai developed a framework for generating transcriptional host response signatures that can robustly detect a pathogen while mitigating cross-reactivity with other, non-target infections. They then used the framework to guide their development of an 11-gene COVID-19 host response signature.
"As we show for COVID-19, the integration of the methods described in both studies helps provide the balance needed for having sensitivity of a diagnostic signature to detect infection while still limiting cross-reactivity with other infections and conditions," coauthor Elena Zaslavsky, a professor of neurology at Mount Sinai, said in a statement. "These new methods can help address the need for more rapid diagnosis of infections so that proper treatment can be initiated earlier."
In their first paper, the researchers searched through NCBI PubMed to identify 24 different previously published transcriptional signatures of host responses to a range of infections, including viral and bacterial.
At the same time, the researchers amassed a collection of datasets of host blood transcriptional responses to pathogen infection. This set included 136 transcriptional datasets of blood transcriptome response to viral, bacterial, parasitic, or fungal infections and another 14 datasets of transcriptomic data from older or obese individual, as age and obesity are associated with an altered immune system.
They developed a framework to systematically assess both the robustness and cross-reactivity of the infection signatures they found in the literature within these datasets.
With their geometric mean scoring approach, the researchers found that the signatures overall could detect the infections they were developed to spot from within independent data. The signatures could also detect asymptomatic and chronic infections, though with reduced performance.
However, many of the signatures had cross reactivity with other non-target infections as well as with age. An additional analysis focused on influenza infection signatures further highlighted a general trade-off between robustness and cross-reactivity.
Then, in their accompanying paper, the researchers used this framework to help generate a COVID-19 host response signature.
To do so, the researchers analyzed human blood transcriptome datasets of individuals with COVID-19 versus healthy controls and COVID-19 versus other pathogens. With a machine learning approach, they evaluated any proposed signature based upon its ability to detect COVID-19, its consistency with ATAC-seq and pathway data, and its cross-reactivity. From this, they selected an 11-gene signature that had consistently high detection of COVID-19 in training and development sets and low cross-reactivity with other viral, bacterial, and noninfectious conditions.
In a validation set of eight additional COVID-19 studies, this signature also had a high detection rate and low cross-reactivity. Further, the researchers compared its performance to those of four previously published gene signatures. While all the signatures had robust detection rates, only their new one exhibited minimal cross-reactivity, they found.
Using immune cell-type specific signatures, the researchers also traced back the cellular origins of their COVID-19 host response signature. The signature overlapped with those of plasmablasts and memory T cells, and further single-cell analysis suggested that the plasmablast-originating part of the signature aids in COVID-19 detection while the memory T cell-originating part of the signature limits cross-reactivity.
"The framework presented in these publications is available online, enabling members of the research community to benchmark the performance of their own signatures in our compendium of immunological conditions," co-first author Daniel Chawla from Yale said in a statement. "With these tools, we hope to enable the discovery of highly specific disease signatures."