A GMR-based assay for quantification of the human response to influenza

Abstract

Detecting and quantifying the host transcriptional response to influenza virus infection can serve as a real-time diagnostic tool for clinical management. We have employed the multiplexing capabilities of GMR sensors to develop a novel assay based on the influenza metasignature (IMS), which can classify influenza infection based on transcript levels. We show that the assay can reliably detect ten IMS transcripts and distinguish subjects with naturally acquired influenza infection from those with other symptomatic viral infections (AUC 0.93, 95% CI: 0.82-1.00). Separately, we validated that the gene IFI27, not included in the IMS panel, has very high single-biomarker accuracy (AUC 0.95, 95% CI: 0.90-0.99) in stratifying patients with influenza. We demonstrate that a portable GMR biosensor can be used as a tool to diagnose influenza infection by measuring the host response, simultaneously highlighting the power of immune system metrics and advancing the field of gene expression-based diagnostics.

Keywords: GMR sensors; Gene expression; Host response; Influenza; Point-of-care diagnostics.

Figure 1:

Workflow for detection of IMS genes on the GMR biosensor array. (a) PBMCs were isolated from healthy human donors. mRNA was isolated, reverse-transcribed to cDNA, and PCR-amplified with biotinylated primers for the IMS genes of interest. The PCR product was denatured at 95°C to create biotinylated ssDNA corresponding to each of the target IMS genes. (b) Sensors on the GMR chip are spotted with ssDNA probes complementary to each of the IMS genes seen in (a), along with a negative control and a positive control, according to the simplified spotting pattern seen. The target ssDNA is added to the sensor surface and allowed to hybridize. Signal is measured after adding streptavidin MNPs. To simplify, only four of the nine IMS genes are shown; for full spotting pattern, see Figure S1.

Figure 2:

GMR levels of IMS expression examined in LPS-treated and IFN-treated healthy-donor PBMCs, each pair of dots connected by a line represents one donor. Fold change was calculated relative to untreated control. (a) For each donor, IMS expression in IFN-treated PBMCs was compared to IMS expression in LPS-treated PBMCs using a paired t-test (p < 0.0001, n=11). (b) For each donor, expression levels of each individual IMS gene were compared between IFN-treated PBMCs and LPS-treated PBMCs using a paired t-test (p < 0.0001, n=11).

Figure 3:

(a) Subjects with flu-like symptoms were classified on the day of symptom onset (day 0) by RT-PCR on nasal wash fluid, as having influenza A or B, HRV, or negative (did not test positive for any of the viruses in the panel). Each dot represents one individual. Using a Mann-Whitney test, the IMS score on day 0 of infection was compared between influenza A or B-infected individuals, HRV-infected individuals (p < 0.0001, n=39), or negative individuals (p < 0.0001, n=41). (b) For influenza-infected individuals, the IMS score was compared between the baseline and day 0 timepoints using a Wilcoxon matched-pairs signed rank test (p < 0.0001, n=26).

Figure 4:

Subjects with flu-like symptoms were classified on the day of symptom onset (day 0) by RT-PCR on nasal wash fluid, as having influenza A or B, HRV, or negative (did not test positive for any of the viruses in the panel). Each dot represents one individual. Expression levels of each individual IMS gene on day 0 of infection was compared between influenza A or B-infected individuals, HRV-infected individuals (p < 0.0001, n=39), or negative individuals using a Mann-Whitney test (p < 0.0001, n=41).

Figure 5:

Logistic regression models were applied to determine the power of the IMS genes in classifying influenza infection. (a) A univariate logistic regression model was applied using the IMS scores on day 0 (day of symptom onset); the AUC was 0.87 (95% CI: 0.79–0.96). (b) A multivariate logistic regression model was applied using the individual IMS gene scores on day 0, the test AUC was 0.93 (95% CI: 0.82–1.00). (c) The IFI27 gene was added to the panel, and a univariate logistic regression model was developed; the AUC was 0.95 (95% CI: 0.90–0.99). (d) A multivariate LASSO logistic regression model was applied using individual IMS genes in conjunction with IFI27, resulting in 4 genes of greater influence. Using just the expression levels of these four genes at day 0 of infection (LASSO regression score) the test set AUC was 0.93 (95% CI: 0.84–1.00).

Figure 6:

(a) IMS score (including IFI27) over time in subjects with flu-like symptoms (b) IFI27 expression (log GMR signal) over time in subjects with flu-like symptoms. (c) (d) Expression time course (log GMR signal) of individual genes in the IMS score, plus IFI27, in two subjects infected with influenza.