An ultra-early, transient interferon-associated innate immune response associates with protection from SARS-CoV-2 infection despite exposure

Abstract

Background: A proportion of individuals exposed to respiratory viruses avoid contracting detectable infection. We tested the hypothesis that early innate immune responses associate with resistance to detectable infection in close contacts of COVID-19 cases.

Methods: 48 recently-exposed household contacts of symptomatic COVID-19 cases were recruited in London, UK between May 2020 and March 2021 through a prospective, longitudinal observational study. Blood and nose and throat swabs were collected during the acute period of index case viral shedding and longitudinally thereafter. Magnitude of SARS-CoV-2 exposure was quantified, and serial PCR and serological assays used to determine infection status of contacts. Whole-blood RNA-seq was performed and analysed to identify transcriptomic signatures of early infection and resistance to infection.

Findings: 24 highly-exposed household contacts became PCR-positive and seropositive whilst 24 remained persistently PCR-negative and seronegative. A 96-gene transcriptomic signature of early SARS-CoV-2 infection was identified using RNA-seq of longitudinal blood samples from PCR-positive contacts. This signature was dominated by interferon-associated genes and expression correlated positively with viral load. Elevated expression of this 96-gene signature was also observed during exposure in 25% (6/24) of persistently PCR-negative, seronegative contacts. PCR-negative contacts with elevated signature expression had higher-magnitude SARS-CoV-2 exposure compared to those with low signature expression. We validated this signature in SARS-CoV-2-infected individuals in two independent cohorts. In naturally-exposed healthcare workers (HCWs) we found that 7/58 (12%) PCR-negative HCWs exhibited elevated signature expression. Comparing gene-signature expression in SARS-CoV-2 Controlled Human Infection Model (CHIM) volunteers pre- and post-inoculation, we observed that 14 signature genes were transiently upregulated as soon as 6 hr post-inoculation in PCR-negative volunteers, while in PCR-positive volunteers gene-signature upregulation did not occur until 3 days later.

Interpretation: Our interferon-associated signature of early SARS-CoV-2 infection characterises a subgroup of exposed, uninfected contacts in three independent cohorts who may have successfully aborted infection prior to induction of adaptive immunity. The earlier transient upregulation of signature genes in PCR-negative compared to PCR-positive CHIM volunteers suggests that ultra-early interferon-associated innate immune responses correlate with, and may contribute to, protection against SARS-CoV-2 infection.

Funding: This work was supported by the NIHR Health Protection Research Unit in Respiratory Infections, United Kingdom, NIHR Imperial College London, United Kingdom (Grant number: NIHR200927; AL) in partnership with the UK Health Security Agency and the NIHR Medical Research Council (MRC), United Kingdom (Grant number: MR/X004058/1). Support for sequencing was provided by the Imperial BRC Genomics Facility which is funded by the NIHR, United Kingdom. The development of the hybrid DABA assay used for quantification of SARS-CoV-2 anti-Spike RBD antibodies was supported by the MRC (MC_PC_19078).

Keywords: Household contacts; Innate immunology; Resistance to infection; SARS-CoV-2.

Fig. 1

Study cohort and sampling schedule. (a) 48 household contacts of COVID-19 cases and 14 unexposed controls were recruited. Serial blood and nose and throat swab samples were collected at study day 0 (D0), and days 7, 14, and 28 post enrolment (D7, D14, and D28 respectively). RNA-Seq and serology analysis was performed on blood samples. PCR was performed on nose and throat swabs. Samples from unexposed controls were collected at a single time point. (b) Extent of exposure to COVID-19 index cases in PCR-positive and PCR-negative contacts. Exposure score is a composite of relationship to index score, room sharing score, and index case viral load. Median and quartiles are displayed. p Value of Wilcoxon test is shown. ∗∗p ≤ 0.01. (c) Viral load trajectories of PCR-positive contacts. Threshold for PCR-positivity is indicated with a red dashed line.

Fig. 2

Identification of a transcriptomic signature of early SARS-CoV-2 infection. (a and b) Volcano plots showing differential gene expression between unexposed control (N = 14) and PCR-positive samples at (a) FP (N = 24) and (b) FP+7 (N = 21). Horizontal dashed line represents the threshold for statistical significance (adjusted p < 0.05). (c) Discordance and Concordance (DISCO) plot comparing Log2 fold changes from comparison of samples from unexposed controls to FP samples from PCR-positive contacts to Log2 fold changes from comparison of samples from unexposed controls to FP+7 samples from PCR-positive contacts. Log transformed DISCO scores are indicated by colour. Diamonds outlined in black represent significantly differentially expressed genes (DEGs) from Fig. 2a and b. R² value of Spearman's test and median DISCO score (mDISCO) are displayed. (d) Hierarchical clustering heatmap displaying expression of 227 genes with a CTV score >0.0002. Samples from all timepoints from PCR-positive contacts and unexposed controls are included. Red denotes high Z-scored gene expression levels, blue denotes low expression. Study timepoints are annotated using colours indicated in the key and pie charts show the relative contribution of samples from each time point to the number of samples per cluster. Ward clustering linkage and Euclidean distances were used. (e) Venn diagram illustrating the overlap between Fig. 2a DEGs and high-variance genes comprising Fig. 2d GC1 identified using CTV analysis. (f) Principal component analysis of samples from PCR-positive contacts and unexposed controls using 96-gene signature expression. (g) Receiver-operating characteristic (ROC) analysis of median normalised expression of 96-gene signature genes as a predictor of infection status. Median expression in unexposed controls was compared to expression in PCR-positive individuals at the FP, FP+7, FP+14, and convalescent time points yielding AUROC values of 0.95, 0.83, 0.85, and 0.73 respectively. Comparisons where p < 0.05 are displayed. (h) Significantly enriched gene ontology terms for the 96-gene signature obtained using ShinyGo. PP = Pre-PCR-positive; FP = Time of First PCR-positive sample; FP+7 = 7 days post FP; FP+14 = 14 days post FP; SC = sample cluster; GC = gene cluster; FDR = false discovery rate.

Fig. 3

Transcriptomic profile of persistently PCR-negative COVID-19 household contacts. (a and b) Volcano plot showing differential gene expression between unexposed controls (N = 14) and PCR-negative contacts at (a) day 0 (N = 24) and (b) day 7 (N = 20). Horizontal dashed line represents the threshold for statistical significance (adjusted p < 0.05). (c) Hierarchical clustering heatmap of 111 genes with a contribution score >0.0002 in CTV analysis. All samples from PCR-negative contacts and unexposed controls are included. Red denotes high Z-scored gene expression levels, blue denotes low expression. Study timepoints of each sample are annotated using colours indicated in the key and pie charts show the relative contribution of samples from each time point to the total number of samples in each of the labelled clusters. Ward clustering linkage and Euclidean distances were used. (d) Significantly enriched gene ontology terms of all 111 CTV analysis-derived high-variance genes from Fig. 3c obtained using ShinyGo. (e) Comparison of median normalised read counts of the 9 genes present in GC3 of Fig. 3c from PCR-negative contacts at each study timepoint, and unexposed controls. Median and quartiles are displayed. Two horizontal black dashed lines represent the median GC3 gene expression in unexposed controls ± 2 median absolute deviations. SC = sample cluster; GC = gene cluster; FDR = false discovery rate.

Fig. 4

An early SARS-CoV-2 infection signature identifies immune resistance to sustained infection in a subgroup of persistently PCR-negative COVID-19 household contacts. (a) Hierarchical clustering of 96-gene signature in PCR-negative contacts and unexposed controls. Red denotes high Z-scored gene expression levels; blue denotes low expression. Study timepoints are annotated using colours indicated in the key and pie charts show the relative contribution of samples from each time point to the number of samples per cluster. Ward clustering linkage and Euclidean distances were used. (b) Comparison of median normalised read counts of 96-gene signature between unexposed samples, PCR-positive contact FP samples and samples from PCR-negative contacts stratified by Fig. 4a sample cluster. Median 96-gene signature expression was compared using Dunn's Test with Bonferroni correction. ∗p ≤ 0.05. (c) Median normalised read counts of 96-gene signature genes over time. PCR-positive contacts and SC1 PCR-negative contacts from Fig. 4a are shown in black. Other PCR-negative contacts are shown in grey. Red bands represent median normalised 96-gene signature gene count in unexposed controls ±2 median absolute deviations. Mixed effects analysis with Tukey's multiple comparisons is displayed. ∗p ≤ 0.05, ∗∗∗p ≤ 0.001. (d–g) Discordance and Concordance (DISCO) plots comparing fold change in gene expression relative to unexposed controls samples between: (d) SC1 PCR-negative contact day 0 samples and PCR-positive contact FP samples; (e) SC1 PCR-negative contact D7 samples and PCR-positive contact FP+7 samples; (f) SC2-4 PCR-negative contact day 0 samples PCR-positive contact FP samples; and (g) SC2-4 PCR-negative contact day 7 samples and PCR-positive contact FP+7 samples. Log transformed DISCO scores are indicated by colour. Genes comprising the 96-gene signature are represented by black outlined circles. Other genes are represented by crosses. R² value of Spearman's tests and median DISCO scores (mDISCO) for 96-gene signature genes are displayed. (h) Comparison of COVID-19 index case exposure score between PCR-negative contacts with samples in Fig. 4a SC1 and those without SC1 samples using Wilcoxon test ∗p ≤ 0.05. Box plots show median and quartiles. PP = Pre-PCR-positive; FP = Time of First PCR-positive sample; FP+7 = 7 days post FP; FP+14 = 14 days post FP; Conv = Convalescent; D0 = Study day 0; D7 = Study day 7; SC = sample cluster; GC = gene cluster; ns = non-significant.

Fig. 5

Early SARS-CoV-2 infection signature identifies cases of immune resistance to sustained infection in a separate, independent validation cohort. (a) Diagrammatic representation of COVIDsortium study design. RNA-seq was performed on whole blood samples from N = 38 PCR-positive and N = 58 PCR-negative healthcare workers (HCWs) with presumed high magnitude exposure to SARS-CoV-2. (b) Volcano plot representing results of Multiple Mann Whitney U tests comparing expression levels of 89 genes from the 96-gene signature between PCR-positive and PCR-negative HCWs. 7 of the 96-gene signature genes were not annotated in this dataset. Horizontal dashed line represents the threshold for statistical significance (FDR < 0.05 after 2-stage multiple hypothesis correction (Benjamini, Krieger, and Yekutieli)). 9 Fig. 3c GC3 genes PI3 are labelled. (c) Hierarchical clustering of the truncated 96-gene signature in PCR-negative HCWs. Genes are displayed in the order generated by hierarchical clustering displayed in Fig. 4c. Red denotes high Z-scored gene expression levels; blue denotes low expression. Black crosses represent the 7 genes from the 96 gene signature that were not present in the COVIDsortium dataset. Ward clustering linkage and Euclidean distances were used. SC = sample cluster; GC = gene cluster. (d) Diagrammatic representation of CHIM study design. RNA-seq was performed on longitudinal whole blood samples from individuals intranasally inoculated with SARS-CoV-2. N = 17 were seronegative at the time of inoculation and became PCR-positive and seroconverted. N = 16 were seronegative at the time of inoculation, remained PCR-negative at serial time points and did not seroconvert. (e) Volcano plot representing results of Multiple Mann Whitney U tests comparing expression levels of 89 genes from the 96-gene signature between PCR-positive and PCR-negative CHIM participants. 7 of the 96-gene signature genes were not annotated in this dataset. Horizontal dashed line represents the threshold for statistical significance (FDR < 0.05 after 2-stage multiple hypothesis correction (Benjamini, Krieger, and Yekutieli)). 9 Fig. 3c GC3 genes and PI3 are labelled. (f) Tile plot displaying the results of multiple paired Mann Whitney U-tests comparing expression of each of the indicated genes pre-inoculation to post-inoculation at each of the indicated time points. Analysis was conducted independently for PCR-positive and PCR-negative participants. Multiple comparisons are corrected for using the Benjamini–Hochberg method. Correction was applied for both comparison of multiple post-inoculation time points and comparisons of multiple genes. Blue tiles represent instances where significant differences (p < 0.05) were observed post-correction. (g) Line graphs displaying the dynamics of expression of genes that were most differentially expressed in PCR-negative contacts between pre-inoculation and 6 hr post-inoculation samples. Grey lines represent expression of genes in individual participants. Bold coloured lines indicate the median gene expression. Red lines = PCR-positive participants; blue lines = PCR-negative participants. Asterisks indicate significant differences in expression compared to pre-inoculation expression (Multiple Mann–Whitney U-test with Benjamini–Hochberg correction. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.) PI = Pre-inoculation, 6 hr = 6 hours post-inoculation.

Fig. 6

Hypothesised model of outcomes to SARS-CoV-2 exposure. (a) Schematic representation of a model of the spectrum of outcomes to SARS-CoV-2 exposure. Exposure and infection exist on a spectrum in which a proportion of contacts resist establishment of sustained replicative infection. In this model, increasing magnitude of viral exposure drives increased innate immune pathway activation. A proportion of contacts avoid viral entry into cells lining the upper respiratory tract, hence experience ‘true non-infection’ and have no measurable innate immune response to exposure. Other contacts experience abortive infection in which virus is rapidly eliminated and is undetectable by conventional PCR, or transient infection in which virus is transiently detectable by conventional PCR but is optimally controlled and does not induce seroconversion. We hypothesise that in these cases, the immune response optimally limits viral replication. This is detectable in the periphery at a magnitude commensurate with the quantity of virus a contact was exposed to. In contacts who fail to optimally control early infection, detectable, sustained replicative infection occurs. In this case intra-host viral replication determines the magnitude of the innate immune response i.e., magnitude of innate immune response correlates positively with peak viral load. We hypothesise that similar immune pathways are induced in those who resist replicative infection and those who become infected, though at different magnitudes. Further, we hypothesise that it is the speed and magnitude of the early, local immune responses that determine whether an individual succumbs to sustained replicative infection. ∗Detectable infection is defined using conventional PCR and serology methods. (b) Schematic representation of the dynamics of viral load and signature gene expression in CHIM participants over time, comparing those who resist detectable infection (blue) with those who succumb to detectable infection (red). Solid lines represent signature gene expression. Dashed lines represent viral load. Horizontal dashed line represents limit of detection for viral load measurement. A rapid increase in expression of signature genes in those with non-detectable infection (blue solid line) is sufficient to limit viral load such that it is undetectable by PCR (blue dashed line). Conversely, slower induction of signature genes in those with detectable infection fails to control early viral proliferation which continues exponentially, driving further expression of signature genes.