Research: Virome-wide detection of natural infection events and the associated antibody dynamics using longitudinal highly-multiplexed serology

Professor Tom Scriba has co-authored: "Virome-wide detection of natural infection events and the associated antibody dynamics using longitudinal highly-multiplexed serology" appearing in the Nature Communications journal.

Background:

Current methods for detecting infections either require a sample collected from an actively infected site, are limited in the number of agents they can query, and/or yield no information on the immune response.

Here we present an approach that uses temporally coordinated changes in highly-multiplexed antibody measurements from longitudinal blood samples to monitor infection events at sub-species resolution across the human virome.

In a longitudinally-sampled cohort of South African adolescents representing >100 person-years, we identify >650 events across 48 virus species and observe strong epidemic effects, including high-incidence waves of Aichivirus A and the D68 subtype of Enterovirus D earlier than their widespread circulation was appreciated. In separate cohorts of adults who were sampled at higher frequency using self-collected dried blood spots, we show that such events temporally correlate with symptoms and transient inflammatory biomarker elevations, and observe the responding antibodies to persist for periods ranging from ≤1 week to >5 years. Our approach generates a rich view of viral/host dynamics, supporting novel studies in immunology and epidemiology. 

Discussion:

In this study, we have developed an approach for the virome-wide detection of infection events within longitudinal high-dimensional antibody measurements and used this system to study viral and host dynamics in individuals and populations over timescales ranging from days to years. Our approach combines a highly-multiplexed serological assay platform with a repurposed statistical tool for the analysis of the resulting time-resolved, high-dimensional reactivity data (Fig. 1). In our analysis of longitudinally-sampled cohorts, we identified 100s of species-specific ‘viral antibody events’ (VAEs) that consist of temporally-co-ordinated changes in reactivity to epitopes across the respective proteomes (Fig. 2). We ascribe these serological events to co-incident viral infections or reactivations, an interpretation supported by the temporal association between VAEs and orthogonal markers of infection (Fig. 5), as well as the strong epidemic effects that we observe in a synchronously-sampled cohort (Fig. 3).

Like all serological assays for infection, our approach is limited by the kinetics of the antibody response, and so becomes sensitive later during the course of infection than direct viral detection. Nonetheless, our detection of significant species-specific signal within as few as 3 days after symptom onset (Fig. 5b, c) suggests that longitudinal sampling in the early symptomatic period may be diagnostic in the setting of a recall response, which likely applies to the majority of natural human infection events. Nonetheless, the advantages of the longitudinal, highly-multiplexed approach are likely strongest in the context of population-level surveillance or retrospective correlative studies that do not require the timely detection of individual cases. While low-dimensional serology has also been used effectively for such purposes, the highly-multiplexed peptide-based approach offers a number of important advantages, including: (i) an ability to simultaneously query many infectious agents, (ii) increased statistical power resulting from the detection of correlated signals across many proteins/epitopes per species target, (iii) increased taxonomic discrimination resulting from epitope-level resolution, and (iv) the generation of rich background distributions that allow more accurate normalization across samples with global differences in signal quality. The latter consideration may be particularly important for cohorts that would otherwise be confounded by sample-to-sample variability due to differences in storage/degradation (e.g. dried blood samples), which may pose particular challenges in longitudinal studies.

While the statistical approach that we describe allows viral events to be inferred with high-confidence and specificity, the natural history cohorts studied here do not allow a precise quantification of the sensitivity with which it detects true infection events. There are, however, several observations that suggest a relatively low false negative rate. First, the overall frequency of events detected in the ACS cohort (~5 per person-year, ~50% of which correspond to respiratory viruses), is broadly consistent with published estimates^3,30,31. Second, the fractions of inflammatory or febrile episodes in the SISCAPA and MyImmunity cohorts for which we detect a VAE—4/9 and 2/4 respectively—is relatively high, particularly in light of the potential for such episodes to have non-viral causes. Third, we detect epidemic waves affecting large subsets (5-30%) of the cohort, including one for Influenza A virus encompassing ~20% of participants, a value that is consistent with estimates for the annual incidence in the United States of this well-studied virus^32,33. Future studies in which the approach described here is paired with the direct detection of viral nucleic acids or proteins will be useful to compare the sensitivities of these approaches, although such studies will need to be designed to account for the transient and often site-specific appearance of viral components during an infection.

The observation that all four VAEs that we detected in the SISCAPA cohort are associated with elevations of both CRP and SAA (Fig. 5) suggests that time-dense measurements of these inflammatory markers can provide a sensitive measure of viral infection. It is notable that the strongest inflammatory event detected here (which also includes elevations in MPO and IgM; marked by red vertical stripe in Fig. 5a) did not associate with an inferred viral event, but instead coincides with an episode of X-ray-confirmed antibiotic-responsive pneumonia. We hypothesize that some or all of the four other inflammatory events that lack a co-incident VAE could likewise be explained by infections with non-viral agents, but without being severe enough to reach clinical attention and involve MPO/IgM elevations. Alternatively, a subset of these events may be attributable to viral infections to which our approach is insensitive, potentially related to the signal loss we observed in DBS samples (Supplementary Fig. 2). These hypotheses should be testable in future studies that apply highly-multiplexed serology, such as PepSeq, to measure antibody reactivity to bacterial and/or fungal antigens in longitudinal series of this type. Future studies in larger cohorts will also be important to determine how the overall association between viral antibody and inflammatory profiles described in Fig. 5 generalizes beyond the single participant shown here.

The potential of our approach to yield valuable epidemiological insights is exemplified by the detection of high-prevalence epidemic waves for Aichivirus A and the D68 subtype of Enterovirus Din the ACS cohort (Fig. 3). The observation of 11 synchronous Enterovirus D68 events in the Western Cape province of South Africa during 2006 is striking and would represent the largest known cluster of infections of its time for this virus³⁴. Enterovirus D68 came to prominence in the autumn of 2014 when it was associated with a large outbreak of pediatric lower-respiratory disease in North America and Europe, along with >100 cases of acute flaccid paralysis in the United States³⁵. It has since caused smaller, biennial outbreaks in the summer-to-autumn season. Although the peak of the wave in our cohort appears to have preceded our sampling window (which began mid-2006), extrapolation from the timing of decreasing reactivity (Fig. 3b) suggests a peak of infections ~April of 2006, coinciding with the Southern Hemisphere autumn season. Our observation of a high-incidence (20%) outbreak in this adolescent surveillance cohort that was unselected for active respiratory disease or symptoms is consistent with the model that Enterovirus D68 is a widely-circulating virus that only comes to clinical attention in a small minority of cases³⁶. Significant circulation in South Africa in 2006, despite the absence of a recognized major clinical outbreak until 2014, might be explained by geographically-related differences in host susceptibility, evolution of the virus during the intervening years, and/or limitations in the available surveillance tools. Similarly, our detection of a wave of Aichivirus A infections affecting ~30% of participants over the Southern Hemisphere summer of 2007-2008 (Fig. 3b) to our knowledge represents the highest incidence wave ever detected for this virus in a general cohort and the earliest documented evidence of its widespread circulation on the African continent^37,38.

Also notable is our detection of (non-epidemically-clustered) VAEs across 7 members of the Human Herpesvirus (HHV) family (Fig. 2a). The HHVs establish life-long infections characterized by long periods of latency, but are known to sometimes reactivate, typically in conditions of stress or immunosuppression, and sometimes with important clinical consequences³⁹. The fact that we detect pre-existing reactivity prior to each HHV event indicates that these VAEs are best explained as viral reactivations, as opposed to primary infections, although re-infections are also possible. Moreover, although the ACS cohort lacks the clinical records necessary to track individual outcomes, the frequencies of these events (e.g. 28 and 11 for Human betaherpesvirus 5 and Human alphaherpesvirus 3, respectively) far exceed the expected incidences of their respective diseases (CMV reactivation disease and Herpes Zoster), indicating that most were subclinical or asymptomatic^40,41. The possibility that serological profiling of frequent, subclinical HHV reactivations could serve as a sensitive and dynamic reporter of a person’s immunological health is an intriguing area for future study. Also remarkable is the detection of VAEs for Measles morbillivirus and Rubivirus rubellae in the ACS cohort, representing species whose circulation is increasingly curtailed by highly-effective childhood vaccination. The detection of 8 Rubella events (1 per ~15 person-years) is particularly notable as it reveals significant circulation in this population, likely reflecting the absence of Rubella in the standard childhood vaccination schedule in South Africa⁴².

We expect the approach developed here to find application in future studies of epidemiological patterns in cohorts sampled longitudinally over timescales of months-years, or more fine-level immune dynamics in individuals sampled over days-weeks. Our adaptation of the PepSeq assay to self-collected dried blood spot samples (Fig. 5) will also enable new, scalable study designs with temporal resolution not readily achieved with traditional clinical collections. Moreover, since highly-multiplexed peptide-based serology platforms like PepSeq are fully-customizable in their antigen content, it should be possible to directly extend the experimental and analytic approaches described here to other areas of epidemiological or clinical interest: for example to enable the broad identification of bacterial infection or allergen exposure events