Periodic shifts in viral load increase risk of spillover from bats

Abstract

Prediction and management of zoonotic pathogen spillover requires an understanding of infection dynamics within reservoir host populations. Transmission risk is often assessed using prevalence of infected hosts, with infection status based on the presence of genomic material. However, detection of viral genomic material alone does not necessarily indicate the presence of infectious virus, which could decouple prevalence from transmission risk. We undertook a multi-faceted investigation of Hendra virus shedding inPteropusbats, combining insights from virus isolation, viral load proxies, viral prevalence, and longitudinal patterns of shedding, from 6,151 samples. In addition to seasonal and interannual fluctuation in prevalence, we found evidence for periodic shifts in the distribution of viral loads. The proportion of bats shedding high viral loads was higher during peak prevalence periods during which spillover events were observed, and lower during non-peak periods when there were no spillovers. We suggest that prolonged periods of low viral load and low prevalence reflect prolonged shedding of non-infectious RNA, or viral loads that are insufficient or unlikely to overcome dose barriers to spillover infection. These findings show that incorporating viral load (or proxies of viral load) into longitudinal studies of virus excretion will better inform predictions of spillover risk than prevalence alone.Significance statementWe present a comprehensive analysis of a high-profile bat-virus system (Hendra virus in Australian flying-foxes) to demonstrate that both prevalence and viral loads can shift systematically over time, resulting in concentrated periods of increased spillover risk when prevalence and viral loads are high. We further suggest that prolonged periods of low-prevalence, low-load shedding may not reflect excretion of infectious virus, resolving the outstanding puzzle of why spillovers have not been observed during periods of low off-season prevalence in subtropical Australia, or more frequently in tropical Australia despite consistent low-prevalence shedding. The consideration of viral loads (or proxies of viral load) along with prevalence may improve risk inference from longitudinal surveys of zoonotic viruses across wildlife reservoir hosts.

Reconstructing prevalence dynamics of wildlife pathogens from pooled and individual samples

“…A second model requirement is that the distribution of biomarker values, which are used to calculate the biomarker probability distribution of pooled samples, is assumed to be constant over time. Although this can be a reasonable baseline assumption, recent work suggests this may not always be the case [25]. Therefore, it is possible to adapt the model using a time-dependent probability distribution when pathogen shedding concentrations are known or suspected to be higher during certain periods.…”

“…Gaussian). Unfortunately, this is typically the case in biological systems, where biomarker distributions are likely to be skewed and non-standard [20, 21], and can change over time [24, 25]. Here, we present an approach that can use any distribution of biomarker values to estimate prevalence from pooled samples.…”

“…The distribution of Ct values in a population rarely follows a uniform distribution, and can instead follow many possible non-standard distributions (e.g., a skewed distribution when low concentrations are more likely). These distributions can also change over time and with changing biological conditions [25]. When this is the case, probability can be calculated by first calculating the total probability of each combination, then taking the sum of the total probabilities of all combinations that result in Ct value C , and dividing this by the sum of all total probabilities of all combinations.…”

“…Then a Ct value was generated for each individual bat, with negative bats receiving a Ct value of 0 and positive bats receiving a Ct value randomly drawn from a non-standard, realistic probability distribution of Ct values. Last, the resulting Ct value of the pooled sample was calculated by first converting each individual Ct value to number of genome copies [25], calculating the mean number of genome copies (including the negative samples), and re-converting this mean of the pooled sample to a Ct value. Note that while a Ct value is generated for individuals contributing to a pooled sample, the individuals used for the “individual sample” model described in the previous paragraph only have a negative or positive status, and not a Ct value.…”

“…There are solutions for this, however. One solution would be to discretize Ct values into larger intervals (e.g., [21][22][23][24], [24][25][26], etc. ), and/or setting all numbers of individuals above a certain maximum value equal to that value.…”