Predicting Potential PRRSV-2 Variant Emergence through Phylogenetic Inference

Abstract

Porcine reproductive and respiratory syndrome (PRRS) is a significant pig disease causing substantial annual losses exceeding half a billion dollars to the United States pork industry. The cocirculation and emergence of genetically distinct PRRSV-2 viruses hinder PRRS control, especially vaccine development. Similar to other viral infections like seasonal flu and SARS-CoV-2, predictive tools for identifying potential emerging viral variants may prospectively aid in preemptive disease mitigation. However, such predictions have not been made for PRRSV-2, despite the abundance of relevant data. In this study, we analyzed a decade's worth of virus ORF5 sequences (n = 20,700) and corresponding metadata to identify phylogenetic-based early indicators for short-term (12 months) and long-term (24 months) variant emergence. Our analysis focuses on PRRSV-2 Lineage 1, which was the predominant lineage within the U.S. during this period. We evaluated population expansion, spatial distribution, and genetic diversity as key success metrics for variant emergence. Our findings indicate that successful variants were best characterized as those that underwent population expansion alongside widespread geographical spread but had limited genetic diversification. Conditional logistic regression revealed the local branching index as the sole informative indicator for predicting population expansion (balanced accuracy (BA) = 0.75), while ancestral branch length was strongly linked to future genetic diversity (BA = 0.79). Predicting spatial dispersion relied on the branch length and putative antigenic difference (BA = 0.67), but their causal relationships remain unclear. Although the predictive models effectively captured most emerging variants (sensitivity = 0.58-0.81), they exhibited relatively low positive predictive value (PPV = 0.09-0.16). This initial step in PRRSV-2 prediction is a crucial step for more precise prevention strategies against PRRS in the future.

Figure 1

Conceptual framework of data generation for systematic predictive modeling. (a) Temporal distribution of PRRSV-2 L1 ORF5 sequences used in this study. As an example, observation time (t) is shown in July 2011 (vertical arrow) with its corresponding pretree (purple bars) and posttree (purple and gray bars) periods. The pretree and posttree were built for each t set as every six months (red bar) from 2011 to 2020. (b) Example pre- and post-timed phylogenetic trees inferred from sequencing data. Tips in purple show sequences from the pretree that are present in both posttrees. (c) Information computed in an example pretree, including designated variants (colored rectangle frames) and early indicators (red circle shows the ancestral node of the blue variant). (d) Success measures (colored oblong shape) are calculated from variants' new descendants in the posttree.

Figure 2

Matrix of Spearman's correlation coefficients (ρ) between all candidate early indicators for the overall data and each prediction scenario data with background color corresponding to the strength of correlation from 1 (red) to −1 (blue) (upper panel), their data density plots (diagonal), and bivariate scatterplots colored by the scenario with LOESS curves fitted (red line) and associated 95% confidence intervals (gray polygon) (lower panel).

Figure 3

Aspects of success, success measures, and distribution of values for success vs. unsuccess of each measure. (a) Distribution of success metrics for population expansion (orange) and genetic diversity (green). (b) Distribution of success metrics for spatial distribution (blue). (c) Venn diagrams tabulating the number of variants that achieved success in one or more of the population, geographic, or genetic diversification aspects (not including success in relative increase in number of states).