Transmission cluster characteristics of global, regional, and lineage-specific SARS-CoV-2 phylogenies

Abstract

The SARS-CoV-2 pandemic has been presenting in periodic waves and multiple variants, of which some dominated over time with increased transmissibility. SARS-CoV-2 is still adapting in the human population, thus it is crucial to understand its evolutionary patterns and dynamics ahead of time. In this work, we analyzed transmission clusters and topology of SARS-CoV-2 phylogenies at the global, regional (North America) and clade-specific (Delta and Omicron) epidemic scales. We used the Nextstrain’s nCov open global all-time phylogeny (September 2022, 2,698 strains, 2,243 for North America, 499 for Delta21A, and 543 for Omicron20M), with Nextstrain’s clade annotation and Pango lineages. Transmission clusters were identified using Phylopart, DYNAMITE, and several tree imbalance measures were calculated, including staircase-ness, Sackin and Colless index. We found that the phylogenetic clustering profiles of the global epidemic have highest diversification at a distance threshold of 3% (divergence of 10, where the tree sampled median is 49). Phylopart and DYNAMITE clusters moderately-to-highly agree with the Pango nomenclature and the Nextstrain’s clade. At the regional and clade-specific scale, transmission clustering profiles tend to flatten and similar clusters are found at distance thresholds between 0.05% and 25%. All the considered phylogenies exhibit high tree imbalance with respect to what expected in random phylogenies, suggesting short infection times and antigenic drift, perhaps due to progressive transition from innate to adaptive immunity in the population.

I. Introduction

The global epidemic of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), causative agent of the coronavirus disease 2019 (COVID-19), has been ongoing since the end of 2019 –declared as pandemic in March 2020– to present days (September 2022), with over 600 million cases and over 6 million deaths recorded in the world (https://covid19.who.int/). Vaccines have been made available for emergency use as early as June 2020 in China, and August 2020 in Russia. In Europe and North America, vaccines received emergency authorization at the end of 2020. As of September 2022, over 12 billion vaccine doses have been administered worldwide.

Across three years, SARS-CoV global and regional spread presented in periodic waves, the virus diversified into different variants, each characterized by a constellation of mutations, with convergent emergence in some cases due to selective pressure from human host immunity [1]. The SARS-CoV-2 evolution during the pandemic seems characterized primarily by purifying selection, but a small set of sites appear to evolve under positive selection [2]. Virus variants have been associated with increased transmissibility, virulence and changes to antigenicity [3]-[5]. Over time, certain variants became dominant by replacing others, e.g., Alpha, Delta, Gamma, and the most recent Omicron. There is also evidence that some circulating SARS-CoV-2 lineages are recombinant [6]. Lineage and variant nomenclature is based on phylogenetic divergence among isolates as well as epidemiological evidence. A systematic definition of lineages, called Pango, has been introduced [7]. The World Health Organization developed a nomenclature for variants of concerns (https://www.who.int/activities/tracking-SARS-CoV-2-variants) and appointed a technical advisory group on virus evolution to develop an early warning system of variant emergence [8].

The enormous collaborative efforts of the scientific community produced very large data SARS-CoV-2 repositories, e.g., GISAID (https://gisaid.org/) with over 13 million sequence submissions, and insightful software tools to view the epidemic in real time, e.g., Nextstrain [9]. Nextstrain utilizes its own clade nomenclature (https://nextstrain.org/blog/2021-01-06-updated-SARS-CoV-2-clade-naming). Dellicour et al. introduced a phylodynamic tool to analyze how dispersal dynamics of lineages could be affected by interventions, e.g., lockdowns [10]. At present, there is a variety of software tools specific for SARS-CoV-2 from sequence analysis to protein structure and interactome prediction [11].

The number of molecular epidemiology studies that analyzed the SARS-CoV-2 pandemic is staggering, and many of them helped shedding light on evolutionary dynamics of the virus [12], [13]. Analyses at the regional level helped identifying introduction events, variant spread, and possible effects of public health measures [14]-[18].

SARS-CoV-2 is still adapting in the human population, thus it is crucial to understand its evolutionary patterns and dynamics ahead of time. In this work, we aimed at probing the following questions: (1) Are the current clade and lineage nomenclatures consistent with intra-inter lineage phylogenetic diversity? (2) Are there differences in transmission cluster shapes among regional epidemics and do they differ between variants? (3) Are the phylogenies imbalanced and do they suggest short infection times with antigenic drift, partial cross-immunity, with differences among regional or clade-specific trees? Accordingly, we analyzed both global and regional phylogenies of SARS-CoV-2, comparing the statistical phylogenetic clustering method Phylopart with the Nextstrain’s clade and Pango lineage nomenclature, assessing how the transmission clusters change from the global, to the regional, to the variant-specific scale, and quantifying several tree imbalance measures with respect to random phylogenies and other known viruses.

II. Methods

We used Nextstrain’s dashboard (https://nextstrain.org/) to select the nCov (i.e., SARS-CoV-2) open global all-time phylogeny scaled in genetic divergence, and then to sub-select the North American, Delta21A, and Omicron20M trees. Nextstrain phylogenies use sequence data and metadata from NCBI GenBank (https://www.ncbi.nlm.nih.gov/genbank/), under the Open Data principles (https://opendatahandbook.org/guide/en/what-is-open-data/). This work does not use original data, but only Nexstrain’s trees, with annotated geography and lineage. For replication purposes, and to acknowledge the data sources (including the effort of researchers who contribute to open data), we uploaded the Genbank accession numbers as well as all metadata to credit all data generators and authors at: https://github.com/DataIntellSystLab/phylopart-sarscov2.

We made two important assumptions in this study: (1) the sampling was representative and uniform; and (2) the tree inferred topologies were correct. Transmission clusters were identified using Phylopart [19] on mid-point rooted trees at multiple distance thresholds over a grid (from 5·10⁻⁵ to 0.25 with 25 steps), estimating the overall patristic distance distribution through sampling of 1,000,000 tip pairs. We also used, DYNAMITE, a refined version of Phylopart that also considers internal nodes into clustering, in a ‘dynamic’ perspective [20]. Agreement among cluster sets was assessed using the adjusted Rand index [21]. All other analyses were performed using R (https://www.r-project.org/), using libraries: ape, mcclust, phytools, and treebalance. The following tree imbalance measures were considered: area per pair index; average leaf depth index; cherry index; Colless index; maximal difference in widths; Rogers index; rooted quartet index; Sackin index; staircase-ness; total cophenetic index [22]. We focused in particular on the staircase-ness [23] that, for every bifurcating tree node, quantifies if one branching contains more tips than the other one. Of note, we used multiple measures since a single one is usually not enough to discriminate imbalance [24], [25]. All indices were compared to distributions obtained from 200 randomly generated trees with the same number of tips as the real phylogenies.

III. Results

Nextstrain’s September 2022 nCov open global all-time phylogeny included 2,698 strains. The regional North American, Delta21A, Omicron20M trees included 2,243, 499, and 543 strains, respectively.

The distribution of nodes at each tree level (from the root) in the global phylogeny deviated from what expected in random trees with the same number of tips, exhibiting slightly more nodes than expected towards the root and the tips of the tree, and significantly less in the central part. This could indicate either a problem with sampling, or different rates, likely from lineages emerging not uniformly from the tree. The median root-to-tip distance increased moderately (0.44 x tree level + 37, ρ=0.73), whilst the median branch length remained constant (Fig. 1).

Fig. 1.

Distribution of number of nodes compared to random trees (left Y axis), median (IQR) root-to-tip distance (right Y axis), and median (IQR) branch length (secondary axis) per tree level (X axis) for the SARS-CoV-2 global phylogeny.

Phylopart identified a clear peak of maximal cluster diversification for both the global and regional North American trees at the 3^rd percentile of the overall distance distribution, i.e., a divergence of 10 where the median tree divergence was 49. For the Delta21A and Omicron20M the highest number of clusters was attained at the 4^th percentile, but there was no clear peak across thresholds (Fig. 2).

Fig. 2.

Number of transmission clusters found by Phylopart on the global, North American, Delta21A, and Omicron20M SARS-CoV-2 phylogenies by varying the patristic distance threshold.

The maximal agreement between Phylopart clustering and Pango lineages was found at a diversity threshold of 8^th percentile, i.e., a divergence of 17, which was higher than the maximal cluster diversification threshold. The adjusted Rand index ranged between 0.48 and 0.55 (considering all tree tips or removing unclustered/singleton isolates, respectively). The maximal agreement between Phylopart and Nextstrain’s clade was found at thresholds between 8^th to 12^th percentile (i.e., diverge of 17 to 19) and the adjusted Rand index ranged between 0.50 and 0.73. Fig. 3 shows the global tree where tips have been colored according to Phylopart’s clustering (3^rd and 8^th distance percentile), Nextstrain’s clade, and Pango lineage. Major discrepancies between Phylopart and Nextstrain’s clade were found in 20B, Delta21J, Epsilon20C, 20A, 20D, Omicron21L, 20G, and Delta21A (agreement between 71.2% and 97.1%). The highest concordance was with Omicron22A, Omicron22B, Omicron22C, Omicron21K, Delta21I, Lambda21G, Eta21D, Theta21E, Kappa21B, EU120E, 20F, 19A, and 19B (agreement >99.7%).

Fig. 3.

Comparison of phylogenetic clustering by Phylopart with Pango lineages and Nextstrain’s clades on the SARS-CoV-2 global tree.

When performing clustering with DYNAMITE, results were similar to those obtained with Phylopart. The best agreement with both Pango lineages and Nextstrain’s clades was found at the 25^th percentile threshold (ARI=0.72 for both Pango and clades). At the 10^th percentile it was similar to Phylopart (ARI=0.41 and 0.46, respectively). Of note, by calculating the DYNAMITE’s threshold using only the portion of the tree from the root to the current nodes being considered for clustering, i.e., in a truly dynamic way that ignores ‘future’ nodes of the tree, the ARI with Pango and clade increased to 0.8. For instance, the SC2 lineage classification was improved by allowing cluster assignment to depend only on what was happening with the outbreak thus far and not on the outbreak/tree in its entirety (i.e., retrospectively).

All the considered phylogenies exhibited high tree imbalance with respect to what expected in random phylogenies. The global tree presented the highest deviation from the random tree distribution –in terms of t-value– for the area per pair index, cherry index, Rogers index, and staircase-ness. The North American tree exhibited the highest deviation for the maximal difference in widths. The Delta21A had the highest deviation for total cophenetic coefficient, Colless index, Sackin index, root quartet index, and average leaf depth index. Of note, every single t-value yielded a p-value below 0.0001, even after adjustment for multiple testing.

IV. Discussion

The transmission cluster characteristics of multi-lineage global and regional SARS-CoV-2 phylogenies are more similar than lineage-specific phylogenies. The clustering of the global tree as estimated by Phylopart is moderately-to-highly correlated with the Nexstrain’s clade and Pango nomenclature, and similar, yet better in terms of ARI, results are obtained when using DYNAMITE. Although lineages and clades incorporate epidemiological evidence besides mere phylogenetic criteria, we recognize some inconsistencies in the divergence thresholds used. Phylopart performs clustering only at the tip level, but DYNAMITE overcomes this problem and better resembles the dynamic lineage/clade assignment. Both Phylopart and DYNAMITE can be biased by sampling rates [26].

All SARS-CoV-2 phylogenies analyzed in this work show high tree imbalance, suggesting short infection times and antigenic drift [27], perhaps due to progressive transition from innate to adaptive immunity in the population [28]. Since there was not a single tree that had the highest imbalance across all measures, we cannot determine if one out of the global, regional or clade-specific epidemics might be subject to higher antigenic drift than the others.

Limitations of this work include assumptions’ violations of a correct phylogeny and of a uniform, representative sampling [29]. A possible solution to overcome sampling bias could be to generate multiple phylogenies using the TARDiS method [30], which optimizes both genetic diversity and temporal distribution, and perform ensemble analyses. Another problem is that we assumed 100% node reliability for the Nextstrain’s trees, whereas poorly supported subtrees might have impacted the cluster agreement calculation.

In conclusion, we foresee improvements in the integration of statistical criteria based on phylogenetic diversity for lineage determination as well as identification of clusters of concerns within sub-epidemics, bearing always the necessity of representative sampling.