key: cord-0869887-6yh82e1w
authors: Mavian, Carla; Pond, Sergei Kosakovsky; Marini, Simone; Magalis, Brittany Rife; Vandamme, Anne-Mieke; Dellicour, Simon; Scarpino, Samuel V.; Houldcroft, Charlotte; Villabona-Arenas, Julian; Paisie, Taylor K.; Trovão, Nídia S.; Boucher, Christina; Zhang, Yun; Scheuermann, Richard H.; Gascuel, Olivier; Lam, Tommy Tsan-Yuk; Suchard, Marc A.; Abecasis, Ana; Wilkinson, Eduan; de Oliveira, Tulio; Bento, Ana I.; Schmidt, Heiko A.; Martin, Darren; Hadfield, James; Faria, Nuno; Grubaugh, Nathan D.; Neher, Richard A.; Baele, Guy; Lemey, Philippe; Stadler, Tanja; Albert, Jan; Crandall, Keith A.; Leitner, Thomas; Stamatakis, Alexandros; Prosperi, Mattia; Salemi, Marco
title: Sampling bias and incorrect rooting make phylogenetic network tracing of SARS-COV-2 infections unreliable
date: 2020-06-09
journal: Proc Natl Acad Sci U S A
DOI: 10.1073/pnas.2007295117
sha: 3b3794fc48e257c10d1130a51f9fef688c28c215
doc_id: 869887
cord_uid: 6yh82e1w

nan

There is obvious interest in gaining insights into the epidemiology and evolution of the virus that has recently emerged in humans as the cause of the coronavirus disease 2019 (COVID-19) pandemic. The recent paper by Forster et al. (1) analyzed 160 severe acute respiratory syndrome coronavirus (SARS-CoV-2) full genomes available (https://www.gisaid.org/) in early March 2020. The central claim is the identification of three main SARS-CoV-2 types, named A, B, and C, circulating in different proportions among Europeans and Americans (types A and C) and East Asians (type B). According to a median-joining network analysis, variant A is proposed to be the ancestral type because it links to the sequence of a coronavirus from bats, used as an outgroup to trace the ancestral origin of the human strains. The authors further suggest that the "ancestral Wuhan B-type virus is immunologically or environmentally adapted to a large section of the East Asian population, and may need to mutate to overcome resistance outside East Asia." There are several serious flaws with their findings and interpretation. First, and most obviously, the sequence identity between SARS-CoV-2 and the bat virus is only 96.2%, implying that these viral genomes (which are nearly 30,000 nucleotides long) differ by more than 1,000 mutations. Such a distant outgroup is unlikely to provide a reliable root for the network. Yet, strangely, the branch to the bat virus, in figure 1 of their paper, is only 16 or 17 mutations in length. Indeed, the network seems to be misrooted, because (see their SI Appendix, figure S4 ) a virus from Wuhan from week 0 (24 December 2019) is portrayed as a descendant of a clade of viruses collected in weeks 1 to 9 (presumably from many places outside China), which makes no evolutionary (2) or epidemiological sense (3) .

As for the finding of three main SARS-CoV-2 types, we must underline that finding different lineages in different countries and regions is expected with any RNA virus experiencing founder effects (2) . According to Forster et al.'s (1) own analysis, a single synonymous mutation (nucleotide change in a gene that does not result in a modified protein) distinguishes type A from type B, while one nonsynonymous mutation (resulting in a protein with a single amino acid change) separates types A and C, and another one separates types B and C. Given SARS-CoV-2's fast evolutionary rate, random emergence of new mutations is entirely expected, even in a relatively short timeframe (4) . When a viral strain is introduced and spreads in a new population, such random mutations can be propagated without them being selected or advantageous, due to founder effects. The fact that SARS-CoV-2 sequences show some geographical clustering is not new and is nicely and interactively shown on Nextstrain (5), but this cannot be used as a proof of biological differences unless backed by solid experimental data (6) . This is particularly true for the work of Forster et al., since their findings are based on a nonrepresentative dataset of 160 genomes, with no significant correlation between prevalence of confirmed cases and number of sequenced strains per country (7, 8) . The essential role of representative sampling is well documented in the literature (9), but was not acknowledged by the authors, who, instead, claim that their "network faithfully traces routes of infections for documented cases," without taking into consideration missing viral diversity, or evaluating multiple transmission hypotheses that would be consistent with sequence data, or even providing any support on the robustness of the branching pattern in their network. Ultimately, no firm conclusion should be drawn without evaluating the probability of alternative dissemination routes.

The inappropriate application and interpretation of phylogenetic methods to analyze limited and unevenly sampled datasets begs for restraint about origin, directionality, and early clade/ lineage inference of SARS-CoV-2. We feel the urgency to reframe the current debate in more rigorous scientific terms, given the dangerous implications of misunderstanding the true dispersal dynamics of SARS-CoV-2 and the COVID-19 pandemic.

Phylogenetic network analysis of SARS-CoV-2 genomes

The Phylogenetic Handbook: A Practical Approach to Phylogenetic Analysis and Hypothesis Testing

The continuing 2019-nCoV epidemic threat of novel coronaviruses to global health-The latest 2019 novel coronavirus outbreak in Wuhan, China

Evolutionary analysis of the dynamics of viral infectious disease

Nextstrain: Real-time tracking of pathogen evolution

We shouldn't worry when a virus mutates during disease outbreaks

A snapshot of SARS-CoV-2 genome availability up to 30 th March, 2020 and its implications

State of GISAID COVID-19 sequence availability

Eight challenges in phylodynamic inference

We are grateful to Paul Sharp, Andrew Rambaut