Bats host major mammalian paramyxoviruses

Drexler, Jan Felix; Corman, Victor Max; Müller, Marcel Alexander; Maganga, Gael Darren; Vallo, Peter; Binger, Tabea; Gloza-Rausch, Florian; Rasche, Andrea; Yordanov, Stoian; Seebens, Antje; Oppong, Samuel; Sarkodie, Yaw Adu; Pongombo, Célestin; Lukashev, Alexander N.; Schmidt-Chanasit, Jonas; Stöcker, Andreas; Carneiro, Aroldo José Borges; Erbar, Stephanie; Maisner, Andrea; Fronhoffs, Florian; Buettner, Reinhard; Kalko, Elisabeth K.V.; Kruppa, Thomas; Franke, Carlos Roberto; Kallies, René; Yandoko, Emmanuel R.N.; Herrler, Georg; Reusken, Chantal; Hassanin, Alexandre; Krüger, Detlev H.; Matthee, Sonja; Ulrich, Rainer G.; Leroy, Eric M.; Drosten, Christian

ICTV currently lists 36 accepted PV species. Gene databases comprise more viruses, many of which are only partially sequenced and not classified into defined species (Supplementary Tables S6 and S7 ). Although 83 clearly distinct PV taxa could be discriminated in our data set based on phylogeny, a classification criterion using a distance threshold comparable to that between HeV and NiV (7.0-7.5% in the L-gene fragment used) identified 66 independent novel taxonomic entities on the rank of tentative species to be represented in our data. Nevertheless, the number of novel taxa added to the PV phylogeny is unlikely to cause a bias toward bat-and rodent viruses in subsequent analyses of host associations. This is foremostly because reference sequences in these analyses were selected to maximise the patristic distance within each genus of PV (Fig. 8a) . The algorithm underlying the parsimony-based ancestral state reconstruction only counts host switches once per resolved phylogenetic root point, irrespective of the number of depending leaves carrying identical traits. In general, probabilistic approaches such as ML-and Bayesian methods are considered more powerful than parsimony models for the reconstruction of character states evolving along phylogenetic branches 34 . However, this is mainly because probabilistic models take branch lengths into account, whereas parsimony-based methods only consider tree topology 34 . For the particular task of reconstructing viral host associations, we favour parsimony approaches out of theoretical considerations. First, we had to assume that host transition happens rarely and is unlikely to take place in a bidirectional manner, because of the fitness valley effect that will prevent host changes to be reversed easily 4, 10 . In particular, viruses conquering a new host will leave behind population immunity in their original AR, preventing re-introduction, and making back and forward transitions (as well as host switching as a whole) a rare process 4, 10 . Another argument was the uncertainty of deeper branch lengths in viral phylogenies 35 . More recent studies on non-retroviral RNA viruses invading mammalian germlines demonstrated tremendous discrepancies between apparent evolutionary rates of extant RNA viruses versus those of phylogenetic stem lineages [36] [37] [38] [39] . Rate differences in deep branches will have great influence on probabilistic models, but will not affect parsimony assumptions. Even though recent findings of viral germline fossils comply with the idea that ancient PV may have existed in mammals [36] [37] [38] [39] , such data are so far unavailable for PV, and we have not attempted in our analysis to determine the evolutionary origins of PV. Such an analysis would require a different approach to sampling, as viruses from taxa other than mammals and birds are currently underrepresented in databases. To attenuate the contribution of deep phylogenetic nodes, we have limited our analysis to an estimation of trait switches along trees, rather than inferring host associations for deep phylogenetic nodes. With this limitation in mind, we can conclude that bats have most often been the donors of those viruses currently encountered in other mammals.