The Arenaviridae family is formed by the genera Mammarenavirus, Reptarenavirus, Hartmanivirus, Antennavirus and Innmovirus (Radoshitzky et al., 2023). The mammarenaviruses, as the name suggests, infect mammals, mostly rodents, and their geographical distribution is related to their natural reservoirs (Salazar-Bravo et al., 2002; Gonzalez et al., 2007). The genus has been divided into two groups based on its place of origin: Old World (OW) and New World (NW) mammarenaviruses. The former are mainly found in Africa, while the latter are native to the Americas. Some members of the Mammarenavirus genus can infect humans and cause hemorrhagic fever. Examples are the OW viruses Lassa (LASV) and Lujo (LUJV) in West and South Africa (Emonet et al., 2006) and the NW viruses Junin (JUNV), Machupo (MACV), Chapare (CHPV), Guaranito (GTOV) and Sabia (SABV) in South America (Sarute and Ross, 2017). The genera Reptarenavirus and Hartmanivirus have been found in snakes (Hepojoki et al., 2015b), and some members of the genus Reptarenavirus cause Boid Inclusion Body Disease (BIBD) in captive snakes (Stenglein et al., 2012; Hetzel et al., 2013). Antennaviruses, for their part, infect striated frogfish and salmon (Shi et al., 2018; Mordecai et al., 2019), while Innmovirus is the only genus whose natural reservoir is still unknown (Chen et al., 2022; Chen et al., 2023).

Arenaviruses are single-stranded ambisense RNA viruses with some differences between the genera. Mammarenaviruses and reptarenaviruses have a bisegmented RNA with an ambisense coding strategy for 4 proteins: GPC and NP are coded in the S segment and L and Z in the L segment. Hartmaniviruses have a bisegmented RNA with an ambisense S segment coding for the GPC and NP proteins and a negative-sense RNA L segment coding for the L protein, but they have no homolog for the Z protein of mammarenavirus and reptarenavirus. Antennaviruses have genomes consisting of three genomic segments, a negative-sense S segment that encodes NP, an ambisense segment that encodes GPC and an unknown protein, and a negative-sense L segment that encodes the L protein but also has no homologous Z protein. Finally, Innmovirus has three negative-sense RNA segments, the S segment that encodes NP, the intermediate segment that encodes GPC and an unknown protein, and the L segment that encodes the L protein (Hepojoki et al., 2015b; Shi et al., 2018; Chen et al., 2022; Radoshitzky et al., 2023).

The genetic code consists of 64 codons, 61 of which code for amino acids and 3 for stop signals during protein synthesis. Since there are only 20 common amino acids, the genetic code is degenerate, i.e., several codons – synonymous codons – are translated into the same amino acids, with the exception of methionine and tryptophan. However, not all synonymous codons are used with the same frequency, with some organisms favoring some codons over others. This unequal use of codons is referred to as codon usage bias or pattern (Ikemura, 1981; Plotkin and Kudla, 2011; Chaney and Clark, 2015). Codon usage bias is a critical measure of genome evolution and has been found in most organisms, from prokaryotes to eukaryotes and viruses (Sharp and Li, 1987; Plotkin and Kudla, 2011; Belalov and Lukashev, 2013; Chaney and Clark, 2015). Several factors have been found to influence biased codon selection, including gene expression levels, translation, base mutations, genetic drift, transcription factors and the external environment (Bulmer, 1991; Butt et al., 2016; Velazquez-Salinas et al., 2016), with natural selection and mutational pressure in combination with genetic drift being the most important factors (Belalov and Lukashev, 2013).

In viral genomes, codon usage is a driving force for their evolution (Dutta et al., 2020). Some researchers suggest that mutational biases are the primary determinant of codon usage patterns in human RNA viruses (Jenkins and Holmes, 2003; Nasrullah et al., 2015; van Hemert et al., 2016; Tort et al., 2020), while others have identified the dominant influence of natural selection (Wang et al., 2016; Baha et al., 2019; Khandia et al., 2019; Luo et al., 2020). As parasitic organisms, viruses have some features in their genomes that differ from those of prokaryotes and eukaryotes, e.g., they rely on the translational machinery of their hosts for gene expression. This interaction between virus and host influences overall viral survival, adaptation, evasion of the host immune response and evolution (Nasrullah et al., 2015; Rahman et al., 2018; Nguyen et al., 2021).

Relevant human RNA viruses that have already been analyzed for codon usage bias include orthomyxoviruses (Luo et al., 2019), flaviviruses (Jenkins et al., 2001), lyssaviruses (Zhang et al., 2018), bornaviruses (He et al., 2014), polioviruses (Zhang et al., 2011), retroviruses (RoyChoudhury and Mukherjee, 2013) and coronaviruses (Mogro et al., 2022), among others. In contrast, research on arenaviruses is limited to one study on codon adaptation index (CAI) (Sharp and Li, 1987) for OW LASV and LCMV (Pontremoli et al., 2017) and one study on codon usage bias in NW CHPV and SABV (Malhotra and Kumar, 2021).

In this study, we performed an analysis of codon usage bias in the Arenaviridae family, focusing on the three proteins that are common to all members. We used different indices such as Effective Number Codons (ENC) (Wright, 1990), Relative Synonymous Codon Usage (RSCU) and CAI to draw general conclusions that could improve the understanding of the evolutionary pressures controlling the evolution of arenaviruses and their adaptability to different hosts and environments.

Our report provides a comprehensive analysis of codon usage bias for all arenavirus genera. By using a concatenated alignment of the L and NP proteins, we have constructed a phylogenetic tree that determines the distance in clustering between specific species within the family with greater precision than trees constructed using NP or L separately. As reported by Forni et al. (2018), our tree also places NW Arenavirus closer to the common ancestor of Mammarenavirus, and both OW and NW mammarenaviruses are in the same clade as Reptarenavirus, with the common ancestors of Hartmanivirus and Antennavirus being the most distant in the phylogeny. The fact that mammarenaviruses and reptarenaviruses are closer to each other in phylogeny is also reflected in their similarity in genome composition. Based on phylogeographic analyses, the possibility that OW and NW mammarenaviruses descended from a common ancestor that occurred in both Africa and South America has been proposed, with a reptilian arenavirus, such as reptarenaviruses, being an attractive candidate (Forni et al., 2018). Since the full range of natural hosts of mammarenaviruses and reptarenaviruses is unknown, another possibility is that their common ancestor infected a different mammalian host and that snakes are not the reservoir of the reptarenavirus, but an accidental host. This is supported by the analogy with the mammarenavirus, which can infect animal species other than natural hosts, at least in experimental settings, often leading to severe disease and even death (Forni et al., 2018; Pontremoli et al., 2019).

The total nucleotide content of the genome can strongly influence codon usage patterns (Jenkins and Holmes, 2003). Furthermore, van Hemert et al. (2016) suggested that nucleotide bias in RNA virus genomes is the primary determinant of specific codon usage, limiting the role of codon selection and translational control. Our analysis of the nucleotide composition of GPC, NP and L genes revealed that most codons containing the dinucleotide CG are highly repressed and those containing the dinucleotide UA are slightly underrepresented, with this difference being smaller in antennaviruses. This was to be expected as it has been reported for most RNA viruses (Karlin et al., 1994). CpG dinucleotides have been largely removed from the human genome (Cooper and Gerber-Huber, 1985), so that human mRNAs are CpG-repressed, a fact that is largely reflected by CpG suppression in viruses that infect humans, even though their genome composition may not have been shaped by DNA methylation/deamination (Cooper and Gerber-Huber, 1985). The selection pressure that has led to CpG suppression in viral genomes remains unknown (Goncalves-Carneiro et al., 2021). However, there are some reasons that could explain this, at least in part. In vertebrates, there are short, interspersed DNA sequences that deviate significantly from the average genomic pattern because they are GC- and CpG-rich and predominantly unmethylated (Deaton and Bird, 2011). These regions, known as CpG islands (CGI), are often sites of transcription initiation. For promoter function, they destabilize nucleosomes and attract proteins (Angeloni and Bogdanovic, 2021). In addition, CGIs have recently been linked to a gene regulatory mechanism in which CGIs are bound by a protein complex that specifically protects genic transcripts from premature termination, effectively distinguishing between genic and extragenic transcription and allowing normal gene expression (Hughes et al., 2023). It could be hypothesized that viral RNA with a low CG content interacts less with host proteins and is therefore more efficient for viral replication and translation. On the other hand, the antiviral host factor zinc finger protein (ZAP) has been reported to selectively inhibit replication of a variety of viruses by recognizing the CG-rich RNA sequences and activating the viral RNA degradation machinery (Pal et al., 2023). Zap and its analog PARP12 genes arose early in vertebrate divergence and originated in an original gene whose progeny are present in some modern invertebrates such as cnidarians but absent in others such as arthropods (Goncalves-Carneiro et al., 2021), suggesting that it may be a mutational pressure in all known arenaviruses. It is hoped that future studies will clarify the role of CpG suppression in viral genomes.

A weak codon usage bias was observed in all Arenaviridae genera (ENC ~50). Although Hartmanivirus had lower ENC values, the corresponding points in the ENC plot were closer to the curve (predicted ENC value in the absence of natural selection), suggesting a greater effect of compositional and mutational biases compared to the other arenaviruses. The ENC analysis also suggests a more important role of selection in the case of Mammarenavirus and Antennavirus, as also shown by the corresponding neutrality plots. Although a weak codon usage bias has already been reported for the mammarenaviruses CHPV and SABV, it was concluded that mutational pressure has the strongest influence on codon usage bias in both hemorrhagic fever pathogens (Malhotra and Kumar, 2021). The difference between our results could be due to the fact that we created the neutrality plots for each protein and grouped the arenaviruses by genus to obtain more information, as we felt that only three proteins were too few to derive a good regression.

The role of mutation and selection was not the same for each protein in all genera. Our results suggest that although selection pressure remains the most important determinant of codon usage bias, it has a stronger influence on the codon usage bias of GPC and NP than on that of L in mammarenaviruses. The L protein has been described as the main driver of evolution of the genus as it is subject to positive selection at multiple sites and there is evidence that changes in its sequence lead to differential replication efficiency and disease phenotypes in rodents (Pontremoli et al., 2017; Forni et al., 2018), and, in the case of LASV, affect transmissibility or severity of the disease in humans (Pontremoli et al., 2019). It has also been hypothesized that selection pressure acting on arenavirus L proteins confers species-specific increased replication capacity or faster spread kinetics to the virus, facilitating escape from immune surveillance and allowing persistent infection (Sullivan et al., 2015). Interestingly, our results suggest that antennaviruses and reptarenaviruses do not follow this pattern and natural selection plays a relatively more important role than mutational pressure for the L protein.

Our CAI analysis showed good agreement with the general use of host codons, with humans and Boidae scoring higher for most viruses. In particular, hartmaniviruses and antennaviruses showed the highest CAI for all proteins when the proteins of their respective hosts were used as reference sets. Hartmanivirus is also positioned differently in the correspondence analysis as it is closer to the Boidae. This suggests an adaptation to host-specific codons, but does not explain the difference to Reptarenavirus, which has the same hosts, which requires further investigation. Remarkably, Mammarenavirus and Reptarenavirus had very similar nucleotide composition, codon usage bias and CAI values, both of which were markedly different from Hartmanivirus, which had the highest CAI within the Boidae for all its proteins. This reinforces the hypothesis that reptarenaviruses may have an unknown mammalian reservoir. In addition, mammalian cells have been successfully infected with reptarenaviruses (Hepojoki et al., 2015a), and recent data also suggest successful infection in the mouse model (Abba et al., 2017). Furthermore, for mammarenaviruses, there is limited evidence of host-virus co-divergence (Forni et al., 2018), suggesting that the shared host of Reptarenavirus and Hartmanivirus does not reflect a closer evolutionary link and providing another suitable explanation for the difference in codon usage behavior between these two genera.

The CAI was also calculated using a set of highly expressed human proteins as a reference. Since CAI correlates with protein expression (Sharp and Li, 1987), our results suggest that the arenavirus proteins analyzed have the potential for efficient translation in humans, at least in terms of optimizing codon usage. However, since we found no significant difference between CAI_HHE and CAI_HLE proteins, further studies should be performed to evaluate the expression levels of arenavirus proteins in human hosts.

In hantaviruses, it has been suggested that less adaptation to their hosts allows controlled and sustained infection in their natural reservoirs (Meyer and Schmaljohn, 2000). Such differences between hosts and natural reservoirs could correlate with the severe disease that certain mammarenaviruses cause when infecting humans, as opposed to the milder symptoms of infection in their reservoirs. In support of this hypothesis, we found that arenaviruses generally have lower CAI and CAI_HHE in Muridae and Cricetidae than in humans. In addition, codon deoptimization of LCMV GPC and NP was found to produce highly attenuated viruses but still provide protection against wild-type LCMV (Cheng et al., 2015; Cheng et al., 2017). However, disease severity of LASV could not be correlated with significantly different nCAI levels, suggesting, at least for LASV, that disease severity is mainly determined either by interindividual human variability or by viral factors other than nCAI (Pontremoli et al., 2017). Future studies need to consider other factors influencing codon usage bias, as has been hypothesized for other viruses, e.g., geographic location and gene function (Chen et al., 2014; Rahman et al., 2018).

Finally, our correspondence analysis (CA) of individual RSCU codons reveals two clusters similar to those described in human mRNA that are rich in GC3 and AT3 and are associated with stabilization or destabilization of mRNA through interaction with interleukin enhancer binding factor 2 (ILF2) and perhaps ILF3 (Hia et al., 2019). The extent to which these mechanisms are effective in arenaviral RNA infections must be clarified in the future.

Here we report the codon usage pattern of the Arenaviridae family common proteins as a whole. We found a weak codon usage bias, as has been reported for other RNA viruses, with ENC and neutrality diagrams indicating an important role of selection in evolution. As with all arenaviruses (and most small eukaryotic viruses), strong repression of CG-containing codons was observed. Antennavirus and Hartmanivirus showed different behavior within the family, both in composition and codon usage pattern. Remarkably, all arenaviruses appear to have codon usage adapted to both H. sapiens and Boidae, and some degree of host adaptation of codon usage patterns was observed for Hartmanivirus and Antennavirus.