The genome of the Human Immunodeficiency Virus type 1 (HIV-1), the major causative agent of the current AIDS pandemic, is relatively small, comprising <10,000 bases in total. Arranged in three different reading frames, it contains only nine canonical genes (Figure 1). Nevertheless, the virus replicates and spreads efficiently in its human host, which expresses about 2500 times more protein-coding genes from a three billion base pair genome. How does a retrovirus with its limited genome size manage to keep pace in this David vs. Goliath struggle? How can such a tiny genome encode all the tools that are required for efficient replication and immune evasion in such a hostile environment? One major advantage of HIV-1 and related retroviruses compared to their host species is certainly their high mutation rate that allows them to quickly adapt to an ever-changing environment. Furthermore, viral proteins are often multifunctional and exert a multitude of immune evasion activities. The paragon of such a multitasking or moonlighting protein is HIV-1 Nef, which has been described to downmodulate a variety of surface receptors including CD4, MHC class I, CD28, and CXCR4, counteracts the host restriction factors SERINC3/5, and upregulates the invariant chain/CD74 to suppress antigen presentation (Pereira and daSilva, 2016). Finally, viral genomes are often very compact, containing overlapping genes that encode for bi- and multi-cistronic mRNAs. As a result, viruses frequently utilize non-canonical translation mechanisms such as internal ribosomal entry, leaky scanning, ribosomal frameshifting, shunting, or reinitiation (Firth and Brierley, 2012). Another important mechanism increasing the coding capacity of viral genomes is alternative splicing. HIV-1 and related lentiviruses contain dozens of splice donor and acceptor sites that allow the generation of more than 100 different mRNA species (Ocwieja et al., 2012; Figure 1). Generation and translation of these mRNAs are tightly regulated throughout the viral replication cycle and enable the coordinated synthesis of structural, regulatory and accessory proteins in an optimized ratio. For example, expression of the HIV-1 regulatory proteins Tat and Rev requires the joining of two exons (tat1/2 or rev1/2) via splicing at donor D4 and acceptor A7, whereas all four accessory proteins (Vif, Vpr, Vpu, and Nef) are encoded by mono-exonic genes. Notably, vpu overlaps with the viral envelope (env) gene and both are expressed from bicistronic mRNA species (Figure 1). Translation of downstream env is enabled by a weak Kozak sequence of vpu (leaky scanning) and/or ribosomal shunting mechanisms that allow to bypass upstream AUG codons (Anderson et al., 2007). While expression of Env as well as all accessory and regulatory proteins requires splicing, Gag and Pol are encoded by the 5′ half of the viral genome and expressed by unspliced viral mRNA. Gag can either be expressed alone or, upon ribosomal frameshifting, as a Gag-Pol poly-protein. The three precursor proteins Gag, Gag-Pol, and Env are proteolytically processed into mature proteins: Gag is cleaved by the viral protease into matrix, capsid, nucleocapsid, and the p6 protein. Similarly, the viral protease generates the mature viral enzymes reverse transcriptase (p51 and p66), protease and integrase from the Pol precursor protein. Finally, the Envelope protein is cleaved by the cellular protease furin into its mature subunits gp120 and gp41.

Considering the multitude of modulatory processes underlying the expression of viral proteins, it is not surprising that several studies have reported the expression of non-canonical fusion proteins by HIV-1. While the majority of these fusion proteins are the result of alternative splicing events joining regular or cryptic open reading frames, two of them are expressed only upon ribosomal frameshifting or gene rearrangements, respectively. The aim of this review article is to provide an overview of previously described fusion proteins of HIV-1. We will summarize our current knowledge on their expression and generation by different HIV-1 strains, discuss possible roles during the retroviral life cycle and critically review a potential relevance for viral replication.

In addition to alternative splicing, ribosomal frameshifting represents another mechanism that may result in the expression of fusion proteins. The most prominent example in HIV-1 and related primate lentiviruses is the Gag/Pol polyprotein, which is the result of a −1 frameshift event in the gag ORF (Figure 3). The frameshift in pol depends on a stem-loop structure stalling the translocating ribosome and an upstream heptameric “slippery site” where ribosomal frameshifting occurs (Dinman et al., 2002). While the slippage heptamer (5′-UUUUUUA-3′) itself results in frameshifting, its frequency is increased to about 5% by the adjacent stem-loop structure (Kobayashi et al., 2010; Mouzakis et al., 2013). Interestingly, a similar combination of slippage sequence and RNA secondary structure can be found within the first exon of tat (Cohen et al., 1990). This second sequence (5′-UAAAAAG-3′) is highly conserved among HIV-1 strains (Steffy and Wong-Staal, 1991) and has been shown to result in the expression of a cryptic reading frame called T that overlaps with rev1 and vpu (Figure 3) (Sonigo et al., 1985; Cohen et al., 1990). Due to a −1 frameshift, this T open reading frame (which does not harbor an initiation codon) is fused to the N-terminus of Tat1, resulting in the expression of a 17 kDa protein called Tat-T or Vpt. Although the frameshift signal is evolutionarily conserved, expression of this fusion protein in primary HIV-1 target cells is questionable. So far, this protein has only been detected upon in vitro translation using reticulocyte extract (Cohen et al., 1990). In fact, expression in infected T cells may be prevented by several splice sites disrupting the T open reading frame. In agreement with this, Tat-T/Vpt was not detectable in Jurkat and COS cells transfected with proviral HXBc2 DNA, and 50 different patient sera failed to detect expression of this fusion protein from an expression plasmid (Cohen et al., 1990). Finally, Tat-T does not exert any detectable Tat or Rev activity (Cohen et al., 1990). Nevertheless, even if Tat-T/Vpt is not expressed in vivo, it remains to be determined whether or how the frameshift sequence in tat1 affects translation of regular Tat.

In the majority of HIV-1 strains, the rev1 and vpu genes lie in different reading frames and/or are separated by an intervening stop codon (Figure 4). However, about 3% of clade A and 20% of clade C viruses as well as some circulating recombinants thereof encode an unusual rev1-vpu fusion gene (Kraus et al., 2010). Analysis of primary HIV-1 isolates harboring this ORF revealed that infected PBMCs express a Rev1-Vpu fusion protein of about 14 kDa (Langer et al., 2015). Although prevalence rates may be skewed by sampling biases, it is tempting to speculate that more than 10% of all circulating HIV-1 strains encode this unusual fusion protein, as subtype C viruses are responsible for about 50% of all infections worldwide (Osmanov et al., 2002; Hemelaar et al., 2006, 2011). Cells infected with rev1-vpu containing viruses, however, still express regular Vpu at much higher levels than Rev1-Vpu as most vpu encoding transcripts lack the initiation codon of rev1 (Kraus et al., 2010; Ocwieja et al., 2012; Langer et al., 2015): in about 75-90% of all vpu/env mRNAs, an intron containing the start codon of Rev1 has been removed due to the usage of splice acceptor A5 (Purcell and Martin, 1993; Ocwieja et al., 2012). Only in 10–25% of the cases, A4 splice acceptors are used and the complete rev1 ORF is retained. The expression of Rev1-Vpu may be further lowered by leaky scanning, in which the Rev1 initiation codon is skipped due to a weak Kozak sequence.

The characterization of virus pairs differing solely in their ability to express Rev1-Vpu revealed that the presence of this unusual fusion gene does not affect Rev-dependent nuclear export of incompletely spliced viral mRNAs. In agreement with the low Rev1-Vpu:Vpu ratio, downmodulation of CD4, tetherin counteraction and inhibition of NF-κB activation by Vpu were not affected either (Langer et al., 2015). Since the presence of rev1-vpu did not enhance viral replication in PBMCs or ex vivo infected tonsillar tissue, this gene arrangement does not seem to confer a selection advantage to HIV-1 per-se. Interestingly, however, mutations in the rev1-vpu intervening region strongly affected Env expression in some viruses (Langer et al., 2015), and the presence of the fusion gene in rev/vpu/env expression cassettes impeded pseudotyping of env-deficient viruses (Kraus et al., 2010). Previous studies demonstrated that HIV-1 optimizes Env expression throughout the course of infection to increase viral infectivity and transmission while minimizing antibody neutralization and immune activation (Parrish et al., 2013; Krapp et al., 2016). Thus, the expression of Rev1-Vpu may be merely an epiphenomenon of adaptive changes modulating Env expression. This hypothesis is in agreement with the description of several regulatory elements in the rev1/vpu region that may modulate leaky scanning and/or induce ribosomal shunting (Schwartz et al., 1990b; Anderson et al., 2007; Krummheuer et al., 2007). Together with the observation that the proportion of rev1-vpu encoding viruses does not seem to increase over time (unpublished data), these findings strongly suggest that the fusion gene itself has a neutral phenotype.

The generation of fusion proteins represents a mechanism of increasing the coding potential of viral genomes and has been identified in diverse viruses including primate lentiviruses, foamy, and papilloma viruses (Lambert et al., 1989; Viglianti et al., 1990; Lindemann and Rethwilm, 1998). By literally piecing together functional domains of different proteins, viruses may generate fusion products that retain or regulate the activity of their parental proteins and/or even exert entirely novel functions. Although HIV-1 is among the best characterized viruses, relatively little is known about its “fuseome,” i.e., the entity of all viral fusion genes and proteins. To date, more than a dozen non-canonical lentiviral fusion proteins have been described (Table 1). While two of them, Tat-T/Vpt and Rev1-Vpu, are the result of ribosomal frameshifting and genetic rearrangements, respectively, the remaining ones are expressed due to alternative splicing events. Although for some of them, expression has been confirmed on both mRNA and protein levels, a relevant role for all of these fusion proteins in lentiviral replication remains questionable for several reasons: (1) Most of the fusion proteins were only identified in a few laboratory-adapted viruses. For example, cryptic exon 6D, which is required for the generation of TNV/p28^tev and p18^6Drev, has only been described for HIV-1 HXB2 and a few closely related viruses (Feinberg et al., 1986; Wright et al., 1986; Benko et al., 1990; Salfeld et al., 1990; Schwartz et al., 1990a; Göttlinger et al., 1992; Neumann et al., 1994; Wentz et al., 1997). Follow-up studies including in vivo transcriptome analyses of patient-derived cells failed to detect 6D transcripts in other HIV-1 strains and subtypes suggesting that they might represent an artifact of laboratory-adapted viruses (Furtado et al., 1991; Smith et al., 1992; Purcell and Martin, 1993; Vega et al., 2016). To our knowledge, Rev1-Vpu is the only unusual fusion protein known to be expressed by intact primary isolates of HIV-1 (Langer et al., 2015). (2) The total cellular levels of many fusion proteins are very low. P17^tev and Tat-T/Vpt, for example, were detectable upon in vitro translation, but not in transfected or infected T cells (Cohen et al., 1990; Furtado et al., 1991). Similarly, Ref was not detectable by Western blotting unless the cells were treated with a proteasome inhibitor (Ocwieja et al., 2012). (3) Although some fusion proteins were shown to exert the activity of their parental proteins, several mutational analyses argue against a crucial role of known fusion proteins in viral replication. Mutation of the splice sites generating exon 6D, for example, revealed that TNV/p28^tev is not required for efficient replication of HIV-1 in CD4+ T cells (Göttlinger et al., 1992). In fact, elevated usage of exon 6D may even be detrimental for viral replication (Wentz et al., 1997). Similarly, the majority of primary HIV-1 isolates seems to do well without a rev1-vpu fusion gene, and gain-of-function mutations did not enhance viral replication in PBMCs or lymphoid tissue (Kraus et al., 2010; Langer et al., 2015).

The observation that fusion proteins are expressed only by a fraction of HIV-1 strains and may be dispensable for viral replication in vivo strongly suggests that their expression is just a tolerated epiphenomenon of other adaptive changes. In line with this hypothesis, several studies suggested that cryptic splice sites such as A6 may have evolved to stabilize adjacent suboptimal splice sites and/or increase mRNA stability to balance the ratio of spliced and unspliced HIV-1 transcripts (Lu et al., 1990; Haseltine and Wong-Staal, 1991; Göttlinger et al., 1992; Lützelberger et al., 2006). Furthermore, novel splice sites may also be an (inevitable) result of adaptive changes in regulatory RNA elements, such as shunting structures or Kozak sequences. Mutations generating a rev1-vpu fusion gene, for example, have been shown to drastically affect env expression (Langer et al., 2015). Finally, fusion proteins may evolve to compensate for detrimental mutations elsewhere in the genome. One striking example has been described by the Berkhout lab, where a Tat-Rev fusion protein evolved to compensate for a mutation of the rev initiation codon (Verhoef et al., 2001). The observed Tat-Rev fusion comprised all domains of Rev and allowed the virus to replicate almost as efficiently as the respective wild type control.

No matter whether HIV-1 fusion proteins represent beneficial helpers, neutral factors or even detrimental byproducts, all of them may potentially be immunogenic and serve as T cell epitopes and/or antibody binding sites. To better assess their relevance for viral replication and immune activation, it is therefore crucial to investigate viral mRNA and protein expression in a broad and unbiased manner. Since viral gene expression seems to depend on the cell type and the viral strain (Ocwieja et al., 2012; Vega et al., 2016) rather than the time point of infection (Saltarelli et al., 1996), it is especially important to perform analyses in primary target cells infected with diverse groups and clades of HIV-1. For example, the recent pyrosequencing of CD4+ CD25+ lymphocytes from individuals infected with non-B subtypes revealed that the diversity of splice site usage and the expression of non-canonical transcripts is substantially higher than previously anticipated (Vega et al., 2016). Furthermore, Ocwieja and colleagues hypothesized that cryptic splice donor sites near the 3′ end of the viral RNA such as SD8955 or D6 may also be joined with adjacent exons of the host and result in the expression of chimeric viral-host proteins as previously described for self-inactivating (SIN) retroviral vectors (Almarza et al., 2011; Ocwieja et al., 2012). Remarkably, even defective proviruses that fail to produce infectious viral particles have recently been shown to express RNA species with unusual exon combinations (Imamichi et al., 2016). Due to large (intron) deletions, these unspliced RNAs may be exported from the nucleus in a Rev/RRE-independent manner, where they are predicted to produce truncated and/or chimeric viral proteins.

Thus, it is very likely that the lentiviral fuseome will further increase, and future analyses will reveal whether some HIV-1 strains express non-canonical fusion proteins with relevant functions in vivo.

SL and DS wrote this review article.