Sequence-specific Sensing of Nucleic Acids

Abstract

Innate immune cells are endowed with many nucleic acid receptors, but the role of sequence in the detection of foreign organisms remains unclear. Can sequence patterns influence recognition? And how can we infer those patterns from sequence data? Here, we detail recent computational and experimental evidence associated with sequence-specific sensing. We review the mechanisms underlying the detection and discrimination of foreign sequences from self. We also describe quantitative approaches used to infer the stimulatory capacity of a given pathogen nucleic acid species, and the influence of sequence-specific sensing on host-pathogen coevolution, including endogenous sequences of foreign origin. Finally, we speculate how further studies of sequence-specific sensing will be useful to improve vaccine design, gene therapy and cancer treatment.

Microbial-specific Sequence Motifs are a Class of Pathogen-associated Molecular Pattern

The innate immune system detects the presence of foreign organisms and initiates a coordinated response to eliminate infectious threats. Among the microbial products sensed by innate immune effectors, efficient recognition of nucleic acids (DNA and RNA) is critical, as suggested by the existence of several families of receptors specific to these ligands (Table 1). However, the detection of nucleic acids also presents a risk for self-recognition, and self-activation in response to host nucleic acids is associated with many autoimmune diseases. Inappropriate detection of self-molecules is prevented through subcellular compartmentalization of receptors, degradation of self nucleic acids by endogenous nucleases, and specialization of innate receptors that detect conserved microbial features absent from the host. Typical microbial features recognized by the innate immune system include distinct products of post-transcriptional processing, such as nucleotide chemical modifications and specific 5′- or 3′-moieties. In addition, microbial genomes display specific nucleic acid patterns at a different frequency compared to the human genome. These variations, such as frequencies of nucleotide words, are sufficient to assign biological sequences to taxonomic spaces, at least broadly [1].

Table 1.

Interactions between viral sequence patterns and innate immune pathways deciphered by large-scale synonymous recoding of viral genomes

Virus	Sequence pattern studied	Innate immune pathway involved	Reference
Influenza	(A/U)CG(A/U)	IFN-I secretion, TLR7-dependent	[23]
Simian Immunodeficiency Virus	Nucleotide bias	IFN-I, IRF3-dependent	[28]
Echovirus	CpG and UpA	Unknown PRR, IRF3-independent	[24]
Infuenza	CpG and UpA	Pro-inflammatory cytokines Unknown PRR	[25].
Foot and Mouth Disease Virus	Codon pair bias	IFN-I Pro-inflammatory cytokines	[26]

As a result, coding and non-coding transcripts pertaining to a single “self” genome display specific sequence properties that can be used to distinguish them computationally from a “non-self”, foreign, genome (Fig. 1) [2]. For example, differential CpG and UpA usage patterns create a metric that computationally segregates the genomes of some human viruses, including positive-, negative-, and double-stranded RNA viruses, DNA viruses and, interestingly, many transcripts only found in high abundance in tumors, into taxonomic spaces that are distinct from the typical human transcriptome (Fig. 1). Thus, differences in sequence patterns theoretically can be used by host cells to distinguish self sequences from non-self.

Figure 1. CpG and UpA Dinucleotide Distributions Among Different Classes of Transcripts.

The distribution of forces on CpG and UpA dinucleotides among non-coding and coding RNA allows one to define a landscape of human transcripts (gray dots). Forces represent the entropy penalty for non-random motif usage, with positive forces indicating over-representation and negative under-representation of the motif. Among this landscape, a particular space is occupied by human coding sequences (yellow dots, each ellipse indicates 1 SD from the mean, 95% of transcripts are distributed inside the median ellipse). The same metric applied to human viruses shows that viral genomes occupy a restricted space, yet most human viruses mimic their human host to within two standard deviations, with a set of exceptions, particularly among the dsRNA viruses. Interestingly, several non-coding RNA upregulated in pancreatic and other cancer cells identified in [77] are clear outliers from normal CpG and UpA use (red dots). This figure is an agglomeration of sequences and analyses from [5, 76, 77].

Here, we describe both the computational approaches to inferring a foreign sequence and the sequence specificity of receptors involved in innate sensing of nucleic acids. We discuss the newly described importance of sequence-specific sensing on the immunostimulatory properties of endogenous elements of foreign origin, along with other mobile elements and non-coding RNAs, and their emerging role in cancer immunity.

Computational Approaches to Define a Foreign Sequence

The earliest computational approaches to defining a foreign sequence used motif usage as a self versus non-self discriminant. A pioneering study showed that a metric based on dinucleotide usage could adequately discriminate a random contiguous segment of human DNA from a random contiguous segment of bacterial DNA [3]. The primary dinucleotide that enforces this discrimination is CpG. Indeed, CpG methylation of DNA, a feature vastly more prevalent in mammals than bacteria, promotes cytosine deamination, leading to accelerated cytosine to thymine mutation. Thus, CpG methylation is thought to account for underrepresentation of CpG motifs in mammalian genomes.

A few years earlier, it was found that a pattern recognition receptor (PRR), Toll-like receptor 9 (TLR9) senses DNA containing unmethylated CpGs in the endosomes of mammalian cells [4]. Hence, the same pattern detected computationally to discriminate mammalian and bacterial genomes is associated with an actual innate receptor performing the same task. Subsequently, new methods were introduced to computationally identify such discriminatory motifs. One approach introduced further constraints on genomes, fixing not just nucleotide frequency but also the coding sequence of a virus and its codon usage. By randomizing the genome sequence while fixing these constraints, it is possible to identify patterns that discriminate pathogen and host genomes, independently of their effect on the codon usage or amino acid sequence [5].

The more constraints introduced, the more computationally taxing randomization becomes. These approaches thus become ill suited to longer genomes or whole transcriptome analyses, where the size of the dataset requires more efficient methods. Motivated by this issue, the problem was recently recast using methods from statistical physics, which avoid tedious genomic randomizations to find atypical patterns [6]. The computational speed associated with this improved conceptual language allows larger datasets, such as eukaryotic transcriptomes, to be efficiently analyzed. In this framework, one defines a force on a genomic pattern as the information entropy cost for a genome to have that pattern at the observed frequency. In other words, a pathogen pays a measurable evolutionary cost to reorganize the information in its genome in a non-random way in order due to the influence of specific sequence patterns targeted by a PRR, reaching an equilibrium with its host that balances between the selective forces on the pattern and randomizing entropic forces. As a consequence when a pathogen changes hosts, potentially encountering a new set of PRRs, it will evolve towards a new equilibrium between PRRs and sequence entropy.

Microbial Avoidance of Sequence Patterns Suggest Immunostimulatory Properties

The first demonstration of sequence-specific immune sensing of foreign nucleic acids came from the discovery that CpG motifs in bacterial DNA trigger B-cell activation [4]. The same immunostimulatory potential of certain dinucleotides was suggested by genomic analyses that identified a pervasive suppression of CpG and UpA dinucleotides in RNA viruses infecting mammals [7] [8]. Later, an evolutionary survey of influenza A upon transition from avian to human hosts suggested a selection pressure against CpG dinucleotides in A/U rich contexts [5]. Interestingly, this motif pattern is highly under-represented in the sequences of several human innate immune genes, such as the type I interferon (IFN-I) family. These RNA, expressed at high quantity almost exclusively during the response to viral infection, may have evolved to avoid further immune stimulation, which could create a positive feedback loop [9]. Similarly, in a study comparing the dinucleotide usage patterns of Flaviviridae viruses infecting vertebrate or invertebrate organisms, the vertebrate-infecting viruses display a decreased CpG frequency not observed in insect-only viruses [10].

Nucleotide word usage is not the only sequence feature that has been associated with sequence-specific immune detection. Yet another feature comprises large secondary RNA structures observed in several RNA virus families. Their presence has been linked to viral fitness, in vivo persistence and escape of innate immune detection [11]. Furthermore, as for many viral families, lentiviruses, such as Human immunodeficiency virus (HIV), possess a genome with a significant nucleotide ratio bias compared to human genes [12]. One hypothesis explaining this biased composition is that, as a retrovirus, HIV is subject to innate pressure from RNA editing enzymes such as APOBEC3G, which can edit cytosine nucleic acids in a motif specific fashion, inducing a mutational bias [13]. Likewise, an analysis of different HIV-1 virus subtypes has uncovered a correlation between HIV-1 nucleotide bias, its ability to elicit an innate immune response, and its pathogenicity [14]. Thus, while there is evidence that some specific immunostimulatory motifs are being selected against, microbial sequences also evolve under both intrinsic and host-related constraints, leading to some pattern usage biases not necessarily identical to those of their hosts.

Large-scale Synonymous Recoding of Viral Genomes Indicates Innate Detection of Specific Sequences

A popular approach for studying the interaction of the immune system with viral sequence patterns is the large-scale synonymous recoding of viral genomes [15]. Taking advantage of the genetic code’s degeneracy, this strategy consists of replacing whole portions of viral genomes with sequences either enriched for or depleted of the pattern of interest, without altering the protein coding sequence. Pioneering studies introduced rare codons in the capsid-coding region of poliovirus genome and observed that it led to viruses with altered replication fitness, lower virulence and infectivity [16] [17]. Similar strategies, including alterations in codon usage, codon pair usage, and/or dinucleotide content also led to viral attenuation, the mechanisms of which are currently under investigation. Depending on the studies, authors have thus far reported inhibition of protein synthesis [17] [18] [19] [20], inhibition of viral cDNA synthesis [21] or reduction of virion infectivity [17] [22]. Interestingly, several of these investigations, listed in Table 1, found that modification of viral genomes modulates their interactions with the innate immune system. Of note, a team generated two non-replicative influenza virus mutants in which the viral PB1 polymerase gene was replaced by a green fluorescent protein (GFP) gene. One virus contained a GFP sequence with 21 (A/U)CG(A/U) motifs, while the other contained only 2. Upon infection of human primary plasmacytoid dendritic cells, the mutant with more CpG motifs elicited higher levels of interferon-α [23].

The immune impact of modifying CpG and UpA dinucleotide frequencies in viral genomes was later confirmed by additional studies. In one study, increasing frequency of CpG and UpA dinucleotides in the genome of the picornavirus echovirus 7 produced mutants with impaired replication kinetics. The authors found no evidence for differential recognition of the virus by an already known PRR. However, experimental use of kinase inhibitors reverted the attenuation phenotype of the mutants, suggesting a role for an unknown PRR in mediating a dinucleotide usage-specific immune response [24]. In a follow-up study on influenza virus, the authors demonstrated that, despite marked attenuation of replication, virus mutants enriched in CpG and UpA induced higher inflammatory cytokines and adaptive responses in vivo [25].

Another team generated a replicative mutant of Foot-and-Mouth Disease Virus, by replacing codons pairs by synonymous codon pairs underrepresented relative to the human genome. The mutant exhibited an increase in its overall CpG content resulting in attenuated replication in vitro and higher levels of type I interferon, IFN-stimulated genes (ISG) and proinflammatory cytokines [26]. A different team produced a mutant of simian immunodeficiency virus (SIV) with optimized nucleotide composition by lowering nucleotide bias in the pol gene, resulting in increased CpG frequency, and overall reduction of dinucleotide bias [27]. Optimized SIV mutant displayed normal replication kinetics but a decreased ability to induce IFN-I expression. Moreover, IFN regulatory factor 3 (IRF3) was required for IFN-I induction, suggesting a role for cytosolic sensors in detecting SIV biased RNA [28].

Sequence Specificity of Known Human Receptors

Several nucleic acid sensors have been proposed or suspected to sense specific microbial sequences, as listed in Table 2. We categorize these receptors depending on two aspects. The first is subcellular localization, since one would hypothesize that sensitivity of PRRs strongly depends on spatial proximity to self-derived ligands. According to this hypothesis, nucleic acid receptors located in endosomes and cytoplasmic DNA sensors should not bind ligands in a highly sequence-dependent manner, and, cytoplasmic RNA PRRs would have developed higher specificity for foreign sequences. This does not always hold true. The second aspect is their sequence specificity and the techniques used to infer them.

Table 2. Known receptors that sense specific nucleic acid sequences.

We listed each receptor according to the nature of its ligand, the sequence it recognizes, its subcellular localization and the methods used to determine the sequence specificity. NGS: Next Generation Sequencing. EMSA: Electrophoretic Mobility Shift Assay.

	Receptor	Sequence/pattern	Method of determination	Localization	Reference
RNA	RIG-I	AU-rich sequence and dsRNA structures	NGS and crystal structure	Cytosol	[41] [42] [43] [48] [49] [50]
	MDA5	AU-rich sequence and dsRNA structures	NGS and crystal structure	Cytosol	[48] [49] [51]
	DDX17	CA- and CT-repeat, (GTA)CATCC(CTA) motif and stem-loop structure	NGS	Cytosol and Nucleus	[53] [54]
	IFIT 2	AU-rich ssRNA sequence	Crystal structure and EMSA	Cytosol	[52]
	TLR7	U(C/G)U or U(U/A)N and (A/U)CG(A/U) motifs. G residues and U-rich ssRNA	Cytokine measure after stimulation	Endosome	[23] [32] [34]
	TLR8	U-rich ssRNA degraded into U residues and short oligonucleotides	Crystal Structure	Endosome	[33]
	TLR13	GAAAGACC and stem- loop-like structure	Crystal Structure	Endosome (mouse)	[81]
DNA	TLR9	RRCGYY	Crystal structure	Endosome	[38]
	cGAS	Short dsDNA flanked with at least 3 G	Cytokine measure after stimulation + receptor binding assay	Cytosol	[57]
	Sox2	(C/T)(A/T)TTGT(T/C)ATG CAAAT	Cytokine measure after stimulation + receptor binding assay	Cytosol	[58]

Nucleic Acid-sensing Endosomal TLRs

TLR3, TLR7, TLR8, TLR9 and TLR13 (Box 1) are endosomal receptors that sense nucleic acids that have been endocytosed or phagocytosed. The crystal structures of most TLRs bound to their respective ligands have been deciphered. TLRs have a ligand-recognition domain facing the lumen of the endosome, a transmembrane domain, and a signaling domain facing the cytosol. Despite compartmentalization, TLR7, TLR8 and TLR9, but not TLR3, demonstrate a certain level of sequence specificity.

Box 1. The Intriguing Specificity of Mouse TLR13.

TLR13 is an endosomal TLR expressed in mice but not humans. TLR13 recognizes its ligand in a stringently sequence-specific manner, sensing a highly conserved bacterial 23S rRNA sequence that contains 5′-GAAAGACC-3′ [82] [83]. Notably, this sequence is found within a region of RNA targeted by certain antibiotics, and clinical isolates of Staphylococcus aureus resistant to these antibiotics are unable to stimulate mouse TLR13 [83]. The structure of mouse TLR13 bound to a 13-nt ssRNA derived from 23S rRNA has also been characterized. When bound to TLR13, the ssRNA folds into a stem-loop-like structure that is different from its usual shape in the bacterial ribosome, and that is also required for activation of TLR13. However, most of the ssRNA nucleotides make base-specific contacts with the surface of TLR13, accounting for its sequence specify [81]. Interestingly, a viral-derived 16-nt ssRNA, containing the related sequence GAAAGACU and predicted to form a similar stem-loop-like structure also induces TLR13 activation [81]. Overall, TLR13 functions as a sequence- and conformation-specific PRR.

TLR7 and TLR8 recognize ssRNA and share a general specificity for U-rich ssRNA [29] [30]. A study showed that a uridine-rich tetramer is the minimal sequence required to elicit TLR7 and TLR8 responses and further suggested TLR8 has increased specificity for AU-rich sequences, while TLR7 has a comparatively heightened specificity for GU-rich sequences [31]. A second investigation found that endogenous nucleic acid sequences induced a level of TLR7 activation in murine B cells directly proportional to their uridine-content, and, particularly, to the frequency of two motifs, U(C/G)U and U(U/A)N [32]. The crystal structure of TLR8 in complex with ssRNAs showed that the receptor contains two ligand-binding sites and that synergistic binding of both sites to their respective ligand is required for signal transduction. The first site binds a uridine residue and the second binds a short guanosine-containing oligoribonucleotide, like UG [33]. After crystallization, intact ssRNA was not observed bound to the TLR8 structure, suggesting TLR8 binds products of ssRNA degradation catalyzed by yet unknown nucleases and phosphatases [33]. A follow-up study performed on TLR7 showed that it also contains two ligand-binding sites. Indeed, TLR7 acts as a dual receptor that binds both guanosine and uridine-containing ssRNA. Interestingly, the ssRNA-binding site is spatially and structurally distinct form that of TLR8. However, as for TLR8, synergistic binding of both ligands is required for efficient activation [34]. Interestingly, use of guanosine derivatives such as Resiquimod (R848) shows chemicals with a minimal nucleotide structure can act as agonist of both TLR7 and TLR8 [35].

TLR9 recognizes bacterial and viral ssDNA containing an unmethylated CpG motif with a strong level of sequence specificity. Indeed, depending on its sequence, a DNA strand will act either as TLR9 agonist or antagonist [36]. The TLR9 immunostimulatory CpG motif, which consists of a hexamer with a central unmethylated CpG, was first described as RRCGYY, with R representing a purine and Y a pyrimidine [4]. In humans, the CpG motif with formula GTCGTT has been proposed as an optimal ligand [37]. The crystal structure of TLR9 has been characterized. Using agonist 12-mer oligonucleotides, it was found that the consensus hexamer sequence, and no other part of DNA, is directly recognized by TLR9 [38]. TLR9’s specificity for CpG motifs and their flanking sequences is determined by precise van der Waals’ interactions and water-mediated hydrogen bonds between multiple TLR9 amino acids and DNA bases. Of note, TLR9 recognizes CpG in DNA sequence only. However, CpG dinucleotide are also suppressed in many RNA viruses, which do not produce DNA at any step of their replication cycle. Thus, it’s likely that TLR9 specificity drives CpG suppression in DNA viruses for example, but is unlikely to be involved in ssRNA and dsRNA viruses genome shaping. Additionally, the backbone phosphate is needed for recognition. In another study, TLR9 has been shown to recognize DNA:RNA hybrids [39]. Curiously, ssDNA fragments isolated from these hybrids were not potent in activating TLR9, presumably due to the absence of CpG motifs. Overall, these results suggest an intriguing dependency of TLR9 sequence specificity to the nature of its ligand.

RIG-I-like Receptors

RIG-I

The best-studied cytosolic RNA sensors are retinoic acid-inducible gene I (RIG-I) and melanoma differentiation factor 5 (MDA5). RIG-I primarily relies on structural features like 5′ triphosphate extremities and dsRNA fragments to detect viral RNA [40]. Despite the high-resolution data of crystal studies, a consensus on the definition of a RIG-I ligand has yet to be reached [41] [42] [43]. RIG-I has been shown to recognize specific sequences of viral genomes, such as poly-U/UC motifs found in the 3′ untranslated region of hepatitis C virus and in the N gene of Hantaan virus [44] [45]. Several studies have also demonstrated that manipulation of RNA ligand sequences, and notably the number of uridine residues, modulate activation [46] [47]. Moreover, precipitation of cellular proteins bound to viral RNA during infection suggests that RIG-I preferentially associates with AU-rich RNAs derived from viral genomes [48] [49] [50]. As expected from its crystal structure determination, RIG-I additionally favors sequences that can form dsRNA structures, such as the sequence in the 3′-UTR of Chikungunya virus and in sequences of RNAs found in Measles, Sendai or Influenza virus defective interfering particles [48] [49] [50].

MDA5

MDA5 binds dsRNA independently of its terminal moieties and, according to structural studies, without any direct sequence specificity. Indeed, MDA5 interacts primarily with the phosphate backbone and 2′ hydroxyl groups of ribose in dsRNA, using it as a platform to stack along dsRNA in a head-to-tail arrangement. The cooperative binding of MDA5 on longer dsRNA increases its affinity, eventually leading to the proper activation of downstream signaling pathways [51]. However, next-generation sequencing (NGS) studies performed in the context of viral infection have revealed that, like RIG-I, MDA5 associates preferentially with AU-rich viral sequences [48] [49].

Other Cytosolic Sensors

IFIT2

Interferon-induced proteins with tetratricopeptide repeats 2 (IFIT2) can directly bind viral RNAs and exert antiviral functions such as inhibition of viral translation and activation of anti-viral signaling pathways. Crystal structure analysis coupled to an electrophoretic mobility shift assay indicates IFIT2 possesses a preference for AU-rich RNAs [52].

DDX17

DEAD-Box Helicase 17 (DDX17) is a multifunctional helicase that binds stem-loop structures of viral RNA in the cytosol of infected cells [53]. In the nucleus, DDX17 binds both CA- and CT-repeat elements found in mature cellular mRNAs and the (GTA)CATCC(CTA) motif found in miRNAs [54]. Thus, DDX17 uses both primary sequence and secondary structure for optimal binding to ligands (see also Box 2).

Box 2. Sequence and Structure: Patterns Inherently Intertwined.

The structure of nucleic acids depends on base-pairing interactions, and therefore is inherently linked to sequence. For instance, GC-rich RNAs tend to fold into more stable structures than AU-rich RNAs, simply because a GC pair has three hydrogen bonds whereas an AU pair only has two. Thus, GC-rich sequences may form robust secondary structures whereas AU-rich fragments should have higher flexibility [84]. In viral genomes in particular, studies have specifically correlated nucleotide composition with secondary structure. In lentiviruses, the preference for adenosine that characterizes this family is increased in single-stranded domains, but absent in double-stranded domains [12]. Similar observations have been made for Flaviviridae [85] and Coronaviridae [86].

This link between the sequence and structure bears significance for several PRRs, which depend on molecular recognition of RNA structures for immune sensing. This is the case for TLR3, which was initially identified as a dsRNA receptor [87]. It has been shown that dsRNA must meet a minimum length of 40 base pairs to activate TLR3 [88]. The crystal structure of human TLR3 complexed to dsRNA demonstrates that TLR3 recognizes the phosphate backbone of dsRNA, but makes minimal contacts with bases, confirming sequence-independent sensing [89]. Yet, several reports have since suggested RNA structures other than perfect stretches of dsRNA activate TLR3, including structures formed by bacterial and host ssRNAs [90]. These studies indicate that specific sequences may form self-RNA duplexes recognized by TLR3. Similarly, Protein Kinase R (PKR) is a PRR activated by several distinct RNA secondary structures including dsRNA, short stem-loop RNAs flanked by single-stranded tails, and misfolded RNA with structural defects, like the bulges and internal loops often found in microbial transcripts [91].

The study of immunostimulatory structures has been limited by technical challenges to determining the secondary and tertiary structures of long RNA molecules in vivo. Recently, the development of several high-throughput techniques has enabled more accurate predictions on how RNA folds inside the cell (Reviewed in [92]). These predictions propose that, contrary to microbial RNAs, cellular self-RNAs are less folded than primarily thought, strengthening the idea that PRRs discriminate pathogens based on structures adopted by their nucleic acids.

cGAS

Crystal studies show that cyclic GMP–AMP synthase (cGAS) senses cytosolic dsDNA predominantly by binding to its sugar–phosphate backbone, suggesting sequence-independent innate sensing [55] [56]. In contrast, additional work on DNA structures observed in early reverse transcripts of HIV-1 as well as endogenous retroviral elements, such as the SL2 stem loop located in the 5′ extremity of HIV’s genome, demonstrated ssDNA stem-loop structures flanked by unpaired guanosines activate cGAS in a sequence-dependent manner [57].

Sox2

Sox2 is a transcription factor that has also been shown to act as a cytosolic dsDNA receptor. To detect microbial DNA, Sox2 binds to sequence motifs in bacterial genomes such as L. monocytogenes. Interestingly, these sequences are similar to the endogenous DNA motifs Sox2 binds when acting as a transcription factor [58].

Sequence Patterns and RNA Decay

Molecular stability is another property linked to nucleic acid sequence (see also Box 3). Unique sequence elements found at the 3′-end of cellular mRNAs can mediate their degradation, such as AU-rich elements (AREs), that are characterized by a tandem repeat AUUUA sequence or a simple U-rich region [59]. AREs are observed in the sequence of many mRNAs related to the immune response and binds specific proteins, such as AU-binding factor 1 (AUF1), which recognizes AREs and targets mRNAs for rapid degradation. [60]. Interestingly, AUF1 can also directly target regions of viral RNA and inhibit viral replication, as observed during enterovirus and human rhinovirus infection [61].

Box 3. Sensing the Codon Bias?

In addition to nucleotide and dinucleotide bias, most human pathogens display a strong codon bias compared to average human codon usage. A study recently performed in yeast shows that the DEAD-box protein Dhh1p (an ortholog of human helicase Ddx6) is able to sense codon bias in cellular mRNA and induce a selective decapping and decay of biased mRNA [93]. The authors propose a mechanism where Dhh1p preferentially binds to mRNAs with high codon bias, owing to an additional interaction of Dhh1p with ribosomes, which are known to stall and accumulate during the translation of non-optimal codons stretches. According to the authors of this study, Dhh1p would thus be able to sense ribosomal speed and induce the selective degradation of codon-biased mRNA. These results echo back an older study showing that Schlafen 11, an interferon stimulated gene, could restrict retroviral infection in a codon-usage-dependent manner, this time by modulating the cellular tRNA levels during HIV infection [94]. Given the strong involvement of DEAD-box proteins in innate immunity and their ability to interact with several innate immune adaptors [95], analyzing the role of Ddx6 and its homologs in sensing microbial sequences may unravel interesting mechanisms linking codon adaptation and pathogen sensing.

Another example of a cellular intrinsic defense mechanism preventing translation of foreign genetic information is the zinc-finger antiviral protein (ZAP). ZAP binds specifically to viral RNAs containing a ZAP responsive element (ZRE) and subsequently recruits cellular RNA decay machinery [62]. The exact nature of ZRE remains to be identified. However, the resolution of the crystal structure of the ZAP RNA binding domain suggests target RNAs adopt a tertiary structure, with certain nucleotides positioned to fit into a three-dimensional cleft formed by ZAP. Interestingly, the authors suggest that the target nucleotides may not be found in a consecutive linear sequence but rather from different regions of the target RNA, accounting for difficulty in identifying a common ZRE motif [63].

Finally, the well-studied ribonuclease L (RNaseL) bridges RNA degradation and innate activation. During viral infection, RNAse L degrades host and viral RNA, preventing viral replication. Interestingly, the degradation products of RNAse L serve as RIG-I ligands, generating an IFN-I response [64]. A structural analysis suggest that RNase L recognizes the pattern UN^N, and cleaves 3′ of UN sequences [65].

Endogenous Silenced Elements: Sequences and Consequences

Overall, sequences of nucleic acids influence their structure, stability and recognition by cellular receptors. Interestingly, many human sequences, which are not transcribed under homeostatic conditions, contain sequence patterns not observed in the rest of human transcriptome (Fig. 1). Recent analysis of their interaction with the innate immune systems has unraveled unexpected new roles for these silenced elements.

During the evolution, sequences of foreign origin, predominantly stemming from retroviruses, have invaded and colonized the human germ line. Once integrated in host DNA, these sequences may have amplified their copy numbers through rounds of reinfection until eventual fixation [66]. Some retroelements, particularly recent integrations, have retained viral characteristics, like sequence-specific features of viral genomes. Under homeostatic conditions, many of these elements are transcriptionally repressed, likely preventing auto-immunity. This involves primarily epigenetic mechanisms such as histone modification and DNA methylation [66]. However, rupture of cellular homeostasis could lead to reactivation, and evidence that this process leads to the transcription of immunostimulatory ligands is beginning to surface (Fig. 2).

Figure 2. A Role for Microbial Sensing Pathways in Detecting Neoplasia.

During episodes of ancient infection, sequences of viral origin have been integrated into the cellular genome and subjected to active epigenetic silencing. This generally coincides with mutational decay that induces functional inactivation. However, certain sequence-specific characteristics of their viral origin may have been maintained over evolutionary time. Cell transformation or immune activation results in the transcriptional reactivation of these sequences. They may then engage receptors of the nucleic acid sensing pathways, leading to the initiation of an antimicrobial innate immune response such as production of IFN-I and pro-inflammatory cytokines. RT: Reverse-Transcriptase. IN: Integrase.

Reactivation of Transposable Elements Is Triggered and Sensed by Innate Immunity

Several studies have shown that the induction of endogenous retroelement transcription activates innate immune pathways. Reactivation of these normally silenced elements has been associated to autoimmune diseases [67]. Studies on Aicardi–Goutières syndrome (AGS) have linked the excessive IFN-I response characteristic of this condition to mutations in nucleases such as TREX1 and RNAseH. These mutations may result in inappropriate accumulation of nucleic acids and subsequent activation of RNA and DNA sensors such as MDA5 and cGAS [68] [69]. In the mouse, artificial activation of Long Interspersed Element-1 (LINE-1) has been shown to increase the expression of IFN-I and ISGs [70]. Interestingly, silenced endogenous sequences can be activated by innate immune responses, suggesting a role for these sequences in signal amplification. For instance, exposure to environmental microbes causes global modulation of endogenous retroelement transcription [71]. Moreover, treatment with IFN-I induces transcription of the short interspersed element (SINE), Alu, which in turn stimulates the secretion of proinflammatory cytokines in a TLR7-dependent manner. The authors of this study were able to identify a specific Alu sequence motif, normally bound to RNA binding protein Ro60, as immunostimulatory [72]. Similarly, activation of the B cell response with T cell-independent type 2 antigens induces transcription of endogenous retrovirus (ERV) RNAs, which are then detected via DNA and RNA sensing pathways [73]. Interestingly, this mechanism can be hijacked by viral infections. Of note, Herpes virus infection leads to the expression of SINE elements, enhancing viral replication and gene expression through activation of the antiviral NF-κB pathway [74]. Thus microbial activation of retroelement expression may function to further alert and amplify the presence of an invading pathogen.

Transcription of Silenced Sequences in Cancer

In addition to infection, the innate immune response can be triggered through expression of endogenous retroelements in tumors. Cell transformation coincides with complex, genome-wide alteration of the epigenetic landscape [75]. Treatment of certain tumor cell lines with inhibitors of DNA methyltransferase induces ERV demethylation and transcription, which correlates with an elevated IFN-I response dependent on RNA-sensing pathways [76] [77]. One is left to wonder how these endogenous retroelements sequences are distinguished from self. Although some of them can differ in sequence from self-RNA (Fig. 1), the connection of specific viral sequence patterns embedded in retroelements to their observed immunostimulatory activity has yet to be established. One particular subclass of pericentromeric repeats, the human repeat human satellite repeat II (HSATII), is greatly overexpressed in certain cancers [78]. It has been shown that HSATII sequences are enriched in the motif pattern of CpG dinucleotides in AU-rich contexts, whose detection induces the production of proinflammatory cytokines [79]. In addition to their atypical dinucleotide composition, pericentromeric repeats display other similarities with retroviral RNAs, including their ability to be reverse-transcribed [80]. Altogether, these results strengthen the proposition that endogenous retroelements bear functional roles in initiating immune responses during cancer development.

Concluding Remarks and Future Perspectives

The mammalian immune system relies on a large array of nucleic acid sensors. In recent years, increasingly refined computational studies of host and microbial genomes have greatly improved our understanding of the features that allow the distinction of self and non-self sequences. In parallel, biochemical and structural analyses of nucleic acid sensors have unraveled various degrees of sequence specificity.

Here, we reviewed both computational and experimental evidence that innate immunity restricts sequence landscapes by targeting specific sequences and sequence patterns primarily found in pathogens. These advances will likely be critical for numerous applications, including the design of optimal nucleic acid adjuvants for use in vaccines. Conversely, only a complete understanding of what constitutes a homoeostatic human transcriptome will allow the design of nucleic acid molecules devoid of immunostimulatory capacity. These should be required for gene therapy and mRNA therapy treatments. Further, we can envision the synthesis of nucleic acid sequences tailored to elicit a specific innate immune response, in order to treat particular conditions in immunotherapy (see Outstanding Questions).

Outstanding Questions Box.

• Can we establish a comprehensive model of the human genome and its transcriptional landscape, sufficient enough to define a “self” -and by exclusion- a “non-self” sequence profile? Under what conditions can that landscape change?

• Are there any additional, unknown innate sensing pathways specialized in sensing foreign sequences? If yes, what are the mechanisms?

• What prevents microbial evolutionary evasion from specific sequences and sequence patterns? Are these constraints one can target?

• Are there evolutionary advantages to maintain or enhance the properties of immunostimulatory sequences in the human genome?

• What epigenetic mechanisms control the expression of immunostimulatory sequences? Can we define an epigenetic signature associated with their transcription?

• Can the expression of immunostimulatory sequences be associated with specific immune markers? For instance, activation of dedicated innate immune pathways and secretion of specific cytokines?

• What are the beneficial contributions of endogenous retroelements in immune activation against infection and cancer development? And what is the clinical outcome associated with their expression?

Finally, it is important to note that some host sequences derived from ancient retroviral infections have maintained characteristics of non-self sequences. Their expression, normally silenced, can be reactivated during inflammation and neoplasia, and may potentiate inflammatory responses. For these reasons, expression of foreign sequence patterns by human cells represents a marker that warrants monitoring. Further work will be required to confidently define what constitutes the landscape of “self” human sequences.

Trends Box.

• Computational analyses of host and pathogen genomes lead to the identification of unique immunostimulatory sequence patterns. In parallel, structural and biochemical analyses of innate receptors confirmed the existence of specificity for unique sequence motifs.

• Large-scale recoding of viral genomes uncovers new interactions between the innate immune system and sequence patterns found in viral genomes.

• The sequence of nucleic acids modulates their structure, stability and direct recognition by cellular pattern recognition receptors.

• Endogenous sequences of foreign origins that are normally silenced can induce innate responses through nucleic acid sensing pathways.

• A better characterization of the innate pathways responsible for sensing foreign sequences will improve computational detection and bring critical insights for gene therapy and vaccine design.