Convergent Losses of TLR5 Suggest Altered Extracellular Flagellin Detection in Four Mammalian Lineages

Abstract

Toll-like receptors (TLRs) play an important role for the innate immune system by detecting pathogen-associated molecular patterns. TLR5 encodes the major extracellular receptor for bacterial flagellin and frequently evolves under positive selection, consistent with coevolutionary arms races between the host and pathogens. Furthermore, TLR5 is inactivated in several vertebrates and a TLR5 stop codon polymorphism is widespread in human populations. Here, we analyzed the genomes of 120 mammals and discovered that TLR5 is convergently lost in four independent lineages, comprising guinea pigs, Yangtze river dolphin, pinnipeds, and pangolins. Validated inactivating mutations, absence of protein-coding transcript expression, and relaxed selection on the TLR5 remnants confirm these losses. PCR analysis further confirmed the loss of TLR5 in the pinniped stem lineage. Finally, we show that TLR11, encoding a second extracellular flagellin receptor, is also absent in these four lineages. Independent losses of TLR5 and TLR11 suggest that a major pathway for detecting flagellated bacteria is not essential for different mammals and predicts an impaired capacity to sense extracellular flagellin.

Toll-like receptor (TLR) proteins are an important component of the innate immune system (Vijay 2018). TLRs comprise a group of membrane-bound pattern recognition receptors and detect pathogen-associated molecular patterns, such as bacterial lipopolysaccharides, peptidoglycans, double-stranded RNA, or flagellin (Vijay 2018). Extracellular flagellin is recognized by TLR5 (toll-like receptor 5) (Hayashi et al. 2001; Mizel et al. 2003; Smith et al. 2003), a receptor that among other tissues and cell types is expressed on the basolateral surface of the intestinal epithelium (Gewirtz et al. 2001; Vijay-Kumar et al. 2008). Upon binding its ligand, TLR5 stimulates the expression of proinflammatory, antibacterial and stress-related genes (Yu et al. 2003; Vijay-Kumar et al. 2008), and hence plays an important role for recognizing pathogenic flagellated bacteria such as Salmonella (Murthy et al. 2004). The flagellin sensor function of TLR5 is likely conserved among vertebrates given that TLR5 of chicken, Anole lizard and rainbow trout also recognizes bacterial flagellin (Tsujita et al. 2004; Iqbal et al. 2005; Voogdt et al. 2016). Importantly, Tlr5 knockout mice do not show an immune response to injected flagellin (Feuillet et al. 2006; Uematsu et al. 2006; Vijay-Kumar et al. 2007), suggesting that Tlr5 is the main extracellular flagellin sensor.

Coevolutionary arms races between pathogens and the host likely explain why pathogen-sensing immune genes frequently evolve under positive selection (Sackton et al. 2007). Consistently, the TLR5 gene experienced positive selection in many primate, mammal and bird lineages, and recurrent positive selection was often detected for sites located near the ligand binding site (Wlasiuk et al. 2009; Alcaide and Edwards 2011; Areal et al. 2011; Smith et al. 2012; Vinkler et al. 2014; Velova et al. 2018). Interestingly, despite its conserved role in detecting flagellin, previous studies revealed that TLR5 is inactivated (lost) in several independent bird lineages (Alcaide and Edwards 2011; Bainova et al. 2014; Velova et al. 2018), in tuatara and in clownfish (Liu et al. 2019). Furthermore, a dominant negative TLR5 polymorphism that introduces a premature stop codon is present in ∼4% of human individuals and has reached a frequency of >20% in some populations (Hawn et al. 2003; Barreiro et al. 2009; Wlasiuk et al. 2009). Therefore, the question arises whether inactivation of TLR5 also occurred in non-human mammals. Here, by analyzing the genomes of 120 mammalian species, we show that TLR5 is lost in four independent mammalian lineages, comprising guinea pigs, Yangtze river dolphin, pinnipeds, and pangolins.

Results

We applied a previously developed approach (Sharma, Hecker, et al. 2018) to screen for gene-inactivating mutations in TLR5 using a genome alignment of 120 placental mammals (supplementary table 1, Supplementary Material online) (Hecker and Hiller 2020). This screen revealed that the single coding exon of TLR5 exhibits stop codon mutations, smaller frameshifting insertions or deletions, or larger deletions in the genome assemblies of 17 species belonging to 14 distinct lineages (gorilla, orangutan, guinea pig, rabbit, pig, alpaca, Yangtze river dolphin, bighorn sheep, Weddell seal, Hawaiian monk seal, walrus, Malayan and Chinese pangolins, little brown bat, elephant, armadillo). Since we have previously shown that apparent gene-inactivating mutations can sometimes be sequencing or assembly errors (Hecker et al. 2017; Sharma, Lehmann, et al. 2018; Hecker, Lachele, et al. 2019; Huelsmann et al. 2019), we carefully validated the correctness of all such putative mutations and performed additional analyses to establish which of these 17 species truly lost TLR5.

First, we used available raw sequencing read data to validate stop codon and smaller frameshifting insertion/deletion mutations. This revealed that frameshifting insertions and deletions that are apparent in the orangutan, rabbit, alpaca, bighorn sheep, little brown bat, elephant and armadillo assemblies are in fact base errors in the assemblies as aligning raw sequencing reads do not validate these putative mutations (supplementary fig. 1A–J, Supplementary Material online). Similarly, sequencing reads show that 1 bp frameshifting deletions in the gorilla (gorGor5) and pig (susScr11) genomes that are assembled from PacBio long sequencing reads (Gordon et al. 2016; Warr et al. 2019) are base errors (supplementary fig. 2, Supplementary Material online). Furthermore, these two base errors are not present in previous assemblies of gorilla (gorGor3) and pig (susScr3), and therefore are likely due to uncorrected errors present in raw PacBio reads (Watson and Warr 2019). In contrast to these nine species, all frameshifts and stop codon mutations present in the domestic guinea pig, Weddell seal, and Pacific walrus were confirmed by several sequencing reads and we found no support for the ancestral, noninactivating allele (fig. 1A, supplementary table 2, Supplementary Material online).

Second, we analyzed the pairwise genome alignments between human and guinea pig, river dolphin, pinnipeds, and pangolins. This confirmed that the remnants of TLR5 occur in a genomic locus with conserved gene synteny (supplementary fig. 3, Supplementary Material online). Furthermore, we found no evidence for the presence of a functional duplicated TLR5 copy in these genomes. Together, this shows that the observed losses of TLR5 are not artifacts arising from aligning paralogous or processed pseudogenes.

Third, since no raw sequencing reads were available to confirm the four TLR5-inactivating mutations observed in the Yangtze river dolphin assembly (fig. 1A), we made use of the recently assembled genomes of the Amazon river and La Plata dolphins, which are sister species of the Yangtze river dolphin (Geisler et al. 2011), and investigated whether any of these inactivating mutations are shared. Indeed, we found that the second stop codon mutation is shared between all three river dolphin species (fig. 1A). The presence of the same mutation in independently sequenced and assembled genomes makes it extremely unlikely that this mutation is a base error. Instead, this shared mutation strongly indicates that TLR5 was already lost in the common ancestor of the three river dolphin species. We also found additional species-specific frameshifting mutations in the Amazon river and La Plata dolphins, which further support that TLR5 is lost in these two species as well (supplementary fig. 4, Supplementary Material online). Similarly, while no sequencing reads are available to confirm mutations in the Hawaiian monk seal assembly, we found that six of seven observed mutations are shared with the Weddell seal (fig. 1A), providing strong support that TLR5 is lost in the Hawaiian monk seal.

To investigate whether TLR5 loss is specific to the domestic guinea pig or whether this gene is also lost in wild guinea pig species, we analyzed the TLR5 locus in the genomes of two additional guinea pigs which are not a part of our genome alignment—Brazilian and Montane guinea pigs. As shown in figure 1A, both stop codons and the 2 bp deletion are shared between all three guinea pig species. Additionally, the large deletion that removes most of the TLR5 coding exon has the same breakpoints in all three guinea pig species (fig. 1B), which makes an assembly error highly unlikely and suggests that this deletion already occurred in their common ancestor. Furthermore, all three guinea pig species share larger structural rearrangements in this locus including several duplications that happened after the deletion (supplementary fig. 5, Supplementary Material online).

TLR5 is entirely deleted in the genomes of the Chinese and Malayan pangolin (fig. 1A). To further investigate whether this deletion could be an assembly error, we generated alignment chains for the Tree pangolin, another recently sequenced pangolin species that was not included in our alignment. We found that TLR5 is also deleted in the tree pangolin and the deletion break points are shared between all three pangolins (fig. 1C), which strongly suggests that the gene deletion already occurred in the common ancestor of the three pangolin species.

The three pinniped species in our data set share several frameshifting deletions (fig. 1A). Since these three species represent only the two of the three basal pinniped lineages, Phocidae (Weddell seal, Hawaiian monk seal) and Odobenidae (walrus), we investigated whether TLR5 is also lost in the third pinniped lineage Otariidae. To this end, we used PCR to amplify TLR5 from tissue samples of the brown fur seal, an otariid seal. Sequencing confirmed that the 25 bp and 17 bp deletion mutations are also present in the brown fur seal (fig. 1D). Furthermore, we found the same mutations in the Antarctic fur seal genome (supplementary fig. 6, Supplementary Material online). Together, the presence of inactivating mutations shared between Otariidae, Phocidae, and Odobenidae suggests that TLR5 loss already occurred in the pinniped stem lineage at least ∼26 Ma (the estimated split of pinnipeds according to TimeTree [Kumar, Stecher, et al. 2017]).

Fourth, we used available transcriptomic data to explore whether there is still expression at the TLR5 locus in the gene loss species. In species that have an intact TLR5 gene, TLR5 expression is detected in several tissues including intestine and lung (supplementary fig. 7, Supplementary Material online). In contrast, we found no evidence for expression in the guinea pig and pangolin (supplementary figs. 8 and 9, Supplementary Material online). However, we did observe clear expression in brain and to a lower extent in lung, placenta and testes at the TLR5 locus in Weddell seal (supplementary fig. 10, Supplementary Material online). Importantly, RNA-seq reads exhibit the frameshifting deletions that are present in the Weddell seal genome assembly, showing that these transcripts cannot encode a functional TLR5 protein. In agreement with previous findings (Sadier et al. 2018), this suggests that the remnants of a lost coding gene can remain to be transcribed, in our case for at least 26 My.

Finally, we used RELAX (Wertheim et al. 2015) to test whether TLR5 evolves under relaxed selection in pinnipeds and the river dolphin where a substantial portion of the gene is still present. We did not apply RELAX to pangolins and guinea pigs where >80% of the gene is deleted. Further supporting TLR5 loss in pinnipeds and Yangtze river dolphin, we found significant evidence for relaxed selection in both lineages (Ka/Ks = 1.17, P value = 0.0002 for Yangtze river dolphin and Ka/Ks = 0.78, P value = 0.0009 for pinnipeds) (supplementary table 3, Supplementary Material online).

With several lines of evidence ranging from signatures of relaxed selection to the presence of validated inactivating mutations, we conclusively show that TLR5 has been lost in four independent mammalian lineages. These mutations either delete the majority or the entire gene (guinea pigs, pangolins) or severely affect the reading frame and domains required for protein function (fig. 1A), suggesting that the remnants of TLR5 cannot encode a functional flagellin-recognition receptor anymore.

Apart from TLR5, mouse Tlr11 has been shown to bind flagellin (Mathur et al. 2012), even though the Tlr11-flagellin interaction is restricted to acidic conditions (Hatai et al. 2016). Therefore, we considered the possibility that TLR5 loss may be compensated by the presence of TLR11, and investigated whether the TLR5-loss species possess an intact TLR11 gene. However, we found that TLR11 is most likely lost in the genomes of guinea pig, river dolphin, pinnipeds, and pangolins (supplementary fig. 11, Supplementary Material online), indicating that this gene cannot compensate for the loss of the flagellin-sensing TLR5.

Discussion

Here, we show that TLR5, the gene encoding the major extracellular flagellin receptor, has been convergently lost four times in mammalian evolution. Our study highlights the importance of carefully validating putative gene-inactivating mutations to detect real gene loss events. All frameshifts detected by our gene loss detection method (Sharma, Hecker, et al. 2018) are really present in the respective assemblies, corroborating the high accuracy of our method; however, our sequencing read validation revealed that many of these frameshifts are actually base errors in genome assemblies that were generated from short or long sequencing reads. While current genome assembly efforts often focus on generating highly contiguous assemblies (Miga et al. 2019), our observations suggest that assembly base accuracy is potentially an underappreciated issue. Indeed, near-perfect assembly base accuracy would be of great importance to automatically analyze genomic data without the need to manually validate putative mutations.

Based on human and mouse studies (Hawn et al. 2003; Feuillet et al. 2006; Uematsu et al. 2006; Vijay-Kumar et al. 2007), loss of TLR5 is predicted to severely impair the capacity to sense extracellular flagellin, in particular because all four TLR5-loss lineages also lack an intact TLR11 gene that may have acted as a compensating factor. An intriguing question is why the evolutionary inactivation of both extracellular flagellin receptors does apparently not have deleterious consequences for guinea pigs, river dolphins, pinnipeds, and pangolins. Similarly, human individuals with TLR5 mutations do not show obvious deleterious phenotypes (Hawn et al. 2003). A potential explanation could be that alternative pathways for flagellin detection compensate for the loss of TLR5 and TLR11. For example, we and others have recently described mammals that lost several human disease-associated genes but do not show the deleterious, disease-resembling phenotypes (Sharma and Hiller 2019; Turakhia et al. 2019), which suggests the existence of either functionally redundant genes or compensatory mechanisms that may have permitted losses of human disease genes in other mammals.

Whether the “natural TLR5 knockout” mammals retain a capacity to sense extracellular flagellin could be tested experimentally. Although research in river dolphins, pangolins, and pinnipeds is generally impractical, guinea pigs have been used in biomedical research including studies of asthma, tuberculosis, Zika virus, and other infectious diseases (Meurs et al. 2006; Clark et al. 2015; Kumar, Krause, et al. 2017), because many aspects of guinea pig physiology and immunology are more similar to humans than to mouse or rat (Padilla-Carlin et al. 2008). Thus, in addition to targeted research in mouse, guinea pigs could be used as another rodent model to study the consequences of natural TLR5 knockout mammals.

Our findings raise questions of which evolutionary and ecological factors are associated or drove the evolutionary losses of TLR5 in non-human mammals. TLR5 also plays an important role in establishing and maintaining a healthy gut microbiome (Vijay-Kumar et al. 2010; Fulde et al. 2018) and gut microbiome composition is influenced by diet (Delsuc et al. 2014; Youngblut et al. 2019), raising the possibility that dietary switches could be one of the involved factors. However, while all four mammalian TLR5-loss lineages are dietary specialists, they specialize in different diets, ranging from carnivorous (river dolphins, pinnipeds), herbivorous (guinea pigs) to myrmecophagous (pangolins that feed on termites and ants) diets. This indicates that specialization to a particular diet was likely not a major factor behind the evolutionary losses of this gene. Furthermore, many other mammals convergently specialized on carnivorous, herbivorous and myrmecophagous diets without losing TLR5. Similarly, the bird lineages that lost TLR5 (Velova et al. 2018) include generalists as well as granivorous or insectivores species, and thus are not different in terms of diet from many other birds that possess an intact TLR5 gene.

In general, loss of a gene can be due to relaxed selection to preserve its function or provide an evolutionary advantage (Albalat and Canestro 2016; Sharma, Hecker, et al. 2018; Hecker, Sharma, et al. 2019; Huelsmann et al. 2019). Interestingly, in humans, the TLR5 stop codon mutation is associated with both positive and negative consequences. On the one hand, TLR5 inactivation is associated with an increased susceptibility to Legionella-induced pneumonia and recurrent urinary tract infections, and constitutes a risk factor for type 2 diabetes (Hawn et al. 2003, 2009; Al-Daghri et al. 2013). On the other hand, TLR5 inactivation is associated with an improved survival in melioidosis patients (West et al. 2013), and in certain ethnic groups with a protection from systemic lupus erythematosus and Crohn’s disease (Hawn et al. 2005; Gewirtz et al. 2006). This indicates that inactivating TLR5 can also have a protective effect against the immune system dysregulation that is involved in these diseases. Such beneficial effects may explain why this stop codon mutation has reached a relatively high frequency of >4% in many human populations, although signatures of recent adaptive evolution have not been detected (Barreiro et al. 2009; Wlasiuk et al. 2009). Since an inflammatory response represents a cost–benefit trade-off that is often optimized for a particular environment (Okin and Medzhitov 2012), adjusting inflammatory responses can be an advantageous evolutionary strategy for species occupying specific environmental niches, exemplified by an attenuated inflammatory response to lipopolysaccharides in blood of deep diving seals (Bagchi et al. 2018). Thus, it remains to be elucidated whether the TLR5 losses in the four non-human mammalian lineages were due to relaxed selection or offered an evolutionary advantage.

Materials and Methods

Investigating and Validating Losses of TLR5

We used our gene loss detection method (Sharma, Hecker, et al. 2018) to investigate whether gene-inactivating mutations (frameshifting insertions/deletions, premature stop codon mutations, splice site disrupting mutations, large deletions) occur in TLR5 in mammals. To this end, we used human as the reference species and projected the human TLR5 gene to a total of 119 other mammals that are part of a 120-mammal genome alignment (Hecker and Hiller 2020). The resulting TLR5 sequences of 119 non-human mammals were re-aligned to the human TLR5 sequence using CESAR (Codon Exon Structure Aware Realigner) (Sharma et al. 2016, 2017) to avoid spurious frameshifts due to alignment ambiguities. Assembly gaps that overlap parts of TLR5 in two species were not taken as evidence for gene loss (see supplementary fig. 12, Supplementary Material online). Since the two TLR5 transcripts in human GENCODE version 32 (Harrow et al. 2012) annotate an identical single coding exon and only differ in 5′ UTR exons, we focused on this single coding exon that encodes the entire TLR5 protein.

Validation of small putative inactivating mutations was done as described before (Jebb and Hiller 2018; Sharma, Lehmann, et al. 2018). Briefly, we extracted a 50 bp genomic context surrounding a mutation, aligned these sequences against sequencing reads stored in SRA (Kodama et al. 2012) (accession numbers for each species are listed in supplementary table 1, Supplementary Material online), and counted how many reads support the inactivating mutation or the ancestral non gene-inactivating allele. To investigate large deletions, we visualized and inspected pairwise alignment chains (Kent et al. 2003) in the UCSC genome browser (Haeussler et al. 2019). In addition, alignment chains were used to investigate whether remnants of TLR5 occur in a context of conserved gene order and to exclude the possibility that a functional TLR5 copy exists elsewhere in these assemblies.

Gene-inactivating mutations that are shared between related species are strong evidence against assembly base errors and indicate that gene loss is fixed in the respective clade (Hecker, Sharma, et al. 2019; Huelsmann et al. 2019; Sharma and Hiller 2019). Therefore, in addition to the 119 non-human mammals included in the 120-mammal genome alignment, we analyzed TLR5 in six additional mammals that are close sister species to TLR5-loss species (two river dolphins, two guinea pigs, tree pangolin, and Antarctic fur seal; NCBI assembly accession numbers are listed in supplementary table 1, Supplementary Material online). For these six species, we computed sensitive pairwise alignment chains to human (hg38) using lastz (Harris 2007) (alignment parameters K = 2,400, L = 3,000, Y = 9,400, H = 2,000, default scoring matrix), axtChain (Kent et al. 2003) (linearGap=loose, otherwise default parameters), and RepeatFiller (Osipova et al. 2019) (default parameters). The presence of shared inactivating mutations was manually confirmed.

Relaxed Selection

We used RELAX (Wertheim et al. 2015) to determine whether the remaining TLR5 coding sequence evolves under relaxed selection in the Yangtze river dolphin and in pinnipeds. To this end, we first extracted the coding sequence of TLR5 from the 120-mammal alignment. Then, we generated pairwise alignments between the human coding sequence and the sequence corresponding to each of the 119 mammals using CESAR. These pairwise alignments were combined into a multiple sequence alignment using maf-join (Kielbasa et al. 2011). The multiple alignment was further postprocessed by removing alignment columns that contained frameshifting insertions and by substituting in-frame stop codons by NNN. We ran the test for relaxed selection twice and specified either the Yangtze river dolphin branch or the branch leading to pinnipeds as the foreground. All other branches leading to species that did not lose TLR5 were specified as background.

Analysis of Transcriptomic Data

RNA-seq data of guinea pig brain (SRP017611) (Fushan et al. 2015), lung (SRP040447) (Davidsen et al. 2014), ovaries and testis (SRP104222) (Bens et al. 2018), Malayan pangolin cerebrum and lung (SRP064341) (Mohamed Yusoff et al. 2016), small intestine, large intestine, and stomach (SRP101596 and SRP156258, Guangdong Institute of Applied Biological Resources), and Weddell seal brain, lung, placenta, and testis (SRP200409, Broad Institute) were downloaded from the Sequence Read Archive (SRA) (Kodama et al. 2012). Next, fastq-dump was used to process the initial read data using parameters for removing technical reads (skip-technical), filtering low-quality reads (read-filter pass), removing tags (clip), converting data into base space (dumpbase), keeping read identifiers (readids), and splitting paired-end reads into separate files (split-files). Reads were mapped to the corresponding genomes with STAR (version 2.4.2a) (Dobin et al. 2013). For Malayan pangolin, we adjusted the number of bins for genomic indices (genomeChrBinNbits: 15). Reads were mapped separately per run as paired-end reads with parameters limiting the ratio of mismatches per mapped read (outFilterMismatchNoverLmax = 0.04) and limiting the number of mapping locations per read (outFilterMultimapNmax = 20), and otherwise defaults. To visualize gene expression in the UCSC Genome Browser (Haeussler et al. 2019), we processed the mapped reads with BEDtools (Quinlan and Hall 2010) and bedGraphToBigWig (Kent et al. 2010).

PCR Analysis of TLR5 in the Brown Fur Seal

Tissue samples of the brown fur seal (Arctocephalus pusillus; Pinnipedia, Otariidae) were obtained from Senckenberg Natural History Collections Dresden, Germany (collection ID: MTD-TB 250) representing a specimen from the Zoological Garden Leipzig (Germany). DNA was extracted using the innuPrep DNA Mini Kit (Analytik Jena, Jena, Germany) following the manufacturer’s instructions (protocol for DNA isolation from tissue samples or rodent tails) except that tissue lysis was performed overnight and DNA was precipitated in two steps using 50 μl elution buffer. Standard PCR reactions were performed in a total volume of 20 µl using 10–20 ng DNA, one unit DFS-Taq polymerase (Bioron, Ludwigshafen, Germany) in the recommended buffer, 3.5 mM MgCl₂, 0.2 mM of each dNTP (Fermentas, St. Leon-Rot, Germany), and 0.37 µM of each primer listed in supplementary table 4, Supplementary Material online. PCR amplification of target TLR5 regions was performed with primer combinations shown in supplementary figure 13, Supplementary Material online, using 35 PCR cycles with denaturation at 94 °C (15 s but 3 min for the first cycle), annealing at 63 °C (20–30 s), and extension at 72 °C (1:30 min but 7 min for the last cycle). PCR products were Sanger sequenced in both directions after enzymatic clean-up with ExoSAP-IT (USB Europe GmbH, Staufen, Germany), cycle sequencing using the PCR primers, the BigDye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems), and an ABI 3130xl Genetic Analyser (Applied Biosystems, Foster City, CA, USA).

Supplementary Material

Supplementary data are available at Molecular Biology and Evolution online.

Acknowledgments

We thank the genomics community for sequencing and assembling the genomes and the UCSC genome browser group for providing software and genome annotations. We also thank Bogdan Kirilenko for computing river dolphin genome alignments, Anja Rauh for assistance in the SGN-SNSD-Mol-Lab, and the Computer Service Facilities of the MPI-CBG and MPI-PKS for their support. This study was supported by the Max Planck Society, the German Research Foundation (HI 1423/3-1), and the Leibniz Association (SAW-2016-SGN-2).

All analyzed genome assemblies and sequencing read data are publicly available on NCBI and SRA (accession numbers listed in supplementary table 1, Supplementary Material online). The multiple genome alignment is available at https://bds.mpi-cbg.de/hillerlab/120MammalAlignment/Human120way/ (last accessed March 15, 2020) and https://genome-public.pks.mpg.de/ (last accessed March 15, 2020). The TLR5 sequence of the brown fur seal has been submitted to NCBI GenBank (accession numbers MT137338-MT137341).

References