Deep-sea microbes as tools to refine the rules of innate immune pattern recognition

See allHide authors and affiliations

Science Immunology  12 Mar 2021:
Vol. 6, Issue 57, eabe0531
DOI: 10.1126/sciimmunol.abe0531

Deep-sea microbes exhibit immunosilence

The innate immune system of mammals uses pattern recognition receptors (PRRs) to detect conserved ligands displayed by potentially pathogenic microbes. To probe the performance of mammalian PRRs when confronted with microbes from a largely foreign ecosystem, Gauthier et al. isolated culturable bacteria from deep-sea Pacific Ocean water samples. Most isolates obtained were Gram-negative Gammaproteobacteria from the Moritella genus. The form of the prototype PRR agonist lipopolysaccharide (LPS) found in the outer membrane of most deep-sea Moritella strains was deficient at engaging with mouse and human LPS-sensing PRRs despite retaining most structural features of LPS from human intestinal E. coli. These findings reveal that the broad recognition powers of PRRs have boundaries that can be violated by a subset of microbes recovered from extreme environments.


The assumption of near-universal bacterial detection by pattern recognition receptors is a foundation of immunology. The limits of this pattern recognition concept, however, remain undefined. As a test of this hypothesis, we determined whether mammalian cells can recognize bacteria that they have never had the natural opportunity to encounter. These bacteria were cultivated from the deep Pacific Ocean, where the genus Moritella was identified as a common constituent of the culturable microbiota. Most deep-sea bacteria contained cell wall lipopolysaccharide (LPS) structures that were expected to be immunostimulatory, and some deep-sea bacteria activated inflammatory responses from mammalian LPS receptors. However, LPS receptors were unable to detect 80% of deep-sea bacteria examined, with LPS acyl chain length being identified as a potential determinant of immunosilence. The inability of immune receptors to detect most bacteria from a different ecosystem suggests that pattern recognition strategies may be defined locally, not globally.


The concept of pattern recognition, initially introduced by Janeway (1), posits that multicellular eukaryotes should have the ability to detect all microbes in the environment. This assumption of near-universal microbial detection is a foundation of modern immunology and relies on the ability of multicellular organisms to detect infections through the actions of a set of cellular proteins known as pattern recognition receptors (PRRs) (2). PRRs recognize potentially infectious agents by detecting specific molecules that are common to large classes of microbes. These molecules, commonly microbial cell wall components or nucleic acids, are known as pathogen-associated molecular patterns (PAMPs). Well-characterized examples of PAMPs include the lipid A region of bacterial lipopolysaccharides (LPSs), the flagellin subunit of bacterial flagella, and double-stranded DNA. Each of these molecules is important for the viability or fitness of the organism that produces them. As such, they are highly stable and prevalent in the microbial world (3).

On the basis of the above-described scenario, PRRs should have the capacity to detect all members of a given class of microbe. The only exceptions to this statement should be host-adapted microbes that evolved strategies to alter their PAMPs to prevent PRR detection. For example, many commensal bacteria coexist with mammalian hosts and avoid immune detection, whereas pathogens avoid detection to exploit the host (46). These exceptions represent the result of coevolution between host and microbe. For bacteria not closely associated with a host, there are adaptive trade-offs to altering PAMP structure that keep these strategies the rarity, rather than the norm (79). As such, outside of the intense host-microbial interface, mammalian PRRs are presumed to detect all bacteria they encounter.

However, virtually all knowledge of microbial detection has been derived from studies of bacteria that overlap ecologically with mammals, either in the same habitat or within mammalian hosts (10). With few exceptions (1113), these studies focused on bacteria that inhabit terrestrial or shallow-depth aquatic environments (14), where mammals abound (15). But what of the bacteria that occupy different ecological niches? Is the selective advantage of PAMP detection exclusive to hosts and microbes from sympatric environments? More simply stated, does the pattern recognition concept have limits?

In this study, we sought to test assumptions of the pattern recognition concept by examining the ability of human and murine cells to recognize bacteria that they would have not had the natural opportunity to encounter. These bacteria were collected and cultivated during an expedition to the remote and deep central Pacific Ocean, where samples were collected from deep-sea seamounts with abundant and diverse invertebrate life, and that are unoccupied by any resident mammals (tables S1 and S2).


We spent 20 days onboard the Schmidt Ocean Institute’s Research Vessel (R/V) Falkor within the boundaries of Kiribati’s Phoenix Islands Protected Area (PIPA), which is a no-take marine protected area and the largest and deepest United Nations Educational, Scientific, and Cultural Organization (UNESCO) World Heritage site (16). We leveraged the opportunity to collect bacteria from the deep sea (>200 m), a distinct ecological niche that is inhospitable to terrestrial life as it is principally devoid of light, under high pressure (>20 atmospheres), and cold (2° to 10°C) (17). In addition, the majority of the deep sea lacks residential mammalian life (15). Although some marine mammals access the deep sea through intermittent diving, the primary habitat of marine mammals is the photic zone of the ocean (tables S1 and S2) (15). Bacteria from these remote, deep-sea ecosystems were therefore of high interest, as mammals would not have had the natural opportunity to interact extensively with bacteria from this ecosystem.

At nine sites within the boundaries of PIPA (Fig. 1A), we collected seawater at stratified depths to assess bacterial community composition by amplicon sequencing of 16S ribosomal RNA genes. There was an observable shift in bacterial community composition between shallow (2 m) and deep (>200 m) seawater samples (Fig. 1, B and C). Thus, distinct microbial communities exist in deep-sea regions where mammalian populations are minimal. Although the amplicon sequencing data obtained provided insights into the bacterial community composition of PIPA, our main goal was to obtain culturable bacteria for experimental analysis.

Fig. 1 Bacterial sample collection during the R/V Falkor’s 2017 expedition to the PIPA, Kiribati.

(A) Site map showing nine sites sampled. (B) Bray-Curtis nonmetric multidimensional scaling ordination of total microbial community composition, based on 16S amplicon data from all stations. (C) Overview of the microbial community composition at stations S1 to S9 as determined by 16S ribosomal RNA (rRNA) amplicon analysis. (D) Streak-purified bacteria strains isolated and sequenced from seawater and substrates collected at stations S2 to S5. (E) Moritella relative 16S amplicon abundance in seawater samples collected from different depths at stations S2 to S5.

To identify culturable bacteria under laboratory conditions, we sampled four sites (S2 to S5) from 200- to 3000-m depth for seawater, sediment, coral tissue, sponge tissue, and the gut contents of corallivorous sea stars. After sample collection at sea, bacteria colonies were grown from deep-sea substrates and seawater to build our “experimental toolbox” of bacterial strains. A total of 117 bacterial colonies were isolated and streak-purified from S2 to S5. 16S ribosomal sequencing of streak-purified strains identified all culturable bacteria to be of the class Gammaproteobacteria, and 70% (82 of 117) of strains were of the genus Moritella (Fig. 1D). Moritella were not detected in surface seawater samples (collected from 2 m) (Fig. 1E). Therefore, the Moritella genus is a common culturable constituent among our samples found in the deeper waters of PIPA. A subset of these Moritella strains served as the foundation for our experimental analyses.

Moritella are Gram negative, which enabled us to determine the ability of mammalian LPS receptors to detect these bacteria. Mouse macrophages display monomers of the LPS receptors CD14 and Toll-like receptor 4 (TLR4; associated with MD-2) at their plasma membrane, where they survey the extracellular space for this PAMP. When CD14 and TLR4 bind LPS, they are lost from the cell surface. CD14 is immediately internalized into endosomes, whereas TLR4 must first dimerize before also being endocytosed (1820). The LPS-induced loss of CD14 and TLR4 monomers from the plasma membrane can be monitored by flow cytometry, which represents a rapid assay that enables LPS-PRR interactions to be tracked for endogenous receptors in their natural setting (on the macrophage surface) (1820).

Using this flow cytometry–based assay, we exposed murine immortalized bone marrow–derived macrophages (iBMDMs) to live bacteria for 20 min and measured the loss of receptors from the cell surface. Escherichia coli LPS interacts efficiently with both receptors (19). Accordingly, live E. coli was used as a benchmark to delineate whether each Moritella strain engaged CD14 and TLR4. We limited the bacterial strains tested to those that were streak-purified while onboard the R/V Falkor (n = 50). Of the strains examined, 88% were of the genus Moritella (n = 44), all isolated from >550-m depth.

Compared to E. coli, which stimulated CD14 and TLR4 loss from the cell surface, specific marine bacterial strains were grouped into three categories (Fig. 2, A to C). Category 1 strains behaved similarly to E. coli, in that loss of both PRRs occurred. Category 2 strains were unable to engage either PRR. Category 3 strains engaged only one receptor. Overall, 10 strains engaged both receptors, 19 engaged neither, and 21 engaged one receptor but not the other (Fig. 2C). These findings were notable for two reasons. First, our finding that any strains of bacteria stimulated CD14 and TLR4 provides direct support for a central tenet of the concept of pattern recognition—that PRRs have the capacity to detect previously unencountered bacteria. Although these findings are consistent with current dogma, it was unexpected that 80% (40 of 50) of bacteria displayed evidence of immune evasiveness, as defined by an inability to stimulate one or both of CD14 and TLR4. Deep-sea microbial species may therefore represent a reservoir of molecules with unpredictable inflammatory activities.

Fig. 2 Grouping of marine bacterial strains into three categories based on engagement with CD14 and TLR4 on mouse macrophages.

(A and B) Surface expression of CD14 (A) and TLR4 (B) as measured by mean fluorescence intensity (MFI) on iBMDMs exposed to live deep-sea bacteria strains was compared to live E. coli [multiplicity of infection (MOI) = 50]. The color of columns represents the predicted acyl chain number for the LPS lipid A expressed, as described in Table 1. Dashed lines are used to delineate whether bacterial strains are stimulatory or silent to CD14 or TLR4, as compared to E. coli. (C) Summary of murine CD14 and TLR4 engagement by strains tested.

A possible explanation for the inability of deep-sea bacteria to stimulate CD14 or TLR4 is that the cell wall may contain features that prevent access of PRRs to immunostimulatory LPS. To address this possibility, we isolated LPS from select Moritella strains to evaluate whether purified LPS behaved similarly to whole bacteria. Twelve Moritella strains over the range of CD14-TLR4 engagement demonstrated in Fig. 2 were selected for this analysis. Specifically, three strains were selected that engaged neither PRR, and nine strains were selected that displayed a range of PRR engagement. In all analyses, purified LPS from each Moritella strain was compared to E. coli LPS (Fig. 3). Dose response curves demonstrated that bacterial cells that were unable to engage CD14 and TLR4 yielded LPS that did not promote PRR loss from the cell surface, as compared to E. coli LPS (Fig. 3A). All strains that had a partial phenotype when bacterial cells were used as stimuli (e.g., those that stimulated TLR4, but not CD14) were fully stimulatory for both PRRs when purified LPS was used (Fig. 3B). This analysis therefore allowed us to classify LPS from Moritella in a binary fashion—either immunostimulatory or immunosilent. Note that we chose the term “silent” rather than “evasive,” as the latter term suggests intent. The immunosilent LPS preparations were isolated from Moritella 9, 28, and 36, the latter two of which were selected for further analyses.

Fig. 3 Dose-response curves testing the ability of purified LPS from Moritella strains to induce PRR loss from the surface of mouse macrophages in comparison to E. coli LPS.

(A) Purified LPS from three Moritella strains predicted to be hexa-acylated did not behave similarly to purified E. coli LPS in flow cytometry assays measuring engagement of CD14 and TLR4 on murine iBMDMs. (B) Purified LPS from nine other Moritella strains predicted to be hexa-acylated behaved similarly to E. coli LPS. Red circles indicate Moritella LPS, and black squares indicate E. coli LPS. Statistical analysis based on comparison of 100 ng/ml dose from Moritella LPS with 100 ng/ml dose of E. coli LPS. *P ≤ 0.01, ns, not significant.

We reasoned that if LPS from Moritella 28 and 36 are incapable of stimulating CD14 and TLR4, then these LPS preparations should be unable to stimulate TLR4-dependent inflammatory responses. Consistent with our findings when assessing CD14 or TLR4 engagement, LPS from Moritella 5 and 24 induced tumor necrosis factor–α (TNFα) production and signal transducer and activator of transcription 1 (STAT1) phosphorylation (Fig. 4, A and B). LPS from strains that did not stimulate CD14 or TLR4 were unable to induce these inflammatory responses (Fig. 4, A and B). All TNFα and STAT1 responses observed upon LPS stimulations were abolished when assays were performed on TLR4-deficient cells, as expected (21).

Fig. 4 Functional analysis of purified LPS from silent (#28, 36) or stimulatory (#5, 24) Moritella strains in murine and human macrophages.

(A and B) Accumulation of TNFα after 3.5 hours (A) and phosphorylation of STAT1 after 2.5 hours (B) measured in wild-type (WT) and Tlr4−/− iBMDMs incubated with LPS (100 ng/ml) from Moritella strains or E. coli. (C) Release of LDH 3 hours after electroporation of wild-type and Casp11−/− iBMDMs with 1 μg of LPS from Moritella strains and E. coli. (D) Cleavage of GSDMD 3 hours after electroporation of wild-type iBMDMs with 1 μg of LPS from Moritella strains and E. coli. (E) Binding of hemagglutinin (HA)–tagged caspase-11 to LPS supplied in excess (5 μg) from Moritella strains or E. coli, as measured by the ability of LPS to compete off biotinylated E. coli LPS (1 μg). (F) Production of pro–IL-1β 2.5 hours after treatment of human THP1 cells incubated with LPS (50 ng/ml) from Moritella strains compared to E. coli LPS. (G) Accumulation of TNFα 4 hours after treatment of human THP1 cells incubated with LPS (100 ng/ml) from Moritella strains compared to E. coli LPS. (H) Engagement of human TLR4 in human TLR4/nuclear factor κB (NF-κB)/SEAP reporter HEK293 cells by LPS from Moritella strains compared to E. coli LPS. a.u., arbitrary units. (I) Release of LDH 2.5 hours after electroporation of human THP1 cells with 1 μg of LPS from Moritella strains compared to E. coli LPS. *P ≤ 0.01 and **P ≤ 0.001.

In addition to CD14 and TLR4, caspase-11 recognizes LPS in the cytosol of macrophages. Engagement of caspase-11 by E. coli LPS results in a lytic form of cell death called pyroptosis (2224). Pyroptosis can be assessed by monitoring the release of the cytosolic enzyme lactate dehydrogenase (LDH) into the extracellular space (25) along with the cleavage of the pore-forming protein gasdermin D (GSDMD) (26). As expected (27, 28), electroporation of iBMDMs with E. coli LPS stimulated pyroptosis, as assessed by LDH release and GSDMD cleavage (Fig. 4, C and D). LPS from Moritella 5 and 24 also stimulated robust LDH release after electroporation. LPS from Moritella strains 28 and 36, which did not stimulate CD14 and TLR4, induced no LDH release from iBMDMs (Fig. 4C). All LDH release observed was abolished when experiments were performed in caspase-11–deficient cells, as expected (2224). Consistent with these results, partial or no cleavage of GSDMD was observed in iBMDMs electroporated with LPS from Moritella strains 28 and 36 (Fig. 4D). Conversely, full cleavage of GSDMD was induced by LPS from E. coli, Moritella 5, and Moritella 24 (Fig. 4D). To determine whether differences in pyroptosis induction related to an ability to interact with caspase-11, in vitro experiments with purified LPS were performed. Using a procedure to assess PRR interactions with E. coli LPS or oxidized lipids (29), we assessed the ability of Moritella LPS to compete with biotinylated E. coli LPS for interactions with caspase-11. Biotinylated E. coli LPS formed a complex with caspase-11 in vitro (Fig. 4E). This interaction was prevented when reactions were performed in the presence of excess nonbiotinylated E. coli LPS or in the presence of immunostimulatory LPS from Moritella 5 or Moritella 24 (Fig. 4E). In contrast, LPS from the immunosilent Moritella strains 28 and 36 did not compete with biotinylated E. coli LPS for binding to caspase-11 (Fig. 4E). These results suggest that only immunostimulatory LPS can interact with caspase-11, a finding that likely explains the pyroptosis-inducing activity of the Moritella strains examined.

To establish whether the phenotypes observed in murine cells extended to other species, studies were performed in human THP1 monocytes and the LPS detection system from horseshoe crab crustaceans. Akin to the observations made in murine systems, LPS from Moritella 28 and 36 did not stimulate the production of pro–interleukin-1β (IL-1β) or TNFα in THP1 monocytes, as compared to LPS from E. coli, Moritella 5, or Moritella 24 (Fig. 4, F and G). To validate these findings, a human embryonic kidney (HEK) 293–BLUE reporter system was used to assess direct engagement of human CD14, TLR4, and MD-2. In these cells, LPS interactions with these PRRs stimulate the secretion of alkaline phosphatase (SEAP). We found that only immunostimulatory LPS induced SEAP production by HEK293-BLUE cells (Fig. 4H), thereby suggesting that silent Moritella LPS from strains 28 and 36 do not engage PRRs in the TLR4 pathway. Last, electroporated LPS from Moritella 5 and 24 induced THP1 monocyte pyroptosis to an extent comparable to that elicited by E. coli LPS (Fig. 4I). LPS from Moritella 28 and 36 induced minimal LDH release from electroporated THP1 monocytes (Fig. 4I).

The horseshoe crab Limulus polyphemus contains a more distinct LPS detection system than terrestrial mammals (30). Upon binding of LPS to the protein factor C from L. polyphemus, a complement-like protease cascade is stimulated that results in a defensive coagulation response. Although it is an aquatic animal, L. polyphemus occupies shallow marine environments (31). LPS preparations that could stimulate detection (or not) by mammalian LPS receptors exhibited the identical pattern of activity of engagement with factor C. LPS from Moritella 28 and 36 did not engage factor C, whereas LPS from Moritella 5 and 24 did so, similarly to E. coli LPS (Fig. 5A).

Fig. 5 Cross-species evasion of LPS receptor activity by deep-sea Moritella strains.

(A) Engagement of L. polyphemus factor C by LPS from Moritella strains compared to E. coli LPS. (B) Inflammatory response to LPS from Moritella strains in vivo. IL-6 and TNFα plasma levels were measured 4 hours after intraperitoneal injection of mice with LPS from stimulatory (#5, 24) or silent (#28, 36) Moritella strains. (C) E. coli LPS (10 μg) was compared to 10 μg of LPS or lipid A (LA) from Moritella strains visualized on a polyacrylamide gel stained with ProQ Emerald LPS staining solution. (D) Accumulation of TNFα after 3.5-hour stimulations of iBMDMs with lipid A (100 ng/ml) compared to LPS from Moritella strains. (E) Release of LDH 24 hours after electroporation of iBMDMs with 1 μg of lipid A or LPS from Moritella strains. (F) Engagement of L. polyphemus factor C by LPS and lipid A from Moritella strains. *P ≤ 0.01, **P ≤ 0.001, and ***P < 0.0001.

To determine whether the immunosilence of select Moritella LPS extended to an in vivo setting, inflammatory responses in mice were examined. Intraperitoneal injections of LPS from Moritella 5 and 24 induced the rapid accumulation of cytokines TNFα and IL-6 in the plasma (Fig. 5B). In contrast, these cytokines were not detected in the plasma of mice injected with Moritella 28 and 36 (Fig. 5B). Overall, symmetrical observations were made in mice, in human and murine cells, and in the in vitro system offered by the horseshoe crab. All these systems revealed immunosilent LPS from Moritella 28 and 36.

The LPS molecule of Gram-negative bacteria is composed of three regions. The hydrophobic lipid A region anchors LPS to the bacterial outer membrane, whereas the water-soluble core oligosaccharide and O-antigen extend from the lipid A anchor into the aqueous extracellular space (32). It was possible that the silent activity of select Moritella strains was due to specific features of the O-antigen region that prevent PRR detection of lipid A. If this possibility was correct, then removal of the O-antigen should render immunosilent molecules stimulatory. Thus, we purified lipid A after hydrolysis of the O-antigen from the LPS preparations of interest. This procedure resulted in the near-complete disappearance of the core and/or O-antigen regions, as assessed by silver staining (Fig. 5C). Removal of O-antigen did not increase the stimulatory activity of the immunosilent lipid A preparations. Lipid A from immunostimulatory Moritella strains remained capable of inducing TNFα production, whereas lipid A from silent Moritella strains 28 and 36 remained incapable of inducing this inflammatory response (Fig. 5D). The same pattern of engagement for Moritella lipid A was observed for murine caspase-11 engagement, as indicated by pyroptosis induction after electroporation (Fig. 5E). Last, neither LPS nor lipid A from Moritella 28 and 36 was able to stimulate L. polyphemus factor C (Fig. 5F). These data indicate that lipid A was not being hidden from LPS receptors by other structural regions of the LPS molecule. We therefore conclude that the silence of some deep-sea bacteria is intrinsic to the lipid A structure of LPS and extends across the evolutionary tree of PRRs.

The features of lipid A that are associated with immunostimulatory activity of E. coli include a hexa-acylated and bisphosphorylated diglucosamine backbone (32). LPS structures that contain more (or less) than six acyl chains are weakly stimulatory to PRRs, and LPS structures containing monophosphorylated LPS are similarly immune-evasive. In addition, specific modifications to lipid A, such as aminoarabinose, phosphoethanolamine (PEtn), or the presence of cardiolipin in LPS preparations, are associated with nonstimulatory activity (10, 33). To determine whether deep-sea LPSs displayed any of these immuno-evasive lipid A features, structural composition analysis was performed. We first performed a large-scale analysis of lipid A structures isolated from all 50 deep-sea bacteria we examined at the start of this study (Table 1). We extracted lipid A and predicted structural composition by matrix-assisted laser desorption/ionization–time-of-flight (MALDI-TOF) mass spectrometry (MS) in the negative ion mode (34). Mass spectral data were used to predict the following within lipid A: (i) the number of acyl chains likely to be present, (ii) the loss of a phosphate from the diglucosamine sugar backbone (monophosphorylation), (iii) the addition of aminoarabinose or PEtn to the diglucosamine sugar backbone, and (iv) the presence of cardiolipin in the lipid A extractions.

Table 1 MALDI-TOF MS structural data and endogenous murine CD14-TLR4 engagement by bacterial strain.

Structural data reported for each strain: (i) the primary m/z peak value, (ii) the number of acyl chains predicted for the m/z peak value reported, (iii) monophosphorylation of the lipid A diglucosamine backbone, and (iv) the addition of PEtn to the lipid A diglucosamine backbone. CD14-TLR4 engagement is reported as yes/no compared to E. coli.

View this table:

These structural analyses revealed that several bacteria contain immunosilent modifications to their lipid A. Specifically, the Colwellia strains examined contained a primary ionizable mass/charge ratio (m/z) peak of >2000 atomic mass units (amu), which is predicted to be hepta-acylated (Table 1). In addition, we observed PEtn additions (Δm/z of 123 amu) to the lipid A backbone of select strains and the presence of cardiolipin (repeating Δm/z of 12 amu) in others (Table 1). These bacteria were identified in Fig. 2 as containing LPS that prevents detection by CD14 or TLR4, an observation that is likely explained by the structural changes to the lipid A we identified.

In contrast to the examples offered above, which are consistent with current views of LPS-PRR interactions, most strains examined (including all 44 Moritella strains) contained a primary ionizable m/z peak ranging from 1650 to 1800 amu (Table 1). These spectra are consistent with hexa-acylated lipid A being present in all Moritella examined. Similar findings were observed for lipid A structures from Halomonas, Shewanella, and Vibrio strains (Table 1). All Moritella strains were also predicted to contain bisphosphorylated lipid A, although some strains had minor populations of monophosphorylation (∆m/z of 80) (Table 1). These features of lipid A—hexa-acylated and bisphosphorylated—are commonly associated with immunostimulatory LPS, such as that from E. coli. Yet, these E. coli–like lipid A structures from deep-sea bacteria were either immunostimulatory or immunosilent. Of the Moritella examined structurally, 77.3% (34 of 44) displayed evidence of immunosilence when bacteria were used to stimulate murine cells (Fig. 2). The immunosilence of Moritella 9 may be explained by the presence of a PEtn on its lipid A diglucosamine backbone (Table 1) (35). Moritella 28 and 36, in contrast, did not have any observable backbone modifications to explain why their LPS was immunosilent across multiple experimental systems.

To confirm the MALDI-TOF MS analyses, we used a newly described gentle extraction method, known as fast lipid analysis technique (FLAT) (36). FLAT enables the detection of very small quantities of lipid A within a bacterial colony, without the need for centrifugation or lyophilization. The mass spectra obtained by this distinct method were consistent with those we obtained using other methods (Table 1), in which all four Moritella strains of interest (two immunostimulatory and two immunosilent) were predicted to be hexa-acylated and bisphosphorylated (Fig. 6A).

Fig. 6 Further characterization of immunosilent and immunostimulatory Moritella strains.

(A) MALDI-TOF MS spectra of Moritella lipid A generated using the FLAT technique. (B) Relative fatty acid content in lipid A derived from silent and stimulatory Moritella as determined by GC-MS. *P ≤ 0.01. (C) Whole-genome phylogeny of Moritella strains. The amino acid phylogeny was inferred using maximum likelihood from a concatenated alignment of 120 single copy genes (56) generated from the four newly sequenced Moritella genomes (green squares) and other publicly available assemblies. Black dots on branches indicate bootstrap support >75%. (D) Degree of sequence conservation for enzymes in the lipid A biosynthesis pathway. The maximum likelihood phylogeny at left is based on a concatenated amino acid alignment of the 10 indicated lipid A biosynthesis enzymes from each genome. The heatmap depicts the % amino acid identity for each individual enzyme in the pathway, as compared to M. oceanus 28. Black dots on tree branches indicate bootstrap support of >75%.

To further assess the structural features of Moritella lipid A, gas chromatography–MS (GC-MS) was performed. This analysis enables the determination of the composition of acyl chains present on lipid A (37). We found that immunosilent strains (Moritella 28 and 36) contained lipid A with the highest amount of C16 acyl chains (Fig. 6B). Conversely, the stimulatory strains contained lipid A with zero C16 chains (Moritella 5) or low amounts of C16 chains (Moritella 24). Instead, these stimulatory strains had mainly C12 and C14 length acyl chains (Fig. 6B). These findings were notable, as C12 and C14 length acyl chains are the optimal lengths for robust activation of TLR4 signaling (38, 39). C16 chains, in contrast, are recognized as being nonoptimal lengths for productive interactions with MD-2 (40). In the context of synthetic lipid A mimics, those with C16 chains cannot bind MD-2 at all and are consequently immunosilent (40). Consistent with this idea, the highest amount of C16 acyl chains was found in the immunosilent Moritella strains 28 and 36. Other than these C16 differences, the GC analysis revealed similar patterns of fatty acid composition in the strains examined, with no strain-specific differences in short or odd-length fatty acids detected. Thus, we propose that immunosilence is not the result of an unusual addition to the lipid A molecules found in the deep sea but rather results from the existence of abundant C16 acyl chains found in select strains, which determines PRR interactions.

We sought to determine how each Moritella strain related to each other and to previously identified Moritella species. A combination of Nanopore and Illumina sequencing was used to determine the genome sequence of the two silent and the two stimulatory Moritella strains (table S3). We determined the phylogenetic relationship between these strains and other sequenced Moritella genomes (Fig. 6C). This analysis indicated that Moritella 5, 24, 28, and 36 strains are evolutionarily distinct from known Moritella. Average nucleotide identity analysis (41, 42) indicated that Moritella 24 represents a distinct species, whereas Moritella 5, 28, and 36 represent substrains of a second new species (table S4). We tentatively propose these species be designated Moritella oceanus (formerly Moritella 5, 28, and 36) and Moritella rawaki (formerly Moritella 24).

Pangenome analysis revealed a set of protein-coding sequences that distinguished the two immunosilent from the two immunostimulatory strains. Specifically, both immunosilent Moritella strains encoded a common set of 192 protein-coding sequences (table S5) not found in either immunostimulatory strain. Fifty-five protein-coding genes were found exclusively in both stimulatory strains (table S6). Most of these genes lack an annotated function. Whether any of these 55 genes affect immunosilence or detection is unknown, as is the degree of sequence conservation across other [unsequenced but behaviorally similar (as in Fig. 2)] Moritella strains.

We identified the lipid A biosynthesis enzymes and their protein-coding sequences (32) within the genomes of the four Moritella strains (fig. S1). We identified one copy of each enzyme, with the exception of LpxL, which had two protein-coding sequences present in each genome. We determined the degree of sequence conservation for enzymes in the lipid A biosynthesis pathway between these four Moritella strains and E. coli (Fig. 6D). Although the E. coli lipid A enzymes were distinct from those present in any Moritella strain, the most notable distinctions came from comparisons within Moritella. The enzymes that build lipid A in immunosilent M. oceanus 28 and M. oceanus 36 are (on a network scale) highly similar to each other. The analogous enzymes in immunostimulatory M. oceanus 5 or M. rawaki 24 are also highly similar to each other, but differed from the corresponding enzymes in immunosilent M. oceanus strains. On the basis of this analysis, the entire lipid A biosynthetic pathway may contribute to the distinct inflammatory activities of select deep-sea Moritella LPS structures.


In this study, we tested the limits of the pattern recognition hypothesis by asking whether mammalian PRRs could detect bacteria from a different ecosystem. Two major conclusions were drawn. First, some deep-sea bacteria can be detected by mammalian LPS receptors, although immunostimulation of both CD14 and TLR4 represented a minor population of all bacteria examined [20% (10 of 50)]. Nonetheless, the identification of even a single bacterium with immunostimulatory activity validates a central principle of pattern recognition—that our innate immune system can detect previously unencountered bacteria. Second, we found that mammalian PRRs were unable to detect most of the LPS displayed on live, cultured bacterial strains from the deep sea (Fig. 2), revealing that the rules of innate immune detection may be more limited than previously appreciated. Thus, we posit a revision to the pattern recognition concept—PRRs should have the capacity to detect all bacteria that exist in the same general habitat as the host. In other words, innate immunity follows local (not global) rules of engagement (fig. S2).

It is unlikely that deep-sea bacteria experience any fitness benefit from evading detection of mammalian PRRs, even those expressed by mammals that occasionally dive to the depths that these microbes inhabit (table S1). Inhabiting an environment devoid (or nearly devoid) of mammals likely creates a scenario where there is no selective advantage to bacterial LPS being immune-evasive, immunosilent, or immunostimulatory. Bacteria from extreme, nonmammalian environments may have cell wall structures that are uniquely optimized to their environment, and when habitats are artificially crossed, bacteria may encounter a mammalian host in which they are accidentally immunosilent. The consequences of such interactions are difficult to predict but are important to consider in this age of increasing globalization and exploitation of deep-sea resources (43).

Last, it is worth considering whether deep-sea invertebrates may have evolved PRRs to detect LPS structures that are common to bacteria found in these habitats. This consideration is notable, as the functions of LPS receptors have been explored almost exclusively in mammals. Genomic analysis of numerous organisms has demonstrated a notable lack of the CD14-MD-2-TLR4 system in marine fish and invertebrates. Although proteins that display homology to TLR4 can sometimes be identified in fish or invertebrates, the significance of this homology is unclear, as CD14 and MD-2 can almost be considered mammal-specific factors. Identification of invertebrate taxa in the deep sea is ongoing, and many species are yet unknown; however, deep-sea corals and sponges are among the most abundant foundational macrofauna (44, 45). Shallow-water cnidarians have an abundance of genes that have been bioinformatically annotated as PRRs (46); however, whether these PRRs also exist abundantly in deep-sea taxa is unknown. One study has examined flagellin signaling in cnidarians (a shallow-water anemone) (47), but nothing is yet known about LPS detection events in any marine invertebrate. As such, it is unclear whether LPS detection systems are common outside of the mammalian lineage. Even the plant Arabidopsis thaliana, which was once thought to use the protein LORE as an LPS receptor, is now recognized to not detect LPS (48). Although these evolution-function considerations remain to be addressed, our data suggest that the rules of innate immune engagement need to be refined, and thus provide a mandate for further exploration of host-microbe interactions in diverse ecosystems.


Study design

We designed this study to evaluate the ability of mammalian PRRs to detect deep-sea bacteria via the PAMP, LPS. Therefore, we cultured bacterial strains from the deep sea of PIPA to screen in a flow cytometry–based assay for their ability to interact with endogenous murine PRRs, CD14 and TLR4. Our results indicated that 80% of live bacteria strains did not engage with murine CD14 and/or TLR4 and were categorized as silent to these PRRs. To confirm that LPS from deep-sea bacteria was silent, we purified LPS from stimulatory and silent Moritella strains to characterize the downstream innate immune response to extracellular and intracellular LPS in murine and human cells. LPS from Moritella strains that were silent to murine and human PRRs did not initiate any downstream innate immune response. LPS from silent Moritella strains also did not engage with murine PRRs in vivo or in L. polyphemus. We then hydrolyzed and purified lipid A from Moritella LPS and confirmed that the activity of silent LPS was a direct result of the structure of the lipid A moiety. To delineate the differences between stimulatory and silent Moritella lipid A, we characterized the structure of lipid A and sequenced the genomes of each strain of interest.

Deep-sea sampling

In October 2017, the R/V Falkor, operated by the Schmidt Ocean Institute, conducted 17 remotely operated vehicle (ROV) surveys in the PIPA (cruise FK171005) [detailed in (45)] using the 4500-m rated ROV SuBastian. On the PIPA expedition, ROV SuBastian was equipped with an Insite Pacific Mini Zeus HD 1080i CMOS (complementary metal-oxide semiconductor) camera for situational awareness and an Insite Zeus Plus or SULIS 4K 12× Zoom camera for science surveys (5.1- to 51-mm-wide angle zoom lens). The vehicle also hosted a Seabird FastCAT CTD Sensor (SBE49) and a Paroscientific 8000 Series Submersible Depth Sensor for measuring depth of observations [as in (49)]. In addition to 17 ROV dives, seven CTD casts were deployed and retrieved in PIPA, from which water was filtered for microbial extraction or culturing, as below.

Isolation of culturable bacteria strains

At four atolls within the boundaries of PIPA (depths of 270 to 2500 m), seawater, soft coral tissue, glass sponge tissue, and sea star gut content were collected to use as substrate to grow deep-sea bacteria. Tissue and gut content were homogenized in 5 ml of phosphate-buffered saline (PBS; pH 7.4) in sterile 15-ml falcon tubes. The substrate (200 μl) was loaded onto individual Difco marine agar 2216 (Becton Dickenson, 212185) plates without antibiotics in a sterile fume hood aboard the R/V Falkor. Plates were incubated at 4°C for 5 to 14 days, protected from light, until colonies formed. After bacteria colony formation, strains were streak-purified, and glycerol stocks (50% marine medium, 25% water, and 25% glycerol) were stored at −80°C in 2.0-ml cryovials. Upon expedition completion, glycerol stocks were transported on dry ice by airplane to be stored at −80°C in Boston, MA.

Amplicon sequencing

Amplicon sequencing is described in Supplementary Methods.

Sequencing bacterial strains from liquid culture

Bacterial gene sequencing is described in Supplementary Methods.

Moritella genome sequencing and analysis

Genome sequencing and analysis are described in Supplementary Methods.

Cell lines

Wild-type, Tlr4−/−, and Casp11−/− iBMDMs were cultured in Dulbecco’s modified Eagle’s medium (DMEM) with 10% fetal bovine serum, 1% penicillin and streptomycin, and 1% supplementation with l-glutamine and sodium pyruvate. This medium is referred to as complete DMEM. Before experimentation, cells were washed with PBS (pH 7.4) and lifted with PBS (pH 7.4) containing 2 mM EDTA. Otherwise, cells were passaged 1:10 every 2 to 3 days. During routine passage, cells were washed with PBS (pH 7.4) and detached with 0.25% trypsin. Trypsin was deactivated by adding serum-containing medium, after which cells were spun down and resuspended in fresh medium.

LPS and lipid A preparations

LPS and lipid A preparations are described in Supplementary Methods.

Lipid A extraction for MALDI-TOF MS

Bacteria strains (n = 50) were grown for 48 to 72 hours to OD600 (optical density at 600 nm) = 1 in 5-ml marine medium at 14.5°C in a shaking incubator (180 rpm). Bacteria were pelleted and frozen for future extraction of lipid A. Frozen bacterial pellets were thawed on ice, and lipid A was extracted from pellets using the isobutyric acid method (50). Briefly, in screw-cap tubes, pellets were resuspended in 70% isobutyric acid and 1 M ammonium hydroxide [5:3 (v/v); 400 μl total]. Tubes were incubated for 60 min at 100°C with occasional vortexing to liberate lipid A. Products were cooled on ice and centrifuged for 5 min at 8000g before harvesting the supernatant into new 1.5-ml tubes containing an equal volume (400 μl) of endotoxin-free ddH2O. Samples were frozen on dry ice for 30 min and lyophilized. Lyophilized samples were washed with methanol, and lipid A was extracted from the remaining pellet with 50 μl of chloroform:methanol:water [3:1.5:0.25 (v/v)]. The final lipid A product was mixed with Dowex beads and centrifuged for 5 min at 5000g. The lipid A product (supernatant) (1 μl) was resuspended with 1 μl of Norharmane matrix (10 mg/ml) [Sigma-Aldrich, N6252; (51)], resuspended in chloroform:methanol [2:1 (v/v)] at a ratio of 1:1, and spotted on a steel reusable MALDI plate. Spots were air-dried, and samples were analyzed on a Bruker Microflex mass spectrometer (Bruker Daltonics, Billerica, MA) in the negative ion mode with reflectron mode. The spectrometer was calibrated with an electron spray tuning mix before every run (Agilent, Palo Alto, CA).

FLAT technique lipid A extraction for MALDI-TOF MS

Microbial colony smears or liquid samples were applied to a target location on a stainless steel MALDI plate. The target plate was incubated in a 70% citric acid buffer in a humidified chamber for 30 min at 110°C. The MALDI plate was washed with deionized water from a squeeze bottle and allowed to air dry, and then 1 μl of Norharmane matrix solution was applied [10 mg/ml in 12:6:1 (v/v/v) chloroform:methanol:water] to each target location (36). Following the method of Leung et al. (52), spectra were acquired from target locations in negative ion mode using a Microflex LRF MALDI-TOF MS (Bruker, Billerica, MA) in reflectron mode with a limited mass range of 1000 to 2400 m/z. Typically, 300 laser shots were summed to acquire each mass spectrum.

MALDI-TOF MS data analysis

Replicate m/z spectra data were generated for each lipid A. Bruker Daltonics flexAnalysis software was used to analyze all spectral data generated. Lipid A was classified as hexa- or hepta-acylated based on the m/z ratio of its primary spectral peak. On the basis of previous MS-based studies on LPS lipid A (35, 5355), peaks with m/z ratios between 1650 and 1800 amu are predicted to be hexa-acylated, whereas peaks above 2000 amu are predicted to be hepta-acylated. Using these parameters, the acyl chain number present in lipid A of each strain was predicted. In addition, the loss or addition of a phosphate to the diglucosamine backbone was indicated by an m/z change of 80 amu, and the addition of PEtn was indicated by an m/z change of 123 amu. The presence of the outer membrane molecule, cardiolipin, was indicated by an m/z ratio value of 1456 amu and single carbon loss/additions.

GC-MS analysis

GC-MS is described in Supplementary Methods.

Murine macrophage stimulations and analysis

Macrophage stimulations and analysis are described in Supplementary Methods.

In vivo LPS challenge

Ten-week-old mice were injected intraperitoneally with LPS (1 mg/kg) purified from Moritella sp. strains 5, 24, 28, or 36. Blood samples were collected 4 hours after injection, and plasma were obtained by centrifugation at 2000g for 10 min. IL-6 and TNFα plasma concentrations were measured by enzyme-linked immunosorbent assay (BioLegend). All animal procedures were approved by the Institutional Animal Care and Use Committee.

Human THP1 cell stimulations and analysis

THP1 stimulations are described in Supplementary Methods.

HEK-BLUE 293T reporter cell line (InvivoGen, hkb-htlr4)

In a 96-well plate, 0.5 × 105 cells per well were plated in culture medium for real-time detection of secreted alkaline phosphatase (InvivoGen, hb-det2). In triplicate, Moritella LPS or E. coli O111:B4 LPS (100 ng/ml) was added to each well. Plates were incubated for 24 hours at 37°C, 5% CO2. Secreted alkaline phosphatase was determined by reading the absorbance at 635 nm (Tecan; model Spark 10M). The statistical analysis was done using Prism. One-way analysis of variance (ANOVA) was performed assuming Gaussian distribution, no pairing, and multiple comparisons to E. coli O111:B4 LPS. Equal SD was not assumed, and Brown-Forsythe and Welch ANOVA tests were performed to determine whether there were differences in absorbance between 293T cells incubated with Moritella LPS and E. coli LPS. Significance was reported to a 99.9% confidence interval (P ≤ 0.001).

Limulus amebocyte lysate Pyrochrome assay

In triplicate, Moritella LPS, Moritella lipid A, or E. coli O111:B4 LPS (20 ng/ml) was sequentially diluted 10-fold from 20 ng/ml to 20 pg/ml in a 96-well plate. The total volume in each well was 50 μl per well. PBS was used as a negative control. Pyrochrome reagent was resuspended in Glucashield buffer according to the manufacturer’s protocol (Associates of Cape Cod, CG1500). Pyrochrome reagent (50 μl) [1:1 (v/v)] was added to each well for a final volume of 100 μl per well. The final concentrations of LPS/lipid A measured in triplicate were as follows: 10 ng/ml, 1 ng/ml, 100 pg/ml, and 10 pg/ml. Absorbance was measured at 405 nm at 37°C for 60 min every 5 min with a plate reader (Tecan, model Spark 10M). Data were analyzed when maximum absorbance was reached at 1 ng/ml. Data presentation and statistical analyses were performed using Prism. One-way ANOVA was performed assuming Gaussian distribution, no pairing, and multiple comparisons to either E. coli LPS or Moritella LPS. Equal SD was not assumed, and Brown-Forsythe and Welch ANOVA tests were performed to determine whether there were significant differences between conditions analyzed. Calculated P values are as indicated in the figure captions.


Supplementary Methods

Fig. S1. Lipid A pathway overview.

Fig. S2. Biological and ecological model of the discoveries made in this study.

Table S1. Reported ranges of Phoenix Islands Region marine mammals.

Table S2. Marine animals from different depths of the ocean.

Table S3. Moritella genomes sequenced in this study.

Table S4. Comparative analysis of all Moritella genomes.

Table S5. Genes found uniquely in immunosilent Moritella.

Table S6. Genes found uniquely in immunostimulatory Moritella.

Table S7. Raw data file (Excel spreadsheet).

References (5776)


Acknowledgments: We wish to thank the Schmidt Ocean Institute and the master and crew of the RV Falkor for ship time to PIPA, cruise FK171005. We are grateful to the PIPA Implementation Office and Conservation Trust for support of this work. We thank the BU Marine Program and C. Johnson (BU) for logistical support and B. R. C. Kennedy for mapping assistance. We thank R. Alegado (U. Hawai’i at Manoa C-MORE) for enabling preparatory materials before sail, and K. Sharp (Roger Williams University) for helpful discussions. Funding: J.C.K. is supported by NIH grants AI133524, AI093589, AI116550, and P30DK34854 and an Investigators in the Pathogenesis of Infectious Disease Award from the Burroughs Wellcome Fund. Our work was conducted under PIPA Research Permit #4/17, funded by NOAA (#NA17OAR0110083 to R.D.R., E.E.C., and T.M.S.). D.R.G. thanks the International Centre for Cancer Vaccine Science project of the International Research Agendas program of the Foundation for Polish Science cofinanced by the European Union under the European Regional Development Fund (MAB/2017/03) at the University of Gdansk. D.R.G. and R.K.E. thank the NIH for funding (AI123820 and AI147314). Author contributions: A.E.G. conceived the idea, designed the study, led experiments, and wrote the manuscript. R.D.R. and J.C.K. conceived the idea, designed the study, and wrote the manuscript. C.E.C., F.M.G., R.S., D.R.G., and R.K.E. designed and conducted structural experiments. V.P. and I.Z. designed in vivo experiments. E.E.C. and T.M.S. performed deep-sea analyses. A.T. designed the deep-sea experiments and conducted the experiments. S.J.B. designed the sequencing experiments and wrote the manuscript. K.S.B. consulted on experimental design and wrote the manuscript. Competing interests: Boston University and Boston Children’s Hospital have filed a patent application entitled “Immunomodulatory LPS compositions” with R.D.R., A.E.G., A.T., and J.C.K. as inventors. J.C.K. holds equity and consults for IFM Therapeutics, Quench Bio, and Corner Therapeutics. None of these relationships influenced the work performed in this study. The other authors declare that they have no competing interests. Data and materials availability: Bacterial strains are available upon request with written permission from the Kiribati Government. Authors will assist in requesting the required permissions. All bacterial genome sequences were deposited in NCBI under BioProject PRJNA639995. Genomic data for four Moritella isolates were deposited in GenBank under accession numbers CP056120 to CP056125 (also see table S3). The 16S amplicon sequence data are available from the NCBI Sequence Read Archive (study SRP267655). All other data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials.

Stay Connected to Science Immunology

Navigate This Article