Transient infection of the zebrafish notochord with E. coli induces chronic inflammation

Osteomyelitis, chondritis and septic arthritis are inflammatory diseases of bones, cartilage and joints, respectively, that are often caused by bacterial infections. These diseases are caused by local or systemic immune responses that lead to leukocyte infiltration, damage to extracellular matrix, compression of vasculature and the progressive destruction of bone and joints (Wright and Nair, 2010). The bacteria that are involved are predominantly Staphylococcus aureus, Streptococcus spp. and members of Gram-negative Enterobacteriaceae, including Escherichia coli (Gutierrez, 2005; Lew and Waldvogel, 2004). Although the pathogen is often present within the infected site, it is occasionally undetectable, despite the persistence of immunological symptoms. Substantial evidence supports the importance of innate immunity in the physiopathology of these diseases; however, its exact role is still poorly understood (Lew and Waldvogel, 2004; Wright et al., 2010).

The innate immune response is the first defence of the host in the case of infection. Macrophages and neutrophils are two major effectors of innate immunity against microbes. Neutrophils are the first to be recruited to the infected tissue; they migrate rapidly by responding to gradients of chemoattractant molecules – such as the contents of damaged cells, bacterial products and host chemokines (Deng et al., 2013; Renshaw and Trede, 2012). Macrophages are recruited later in order to aid the removal of dead cells and microbes; they play a role in remodelling the injured tissue and activating adaptive immunity. Neutrophils and macrophages are both professional phagocytes, but the efficiency of phagocytosis depends on the site at which they encounter bacteria (Colucci-Guyon et al., 2011).

Although leukocytes can migrate rapidly and specifically to an infection site, most tissues have specific resident macrophages permanently patrolling the tissue, circulating through tissues to eliminate cellular debris and microorganisms. From this immunological perspective, bone and cartilage are very specific tissues in which, owing to their highly structured and resistant extracellular matrix, no specialised phagocytes patrol. The best-documented example is that of bones. Resident bone macrophages, OsteoMacs, are not present within the bone matrix but, with their stellate morphology, do extensively cover bone surfaces (Pettit et al., 2008). They probably act as guards to prevent microbes from entering or attacking the bone. By contrast, osteoblasts (that make bone) and osteoclasts (monocyte-macrophage lineage cells that resorb bone) are capable of phagocytosis, probably to compensate for the absence of professional phagocytes within the tissue. Some pathogens have even developed specific strategies to colonise these cells (Wright and Nair, 2010). Although documented to a lesser extent, the situation in cartilage is similar, with chondrocytes displaying phagocytic activity and a highly phagocytic cell population that specifically expresses the macrophage marker CD163, present below the superior zone at the periphery of cartilage (**Castillo and Kourí, 2004; Jiao et al., 2013).

To address the complexity of the chronic inflammation of bones and joints, a number of mammalian infection models have been developed (Patel et al., 2009). Most of them rely on the injection of Staphylococcus aureus, or other bacteria, directly into the bone in combination with an adjuvant or a trauma. Owing to these models, advances have been made in understanding the disease and in tuning antibiotic therapies; however, to understand the complexity of dynamic interactions during inflammation, non-invasive and tractable models in which pathogens and immune cells can be imaged in the context of three-dimensional tissue architecture in the host are required. Owing to its exceptional transparency, the zebrafish embryo is an exquisite vertebrate model for the study of tissue-specific infections and the inflammatory consequences within a homologous tissue.

At the embryonic and early-larval stages, zebrafish do not possess bone. However, the notochord, a transient embryonic structure that is present in all chordates, in addition to its essential roles in vertebrate patterning during development, serves as an axial skeleton of the embryo until other structures, such as the vertebrae, form (Stemple, 2005). The notochord comprises vacuolated cells that are encased in a sheath of collagen. Osmotic pressure within the vacuoles exerts such a pressure on the thick collagenous sheath that it forms a hydrostatic skeleton in the embryo and young larva. Notochordal cells that express many cartilage-specific genes are most closely related to chondrocytes, but as development proceeds, they become ossified in regions of developing vertebrae and contribute to the formation of the centre of the intervertebral discs in a structure called the nucleus pulposis (Stemple, 2005). The notochord can be studied to dissect the molecular mechanisms that are involved in the immune response to infection in a chrondrocyte-like tissue. Previous results from our laboratory have shown that the injection of an attenuated mutant of Mycobacterium marinum, TesA:Tn, into the notochord allows active replication of the bacteria, whereas the infection is easily controlled by professional phagocytes when injected at other locations (Alibaud et al., 2011). The drawback of this infection model is that the active replication of the mycobacteria within the notochord ultimately leads to the death of the host. As neither macrophage nor neutrophil infiltration is observed (by using electron microscopy), we hypothesised that the notochord is an organ that is inaccessible to leukocytes and that this organ could be a good model in which to study the outcome of an infection in a tissue where leukocytes cannot reach the pathogen.

Here, we show that the injection of non-pathogenic Gram-negative bacteria Escherichia coli K12 into the notochord of zebrafish embryos induces a strong immune response without compromising the survival of the host. Using a combination of imaging and genetic techniques, we investigate the fate of the bacteria in this organ, analyse the interaction between the bacteria and the leukocytes, describe the inflammation induced by the infection and analyse the consequences of this inflammation.

To characterise the immune response to notochord infection, we first injected 3×10³ bacteria of an Escherichia coli K12 strain that expressed the fluorescent protein DsRed (E. coli-DsRed) into the notochord of zebrafish embryos at 45 hours post-fertilisation (hpf) (Fig. 1A). Injection through the collagen sheath, just posterior to the yolk extension, leads to the anterior sliding of the bacteria around the notochordal cells inside of the collagen sheath. Fifteen minutes after injection, the embryos were fixed, and the localisation of the bacteria was checked by using transmission electron microscopy (TEM). Bacteria were found inside the notochord, mainly at the periphery, pinned against the collagen sheath by the turgescent notochordal cells (Fig. 1B,C). The fate of the bacteria in the notochord was monitored by live-imaging microscopy and compared to the fate of bacteria that were injected inside muscles. When injected into muscles, bacteria were restricted to the injected somite for the first hours, but after 24 hours, their fluorescent remnants were observed within motile cells that migrated away from the site of injection (Fig. 1D). By contrast, in most cases (73%, 57 out of 78), the day following injection into the notochord, fluorescent bacteria were not observed in the notochord, suggesting that bacteria had been removed (Fig. 1E). In the other 27% (21 out of 78) – presumably those embryos that received the higher bacterial load – bacteria proliferated within the notochord (Fig. 1F–I) and led to larval death within 3 days, as illustrated by the reduced survival of the embryos in which bacteria had been injected into the notochord (Fig. 1J). Further experiments were performed using lower bacterial loads (2×10³); at this load, all embryos survived and bacterial division was not observed (by using TEM; data not shown). The clearance of bacteria from the notochord of most of the infected fishes was confirmed by counting colony-forming units (CFUs) (Fig. 1K), indicating that bacteria, if not too numerous, are efficiently cleared from the notochord within the first 24 hours of infection.

During the first weeks of development, the zebrafish larva has no functional adaptive immune system, relying mostly on macrophages and neutrophils, to fight infections (Lieschke et al., 2001; Meijer and Spaink, 2011; Renshaw and Trede, 2012). Therefore, we determined whether these cells were able to reach the bacteria in the notochord in order to phagocytose and/or kill them. To study neutrophil-microbe interactions, we injected fluorescent E. coli into Tg(mpx:GFP) embryos, which expressed green fluorescent protein (GFP) under the control of the neutrophil-specific promoter of myeloperoxidase (mpx) (Renshaw et al., 2006), and we then imaged neutrophil behaviour by using confocal time-lapse microscopy from 2 hours post-infection (hpi). In contrast to PBS-injected controls (data not shown), neutrophils were primarily attracted to the injection site, then, arriving from the ventral side of the larvae (Fig. 2A,B; supplementary material Movies 1, 2), they crawled along the outside of the infected notochord, spreading towards the anterior region (Fig. 2B; supplementary material Movie 1). Although neutrophils appeared to be attempting to try to phagocytose bacteria (i.e. crawling forward and backward in the vicinity of bacteria, emitting cytoplasmic protrusions), they failed to engulf them (supplementary material Movie 3). Only a few neutrophils contained engulfed bacteria, but these leukocytes were carefully, and retrospectively, traced from the injection site, suggesting that they phagocytosed a few E.coli that were deposited at the injury site of the notochord (Fig. 2B; supplementary material Movie 1). This behaviour is in stark contrast with that observed in muscles, where many neutrophils displayed phagosomes that were full of bacteria at 2 hpi and still efficiently engulfed bacteria (data not shown; Colucci-Guyon et al., 2011).

To analyse macrophage-bacteria interactions, we constructed the Tg(mpeg1:mCherryF) line that expressed red fluorescent macrophages. Transgenic larvae were infected either in the bloodstream or in the notochord, and macrophage behaviour was recorded at 1 hpi. Similar to neutrophils, although less mobile, macrophages were recruited to the infected notochord and extended pseudopodal protrusions in attempts to engulf bacteria (Fig. 2A,C; supplementary material Movies 4, 5). In contrast to the numerous macrophages that were laden with E. coli after bloodstream infection (supplementary material Movie 6), macrophages were inefficient at engulfing the bacteria present in the notochord. It is noteworthy that no significant difference in the time to the first recruitment was detected between neutrophils and macrophages (data not shown).

The fact that leukocytes easily engulfed bacteria at the injection site, where the notochordal collagen sheath had been damaged by the injection, but not in the undamaged upstream region suggests that bacteria might not be accessible to leukocytes because of the physical properties of the notochord (data not shown). We tested this hypothesis by performing three-dimensional reconstruction of confocal acquisitions. From 2 hpi onwards, bacteria were found in the phagosomes of neutrophils and macrophages after injection into muscle (Fig. 3A) and the bloodstream, respectively (Fig. 3C); however, they remained next to, but outside of, immune cells after notochord infection (Fig. 3B,D). Indeed, many leukocytes accumulated around the notochord but could not infiltrate it, even at 24 hpi (Fig. 3E–G).

To confirm further that leukocytes could not reach the bacteria, and engulf them, in the notochord, we analysed the ultrastructure of the notochord of infected larvae by using TEM at different timepoints following infection. At 15 minutes post-infection, bacteria were found in the periphery of the notochord, between notochordal cells and the collagen sheath as mentioned above (supplementary material Fig. S1A–C). At 2 hpi and 4 hpi, bacteria remained in the same location – in the extracellular space – although they were less numerous at 4 hpi (supplementary material Fig. S1D–F; data not shown for the 4 hpi timepoint). Replicating bacteria were never observed under these conditions. This correlated with the observation that DsRed fluorescence in E. coli remained constant in the notochord for a few hours following injection before diminishing completely. Although a few leukocytes were detected outside of the notochord (data not shown), none were found inside, confirming that leukocytes cannot phagocytose bacteria in this structure (supplementary material Fig. S1). Strikingly, numerous E. coli displayed an altered morphology (supplementary material Fig. S1B,C,F), suggesting they are eliminated by a humoral lysis mechanism. No altered morphologies were observed when higher doses of bacteria were used, suggesting that the humoral lysis mechanism is overpowered when the notochord is injected with higher doses of E. coli.

We analysed the kinetics of immune cell recruitment using Tg(mpx:GFP) and Tg(mpeg1:mCherryF) lines by counting the leukocytes that accumulated around the infected notochord. As for muscle infections (Colucci-Guyon et al., 2011), notochord infections led to neutrophil recruitment within the very first hours following the injection. After injection of bacteria into the muscle, the number of neutrophils that had been recruited started decreasing at 24 hpi (Fig. 4A,C), whereas, in the notochord, neutrophil recruitment continued to increase at 48 hpi and remained high up to 6 days post-infection (dpi), forming one to four clusters that were towards the anterior of the injection site (20 out of 26 fish; Fig. 4A,C). Macrophages migrated to both infected muscle and notochord in a similar manner (Fig. 4B) – in both cases, the recruitment of macrophages lasted for 6 days, but more macrophages were recruited to the infected notochord (23 out of 23 fish; Fig. 4B,D). The observed immune response was not a consequence of the injection per se in the notochord, because the injection of PBS or fluogreen Dextran caused neither neutrophil nor macrophage recruitment (Fig. 4C,D; supplementary material Fig. S2).

We conclude that infection of the notochord induces a massive recruitment of macrophages and neutrophils to its periphery; remarkably, this inflammatory response is not properly resolved, as leukocyte accumulation persists for days, despite the clearance of bacteria in a matter of hours.

Inflammation is fundamental in order to control bacterial infections. However, it is crucial that this inflammation resolves in a timely manner in order to prevent damage to the surrounding tissues. In order to analyse the consequences of notochord inflammation, we imaged larvae, in which bacteria had been injected into the notochord, for 10 days. At 5 dpi, most of the infected larvae presented curved malformation and lesions of the notochord, upstream of the injection site (43 out of 49 fish; supplementary material Fig. S3A,B). Although the defects were lessened, we still observed this malformation at 10 dpi (supplementary material Fig. S3C). Histological analysis of PBS-injected larvae at 5 dpi revealed normal architecture with highly vacuolated cells that were surrounded by a regular collagen sheath. By contrast, the infected larvae displayed swelling of the notochord at various locations along the antero-posterior axis (Fig. 5A,B; supplementary material Fig. S3D–K). At these spots, we observed an ectopic collagen-rich extracellular matrix, and an increased number of cells and vacuoles inside the notochord, suggesting the presence of supernumerary notochordal cells, rather than a phenomenon of fragmented vacuoles. Cells of a smaller size were also found clustered either outside or inside of the notochord; they were often trapped within the matrix, suggesting that they are infiltrating leukocytes (supplementary material Fig. S3J–K). Analysis by using confocal microscopy revealed that, although leukocytes were positioned at the margin of the myotome in PBS-injected larvae, they started infiltrating the notochord in E. coli-injected larvae from 2 dpi (Fig. 5C,D shows the 5 dpi timepoint; the data at 2 dpi is not shown), a timepoint that is long after bacterial clearance. Because notochord defects colocalised with neutrophil clusters (Fig. 5E,F), we asked whether neutrophil infiltration contributes to the destruction of host tissues. Indeed, this infiltration was concomitant with neutrophil degranulation – 35% of neutrophils in the inflammation region harboured no myeloperoxidase-containing granules, whereas all neutrophils in the control retained them (Fig. 5G–J). Taken together, these results show that infection of the notochord eventually leads to severe damage to the notochord tissue, which correlates with late leukocytic infiltration and neutrophil degranulation.

As the notochord participates to vertebral column formation we asked whether the inflammation of the notochord could trigger vertebral column defects. We thus examine the axial skeleton of juvenile fishes at 60 days post-fertilization (dpf) using Alizarin red staining. Zebrafish that had been previously injected with E. coli in the notochord developed scoliosis and dysmorphic vertebral column in their anterior region (upstream the injection site), characterised by fused and misshaped vertebral bodies and absence of intervertebral disc (11 out of 11 fish; Fig. 5K). No defects were observed in the PBS injected controls, excepting a single vertebral body that appeared more calcified at the injection site in few fishes (two out of 19 fish; Fig. 5K; data not shown).

In order to decipher the molecular mechanisms that orchestrate the inflammatory response to notochord infection, we analyzed the expression of many cytokines and chemokines in larvae that had been infected in the notochord (supplementary material Table S1). The expression of most of the chemokines that we examined was unchanged, such as that of the chemokine Cxcl-c5c; however, transcripts of the proinflammatory cytokine Il1b were strongly upregulated from 1 dpi onwards (Fig. 6A,B).

To study the spatiotemporal behaviour of the cells that produce the il1b transcripts, we established a transgenic line that expressed farnesylated GFP (membrane-targeted GFP; GFP-F) under the control of the il1b promoter Tg(il1b:GFP-F). Uninfected Tg(il1b:GFP-F) developing embryos expressed GFP-F during the segmentation period in the elongating tail bud (supplementary material Fig S4). At 24 hpf, GFP expression appeared in tail keratinocytes, the olfactory epithelium and individual cells on the yolk sac that behave as primitive leukocytes. At 35 hpf, when leukocytes appear in the caudal hematopoietic tissue (CHT), scattered GFP⁺ cells were also found in the CHT region. From 50 hpf, GFP was expressed in keratinocytes at the tip of the caudal fin, in fin buds, retina, neuromasts, gills and thymus (supplementary material Fig. S4; data not shown). Whole-mount in situ hybridisation analysis confirmed the expression of endogenous il1b mRNA in the different organs that are cited above, showing that Tg(il1b:GFP-F) recapitulates il1b transcriptional expression (supplementary material Fig. S5A,C). Furthermore, upon stimulation by ‘danger signals’, both the neutrophils and macrophages of Tg(il1b:GFP-F) larvae strongly expressed GFP, correlating with the upregulation of the endogenous il1b mRNA that was observed in both leukocyte populations (supplementary material Fig. S6).

To study il1b transcription upon infection, we injected fluorescent bacteria (E. coli-DsRed) into the notochord of Tg(il1b:GFP-F) embryos. Specific expression of GFP was detected along the notochord as early as 18 hpi, which increased significantly by 48 hpi and was maintained up to 5 to 6 dpi before decreasing (24 out of 36 fish; Fig. 6D; supplementary material Fig. S5D). The induction of GFP upon infection correlated with the upregulation of il1b mRNA (supplementary material Fig. S5A–C). Conversely, control PBS and muscle injections, respectively, induced no and low expression of GFP around the injection site at 24 hpi and 48 hpi (Fig. 6C,E; supplementary material Fig. S5D). The analysis of larvae that had been injected in the notochord revealed that GFP was expressed by leukocytes that were clustered around the notochord but not by notochordal cells (Fig. 6F). Additional GFP expression was detected in leukocytes that were scattered throughout the body and in the keratinocytes surrounding the inflammation site (supplementary material Fig. S5E,F; Fig. 6J).

By early timepoints (18 hpi) 30% of GFP-positive cells (il1b⁺) also expressed the macrophage marker mpeg1 (mpeg1⁺), whereas only 10% expressed the neutrophil marker lysosymeC (lyz⁺) (Fig. 6G,L). During the late stages of inflammation (48 hpi), two distinct populations of GFP-expressing cells were observed around the inflamed notochord – cells that expressed a high level of GFP [il1b^high, 60±6.5% (mean±s.e.m.)] and those that expressed a low level of GFP (il1b^low, 40±6.5%). At 48 hpi, no macrophages were il1b^high, whereas neutrophils (lyz⁺) comprised 68% of these il1b^high cells. The other 32% of il1b^high cells were mpeg1⁻ and Lyz⁻ and unidentifiable. The remaining il1b^low cell population comprised 35% neutrophils (il1b^low Lyz⁺) and 27% macrophages (il1b^low mpeg1⁺) (Fig. 6H–L).

In conclusion, our data suggest that Il1b production dynamically shifts from early timepoints of the inflammatory response – at which both macrophages and neutrophils contribute to Il1b production – to late phases, when neutrophils are the main source of Il1b.

Morpholino-mediated gene knockdown was used to investigate the role of zebrafish Il1b during severe notochord inflammation. Reverse transcription (RT)-PCR analysis, performed using primers on both sides of the splice site that is targeted by the morpholino, provided evidence of strong blocking of splicing, whereas overall morphology was not noticeably affected (Fig. 7A; the general morphology of the morphants is not shown). The il1b morpholino was injected into Tg(mpx:GFP ; mpeg1:mCherryF) embryos, and the morphants were infected by the injection of E. coli into the notochord. In embryos that had not been injected with morpholino, or those that had been injected with control scrambled morpholino, both neutrophils and macrophages were recruited to the inflamed notochord in a normal manner. In il1b morphants, neutrophil recruitment was slightly reduced at 4 hpi and strongly reduced at 48 hpi compared with controls (Fig. 7C,E–G); only macrophages localised around the notochord of il1b morphants at 4 and 48 hpi (Fig. 7G; supplementary material Fig. S7). The decreased neutrophil recruitment was not caused by a general reduction of the neutrophil population under steady-state conditions because the total number of neutrophils in il1b morphants was similar to that in uninjected embryos or embryos that had been injected with the control scrambled morpholino at 4 hpi (Fig. 7D for il1b morpholino compared to scrambled morpholino; data not shown for il1b morphants compared to uninjected larvae). Additionally, a reduction of the overall inflammatory response was excluded because macrophage recruitment was unaffected (Fig. 7G). Long-term examination of the infected il1b morphants revealed that loss of il1b prevented formation of the lesions that were often detected in the rostral part of the notochord after infection (Fig. 7H–J, lesions were found in one out of 12 il1b morphant fish and seven out of 11 scrambled morpholino fish). Taken together, these results establish an important and specific role for Il1b in the recruitment of neutrophils during chronic inflammation of the notochord and subsequent sequelae.

This article describes a model of zebrafish notochord infection with E. coli that recapitulates several of the inflammatory aspects of bone and cartilage infection in human. We thus propose that the specific advantages of the larval zebrafish model system – transparency, genetic tractability – should facilitate the study of the early cellular and molecular inflammatory mechanisms that are related to matrix-tough tissue infections.

The similarity of the notochord with cartilage (in terms of transcriptional profile) and vertebral bone (in terms of structure and function) (Fleming et al., 2001; Stemple, 2005) prompted us to establish a model of the infection of this organ in order to mimic the inflammatory aspects of human osteomyelitis and osteoarthritis. As in some infections of human bones and joints, infection of the zebrafish notochord induces a severe inflammation that greatly outlasts the presence of live microbes and thus can be considered ‘chronic’ – taking into account due allowances. Indeed, injection of E. coli into the notochord leads to the rapid recruitment of both neutrophils and macrophages to the infected notochord within the first hour. The injection needle locally disrupts the collagen sheath, creating an opening for the leukocytes to phagocytose some bacteria in the notochord. However, probably because of the turgidity of the notochordal cells within the thick collagen sheath, neutrophils and macrophages fail to enter the notochord at sites distant from the injection site. This finding suggests that, in contrast to what is observed for transendothelial migration, leukocytes do not easily pass through all matrixes. Even though the injected bacteria are cleared in less than one day, neutrophil and macrophage recruitment along the notochord persists for days, preventing the resolution of inflammation. Although they, initially, randomly spread around the notochord at 2 to 24 hpi, leukocytes accumulate in patches in the later stages of the response (from 48 hpi), suggesting that they are involved in the activation and recruitment of more leukocytes. Granules, the hallmarks of granulocytes, are stores of enzymes that participate in killing microbes and digesting tissues. We found that, from 2 dpi, despite the fact that injected microbes have been cleared by this point, neutrophils degranulate next to the notochord. Thus, the release and activation of proteolytic enzymes that are able to disassemble the collagen fibrils that impede neutrophil-bacteria contacts might explain the infiltration of neutrophils into the notochord, which is observed in the late phases of inflammation (Nathan, 2006). This neutrophil degranulation might be responsible for the tissue damage that is observed in notochord-infected larvae – including perturbation of extracellular matrix deposition, swelling and curvature of the notochord, and cellular clusters within the notochord that are reminiscent of the abscesses observed in human osteomyelitis (Lew and Waldvogel, 2004; Wright and Nair, 2010). The notochord lesion, which is induced by the severe inflammatory episode might contribute to the dismorphic vertebrae that were observed during adulthood. Although extracellular matrix deposition, neutrophil infiltration and tissue damage are the hallmarks of chronic infection, which occur in some osteomyelitis and septic arthritis individuals, curvature of the axial skeleton is reminiscent of that of observed in individuals that have bone tuberculosis. Indeed, also known as Pott disease, which accounts for 1–2% of overall tuberculosis cases, it is usually characterised by the spread of tuberculosis bacilli into the disc space from the vascular system, resulting in the destruction of the disc and progression to the bone of adjoining vertebral bodies. As the vertebrae degenerate and collapse, a kyphotic deformity results (Moon, 1997). The chronic inflammation of the zebrafish notochord after bacterial clearance described here is also reminiscent of autoinflammatory disorders – such as chronic non-bacterial osteomyelitis, deficiency of IL-1β receptor antagonist or neonatal-onset multisystem inflammatory disease. Although the etiology of chronic non-bacterial osteomyelitis is still unknown, deficiency of IL-1β receptor antagonist and neonatal-onset multisystem inflammatory disease are caused by mutations in the IL1RN and NLRP3 genes, respectively, and result in over-activation of the IL-1β pathway (Morbach et al., 2013). Accordingly, neutrophil recruitment to the infected zebrafish notochord was found to be dependent on Il1b.

Many bacteria avoid the immune response by hiding inside of cells of their host. By doing so, they are out of reach of the immune system and multiply within these cells before further invasion of the body. We have previously shown that, when injected into the notochord, non-pathogenic Mycobacterium marinum Tes:A mutant bacteria are internalised by notochordal cells as revealed, using TEM, by the presence of the bacteria in endosomes (Alibaud et al., 2011). Interestingly, similar bacterial internalisation has previously been reported for osteoblasts and chondrocytes (Diaz-Rodriguez et al., 2009; Wright and Nair, 2010; Jiao et al., 2013, **Castillo and Kourí, 2004). All of these observations suggest that osteoblasts, chondrocytes and notochordal cells are non-professional phagocytes that engulf invader microbes. However, their phagocytic ability depends on the nature of these microbes. We observe here that notochordal cells do not phagocytose E. coli; bacteria remain extracellular for a few hours before being eliminated, as demonstrated by CFU assays, and fluorescence and electron microscopy. How are they killed? As leukocytes cannot phagocytose them, another mechanism must be used. The ‘unhealthy’ shape of some E. coli within the notochord, as observed by electron microscopy, suggests death by a humoral lysis mechanism that remains to be identified. The complement system is an obvious candidate, and we are currently studying the contribution of complement genes to the observed bacterial clearance.

IL-1β is a critical pro-inflammatory mediator of the host response to microbial infection. It acts on almost all cell types and has many functions – including growth and proliferation, the induction of adhesion molecules in epithelial cells, the activation of phagocytosis and release of proteases in macrophages and neutrophils, and the activation of B and T cells (Dinarello, 2009). Using transgenic zebrafish larvae that expressed farnesylated GFP under the control of the il1b promoter, we show here, for the first time, the dynamic transcriptional activity of il1b in a whole organism. Its expression during zebrafish development, especially in the keratinocytes of the caudal and pectoral fin bud, correlates with the regions of the body where active keratinocyte proliferation occurs. This is in agreement with findings in mammals where keratinocytes express IL-1β (Kupper et al., 1986). However, resting keratinocytes express IL-1β in its unprocessed biologically inactive form – pro-IL-1β. Indeed, to be secreted as an active cytokine, pro-IL-1β must be cleaved by caspases (caspase-1, caspase-11 and others), which are themselves activated through inflammasome assembly (Kayagaki et al., 2011). Although il1b transcription is low in most cell types and strongly inducible following the detection of microbial molecules – notably in myeloid cells – constitutive expression of pro-IL-1β by some tissues should allow for a faster response to external injuries. This is logical for cells that are directly exposed to the environment – such as keratinocytes, olfactory epithelial cells, enterocytes or neuromast cells – all cell types that were seen to spontaneously express endogenous il1b mRNA and GFP in the Tg(il1b:GFP-F) line. In zebrafish, Il1b has been shown to be structurally conserved compared with its human counterpart, and the Il1b pathway is also dependent on caspases (Ogryzko et al., 2014). Whether the active form of Il1b is produced in these zebrafish organs remains to be determined. Notably, spontaneous GFP expression in some cell types, such as those of neuromasts and the gut, mimics the spontaneous expression of GFP in the Tg(NFκB:EGFP) reporter transgenic zebrafish line (Kanther and Rawls, 2010), which is in agreement with the well-known contribution of NF-κB to il1b induction (Ogryzko et al., 2014).

In addition to its basal expression during development, GFP is also induced following E. coli infection in Tg(il1b:GFP-F) fish, which parallels endogenous il1b induction. This induction is strictly correlated in intensity and time with the recruitment of leukocytes. Indeed, infection in muscles induces a low level of expression that quickly resolves, whereas infection of the notochord induces strong and persistent expression. The GFP expression level decreases one day before the resolution of inflammation. In most mammalian systems, IL-1β has been reported to be mainly produced by circulating monocytes, tissue macrophages and dendritic cells (Dinarello, 2009). Here, we describe that, in our zebrafish notochord infection model, il1b is first expressed in macrophages during the early phase of inflammation and is mainly expressed by neutrophils during the late phases. This is reminiscent of the recent work of Karmakar and collaborators, which shows that neutrophils are the main source of Il1b in response to Pseudomonas aeruginosa in mouse, highlighting the importance of activated neutrophils in mediating inflammation (Karmakar et al., 2012).

One interesting question arising from the present work is which signal stimulates the recruitment of leukocytes to the infected notochord? Is it a primary signal that is actively emitted by the notochordal cells, does it come from the injected bacteria or does it correspond to bacterial leftovers after destruction within the notochord? We are currently investigating this latter hypothesis through the injection of different bacterial products into the notochord of larvae and evaluating their ability to recruit leukocytes.

Our data demonstrate that il1b is not expressed by notochordal cells after E. coli injection; however, it is partially required for the initial recruitment of neutrophils, even though a low number of neutrophils is still recruited to the infected notochord in the absence of il1b. Il1b is also required for the persistence of neutrophils at the inflammation site, but it has no effect on macrophage recruitment, suggesting that Il1b is a signal produced by recruited leukocytes themselves that amplifies the activation of neutrophils and prevents resolution of inflammation. If neutrophils do not encounter a bacterium, they are thought to degranulate to promote tissue digestion in order to gain access to bacteria (Nathan, 1987). Indeed, we show that the notochord is not directly accessible to neutrophils, thus degranulation might produce cellular and matrix debris that maintain inflammation around the notochord. Therefore, multiple different signals probably induce and maintain the inflammatory response.

IL-1β is a key player in the inflammatory process that occurs in bone diseases (Wright and Nair, 2010). The neutralisation of IL-1β signalling appears to be crucial for the potential therapy of osteomyelitis, but also for other inflammatory diseases, such as rheumatoid arthritis. Several specific drugs that inhibit this pathway are currently used to improve the symptoms of these diseases (Dinarello, 2004). Interestingly, in our model, knocking down il1b suppressed the long lasting inflammation. We therefore consider the zebrafish model to be perfectly adapted to the challenge of dissecting the exact function of il1b during chronic inflammation of bone and joints.

All animal experiments that are described in the present study were conducted at the University Montpellier 2 according to European Union guidelines for the handling of laboratory animals (http://ec.europa.eu/environment/chemicals/lab_animals/home_en.htm) and were approved by the Direction Sanitaire et Vétérinaire de l’Hérault and Comité d’Ethique pour l’Expérimentation Animale under reference CEEA-LR-13007.

Fish maintenance, staging and husbandry were as described previously (Alibaud et al., 2011) under standard conditions (Westerfield, 2007). Experiments were performed using the F1 golden zebrafish mutant (Lamason et al., 2005), originally purchased from Antinea, and using the transgenic line Tg(mpx:GFP) (kind gift from Steve Renshaw, MRC Centre for Developmental and Biomedical Genetics, Sheffield, UK) to visualise neutrophils. Embryos were obtained from pairs of adult fish by natural spawning and raised at 28.5°C in tank water. Embryos and larvae were staged as described previously (Kimmel et al., 1995).

Student’s t-tests (two-tailed, unpaired) were performed in Fig. 1K; Fig. 4C,D; Fig. 5J; Fig. 6L and Fig. 7D–G.

For statistics on the survival of larvae that had been infected with E. coli, the survival rate of each of the populations of injected embryos was compared with that of the uninjected embryos using the log-rank (Mantel–Cox) test using GraphPad Prism 4 (Graphpad Software).

Based upon the work of Ellet et al. (Ellett et al., 2011), we cloned a new version of the mpeg1 promoter to drive the specific expression of a red fluorescent membrane-targeted protein. A 1.9-kb fragment of the mpeg1 promoter was amplified from genomic DNA using primers zMpeg1P4 (5′-TTGGAGCACATCTGAC-3′) and zMpeg1E2N2 (5′-TTATAGCGGCCG -CGAAATGCTCTTGACTTCATGA-3′), digested by NotI and then ligated to the reading frame of the farnesylated mCherry (mCherry-F) protein so that the mpeg1 AUG was in-frame with the downstream mCherry-F open reading frame. The plasmid was injected in one-cell stage embryos together with the Tol2 transposase RNA. F0-microinjected embryos were screened using the expression of the transgene in macrophages, as described previously (Ellett et al., 2011), and stable lines were established. The same strategy was used to clone a 3.4-kb fragment of the il1b promoter using primers P2 (5′-ATGAGCTCCGAAATTCAGCTGG-3′) and E2N (5′-TTATAGCGGCCGCTTGCCCGCATGCCATCATTTCTA-3′).

To determine the relative expression of il1b and cxcl-c5c, total RNA from infected larvae and controls (pools of ~10 larvae each) was prepared at different timepoints post-infection. RNA preparation, reverse transcription and quantitative real-time PCR, including statistical analysis, have been described elsewhere (Aggad et al., 2009). The primers used were as follows: zIL1b.fw, 5′-TGGACTTCGCAGCACAAAATG-3′; zIL1b.rv, 5′-GTTC -ACTTCACGCTCTTGGATG-3′; zEF1a.fw, 5′-TTCTGTTACCTGGCAAAGGG-3′; zEF1a.rv, 5′-TTCAGTTTGTCCAACACCCA-3′; CXCL-C5C.fw, 5′-AACCAAGTCCATCCTGCTCA-3′; CXCL-C5C.rv, 5′-CGTTAGGATCCAAACACACCT-3′.

For il1b loss-of-function assays, we designed morpholino antisense oligonucleotides that specifically hybridised with the splicing site between intron 2 and exon 3 – morpholino sequence 5′-CCCACAAACTGCAAAATATCAGCTT-3′. The oligonucleotides against il1b (3 nl at a final concentration of 0.5 mM), or scrambled morpholinos (5′-AATCACAAGCAGTGCAAGCATGATG-3′, 3 nl at a final concentration of 0.5 mM) were injected into one-cell stage embryos. No side effect was observed. Efficiency was tested by using RT-PCR with primers for both sides of the morpholino target sequence: zil1b.30, 5′-TGCCGGTCTCCTTCCTGA-3′ and zil1b.50, 5′-GCAGAGGAACTTAACCAGCT-3′ (Fig. 7A). Sequencing confirmed the retention of the second intron of il1b, leading to a frame shift and premature stop codon. Larvae were infected with Escherichia coli K12 bacteria that harboured DsRed (van der Sar et al., 2003), Crimson (Clontech), or a GFP expression plasmid and had been grown in Luria broth (LB) with appropriate antibiotics. Overnight stationary-phase cultures were centrifuged (1 minute, 10,000 g), and the bacteria were resuspended in sterile PBS (2×10⁹–4×10⁹ CFU/ml) and then injected into the muscle, caudal vein or notochord of 48- to 50-hpf larvae (Alibaud et al., 2011; Colucci-Guyon et al., 2011).

Larvae were anaesthetised in 0.01% 3-amino benzoic acidethylester (Tricaine), positioned on 35-mm glass-bottomed dishes (WillCo-dish^®), immobilised in 0.8% low-melting-point agarose and covered with 2 ml of embryo water containing Tricaine. Epifluorescence microscopy was performed by using an MVX10 Olympus MacroView microscope that was equipped with MVPLAPO 1× objective and XC50 camera. Confocal microscopy was performed using a Leica SPE upright microscope with 40× HCX APO L 0.80 W and 20× CHX APO L 0.5 W objectives. Time-lapse acquisitions were performed at 23–26°C. The 4D files generated by the time-lapse acquisitions were processed using ImageJ. Three-dimensional reconstructions and optical sectioning were performed on the four-dimension files using Imaris (Bitplan AG, Zurich, Switzerland).

Individual larvae were lysed in 1% Triton-X100 and homogenised with a 26G syringe five to six times. The resultant debris and notochords were washed twice in PBS and incubated for 30 minutes at 30°C with agitation in the presence of 10 mg/ml collagenase that had been dissolved in PBS supplemented with 1 mM CaCl₂. A fraction of the total lysate was plated on selective LB agar plates.

Larvae were fixed at 4-5 days post-infection in Bouin’s solution and washed in 70% ethanol. After successive dehydration, butanol and Paraplast baths, larvae were embedded in Paraplast X-TRA (Sigma), sectioned (using a Leitz Wetzlar Microtome), stained with Masson’s trichrome and imaged (using a Nanozoomer Slide scanner with 40× objective). For bone staining, juvenile fish were fixed in 4% paraformaldehyde in PBS for 48 hours at 4°C, dehydrated with 100% ethanol over 4 days, stained with Alizarin red (Javidan and Schilling, 2004) and transferred to 100% glycerol over the course of 1 week before analysis and storage.

Double transgenic larvae Tg(mpx:eGFP/mpeg1:mCherry-F) were injected with E. coli K12 or PBS into the notochord at 2 dpf. At 2 dpi, 40 anaesthetised larvae per sample were allocated to each well of a 6-well plate and dissociated in 2 ml of trypsin 0.25% with 0.5 mM EDTA for 1 hour at 33°C with pipetting every 10 minutes for correct dissociation. Reactions were stopped by the addition of CaCl₂ (final concentration 1 mM) and foetal calf serum (final concentration 10%). The lysates were passed through a 40-μm mesh filter, washed twice with 1% FCS in PBS and sorted by fluorescence-activated cell sorting (MoFlo Astrios EQ) for GFP-labelled neutrophils and mCherry-labelled macrophages for collection directly into lysate buffer solutions for mRNA extractions using the RNAqueous-micro kit (Ambion).

Double transgenic larvae Tg(il1b:eGFP/mpeg1:mCherry-F) or Tg(il1b:eGFP/LysC:mCherry) at 4 dpf were anaesthetised and trypsinised individually in 6-well plates, as described above. After washing with 1% FCS in PBS, the cell suspensions were cultured in Leibovitz L-15 medium (21083-027, Gibco) for 1 hour at 30°C. Non-attached cells were then removed, and the attached macrophages and neutrophils were supplied with fresh L-15 medium and incubated at 30°C for an additional 3 hours before confocal imaging.

We thank Karima Kissa, Paul Guglielmi and the members of DIMNP for helpful discussions; Evelyse Grousset for her help with histological sectioning; Pierre-Henri Commère for fluorescence-activated cell sorting, Steve Renshaw for providing the Tg(mpx:GFP) line and Vincent Domergue for helping to generate the Tg(il1b:GFP-F) line.