The CXCR3-CXCL11 signaling axis mediates macrophage recruitment and dissemination of mycobacterial infection

Macrophages are extremely dynamic phagocytic cells, able to integrate and respond to a wide spectrum of signals from infected tissues. A variety of receptors on their cell membrane can sense pathogen-associated molecular patterns (PAMPs), which induce the innate immune response (Medzhitov and Janeway, 2000). Some of these PAMPs, such as N-formylated bacterial peptides, have direct chemoattractant activity on phagocytes (Schiffmann et al., 1975). Moreover, a crucial contribution to efficient phagocyte recruitment is provided by lipidic and peptidic chemoattractant factors, produced or activated directly by the host locally at the infection site (Ford-Hutchinson et al., 1980; Lira and Furtado, 2012; Sun and Ye, 2012). In this group of compounds, the inflammatory chemokines play a major role. This subclass of small chemotactic proteins is induced upon infection and is able to exert target-specific activities towards subsets of leukocytes, both myeloid and lymphoid (Groom and Luster, 2011). In humans, CXCL9 [also known as MIG (monokine-induced by IFN-γ)], CXCL10 [IP-10 (IFN-γ-inducible protein 10)] and CXCL11 [I-TAC (T cell α chemoattractant)] are IFN-inducible chemokines and mediate recruitment of T cells, natural killer (NK) cells and monocytes/macrophages at the infection site, predominantly through their cognate G-protein coupled receptor, CXCR3 (Janatpour et al., 2001; Loetscher et al., 1996). This signaling axis has been implicated in several physiological activities, including maturation of T cells and vasculogenesis (Liu et al., 2005; Zhou et al., 2010). Additionally, CXCR3 and its ligands have been linked to inflammatory and immune-related diseases, of autoimmune (Bondar et al., 2014; Lacotte et al., 2009; Liu et al., 2005; Müller et al., 2010), infectious (Chakravarty et al., 2007; Cohen et al., 2013; Rosas et al., 2005; Seiler et al., 2003) or malignant (Fulton, 2009; Kawada et al., 2007; Oghumu et al., 2014; Pan et al., 2006) nature. Most of the literature on mammalian systems focuses on the role of this receptor in maturation, priming, activation and migration of T cells (Bondar et al., 2014; Liu et al., 2005; Slütter et al., 2013). However, recent studies have demonstrated that CXCR3 also plays an important role in directing macrophage activities, both under physiological and under pathological conditions (Cuenca et al., 2011; Kakuta et al., 2012; Oghumu et al., 2014; Zhou et al., 2010).

The zebrafish embryo model provides a useful platform to study chemokine-dependent cell migration, combining excellent possibilities for intravital imaging with the availability of a vast array of genetic tools (Raz and Mahabaleshwar, 2009). Homologous relationships between mammalian and zebrafish CXCR4-CXCL12 and CXCR2-CXCL8 receptor-ligand pairs have been well established and studies in zebrafish have contributed significantly to the understanding of the role of these signaling axes in developmental processes, neutrophil motility, long-range neutrophil mobilization and infection-induced chemotaxis (David et al., 2002; Deng et al., 2013; Doitsidou et al., 2002; Sarris et al., 2012; Walters et al., 2010). Based on phylogeny reconstructions, the CXCR3-CXCL11 axis emerged for the first time in a common ancestor of zebrafish and mammals (Xu et al., 2014). In placental mammals, amphibians and reptiles, a single copy per haplotype of CXCR3 is generally present, whereas CXCR3 was lost in the divergence of avian and marsupial mammalian clades. Several teleost fish show an expansion of the CXCR3 family (Aghaallaei et al., 2010; Chang et al., 2007; Xu et al., 2014), including zebrafish, where three copies, namely cxcr3.1 (ENSDARG00000007358), cxcr3.2 (ENSDARG00000041041) and cxcr3.3 (ENSDARG00000070669), are located in tandem on chromosome 16 (Nomiyama et al., 2013). The CXCL9-CXCL10-CXCL11 triplet of CXCR3 ligands in mammals is likely to have originated from a relatively recent common ancestor (O’Donovan et al., 1999). The situation in fish is variegated and, in some cases, specific expansions have taken place. In zebrafish, a cluster of seven putative cxcl11 genes, which are grouped together in a single locus on chromosome 5, share both homology and synteny with human CXCL11 (Nomiyama et al., 2013). However, an association between the different isoforms of Cxcl11 ligands and Cxcr3 receptors has not been described, and the in vivo relevance of this signaling axis in the zebrafish model has not been addressed.

In previous work we have shown that one of the three CXCR3 paralogs, cxcr3.2, is expressed in macrophages of 1-day-old zebrafish embryos (Zakrzewska et al., 2010). In the present study we used a cxcr3.2 mutant to investigate the role of Cxcr3 signaling in macrophage mobilization and function. In agreement with previous morpholino knockdown results, the receptor loss-of-function resulted in the attenuation of macrophage recruitment to local infection with Salmonella typhimurium. Moreover, we identified two infection-inducible CXCL11-like chemokines, which act as functional ligands of Cxcr3.2 with chemoattractant activity on macrophages. Finally, we demonstrate here that cxcr3.2 is required for efficient recruitment of macrophages to Mycobacterium marinum infection and for the dissemination of this pathogen into host tissues, which is driven by macrophages. The zebrafish-M. marinum host-pathogen pair is widely used to model human tuberculosis and has provided important insights into the interaction of mycobacteria with host macrophages (Cambier et al., 2014; Clay et al., 2007; Davis et al., 2002; Roca and Ramakrishnan, 2013; Torraca et al., 2014; van der Vaart et al., 2014). M. marinum is closely related to the human pathogen Mycobacterium tuberculosis, and the zebrafish model replicates the formation of granulomas, the typical histopathological hallmark of human tuberculosis (Cronan and Tobin, 2014; Ramakrishnan, 2013). The results presented here demonstrate a novel function for the CXCR3-CXCL11 signaling axis in macrophage responses that drive the initiation and expansion of these granulomatous lesions that are crucial for the dissemination of mycobacterial infection.

We previously reported that cxcr3.2 expression could be detected by fluorescent in situ hybridization at 1 day post fertilization (dpf) in phagocytes expressing the macrophage marker csf1r (colony stimulating factor 1 receptor) and not in cells positive for the neutrophil marker mpx (myeloperoxidase) (Zakrzewska et al., 2010). However, we were unable to detect its expression with the same method at later stages. To determine whether cxcr3.2 continues to be expressed in macrophages during the embryonic and larval development, we analyzed RNA expression levels from FACS-sorted mpeg1:mcherryF⁺ and mpx:eGFP⁺ cells from the double-transgenic line Tg(mpeg1:mcherryF/mpx:eGFP). These data show that macrophages (mpeg1:mcherryF⁺ population) maintain cxcr3.2 expression at 2 and at 6 dpf (Fig. 1A–C and supplementary material Fig. S1). Expression of cxcr3.2 could also be detected in neutrophils (mpx:eGFP⁺ population). In addition, cxcr3.3 could be detected in both phagocyte types, whereas cxcr3.1 was not specifically enriched in the sorted cell populations (Fig. 1C and supplementary material Fig. S1).

Sequencing of an ENU (N-ethyl-N-nitrosourea)-mutagenized zebrafish library resulted in the identification of a cxcr3.2 mutant allele, cxcr3.2^hu6044, which carries a T-to-G (deoxythymidine to deoxyguanosine) substitution, creating a premature stop codon. This mutation leads to the interruption of the protein translation after 15 amino acids, before the region that encodes all the transmembrane domains that are essential for the function of the receptor (Fig. 1D and supplementary material Fig. S2). The nonsense cxcr3.2^hu6044 mutation is not likely to lead to a functional truncated protein by using a downstream AUG codon as a signal for translation initiation. The second AUG in frame is located 354 nucleotides (118 amino acid residues) downstream from the canonical start codon and use of this codon as a translation start would lead to a truncated product lacking both the most N-terminal extracellular domain and the first two transmembrane domains. Furthermore, because the mutation occurs downstream of all the splicing sites, the possibility of alternative splicing and/or altered pre-RNA maturations seems unlikely and this was excluded by sequencing of the cDNA of cxcr3.2 in mutants and wild types (wt). The cxcr3.2 locus is closely linked to the loci of cxcr3.1 and cxcr3.3 owing to their chromosome proximity. To evaluate the presence of additional alterations in these genes as a consequence of the ENU mutagenesis, we sequenced their genetic loci in the AB/TL wt strain in our facility (used to outcross the mutant) and in two families of cxcr3.2^+/+ and cxcr3.2^−/− fish. We did not identify additional nonsense mutations, although we could detect several non-synonymous single-nucleotide polymorphisms (nsSNPs), which are described in supplementary material Table S1. However, all the nsSNPs that were found in the cxcr3.2^−/− line were also present in the AB/TL fish line, indicating that these changes are likely to be natural wt polymorphisms and not an effect of the ENU mutagenesis. To address the possible relevance of these nsSNPs with respect to the protein function, we used the PROVEAN software tool (Protein Variation Effect Analyzer; http://provean.jcvi.org) (Kumar et al., 2009; Choi et al., 2012). None of the nsSNPs was predicted to impact on the protein functionality (supplementary material Table S1). Therefore, it is unlikely that single amino acid replacements in the protein sequences of Cxcr3.1 or Cxcr3.3 would affect the phenotype of fish carrying the cxcr3.2^hu6044 nonsense allele.

Crossing of heterozygous cxcr3.2^hu6044 carriers generated homozygous embryos in Mendelian proportions with their heterozygous and wt siblings. Screening of the adult offspring from a heterozygous incross showed that, in laboratory conditions, cxcr3.2^−/− fish do not exhibit differences in survival and development compared with their wt siblings (Fig. 1E) and could only be distinguished by genotyping.

Based on combined immunostaining for the pan-leukocyte marker Leukocyte-plastin (Lp) and Mpx activity staining (Cui et al., 2011), cxcr3.2^−/− embryos showed similar numbers of macrophages and neutrophils as their wt siblings (Fig. 1F,G). With the aim of investigating the relevance of cxcr3.2 expression in macrophage behavior, we crossed the cxcr3.2 mutation into the Tg(mpeg1:gal4/UAS:kaede) background. Labeled macrophages showed similar numbers and spatial distribution in mutant and wt (Fig. 1H,I). However, a basal macrophage migratory deficiency was observed in the mutants (Fig. 1J–L and supplementary material Movies 1, 2). The aberrant motility might be explained by the presence of constitutive quantities of Cxcr3.2 ligands in the macrophage microenvironment, which could contribute to a higher basal activity of cxcr3.2^+/+ macrophages.

To test whether the basal motility defect of cxcr3.2^−/− macrophages affected the stimulus-directed chemoattraction to cxcr3.2-independent factors, we locally injected the chemoattractant factors leukotriene B4 (LTB₄) and N-formyl-methionyl-leucyl-phenylalanine peptide (fMLP) into the hindbrain ventricle of embryos at 30 hpf (hours post fertilization). At this developmental stage, the neutrophil population is not fully differentiated (Henry et al., 2013; Lieschke et al., 2001) and the population of phagocytes infiltrating the hindbrain upon chemotactic stimulation consists predominantly of macrophages (supplementary material Fig. S3); therefore, we could use Lp immunostaining to identify recruited macrophages. No significant difference was observed in the numbers of cxcr3.2^+/+ and cxcr3.2^−/− macrophages accumulated to either stimuli in 3 hours (Fig. 2A–H). In addition, we also employed a previously described chemically induced inflammation (ChIn) assay (d’Alençon et al., 2010), using copper sulphate treatment of embryos at 3 dpf to induce acute inflammation of lateral line neuromast hair cells. At this stage, macrophages were counted as Lp-positive and Mpx-negative cells using the combined Lp/Mpx staining. Also in this assay, no significant difference in the numbers of macrophages recruited to the inflamed neuromasts was observed between cxcr3.2^+/+ and cxcr3.2^−/− larvae (Fig. 2I). Therefore, we concluded that cxcr3.2 mutation does not affect the capability of macrophages to respond to stimulatory sources independent of Cxcr3.2 signaling and that the basal motility defect does not influence the experimentally induced macrophage recruitment.

To determine whether Cxcr3.2 signaling contributes significantly to the recruitment of macrophages to different types of bacterial infections, we injected either M. marinum or S. typhimurium into the hindbrain ventricle of embryos at 30 hpf. In both infection models, a significant reduction of the number of macrophages accumulating at the infected site was detected in cxcr3.2^−/− embryos at 3 hpi (hours post injection) (Fig. 3A). A similar reduction of macrophage recruitment was also observed when M. marinum was locally injected into the otic vesicle at 3 dpf (Fig. 3B). To visualize the dynamics of the macrophage migration in vivo, we used the combined mutant-transgenic line Tg(mpeg1:gal4/UAS:kaede)/cxcr3.2^−/− and followed the early response of mpeg1-positive cells to M. marinum infection in the otic vesicle of 4 dpf larvae by confocal time-lapse imaging. In agreement with the previous results, a difference in the trend of macrophage recruitment was observed between the mutant and the wt over a time course of 5 hours (Fig. 3C–H; supplementary material Movies 3, 4). Furthermore, at locations distal from the infection site, macrophages in wt larvae showed more frequently an activated morphology with formation of branched protrusions (Fig. 3E) when compared with the mutant line (Fig. 3H). To quantify this phenomenon, we classified the distal macrophages of locally infected larvae according to their circularity index (CI), which estimates by an index between 0 and 1 the level of divergence of the cell shape projection from a perfect circle (CI=1). The different intervals of circularity were differently populated in wt and mutant larvae, with the classes of reduced circularity (0 to 0.4) being more populated in cxcr3.2^+/+ larvae and the classes of higher circularity (0.6–0.8) being more populated in cxcr3.2^−/− larvae (Fig. 3I–L). These results provide evidence that the Cxcr3.2-dependent signaling pathway mediates a significant component of the macrophage recruitment to pathogens in the early phase of the infection. To determine whether the infection-dependent macrophage recruitment can also be modulated pharmacologically, we tested a chemical inhibitor of human CXCR3, NBI74330 (Scholten et al., 2014), which binds with high affinity to a pocket formed by the transmembrane domains of CXCR3. Key amino acid residues in this pocket are conserved between the human and the zebrafish receptors (supplementary material Fig. S4). Treatment with this CXCR3 antagonist attenuated the macrophage recruitment to local M. marinum infection in cxcr3.2^+/+ embryos to a similar level as that of the vehicle-treated cxcr3.2^−/− embryos and did not show a cooperative effect with the cxcr3.2 mutation (Fig. 3M). These results support the conservation of CXCR3 signaling between fish and mammals.

Although our analysis of the cxcr3.2 mutant supports the role of Cxcr3.2 in macrophage chemotaxis to infections, the chemokine ligands that signal via this receptor are unknown. The assignment of ligand-receptor pairs is complicated by the relatively poor conservation of chemokine sequences among vertebrates and the species-specific expansions of the chemokine gene clusters (Nomiyama et al., 2013). However, systematic study of the orthologous relationships between vertebrate chemokines indicated that seven CXCL11-like chemokine genes, located in tandem on chromosome 5 (cxcl11aa, cxcl11ac, cxcl11ad, cxcl11ae, cxcl11af, cxcl11ag, cxcl11ah), have evolved in zebrafish as a counterpart to the mammalian CXCL9, CXCL10 and CXCL11 genes (Nomiyama et al., 2013). The amino acid similarity between the CXCL11-like chemokines in zebrafish and human CXCL11 exceeds the similarity that human CXCL9, CXCL10 and CXCL11 show among each other (supplementary material Fig. S5; Table S2). We therefore considered the zebrafish CXCL11-like chemokines as putative ligands for the Cxcr3.2 receptor. Because Cxcr3.2 was clearly involved in the early phase of the infection response, we reasoned that the ligands that induce Cxcr3.2-mediated chemotaxis should be promptly upregulated upon local infection. For this reason, we collected RNA samples from whole embryos infected in the hindbrain with 200 colony-forming units (CFU) of either S. typhimurium or M. marinum at 1 and 3 hpi and designed gene-specific primers for the members of the cxcl11 gene cluster. Because of the high level of sequence conservation between cxcl11af and cxcl11ag (only 2 bp difference on the cDNA leading to a single semiconservative residue change of an aspartic acid with a glutamic acid), a promiscuous primer pair was used that can amplify both gene transcripts (cxcl11af/ag). Analysis by qRT-PCR revealed that, at 1 hpi, cxcl11af/ag showed twofold upregulation with M. marinum infection and fourfold upregulation with S. typhimurium infection, although no statistical significance was observed compared with the mock-injected controls (Fig. 4A). At 3 hpi the expression of cxcl11aa, cxcl11ae and cxcl11af/ag was significantly upregulated to levels of two- to 4.5-fold (Fig. 4B). In particular, cxcl11aa displayed the highest levels of induced transcription (~4.5-fold induction) with both the pathogens tested, suggesting this chemokine as an effective signaling ligand of Cxcr3.2 involved in the response to infection. We verified that the genes induced by local infection were also responsive to systemic infection with M. marinum. Upregulation of cxcl11aa and cxcl11ae was detected both at 4 hpi and at 4 dpi (days post injection) during M. marinum systemic infection, whereas cxcl11af/ag was significantly induced to levels of ~fourfold only at the later stage of infection (Fig. 4C,D). Additionally, at this time point, cxcl11ac and cxcl11ad were also significantly upregulated to levels of 1.5- to two-fold (Fig. 4D).

To assess the chemoattractant properties of the infection-inducible chemokines Cxcl11aa, Cxcl11af and Cxcl11ae, we used Pichia pastoris strain X-33 to express recombinant proteins. As a control, we also expressed zebrafish Cxcl8a (Il8), known to be a potent and neutrophil-specific chemoattractant (Deng et al., 2013; Sarris et al., 2012). All three purified CXCL11-like chemokines showed chemoattractant capabilities towards macrophages when locally injected in vivo into the hindbrain at 30 hpf (Fig. 5), whereas no significant macrophage recruitment was exerted by Cxcl8a (supplementary material Fig. S6). Similar levels of these chemokines were injected in the otic vesicle at 54 hpf to evaluate their chemoattractant capabilities towards neutrophils (supplementary material Fig. S6B,C). Cxcl11aa and Cxcl11af did not show chemoattraction of neutrophils under these conditions, whereas Cxcl11ae and Cxcl8a exerted significant neutrophil chemoattraction. To determine whether the macrophage chemoattraction is dependent on cxcr3.2, hindbrain injections of the recombinant proteins were performed in both wt and cxcr3.2 mutants. Both Cxcl11aa and Cxcl11af did not stimulate recruitment upon local injection in cxcr3.2 mutants when compared with their mock controls (Fig. 5A,C). In contrast, the chemoattraction of phagocytes mediated by Cxcl11ae was independent of cxcr3.2 mutation (Fig. 5B; supplementary material Fig. S6C). Taken together, these results support a direct ligand-receptor interaction between Cxcr3.2 and the chemokines Cxcl11aa and Cxcl11af that mediates the chemoattraction of macrophages. Differently, Cxcl11ae, which exerted a Cxcr3.2-independent phagocyte chemoattraction, is likely to signal via a yet-unidentified receptor.

Pathogenic mycobacteria have the ability to resist intracellular macrophage digestion and they can use the macrophages as a vector for distal dissemination of the infection (Clay et al., 2007). We hypothesized that cxcr3.2 depletion, preventing a high level of macrophage accumulation to the local infection site, might also prevent extensive dissemination and help to locally restrict the infection. To test this hypothesis we followed M. marinum hindbrain infection for 24 hours and evaluated the frequency of infection dissemination in cxcr3.2^+/+ and cxcr3.2^−/− zebrafish embryos. At 24 hpi, almost 50% of the wt embryos displayed dissemination of the infection from the head to the trunk and tail, whereas, in more than 80% of the mutants, the infection remained locally confined (Fig. 6A–F). Dissemination to other areas of the head could be seen already as early as 6 hpi and also this phenotype was attenuated in cxcr3.2^−/− embryos (Fig. 6G–I). Disseminated bacteria outside the hindbrain and/or midbrain were residing in phagocytes and, in time course experiments, we could visualize that egression of mycobacteria from the ventricles is facilitated by macrophages (Fig. 6J), in agreement with previously published results (Clay et al., 2007). When dissemination to the tail and trunk occurred, one to five dissemination foci could be detected in the cxcr3.2^+/+ embryos, whereas cxcr3.2^−/− embryos never showed more than one or two bacterial clusters distally from the original injection point. At 5 dpi, the bacterial burden in the hindbrain was similar between wt and mutant larvae, but mutants still showed lower levels of dissemination of the infection towards distal areas (Fig. 7A,B). The infection foci generated distally developed into typical granuloma-like aggregates, as previously described for the zebrafish-M. marinum model (Davis et al., 2002). The size of these granulomatous lesions was significantly reduced in the cxcr3.2^−/− mutant larvae (Fig. 7C–F). Therefore, we concluded that cxcr3.2-mediated signaling strongly influences the dynamics of the infection progression and of granuloma formation. To further investigate the relevance of cxcr3.2 signaling in the formation of granulomas, we injected 200 CFU of M. marinum systemically in 1 dpf embryos via the caudal vein, which, in the wt leads to many granulomatous lesions (Clay et al., 2007). Images of single granulomas at 5 dpi, stained for both macrophages and neutrophils, revealed that granuloma-like aggregates could still be formed in cxcr3.2 mutants. Similar structures and phagocyte compositions were observed when lesions of similar sizes in wt and mutant were compared (Fig. 8A,B). However, it must be noted that a large variation in granuloma structure and composition already exists when comparing different granulomas within the same larva or between different wild types, and this makes it very difficult to assess the effect of a mutation on the general architecture of the granulomas. Despite this, we observed that Cxcr3.2 deficiency provided partial protection against mycobacterial infection. Not only did mutants exhibit reduced levels of infection burden (Fig. 8C–E), but also a reduced number of bacterial clusters (Fig. 8F) and smaller average bacterial cluster size (Fig. 8G). Taken together with the results of hindbrain infection, these data demonstrate the important role of Cxcr3.2-dependent signaling in guiding macrophage-mycobacteria interactions, and show how this signaling leads to direct effects on the infection progression.

The chemokine receptor CXCR3 and its ligands play important roles in the pathogenesis of infectious diseases, autoimmune disorders and cancer (Bondar et al., 2014; Chakravarty et al., 2007; Cohen et al., 2013; Cuenca et al., 2011; Kakuta et al., 2012; Kawada et al., 2007; Oghumu et al., 2014; Pan et al., 2006; Rosas et al., 2005; Seiler et al., 2003; Slütter et al., 2013). In this study, we report on the function of CXCR3 signaling in macrophage recruitment to infection foci and in the early establishment of mycobacterial granulomas. We found that the Cxcr3.2 receptor, one of the three zebrafish homologs of human CXCR3, interacts with infection-inducible zebrafish homologs of the CXCL11 ligand family and is required for the mobilization of macrophages to different pathogens, such as locally delivered M. marinum or S. typhimurium. Furthermore, mutation of cxcr3.2 reduced the macrophage-mediated dissemination of M. marinum, leading to attenuation of the formation and expansion of granulomatous lesions in both local and systemic models of mycobacterial infection.

CXCR3 is best known as a canonical marker for Th1 cells, but several recent studies have raised interest in the expression of this receptor by macrophages. These studies have implicated CXCR3 signaling in processes as diverse as the recruitment of macrophages to allografts (Kakuta et al., 2012), the macrophage-mediated remodeling of blood vessels (Zhou et al., 2010) and the polarization of macrophages towards an M2 phenotype that promotes tumor progression (Oghumu et al., 2014). Furthermore, CXCR3 signaling has been shown to play a crucial role in the murine neonatal response to sepsis (Cuenca et al., 2011). Like murine neonates, zebrafish embryos and early larvae rely heavily on their innate immune system for defense against infection. During zebrafish embryogenesis, macrophages are the first leukocyte cell type to develop and they express cxcr3.2 from day 1 (Zakrzewska et al., 2010). In mutants of cxcr3.2, or in wt embryos treated with a human CXCR3 antagonist (NBI74330), we observed a significant reduction in the recruitment of macrophages to local bacterial infection in the hindbrain. In contrast, Cxcr3.2-deficient macrophages were able to normally migrate in response to chemically induced wounding or towards Cxcr3.2-independent chemoattractants, such as LTB₄ and fMLP. These data suggest that Cxcr3.2 signaling is specifically activated by pathogen-induced chemokine signals. We considered a cluster of CXCL11-like chemokines as the putative ligands of Cxcr3.2 and confirmed that two of these, Cxcl11aa and Cxcl11af, exerted chemoattractant activity on macrophages following hindbrain injection of the recombinant proteins. Most likely, Cxcl11ag, which is near-identical to Cxcl11af, also signals through Cxcr3.2. It is currently unknown whether the cxcl11 genes in zebrafish are IFN-γ-inducible like their mammalian counterparts, but IFN-γ responsive elements are present in the promoters of these genes (van der Aa et al., 2012). In addition, because we were unable to detect expression of the zebrafish cxcl11 genes in situ, the cell types producing these chemokines remain to be established. However, qRT-PCR showed rapid upregulation of cxcl11aa and cxcl11af/ag gene expression following infection, supporting their function as the ligands mediating the infection-dependent recruitment of Cxcr3.2-positive macrophages.

Expression analysis on FACS-sorted phagocyte populations showed that also cxcr3.3 is expressed in macrophages, but macrophage motility and recruitment defects in the cxcr3.2 mutant line indicates that expression of cxcr3.3 cannot compensate for the loss of function of cxcr3.2. In addition, the expression analysis revealed that also neutrophils express cxcr3.2 at 2 and 6 dpf. Injection of Cxcl11aa or Cxcl11af into the otic vesicle at 2 dpf did not chemoattract a higher number of neutrophils within 3 hours than mock injections, whereas comparable concentrations of these chemokines were able to recruit macrophages into the hindbrain, and comparable concentrations of Cxcl8a and Cxcl11ae mobilized neutrophils when delivered in the otic vesicle. Different explanations can be given for this effect. Firstly, it is possible that different concentrations of chemokines are required to efficiently chemoattract different cell types. Secondly, the requirement of co- stimulatory signals or cell-specific co-receptors might be different between the phagocyte populations. Thirdly, although macrophages and neutrophils at 2 dpf seem to express comparable levels of cxcr3.2 mRNA, it remains unknown whether similar protein levels of Cxcr3.2 are exposed on their membranes. It should be noted that macrophages and their progenitors are marked by cxcr3.2 expression already at 1 dpf, whereas its expression could not be detected at this time point on neutrophil progenitors (Zakrzewska et al., 2010). In line with this consideration, it is possible that this different timing in messenger expression impacts the protein levels at 2 dpf.

Macrophages are essential for the dissemination of pathogenic mycobacteria and mediate the formation of both primary and secondary granulomas in the zebrafish host following infection with M. marinum (Clay et al., 2007; Davis et al., 2002). When M. marinum was locally injected into the hindbrain ventricle of 1-day-old embryos, almost half of the embryos exhibited dissemination within 24 hours, where single infected macrophages migrated out of the ventricle and localized distally. In cxcr3.2 mutants, this dissemination of the infection was significantly reduced, which might be a consequence of the diminished macrophage attraction to the primary infection source or a direct effect on the retromigration ability of cxcr3.2 mutant macrophages. When the bacteria were injected intravenously, cxcr3.2 mutation reduced the formation and the expansion of granulomas, thereby attenuating the dissemination of bacteria and the overall burden of systemic infection. This phenotype might be explained by the reduced motility of macrophages in cxcr3.2 mutants, because it has been shown that early granulomas in zebrafish larvae expand by spreading of the infection to newly recruited macrophages (Davis and Ramakrishnan, 2009). In agreement, the phenotype of cxcr3.2 mutant larvae resembles those caused by deficiency in other host (mmp9) or bacterial (ESAT-6) factors that also impair macrophage recruitment (Volkman et al., 2010; Davis and Ramakrishnan, 2009).

The zebrafish larval tuberculosis model is limited to the study of the initial stages of granuloma formation by macrophages in a context where the adaptive immune system is not yet functional. A beneficial effect of CXCR3 mutation has also been observed during chronic infection of BALB/c mice with Mycobacterium tuberculosis (Chakravarty et al., 2007). In this model the resistance of CXCR3-deficient mice was attributed to the function of CXCR3 in T-cell priming. Another study using C57BL/6 mice showed that CXCR3 mutation affected early granuloma formation after aerosol M. tuberculosis infection and correlated this with the invasion of polymorphonuclear neutrophils that produce chemokine signaling via CXCR3 (Seiler et al., 2003). Together, the studies in mice and zebrafish models support further investigation of the CXCR3 signaling axis as a host therapeutic target for tuberculosis. Our study is the first to implicate this signaling axis in macrophage responses that drive the initiation and expansion of mycobacterial granulomas. In future work it will therefore be of great interest to investigate how macrophage and T-cell responses determined by CXCR3 signaling cooperate in the control of mycobacterial infections, using adult zebrafish or mammalian models of tuberculosis.

Recently, another chemokine receptor, Ccr2, has also been shown to mediate macrophage recruitment following hindbrain infection of M. marinum in zebrafish embryos (Cambier et al., 2014). This Ccr2-mediated pathway is dependent on the presence of phenolic glycolipids on the mycobacterial cell surface and it recruits a population of macrophages that are permissive for mycobacterial growth, because activation of the host immune response is largely avoided owing to the presence of other cell surface lipids in virulent mycobacteria (phthiocerol dimycoceroserate lipids), which physically mask the underlying PAMPs. M. marinum bacteria lacking phenolic glycolipids were still able to recruit macrophages, and morpholino knockdown of either Ccr2 or its ligand Ccl2 attenuated recruitment but did not fully abolish it. These observations indicate that redundant and/or synergistic mechanisms are cooperating in macrophage mobilization. Combined experiments will be necessary to reveal whether the Ccr2-Ccl2 axis is (partially) redundant or synergistic with the Cxcr3-Cxcl11-mediated macrophage recruitment shown here.

Interestingly, we found that Cxcr3.2 is also involved in the basal motility of macrophages under physiological conditions. We hypothesize that the lower basal motility of macrophages in cxcr3.2 mutants could be due to the inability to sense small amounts of Cxcr3.2 ligands secreted in the macrophage microenvironment. Possibly, the macrophages themselves could be involved in an autocrine or paracrine secretion of these ligands. Similar mechanisms acting via CXCR3 signaling have already been described in the literature. Keratinocytes have been shown to express CXCL10 and CXCR3 to guide their own migration for re-epithelialization in a wound-healing response (Kroeze et al., 2012). Similarly, synovial fibroblasts use this ligand-receptor pair to regulate their invasion of joints in rheumatoid arthritis (Laragione et al., 2011). Furthermore, myeloid cells and hematopoietic progenitors secrete many different chemokines, including CXCR3 ligands, to regulate hematopoiesis in an autocrine or paracrine manner (Majka et al., 2001). The autocrine or paracrine production of Cxcr3.2 ligands could potentially work as a local macrophage stimulator, which might significantly contribute to the surveillance activities that macrophages exert in tissues. During mycobacterial disease, similar mechanisms might be stimulated within the core of the granulomatous lesions, and could be involved in the in-and-out trafficking properties of macrophages, characteristic of these dynamic structures (Chensue, 2013; Fuller et al., 2003). In various animal models of tuberculosis, including the most clinically relevant macaque model, abundant expression of CXCR3 ligands is detected in the core and in the direct neighborhood of the granulomatous lesions (Aly et al., 2007; Fuller et al., 2003; Khader et al., 2009). Our study suggests that this is relevant not only for the recruitment of T cells, but also for regulating macrophage activities in the immunopathology of the granulomatous lesion. A number of studies with selective agonists or antagonists of CXCR3 have already shown beneficial effects on inflammation-associated diseases (O’Boyle et al., 2012; van Wanrooij et al., 2008), and the zebrafish model might be a suitable model to test their effectiveness on mycobacterial infections.

Concluding, here we propose a dual biological role of Cxcr3.2-Cxcl11aa/af ligand-receptor signaling. First, our results implicate Cxcr3.2 and its ligands in surveillance against pathogens by promoting the random patrolling of inactive macrophages. Second, we show that this pathway is involved in the mobilization of macrophages during infection. Depending on the specific interactions of different pathogens with their hosts, a Cxcr3.2-dependent response could be beneficial for the resolution of infection or have an unfavorable effect because, on the one hand, it can sustain the recruitment of macrophages to the infection site, but, on the other hand, it can promote the dissemination of bacteria, as in the case of mycobacterial infection.

Zebrafish lines were handled in compliance with the local animal welfare regulations and maintained according to standard protocols (zfin.org). The breeding of adult fish was approved by the local animal welfare committee (DEC) of the University of Leiden (license number: 10612) and adhered to the international guidelines specified by the EU Animal Protection Directive 2010/63/EU. Adult zebrafish were not sacrificed for this study. All experiments in this study were performed on embryos/larvae before the free-feeding stage and did not fall under animal experimentation law according to the EU Animal Protection Directive 2010/63/EU.

Fish lines used in this work were the following: wild-type (wt) strain AB/TL, double-transgenic line Tg(mpeg1:mcherryF/mpx:eGFP) (Bernut et al., 2014; Renshaw et al., 2006), homozygous mutant (cxcr3.2^−/−) and wt siblings (cxcr3.2^+/+) of cxcr3.2^hu6044, Tg(mpeg1:gal4-VP16/UAS-E1b:kaede), in short referred to as Tg(mpeg1:gal4/UAS:kaede) (Ellett et al., 2011), and the combination of Tg(mpeg1:gal4/UAS:kaede) with the cxcr3.2 mutant strain. The cxcr3.2^hu6044 allele was identified by sequencing of an ENU (N-ethyl-N-nitrosourea)-mutagenized zebrafish library and was obtained from the Hubrecht Laboratory and the Sanger Institute Zebrafish Mutation Resource. Heterozygous F2 carriers were outcrossed twice against wt, and were subsequently incrossed. Resulting cxcr3.2^−/− and cxcr3.2^+/+ siblings were raised and used to obtain embryos for all the experiments. The combined mutant-transgenic line Tg(mpeg1:gal4/UAS:kaede/cxcr3.2^−/− or cxcr3.2^+/+) were obtained by crossing heterozygous carriers with the original transgenic line and subsequently incrossing the heterozygous offspring. For genotyping, genomic DNA was amplified using forward primer 5′-GGCATCTTTTTTGTTACAGCCTACAGCTTA-3′ and reverse primer 5′-TGGCGATATCGGCGGATAACA-3′, amplifying a 201 base pair (bp) product containing the mutation. The forward primer introduces an additional base change, which only in combination with the mutant allele generates the consensus for DdeI restriction enzyme. Therefore, the mutant allele was distinguished from the wt by specific digestion into a 174 fragment that can be separated from the undigested wt amplicon on a 2.5% agarose gel. Alternatively, genotyping was performed by KASP assay using the primers 5′-CATCATAGGAAGTACTGTTGTAGTCA-3′, 5′-CATCATAGGAAGTACTGTTGTAGTCC-3′ and 5′-GGCATCTTTTTTGTTACAGCCTACAGATT-3′. Robustness of both methods was verified several times by sequencing of the amplicons.

Embryos were grown at 28.5°C in egg water (60 μg/ml sea salt, Sera marin, Heinsberg, Germany). For live-imaging or injection assays, larvae were anesthetized in egg water medium containing 0.02% buffered Tricaine (3-aminobenzoic acid ethyl ester; Sigma-Aldrich, St Louis, MO, USA). To prevent melanization, larvae were generally maintained in egg water supplemented with 0.003% PTU (1-phenyl-2-thiourea; Sigma-Aldrich).

Sequencing of the full coding sequence of cxcr3.1, cxcr3.2 and cxcr3.3 was obtained by amplification with primers described in supplementary material Table S3. For cxcr3.2 and cxcr3.3, both genomic and cDNA templates extracted from pools of 15–20 embryos were used. Amplification was performed with Phusion high-fidelity DNA polymerase (Thermo-Scientific, Pittsburgh, PA, USA). DNA amplicons were then gel-extracted on 1.5% agarose and column-purified with PureLink quick gel extraction and PCR purification kit (Invitrogen, Life Technologies, Carlsbad, CA, USA). Sequencing with M13Fw, M13Rv universal primers (incorporated in the amplification primers) or with custom-made primers was outsourced to Baseclear (Leiden, The Netherlands). For cxcr3.1, sequencing results derive exclusively from genomic DNA amplifications. Amplification of cDNA templates for cxcr3.2 resulted in a band of identical size in mutant, wt and AB/TL, thereby excluding altered exon/intron arrangements attributable to the ENU-mutagenesis per se or to the cxcr3.2^hu6044 allele.

Approximately 200 CFU (1 nl) of M. marinum strain Mma20 expressing mCherry (van der Sar et al., 2004), or Salmonella enterica serovar Typhimurium (S. typhimurium) strain SL1027 expressing DsRed (van der Sar et al., 2003), were grown and harvested as described previously (Benard et al., 2012; Cui et al., 2011). Embryos were staged at 30 hpf and bacteria or mock control [phosphate buffer saline (PBS) supplemented with 0.1% phenol red (Sigma-Aldrich) and 2% polyvinylpyrrolidone-40 (Sigma-Aldrich)] were locally injected in the hindbrain cavity as described previously (Benard et al., 2012; Cui et al., 2011). Injections of bacteria in the otic vesicle, as shown in Fig. 3, were performed either at 3 dpf (Fig. 3B,I–L) or at 4 dpf (Fig. 3C–H). When infection was delivered systemically, the same dose was instead injected in the caudal vein as in Benard et al. (Benard et al., 2012). As a control, the same dose was spotted onto plates, incubated and counted. Embryos were kept into fresh PTU egg water, incubated at 28.5°C, and collected for qRT-PCR or used for imaging at 1–6 dpf. In Fig. 8A,B, embryos were fixed at 6 dpf in 4% paraformaldehyde in PBSTx (1× PBS supplemented with 0.8% Triton X-100; Sigma-Aldrich) and prepared for Myeloperoxidase (Mpx) activity stain with TSA staining kit (PerkinElmer Inc., Waltham, MA, USA), followed by immunostaining against the pan-leukocyte marker Leukocyte-plastin (Lp) as described previously (Cui et al., 2011).

Mpeg1:mcherryF-positive, mpx:eGFP-positive and unlabeled cells were sorted from Tg(mpeg1:mcherryF/mpx:eGFP) at 2 and 6 dpf. FACS-protocol and RNA isolation were performed according to Rougeot et al. (Rougeot et al., 2014). To evaluate the induction of the cxcl11 genes upon infections, pools of 18–20 embryos were collected for RNA isolation, snap-frozen in liquid nitrogen and subsequently stored at −80°C. RNA was extracted using Qiazol reagent (Qiagen, Valencia, CA, USA) according to the manufacturer’s guidelines. Residual genomic DNA was removed by DNA-free kit (Ambion, Life Technologies). The cDNA was prepared using the iScript cDNA-synthesis kit (Invitrogen, Life Technologies) and was used as a template for qRT-PCR reaction with iQ SYBR Green Supermix according to the manufacturer’s instructions (Bio-Rad Laboratories, Munich, Germany). Specificity of the amplification reaction was analyzed using dissociation curves. Each qRT-PCR was performed in technical duplicate and on biological replicates as indicated in the figure legends. Reference genes were eif4a1b or eif5 (eukaryotic translation initiation factor 4a isoform 1b or 5) for FACS-sorted cells and ppiab (peptidylprolyl isomerase ab/cyclophilin a) for infection experiments. Fold changes were determined using the ΔΔ comparative threshold method. Primers are reported in supplementary material Table S3.

Synthetic coding sequences for Cxcl11aa, Cxcl11af, Cxcl11ae and Cxcl8a (included as negative control) were generated (Baseclear) according to database accessions (supplementary material Table S2). To enable secretion in yeast, the sequences were codon optimized and the predicted zebrafish signal peptide was replaced with yeast alpha-factor secretion signal, as a result from cloning into pPICZα expression vector (Invitrogen, Life Technologies). Additionally, a HA (human influenza hemagglutinin)-tag was added at the C-terminus to facilitate the purification process and identification. The recombinant chemokines were produced by Pichia pastoris strain X-33 transformed with the chemokine vectors as described previously (Wu and Letchworth, 2004). Proteins were purified via Fast Protein Liquid Chromatography in NaCl salt gradient and finally desalted and concentrated by membrane filtrations on Amicon Ultra Centrifugal filter devices with a nominal molecular weight limit of 3 kilodaltons (Amicon, Merck KGaA, Ireland), using 50 mM sodium phosphate buffer pH 6.5 as a washing and suspension vehicle. Purity and identity of the proteins were confirmed by trypsinization and electrospray mass-spectrometry. The recombinant chemokines (0.5–1.5 mg/ml), LTB₄ (leukotriene B4; Santa Cruz Biotechnology, Santa Cruz, CA, USA; 10.1 ng/ml), fMLP (N-formyl-methionyl-leucyl-phenylalanine; Sigma-Aldrich; 0.2 mg/ml) or mocks [sodium phosphate buffer pH 6.5 for the chemokines, 5% DMSO (Sigma-Aldrich) in PBS for fMLP and 0.02% ethanol (Sigma-Aldrich) in PBS for LTB₄] were supplemented with 0.1% phenol red and injected at 30 hpf in the hindbrain ventricle (1 nl) or at 52 hpf in the otic vesicle (0.5 nl) as described previously (Benard et al., 2012; Cui et al., 2011). In both cases, embryos were fixed at 3 hpi in 4% paraformaldehyde in PBSTx and prepared for Lp immunostaining as in Cui et al. (Cui et al., 2011). At 30 hpf, the population of fully differentiated leukocytes is represented almost exclusively by macrophages (Henry et al., 2013; Lieschke et al., 2001); thereby, we could assume that nearly all the Lp-stained cells able to migrate and infiltrate in the ventricle represented macrophages at this developmental stage. As is shown in supplementary material Fig. S3, only one to two mpx-positive cells [mpx-whole mount in situ hybridization as in reference (Cui et al., 2011)] could be counted within the perimeter of the hindbrain in this experimental setting at 3 hours post local bacterial infection, which is less than 10% of the cells positive for the macrophage marker mfap4. At later developmental stages, in order to discern between neutrophils and macrophages, samples were processed also with a neutrophil-specific Mpx activity staining as described previously (Cui et al., 2011), by using the leukocyte peroxidase (Myeloperoxidase) staining kit (Sigma-Aldrich) for the histochemical detection of the enzymatic activity of Mpx. Leukocytes accumulated at the injected cavity (macrophages: Lp-positive and Mpx-negative; neutrophils: Mpx-positive) were counted using a Leica MZ16FA fluorescence stereomicroscope (Leica Microsystems, Rijswijk, The Netherlands).

3-dpf larvae were exposed to 10 μM copper sulphate (CuSO₄; Sigma-Aldrich) for 2 hours as described previously (d’Alençon et al., 2010). Treated larvae were then fixed and used for combined Mpx activity staining and Lp immunostaining as described above.

Bath-treatment with the CXCR3 high-affinity antagonist NBI74330 or vehicle treatment (0.5% DMSO) was started at 27 hpf by exposing dechorionated embryos to 50 μM of the drug in medium. Embryos were incubated for 3 hours at 28.5°C and then injected in the hindbrain with mock or M. marinum as described above. Injected embryos were maintained for an additional 3 hours in 50 μM NBI74330 or vehicle alone and then fixed in 4% paraformaldehyde/PBSTx and prepared for Lp immunostaining as described previously (Cui et al., 2011).

Fixed or live embryos and larvae were imaged using a Leica MZ16FA fluorescence stereomicroscope. For time-lapse experiments, samples were mounted in 2% low-melting-point agarose (SphaeroQ, Burgos, Spain) and images were acquired with a laser-scanning confocal microscope (Leica TCS SPE, Leica Microsystems or Zeiss Observer 6.5.32, Carl Zeiss, Sliedrecht, The Netherlands). To assess the average speed of macrophages (Fig. 1J,K), a time-lapse experiment was performed and quantification was obtained on overlaid z-stacks by Fiji/ImageJ software (NIH, Bethesda, MD, USA) using the ManualTrack plug-in as described elsewhere (Meijering et al., 2012). The average speed was calculated as the average of all the speeds assumed by every single macrophage at each time point. Analysis was performed by cumulating three experiments in which 15–21 macrophages per embryo were followed. To quantify the morphological differences between macrophages in cxcr3.2 mutants and wt, bacteria were injected into the otic vesicle of Tg(mpeg1:gal4/UAS:kaede/cxcr3.2^+/+ or cxcr3.2^−/−) larvae at 3 dpf and fixed at 4 hpi. Images of macrophages were acquired in the trunk. Perimeter and area of the cells were obtained by Fiji/ImageJ using the Analyze Particles plug-in. The circularity index (CI) corresponding to each cell was obtained by the formula: CI=4π (area/perimeter²), resulting in an index that ranges from 0 (infinitely branched structure) to 1 (perfect circle). Macrophages were classified in five different intervals of circularity based on their CI (0.0 to 0.19, 0.2 to 0.39, 0.4 to 0.59, 0.6 to 0.79, 0.8 to 1.0) and the average percentages of macrophages in each interval were estimated for cxcr3.2^+/+ or cxcr3.2^−/− (Fig. 3I). To estimate the divergence of distribution of cxcr3.2^+/+ and cxcr3.2^−/− macrophages from the overall mean, the percentages in each interval were divided by the average percentage of mutants and wt assumed in that interval, using the formulas: Deviation_(cxcr3.2+/+)=%cxcr3.2^+/+/[(%cxcr3.2^+/+ + %cxcr3.2^−/−)/2] and Deviation_{(cxcr3.2−/−)}=%cxcr3.2^−/−/[(%cxcr3.2^+/+ + %cxcr3.2^−/−)/2] (Fig. 3J). To quantify the dissemination of bacterial infection (Fig. 6F,I), the presence or absence of infection distally from the infected site was evaluated, giving a score of 1 in case of dissemination and a score of 0 in case of absent dissemination. Quantification of total bacterial pixels (Fig. 7A,B and Fig. 8E) was obtained using dedicated bacterial pixel count program as in Stoop et al. (Stoop et al., 2011). Total bacterial cluster count (Fig. 8F) was performed manually from images. Quantification of the area of single distal clusters (Fig. 7C) and average area of disseminated granulomas (Fig. 8G) were performed using ImageJ quantification tools as in Elks et al. (Elks et al., 2013).

In the survival test (Fig. 1E), non-significant deviation from Mendelian rate was evaluated by χ² test on four independent replicates. For qRT-PCR, statistical significance was estimated on five (Fig. 1A–C), four (Fig. 4A,B) or three (Fig. 4C,D) biological replicates by two-tailed t-tests on ln(n)-transformed relative induction folds. All the other experiments were statistically analyzed using GraphPad Prism 4 or 5 (GraphPad Software, La Jolla, CA, USA). Where correction for non-parametric distribution was required (Fig. 1F,G; Fig. 2; Fig. 3A,B; Fig. 3M; Fig. 5; Fig. 6; Fig. 7A,B; Fig. 8E,F), comparisons between two groups were performed with two-tailed Mann-Whitney test and comparisons among more than two groups were performed with Kruskal-Wallis test, followed by Dunn’s multiple comparison test. When a parametric distribution was assumed (Fig. 1A–C; Fig. 1L; Fig. 3I,J; Fig. 4; Fig. 8G), comparisons between two groups were performed with two-tailed t-test. In Fig. 7C, significance was estimated with an unpaired t-test with Welch’s correction, suitable to compare parametric data having different variances. Significance (P-value) is indicated with: ns, non-significant; *P<0.05; **P<0.01; ***P<0.001.

We thank Ewa Snaar and Claudia Tulotta for valuable discussions. We are grateful to Fons Verbeek and Alex Nezhinsky (Leiden Institute of Advanced Computer Science) for making their pixel quantification software available prior to publication and to Erica Benard for RNA samples of M. marinum systemically infected embryos. For providing the zebrafish knockout allele cxcr3.2^hu6044 we thank the Hubrecht Laboratory and the Sanger Institute Zebrafish Mutation Resource.