Abstract
Innate lymphoid cells (ILCs) were originally classified based on their cytokine profiles, placing natural killer (NK) cells and ILC1s together, but recent studies support their separation into different lineages at steady-state. However, tumors may induce NK cell conversion into ILC1-like cells that are limited to the tumor microenvironment and whether this conversion occurs beyond this environment remains unknown. Here we describe Toxoplasma gondii infection converts NK cells into cells resembling steady-state ILC1s that are heterogeneous and distinct from both steady-state NK cells and ILC1s in uninfected mice. Most toxoplasma-induced ILC1s were Eomes-dependent, indicating that NK cells can give rise to Eomes− Tbet-dependent ILC1-like cells that circulate widely and persist independent of ongoing infection. Moreover, these changes appear permanent, as supported by epigenetic analyses. Thus, these studies markedly expand current concepts of NK cells, ILCs, and their potential conversion.
Introduction
Innate lymphoid cells (ILCs) comprise diverse populations. Like T and B cells, they are derived from common lymphoid progenitors (CLPs) but do not undergo antigen receptor recombination (Spits and Cupedo, 2012). ILCs are tissue-resident within mucosal sites where they participate in tissue homeostasis and survey for pathogens (Sojka et al., 2014a, Gasteiger et al., 2015), although recent studies showed that inflammation can induce ILC2 mobilization (Huang et al., 2018). ILC classification was initially based on their cytokine production, with type 1 ILCs (ILC1s) producing IFNγ, ILC2s producing IL-5 and IL-13, and ILC3s producing IL-17 and IL-22. ILC classification was subsequently corroborated by distinctive markers and developmental requirements.
Prior to the discovery of ILCs, NK cells were the only known innate CD3− lymphocytes. Their classification within the larger scheme of ILCs has been challenging, in particular due to their strong resemblance to ILC1s while also maintaining important distinctions. For example, mouse ILC1s and NK cells both express NK1.1 and NKp46 and produce IFNγ, and as such were initially classified together as Group 1 ILCs (Spits et al., 2013). However, grouping NK cells with ILC1s was confounded by several discordant features.
Whereas ILC1s are restricted to organs, and CD49a expression is a faithful marker of tissue-resident populations under steady-state conditions (Peng et al., 2013, Sojka et al., 2014a), most NK cells freely circulate. These cells also differ in their developmental requirements, with NK cells requiring Eomes for development, while Eomes is not expressed in ILC1s and is dispensable for development (Daussy et al., 2014). Moreover, ILC1s are completely Tbet-dependent (Daussy et al., 2014, Sojka et al., 2014a), whereas NK cells require Tbet only for maturation (Townsend et al., 2004, Gordon et al., 2012). Consistent with their differential requirements for T-box transcription factors, NK cells and ILC1s arise at different branch points downstream of the CLP. While NK cells originate from NK cell precursor cells (Rosmaraki et al., 2001), other ILCs further differentiate along a pathway that includes the common helper innate lymphoid progenitor (CHILP) and ILC precursor (ILCP), which do not have NK cell potential (Klose et al., 2014, Constantinides et al., 2014, Yang et al., 2015). Given the many ways in which NK cells and ILC1s differ, ILC classification was recently revised so that NK cells and ILC1s are now considered distinct ILC lineages (Vivier et al., 2018).
Constructing clear definitions of NK cells and ILC1s is complicated, however, by their plasticity and heterogeneity upon stimulation, which can alter the features that discriminate between them under steady state conditions. For example, within the tumor microenvironment or upon in vitro culture with TGF-β, NK cells may downregulate Eomes and resemble ILC1s (Gao et al., 2017, Cortez et al., 2016). However, these cells are found only in the tumor microenvironment in vivo and it is not clear if they can persist in the absence of tumor. Moreover, NK cells and ILC1s become activated under similar circumstances, such as murine cytomegalovirus (MCMV) infection and tumorigenesis (Weizman et al., 2017, Dadi et al., 2016, Yokoyama, 2013). As it is currently difficult to assess NK cells independently of ILC1s, changes undergone by either population may be masked by the presence of the other. Independent examination of each population, especially in settings known to stimulate NK cells and ILC1s, is required to better understand these cells and their relationship to each other.
Both NK cells and ILC1s respond to infection with Toxoplasma gondii, an intracellular parasite that infects a third of the global population, and is a natural pathogen in mice (Montoya and Liesenfeld, 2004). During acute infection, rapidly replicating tachyzoites disseminate systemically, triggering dendritic cells to produce IL-12, which subsequently stimulates NK cells to produce IFNγ and ILC1s to produce IFNγ and TNFα(Klose et al., 2014, Goldszmid et al., 2012). Tachyzoites eventually differentiate into slowly replicating bradyzoites, which are primarily encysted within cells of the central nervous system and skeletal muscle (Yarovinsky, 2014). As T. gondii infection elicits activation of both NK cells and ILC1s, herein we sought to investigate how NK cells and ILC1s respond to gain better insight into inflammation-induced changes. Indeed, we found that ILC1s become permanently heterogeneous after infection, largely owing to the surprising conversion of NK cells into ILC1s.
Results
T. gondii infection results in ILC1 expansion
Following administration of α-NK1.1, acute infection with the type II Prugniaud (Pru) strain of T. gondii resulted in increased parasite load and higher mortality rates (Supplementary Fig. 1a-c), consistent with previous reports (Goldszmid et al., 2007). Since α-NK1.1 affects both NK cells and ILC1s and both of have been previously implicated in the immune response to T. gondii (Goldszmid et al., 2012, Klose et al., 2014), we sought to investigate how NK cells and ILC1s respond to infection. Here we assessed these populations by following expression of Eomesodermin (Eomes) and CD49a among CD3− CD19− NK1.1+ NKp46+ cells, as NK cells express Eomes and not CD49a while ILC1s express CD49a but not Eomes. In the uninfected spleen, the vast majority of NK1.1+ NKp46+ cells were NK cells (Fig. 1a), although there was a small population of ILC1s, which are primarily found in other organs including the liver and small intestine (Sojka et al., 2014a, Fuchs et al., 2013). Interestingly, over the course of infection, NK cells decreased both as a proportion of NK1.1+ NKp46+ cells and in absolute number (Fig. 1a,b). By contrast, there was an increase in ILC1s that was clearly evident at 21 d post-infection (p.i.) (Fig. 1a,c). Notably, most ILC1s from infected mice expressed Ly6C, unlike ILC1s present under steady state conditions (Fig. 1a,d). The expansion of splenic ILC1s persisted for at least 4 months p.i. (Fig. 1e).
ILC1 expansion persists in the absence of ongoing infection
Two major possibilities could account for the sustained expansion of ILC1s. They may represent a response to ongoing parasite replication, as bradyzoites develop following acute infection with Pru T. gondii infection and parasite reactivation can occur. Alternatively, the increased ILC1s might be a permanent change that persists even after infection subsides. To test whether ongoing parasite replication plays a role, we infected mice with T. gondii, then suppressed growth of tachyzoites with sulfadiazine at various time points p.i. (Eyles and Coleman, 1955). Interestingly, infection induced expansion of ILC1s, as seen at 35 d.p.i, even in sulfadiazine-treated mice when treatment began 7 or 10 d.p.i., with ILC1 numbers comparable to those of untreated, infected mice (Fig. 2a). ILC1 expansion also occurred following repeated administration of the Cps1-1 T. gondii strain (Fig. 2b), which does not replicate in vivo and therefore does not persist in a chronic form (Fox and Bzik, 2002). Moreover, the ILC1 number was similar after infection with a single inoculum of Δgra4 parasites, which are defective in cyst formation (Fox et al., 2011, Jones et al., 2017), as compared to the WT parental Pru strain (Fig. 2c). BALB/c mice, which have fewer cysts during chronic T. gondii infection (Suzuki et al., 1993), also had increased numbers of ILC1s after Pru infection (Fig. 2d), also indicating that expanded ILC1s were not confined to infected C57BL/6 mice. In summary, we found that T. gondii infection induced ILC1 expansion, which persisted independent of ongoing parasite replication, suggesting a permanent change in ILC1s; in subsequent experiments we further examined the impact of infection with the Pru strain of T. gondii.
Expansion of ILC1s among circulating cells
ILC1s have previously been established as being tissue-resident under steady state conditions (Gasteiger et al., 2015, Sojka et al., 2014a, Peng et al., 2013). However, surprisingly, we found that infected mice displayed ILC1 expansion even in the blood, and a highly vascularized organ, i.e., the lung, indicating that at least some of these cells circulate (Fig. 3a,b). Consistent with our previous findings in the spleen (Fig. 1a,d), the ILC1s in the blood and lungs of infected mice also expressed Ly6C (Supplementary Fig. 2a). However, there was no increase of ILC1s in the mesenteric lymph node, brain, peritoneum, bone marrow, uterus, or salivary gland (Fig. 3c, Supplementary Fig. 2b). Nonetheless, these data indicated that T. gondii infection induced ILC1 expansion, even in the circulation.
Parabiosis experiments could provide supportive evidence of circulating versus tissue-resident cells, but the possible reactivation and spread of T. gondii following surgery, as seen clinically (Bosch-Driessen et al., 2002), could confound analysis. Instead, we utilized Eomes-GFP reporter mice to assess how ILC1s and NK cells from infected mice behave following transfer into uninfected mice (Daussy et al., 2014). We isolated NK cells and ILC1s from congenically distinct Eomes-GFP mice 35 d.p.i. for adoptive transfer into naïve Rag2−/− Il2rg−/− mice (Fig. 3d). When we assessed the recipient mice 24 d later, we detected both transferred populations in the spleen and liver (Fig. 3e). Although NK cells upregulated CD49a upon transfer into Rag2−/− Il2rg−/− mice, as previously reported (Gao et al., 2017), the Eomes expression generally remained the same (Fig. 3f). Therefore, in the setting of homeostatic expansion, transferred NK cells and ILC1s from infected mice persist with generally stable Eomes expression, and transferred ILCs from infected mice can circulate, providing complementary evidence, in addition to above (Fig 2a-d), that they persist in the absence of ongoing infection.
Interestingly, however, the ratio of transferred ILC1s to NK cells was higher in the liver than in the spleen, with ILC1s comprising nearly all transferred cells in the liver (Fig. 3e). This suggested that the ILC1s from infected mice may not uniformly circulate and rather they might preferentially migrate to the liver despite originating in the spleen. Alternatively, the adoptive transfer studies may be revealing properties of heterogeneous populations of cells. To further evaluate these issues, we studied the liver, which contains a much larger population of ILC1s than the spleen, in more detail directly following primary infection. Infection decreased the frequency of liver NK cells among NK1.1+ NKp46+ cells, though the total NK cell number did not change (Fig. 3g,h). Notably, we observed that infection resulted in a striking population of liver Eomes+ CD49a+ cells (Fig. 3g,i). The frequency and absolute number of ILC1s also did not change (Fig. 3j), although this was potentially confounded by the sizeable ILC1 population already present at steady state (Sojka et al., 2014a, Peng et al., 2013). Indeed, we found that Ly6C was expressed by only half of the ILC1s in the livers of infected mice at 35 d.p.i. (Fig. 3g,k) while Ly6C was expressed by a higher percentage (~80%, Fig 1d). Accordingly, the number of liver Ly6C+ ILC1s was increased relative to uninfected mice, whereas the number of Ly6C− ILC1s remained comparable (Fig. 3k). This suggested that at 35 d.p.i. not all ILC1s displayed effects of T. gondii infection and that Ly6C expression may distinguish between steady-state ILC1s and ILC1s induced by infection. Thus, the data suggest that in infected mice, steady-state ILC1s are maintained while an additional ILC1 subset arises.
T. gondii-induced ILC1s are distinct from NK Cells and ILC1s
Ly6C is expressed by ILC1s only in the spleen, blood, and liver of infected mice, raising the possibility that T. gondii induces the development of ILC1s that do not exist under steady state conditions. To assess the possibility of a novel ILC1 subset, we first performed more detailed comparisons of NK cells and ILC1s in the spleens of uninfected and infected mice. As circulatory capacity was specific to ILC1s in infected mice, we examined expression of molecules typically associated with ILC1 tissue-residency (Sojka et al., 2014a, Daussy et al., 2014). ILC1s in both uninfected and infected mice did not express CD62L and expressed higher levels of PECAM-1 and CXCR6 compared to NK cells (Fig. 4a). However, infection resulted in NK cells and ILC1s gaining expression of CCR8 and CX3CR1 compared to NK cells and ILC1s in uninfected mice. Additional differences included ILC1s from infected mice expressing Neuropilin-1 and KLRG1 at higher levels than other populations, and DNAM-1 more highly than NK cells (Fig. 4b). Moreover, while ILC1s from uninfected mice were predominantly CD27+ CD11b− and NK cells from uninfected and infected mice encompassed all subsets, ILC1s from infected mice were mainly CD27− CD11b+ (Fig. 4c). Most ILC1s in both uninfected and infected mice expressed NKG2A and expressed Ly49 receptors at low frequencies (Fig. 4d). NK cells also displayed decreased frequency of Ly49D+ and Ly49H+ subsets following infection. Lastly, NK cells and ILC1s in both uninfected and infected mice produced IFNγ in response to stimulation (Fig. 4e), a hallmark of both cell types (Sojka et al., 2014b). However, only ILC1s from uninfected mice could produce TNFα (Fig. 4f). Similar results were obtained from studies of ILC1s in blood of infected mice (Supplementary Fig. 3a,b). Thus, the spleens and blood of infected mice contained a novel ILC1 subset.
Studies of the infected liver solidified the notion that infection induced the formation of a distinct ILC1 subpopulation. In the liver, we separately assessed the Ly6C− ILC1s that are present in uninfected mice, as well as both Ly6C− ILC1s and Ly6C+ ILC1s in infected mice (Supplementary Fig. 3c). Ly6C+ ILC1s expressed higher levels of CX3CR1, Neuropilin-1, KLRG1, and CD11b, lower levels of CXCR6 and CD27, and produced less TNFα than Ly6C− ILC1s in uninfected and infected mice, which were similar (Supplementary Fig. 3d-f). Therefore, the Ly6C+ ILC1s in the liver therefore resembled those in the infected spleen (Fig. 4a-c,f), suggesting that while ILC1s in uninfected and infected mice share the Eomes− CD49a+ phenotype, they can be distinguished by marker expression and cytokine production profile.
T. gondii infection induces heterogeneity of NK Cells and ILC1s
ILC1s in infected mice expressed high levels of Tbet (Fig. 5a), which was previously identified as a requirement for ILC1 development (Daussy et al., 2014, Sojka et al., 2014a). We therefore hypothesized that Tbet may be required for ILC1 expansion after infection. As Tbx21−/− mice succumb to T. gondii infection (Harms Pritchard et al., 2015), we studied WT:Tbx21−/− mixed bone marrow chimeras. In spleens of both uninfected and infected chimeras, NK cells were present though they were biased towards WT origin (Fig. 5b, Supplementary Fig. 4a), consistent with a requirement for Tbet to complete NK cell maturation (Gordon et al., 2012). However, no ILC1s (Eomes− CD49a+) were found of Tbx21−/− origin in infected WT:Tbx21−/− chimeras, as with ILC1s in naïve chimeras. Thus, splenic ILC1s are Tbet-dependent, even after T. gondii infection.
In the livers of infected WT:Tbx21−/− chimeras, the picture was more complex. There were no ILC1s (Eomes− CD49a+) of Tbx21−/− origin in the infected liver, as with infected spleens (Fig. 5b). However, the livers of infected WT:Tbx21−/− chimeras displayed an accumulation of Tbx21−/− Eomes+ CD49a+ cells that were not present in the spleen (Fig. 5b). Eomes+ CD49a+ cells were also present among counterpart WT cells, though at lower frequencies. These data suggested that ILC1s in infected mice may be derived from Eomes+ cells, i.e., NK cells.
To better assess the heterogeneity induced by T. gondii infection and the role of Tbet in this process, we performed single-cell RNA-seq (scRNA-seq) on sorted WT (CD45.1+) or Tbx21−/− (CD45.2+) NK1.1+ NKp46+ cells from uninfected and d35-infected WT:Tbx21−/− chimeras. We chose to examine the liver for this analysis for several reasons: 1) infection appears to induce a novel ILC1 subset that is common to the liver and spleen; 2) the liver contains ILC1s under steady state, allowing for direct comparison of steady-state ILC1s and T. gondii-induced ILC1s; and 3) the infected liver uniquely contains a Tbx21−/− Eomes+ CD49a+ population. Using t-distributed stochastic neighbor embedding (t-SNE) analysis, we grouped the cells into 17 clusters, C1-C17 (Fig. 5c). In the uninfected WT sample, C1 was the largest cluster, whereas C1 was diminished in the infected WT and Tbx21−/− samples (Fig. 5d,e). C1 cells expressed Eomes and Itgam but did not express Cd27 or Itga1 (Fig. 5f,g), consistent with C1 being mature NK cells, which are indeed the predominant NK1.1+ NKp46+ cells present in the steady-state liver (Supplementary Fig. 5b-d). C3 was the second largest cluster in the uninfected WT sample, and C3 cells did not express Eomes or Ly6c2, expressed Itga1, Cd27, and Cxcr6, and were absent in Tbx21−/− samples, indicating that C3 comprised ILC1s (Fig. 5f,g, Supplementary Fig. 4d).
Two closely related clusters, C10 and C11, were unique to the infected WT sample (Fig. 5d,e). Cells in C10 and C11 expressed Itga1, Ly6c2, and Klrg1, and did not express Eomes (Fig. 5f,g). This specificity to infection, Tbet-dependence, and gene expression identified these cells as T. gondii-induced ILC1s. Interestingly, these two clusters indicated heterogeneity even within the T. gondii-induced ILC1 population. The t-SNE analysis placed C10 closer to mature NK cells (C1) and C11 closer to ILC1s (C3). Comparison of C10 and C11 revealed that C10 expressed signature NK cell genes at higher levels (Robinette et al., 2015), whereas C11 expressed higher levels of signature ILC1 genes (Supplementary Fig. 4e). Interestingly, C10 also highly expressed Cx3cr1 whereas C11 did not (Fig. 5f,g). We corroborated these findings with flow cytometric analysis that showed that Ly6C+ ILC1s in the infected liver contained both a CX3CR1+ CXCR6– population that expressed higher levels of Zeb2, mirroring the C10 phenotype, and also a CX3CR1– CXCR6+ population that mirrors C11 (Supplementary Fig. 4f,g). C5 was also increased in the infected samples, and these cells co-expressed Eomes and Itga1, and were especially prominent in the infected Tbx21−/− sample (Fig. 5e,g), suggesting that this cluster includes the Eomes+ CD49a+ cells that arose in the liver after infection (Fig. 5b). In summary, scRNA-seq analysis revealed that T. gondii infection induces heterogeneous NK cell and ILC1 populations, including two distinct ILC1 subpopulations that are unique to infected mice.
NK Cells downregulate Eomes during T. gondii infection
scRNA-seq analysis revealed heterogeneity of NK1.1+ NKp46+ cells and showed that some T. gondii-induced ILC1s have core NK cell traits. To assess if NK cell conversion into ILC1s might account for this population by means of Eomes downregulation, we examined Ncr1iCre Eomesf/f (Eomes conditional KO or cKO) mice, in which NK cells are absent owing to their requirement for Eomes, as shown by a drastic reduction of NK1.1+ NKp46+ cells (Supplementary Fig. 5a) with normal numbers of CD49a+ ILC1s (Fig. 6a,b) (Gordon et al., 2012, Pikovskaya et al., 2016). Following T. gondii infection, the numbers of CD49a+ and CD49a+ Ly6C+ cells in the spleen were significantly reduced in Eomes cKO compared to Eomesf/f controls (Fig. 6a,b), showing that expansion of these cells requires Eomes. However, while the overall number of CD49a+ cells did not change between uninfected and infected Eomes cKO mice, the number of CD49a+ Ly6C+ cells increased slightly (Fig. 6b), and a greater frequency of the CD49a+ cells expressed Ly6C (Fig. 6c), showing that mild ILC1 expansion occurs in the absence of NK cells. However, approximately three times as many CD49a+ Ly6C+ cells were derived from Eomes-dependent cells, i.e., NK cells (Fig. 6b), suggesting Eomes expression by NK cells contributes to the development of most T. gondii-induced ILC1s.
The role of Eomes-expressing NK cells in formation of T. gondii-induced ILC1s could be either cell-intrinsic or cell-extrinsic. Though an Eomes fate-mapping mouse, i.e., EomesCre, could be used to identify a cell-intrinsic role for Eomes, this was not possible due to the ubiquitous expression of Eomes during development (data not shown). Another approach would be to examine Eomes downregulation following adoptive transfer of congenic NK cells prior to infection but this was hindered due to the inability to find transferred cells after 14 d.p.i. (data not shown), possibly due to the high turnover of cells during infection. Rather, we used mixed bone marrow chimeras to examine a cell-intrinsic or cell-extrinsic requirement for Eomes during T. gondii-induced ILC1 expansion. In uninfected mixed bone marrow chimeras reconstituted with WT cells and either Eomesf/f or Eomes cKO cells, the contributions of Eomesf/f and Eomes cKO cells to CD49a+ cells were comparable, consistent with CD49a+ cells being Eomes-independent ILC1s (Fig. 6d). This degree of chimerism did not change upon infection of WT:Eomesf/f chimeras. However, infection of WT:Eomes cKO chimeras resulted in a significant reduction of CD49a+ cells of Eomes cKO origin. Although there were a few CD49a+ ILC1s of Eomes cKO origin, consistent with Eomes-independent ILC1 origin, T. gondii infection resulted in ILC1s that were mostly dependent on a cell-intrinsic effect of Eomes in NKp46-expressing cells, i.e., NK cells. Therefore most T. gondii-induced ILC1s are ex-NK cells.
We postulated that STAT4 may play a role in the expansion of T. gondii-induced ILC1s because STAT4 and IL-12 signaling is critical for NK cell and ILC1 activation during T. gondii infection (Hunter et al., 1994, Cai et al., 2000, Klose et al., 2014). To test the relevance of this signaling in vivo, we infected WT:Stat4−/− mixed bone marrow chimeras. Indeed, STAT4-deficient ILC1s were diminished following infection, while chimerism of NK cells and ILC1s remained equal between uninfected and infected chimeric mice (Fig. 6e), suggesting that T. gondii-induced ILC1s were dependent on STAT4. Additionally, in vitro culture of splenocytes from infected mice resulted in an increased frequency of ILC1s following culture in IL-2 and IL-12, compared to culture in IL-2 alone (Supplementary Fig. 5b), suggesting that IL-12 may contribute to ILC1 expansion. However, culture of purified NK cells from infected mice did not result in Eomes downregulation (Supplementary Fig. 5b,c), showing there may be IL-12 cell-extrinsic effects involved in this conversion. Notably, T. gondii-induced ILC1s stably maintained their phenotype after 2 days in culture (Supplementary Fig. 5d). Taken together, these results suggest that STAT4 and IL-12 contribute to ILC1 expansion in the context of T. gondii infection, with contributions from additional factors as well.
Eomes downregulation within NK cells accompanies extensive transcriptional and epigenomic changes
Our scRNA-seq data suggested that T. gondii infection drastically rewires NK cells and ILC1s to induce permanent changes. To provide confirmatory evidence, we performed bulk RNA sequencing (RNA-seq) to inform epigenomic analyses, even though we realize that these approaches lose single cell resolution. For this analysis, we used splenocytes, sorted into NK cells (Ly6C− CD49a−) from uninfected and infected mice, and T. gondii-induced ILC1 (Ly6C+ CD49a+) populations, which were accurately identified Eomes+ and Eomes− cells, respectively (Supplementary Fig. 6a). With respect to our scRNA-seq analysis, these cells represent Eomes+ (conventional) NK cells from uninfected and infected mice, and T. gondii-induced ILC1s (both converted NK cells and Ly6C+ ILC1s, of which most were converted NK cells, Fig. 5, 6b). Principal component analysis and hierarchical clustering showed a higher degree of similarity between NK cells from uninfected and infected mice, relative to T. gondii-induced ILC1s (Fig. 7a,b). As expected, Ly6c1, Ly6c2, Itga1, and Eomes were differentially expressed (Fig. 7c). We found extensive differences between NK cells and T. gondii-induced ILC1s, including differences in the expression of surface markers, transcription factors, secreted factors, adhesion molecules, chemokine receptors, and signaling molecules (Fig. 7d).
Since NK cells and ILC1s display distinct patterns of chromatin accessibility (Shih et al., 2016), and infection can induce long-lasting epigenomic changes in NK cells (Lau et al., 2018), we performed the assay for accessibility for transposase-accessible chromatin using sequencing (ATAC-seq) (Buenrostro et al., 2015) to compare the NK cells and ILC1s present in T. gondii-infected mice at the epigenomic level. We found widespread chromatin remodeling in T. gondii-induced ILC1s, with T. gondii-induced ILC1s clustering separately from NK cells (Fig. 7e,f). Most differentially accessible regions fell within intronic and intergenic regions, consistent with categorization as putative enhancers (Supplementary Fig. 6b).
We organized differentially accessible regulatory elements (REs) into a non-redundant peak set containing 6 clusters (Supplementary Fig. 6c). Clusters 1, 3, and 5 contained REs that were uniquely more accessible in uninfected NK cells, NK cells from infected mice, and T. gondii-induced ILC1s, respectively. Clusters 2, 4, and 6 contained REs that were shared between two groups. Cluster 2 encompassed REs that were more accessible in both NK cell groups relative to T. gondii-induced ILC1s, and accounted for the largest fraction of peaks, showing that NK cells share many epigenetic features regardless of infection. Somewhat surprisingly, Cluster 4, which contained REs that were more accessible in both NK cells and ILC1s from infected mice relative to uninfected NK cells, was the second largest cluster, suggesting that infection may induce epigenetic changes common to both populations.
We directly compared the REs in Clusters 3 and 5 to identify additional differences between NK cells and T. gondii-induced ILC1s following infection. Analysis of transcription factor binding motifs revealed motifs that were significantly enriched in each population, such as NF-κb in Cluster 3, whose motif was present in 13.2% of REs, and Klf4 in Cluster 5, whose motif was present in 25.8% of REs (Supplementary Fig. 6d). We also identified the specific loci that contained the most differentially accessible REs, and noted that the presence of more accessible REs generally correlated with greater gene expression. Notably, many of the epigenetic features shared by memory NK cells and memory CD8+ T cells (Lau et al., 2018) were among the loci that gained accessibility in T. gondii-induced ILC1s (Supplementary Fig. 6e). Among the differentially accessible regions were the Eomes and Tbx21 loci, supporting our data above detailing expression changes of these transcription factors (Fig. 7g, Supplementary Fig. 6f). Overall, our data suggest that T. gondii induces the conversion of NK cells into ILC1s in a process that encompasses changes in gene expression and chromatin accessibility.
Discussion
Infection with T. gondii, a natural mouse pathogen, causes the expansion of a novel ILC1 subpopulation that stably persists even after acute infection resolves. These ILC1s are heterogeneous, due to contributions from both ILC1s and NK cells, though NK cells give rise to most T. gondii-induced ILC1s by means of Eomes downregulation. These findings disrupt current notions surrounding NK cells and ILC1s, which were recently established as discrete ILC lineages (Vivier et al., 2018), and exemplify the limitations of categorizing NK cells and ILC1s based solely on their phenotypes under steady state conditions.
Previous studies indicated that Eomes downregulation may occur in NK cells within the tumor microenvironment (Gill et al., 2012, Gao et al., 2017) and in vitro (Cortez et al., 2016). Akin to the intermediate ILC1s (intILC1s) described within tumors (Gao et al., 2017), we found T. gondii infection induced Eomes+ CD49a+ NK cells in the liver, which are possibly intermediates between NK cells and ex-NK cells. However, our findings differ in several significant ways. The tumor studies indicated that Eomes− NK cells were restricted to tumors and emphasized the importance of the microenvironment (Gao et al., 2017, Gill et al., 2012), concluding that within tumors, Eomes downregulation transiently silences NK cells to hamper immunosurveillance (Silver and Humbles, 2017). There was no evidence presented that these Eomes− cells can circulate, though they proliferate when adoptively transferred into lymphocyte-deficient hosts. Moreover, it is not clear if these cells persist in the absence of tumor. By contrast, T. gondii-induced ILC1s are disseminated throughout the circulation, retain the ability to produce IFNγ, do not produce TNFα, and are maintained in the absence of ongoing stimulus.
Our findings further indicate a permanent transformation of T. gondii-induced ILC1s rather than a transient response to inflammation, a notion that is further bolstered by the observations that T. gondii-induced ILC1s possess a unique gene expression profile. Moreover, T. gondii infection induced epigenetic modifications in NK cells and ILC1s, though our analysis did not specifically examine each of the different individual NK and ILC1 subsets we identified by scRNA-seq. However, these modifications were extensive and could not be readily explained by heterogeneity in the T. gondii-induced ILC1s that we discovered by scRNA-seq. Nonetheless, our studies go beyond just the differences between tumor-induced and T. gondii-induced Eomes downregulation. Our findings indicate that DX5 and CD49a expression to identify NK cells and ILC1s may not always faithfully reflect Eomes expression, as is the case during early T. gondii infection. Moreover, although there is clear evidence supporting conversion from NK cells to Eomes− NK cells, such as intILC1s (Gao et al., 2017), there is a general inability to distinguish between preexisting ILC1s and de novo converted ILC1s. Indeed, this may explain discrepancies that currently exist within the field, such as the conflicting report that ILC1-like cells confer protection against certain tumors (Dadi et al., 2016). Thus, our studies suggest that ILC1s may be more heterogeneous than currently thought, as they can arise from distinct precursors (Vivier et al., 2018).
In T. gondii infection, IL-12 is critical for NK cell activation (Gazzinelli et al., 1993) and our data suggest that it can have long-lasting effects. This parallels the requirements for development of memory and memory-like NK cells induced by MCMV and cytokine stimulation (Cooper et al., 2009, Sun et al., 2012, Romee et al., 2012). In addition to IL-12 dependence, MCMV-induced memory NK cells and T. gondii-induced ILC1s share Ly6C expression, as well as alterations in their receptor repertoire, although T. gondii-induced ILC1s primarily express the inhibitory receptor NKG2A while MCMV-induced memory NK cells are Ly49H+. Moreover, memory and memory-like NK cells have few markers that differentiate them from resting NK cells, whereas NK cells converted by T. gondii infection lose expression of Eomes and gain expression of Ly6C, KLRG1, CX3CR1, and Nrp-1, with scRNA-seq revealing global transcriptional differences that identify them as a discrete subpopulation.
The persistence of T. gondii-induced ILC1s after clearance of infection is reminiscent of classical immune memory and suggests that ex-NK cells play a role beyond what has been attributed to NK cells and ILC1s in T. gondii infection thus far. One possibility is that T. gondii-induced ILC1s are an inflammatory subpopulation, such as the inflammatory ILC2s that gain circulatory capacity in response to activation by cytokines or infection (Huang et al., 2018). These inflammatory ILC2s can be distinguished from tissue-resident ILC2s based on their increased expression of KLRG1 and S1P receptors, features that are also characteristic of T. gondii-induced ILC1s, but not steady-state tissue-resident NK cells. Most T. gondii-induced ILC1s express CX3CR1, which has previously been shown to regulate NK cell recruitment into the circulation (Ponzetta et al., 2013, Sciume et al., 2011). Another possibility is that ex-NK cells may provide a protective role in subsequent T. gondii infections, as shown previously (Denkers et al., 1993), although this finding remains controversial (Goldszmid et al., 2007). Moreover, we have not been able to show that repeated injection of an avirulent T. gondii strain confers NK1.1-dependent protection against subsequent lethal re-challenge. Regardless, we have observed ILC1 expansion following repeated exposure to avirulent T. gondii strains, and future studies may illuminate how gaining circulatory capacity affects the functionality of ILC1s and how ex-NK cells contribute to subsequent immune responses to T. gondii.
In summary, we found that T. gondii infection induced permanent changes to NK cells and ILC1s, including conversion of NK cells into ILC1s. Our studies indicate that the current system of ILC classification, based on their phenotype and development under steady state conditions, may not apply following inflammation. Rather, plasticity may give rise to populations that resemble one another at first glance, but actually represent the convergence of multiple developmental paths to form interrelated populations.
Materials and Methods
Mice
Eomesf/f, Tbx21−/−, and Stat4−/− were purchased from The Jackson Laboratory. C57BL/6, CD45.1+, and BALB/c mice were purchased from Charles River. Ncr1iCre mice were a kind gift from Eric Vivier. Zeb2-GFP mice were a generous gift from Kenneth Murphy (Wu et al., 2016). Eomes-GFP mice were graciously provided by Thierry Walzer (Daussy et al., 2014). Ncr1iCre were bred to Eomesf/f mice to generate Eomes cKO mice and Cre-negative littermate controls. Rag2−/− were bred to Il2rg−/− to generate Rag2−/− Il2rg−/− mice. Age and sex-matched animals were used in all experiments. Mice were infected at 6-12 wk of age. All protocols were approved by the Institutional Animal Care and Uses Committee (Washington University School of Medicine, St. Louis, MO) under animal protocol number 20160002.
Parasites
PRU-FLuc-GFP Type II strain of T. gondii was a generous gift from John C. Boothroyd and used in all infections unless otherwise specified. Pru∆ku80∆hx∆gra4::HX/mCherry and the wildtype parental strain Pru∆ku80∆hx were previously described (Jones et al., 2017). The attenuated Cps1-1 mutant on the type I RH background was described previously (Fox and Bzik, 2002).
All T. gondii strains used in this study were maintained in human foreskin fibroblast (HFF) monolayers grown in D-10 medium (Dulbecco’s modified Eagle medium, 10% fetal bovine serum (FBS), 2 mM glutamine, 10 mM HEPES pH 7.5, 20 μg/mL gentamicin), maintained at 37 °C with 5% CO2. The Cps1-1 mutant was grown as described above but supplemented with 200 μM uracil (Sigma). Mature parasites were lysed from host cells by vigorous pipetting and egressed parasites were filtered through 3 μm polycarbonate membranes and resuspended in HHE medium (Hanks’ balanced salt solution, 10 mM HEPES, 0.1 mM EGTA). Cell cultures were determined to be mycoplasma-negative using the e-Myco plus kit (Intron Biotechnology).
Infection
Mice were injected intraperitoneally (i.p.) with 200 PRU-FLuc-GFP, Pru∆ku80∆hx, or Pru∆ku80∆hx∆gra4::HX/mCherry tachyzoites that were propagated in culture as described above. In indicated experiments, mice were treated with 0.5 g/L sulfadiazine in drinking water to suppress parasite growth. To generate non-persistent infections, 1×105 Cps1-1 tachyzoites were injected 3 times, 2 wk apart.
Luciferase Imaging
Mice were injected i.p. with 150 mg D-luciferin (Gold Biotechnology) per kg body weight, incubated for 10 min, then anesthetized with continuous isoflurane anesthesia at a flow rate of 1 L/min. Images were captured using an IVIS Spectrum In Vivo Imaging System (Perkin Elmer). Luminescence was quantitated using Living Image software (Perkin Elmer). NK1.1-depleted mice were injected i.p. with 100 μg of α-NK1.1, 3 d and 1 d prior to infection. Purified α-NK1.1 antibody was generated at the Rheumatic Diseases Core Center Protein Purification and Production Facility using the PK136 hybridoma (ATCC).
Cell Isolation
Spleens were mashed through a 70 μm cell strainer and treated with Tris-NH4Cl to lyse red blood cells. Livers were mashed through a 70 μm cell strainer, resuspended in isotonic 38.5% Percoll (Sigma-Aldrich), centrifugated at 325 x g for 20 min, and treated with Tris-NH4Cl. Peritoneal cells were isolated by lavage of the peritoneal cavity with PBS. Bone marrow was flushed from femurs and tibias, mashed through a 70 μm cell strainer, and treated with Tris-NH4Cl. Lymph nodes were mashed through a 70 μm cell strainer. Brains were mashed through a 70 μm cell strainer, resuspended in 38.5% isotonic Percoll, and centrifugated at 325 x g for 20 min. Lungs were perfused with PBS, minced, digested in RPMI-1640 containing 2% FBS, 1 mg/mL Collagenase Type IV (Sigma-Aldrich), and 0.2 mg/mL DNase (Sigma-Aldrich), mashed through a 70 μm cell strainer, resuspended in 38.5% isotonic Percoll, and centrifugated at 325 x g for 20 min. Uterus and salivary gland were minced, digested in RPMI-1640 containing 0.17 mg/mL Liberase TL (Sigma-Aldrich) and 0.1 mg/mL DNase (Roche), mashed through a 70 μm cell strainer, resuspended in 38.5% isotonic Percoll, and centrifugated at 325 x g for 20 min.
Flow Cytometry and Cell Sorting
Cells were stained with Fixable Viability Dye and Fc receptors were blocked with 2.4G2 hybridoma (ATCC) culture supernatants prior to staining in PBS containing 2% FBS and 2.5 mM EDTA on ice. Data was acquired using a FACSCanto instrument (BD Biosciences) or FACSAria instrument (BD Biosciences) using FACSDiva software (BD Biosciences). Data was analyzed with Flowjo v10 (Treestar).
For cell sorting, splenocytes were enriched for NK1.1+ NKp46+ cells by negative selection. Splenocytes were incubated with 0.2 μg/mL biotin anti-mouse CD4, 0.2 μg/mL biotin anti-mouse CD8, and 1 μg/mL biotin anti-mouse CD19. One hundred μL of EasySep Mouse Streptavidin Rapidspheres (BD Biosciences) were added per 1×108 splenocytes. Cells were placed in an EasySep magnet (BD Biosciences) and enriched cells were poured off and subjected to extracellular staining. Sorting was performed using a FACSAria instrument (BD Biosciences) using FACSDiva software (BD Biosciences). NK cells were identified as CD3− CD19− NK1.1+ NKp46+ Eomes-GFP+ CD49a− and T. gondii-induced ILC1s were identified as CD3− CD19− NK1.1+ NKp46+ Eomes-GFP− CD49a+.
Stimulations and Intracellular Staining
Following extracellular staining, cells were fixed with FoxP3/Transcription Factor Staining Buffer Set for 30 min at room temperature and washed with 1x Permeabilization buffer (eBiosciences). Transcription factor antibodies were diluted in 1x FoxP3 permeabilization buffer and cells were incubated for 30 min at room temperature.
To assess Eomes downregulation after culture in IL-12, 5×106 splenocytes were cultured in R-10 medium (RPMI-1640 medium containing 10% FBS, 2 mM glutamine, 100 U/mL penicillin, and 100 μg/mL streptomycin) with 300 IU/mL IL-2 with 20 ng/mL IL-12 (Pepro Tech). Unstimulated control wells contained only 300 IU/mL IL-2.
For detection of IFNγproduction, 5×106 splenocytes were cultured for 5 h in R-10, with the addition of 20 ng/mL IL-12, 20 ng/mL IL-12 + 5 ng/mL IL-18, or 0.5 μg/mL PMA + 4 μg/mL Ionomycin. Brefeldin A was added after 1 h.
Generation of Bone Marrow Chimeras
Donor bone marrow was harvested by flushing femurs and tibias and mashing through a 70 μm cell strainer. Cells were mixed in a 1:1 ratio with CD45.1+ bone marrow, and intravenously injected into irradiated mice. Recipient mice were lethally irradiated (900 rad) and reconstituted with a 1:1 ratio of CD45.1+ bone marrow cells and CD45.2+ Eomes cKO, Eomesf/f, Tbx21−/−, or Stat4−/− bone marrow cells (Kaplan et al., 1996, Zhu et al., 2010, Finotto et al., 2002). Mice were maintained on sulfamethoxazole/trimethoprim oral suspension added to the drinking water for 2 wk after reconstitution and used for experiments 6-8 wk after reconstitution. To prevent chimeric mice from succumbing to acute T. gondii infection, bone marrow chimeras were treated with 0.5 g/L sulfadiazine in drinking water beginning 10 d.p.i. and maintained on sulfadiazine until the end of the experiment.
Single-Cell RNA-Seq
Sorted cells were subjected to droplet-based 3’ end massively parallel single-cell RNA sequencing using Chromium Single Cell 3’ Reagent Kits as per manufacturer’s instructions (10x Genomics). The libraries were sequenced using a HiSeq3000 instrument (Illumina).
Sample demultiplexing, barcode processing, and single-cell 3’ counting was performed using the Cell Ranger Single-Cell Software Suite (10x Genomics). Cellranger count was used to align samples to the mm10 reference genome, quantify reads, and filter reads with a quality score below 30.
The Seurat package in R was used for subsequent analysis (Butler et al., 2018). Genes expressed in fewer than 3 cells and cells that expressed less than 400 or greater than 3500 genes were removed for downstream analysis. Data was normalized using a scaling factor of 10,000s and nUMI was regressed with a negative binomial model. Principal component analysis was performed using the top 3000 most variable genes and t-SNE analysis was performed with the top 40 PCAs. Clustering was performed using a resolution of 0.8. For heatmaps, the mean expression by all cells within the cluster was used.
RNA Sequencing
RNA was isolated from 5×104 sorted cells using Trizol. Libraries were prepared using the Clontech SMARTer Kit. Sequencing was performed using 1×50 single-end reads with a HiSeq3000 instrument (Illumina). Reads were quantified using kallisto and differential expression was assessed using the DESeq2 package in R (Bray et al., 2016, Love et al., 2014). Using a log2FC cutoff of 1 and an FDR threshold of 0.1, we identified 657 differentially expressed (DE) genes between the NK cells from uninfected and d35-infected mice, 2288 DE genes between NK cells from infected mice and T. gondii-induced ILC1s, and 1685 DE genes between NK cells from uninfected mice and T. gondii-induced ILC1s. Variance-stabilized transform values were used for subsequent analysis. All plots were generated in R using the ggplot2, pheatmap, and hclust packages. GSEA was performed using the GSEA Preranked module in GenePattern (Subramanian et al., 2005, Mootha et al., 2003, Reich et al., 2006).
ATAC-seq
Samples were prepared as previously described (Buenrostro et al., 2015), then purified with MinElute spin columns (Qiagen). DNA fragments were amplified using Nextera index adapters as per manufacturer’s instructions (Illumina). Libraries were purified with AMPure XP beads (Beckman Coulter). Three libraries were pooled and sequenced with 2×50 paired end reads using a HiSeq3000 instrument (Illumina).
Sequences were aligned to the mm10 reference genome using Bowtie2 (Langmead and Salzberg, 2012). Reads with a quality score below 30 were removed with Samtools (Li et al., 2009) and duplicate reads were filtered with PicardTools (Broad Institute). Peaks were called using MACS2 (Zhang et al., 2008) with an FDR cutoff of 0.05. Narrowpeak files generated from MACS2 were converted to Bigwig files with deepTools (Ramirez et al., 2014), and visualized using the UCSC genome browser. Differential peaks were identified using Homer (Heinz et al., 2010) and heatmaps of differentially accessible regions were generated with deepTools. Transcription factor binding motifs were identified de novo with Homer.
Across all samples, we identified 71,504 peaks. We further analyzed 9,640 discrete peaks that displayed two-fold changes between at least two groups. Between NK cells from control and infected mice, there were 1,809 differential peaks (1445 were larger in the NK cells from infected mice and 364 were larger in NK cells from uninfected mice), between NK cells from infected mice and T. gondii-induced ILC1s, there were 6,710 differential peaks (4572 were larger in the cNK cells and 2,138 were larger in T. gondii-induced ILC1s), and between NK cells from uninfected mice and T. gondii-induced ILC1s, there were 5,938 differential peaks (3065 were larger in the uninfected NK cells and 2,973 were larger in T. gondii-induced ILC1s).
Statistical Analysis
Prism (GraphPad) was used for statistical analysis. Student’s t-test was used for comparison of 2 groups and 2-way ANOVA with Bonferroni correction was used in analyses involving multiple comparisons. P-values are shown in figures and figure legends. In all graphs, points represent biological replicates, bar position represents the mean, and error bars represent + s.e.m.
Data Availability
The sequencing data in this publication are available on GEO. The RNA-seq and ATAC-seq data are accessible under the accession number GSE124313. The scRNA-seq data are available under accession number GSE124577.
Competing Interests
The authors declare no competing interests.
Acknowledgements
This work was supported by NIH grants F30DK108472 (E.P.), AI128845 (W.M.Y.), AI120606 (E.M.O), AI134035 (E.M.O., M.C.), AI11852 (E.M.O.), and CA188286 (E.M.O.). We thank members of the Yokoyama Lab, Ilija Brizic, Stipan Jonjic, and Victor Cortez for helpful discussions, Jennifer Barks and Reeha Savari Dhason for HFF cells, Nathaniel Jones, Sumit Kumar, Josh Radke, and Kevin Brown for T. gondii strains, Suzanne Hickerson for technical assistance, and Lacey Feigl, Lorraine Schwartz, and Shirley McTigue for administrative assistance.
We thank the Genome Technology Access Center in the Department of Genetics at Washington University School of Medicine for help with genomic analysis. The Center is partially supported by NCI Cancer Center Support Grant #P30 CA91842 to the Siteman Cancer Center and by ICTS/CTSA Grant# UL1 TR000448 from the National Center for Research Resources (NCRR), a component of the National Institutes of Health (NIH), and NIH Roadmap for Medical Research. This publication is solely the responsibility of the authors and does not necessarily represent the official view of NCRR or NIH. We also thank the McDonnell Genome Institute, especially Catrina Fronick, Robert Fulton, and Amy Ly, for assistance and advisement regarding genomic analysis.