The microbiota maintain homeostasis of liver-resident γδT-17 cells in a lipid antigen/CD1d-dependent manner

The microbiota control regional immunity using mechanisms such as inducing IL-17A-producing γδ T (γδT-17) cells in various tissues. However, little is known regarding hepatic γδT cells that are constantly stimulated by gut commensal microbes. Here we show hepatic γδT cells are liver-resident cells and predominant producers of IL-17A. The microbiota sustain hepatic γδT-17 cell homeostasis, including activation, survival and proliferation. The global commensal quantity affects the number of liver-resident γδT-17 cells; indeed, E. coli alone can generate γδT-17 cells in a dose-dependent manner. Liver-resident γδT-17 cell homeostasis depends on hepatocyte-expressed CD1d, that present lipid antigen, but not Toll-like receptors or IL-1/IL-23 receptor signalling. Supplementing mice in vivo or loading hepatocytes in vitro with exogenous commensal lipid antigens augments the hepatic γδT-17 cell number. Moreover, the microbiota accelerate nonalcoholic fatty liver disease through hepatic γδT-17 cells. Thus, our work describes a unique liver-resident γδT-17 cell subset maintained by gut commensal microbes through CD1d/lipid antigens.

T he liver is situated in a unique systemic circulation system that receives blood from both the hepatic artery and the portal vein, making this organ a prime location for both metabolic and immune function [1][2][3] . However, the precise mechanism that connects the microbiota and the hepatic immune response is seldom reported. Bacterial translocation and pathogen-associated molecular pattern (PAMP) transport are the two main events that have been observed in the liver-gut axis 4,5 . However, the proposed mechanisms will remain elusive until the soluble factors from the microbiota and their cellular targets in liver-gut axis are determined.
The liver is enriched in innate immune cells, including gdT cells at a frequency of 3-5% (5 to 10-fold greater than in other tissues or organs) within total liver lymphocytes 1 . gdT cells function as a bridge between innate and adaptive immunity because they express a rearranged T-cell receptor (TCR) that recognizes certain antigens and can also rapidly secrete pro-inflammatory cytokines including interleukin (IL)-17A upon stimulation 6 . By producing IL-17A to recruit neutrophils and enhance adaptive immunity, IL-17A-producing gdT (gdT-17) cells have an important role in host defence against bacterial, fungal and viral infections, as well as stress, tumour surveillance and autoimmune diseases 7 . However, although hepatic gdT cells are involved in several liver immune diseases 8 , their physiological characteristics, and why the liver contains such high levels of gdT cells, are unknown.
CD1d, a typical lipid presentation molecule for natural killer T (NKT) cells 9 , can also present lipid antigens to the gdTCR and activate gdT cells 10 . A gdT cell subset in human blood can respond to CD1d-presented sulfatide, a lipid antigen present in both hosts and bacteria 11 . Another gdT cell subset in the mouse duodenum can respond to exogenous lipid antigens including phosphatidylcholine, phosphatidylethanolamine (PE) and phosphatidylglycerol (PG) presented by CD1d 12 . The liver constantly encounters microbial lipid components, and crosstalk occurs between CD1d and liver NKT cells [13][14][15][16] ; however, little is known regarding the role of gdT cells in this process.
Here we compare gdT cells originating from several organs and identify a liver-resident gdT-cell population that predominantly produces IL-17A. The microbiota maintain hepatic gdT-17 cell homeostasis, the underlying mechanism of which involves microbiota lipid antigens presented by hepatocyte-expressed CD1d, but not PAMPs or cytokine signals. Moreover, liverresident gdT cells responding to the microbiota contribute to nonalcoholic fatty liver disease (NAFLD).

Results
Hepatic cdT cells produce IL-17A. Compared with other immune organs and tissues, hepatic gdT cells predominantly produced high levels of IL-17A, similar to gdT cells from the peritoneal cavity (PC) and lung and significantly higher than those from inguinal lymph nodes (iLNs), the spleen, the thymus, small intestine intraepithelial lymphocytes (IEL), colon IEL and mesenteric LN (mLN) (Fig. 1a,c). In terms of phenotype, hepatic gdT cells exhibited mixed Vg chain usage, which was also distinct from gdT cells of other organs (Fig. 1b). They were in a more active and mature state, as indicated by higher percentages of CD44 high CD62L À cells and lower CD24 expression (Fig. 1c). Corresponding with their high IL-17A expression levels, hepatic gdT cells expressed low levels of CD27 (Fig. 1c), which is a fate determinant of gdT cells to express IFN-g (gdT-1) but not IL-17A (gdT-17) 17 . However, unlike gdT cells of the PC and lung, hepatic gdT cells rarely expressed cytokine receptors including CD121, CD25 and CD127 (Fig. 1c). Interestingly, neonatal mice had low levels of gdT-17 but high levels of gdT-1 cells in the liver (Fig. 1d). As the mice aged, the hepatic gdT-17 cell frequency increased, while that of gdT-1 cells decreased, suggesting that hepatic gdT-17 cells might be induced after birth (Fig. 1d). Overall, hepatic gdT cells exhibited a unique composition and phenotype, indicating that they represent a distinct gdT-cell subtype.
We further characterized the trafficking and homing tendencies of hepatic gdT cells. GFP þ splenic and CD45.1 þ hepatic lymphocytes were transferred separately or together into Cd45.2 þ Tcrd -/mice ( Supplementary Fig. 1a). One day after transfer, splenic gd T cells travelled randomly into both the spleens and the livers of recipient mice. In contrast, hepatic gdT cells selectively homed only to the liver, and co-transfer with splenic lymphocytes did not change this homing tendency. We confirmed this result by transferring purified hepatic gdT cells into Rag1 -/mice and found that hepatic gdT cells, particularly gdT-17 cells, homed back to the liver but not to other organs ( Supplementary Fig. 1b). To further explore whether hepatic gdT cells reside in the liver, CD45-congenic mice were surgically joined by parabiosis and evaluated for chimerism in various cell populations 2 weeks later. While conventional T cells exhibited substantial chimerism (Fig. 1e), the livers of CD45.1 parabiont mice contained almost all CD45.1 þ gdT cells with very few, if any, CD45.2 þ gdT cells; vice versa was observed in CD45.2 parabiont mice (Fig. 1e). Interestingly, PC and thymic gdT cells were also locally retained, but lung, spleen and iLN gdT cells were mutually exchanged ( Supplementary Fig. 1c).
Compared with circulating gdT cells, most of the chemokine receptors and integrin molecules were expressed at lower levels on hepatic gdT cells, but C-X-C chemokine receptor 6 (CXCR6) was expressed at extremely high levels on hepatic gdT cells ( Supplementary Fig. 2a). However, it was not involved in the liver homing of hepatic gdT cells because the numbers of hepatic gdT cells and gdT-17 cells were not changed in Cxcr6 À / À mice ( Supplementary Fig. 2b), although CXCR6 was important for the liver homing of natural killer (NK) and NKT cells 18,19 . Moreover, the liver residency of hepatic gdT cells was absent in Il17a À / À mice. WT and Il17a À / À mice were surgically joined by parabiosis; 2 weeks later, the WT mouse liver contained few Il17a À / À mouse-derived gdT cells, but the Il17a À / À mouse liver contained nearly 50% gdT cells derived from the WT mouse ( Supplementary Fig. 2c). These results suggest that gdT cell liver homing might be related to their IL-17 expression ability in unknown mechanism.
In summary, these data collectively indicate that liver-resident gdT cells are indeed a unique gdT-cell subset with low expression levels of cytokine receptors, and that predominantly produce high levels of IL-17A and are retained in the liver.
The microbiota maintain hepatic cdT-17 cell homeostasis. To explore if the microbiota might influence liver-resident gdT-17 cells, antibiotics were used to clear mouse gut bacteria as previously reported 20 . The antibiotic (Abx)-treated mice had a sharp decrease in commensal microbes, as indicated by the 99% reduction in cultivable bacteria, 95% reduction in bacterial DNA and elongated caecum ( Supplementary Fig. 3a-c). Moreover, Abx-treated mice displayed a normal peripheral immune response ( Supplementary Fig. 3d). Although the total number and frequency of hepatic B, T, NK, NKT and regulatory T (Treg) cells were normal in Abx-treated mice ( Supplementary Fig. 3e,f), the accumulation of liver-resident gdT-17 cells during ontogeny was blocked in mice treated with antibiotics beginning in utero through the pregnant mother (Fig. 2a). The reduction in CD27 expression on hepatic gdT cells was also arrested, but the decline of hepatic gdT-1 cells was not affected (Fig. 2a).
To directly observe the interaction between hepatic gdT-17 cells and the microbiota, germ-free (GF) mice were used, which had even more significantly decreased numbers of hepatic gdT-17 cells than Abx-treated SPF mice (Fig. 2b). In addition, treating GF mice with antibiotics did not further reduce their hepatic gdT-17 cell numbers, suggesting that antibiotics were not directly toxic to gdT-17 cells (Fig. 2b). Reconstituting Abx-treated mice with gut commensal microbes was sufficient to restore their reduced hepatic gdT-17 cell numbers to a normal level (Fig. 2b). The kinetics of hepatic gdT-17 cell recovery in commensal microbereconstructed GF mice were also studied. Commensal microbe reconstitution recovered the hepatic gdT-17 cell number gradually, reaching a similar level to that in SPF mice after 4 weeks of reconstruction (Fig. 2c). These results indicate that the accumulation of hepatic gdT cells relies on microbiotas.
The reduced hepatic gdT-17 cells in Abx-treated mice and GF mice were not restricted by Vg chain usage (Fig. 2d). Consistent with the reduced IL-17A expression by gdT cells from Abx-treated mice and GF mice, they were less differe-ntiated and less activated, as indicated by the lower proportions of CD27 -CCR6 þ and CD44 high CD62Lcells (Fig. 2e). Moreover, enhanced apoptosis and reduced proliferation of hepatic gdT cells were also observed after Abx treatment (Fig. 2e). To better understand the specific influence of the microbiota on hepatic lymphocytes, an in vivo BrdU incorporation method was used to assay the cell proliferation. Only gdT-17 cells, but not NK, abT or even IL-17A-negative gdT cells, displayed reduced BrdU incorporation and Ki67 expression in Abx-treated mice, indicating that the microbiota specifically promoted the proliferation of hepatic gdT-17 cells, but not all lymphocytes, in the liver (Fig. 2f). Together, these data indicated that the microbiota sustain liverresident gdT-17 cell homeostasis by maintaining their normal activation, survival and proliferation.
Global commensal load affects hepatic cdT-17 cell number. To screen for bacteria that influence liver-resident gdT-17 cell numbers, different combinations of antibiotics (A, ampicillin; V,    vancomycin; N, neomycin; M, metronidazole) were used to treat mice. Different antibiotic mixtures had different targets ( Supplementary Fig. 4a,b); thus, we produced a series of mice with commensal microbes at various clusters ( Supplementary  Fig. 4c) and diversities ( Supplementary Fig. 4d). However, although different compositions of microbes were induced in the mice (Fig. 3a), there was almost no correlation between the gdT-17 cell number and the antibiotic type ( Fig. 3b) or the bacteria species diversity (Fig. 3c). Interestingly, the mice with low levels of microbes always had small numbers of hepatic gdT-17 cells, and the mice with high levels of microbes always had large numbers of hepatic gdT-17 cells (Fig. 3b). Indeed, both the numbers and the proliferation percentages of hepatic gdT-17 cells positively correlated with the global microbe DNA loads (Fig. 3d). These results suggest that the global bacterial load is a key factor in the homeostasis of liverresident gdT-17 cells.
E. coli, a typical type of commensal microbe, was chosen to further explore this hypothesis. E. coli was completely deleted in eight groups of mice, including the A, V, AM, VN, AVM, AVN, VNM and AVNM groups (Fig. 4a), but some mice without E. coli (V, AVM and AVN groups) still had a comparable level of hepatic gdT-17 cells as normal mice (Fig. 3b), suggesting that E. coli is not an irreplaceable bacterium for liver-resident gdT-17 cell homeostasis; indeed, there was no correlation with gdT-17 cell numbers (Fig. 4b). However, similar to the transfer of fresh faeces, transferring E. coli alone recovered the decline in hepatic gdT-17 cells in Abx-treated mice, with dose dependency in single E. coli-transferred mice (Fig. 4c). Thus, these results indicate that the global bacterial load, regardless of species specificity, is important for hepatic gdT-17 cell homeostasis.

expressions on hepatic gdT cells (Supplementary
The liver is rich in lipid antigens and CD1d; thus, we speculated that CD1d/lipid antigen complexes interact with liverresident gdT-17 cells in the liver. WT and Cd1d À / À mice were co-housed to share the same microbiota, but Cd1d À / À mice had a decreased number of hepatic gdT-17 cells with lower proportions of CD27 À CCR6 þ cells compared with WT mice, similar to the result observed in Abx-treated WT mice (Fig. 5a). Co-housed NKT-deficient Ja18 À / À mice had similar hepatic gdT-17 cell numbers to WT mice, demonstrating that CD1d sustained hepatic gdT-17 cells directly and independently of NKT cells (Fig. 5b). Moreover, Abx-treatment did not further downregulate the number and frequency of hepatic gdT-17 cells in Cd1d À / À mice (Fig. 5a). Collectively, these data indicate that the microbiota maintain the homeostasis of liver-resident gdT-17 cells mainly through CD1d signalling.
To explore whether CD1d-associated lipid antigens indeed have a role in this process, Abx-treated WT mice were injected with previously reported CD1d-presenting bacterial lipid antigens, including E. coli cardiolipin (CL), PG, PE and total polar lipid extract from E. coli. After receiving exogenous E. coliderived lipids, the liver-resident gdT-17 cell number, proliferation and CD27 À CCR6 þ expression levels partially but markedly recovered in Abx-treated WT mice but not in Cd1d À / À mice (Fig. 5c), indicating that lipid antigens act in a CD1d-dependent manner. Moreover, lipid antigens still induced the recovery of the hepatic gdT-17 cell numbers in Abx-treated IL-1/IL-23-neutralized mice ( Supplementary Fig. 5h) and TLR2/TLR4/TLR9deficient mice ( Supplementary Fig. 5i), which further suggested that lipid antigens maintain the pool of liver-resident gdT-17 cells in a TLR/cytokine signal-independent manner. Some microbiota metabolic molecules may reach the liver via the portal vein 22 . To explore if microbiota lipid antigens could reach the liver, 14 C-glucose was used to label E. coli that was then intragastrically delivered into mice. After 6 h, there was high radioactivity in the lipids extracted from the livers of mice transferred with 14 C-glucose labelled E. coli but not in the mice transferred with unlabelled E. coli (Fig. 5d). This suggest that microbial lipids can enter the liver at steady state, which allows them to encounter hepatic gdT-17 cells in the liver. gdT-17 cells from both the liver and the spleen were stained with CD1d tetramers loaded with different lipid antigens (PE, PG, CL and PBS57), and liver, but not spleen, gdT-17 cells specifically recognized E. coli-derived lipid antigens (PE, PG and CL) and a model antigen PBS57 (ref. 23; Fig. 5e). In addition, these CD1d tetramer-positive gdT-17 cells in the liver decreased markedly after microbiota depletion (Fig. 5f). These results indicate that liver gdT-17 cells specifically recognize microbiota-derived lipid antigens.
Next, we asked which cells expressed CD1d stimulate hepatic gdT-17 cells. Using fetal liver chimeric mice, we found that the number of hepatic gdT-17 cells decreased only when CD1d was deficient in non-hematopoietic cells, but not in hematopoietic cells (Fig. 6a). Indeed, when using clodronate to deplete macrophages and dendritic cells (DCs), both of which are major hematopoietic cells expressing CD1d, the number of hepatic gdT-17 cells was not reduced but rather slightly increased (Fig. 6b). Interestingly, the hepatocytes expressed a high level of CD1d (Fig. 6c), as previously reported 24 . In the in vitro co-culture system, WT hepatocytes displayed a higher ability to promote gdT-17 cells than hepatocytes from Abx-treated mice and Cd1d À / À mice (Fig. 6d); this difference might arise from the reduced number of endogenous lipid antigens presented by CD1d on Abx-treated hepatocytes. After supplying exogenous lipid antigens, WT or Abx-treated, but not Cd1d À / À , hepatocytes induced an approximately twofold increase in the gdT-17 cell number compared with those without exogenous lipid antigens added (Fig. 6e). These results indicate that hepatocytes can directly present commensal/E. coli lipid antigens through CD1d to promote liver-resident gdT-17 cell expansion.
Hepatic cdT-17 cell accelerate HFD-induced NAFLD progression. Using an Il17ra À / À mouse, Harley et al. 25 showed that IL-17A signalling could accelerate NAFLD through recruiting neutrophils and inducing nicotinamide adenine dinucleotide phosphate (NADPH) oxidase-dependent ROS, which ultimately induced hepatocellular damage. We wondered if the predominant liverresident gdT-17 cell subtype was the source of IL-17A during NAFLD, and, importantly, whether the microbiota promoted NAFLD through these cells.
High-fat diet (HFD)-fed mice displayed elevated numbers of gdT-17 cells in several organs, including the liver (Fig. 7a,  Supplementary Fig. 6a), suggesting that hepatic gdT-17 cells might be one of the main sources of IL-17A in the liver during NAFLD. Though Tcrd À / À mice a body weight increase comparable to that of WT mice ( Supplementary Fig. 6b), they were protected from NAFLD by displaying reduced steatohepatitis, reduced liver damage and a catabatic glucose dysmetabolism (Fig. 7b-d). The reduced NAFLD symptoms in Tcrd À / À mice were also displayed in another mouse NAFLD model induced by high-fat/high-carbohydrate diet (HFHCD) (Fig. 7e), and transfer of HFHCD-fed mice with WT hepatic gdT cells, but not with Il17a À / À hepatic gdT cells, accelerated NAFLD in Tcrd À / À mice (Fig. 7e). These results demonstrate that hepatic gdT-17 cells can accelerate NAFLD. Next, we asked whether the microbiota were involved in hepatic gdT-cell expansion during NAFLD. Both HFD and HFHCD triggered the increase and proliferation of hepatic gdT-17 cells (Fig. 8a,b). However, after depleting commensal microbes, Abx-treated mice that started with lower hepatic gdT-cell numbers had only slight increases compared with the baseline of control mice, and the numbers was far lower than the levels observed in HFD/HFHCD-induced NAFLD mice ( Fig. 8a,b). Another candidate for microbiota associated-IL-17A production was Th-17 cells; unlike Th-17 cells in the intestine 26 and skin 27 , which can be induced to proliferation by their local microbiota, the numbers of hepatic Th-17 cells did not decrease in Abx-treated mice, and HFD and HFHCD did not trigger increased hepatic Th-17 cells, indicating the irrelevant role of hepatic Th-17 cells in this process (Fig. 8c). The downstream effectors of IL-17A during NAFLD 25 , the numbers of neutrophils (Fig. 8d), and the mRNA levels of NADPH oxidase enzymes (Fig. 8e) also decreased in the livers of Abx-treated mice.
More importantly, Abx-treated mice exhibited a similar alleviation of NAFLD as Tcrd À / À mice fed with HFD/HFHCD, as indicated by the reduced steatohepatitis (Fig. 9a,b), the reduced ALT level (Fig. 9c), the elevated body weight (Fig. 9d) and the catabatic glucose dysmetabolism (Fig. 9e). The reduced liver damage, elevated hepatic triglyceride level, elevated body weight and catabatic glucose dysmetabolism in Abx-treated mice could be reversed by transferring gdT cells or IL-17A protein ( Fig. 9f-i). Together, these data suggest that the microbiota function as a co-factor to accelerate HFD/HFHCD-triggered NAFLD via increasing hepatic gdT-17 cells.

Discussion
To our knowledge, our study is the first to describe the unique characteristics and mechanisms of liver-resident gdT cells controlled by the microbiota. First, they are liver resident without exchanging with circulating gdT cells ( Fig. 1e and Supplementary  Fig. 1a-c); second, hepatic gdT cells predominantly produce a high level of IL-17A (Fig. 1a-d); third, they specifically recognize (Fig. 5e) and are promoted by (Figs 5c and 6d,e) CD1d/commensal lipid antigen. gdT-17 cells originate from fetal gd thymocytes and are preferentially located in barrier tissues 28 . The heterogeneity of gdT cells in different locals is more common than previously thought. Skin gdT-17 cells are compartmentally controlled by the skin, but not gut, microbiota 27 ; IL-1R signalling has an important role in microbiota-mediated IL-17A production by PC gdT cells 21 ; IL-17A production by lung-resident gdT cells requires IL-6 signalling instead 29 ; dermal gdT cells preferentially rely on IL-1 and MyD88 signalling to produce IL-17A 30 ; the gdT-cell subtype in the colonic lamina propria produces IL-17A in an IL-23-independent manner 31 . Thus these findings together with the mechanism of liver gdT-cell homeostasis found by us, further indicate that the host-microbiota interaction at distinct barrier sites occurs in a tissue-specific manner 32 .
IL-17A production by gdT cells is often elicited by cytokines or TLRs without prior antigen exposure (termed 'natural gdT-17' cells) 33 , and we showed that TLR and IL-1/IL-23 signalling was not responsible for liver gdT-17 cell homeostasis in a physiological state (Supplementary Fig. 5). These two seemingly contradictory facts together exposed the mystery between gdT-cell antigen recognition and TLR/cytokine recognition. Increasing numbers of studies have indicated that encountering antigen is a prerequisite for gdT cells to respond to inflammatory cytokines 34,35 . Until now, the antigen specificity or antigen requirement of the 'natural gdT-17' cells has remained unknown or controversial 10,28 . However, although gdT cells in the mucosa and epithelium are well-suited to recognize microbial components, their recognized antigens remain scarce 28 . In our study, we demonstrate for the first time that commensal lipid antigens can reach the liver (Fig. 5d) and be presented by CD1d to the gdTCR of liver-resident gdT cells (also a type of 'natural gdT-17') (Figs 5e and 6d,e), which supports the development and homeostasis of this cell subset.
The lipid antigens that specifically link CD1d and the gdTCR include bacterial CL 36 , synthetic a-GalCer 37 , sulfatide 11,38 , synthetic phosphatidylcholine, PE, PG 12 and synthetic and natural PE 39 . Among these, CL, PG and PE can be derived from both bacteria and stressed host cells 40,41 . Indeed. E. coli CL,  PG and PE could partly recover the decreased hepatic gdT-17 cells in Abx-treated mice (Fig. 5c); nevertheless, none of these antigens could completely recover the decline, suggesting that other lipid antigens may also have roles in this process. We showed that liver gdT-17 cells could also be stained by CD-1d-PBS57 tetramer (an improved form of a-GalCer 23 ) (Fig. 5e), which may be a result of the cross-reactivity of the CD1d-lipid antigens recognized by the gdTCR. The cross-reactivity of CD1dphospholipids and CD1d-a-GalCer 37 , as well as that of CD1d-CL and CD1d-phospholipids 42 , were also observed in other gdTCR studies. Several types of hepatic cells, including endothelial cells, Kupffer cells and DCs, can express CD1d, but the strongest CD1d-expressing cell is hepatocytes, which are also critical lipid metabolic factories 24 . Furthermore, CD1d expression on hepatocytes is the main component of hepatic T-cell-lipid antigen recognition 13 . Investigators previously observed that CD1d on hepatocytes could activate NKT cells [14][15][16] . Our study demonstrated that hepatocytes can also promote gdT-17 cells in the presence of lipid antigens (Fig. 6c-e). The difference in lipid antigen recognition by the TCR of gdT cells and that of NKT cells is an interesting topic. For example, although CD1d can present a-GalCer to both gdT cells and NKT cells, the mode of gdTCR-CD1d-a-GalCer recognition appears to be markedly different from that of NKT recognition 37 . Thus, the difference between NKT and gdT cells in CD1d recognition of microbiota-derived lipid antigens requires further investigation.
NAFLD is induced by chronic inflammation in obesity, and the progression of NAFLD is the comprehensive result of hepatic immune cells, including Kupffer cells 43 , NK cells 44 and NKT cells 45 , and liver metabolic cells, including hepatocytes and hepatic stellate cells 44,46 . However, although the metabolic relationship between the microbiota and fatty liver has been shown, how the microbiota influences the hepatic immune response during NAFLD remains unclear [47][48][49] . In this study, we showed that microbiota-maintained liver-resident gdT-17 cells were the main source of IL-17A and could significantly accelerate NAFLD (Fig. 7). An interesting finding was the uncoupling of decreased liver inflammation (Fig. 9a-c) and increased body weight (Fig. 9d) in Abx-treated mice, which was also found by other groups in Il17ra À / À mice 25 and Abx-treated mice 50 . However, this uncoupling was not found in Tcrd À / À mice, who  4,9). (e) HFHCD-fed Tcrd À / À mice were either i.v. transferred with WT hepatic gdT cells or Il17a À / À hepatic gdT cells (2 Â 10 4 , once per week) from the 4th week to the 10th week during HFHCD treatment, and the serum ALT level and GTT curve were evaluated (n ¼ 4 per group). The data are representative of three independent experiments and shown by the mean±s.e.m. (*Po0.05; **Po0.01; ***Po0.001 unpaired Student's t-test (a, b, c), one-way ANOVA post hoc test (e, left panel), two-way ANOVA test (d, e right panel). displayed reduced liver inflammation (Fig. 7b-e) but comparable body weights to those of WT mice, as shown by us (Supplementary Fig. 6b) and others 51 . This suggests that different mechanisms are used by IL-17A to promote hepatic inflammation and lipid metabolism and that the lipid metabolism pathway might be compensated for by other signals in Tcrd À / À mice.
After depleting commensal microbes, hepatic gdT-17 cells did not increase (Fig. 8a), but they did increase in the presence of microbiota during NAFLD. Nevertheless, same as gdT-17 cells in the liver, gdT-17 cells in the gut and other tissues also increase during NAFLD (Supplementary Fig. 6a), mainly because NAFLD is a systemic disease with inflammation in multiple organs 52 . Thus, though there is no mutual exchange or circulation between hepatic gdT cells (Fig. 1e) and gut gdT cells 53 at steady state, we still cannot fully exclude the possibility that a portion of gut gdT-17 cell would traffic to the liver in the disease state. However, instead of the cellular traffic model, we depict a molecular traffic model by displaying that E. coli lipid can reach the liver from gut (Fig. 5d); then the lipid antigen can be presented by hepatocyteexpressed CD1d to directly stimulate hepatic gdT cells (Fig. 6c-e) and ultimately alleviate hepatic gdT-17 cell clones that are specific to CD1d-lipid antigen tetramer (Fig. 5e). Nevertheless, four possible mechanisms exist for the accumulation of hepatic gdT-17 cell during NAFLD from our analysis. First, the excessive accumulation of fat-related lipid antigens during NAFLD may further affect hepatic gdT-17 cells; because it is difficult to distinguish the sources of lipids in vivo, there is still more work to do. Second, because NAFLD is always accompanied by the over growth of commensal microbes [47][48][49] , an elevated global microbe load or several specific bacteria may induce a higher number of hepatic gdT-17 cells as described in our study. Third, HFDinduced inflammation may stimulate the microbiota-maintained hepatic gdT-17 cells as a bystander effect. Forth, there may be a portion of gdT-17 cells activated in other tissues (for example, gut) traffic into the liver. Similar to that observed in the mouse liver, healthy human livers also contain a distinct Vd3 þ gdT-cell subset 54 ; interestingly, they recognize CD1d and release IL-17A after activation 55 . However, their detailed characteristics and precise mechanisms are unclear. Using a mouse model, our work demonstrates for the first time that hepatic gdT cells are a unique liver-resident subset. Hepatocyte CD1d could present gut commensal lipid antigens to hepatic gdT cells, which drove them to predominantly produce IL-17A and maintained their haemostasis. Our results reveal a novel crosstalk between metabolism and immunity in the liver-gut axis and also suggest a tissue-specific interaction between the microbiota and immune cells in the liver (Supplementary Fig. 7).

Methods
Mice. C57BL/6 mice were purchased from the Shanghai Laboratory Animal Center (SLAC, Chinese Academy of Sciences); Rag1 À / À mice were obtained from the Model Animal Research Center (Nanjing University); Cxcr6 gfp/gfp and CD45. mice were purchased from the Jackson Laboratory; Tcrd À / À mice were a gift from Dr Zhinan Yin (Nankai University); Tlr2 À / À , Tlr4 À / À and Tlr9 À / À mice were a gift from Dr Shaobo Su (Sun Yat-sen University); Il17a À / À mice were a gift from Dr Zhexiong Lian (University of Science and Technology of China (USTC)); and Ja18 À / À and Cd1d À / À mice were a gift from Dr Li Bai (USTC). All of the above mice were housed in a specific pathogen-free facility and all of the animal protocols were approved by Local Ethics Committee for Animal Care and Use at University of Science and Technology of China. Germ-free mice were purchased from SLAC and housed in the germ-free facility at SLAC according to their animal care regulations. The sample size was determined by the 'resource equation' method, taking into account the possible reduction of diet/drink treatment.
Mouse treatment. Six-to ten-week-old male mice were used in most of the experiments with exceptions, such as the use of neonatal mice in the ontogeny experiment and the use of 30-week-old mice in the HFD/HFHCD-induced NAFLD experiment. Commensal microbes were depleted using antibiotics as previously reported 20 , the detailed method are followed. Four kinds of antibiotics including ampicillin (1 g l À 1 ), vancomycin (0.5 g l À 1 ), neomycin sulfate (1 g l À 1 ) and metronidazole (1 g l À 1 ) were dissolved in sterile water and stored in 4°C no more than a week before using. This antibiotic-contained water was supplied as drinking water to adult mice and pregnant mice for more than 4 weeks and was changed every 3 days. The volume of drinking water and the body weights of the mice were monitored twice a week. Mice exhibiting more than a 30% decline in body weight were removed. After antibiotic treatment stopped, Abx-treated mice were co-housed with normal mice for 4 weeks to reconstitute commensals.
For the PAMP restoration experiment, WT mice and Abx-treated mice were intraperitoneally (i.p.) injected with 50 mg curdlan, 50 mg Pam3csk4, 100 mg LPS, 100 mg poly(I:C) or 50 mg CpG (all from Sigma) and harvested 1 day later. For cytokine neutralization, the mice were intravenously (i.v.) injected with 50 mg control IgG, anti-IL-1b or anti-IL-23 antibody twice at 3-day intervals and harvested 3 days later. For the lipid antigen restoration experiment, the mice were i.p. injected with 20 mg E. coli CL, PG or PE or 50 mg E. coli polar lipid extract (all from Avanti Polar Lipids, Alabama) six times at 2-day intervals and harvested 2 days after the last injection. For macrophage and DC depletion, the mice were i.v. injected with 200 ml Cl2MDP-Lip (Vrije Universiteit) twice at 3-day intervals and harvested 3 days later. All of the mice receiving the above treatment were randomly assigned to different groups, but the investigators were not blinded to the treatments or genotypes.
Bacterial diversity analysis. Fresh stool samples were collected and weighed, and bacterial DNA was extracted from the stool using a QIAamp Fast DNA Stool Mini Kit (Qiagen). The 16S rRNA gene was analysed to determine the bacterial composition and diversity using an Illumina MiSeq (Novogene Bioinformatics Technology Co., Ltd).
For bacterial titre analysis, fresh stools were collected and homogenized in sterile PBS. The serially diluted homogenates were plated onto blood agar plates (for total cultivable bacterial determination) or eosin methylene blue (EMB; for E. coli determination) plates at 37°C for 24 h. The colonies were distinguished by their biochemical reactions and counted.
Faeces/E. coli transfer. Mice were treated with antibiotics for 4 weeks, and their antibiotic-containing water was replaced with antibiotic-free water. Then, the mice were intragastrically administered 0.2 ml fresh faeces (50 mg ml À 1 ) or 10 8 /10 9 /10 10 c.f.u. of E. coli that was isolated from an EMB plate and expanded in LB. The mice were analysed 3 weeks later. 14 C labelling of E. coli. E. coli was 14 C labelled as previously reported 56 with a slight modification. Briefly, E. coli was isolated from fresh mouse faeces using an EMB plate, expanded in M9 minimal medium to OD 600 ¼ 0.4 and spiked with 25 mCi/L 14 C-glucose (NEC042V250UC, PerkinElmer) for 16 h. A total of 10 10 c.f.u. of unlabelled or 14 C-labelled E. coli was intragastrically delivered to the mice. The livers were collected 6 h later, the lipids in the liver were extracted using a methanol/chloroform (2:1) mixture and radioactivity was assessed.
Parabiosis. CD45.1 þ and CD45.2 þ mice were joined by parabiosis for 2 weeks as previously described 57 . Age matched and co-housed mice were anaesthetic and shaved. An incision along the lateral aspect of each mouse was made. The mice were then sutured together at the elbow and knee as well as at the skin around the incision. Mice were i.p. injected with 5% glucose and 0.9% sodium chloride to recover energy and water. Buprenex was used to relieve pain. Sulfatrim was added to the drinking water for 7 days.  #TP26301 þ carbohydrates (18.9 g l À 1 sucrose þ 23.1 g l À 1 fructose) in drinking water) and LFLCD (fat 12.5% kcal, carbohydrate 68.1% kcal, protein 19.4% kcal; TROPHIC Animal Feed High-Tech Co. #TP26323 þ carbohydrate-free drinking water). Fresh food was supplied twice a week, and the food and drink consumption were quantified throughout the experiment. The body weights of the mice were monitored weekly.
For the IL-17A and hepatic gdT-cell transfer experiment, Abx-treated mice were i.v. injected with IL-17A protein (PeproTech, 500 ng, once per week) or purified hepatic gdT cells (2 Â 10 4 , once per 2 weeks) from the 4th week on an HFD until harvesting at the 24th week. Tcrd À / À mice were i.v. injected with purified hepatic gdT cells (2 Â 10 4 , once per week) from the 4th week on a HFHCD until collecting at the 10th week.
For the GTT test, mice were fasted for 12 h and i.p. injected with 1 g kg À 1 glucose; for the ITT test, mice were fasted for 3 h and i.p. injected with 0.5 U kg À 1 human fast-acting insulin (Lilly France). Tail vein blood was collected both pre-injection and at various times post injection of glucose or insulin. The glucose levels in the blood were assayed by an automated glucometer (LifeScan). Serum ALT and AST were determined using an automated Chemray 240 clinical analyzer (Rayto, Shenzhen, China). Liver samples that were fixed in 4% paraformaldehyde and paraffin-embedded were sliced into 5-mm-thick sections and stained with H&E. Lipids in the frozen liver tissue were extracted using a methanol/chloroform (2:1) mixture and dissolved in isopropyl alcohol. The triglyceride level was detected using a kit (Huili, China).
Cell preparation. Mice were killed, and the iLNs, spleens, livers, lungs, thymi, intestines, colons and mesenteric LNs were collected. Mononuclear cells (MNCs) from each organ were then separated as previously described with slight changes 58 . Briefly, LNs and thymi were passed through a 200-gauge steel mesh and washed with PBS. Spleens were first passed a 200-gauge steel mesh then RBC lysed and washed with PBS. Livers were passed a 200-gauge steel mesh and the cell pellet were collected in the flow through, the MNCs in the pellets were isolated by gradient centrifugation with 40 and 70% Percoll. Lungs were first excised and minced to small pieces, then digested with 0.1% collagenase I for 1 h at 37°, the large pieces of lung were removed by filtering and the MNCs in the flow through were obtained by gradient centrifugation with 40 and 70% Percoll. Intestines and colons were surgically exclude the peyer's patch and then excised to small pieces, digested with 1 mM DTT for 15 min at 37°C and passed a 200-gauge mesh, IEL in the flow through were isolated by gradient centrifugation with 40 and 70% Percoll, the unfiltered tissue was further digested with collagenase IV for 1 h at 37°C and filtered with 200-gauge mesh, LPL in the flow through were isolated by gradient centrifugation with 40 and 70% Percoll. PC MNCs were obtained by peritoneal lavage using cold PBS. Hepatic gdT cells were purified using a mouse TCRgd þ T-Cell Isolation Kit (Miltenyi Biotec).
In vitro co-culture. Hepatocytes were separated using a previously described two-step perfusion method 59 . Briefly, mice were anaesthetized and the portal vein was cannulated. The liver was first digested with 0.5 mM EGTA and then digested with 0.05% collagenase IV, the digested liver resuspension was passed a 200-gauge steel mesh and the hepatocytes in the flow through were isolated by 40% Percoll. Hepatocytes (5 Â 10 4 per well) were pre-cultured for 24 h to adhere, during which mixed lipid antigens (2.5 mg ml À 1 PG, PE and CL, dissolved in ethanol and chloroform) were loaded or not loaded onto the cells. Then, the hepatocytes were washed and co-cultured with purified hepatic gdT cells (1 Â 10 4 per well) for 3 days. IL-17A expression was detected by FACS, and the cell number was counted.
Flow cytometry analysis. Freshly isolated MNCs were blocked and incubated with the indicated fluorescent mAbs for 30 min at 4°C. The cells were stimulated with 50 ng ml À 1 PMA (Sigma) and 1 mg ml À 1 ionomycin (Sigma) and treated with 10 mg ml À 1 monensin (Sigma), and the cells were then blocked with rat serum and stained for surface markers, fixed, permeabilized and labelled with the indicated intracellular antibody. Antibody information is summarized in Supplementary  Table 1. All samples were collected on an LSRII flow cytometer (BD Biosciences) and analysed using FlowJo software (Tree Star). The gating strategy is showed in Supplementary Fig. 8.
CD1d-lipid antigen tetramer. Brilliant Violet 421-labelled PBS-57-loaded and unloaded mCD1d tetramers were kindly provided by the NIH tetramer core. Unloaded mCD1d tetramer (1 mg ml À 1 , 100 ml) was conjugated with lipid antigens (1 mg ml À 1 PE, PG or CL, 20 ml) in PBS buffer containing 1 mM pepstatin, 1 mg ml À 1 leupeptin and 2 mM EDTA, which was then washed and concentrated using a 30 K microconcentrator (Amicon Ultra-15, Millipore UFC903024). The cells were stained with other antibodies as described in the flow cytometry analysis section, collected on a SP6800 spectral analyzer (Sony Biotechnology Inc.), and analysed using FlowJo software with a gating strategy showed in Supplementary  Fig. 8.
BrdU incorporation. Mice were i.p. injected with 1 mg BrdU three times at 2-day intervals. The BrdU þ cell frequency was evaluated according to the FITC BrdU Flow Kit instructions (BD Pharmingen).
Quantitative RT-PCR. Total RNA from liver tissue was extracted using TRIzol reagent (Invitrogen). Gene expression was analysed according to the instructions of the SYBR Premix Ex Taq kit (Takara) and quantified using the DDCt method. All primers (Supplementary Table 2) were synthesized by Sangon (Shanghai, China).
Statistics. Student's t-test for two groups and a one-way ANOVA for more than two groups were used to determine statistically significant differences. A two-way ANOVA test was used to determine differences in the GTT and ITT tests. Differences achieving values of Po0.05 were considered statistically significant.
Data availability. The data that support the findings of this study are available within the article and its Supplementary Information files or from the corresponding authors on request.