Introduction

T cell immunoglobulin and mucin domain-containing protein 3 (TIM3), first discovered in 2002 (ref.1), is a member of the TIM family of immunoregulatory proteins. These are characterized by a common structural organization consisting of an amino-terminal immunoglobulin variable domain (V domain) with five noncanonical cysteines, a mucin stalk, a transmembrane domain and a cytoplasmic tail. Members of the TIM family are encoded by three genes in humans (HAVCR1, HAVCR2 and TIMD4, encoding TIM1, TIM3 and TIM4, respectively) and eight genes in mice, located on chromosome band 5q33.2 and chromosome band 11B1.1 in humans and mice, respectively. The genes encoding the TIM family are located within a genomic region that has been associated with allergy and asthma2, towards the centromeric end of the IL4, IL5 and IL13 gene locus3. Of the TIM family, TIM3 has received the most attention because of its association with the regulation of immune responses in autoimmunity and cancer. Although it was originally identified as a molecule expressed by interferon-γ (IFNγ)-producing CD4+ and CD8+ T cells1, many other cell types, including regulatory T cells (Treg cells)4, myeloid cells5, natural killer (NK) cells6 and mast cells7, have been shown to express TIM3. Thus, therapeutic targeting of TIM3 likely modulates immune responses by acting on multiple cell types. Blockade of TIM3 is currently being investigated in clinical trials for treatment of cancer alongside the inhibition of checkpoint receptors such as lymphocyte activation gene 3 protein (LAG3) and T cell immunoreceptor with Ig and ITIM domains (TIGIT)8. Here, we discuss the advances in our understanding of TIM3 biology.

Molecular mechanism of TIM3 function

Structure and signalling

A unique feature of TIM3 is its lack of known inhibitory signalling motifs in its cytoplasmic tail (Fig. 1). Unlike more ‘classic’ checkpoint receptors such as programmed cell death 1 (PD1) and TIGIT, its cytoplasmic tail contains five tyrosines which are conserved between humans and mice. Although the precise intracellular signalling mechanism has not been fully elucidated, it is known that Tyr256 and Tyr263 allow interactions with HLA-B-associated transcript 3 (BAT3)9 and the tyrosine kinase FYN10. TIM3 can be found in lipid rafts and is recruited to the immunological synapse on T cell activation, where it can interact with both BAT3 and the tyrosine kinase LCK11. When TIM3 is not bound by a ligand, BAT3 is bound to its cytoplasmic tail and recruits the active, catalytic form of LCK. The current hypothesis is that in this state, TIM3 is permissive to T cell activation. Both the soluble lectin galectin 9 and the adhesion molecule carcinoembyronic antigen-related cell adhesion molecule 1 (CEACAM1), two ligands described for TIM3 (see later), were shown to trigger phosphorylation of Tyr256 and Tyr263 by the tyrosine kinase ITK12,13. Upon phosphorylation, BAT3 is released from TIM3, thereby allowing TIM3 to exert its inhibitory function. Bat3-knockout T cells have an elevated expression of co-inhibitory receptors and exhibit reduced pathogenicity in the context of autoimmunity, indicating feedforward negative regulation through BAT3 (ref.9). FYN can potentially contribute to TIM3-mediated inhibitory signalling as it is known to bind the same region of TIM3 as BAT3, indicating potential competition between FYN and BAT3 for TIM3 binding. Moreover, FYN is known to induce T cell anergy by activating phosphoprotein associated with glycosphingolipid-enriched microdomains 1 (PAG), leading to the recruitment of the tyrosine kinase CSK, which then phosphorylates LCK on an inhibitory residue14, resulting in the suppression of T cell receptor (TCR) signalling. TIM3 can also associate with the receptor phosphatases CD45 and CD148 and disrupt the immunological synapse, which has been proposed as an additional mechanism of T cell inhibition11.

Fig. 1: Models for TIM3–ligand interactions.
figure 1

a | In its ligand-unbound form, T cell immunoglobulin and mucin domain-containing protein 3 (TIM3) interacts with HLA-B-associated transcript 3 (BAT3) and maintains T cell activation by LCK recruitment. b | Galectin 9 can be bound to the surface of tumour cells and can also be secreted by tumour cells, antigen-presenting cells (APCs) and other cells in the parenchyma, or can be secreted in autocrine fashion by TIM3-expressing cells. Its carbohydrate recognition domains bind to the glycan-binding site on TIM3. Galectin 9 has two carbohydrate recognition domains and can promote the oligomerization of TIM3, thus potentially facilitating the formation of other TIM3–ligand complexes such as carcinoembyronic antigen-related cell adhesion molecule 1 (CEACAM1)–TIM3. In addition, phosphatidylserine (PtdSer) released from apoptotic cells can bind the FG–CC′ cleft-binding site on TIM3. On binding to galectin 9 or CEACAM1, Tyr256 and Tyr263 in the intracellular domain of TIM3 are phosphorylated; this releases BAT3 and allows recruitment of the tyrosine kinase FYN. This results in the disruption of immune synapse formation and in phosphatase recruitment. Ultimately, the cell becomes anergic or undergoes apoptosis, which is mediated by intracellular calcium release. c | Most of the TIM3-targeted antibodies that facilitate antitumour immunity interfere with either CEACAM1 or PtdSer binding to TIM3, thus maintaining the TIM3–BAT3 interaction. NK cell, natural killer cell.

On the other hand, ectopic expression of TIM3 in Jurkat cells in vitro in the context of short-term TCR stimulation was shown to promote TCR signalling10. In addition, TIM3 expression has been associated with increased AKT–mechanistic target of rapamycin (mTOR) signalling, which was considered a sign of co-stimulation by TIM3 (ref.15). These observations led to an ongoing discussion of whether TIM3 is a co-stimulatory or a co-inhibitory receptor, a debate which may be resolved following acquisition of new data regarding the effects of TIM3 loss-of-function mutations in human disease and initial clinical trials with TIM3-blocking antibodies in cancer patients (discussed later).

TIM3 is also found in a soluble form that lacks the mucin and transmembrane domains. In mice, this form may be generated by alternate splicing16,17; however, the existence of this splicing variant in humans is debated18. Instead, it has been proposed that soluble TIM3 is a result of metalloproteinase-dependent cleavage that facilitates its shedding from the cell surface19. Since elevated plasma levels of TIM3 have been reported in patients with graft-versus-host disease20 and in patients with osteosarcoma21, soluble TIM3 may serve as a biomarker for these maladies. Elevated plasma levels of TIM3 also appear to correlate with enhanced viral load in patients with HIV infection18. However, the exact function of soluble TIM3 is not known at present. One can postulate that it may serve as a decoy receptor and act as a ‘molecular sink’ for TIM3 ligands, thereby interfering with the inhibitory function of TIM3. By contrast, in vitro experiments show that soluble TIM3 can reduce IL-2 production by T cells22.

Ligands of TIM3

TIM3 has been reported to have multiple different ligands (galectin 9, phosphatidylserine (PtdSer), CEACAM1 and high mobility group protein B1 (HMGB1)), which bind to different regions on the TIM3 extracellular immunoglobulin V domain. How TIM3 interacts with each of these ligands and the biological consequences are discussed in the following subsections.

Galectin 9

The first ligand attributed to TIM3 was galectin 9, a C-type lectin broadly expressed and secreted by many haematopoietic cells that binds to carbohydrate moieties on cell surface proteins. On TIM3, it binds to carbohydrate motifs on the immunoglobulin V domain and can induce intracellular calcium influx and cell death in TIM3+ T cells17 (Fig. 1). The precise binding site for galectin 9 on TIM3 has not been clearly defined as it requires the prediction of glycosylation sites that are different between mouse and human TIM3; however, predicted residues that can undergo glycosylation in mouse TIM3 include Thr44, Asn100 and Asn74 (ref.23). Galectin 9 binding results in the oligomerization of TIM3 on the cell surface, inducing the release of BAT3 from the intracellular tail of TIM3. This, in turn, leads to T cell inhibition and ultimately to cell death9.

Multiple lines of evidence highlight the role of the TIM3–galectin 9 interaction in suppressing immune responses. For example, the administration of soluble galectin 9 ameliorated disease in the experimental autoimmune encephalomyelitis (EAE) mouse model of multiple sclerosis, whereas knockdown of galectin 9 expression with siRNA exacerbated disease in this model17. Indeed, in patients with multiple sclerosis, the number of TIM3+ and galectin 9-positive CD4+ T cells in the blood is increased during the benign stages of multiple sclerosis compared with more advanced stages24.

In TIM3+ Treg cells, which have superior suppressor function compared with TIM3 Treg cells, blocking of the TIM3–galectin 9 pathway decreases their suppressive capacity25. TIM3 transgenic mice exhibit accelerated tumour growth in an EL4 lymphoma model and decreased severity of disease in an EAE model26. Moreover, CT26 colorectal carcinoma tumour cells can secrete galectin 9 and thus inhibit TIM3+ T cells in trans, leading to cell death27. In human acute myeloid leukaemia (AML), an autocrine TIM3−galectin 9 loop drives the self-renewal of AML stem cells by activating the nuclear factor-κΒ (NF-κB) and β-catenin pathways28 and by secreting both TIM3 and galectin 9, cancer cells can evade immune surveillance22. Moreover, galectin 9 expression may be induced by TIM3 as an inhibitory feedback mechanism. Transgenic mice that overexpress TIM3 in T cells have increased numbers of CD11b+Gr-1+ myeloid-derived suppressor cells that express galectin 9; moreover, the same increased expansion was seen in mice bearing transgenic galectin 9, and introduction of TIM3 deficiency by crossing galectin 9 transgenic mice with Tim3-knockout mice abrogates this expansion26. Collectively, these data show that the TIM3–galectin 9 interaction can act by different mechanisms and on different cell types to suppress immune responses.

The TIM3–galectin 9 interaction is also important in both bacterial and viral infections. During Mycobacterium tuberculosis infection, both lung CD4+ T cells and lung CD8+ T cells express TIM3, which allows them to interact with galectin 9-positive macrophages, leading to a restriction of bacterial proliferation within the macrophages (discussed later)29. In the context of HIV infection, in vitro experiments have shown that the binding of galectin 9 to TIM3 on CD4+ T cells lowers the expression of the HIV co-receptors CCR5, CXCR4 and α4β7 on the T cells, thus enabling them to resist HIV infection30. However, the signalling mechanisms in this context are as yet unexplored.

Importantly, there is also evidence that galectin 9 can exert effects that are independent of TIM3. For example, in vitro experiments demonstrated that galectin 9 can enhance cytokine production in both T helper 1 cells (TH1 cells) and TH2 cells31 and suppress TH17 cell differentiation. This was independent of TIM3 expression but required O-linked glycosylation of an unknown glycoprotein and IL-2 (ref.32). By virtue of binding N-linked sugars of a defined configuration, galectin 9 can also bind to the cell-surface glycoprotein CD44 on Treg cells and thereby regulate their function33. On the basis of binding assays and functional assays performed with HIV-1-infected human T cells, some investigators have also argued that TIM3 does not bind galectin 9 (ref.34), which is in stark contrast to most of the TIM3 literature. Since galectin 9 does not bind in the unique pocket framed by the FG and CC′ loops of TIM3, it was predicted that TIM3 must have additional ligands that bind to this pocket23.

Phosphatidylserine

The elucidation of the crystal structures of TIM1, TIM3 and TIM4 led to the discovery of PtdSer, a phospholipid that acts as a surface marker for apoptotic cells (reviewed in ref.35), as a ligand for all TIM family members36,37. PtdSer binds to the pocket framed by the FG and CC′ loops in the TIM3 immunoglobulin V domain37, with coordinated calcium binding23 involving the amino acids Asn119 and Asp120 (ref.38). However, TIM3 binds PtdSer with at least five times lower affinity compared with other TIM family members39. Notably, the binding site for PtdSer is on the opposite side of the immunoglobulin V domain relative to the predicted sites for galectin 9 binding35 and appears to be affected by allelic polymorphisms as BALB/c allelic variants have a stronger capacity to bind PtdSer than congenic C.D2 Es-Hba variants37. Although binding of PtdSer to TIM3 does not lead to engulfment and phagocytosis of apoptotic material by T cells, it is postulated that it is important for antigen cross-presentation by TIM3+ dendritic cells (DCs)39. Importantly, all mouse and human anti-TIM3 antibodies that have demonstrated functional efficacy in vivo and in vitro interfere with TIM3 binding to PtdSer and also to CEACAM1 (discussed later)40.

HMGB1

The alarmin HMGB1 was reported as a ligand for TIM3+ DCs, but its binding site has not been determined41. HMGB1 binds to DNA released from dying cells and facilitates nucleic acid sensing by Toll-like receptors (TLRs). HMGB1 can also be secreted by tumour cells42. Thus, TIM3+ myeloid cells in the tumour microenvironment may act as a molecular sink for HGMB1, thereby interfering with its function in innate immune activation (Fig. 2). Indeed, in models of MC38 colon carcinoma, TIM3 blockade was shown to increase both the efficacy of DNA-based tumour vaccines and responsiveness to cisplatin41, which is known to trigger release of HMGB1 (ref.43). This effect was unchanged by the administration of galectin 9-binding antibodies or annexin V, which binds PtdSer, suggesting that the HMGB1–TIM3 interaction is independent of other TIM3 ligands. It will be interesting to test whether TIM3 blockade also increases the efficacy of other chemotherapeutic drugs that trigger HMGB1 release. Whether these effects depend on the interaction of TIM3 with HMGB1 or with PtdSer displayed on the surface of dying cells remains to be addressed. Although intriguing, there has been little follow up of the TIM3–HMGB1 interaction and its relevance to TIM3-mediated regulation beyond the initial report41.

Fig. 2: TIM3 function in dendritic cells.
figure 2

a | Endosomal Toll-like receptor 3 (TLR3), TLR7 and TLR9 sense viral double-stranded RNA, single-stranded RNA and bacterial or viral DNA, respectively, and activate several downstream transcription factors, such as nuclear factor-κB (NF-κB) and several interferon regulatory factors (IRFs), giving rise to the secretion of IL-12 and interferon-γ (IFNγ). In dendritic cells, the activation of TLR3, TLR7 and TLR9 can be suppressed by T cell immunoglobulin and mucin domain-containing protein 3 (TIM3)-mediated sequestration of high mobility group protein B1 (HMGB1), a protein that binds nucleic acids and facilitates their uptake, leading to TLR signalling. Thus, TIM3 serves as a molecular sink and prevents the entry of the nucleic acids into the cell, thereby dampening both antiviral and antitumour immune responses. Antibody-mediated blockade of TIM3 can relieve this suppression. b | TIM3 is highly expressed in type 1 conventional dendritic cells in tumours, where its blockade promotes the expression of CXC-chemokine ligand 9 (CXCL9) by type 1 conventional dendritic cells and the recruitment of CD8+ T cells that can respond to tumour antigens released by chemotherapy-induced apoptosis. More findings on TIM3 function in dendritic cells and other non-T cell leukocytes are described in Table 1. PtdSer, phosphatidylserine.

CEACAM1

The most recently discovered ligand for TIM3 is the glycoprotein CEACAM1 (refs13,44), which is known to have a role in the regulation of antiviral responses45. In vitro, CEACAM1 is co-expressed with TIM3 on ovalbumin peptide-stimulated and Staphylococcus aureus endotoxin B-stimulated T cells following repeated exposure to these antigens, and is therefore thought to contribute to establishing T cell tolerance13.

CEACAM1 is thought to bind to the CC′ and FG loops of TIM3 (ref.13). CEACAM1 has also been found to be able to bind TIM3 intracellularly, which appears to be important for the maturation of TIM3, as mutant forms of either TIM3 or CEACAM1 co-expressed in HEK293 cells resulted in intracellular TIM3 accumulation and TIM3 hypoglycosylation13. Accordingly, in a mouse model of colitis, CEACAM1−/− T cells expressed reduced surface levels of TIM3 concomitant with greater production of the effector cytokines IFNγ, tumour necrosis factor (TNF), and IL-17A13. CEACAM1 binding can trigger the release of BAT3 from TIM3, thus allowing TIM3-mediated inhibition of TCR signalling13. In addition to its expression by T cells, CEACAM1 is expressed by DCs46, monocytes47 and macrophages48. Thus, the TIM3–CEACAM1 axis can potentially inhibit immune responses either in cis or in trans in both T cells and myeloid cells. The cis interaction promotes the stability of mature TIM3 on the cell surface, and both cis and trans interactions mediate the inhibitory function of TIM3.

The TIM3–CEACAM1 axis has the potential to be an important target for immunotherapy for cancer. As noted earlier, all antibodies directed against mouse TIM3 that have demonstrated functional efficacy in vivo in mouse cancer models interfere with both PtdSer binding and CEACAM1 binding40. CEACAM1 is also highly expressed by some tumour cells, such as melanoma cells, where it is thought to dampen T cell responses49. However, CAECAM1 may also have tumour-suppressive function under some circumstances, since it was found that CEACAM1−/− mice develop a higher tumour burden in a chemically induced model of colorectal cancer50. Thus, TIM3–CEACAM1 interactions in the tumour microenvironment are highly complex and require careful molecular dissection to delineate regulatory mechanisms.

Multiple ligand interactions

Notably, all anti-TIM3 antibodies with antitumour activity have the capacity to interfere with PtdSer and CEACAM1 binding but not galectin 9 binding to TIM3 (ref.40). This leaves open the question of whether and how galectin 9 may play a role in TIM3 signalling in tumours. It has been shown that galectin 9, which has two carbohydrate recognition domains, can form a molecular bridge between TIM3 complexed with other ligands, such as CEACAM1 and PtdSer, and the phosphatases CD45 and CD148 at the immunological synapse11.

Given the variety of TIM3 ligands, it is difficult to speculate as to how TIM3+ cells can discriminate between these ligands. Since the affinity of TIM3 for all of its ligands is unknown at present, it is possible that the distribution and individual expression level of each ligand in a given tissue microenvironment determines the signalling through TIM3. Indeed, some tissues, such as those of the gastrointestinal tract and specifically the colon, contain cells that express high levels of CEACAM1 (ref.51), whereas others, such as those of the liver, are enriched in cells expressing galectin 9 (refs52,53). Notably, galectin 9 binding does not preclude binding to other ligands in different binding sites. Moreover, we predict that galectin 9 binding might facilitate TIM3 binding to other ligands by inducing an allosteric change that makes the cleft framed by the FG and CC′ loops more accessible to other ligands (Fig. 1).

The functions of TIM3 in immune regulation

Role of TIM3 in transplant tolerance and autoimmunity

TIM3 was initially identified as a marker for IFNγ-secreting effector CD4+ TH1 cells and CD8+ type 1 cytotoxic T cells1, and was later found to be expressed by human TH1 cells54. The TH1 cell transcription factor T-bet was found to regulate TIM3 expression55 in cooperation with other transcriptional regulators, including NFIL3 (ref.56), PRDM1 and MAF57. Treatment of mice with TIM3-blocking antibodies or a soluble TIM3–immunoglobulin fusion protein, which acts as a molecular sink for TIM3 ligands, exacerbates disease in the EAE model and results in enhanced TH1 cell immunity and macrophage hyperactivation. This gave the first indication that TIM3 might function as an inhibitory molecule that serves to limit IFNγ-driven inflammation. In keeping with the expression of TIM3 on IFNγ-secreting TH1 cells, TIM3 deficiency leads to TH1 cell dysregulation but not TH17 cell dysregulation in EAE58. Later it was confirmed that TIM3 is also expressed in human TH17 cells, albeit to a lesser extent relative to TH1 cells59. Notably, the surface expression of TIM3 on T cells is low in patients with ulcerative colitis60, multiple sclerosis61,62, rheumatoid arthritis63 and psoriasis64 compared with that on T cells from healthy controls; this is consistent with the limited regulation of the inflammatory properties of these T cells in patients with these autoimmune diseases. Indeed, TIM3 blockade is known to worsen disease in multiple preclinical models of autoimmune inflammation, including inflammatory bowel disease65 and diabetes66. Moreover, both TIM3-deficient mice and mice treated with a TIM3–immunoglobulin fusion protein exhibit defects in the induction of antigen-specific tolerance in two different models: administration of high-dose soluble antigen and transfusion of allogeneic donor splenocytes together with CD154 (CD40L) blockade16,66. In line with these data, administration of galectin 9 suppresses allogenic skin graft rejection67.

It is now recognized that TIM3 is also expressed by FOXP3+ Treg cells, specifically those at tissue sites. In vitro experiments have shown that TIM3+ Treg cells are enriched in effector molecules such as IL-10, perforin, and granzyme A53,68,69 and have superior suppressive function when compared with TIM3 Treg cells68,69. Indeed, TIM3+ Treg cells appear to be instrumental in the suppression of allograft rejection as their numbers in draining lymph nodes and spleens are significantly increased following allogeneic but not syngeneic grafts. They also show a more suppressive phenotype, as well as higher proliferation rate, compared with allograft-induced TIM3 Treg cells69. An open question is whether TIM3+ Treg cells are highly active effector Treg cells or are a unique subset of Treg cells with specialized effector functions directed at distinct cell types or tissue sites.

Role of TIM3 in M. tuberculosis infection

Infection with M. tuberculosis remains a serious threat to human health. During chronic M. tuberculosis infection in mice, TIM3 expression on CD8+ T cells is elevated70,71. Similarly, elevated TIM3 expression is found on total CD8+ T cells and antigen-specific CD8+ T cells in patients with tuberculosis72,73. In vivo administration of TIM3–immunoglobulin fusion protein reduced the M. tuberculosis burden in M. tuberculosis-infected mice29. Further in vitro analyses demonstrated that TIM3+ T cells interact with galectin 9 on macrophages, which triggers IL-1β production by macrophages and thereby limits intracellular M. tuberculosis replication in macrophages29,74. This is likely mediated by a yet unknown, indirect mechanism, as galectin 9 itself does not contain a signalling motif. A similar role for the TIM3–galectin 9 pathway in triggering antibacterial activity in M. tuberculosis-infected human macrophages has been identified75. Thus, during M. tuberculosis infection, TIM3 seems to have a dual role, mediating the activation of macrophages by galectin 9 to inhibit intracellular bacterial replication while restraining TIM3+ effector T cell responses.

Role of TIM3 in cancer and chronic viral infections

A key obstacle to successful antitumour immunity is T cell dysfunction or T cell exhaustion, a cellular state characterized by compromised proinflammatory cytokine production and cytotoxicity that develops in response to continuous exposure to antigen, most notably in chronic viral infections and cancer76. TIM3 serves as a marker of a highly dysfunctional subset of CD8+ T cells in patients infected with HIV, and in vitro experiments showed that TIM3 blockade with antibodies can restore the proliferative response of TIM3+ CD8+ T cells to HIV-1 peptides77. Subsequently, TIM3 expression was found to be a marker for the most dysfunctional T cell population within the CD8+ PD1+ T cell pool in humans with chronic viral infections (hepatitis C virus infection, hepatitis B virus (HBV) infection) and also in mouse models of infection (lymphocytic choriomeningitis virus infection, HBV infection, Friend virus infection)78,79,80,81. Moreover, TIM3 levels are positively correlated with viral load and disease progression in many viral infections, which is consistent with its proposed inhibitory role77,78,82.

TIM3 also marks the most dysfunctional subset among tumour-infiltrating CD8+PD1+ T cells in cancer79,83,84. Accordingly, antibody blockade of both TIM3 and PD1 was shown to have a synergistic effect with regard to inhibiting tumour growth84 and improving tumour antigen-specific CD8+ T cell responses of patient-derived T cells83,85. In contrast, TIM3 overexpression on T cells exacerbates tumour progression in the EL4 mouse model of lymphoma26. In addition to TIM3+CD8+ T cells, TIM3+ Treg cells might also be targeted by anti-TIM3 antibody therapy as TIM3+ Treg cells are the predominant Treg cell population in tumours4,53,68 and their presence correlates with tumour severity and progression4. For example, in patients with lung cancer, 60% of Treg cells were found to express TIM3 (ref.4), and in patients with hepatocellular carcinoma, 70% of Treg cells were found to express TIM3 (ref.53). In a mouse model of colorectal cancer, TIM3+ Treg cells accounted for more than 50% of total Treg cells, a proportion that far exceeds their relative frequency in tumour-draining lymph nodes, spleen or blood. Moreover, in tumour models such as the MC38 colon carcinoma mouse model, the vast majority of CD4+TIM3+ T cells also expressed FOXP3 (~90%), indicating that they are Treg cells68. Indeed, TIM3 blockade in combination with PDL1 blockade results in a downregulation of Treg cell effector molecules in TIM3+ Treg cells in vivo68, and TIM3 blockade lowered both TIM3+ Treg cell numbers and tumour growth in a mouse model of head and neck cancer86.

TIM3 can also serve as a prognostic marker for solid tumours such as colon, gastric and cervical cancer, non-small-cell lung cancer and clear cell renal carcinoma, with high levels of TIM3 expression associated with lower survival and vice versa87. Furthermore, elevated TIM3 levels on cells in both the circulation and the tumour tissue of patients with colorectal cancer correlated with an unfavourable disease course88. Similar findings were obtained in patients with HBV-associated hepatocellular carcinoma, where survival is negatively correlated with the presence of TIM3+ T cells89. In patients with hepatocellular carcinoma negative for HBV, TIM3 expression on both CD4+ T cells and CD8+ T cells is minimal89.

TIM3 expression is also observed on malignant haematopoietic cells, specifically AML cells (reviewed in ref.90). TIM3 is highly expressed by leukaemic stem cells but not by normal haematopoietic stem cells in these tumours91,92,93. Most of the disease-initiating cells appear to express TIM3, since xenograft models show that TIM3+ cells from patients with AML, but not TIM3 cells, can establish AML when administered to immunodeficient mice94. The exact role of TIM3 on leukaemic stem cells is not known; however, anti-TIM3 antibody blockade can ameliorate disease in a xenograft model of AML, indicating that TIM3 can act as both a biomarker and a clinical target94.

One feature of TIM3 that makes it an attractive target for checkpoint-targeted cancer immunotherapy is its co-expression with PD1 (refs83,84). In preclinical models, co-expression of TIM3 and PD1 is observed in both CD4+ T cells and CD8+ T cells and TIM3+PD1+ T cells appear to be most dysfunctional, expressing significantly lower amounts of the effector cytokines IFNγ, TNF and IL-2 compared with their TIM3 counterparts84. Moreover, TIM3 can be upregulated as a result of PD1-directed therapy; this has been shown to be mediated by the phosphoinositide 3-kinase (PI3K)–AKT pathway95. Co-blockade of the TIM3 and PD1 pathways results in greater restoration of T cell responses than PD1 pathway blockade alone in mouse models of melanoma, colorectal cancer and AML84,96,97,98, with similar outcomes in mouse models of hepatitis C virus infection, HBV infection, and lymphocytic choriomeningitis virus infection78,79,99,100.

The immunosuppressive cytokine IL-27 was recently shown to be one of the key drivers of TIM3 expression on both CD4+ T cells and CD8+ T cells56. IL-27 induces transcriptional regulators that activate a module of co-inhibitory receptors that includes not only PD1 and TIM3 but also TIGIT and LAG3 (ref.57). Thus, there is a common link between TIM3 and other checkpoint receptors that regulate effector T cell function. As a single agent, TIM3-blocking antibodies do not have a substantial clinical benefit; however, they elicit stronger immune responses when combined with other checkpoint molecules. The advantage of targeting TIM3 with checkpoint inhibitors is that this may circumvent some of the toxic effects that have been observed with CTLA4- or PD1-directed checkpoint blockade, since its expression on T cells is limited to terminally differentiated IFNγ-producing T cells, unlike CTLA4 and PD1, which are globally expressed on all T cells on activation. Moreover, TIM3 expression is restricted to intratumoural Treg cells, whereas CTLA4 is broadly expressed by all Treg cells. Finally, both CTLA4-deficient mice101,102 and PD1-deficient mice103,104 develop spontaneous autoimmunity, whereas TIM3-deficient mice do not. Thus, blockade of TIM3 in combination with CTLA4 or PD1 is likely to be both safer and more specifically targeted to intratumoural T cells than the combination of CTLA4 and PD1 blockade. Recently, co-blockade of TIM3 and the ectonucleotidase CD39, with a TIM3-blocking antibody and the small molecule POM-1, was shown to significantly reduce tumour growth in the B16F10 melanoma mouse model. Both TIM3 and CD39 were found be highly expressed by CD8+ T cells from cancer patients who were unresponsive to anti-PD1 therapy105, suggesting that TIM3-blocking antibodies can be used as cancer therapeutics in combination with drugs other than anti-PD1 antibodies.

In line with the observation that TIM3 is a marker for the most dysfunctional T cell subset in both cancer and chronic viral infection, it has recently been noted that TIM3 expression is a marker for T cells that have lost the expression of Tcf7, which encodes TCF1, a transcription factor that maintains stemness and restrains effector differentiation106. Thus, there is a negative correlation between TIM3 and TCF1 expression, where TCF1 expression can serve as a marker for stem-like cells and TIM3 expression can serve as a marker for terminally differentiated effector and exhausted cells that cannot be reinvigorated by checkpoint blockade106,107,108,109,110,111. Indeed, the presence of PD1 T cells and PD1+ T cells that express TCF1 is essential for the response to checkpoint blockade107,109. Whether there is a regulatory relationship between TIM3 and TCF1 and whether this occurs at the transcriptional or post-translational level is currently not known.

Role of TIM3 in non-T cells

Although it was initially thought that TIM3 is expressed only by T cells, it is now appreciated that TIM3 is expressed by multiple cell types, including DCs, macrophages, myeloid-derived suppressor cells and NK cells (Table 1). Although the role of TIM3 in some of these cell types is still being investigated, accumulating data indicate that TIM3 also functions as an inhibitory receptor in these cells. The current knowledge of TIM3 biology in these non-T cell populations is limited.

Table 1 The role of TIM3 in non-T cell leukocytes

There appears to be a close connection between TIM3 expression and the modulation of TLR activation in myeloid cells. In vitro experiments showed that TIM3 antibody blockade or TIM3 knockdown with siRNA in monocytes and macrophages that were stimulated through multiple different TLRs increases the production of many cytokines and the expression of TLR2 and TLR4, indicating that TIM3 is a negative regulator of the TLR response112,113. Accordingly, TIM3 blockade can worsen experimental sepsis in wild-type but not Tlr4−/− mice113. In a model of colitis induced by dextran sodium sulfate treatment, colonic macrophages expressed lower levels of TIM3 and showed increased macrophage polarization towards a proinflammatory response, with excessive production of TNF, IL-1β, NOS2, IL-6 and IL-12. In contrast, overexpression of TIM3 in macrophages both reduced the expression of genes encoding proinflammatory mediators and attenuated colitis in this model114. TIM3 overexpression on macrophages appears to attenuate the massive inflammatory response to dextran sodium sulfate and thereby prevent undesired tissue disease. In line with this, TIM3 expression is found to be upregulated on peripheral blood mononuclear cells in patients with acute sepsis but is found to be suppressed in patients with severe sepsis113. TIM3 is also highly expressed on human peripheral blood mononuclear cells during pregnancy, where TIM3 appears to blunt the response to TLR4 stimulation115. An analysis of how TIM3 inhibits signalling through TLR4 suggested a mechanism of NF-κB inhibition that is dependent on the activation of the PI3K–AKT pathway and the NF-κB inhibitor protein A20 (ref.113). Lastly, as discussed earlier, TIM3 expression on DCs may inhibit nucleic acid sensing through TLRs41.

TIM3 has also been reported to have TLR-independent effects in macrophages. TIM3 can act as an adaptor protein for the transcription factor STAT1 (signal transducer and activator of transcription 1), which binds to Tyr256 and Tyr263 of the intracellular tail of TIM3. This leads to the suppression of transcription of the inflammatory microRNA miR-155 and enhances the function of SOCS1, a negative regulator of Janus kinase (JAK) signalling. This, in turn, promotes M2 macrophage polarization116. TIM3 has also been shown to interact with the transcription factor NRF2 (nuclear factor erythroid 2-related factor 2), which leads to the downmodulation of the scavenger protein CD36 and the haem oxygenase 1–IL-1β pathway. In a model of Listeria monocytogenes infection, this results in lower levels of bacterial ingestion by macrophages117.

Given that TIM3 mediates the capture of apoptotic material by binding to PtdSer, its expression on mouse macrophages and CD8+ DCs can facilitate cross-presentation39. Of note, other members of the TIM family, such as TIM1 and TIM4 (the latter being highly expressed on several tissue-resident macrophages and DCs), have a much higher affinity for PtdSer and thus are more potent in allowing the uptake of apoptotic material118. The lower binding affinity of TIM3 for PtdSer is because TIM3 has only two of the four conserved residues in the FG and CC′ loops that mediate the interaction with PtdSer35,37. How the roles of TIM3 in antigen cross-presentation39 and suppression of myeloid cell responses to inflammatory stimuli117 relate to each other is incompletely understood. Which one of these functions dominates could be a critical determinant of T cell activation and disease outcome.

A recent study showed that TIM3 blockade improved the antitumour response to chemotherapy in a mouse model of breast cancer119. Anti-TIM3 antibodies promoted the production of the chemokine CXCL9 by BATF3+CD103+ DCs (also called type 1 conventional DCs (cDC1s)) following exposure to chemotherapy-induced tumour debris. cDC1s are known for their ability to cross-present tumour antigens and license CD8+ T cells to eliminate tumours and were required for the effect of anti-TIM3 antibodies in this model (Fig. 2). In this study, antibodies directed at HGMB1 or CEACAM1 did not facilitate CXCL9 secretion by tumour DCs, but antibodies against galectin 9 did, suggesting that blockade of galectin 9–TIM3 interactions might have antitumour potential. To gain a better understanding of the role of TIM3 in DCs, how it affects antigen cross-presentation and how it inhibits the inflammatory phenotype in autoimmunity and cancer, will require genetic experiments involving the DC-specific deletion of TIM3 using the CD11ccre (ref.120), ZBTB46cre (ref.121) or cDC1-specific XCR1cre (ref.122) constructs.

In models of brain ischaemia, TIM3 is upregulated in glial cells in a hypoxia-inducible factor 1α (HIF1α)-dependent manner, and TIM3 blockade has been shown to limit infarction size and the local recruitment of neutrophils into the ischaemic lesion123. Microglia express high levels of TIM3 (ref.5), and both neurons124 and astrocytes125 express galectin 9 and can facilitate TIM3–galectin 9 interactions with microglia. The importance of the TIM3–galectin 9 axis in the central nervous system and its importance in neurological diseases is yet to be determined. Collectively, these data support the hypothesis that TIM3 is an inhibitory receptor on myeloid cells and that modulating TIM3 signals may be of therapeutic value in treating many clinical disorders ranging from autoimmunity to cancer.

TIM3 is also expressed by mature resting CD56dim NK cells and is further upregulated on activation in response to cytokine stimulation6,126. High expression of TIM3 is a marker of effector NK cells that are producing IFNγ and are undergoing degranulation6. However, similarly to its expression by T cells in cancer, TIM3 might be a marker for dysfunctional NK cells. Accordingly, TIM3 blockade has been shown to reverse NK cell dysfunction in NK cells purified from the blood of patients with melanoma127.

Recent data also describe an important role for TIM3 expression in NK cells at the maternal–fetal immune interface (reviewed in ref.128). NK cells constitute the vast majority of the total pool of decidual immune cells (60–80%)129. The numbers of TIM3+ NK cells are significantly increased in human placentae during the first trimester of pregnancy, and these cells exhibit a more immunosuppressive phenotype compared with TIM3 NK cells. In patients with recurrent miscarriages, these TIM3+ NK cells were found to have weaker immunosuppressive properties, and pregnant mice treated with anti-TIM3 antibodies have smaller litter sizes due to the disruption of maternal–fetal tolerance130. In addition to the reduced immunosuppressive capacity of TIM3+ NK cells in patients with recurrent miscarriages, the actual numbers of TIM3+ NK cells were found to decrease following miscarriage131. It has been shown that TIM3 inhibits NK cell cytotoxicity towards human embryonic trophoblasts in a manner that is dependent on galectin 9 (ref.132), which is expressed by decidual Treg cells133 and/or the trophoblast cells131.

On mast cells, TIM3 has been suggested to have an activating function, which is in contrast to its role in other immune cells. Antibodies against TIM3 enhance the secretion of IL-3, IL-4, IL-6 and IL-13 in mast cells following IgE sensitization and antigen-dependent activation in vitro134. Moreover, in vitro experiments showed that both genetic deletion or siRNA knockdown of TIM3 significantly diminished IL-6 production following mast cell activation, and genetic deletion of TIM3 abrogated antigen-mediated glycolysis, an effect mediated by the FcεRII receptor7,135. Further investigation is needed to decipher the role TIM3 signalling in mast cells and its importance with regard to allergies and mast cell biology.

Genetic TIM3 alterations in human disease

Polymorphisms

In humans, polymorphisms in the coding and non-coding regions of HAVCR2 have been linked to allergic diseases and autoimmunity (reviewed in ref.136). Three main polymorphisms in HAVCR2 have been associated with multiple clinical conditions: −1516G/T (rs10053538) and −574G/T (rs10515746) in the promoter region and +4259T/G (rs1036199) in the coding region. These polymorphisms were associated with rhinitis137, cancer of the gastrointestinal tract138, renal cell carcinoma and pancreatic cancer139,140, non-small-cell lung cancer141 and in patients infected with HIV with non-Hodgkin lymphoma141. Although it is not known how these polymorphisms affect TIM3 expression and function, one study reported that two polymorphisms in PD1 (+8669AA (rs10204525)) and HAVCR2 (−1516G/T) are correlated with high expression of TIM3 by liver-infiltrating lymphocytes in patients with hepatocellular carcinoma who are infected with HBV142. A more direct link between HAVCR2 polymorphism and T cell function was seen in patients with osteoarthritis carrying the HAVCR2 +4259T/G allele, who showed high levels of IFNγ production by their CD4+ T cells143. Importantly, polymorphisms in seven amino acids in the TIM3 immunoglobulin V domain exist between the TH1 cell-biased DBA mice and TH2 cell-biased BALB/c mice, which may contribute to the differential cytokine profiles of the two mouse strains2.

TIM3 germline mutations

With regard to the debate whether TIM3 acts as an inhibitory or a stimulatory molecule, new evidence linking germline HAVCR2 mutations to human disease indicates that TIM3 acts as an inhibitory molecule144,145,146. It was shown that mutations in the TIM3 immunoglobulin V domain, c.245A>G (Tyr82Cys) and c.291A>G (Ile97Met), result in misfolding of TIM3 and loss of its expression on the cell surface of both T cells and myeloid cells. As a result of the mutation, TIM3 aggregates intracellularly. Patients harbouring these loss-of-function mutations exhibit a severe autoinflammatory and autoimmune phenotype with an excessive production of the proinflammatory molecules CXCL10, IL-1 β, IL-18 and soluble CD25. Macrophages are hyperactivated, resulting in excessive IL-1 and TNF levels in the serum and, finally, similarly to what is found in patients with systemic lupus erythematosus, B cells generate autoantibodies against double-stranded DNA. The dysregulated inflammatory state in these patients further promotes the development of subcutaneous panniculitis-like T cell lymphoma (SPTCL)145 (Fig. 3), which can be associated with haemophagocytic lymphohistiocytosis147,148. Strikingly, the onset of SPTCL is much earlier in patients with HAVCR2 mutations as compared with patients with normal HAVCR2, and the disease is more severe, with most patients with mutant TIM3 developing haemophagocytic lymphohistiocytosis.

Fig. 3: TIM3 gene single-point mutations give rise to loss of function in SPTCL.
figure 3

a | Three mutations in the immunoglobulin V domain of T cell immunoglobulin and mucin domain-containing protein 3 (TIM3) have a different epidemiology in the general world population. The Tyr82Cys and Ile97Met mutations can cause disease either as homozygous or compound mutations, and the Thr101Ile mutation may cause disease as a compound mutation with the Tyr82Cys mutation, which suggests a similar function of these residues. b | The TIM3 mutations Tyr82Cys, Ile97Met and Thr101Ile lead to protein misfolding, resulting in TIM3 intracellular aggregates, which abrogate TIM3 signalling. c | Due to the broad expression and function of TIM3 in different immune cells, the absence of TIM3 signalling leads to systemic immune hyperactivation, resulting in myeloid cell dysregulation, a reduction in the number of regulatory T cells (Treg cells) in subcutaneous adipose tissue, excessive proliferation and malignancy of CD8+ T cells in the adipose tissue, which gives rise to subcutaneous fat lesions, and an indirect promotion of autoantibody production by B cells, resulting in autoimmune symptoms with lupus-like disease. dsDNA, double-stranded DNA; HLH, haemophagocytic lymphohistiocytosis; SPTCL, subcutaneous panniculitis-like T cell lymphoma; TNF, tumour necrosis factor.

The Tyr82Cys mutation is predominantly found in populations with Polynesian and East Asian origin, whereas the Ile97Met mutation is characteristic of patients of European origin. Sixty percent of patients with SPTCL examined possessed at least one of these mutations145. An additional study in an independent cohort of Japanese patients with SPTCL detected a homozygous Tyr82Cys mutation in 10 of 13 patients. Since the frequency of a homozygous Tyr82Cys mutation in the general population is estimated to be 1 in 35,000, this mutation likely confers high susceptibility to SPTCL146. Moreover, one patient with SPTCL carried a heterozygous Tyr82Cys mutation and an additional heterozygous mutation, Thr101Ile. Both studies found that HAVCR2 mutations occur mainly in patients of Asian ethnicity, which correlates with the high prevalence of known HAVCR2 single-nucleotide polymorphisms in Korean137 and Chinese Han138 populations. Importantly, the Thr101Ile mutation is a known variant of HAVCR2; when this variant is expressed ectopically in human embryonic kidney 293 (HEK293) cells, TIM3 is hypoglycosylated and remains mainly in the endoplasmic reticulum and Golgi apparatus, with minimal surface expression13. Notably, homozygous Thr101Ile mutations were also found in patients with inflammatory bowel disease13. Together these data demonstrate that a germline loss-of-function mutation in HAVCR2 results in hyperimmune activation and autoinflammatory-like disease. Although SPTCL is considered to be a T cell lymphoma, treatment with the immunosuppressive IL-1β inhibitor anakinra ameliorates disease symptoms, indicating that it is the excessive inflammation resulting from loss of TIM3 that gives rise to malignancy145. These new data provide a strong indication that TIM3 is indeed an inhibitory molecule and that its aberrant folding and lack of surface expression can give rise to autoimmunity and inflammation. Whether the malignancy observed in patients is due to the excessive inflammation or vice versa is unknown at present.

How these mutations result in TIM3 misfolding is also not known. Since the mutations occur in the vicinity of predicted glycosylation sites, the Tyr82Cys and Thr101Ile mutations may both hinder glycosylation of the N-linked or O-linked glycosylation sites. These mutations are predicted to be in a pocket where galectin 9 is thought to bind. Since glycosylation is important for appropriate protein folding149 and stability150, one can postulate that interference with intracellular galectin 9 binding in a yet undetermined mechanism might result in the misfolding of TIM3 and thereby prevent its cell-surface expression. Another possibility is that these mutations affect the TIM3–CEACAM1 interaction, which is important for TIM3 glycosylation, maturation and trafficking to the cell surface13. The onset of SPTCL occurs usually during the fourth decade of life and might result from the accumulation of multiple inflammatory insults throughout life, which, beyond a certain threshold, result in disease. Genetically engineered animals carrying these point mutations, together with modelling and determination of the crystal structure of the aberrant, misfolded protein, will allow a better understanding of how these mutations alter TIM3 biology and its relation to SPTCL.

TIM3 inhibition in the clinic

The vast body of evidence supporting an inhibitory function for TIM3 and the prospect of TIM3 blockade enhancing anti-PD1 therapy in cancer has led to the development of anti-TIM3 blocking antibodies for clinical application. Several clinical trials using anti-TIM3 antibodies in cancer patients are currently under way (Table 2). Most trials include a combination arm that includes anti-PD1 treatment once dose escalation and safety criteria have been met for single-agent treatment with anti-TIM3 blocking antibodies. At least one bispecific antibody targeted at both TIM3 and PD1 is being tested151. The first trial data to be reported were for TSR-022, a humanized anti-TIM3 IgG4 antibody developed by Tesaro152. Single-agent administration of TSR-022 resulted in stable disease in 31 patients and one partial response in a patient with leiomyosarcoma, with no dose-limiting toxicity observed153. This opened the door to the combination treatment with the anti-PD1 antibody TSR-042. Dose escalation of TSR-022 with a fixed dose of TSR-042 showed increased clinical activity (stable disease and partial responses) correlating with increased doses of TSR-022 in 202 patients with non-small-cell lung cancer that was refractory to treatment with anti-PD1 or anti-PDL1 antibodies154. Importantly, the combination was well tolerated, with no dose-limiting toxicity observed. Eli Lilly and Company recently reported that their TIM3 blocking antibody, LY3321367, successfully passed phase I clinical trials155, and Novartis also reported the successful completion of a phase I clinical trial with the anti-TIM3 antibody MGB453 (ref.156). These data are in line with data in preclinical models showing the superior efficacy of TIM3 pathway and PD1 pathway co-blockade over single-agent treatments84,97. Equally important, and similarly to antibodies with antitumour efficacy in preclinical models that were targeted at mouse TIM3, many of the current antibodies tested in humans are directed at the PtdSer binding site (Table 2).

Table 2 Ongoing and planned clinical trials with anti-TIM3 reagents

Conclusion and perspective

TIM3 biology is complex in terms of its non-canonical signalling, broad expression by different types of immune cells and multiple ligands. Despite some debate about whether it acts as an inhibitory receptor or a stimulatory receptor, naturally occurring germline loss-of-function mutations in HAVCR2 argue for the function of TIM3 as an inhibitory receptor with a key role in regulating inflammation. Ample preclinical data support the therapeutic modulation of TIM3 in multiple disease contexts, including autoimmunity, infection, and cancer. Currently, the therapeutic potential of targeting TIM3 is being tested in cancer, where its co-blockade with other checkpoint receptors is being investigated in the clinic, with promising results in patients with anti-PD1-refractory disease. Although much remains to be learned about the circuitry by which TIM3 operates to mediate its effects in different contexts and cell types, its potential as a drug target in diverse human diseases will fuel further investigations into its biology.