The correlation between patient survival and T cell-infiltrate in primary melanoma has been examined in about 20 papers published since the late seventies [27-47]. In Table 1, we have summarized the eight studies that included more than 180 patients.

In the first study, Larsen and Grude reported in 1978 a better overall survival of the patients with an intense lymphocytic infiltration “completely surrounding the part of the melanoma that invaded the normal tissues” [27]. Ten years later, Clark introduced his melanoma classification based on thickness and invasion that became a widely accepted prognostic model [31]. Clark also introduced the notion of brisk T cell infiltrate: “TIL were present throughout the substance of the vertical growth phase or present and infiltrating across the entire base of the vertical growth phase”. Since then, the general view was that a brisk T cell infiltration was a good clinical prognostic factor in melanoma.

In the nineties, the systematic biopsy of the sentinel lymph node (SLN) was introduced, which is the hypothetical first lymph node or group of nodes reached by metastasizing cancer cells from a primary melanoma [48]. Absence of TIL in the primary melanoma is predictive of the presence of metastatic tumor cells in the SLN [43,49]. The detection of metastasis in the SLN is a bad prognosis factor, decreasing 5-year survival from 89% (negative SLN) to 43% (positive SLN). In the latest multivariate analyses, which now include the SLN status, the presence of TIL in the primary melanoma is not an independent predictive factor for survival [43,45].

In melanoma metastases, higher numbers of TIL have been associated with a better survival [34,36,46]. Considering that most of the data were obtained with lymph node metastases, where it is difficult to distinguish between tumor-infiltrating lymphocytes and lymph node lymphocytes invaded by tumor cells, interpretation of these results requires caution.

Beyond melanoma, there is growing evidence that dense TIL infiltrate is associated with a longer survival (Table 2).

In various histological types of tumors, the presence of TIL is unanimously considered as a good clinical prognostic factor: colon adenocarcinoma [50-59], ovarian carcinoma [60-63], endometrial carcinoma [64], non-small-cell lung carcinoma [65-67], lymphoma [68-71], urothelial carcinoma [72-74], esophageal carcinoma [75,76], hepatocellular carcinoma [77,78], oral squamous cell carcinoma [79-81], nasopharyngeal carcinoma [82-84], glioblastoma [85-87], breast carcinoma [88,89], and prostate carcinoma [90].

Colon adenocarcinomas without any sign of metastasis had more T cell infiltrates than tumors with pathological signs of early metastatic invasion such as vascular emboli, lymphatic invasion or perineural invasion [53].

Presence of TIL is not always of good prognosis, e.g., in squamous cell carcinoma [91] and renal cell carcinoma [92,93]. Considering that renal cell carcinoma patients were often treated by immunotherapy, i.e., injections of IL-2 and IFN-α, the fact that tumor infiltration by T cells is of bad prognosis seems contra-intuitive, but considering that TIL numbers usually increase with the tumor grade, the simplest explanation for this correlation is that the grade of the tumors were not taken into account. However, the positive impact of immune cells on tumor growth cannot be excluded.

Noteworthy, TIL populations contain not only CD8 T cells able to kill tumor cells and CD4 T helper cells that secrete cytokines supporting proliferation of other T cells, but also regulatory T cells (Treg) able to inhibit function and proliferation of other T cells, and T cells that secrete, e.g., chemokines attracting macrophages and vasoendothelial growth factor (VEGF) promoting neoangiogenesis. In other words, TIL could also indirectly promote tumor progression [94].

T cell receptor (TCR) genomic loci undergo somatic recombination, plus the addition and removal of bases at recombination junctions, in order to generate the repertoire of structurally diverse TCR necessary for antigen recognition. The different TCR-β subunit chains, which have been estimated at 1 million per individual [95], can be unambiguously identified by their hypervariable “Complement Determining Region 3” sequence, which is the principal site of antigen contact. Upon antigen recognition, a T cell can proliferate and thus, by clonal amplification, the TCR-β transcript expressed by this T cell is multiplied. Evidence of focal TCR-β enrichment has been interpreted as the result of accumulation of T cells recognizing tumor antigens, within the tumor, despite the canonical view that tumor-specific T cells, which have encountered their cognate antigen, proliferate in a secondary lymphoid organ, and are subsequently recruited at the tumor site. Such enrichment was reported for melanoma [96-102], and for other tumors such as non-small-cell lung carcinomas, hepatocellular carcinomas, renal cell carcinomas, gliomas and oral squamous cell carcinomas [103-107]. Analysis of the TCR-γ transcripts was also used as an approach to evidence focal T cell enrichment [101].

Evidence for enrichment in one TCR transcript was followed for a few patients by molecular identification of the antigen recognized by the corresponding CTL, which was isolated either from the tumor or from the blood. This led to the identification of four MAGE-C2 peptides, one MAGE-6 peptide, one gp100 peptide and one peptide encoded by a mutated gene encoding myosin class I [108-112].

Clearly in favor of TIL proliferation in situ is the observation of tertiary lymphoid structures in some tumors. Tertiary lymphoid structures are lymph node-like structures inside the tumor mass composed of dendritic cells, T cells and B cells organized in germinal centers that could be sites of antigen presentation [113,114]. In non-small-cell lung cancer patients, the presence of tertiary lymphoid structure was even correlated with a better survival [114]. More information about the concept of tertiary lymphoid structures can be found in Pages et al. [113].

Study of the TCR diversity indicates accumulation of TCR clonotypes, suggesting the presence of anti-tumor T cells, but often without indication about the antigens recognized by these T cells. Hence the use of HLA-peptide multimers was introduced to identify some of the antigens targeted by TIL. We have only considered experiments with TIL that were not expanded before analysis. Using HLA-A2 multimers, which were folded with four peptides derived from melanoma differentiation proteins, the cumulative frequency of multimer-positive TIL ranged from 0 to >2% of the TIL (4/16 and 6/16 patients, respectively). Interestingly, these frequencies of anti-tumor T cells were close to the frequencies observed for anti-hepatitis CD8 T cells detected in liver biopsies of patients chronically infected with hepatitis C virus. Using a pool of HLA-A2 multimers, which were folded with four peptides derived from non-structural proteins of hepatitis C, frequencies of multimer-positive T cells ranged from 0 to >2% of the TIL (2/6 and 2/6 patients, respectively) [115]. In other reports, similar analyses were performed with TIL expanded for a few days in vitro [116-118]. In these reports, however, frequencies cannot be estimated.

The frequencies of anti-vaccine T cells infiltrating metastases of a melanoma patient, who was vaccinated with MAGE-3.A1 and experienced tumor regression, were analyzed in details. These frequencies were compared with frequencies of TIL directed against tumor antigens other than the vaccine antigens (referred to as “anti-tumor TIL”). No anti-vaccine T cells were detected before vaccination in the blood or in the tumor. After vaccination, the frequency of anti-vaccine T cells was 1/67,000 of CD8 T cells in an invaded lymph node, 6-fold higher than in the blood [110]. After vaccination, anti-tumor TIL were about 10,000 times more frequent than anti-vaccine T cells inside metastases.

Taken together, these results suggest that the anti-vaccine CTL are clearly not the main effectors that kill the bulk of the tumor cells, but that their interaction with the tumor generates conditions enabling at least part of the numerous anti-tumor CTL to participate to the destruction of the tumor cells. Naive T cells appear to be stimulated in the course of this process as new anti-tumor clonotypes arise after vaccination. Indeed, one CTL clonotype, specific for a peptide encoded by a mutated caseinolytic protease, represented up to 7% of the TIL in metastases, whereas the corresponding TCR clonotype was not detected before vaccination [110,119,120]. Similar observations were made in another vaccinated patient [111].

An example of clonotype spreading—a diversification of the T cell clonotypes recognizing the same antigen—was also observed. An anti-MAGE-C2_336-344.A2 T cell clonotype was detected in the blood and in the tumor before and after vaccination, but another anti-MAGE-C2_336-344.A2, which recognized the same peptide/HLA was found at 1/11,000 in the blood only after vaccination and represented up to 9% of the CD8 TIL [110,120]. Clonotype spreading was also observed in a metastatic melanoma patient who experienced a complete clinical response after adoptive transfer of an anti-Melan-A.A2 CTL clone [121]. This clinical response correlated with an increased frequency of anti-Melan-A.A2 CTL in the blood, but the TCR sequence of this blood clonotype was different from the TCR sequence of the injected T cell clone.

Expression of activation markers CD137 or CD25 on TIL was used as markers of TIL that have recently encountered their antigen. Expression of these markers in the peritumoral region was associated with a better survival [39,122]. Considering that regulatory T cells also express these receptors, the significance of the expression of these two markers is questionable.

To detect TIL proliferating at the tumor site, the presence of nuclear protein Ki-67 was used as a marker of cell division [123]. Higher numbers of TIL expressing Ki-67 were associated with a better survival in renal cell carcinoma and in hepatocellular carcinoma [78,93].

The TUNEL method labels fragmented DNA, which is a characteristic of apoptotic cells. In seminoma, 2% of tumor cells were considered apoptotic by this method and half of them were in contact with at least one CD3 cell [124-126]. Co-localization of TIL with apoptotic cells was also examined in four gliomas. Two-thirds of the tumor cells that were in close contact with TIL, i.e., which form immunologic synapses, were in apoptosis whereas only 25% of the tumor cells that were not in contact with TIL displayed signs of apoptosis [127].

The consensus that seems to emerge from studies on colon adenocarcinoma, hepatocellular carcinoma, melanoma and lymphoma, is that the presence of granzyme B⁺ CD8 TIL is a favorable prognosis factor [44,58,70,77]. In colon carcinoma, the frequency of granzyme B⁺ cells among the CD8 TIL can reach 30% [50]. This frequency is comparable to the 10% of granzyme B⁺ cells among the CD8 T cells in the liver of patients with hepatitis B, but it can reach 40% in hepatitis B patients that have an autoimmune disease characterized by T cell hyperreactivity [128-130].

Upon activation, CTL express CD107a at their surface for a few hours. This lysosomal-associated membrane protein (LAMP-1 or CD107a) has been described as a marker of degranulation of CD8 T cell upon stimulation. This marker was found at the surface of 1% to 9% of the TIL in colon adenocarcinoma metastases and seminoma [131-133]. Considering that no comparison was made between tumor tissues and non-cancerous tissues with an ongoing efficient cytolytic T cell response, one may wonder if this level of CD107a expression has to be considered as a sign of good or poor lytic capacity.

The optimal cytokine pattern for an efficient anti-tumor immune response may differ in various tumor types. Cytokines have not been examined directly in the tumors but the subtypes of infiltrating cells have been studied by detection of specific transcription factors. The transcription factors that specify lineage commitment in T helper cells have started to be deciphered and the generally accepted model considers T-bet and GATA-3 as the master regulators of the so-called Th1 and Th2 differentiation, respectively, with c-Maf as the downstream factor that selectively controls IL-4 gene transcription [134,135]. TIL were analyzed for GATA-3 and T-bet expression by immunohistochemistry in 69 pancreatic tumor tissues. The median value of GATA-3/T-bet ratio was 5.2—this should promote a Th2 response. A 36-month survival analysis indicated a shorter survival by 15 months for patients with a GATA-3/T-bet ratio above the median value [136]. On the contrary, the 10-year survival of patients with Hodgkin lymphoma was significantly longer if the CD4 TIL infiltrate contains a higher frequency of c-Maf cells, which should produce IL-4 and be considered as Th2 [71].

TNF-α can be secreted by several types of tumor-infiltrating cells, e.g., fibroblasts, macrophages, NK cells, T cells, and could exert cytotoxic functions on tumor cells (reviewed in Bazzoni et al. [137]). TNF-α mRNA or protein was detected in non-small-cell lung cancers, hepatocellular carcinomas, colon adenocarcinomas or in the serum of gastric cancer patients, and was associated with a better survival [78, 138-140]. However, in terminal cancer patients, TNF-α is also implicated in cachexia, a loss of adipose and skeletal muscle mass, and accordingly elevated TNF-α in the serum of lymphoma and breast cancer patients was correlated with poor survival [141-145].

Taken together, these results give the impression that TIL are functional at tumor sites. This is in contrast with the results of the T cell function assays described in the next section. One possible explanation is that most anti-tumor TIL are anergic and that only few functional TIL are detected by immunohistochemistry. Another explanation is that the functional TIL detected by immunohistochemistry have been activated outside the tumor, e.g., by antigens derived from Epstein-Barr virus or cytomegalovirus, and have been recently attracted into tumors by chemokines.

T cells isolated from chronic lymphocytic leukemia patients showed impaired immunological synapse formation with leukemic B cells associated with defective actin polymerization [150]. Defective synapse formation between T cells and tumor B cells was also observed with cells isolated from patients with acute myeloid leukemia or follicular lymphoma [151,152].

Anti-melan-A^MART-1 specific CD8 T cells were isolated with HLA-peptide multimers from melanoma or blood samples and short-term stimulated with peptide-pulsed cells. Eight percent of these TIL contained IFN-γ, whereas 44% of the multimer⁺ T cells isolated from blood did contain IFN-γ [148]. Anti-cytomegalovirus CD8 TIL were also isolated with HLA-peptide multimers, from the same tumors. Thirty percent of them contained IFN-γ upon stimulation with peptide-pulsed cells, whereas 38% of anti-cytomegalovirus blood CD8 T cells contained IFN-γ. In another study with one melanoma patient, polyclonal CD8 T cells isolated from ascites did not secrete IFN-γ upon short-term stimulation with tumor cells but were able to secrete IFN-γ upon stimulation with the cells pulsed with an EBV peptide [153].

How to explain the observations that anti-virus T cells were functional, whereas the anti-tumor cells were not? If the tumor environment is suppressive, why were the anti-virus T cells not sensitive to suppression? One explanation is that anti-virus T cells were residing in the tumor only for a short time. Another explanation is that anti-tumor T cells have been repetitively stimulated by tumor antigens and are therefore more sensitive to immunosuppression.

We have tested CD8 and CD4 TIL, which were isolated from melanoma and carcinomatous ascites of the ovary, colon, and pancreas. TIL were stimulated with beads coated with anti-CD3 and anti-CD28 antibodies, which is a strong stimulus. Compared to donor blood T cells, CD8 and CD4 TIL secreted low levels of IFN-γ and TNF-α [154].

CD8 TIL isolated from renal cell carcinoma, malignant pleural effusions, ovarian and pancreatic carcinoma ascites failed to lyse cells coated with anti-CD3 antibodies in an assay known as a redirected killing assay, whereas CD8 blood T cells were cytolytic [154,155]. Presence of perforin was used in several studies as a marker for the cytolytic potential of TIL [148,156], but one could also argue that TIL that have very recently lysed tumor cells have lost most of their perforin.

Taken together, the few results reported above give the impression that most TIL are dysfunctional in ex vivo functional assays whereas the reports on signs of TIL activity in situ gave the opposite impression. Because the ex vivo functional assays are rather straightforward, we are tempted to conclude that most TIL are actually dysfunctional. We hope that additional experiments from different groups will help strengthen this conclusion. There is also a need for functional assays with control T cells isolated from non-cancerous tissues such as acute inflammatory sites.

Several molecules expressed by TIL could down-modulate their activation upon antigen recognition. A more complete set of such molecules is described in Table 3, together with potential drugs tested in vitro or in clinical trials.

Numerous immune cell types have some level of immunosuppressive activity. Among them are Treg, myeloid-derived suppressor cells, mesenchymal stem cells and tumor-associated macrophages. Only the presence of Treg and macrophages has been associated with survival prognosis and will be discussed below.

We describe below soluble molecules that can modulate TIL activation and function. A more complete set of molecules is shown in Table 4, together with potential drugs tested in clinical trials or in vitro.

Galectins belong to a family of lectins, i.e., sugar- or glycan-binding proteins. Galectins share the ability to bind β-galactosides and contain homologous carbohydrate recognition domains (CRD). Their binding affinity for simple β-galactosides, such as disaccharides or trisaccharides, is relatively weak. However, galectin binding to natural glycoconjugate ligands expressed on cell surfaces or in the extracellular matrix is usually of higher affinity, in the submicromolar range. We will discuss below the role in the tumor microenvironment of the sole galectin-1, -3, and -9. Further reading on galectins can be found in reviews by Rabinovich [324] and Cummings [325].

Galectins have been classified in three structural groups: (a) the prototype galectin-1 with a single CRD, which can form homodimers via non-covalent interactions; (b) the chimeric galectin-3 with a single CRD and a large amino-terminal domain, which contributes to self-aggregation into pentamers; and (c) the tandem-repeat galectins with two different CRD and a 5 to 70 amino-acid linker, e.g., galectin-9.

Galectins are found in the cytosol and in the nucleus where they interact with proteins to regulate growth, cycle progression and apoptosis [326,327]. Galectins lack a signal peptide but can nevertheless be secreted by an unknown mechanism. Extracellularly, the different roles of extracellular galectins are related to their ability to bind sugar moieties on surface O- and N-glycoproteins. The interactions of galectins and glycoproteins are complex and cannot be interpreted in terms of interactions between a ligand and a receptor. Several factors contribute to high-affinity binding to natural glycoconjugate ligands: the natural multivalency, the oligomeric state of the galectins and the multivalency of their natural glycoconjugate ligands [325].

Galectins participate in both glycoprotein clustering and glycoprotein/galectin lattices, and could therefore modify receptor signaling and increase the surface half-life of some glycoproteins [328]. In tumors, galectin-1, -3, -9 are often secreted by tumor cells and monocyte-derived cells, but can also be secreted by activated B or T cells [329-335]. The inhibition of T cell proliferation by CD4⁺CD25^high cells, possibly Treg cells, was reported to be reversed by an anti-galectin-1 antibody [336]. T cell proliferation was also inhibited in the presence of mesenchymal stromal cells, which secreted high levels of galectin-1 and galectin-3. T cell proliferation was restored by adding thiodigalactoside, a ligand competitor of both galectins, or by inhibiting galectin-3 expression in mesenchymal stromal cells by RNA interference [337-340].

Sera of melanoma and colon adenocarcinoma patients were reported to contain more galectin-3 than sera of healthy volunteers [341,342]. Melanoma patients with more than 10 ng/mL were reported to have a median survival of less than 5 months, whereas the median survival of patients with less than 8 ng/mL was 60 months [343]. No correlation was found, however, between tumor regression in melanoma patients vaccinated with tumor antigens and concentration of galectin-3 [344]. In head and neck squamous cell carcinoma patients, elevated galectin-3 and galectin-1 in serum were also correlated with a worse survival [345]. Is the presence of galectins a sign of immune suppression and better conditions for metastasis or the mere sign of an advanced disease and high tumor burden?

Blood T cells, T cell lines or Jurkat T cells were stimulated in the presence of galectin-1, -3, or -9, usually added in the culture at μM concentrations. Treated cells either entered in apoptosis [346-349], proliferated less or secreted more IL-10 [350]. It was also reported that a Hodgkin lymphoma cell line, which expressed galectin-1, had a negative effect on proliferation of CD4 T cells [351,352] and that this effect diminished when the lymphoma cells knocked down for galectin-1 were added to the co-culture. T cell apoptosis mediated by galectins could be a physiological mechanism taking place in the thymus during early development of T cells, as thymic epithelial cells can secrete galectin-1 and in vitro exposure to galectin-1 induced apoptosis of subsets of CD4^lo CD8^lo thymocytes [353].

Galectin-1, -3 and -9 added at μM concentrations also induce apoptosis of prostate carcinoma, melanoma, B-lymphoma and leukemia cells, provided they harbor the adequate glycosylation pattern [346,354-357]. We did not observe apoptosis of blood T cells, T cell clones or tumor cell lines when galectin-3 was added at 10 nM, which is the highest concentration of galectin-3 that we measured in carcinoma ascites. It is nevertheless possible that galectins could accumulate in solid tumors in confined microenvironments and reach concentrations in the μM range.

The apoptosis induced by galectin-1 and galectin-3 seems to be mediated by their interactions with highly glycosylated surface molecules, such as CD43 and CD45, as these molecules co-purified with these galectins and were associated in patches at the T cell surface [329,358-360]. Modification of the glycosylation pattern of T cells rendered them resistant to galectin-1-induced apoptosis [361-363]. The apoptosis induced by galectin-9 requires its binding to glycan moieties on Tim-3, a negative costimulatory receptor expressed on some activated T cells [189,298].

We have reported that treatment of TIL with galectin ligand LacNAc boosted their secretion of cytokines upon stimulation. Why do galectin-3 ligands improve TIL function? Our working hypothesis is that TIL have been stimulated by antigen recently, and that the resulting activation of T cells could modify the expression of enzymes of the N-glycosylation pathway and change the structure of N-glycans exposed at the cell surface, as shown for murine T cells [364]. We surmise that the recently activated TIL, compared to resting T cells, harbor a set of glycans that are either more numerous or better ligands for galectin-3. Galectin-3 is abundant in many solid tumors and carcinomatous ascites, and can thus bind to surface glycoproteins of TIL and form lattices that would thereby reduce mobility of surface receptors implicated in T cell activation. This could explain the impaired function of TIL. The release of galectin-3 by soluble competitor ligands would restore mobility of surface receptors and boost IFN-γ secretion by TIL. We recently strengthened this hypothesis by showing that CD8 TIL treated with an anti-galectin-3 antibody had an increased IFN-γ secretion [154].

Ex vivo treatment of TIL with galectin-3 ligand LacNAC boosted their functions [323]. We have thus searched for galectin-3 ligands available for clinical use. GCS-100^®, a modified citrus pectin with cytotoxic properties on tumors, was administered by i.v. injection to 24 multiple myeloma patients [365-368]. Side effects were minor. In vitro, GCS-100 detached galectin-3 from TIL and boosted cytotoxicity and secretion of different cytokines [154]. Pectasol^®, another modified citrus pectin, was given orally to 10 prostate cancer patients [369]. Seven of them experienced an increased doubling time of blood PSA, which suggests a decreased tumor growth.

DAVANAT^®, a galactomannan derived from guar gum, has been shown to bind to galectin-1 [277], and to increase the anti-tumor activity of chemotherapy drug 5-fluorouracil in mice. It was injected together with 5-fluorouracil in 25 patients with solid tumors without major side effects [276]. Short ex vivo treatment of TIL with DAVANAT boosted secretion of cytokines and lytic activity (our own unpublished data). We intend to launch a clinical trial with metastatic melanoma patients where MAGE3.A1 and Na17.A2 peptides, which correspond to tumor-specific antigens expressed by various types of tumor, will be administered together with systemic and local injections of DAVANAT.