Liver sinusoidal endothelial cells contribute to CD8 T cell tolerance toward circulating carcinoembryonic antigen in mice


  • Potential conflict of interest: Nothing to report.

  • Supported by a grant from the German Cancer Aid (to P. K. and L. D.).


Immunity against cancer is impeded by local mechanisms promoting development of tumor-specific T cell tolerance, such as regulatory T cells, myeloid-derived suppressor cells, or immunosuppressive factors in the tumor microenvironment. The release of soluble antigens, such as carcinoembryonic antigen (CEA) from colorectal carcinoma (CRC) cells, has been investigated for diagnostic purposes, but not for its immunological consequences. Here, we address the question of whether soluble CEA influences tumor-specific immunity. Mice were injected with soluble CEA protein, and CEA-specific CD8 T cells were analyzed for their phenotype and functionality by means of restimulation ex vivo or antitumor efficacy in vivo. We furthermore characterized the CD8 T cell population in peripheral blood mononuclear cell (PBMCs) from healthy donors and colorectal carcinoma patients. In mice, circulating CEA was preferentially taken up in a mannose receptor–dependent manner and cross-presented by liver sinusoidal endothelial cells, but not dendritic cells, to CD8 T cells. Such systemically circulating CEA promoted tolerization of CEA-specific CD8 T cells in the endogenous T cell repertoire through the coinhibitory molecule B7H1. These CD8 T cells were not deleted but were rendered nonresponsive to antigen-specific stimulation and failed to control growth of CEA-expressing tumor cells. These nonresponsive CD8 T cells were phenotypically similar to central memory T cells being CD44highCD62LhighCD25neg. We found T cells with a similar phenotype in PBMCs of healthy donors and at increased frequency also in patients with colorectal carcinoma. Conclusion: Our results provide evidence for the existence of an unrecognized tumor immune escape involving cross-presentation of systemically circulating tumor antigens that may influence immunotherapy of cancer. (HEPATOLOGY 2012;56:1924–1933)

The interaction of the immune system with cancer cells is complex and may affect both cancer development and progression. The concept of cancer immune surveillance has been refined and termed “cancer immunoediting.”1 This conceptual framework of the role of the immune system in cancer includes three distinct phases: an elimination phase, in which tumor cells are killed; an equilibrium phase, in which the immune system controls growth of cancer cells; and finally, the escape phase.1

Key to the understanding of immune control of cancer was the identification of tumor antigens that allow development of adaptive T cell immunity, which is essential for the equilibrium phase of cancer immunoediting. Presently, it is believed that the most critical step in cancer progression is the escape phase, where numerous mechanisms may cooperatively impede immune surveillance of cancer cells. These mechanisms can be divided into direct immune escape of cancer cells becoming invisible to the immune system, e.g. by mutational loss of major histocompatibility complex class I molecules and tumor antigens,2 and into mechanisms that establish an immunosuppressive state within the microenvironment of the tumor. Such mechanisms include induction of regulatory mediators such as transforming growth factor β, indoleamine 2,3-dioxygenase, or galectin by tumor cells3–5 and also recruitment of immunosuppressive cells such as regulatory T cells or myeloid-derived suppressor cells.6, 7

So far, the escape mechanisms of cancer cells have been attributed to local modulation of immune responses in the direct vicinity of cancer cells. The release of tumor antigens from cancer cells into the bloodstream has been used for diagnostic purposes, such as carcinoembryonic antigen (CEA) for colorectal, gastric, and ovarian carcinoma.8–10 Systemic distribution of such antigens in the absence of a strong innate immune stimulus11 can lead to the elimination of antigen-specific CD8 T cells by cross-tolerance.12 However, the relevance of circulating tumor antigens for the modulation of tumor-specific immunity has not been addressed. Here, we demonstrate that circulating CEA is preferentially taken up by liver sinusoidal endothelial cells (LSECs). CEA is cross-presented by LSECs to CD8 T cells and circulating CEA in vivo induces a state of nonresponsiveness in antigen-specific CD8 T cells that in turn promotes immune escape of CEA-expressing tumor cells.


APC, antigen-presenting cell; CEA, carcinoembryonic antigen; CRC, colorectal carcinoma; CTL, cytotoxic T lymphocyte; DC, dendritic cell; IFNγ, interferonγ; IL, interleukin; LSEC, liver sinusoidal endothelial cell; MR, mannose receptor; OVA, ovalbumin; PBMC, peripheral blood mononuclear cell; PBS, phosphate-buffered saline.

Materials and Methods

Patients and Healthy Donors.

Blood samples were collected from patients with colorectal carcinoma (CRC) (n = 15) prior to therapy start or healthy volunteers (n = 14). The study was approved by the ethics committee of the University of Bonn, and written consent was obtained from all patients before blood sampling. Peripheral blood mononuclear cells (PBMCs) were isolated from freshly obtained blood by way of density gradient centrifugation.


C57BL/6 mice were obtained from Elevage Janvier (Le Genest-Saint-Isle, France). B7H1−/− (a gift from L. Chen), MR−/− (a gift from S. Burgdorf), RAG2−/−, TLR4−/−, and HLA-A*0201×C57BL/B6 transgenic mice were bred under specific pathogen-free conditions in the central animal facility in Bonn according to the Federation of European Laboratory Animal Science Association and were used for experiments at 6-10 weeks of age. All mouse experiments were approved by the Animal Care Commission of North Rhine–Westphalia.

Ex Vivo Cross-presentation Assay.

HLA-A*0201×C57BL/6 and C57BL/6 or MR−/− mice were injected intravenously with either 100 μg CEA or ovalbumin (OVA) protein or left untreated. After 15 minutes, CD11b+ or CD11c+ cells were isolated from the spleen or liver nonparenchymal cells using CD11c or CD11b Macs beads, or LSECs were isolated from liver nonparenchymal cells with CD146 Macs beads according to the manufacturer's protocol using an Auto-MACS. A total of 1 × 105 CD11b+, CD11c+, or CD146+ antigen-presenting cells (APCs) were incubated with 1 × 105 purified CD8 T cells from pGT64-CEA–vaccinated HLA-A*0201×C57BL/6 mice and cocultured in the presence of brefeldin A and monensin for 8 hours. As a positive control APC from nontreated HLA-A*0201×C57BL/6 mice were loaded in vitro with 1.5 μg/mL of the CEA-derived HLA-A*0201 binding peptide YLSGANLNL. APCs from C57BL/6 and MR−/− mice were loaded with the H-2Db binding CEA-derived peptide CGIQNSVSA. After 8 hours, T cells were stained intracellularly for IFNγ and analyzed by way of flow cytometry.

Statistical Analysis.

The Student t test was used to determine statistical significance of the results. Data are presented as the mean ± SEM, and P ≤ 0.05 was considered significant.

Additional Materials and Methods are presented in the Supporting Information.


LSECs Take Up the Tumor Antigen CEA in a Mannose Receptor–Dependent Fashion.

We have shown that LSECs efficiently scavenge the soluble protein ovalbumin from the circulation and cross-present it to CD8 T cells.13 Because CEA, when released from tumor cells, gains access to the systemic circulation, we investigated whether LSECs and other hepatic and splenic APCs were capable of taking up CEA from the circulation. To this end, C57BL/6 mice were injected intravenously with fluorescently labeled CEA. Within 15 minutes, CD146+ LSECs internalized large amounts of CEA. No significant uptake of CEA by hepatic or splenic CD11c+ dendritic cells (DCs) was observed (Fig. 1A and C, Suppl. Fig. 1A). Compared with LSECs, CEA uptake by hepatic CD11b+ or F4/80+ cells occurred to a lesser extent (Fig. 1A). Because CEA is heavily glycosylated, we analyzed the involvement of the mannose receptor (MR), which participates in uptake of glycosylated antigens, for endocytosis of CEA by LSECs. Intravenous injection of CEA−Alexa647 into wild-type C57BL/6 or MR−/− mice showed that CEA internalization into LSECs mainly depended on the MR (Fig. 1B,C), whereas uptake into DCs or Kupffer cells (Fig. 1C) was not altered in MR−/− mice. This is different from the uptake of OVA by LSECs, which was not influenced by the absence of the MR, whereas efficient ovalbumin uptake by hepatic macrophages and DCs requires the MR (Fig. 1D). Furthermore, we detected uptake of neither CEA nor OVA into splenic CD11c+ or CD11b+ cells (Supporting Fig. 1). Taken together, these data show that LSECs are proficient in uptake of circulating CEA via the MR.

Figure 1.

Uptake of CEA by LSECs is MR-dependent. Alexa647-labeled CEA (A-C), PBS (B), or OVA (D) protein was injected intravenously into C57BL/6 (A-D) or MR−/− (B-D) mice. Fifteen minutes after injection, liver nonparenchymal cells were isolated and stained for CD146, CD11b, CD11c, or F4/80 and analyzed using flow cytometry. (A) Dot plots of viable cells. Numbers indicate percentages of Alexa647+ cells within the gated cell population. (B) Histograms show CD146± C57BL/6 LSECs (black line) or MR−/− LSECs (dotted line) from CEAAlexa647-injected mice or LSECs from PBS-injected mice (filled gray line). (C, D) Mean fluorescence intensity of liver CD146+, CD11c+, CD11b+ or F4/80+ cells from C57BL/6 or MR−/− mice previously injected with CEAAlexa647 (C) or OVAAlexa647 (D). The cumulative results of three independent experiments are shown. *P ≤ 0.05.

CEA Is Efficiently Cross-presented by LSECs In Vivo to CD8 T Cells.

Because LSECs are capable of cross-presenting soluble antigens to antigen-specific naïve CD8 T cells in vivo,13–15 we analyzed whether the CEA protein is cross-presented to CD8 T cells after uptake by LSECs. Because there are no CEA-specific TCR transgenic animals available, we generated CEA-specific cytotoxic T lymphocytes (CTLs) by way of DNA vaccination of C57BL/6 mice, which were then incubated in vitro with CEA and OVA protein or CEA and OVA peptide–pulsed LSECs. Both CEA protein and CEA peptide–loaded LSECs induced IFNγ and interleukin (IL)-2 release by the CEA-specific CTLs (Fig. 2A,B). No IFNγ or IL-2 was produced when an irrelevant antigen (OVA or SIINFEKL peptide) was presented by LSECs. We excluded the possibility that our CEA protein preparation altered LSEC function due to contamination with lipopolysaccharide or other Toll-like receptor ligands, because LSECs did not produce any IL-6 after incubation with CEA protein (Supporting Fig. 2).

Figure 2.

LSECs cross-present CEA protein to CTLs in vitro and in vivo. CEA571-579 peptide–specific CTLs were generated as described. C57BL/6 LSECs were incubated with 100 μg/mL CEA protein, 100 μg/mL OVA protein, 1.5 μg/mL CEA571-579 peptide (CGIQNSVSA), or 1.5 μg/mL OVA257-264 peptide (SIINFEKL) for 2 hours. After washing, 1 × 106 CEA-specific CTLs were added, and after 24 hours IFNγ (A) and IL-2 (B) secretion into the supernatant was measured via enzyme-linked immunosorbent assay. The cumulative results of three independent experiments are shown. (C) C57BL/6 or MR−/− mice were injected intravenously with 100 μg CEA protein or were left untreated. After 15 minutes, LSECs were isolated and incubated with CEA-specific CTLs. LSECs from noninjected C57BL/6 or MR−/− mice were loaded with 1.5 μg/mL CEA571-579 peptide (CGIQNSVSA). The amount of IFNγ-producing CD8 T cells was determined by intracellular staining. (D) HLA-A*0201×C57BL/6 mice were injected intravenously with 100 μg CEA or OVA protein or PBS. After 15 minutes, various APC populations were isolated. APCs from PBS-injected mice were loaded with CEA peptide as a positive control. CD11b+, CD11c+, or CD146+ APCs were then cocultured with CEA-specific CTLs from HLA-A*0201×C57BL/6 mice in the presence of brefeldin A/monensin and stained for intracellular IFNγ 8 hours later. The percentage of IFNγ-producing CEA-specific CD8 T cells within the total CD8 population is shown. The results are representative of two independent experiments. *P ≤ 0.05; **P ≤ 0.01.

We next investigated whether cross-presentation of CEA was dependent on the MR. Indeed, only LSECs that were isolated from C57BL/6 but not MR−/− mice that were intravenously injected with CEA protein induced IFNγ production in CEA-specific CTLs in vitro (Fig. 2C). MR−/− LSECs were in principle capable of stimulating CD8 T cells, because CEA peptide–loaded MR−/− LSECs induced amounts of IFNγ+ CD8 T cells similar to that of CEA peptide–loaded C57BL/6 LSECs. More importantly, we assessed whether other APC populations from the liver and spleen were capable of cross-presenting CEA protein when administered in vivo. To this end, we isolated various APC populations from CEA of OVA protein or phosphate-buffered saline (PBS)-injected HLA-A*0201×C57BL/6 transgenic mice, and cocultured them ex vivo together with CEA-specific CTLs generated by DNA vaccination in HLA-A*0201×C57BL/6 transgenic mice. All APCs were capable of inducing IFNγ production by these CTLs when they were loaded in vitro with the CEA-derived peptide YLSGANLNL, indicating that the cells were viable after isolation (data not shown). However, only CD146+ LSECs from CEA-injected mice also induced IFNγ production in CEA-specific CTLs directly ex vivo (Fig. 2D). In contrast, both splenic and hepatic CD11c+ and CD11b+ APCs did not cross-present circulating CEA ex vivo, indicating that CEA is not only preferentially taken up but also preferentially cross-presented by LSECs in the liver and not by DCs or macrophages in the liver or spleen.

Cross-Presentation of Systemically Circulating Soluble CEA Leads to the Development of Tolerant CEA-Specific CD8 T Cells in HLA-A*0201×C57BL/6 Transgenic Mice.

Next, we investigated whether circulating CEA would lead to priming of naïve CD8 T cells. To this end, CEA-specific CD8 T cells from the endogenous naïve CD8 T cell pool in HLA-A*0201×C57BL/6 transgenic mice were analyzed 2 weeks after injection of CEA protein or after pGT64-CEA plasmid DNA vaccination. After DNA vaccination, we detected between 7% and 11% of CEA-specific CTLs among total CD8 T cells in the liver and spleen by staining with HLA-A*0201YLSGANLNL dextramers (Fig. 3A). Importantly, after intravenous injection of purified CEA protein, we detected CEA-specific CD8 T cells in the liver and spleen that accounted for approximately 4% of total CD8 T cells (Fig. 3A). We have reported that stimulation of naïve CD8 T cells by cross-presenting LSECs does not cause clonal deletion but leads to the induction of a population of CD8 T cells with a surface marker phenotype clearly distinct from both naïve and DC-activated CD8 T cells (Fig. 3B; von Oppen et al.13 and Diehl et al.15). Such LSEC-stimulated CD8 T cells are CD44high, CD62Lhigh, and CD25neg. Interestingly, the expression pattern of the CEA dextramer–positive T cells in the CEA protein–treated mice matched that of LSEC-tolerized T cells being CD44high, CD62Lhigh, and CD25neg, whereas the CEA tetramer–positive T cells generated by DNA vaccination showed an activated phenotype being CD44high, CD62Llow, and CD25pos (Fig. 3C). Thus, systemically circulating CEA protein induces CEA-specific CD8 T cells with a surface marker expression profile similar to that of CD8 T cells primed by tolerogenic cross-presenting LSECs. Furthermore, CEA dextramer–positive CD8 T cells from mice that received CEA protein intravenously did not produce IFNγ upon restimulation, whereas CD8 T cells from pGT64-CEA–vaccinated animals did (Fig. 3D,E). Taken together, these results suggest that cross-presentation of systemically circulating CEA protein in vivo does not lead to clonal deletion but induces phenotypically and functionally tolerized CD8 T cells.

Figure 3.

CEA-specific CD8 T cells in HLA-A*0201×C57BL/6 mice after systemic CEA administration are phenotypically and functionally tolerant. (A, C, D, E) HLA-A*0201×C57BL/6 mice were injected intravenously with CEA protein, vaccinated with pGT64 CEA plasmid, or were left untreated. Splenic and hepatic lymphocytes were isolated, stained with CD8α, a CEA-peptide YLSGANLNL-loaded HLA-A*0201 dextramer, or a control peptide-loaded dextramer, and analyzed by flow cytometry. The percentage of dextramerpos cells in total CD8 T cells is shown. (B) Naïve OT-1 T cells were transferred into C57BL/6 mice that subsequently received OVA protein intraperitoneally (tolerized OT-1 T cells) or were injected subcutaneously with 200 μg OVA in incomplete freund's adjuvant (IFA) (activated OT-1 T cells). Splenic OT-1 T cells were analyzed via flow cytometry. (C) Histograms show the expression levels of CD44, CD62L, and CD25 on CD8α pos and YLSGANLNL-dextramerpos T cells from the livers and spleens of pGT64-CEA, CEA-injected, or control HLA-A*0201×C57BL/6 transgenic mice. (D) Splenocytes were restimulated with PMA/ionomycin, stained with CD8α, YLSGANLNL-dextramer, and intracellularly with IFNγ or control antibody. Dot plots are gated on CD8pos cells, and numbers are the percentage of cells in each quadrant. (E) Enumeration of CEA-specific IFNγ-producing cells in the liver and spleen of plasmid-vaccinated or CEA protein–treated animals. Data are representative of three independent experiments. *P ≤ 0.05; **P ≤ 0.01.

CD8 T Cells Fail to Control Growth of Tumors Expressing CEA in the Presence of Systemically Circulating CEA.

To further characterize the functional capacity of CEA-specific CD8 T cells generated as a consequence of circulating CEA, we investigated the ability of these cells to control the growth of a CEA expression tumor in vivo. To this end, Rag2−/− mice received purified splenic CD8 T cells isolated either from C57BL/6 mice that were injected every second day with 2 mg CEA intravenously for 2 weeks or that were vaccinated with pGT64-CEA DNA or from untreated mice. Three days after CD8 T cell transfer, Rag2−/− mice were injected subcutaneously with the murine tumor cell lines MC38 and MC38-CEA on different flanks. The mice that received CD8 T cells from DNA-vaccinated animals controlled MC38-CEA growth compared with mice that received naïve CD8 T cells (Fig. 4A,D). However, Rag2−/− mice that received CD8 T cells from CEA protein–treated mice failed to control MC38-CEA tumor growth; instead, tumors grew even faster than in Rag2−/− mice that received naïve CD8 T cells (Fig. 4A,D). This indicates that CEA-specific CD8 T cells generated after circulating CEA were tolerized and did not have a protective function in antitumor defense. The tolerization of T cells was antigen-specific, because the MC38 tumors not expressing CEA grew equally fast in all three groups (Supporting Fig. 3), suggesting that there is no bystander effect of tolerized T cells on T cells with different antigen specificity. Because we found the uptake of CEA to be MR-dependent (Fig. 1B), we further tested whether the lack of uptake of soluble CEA by LSECs in MR−/− would result in the lack of tolerization of CEA-specific CD8 T cells in the naïve endogenous T cell repertoire. Indeed, as shown in Fig. 4C, tumor growth in Rag2−/− mice receiving CD8 T cells from CEA-injected MR−/− mice did not show accelerated growth, as was the case with CD8 T cells from C57BL/6 mice (Fig. 4A), but instead had similar kinetics as CD8 T cells from naïve C57BL/6 (Fig. 4A) and MR−/− (Fig. 4C) mice.

Figure 4.

CD8 T cells from mice that received CEA protein systemically are incapable of controlling CEA-expressing tumor cells. C57BL/6 mice and B7H1- and MR-deficient mice received 2 mg of CEA protein intravenously every second day, were immunized with pGT64 CEA plasmid, or were left untreated. A total of 5 × 106 purified splenic CD8 T cells from C57BL/6 (A, D), MR−/− (B, D), or B7H1−/− (C, D) were adoptively transferred into Rag2-deficient mice. Three days later, 5 × 106 MC-38-CEA or MC-38 cells were injected into the right flank or left flank, respectively. Graphs show tumor growth of MC38-CEA in Rag2−/− mice that received CD8 T cells from C57BL/6 (A, D), MR−/− (B, D), or B7H1−/− mice (C, D). D: Specific MC38-CEA tumor volumes at day 15 of mice receiving CD8 T cells from C57BL/6, MR−/− and B7H1−/− mice. *P ≤ 0.05; ***P ≤ 0.001; ns, not significant.

LSEC-induced tolerization of OVA-specific CD8 T cells is B7H1-dependent in vitro.15 To investigate whether tolerance induction in endogenous CEA-specific CD8 T cells in vivo in response to circulating CEA-protein is also B7H1-dependent, we adoptively transferred CD8 T cells (which were generated in B7H1-deficient mice after CEA protein injection or DNA vaccination as described, into Rag2−/−) that were subsequently challenged with MC38 or MC38-CEA tumor cells (Fig. 4C,D). The CEA-specific CD8 T cells generated in B7H1−/− mice after systemic CEA were better at controlling MC38-CEA tumor growth (Fig. 4C) than T cells from untreated B7H1−/− mice and T cells from CEA protein–treated C57BL/6 mice. Taken together, these data show that systemically circulating CEA protein results in a B7H1-mediated tolerization of antigen-specific CD8 T cells that fail to eliminate CEA-expressing tumors.

CD8 T Cells with Phenotypic and Functional Characteristics of LSEC-Tolerized CD8 T Cells Are Detected in Healthy Individuals and Are Increased in Patients with Colorectal Carcinoma.

Given the ability of LSECs to cross-present circulating antigens in mice, we reasoned that LSECs in humans might similarly function as APCs that tolerize CD8 T cells. This should lead to a population of CD8 T cells with similar phenotypic and functional characteristics. Indeed, we identified a sizeable population of CD8pos, CD45ROpos antigen-experienced T cells in PBMCs from healthy individuals that were CD25neg and CD62Lhigh (Fig. 5A). These cells did not differ from CD25pos T cells with respect to expression of Foxp3, IL-10, granzyme B, perforin, and PD-1, indicating that no regulatory or exhausted T cells were present in this cell population (Fig. 5B). Murine LSEC-tolerized CD25neg CD8 T cells do not respond to secondary stimulation through the T cell receptor.14, 15 To investigate whether the human CD25neg CD8 T cells were responsive to restimulation, we sorted CD62LhighCD25pos or CD62LhighCD25neg T cells from CD8pos, CD45ROpos PBMCs and restimulated the T cells with either anti-CD3/CD28 beads or PMA/ionomycin or left them untreated. CD62LhighCD25pos CD8 T cells produced IFNγ upon stimulation (Fig. 5C). However, CD62LhighCD25neg T cells did not secrete IFNγ (Fig. 5C). The CD25neg population neither proliferated nor degranulated (Supporting Fig. 4), whereas their CD25pos counterpart did, suggesting that both phenotypically and functionally CD45ROposCD62LhighCD25neg CD8 T cells resemble tolerized murine CD44highCD62LhighCD25neg CD8 T cells. Importantly, when we analyzed CRC patients (Supporting Table 1) for the frequency of this particular population of CD8 T cells in peripheral blood, we found that it was significantly increased (Fig. 5D). From these 15 patients, who did not receive prior treatment, seven were HLA-A2–positive by way of flow cytometry, and one of these patients had an elevated serum CEA value (>2,400 mg/L). In PBMCs of this patient, 3.6% of the CD8pos, CD45ROpos, CD62Lhigh, and CD25neg T cells were CEA-dextramer positive (data not shown), indicating that circulating CEA protein can generate tolerized CD8 T cells in CRC patients. In conclusion, in peripheral blood of healthy individuals, CD8 T cells with a phenotype and functional repertoire similar to that of murine CD8 T cells tolerized to systemically circulating antigens are present, and this CD8 T cell population was found in increased frequency in cancer patients. This finding suggests that this novel immunosuppressive mechanism may hamper anticancer immunotherapy strategies.

Figure 5.

Detection of phenotypically tolerant CD8 T cells in healthy individuals and CRC patients. PBMCs from healthy individuals (A, B, C, D) or CRC patients (D) were stained for CD8, CD45RO, CD62L, and CD25. (A) Dot plot shows CD62L and CD25 expression on viable CD8+CD45RO+ T cells of a healthy donor. Numbers indicate the percentage of cells in each quadrant. (B) Flow cytometric analysis of Foxp3, IL-10, PD-1, granzyme B, and perforin expression in CD8posCD45ROposCD62Lhigh and CD25neg (solid line) or CD62Lhigh and CD25pos (dotted line) T cells from healthy donors. (C) CD8posCD44highCD45ROpos T cells were sorted into CD62Lhigh and CD25pos or CD62Lhigh and CD25neg fractions. A total of 10,000 cells were stimulated as indicated or were left untreated. IFNγ secretion was analyzed via enzyme-linked immunosorbent assay. The data for two individual donors are shown. (D) Enumeration of CD8posCD45ROposCD25neg and CD62Lhigh T cells in PBMCs from healthy donors (n = 14) and CRC patients (n = 15). ***P ≤ 0.001.


Several immunoregulatory mechanisms have been identified that impede the development of antitumor immunity and can impair the success of cancer immunotherapy.6, 7, 16 Certain tumors such as CRC release substantial amounts of antigens such as CEA into the circulation. The significance of the release of such tumor antigens for tumor-specific immunity is not clear. Here, we describe a novel mechanism by which the presence of systemically circulating tumor antigens can additionally circumvent the development of tumor-specific CD8 T cell immunity. LSECs efficiently scavenge circulating antigens, cross-present them to naïve CD8 T cells, and induce nondeletional T cell tolerance17 and may thus skew tumor-specific CD8 T cell responses.18

We show that systemically circulating CEA protein was efficiently taken up by LSECs, whereas there was very little uptake of CEA by DCs and macrophages. In LSECs, but not DCs or macrophages, uptake depended on MR expression. MR-mediated uptake of antigens is essential for efficient cross-presentation in DCs because of routing of this receptor into endocytic compartments competent for cross-presentation.19 Thus, circulating CEA was efficiently cross-presented to CEA-specific CD8 T cells by LSECs in vivo. DCs and macrophages from both the liver and spleen did not cross-present CEA, which might be attributable to the modest uptake of CEA combined with the lack of MR dependency.

It is therefore likely that circulating CEA is predominantly cross-presented by LSECs for presentation to circulating naïve CEA-specific CD8 T cells. LSECs form a large population of cells (>1 × 1010 in human liver) with a substantial cumulative surface within the hepatic sinusoids where circulating T cells can easily establish cognate interaction because of slow blood flow and narrow vessel diameter.20 It is thus conceivable that LSECs can compete for functional differentiation of naïve CD8 T cells with professional APCs that exert their function in a distinct anatomic location (i.e., secondary lymphatic tissue). Furthermore, we provide evidence that the consequence of cross-presentation of circulating CEA is induction of tolerant CEA-specific CD8 T cells that cannot execute CTL effector function in response to antigen-specific restimulation. Such CEA-specific yet tolerant CD8 T cells fail to control the growth of a CEA-expressing CRC cell line in vivo. This suggests that tolerogenic functional differentiation of naïve CEA-specific CD8 T cells in the context of systemic CEA protein distribution determines their subsequent nonresponsiveness when encountering their cognate antigen on tumor cells, in other words establishing an additional form of tumor immune escape. Many tumor-associated antigens (e.g., α-fetoprotein, CA-19-9, CA-125) are present in the circulation of tumor patients.21, 22 Therefore, the immunotolerizing effects of circulating tumor antigens on the endogenous CD8 T cell repertoire may be relevant for additional types of tumor antigens as well as other types of cancer in addition to CRC.

The key coinhibitory molecule B7H1 involved in LSEC tolerance induction in vitro15 is also relevant for development of CEA-specific tolerant CD8 T cells upon systemic administration of CEA in vivo. Clearly, CEA-specific CD8 T cells generated in B7H1−/− but not C57BL/6 mice after cross-presentation of circulating CEA protein showed potent antitumor activity, demonstrating the essential role of B7H1 for tolerance induction toward CEA (Fig. 4). This finding demonstrates the essential role of B7H1 in this novel form of tumor immune escape and reiterates the similarity between LSEC-mediated tolerance induction and CD8 T cell tolerance developed in the presence of circulating CEA. In addition, CEA-specific CD8 T cells rendered tolerant after systemic application of CEA protein showed phenotypic characteristics similar to those of LSEC-tolerized CD8 T cells that were CD44highCD62LhighCD25neg and nonresponsive to PMA/ionomycin stimulation.

Importantly, we detect this particular antigen-experienced yet tolerant CD8 T cell population in normal healthy human individuals indicating that induction of CD8 T cells nonresponsive to circulating antigens may be occurring normally in a physiological situation. This particular surface phenotype is also found in central memory CD8 T cells.23 However, these memory T cells are easily activated upon re-encounter with their specific antigen.23 Intriguingly, in patients with CRC, we found increased frequencies of this antigen-experienced but nonresponsive CD8 T cell population and detected CEA-specific CD8 T cells with this particular phenotypical population, which suggests that tumor cells may have initiated CTL immune escape by the mechanism observed in the experimental murine tumor model.

The blockade of B7H1/PD-1 interactions has proven successful in re-activating PD-1–positive exhausted T cells in chronic viral infections.24 Studies in mice have shown that PD-1 blockade can also have beneficial effects in antitumor therapy25 and first studies in cancer patients have yielded positive results.26 The blockade of coinhibition may allow naïve precursor CD8 T cells with tumor specificity to be stimulated by normally tolerizing APCs, but because of the low frequency of such T cells, the generation of a sizeable population of tumor-reactive CTLs will require time. The blockade of coinhibitory molecules will likely not reinvigorate the already tolerized CD8 T cells to increase tumor-specific immunity, because these cells already resist stimulation via the TCR. The conditions necessary for the activation of antigen-experienced but tolerant tumor-specific CD8 T cells will require further experimental efforts and may further optimize the effectiveness of immunotherapy.


We thank A. Dolf, P. Wurst, and E. Endl of the flow cytometry core facility for technical assistance.