Two Mechanisms for Tumor Evasion of Preexisting Cytotoxic T-Cell Responses: Lessons from Recurrent Tumors¹

The immune system has evolved to combat infectious agents (1, 2). Tumors, on the other hand, are not efficiently recognized and eliminated by the immune system. The poor recognition of tumors is manifested at both the induction and effector phases of antitumor immune responses. Recent studies have revealed multiple mechanisms by which tumors have avoided immune destruction: (a) it is clear that optimal induction of T-cell responses requires the expression of both tumor antigens and costimulatory molecules (3, 4, 5, 6, 7). Although many tumor cells have recognizable antigen (8), they lack costimulatory molecules (9) and are thus poor inducers for T-cell activation. Whereas priming of antitumor CTLs can take place without costimulatory molecules on the tumor (10), our recent study indicated that effector T cells are not produced unless the costimulatory molecule is expressed on the tumor cells (11); (b) many established tumors have defective antigen presentation machinery (12, 13, 14, 15, 16). Whereas requirements for the tumor expression of MHC:peptide complex for induction of antitumor effector CTLs remain to be demonstrated, tumor lysis by CTLs requires expression of target antigens on the tumors; (c) established tumors may thrive despite the presence of antitumor immune responses. Wick et al. (17) and Singh et al. (18) have elegantly demonstrated that although established tumors can induce immunity for subsequent challenge by the same tumor cells, the primary tumors are spared from immune rejection; and (d) large tumors may nonspecifically affect the function of T cells by changing the composition of TCR/CD3/ζ-chain complex (19). Each of these mechanisms can partially explain poor T-cell-mediated tumor rejection.

Immune-mediated rejection involves both the induction and effector function of T cells, which makes it difficult to discern whether the failure is due to induction or to effector function. This obstacle can be overcome by analyzing the phenotype of the recurrent tumors. If the primary tumor rejection was mediated by T cell immunity, which is generally long lived, then the recurrent tumors must have arisen in the presence of preexisting antitumor immunity. Uyttenhove et al. (20) isolated P1A⁻ recurrent P815 tumors in mice that had rejected the wild-type P815 tumor. This study suggested that tumors may evade CTL recognition by changing their antigenic composition, much as viruses do. Unfortunately, the mechanisms of tumor recurrences have not been studied systematically, perhaps due to the lack of proper experimental models.

The discoveries of the B7 family members, B7-1 and B7-2, as the prototypic costimulatory molecules have inspired a novel approach to enhance antitumor immunity. Several laboratories have demonstrated that expression of B7-1 and B7-2 on tumor cells induces CD8 T-cell-mediated rejection of the challenging tumors (4, 5, 6, 7). In the plasmacytoma model that we have studied previously (5, 6, 11, 21), we have observed several cases of tumor recurrence during the course of study, whereas the majority of the mice remain tumor free for more than a year. Because the mice bearing the recurrent tumors have strong antitumor CTL responses, the recurrent tumors must have evaded host immunity. We therefore analyzed the immunological properties of the recurrent tumors. Our results demonstrate that the recurrent tumors can be divided into two groups. One group lacks costimulatory molecule B7, whereas the other group has severely reduced levels of cell surface MHC class I with a concurrent reduction of multiple genes devoted to MHC class I antigen presentation. Both groups of tumor cells are drastically less susceptible than the parental J558-B7 tumor to cytolysis by the ex vivo antitumor CTLs. Our results revealed two independent mechanisms that allow tumor evasion of preexisting antitumor immunity. The requirement of both antigen and costimulatory molecules on target cells for CTL recognition raised a fundamental question regarding the nature of the effector mechanism in antitumor immunity and has important implications for tumor immunotherapy.

Syngeneic BALB/c mice were purchased from the National Cancer Institute (Bethesda, MD). Male mice between 6 and 18 weeks of age were used for the study.

Tumor cell line J558-B7 has been described previously (5). All other tumor cell lines were isolated from recurrent tumors. Briefly, BALB/c mice were injected with 5 × 10⁶ J558-B7 tumor cells in the left flank. Tumors normally became palpable within the first 2 weeks. Approximately 80% of the mice rejected tumors between the third and fourth week. Whereas the overwhelming majority of the mice that had rejected the tumors remained tumor free, some of them developed recurrent tumors at the initial site of injection. The history of the recurrent tumors is summarized in Table 1. The recurrent tumors were removed and ground into a single cell suspension. The viable cells were either analyzed directly or cultured for a given period of time before analysis. In some experiments, the recurrent tumors were cloned by limiting dilution, and a representative clone from each recurrent tumor was amplified and used for biochemical or immunological analysis.

Viable J558 tumor cells were incubated with either anti-B7-1 mAb⁵ 10.16A.1 (22) or anti-MHC class I mAbs HB27 (H-2L^d specific; Ref. 23), HB76 (H-2D^d specific; Ref. 23), and HB159 (H-2K^d specific) at 4°C for 2 h. After three washes with PBS containing 1% FCS, the bound mAbs were detected by FITC-labeled goat antimouse IgG (Caltag, Mountainview, CA).

In some experiments, anti-B7-1 mAb and biotinylated anti-L^d mAb (Phar-Mingen, San Diego, CA) were used for two-color flow cytometry. The anti-B7-1 mAb bound to the cell surface was detected by FITC-conjugated goat antihamster mAb (mouse/rat immunoglobulin adsorbed), whereas biotinylated HB27 was measured by phycoerythrin-conjugated streptavidin.

Two types of effector T cells were used for the study. The tumor antigen-specific CTL cell line P1CTL has been described previously (24). The ex vivo CTLs, which were isolated from J558-B7 tumors at 3 weeks after injection as described previously (5, 6), were used as effectors without in vitro restimulation. J558-B7 or various recurrent tumor cell lines were labeled with ⁵¹Cr and used as a target. The data presented and the specific lysis percentage were calculated as described previously (11).

To quantify the number of precursor cells, 24 replicates of given numbers of spleen cells were stimulated with 1 μg/ml P1A peptide in the presence of 1% supernatants from activated EL4 cells and 2 × 10⁵ cells/well of irradiated syngeneic feeder cells. After 7 days, the cultures were split into two and washed with medium. A total of 5 × 10³ P1A-pulsed or control NP peptide-pulsed, ⁵¹Cr-labeled target cells were added to the wells to determine both the specificity and cytotoxicity of the microcultures. The wells with a specific lysis percentage equal to or greater than 3 SDs of medium release were scored as positive. The small number of wells that scored positive for both P1A and control NP peptides were excluded from the analysis. The frequencies of the CTLp were determined by linear regression analysis.

The GAPDH, H-2D^b, β-2 M, LMP-2, LMP-7, TAP-1, and TAP-2 probes were amplified from either spleen or tumor RAW8.1 RNA (16). The sequence of the primers used is listed in Table 2. All PCR products were verified by DNA sequencing. The hybridization condition is essentially identical to that described previously (16), except that total RNA was used instead of polyadenylated RNA.

We have reported that in the majority of the syngeneic mice, plasmacytoma J558-expressing costimulatory molecule B7-1 (referred to hereafter as J558-B7) is rejected by a CD8-dependent mechanism (5, 6, 11, 21). Whereas the majority of the mice remain tumor free, several recurrent tumors were observed at the site of initial injection at various time points after the initial rejection of the tumors, as shown in Table 1. The recurrent tumors were surgically removed. The single-cell suspensions were cultured in vitro and subjected to further analysis.

Our previous studies have established that a major T-cell antigen in the J558-B7 tumor is the P1A antigen (5, 6, 11). To determine whether mice bearing the recurrent tumor maintain immunity induced by the primary J558-B7 challenge, we carried out limiting dilution analysis to quantitatively compare the number of P1A-specific precursors in a mouse that rejected J558 tumors with that in a mouse that bore the recurrent tumor ReB7. We stimulated given spleen cells with irradiated syngeneic feeder cells and P1A peptide for 7 days, split each well into two, and measured CTL activity against tumor antigen P1A and a control peptide for influenza virus. As shown in Fig. 1, spleen cells from the mouse that had rejected J558-B7 tumor with no tumor recurrence had expanded numbers of CTLp for tumor P1A, as did those from the mouse that bore the recurrent tumor, ReB7. In contrast, naive mice had a barely detectable level of P1A-specific CTLp.

Because limiting dilution analysis does not reveal the existence of effector CTLs, we tested whether tumor recurrence can take place in the presence of effector CTLs. We isolated TILs from the recurrent BR-2 tumors and determined whether they contained cytotoxic T cells. As shown in Fig. 2, TILs from BR-2 tumors kill J558-B7 tumor cells, but not mutant cell line ReB7 that lacks cell surface MHC class I (16). It is therefore likely that the effector T cells are CTLs rather than NK cells. Because the ex vivo TILs were used without further restimulation in vitro, mature effector CTLs must be present in the BR-2 tumor. The coexistence of the effector CTLs and the BR-2 tumor indicated that BR-2 tumor cells evaded the cytolysis of the CTLs. Moreover, because all six mice with tumor recurrence have rejected the initial tumor challenge and because we have established that B7-1-transfected tumors are rejected by a CD8 T-cell-dependent mechanism (5), it is likely that all recurrent plasmacytomas can evade the preexisting immunity.

To understand the mechanism of tumor evasion, we analyzed the cell surface expression of MHC and costimulatory molecule B7-1. The profiles of five recurrent tumors are shown in Fig. 3. Parental J558-B7 expresses high levels of B7-1, H2-L^d, and H2-K^d. ReB7, on the other hand, expresses essentially undetectable cell surface H-2D^d (data not shown), H-2L^d, and H-2K^d, although high levels of B7-1 were detected on the surface. Re#3, Re#6, and Re#11 have phenotypes similar to those of ReB7, although the reduction of MHC class I is somewhat less pronounced. In contrast, Re#10 has lost B7-1 but retains cell surface MHC class I. Interestingly, the recurrent tumor BR-2 is comprised of two populations. As shown in Fig. 4, approximately 60% of the BR-2 cells express B7-1, whereas 40% of the cells express MHC class I. To test whether reductions in B7-1 and H-2L^d are mutually exclusive, we carried out two-color flow cytometry. As shown in Fig. 5 a, optimal expression of B7-1 and H-2L^d is indeed mutually exclusive, because B7⁺MHC^low and B7⁻MHC^high populations can be obtained by sorting.

Because the BR-2 tumor had been infiltrated with effector CTLs (Fig. 2), it is likely that both populations of recurrent tumor cells can avoid T-cell-mediated immunity in vivo. To substantiate this notion, we tested the susceptibility of B7⁺MHC^low and B7⁻MHC^high populations (Fig. 5,a) to cytolysis by either ex vivo CTLs (5, 6, 11) or an established CTL cell line, P1CTL (24). As shown in Fig. 5,b, both B7⁺MHC^low and B7⁻MHC^high populations are resistant to cytolysis by ex vivo CTLs. Interestingly, the established CTL cell line P1CTL lysed B7⁺MHC^low and B7⁻MHC^high populations to a similar extent (Fig. 5 c). Because ex vivo CTLs (5, 6, 11) and the P1CTL cell line (24) recognize the same antigenic peptide, we conclude that recognition by the established CTL cell line is independent of B7-1, whereas ex vivo CTLs require expression of B7-1 on the target cells for optimal cytolysis. Moreover, lysis of B7⁻MHC^high tumor cells by the P1CTL cells indicates that the lower susceptibility of the B7⁻ targets to ex vivo CTLs is not due to defective processing of tumor antigens.

We also tested all recurrent tumor cell lines for their susceptibility to cytolysis by TILs from J558-B7 tumors. As shown in Fig. 6,a, ex vivo TILs from J558-B7 tumors efficiently lysed J558-B7. However, none of the recurrent tumor cell lines are lysed efficiently. To determine whether any recurrent tumors evade CTLs by down-regulating the tumor antigen P1A, we analyzed its expression among the recurrent tumor lines. As shown in Fig. 6 b, all recurrent tumors and their parent J558-B7 tumors express a comparable amount of tumor antigen P1A. The fact that the recurrent tumors, which lack either MHC class I or B7-1, have reduced susceptibility to ex vivo CTLs indicates that efficient recognition of the tumor cells by the ex vivo CTLs requires the presence of both costimulatory molecules and target antigen.

Cell surface expression of MHC class I antigens requires intact antigen processing machinery (25). Lack of cell surface MHC can be due to defects in proteosome degradation of cytosolic proteins (26, 27), transportation of peptides into the ER (28, 29, 30, 31), or the assembly of MHC class I heavy chain, β-2 M, and the antigenic peptides (32, 33). The specificity of the proteosome is modified by MHC-linked genes LMP-2 and LMP-7 (34, 35, 36), whereas peptide transport requires functional TAP-1 and TAP-2 genes (28, 29, 30, 31). To understand the molecular basis of the antigen presentation defects, we isolated RNA from the recurrent tumor cell lines and determined the expression of LMP-2, LMP-7, TAP-1, TAP-2, tumor antigen P1A, β-2 M, and MHC class I heavy chain. As shown in Fig. 7, whereas expression of P1A is normal on all cell lines, considerable down-regulation in MHC class I, β-2 M, LMP-2, LMP-7, TAP-1, and TAP-2 was observed among the MHC^low recurrent tumors. As described previously (16), ReB7 expressed barely detectable levels of LMP-2, LMP-7, TAP-1, and TAP-2 and significantly reduced MHC class I. Here we have also observed a significant reduction in β-2 M. All other MHC^low cell lines have a concurrent reduction of the genes involved in MHC class I antigen presentation. In contrast, B7-1⁻ clone 10 retains a high level of expression of all of these antigen presentation genes. Because inactivation of TAP-1, TAP-2, or β-2 M is sufficient to avoid antigen presentation, simultaneous down-regulations of multiple genes strongly suggest that defects in a key control element may be responsible for the antigen presentation defects.

Multiple mechanisms have been described that enable tumor evasion of host immunity. Most investigators have focused on the mechanism by which tumors avoid induction of antitumor immunity. The mechanism for tumor evasion of preexisting immunity is less clear and has been difficult to address due to a lack of experimental models. We have analyzed six recurrent tumors in mice that have rejected the B7-1-transfected plasmacytoma J558 and found that they belong to two groups: (a) one has down-regulated cell surface MHC; and (b) the other has lost the costimulatory molecule B7-1. Because the mice that bear the recurrent tumors have developed a strong antitumor immunity, the recurrent tumors must have evaded the preexisting immunity. Thus our study reveals two mechanisms for tumor evasion of host immunity: (a) the down-regulation of MHC class I; and (b) the loss of costimulatory molecule B7-1.

MHC class I antigen presents antigenic peptides to cytotoxic T cells and is therefore essential for T-cell recognition. Whereas poor expression of MHC class I and other genes devoted to antigen presentation has been described in malignant human tumors and experimental tumors (12, 15, 37, 38, 39, 40, 41), its contribution to tumor evasion of host immunity has not been clearly established. A major issue that was not resolved is whether poor antigen presentation reflects the tissue origin of the malignant tumor or, alternatively, represents an active mechanism of tumor evasion. These two hypotheses are difficult to differentiate because precursors for the malignant tumors are generally not available for comparison. Here we showed that the majority of the recurrent tumors have down-regulated cell surface MHC class I. In comparison to J558-B7, the precursor of the recurrent tumors, the recurrent tumors with reduced cell surface MHC are much less susceptible to cytolysis by ex vivo CTLs. Moreover, we have recently reported that in one of the recurrent tumors isolated, restoration of cell surface MHC by the wild-type PML gene restored host immunity (16). Taken together, these results formally establish that down-regulation of MHC class I on tumors allowed tumor evasion of host immunity.

Optimal cell surface expression of MHC class I involves multiple steps including proteolysis by proteosome (26, 27), transportation of peptides into the ER (28, 29, 30, 31), and assembly of MHC class I heavy chain, β-2 M, and the antigenic peptides (32, 33). The specificity of proteosome-mediated protein degradation is modified by MHC-encoded products LMP-2 and LMP-7 (36), whereas peptide translocation into the ER is strictly dependent on both TAP-1 and TAP-2 (42). Genetic experiments revealed that cells lacking either LMP-2 or LMP-7 can have reduced cell surface MHC class I (26, 27), whereas cells deficient in either TAP-1 or TAP-2 are essentially devoid of intact cell surface MHC molecules (31, 43). Our analysis of the recurrent tumor cells revealed that MHC class I-deficient tumors have significantly reduced expression of LMP-2, LMP-7, TAP-1, TAP-2, MHC class I heavy chain, and β-2 M. Given the fact that dysfunction of a single gene such as β-2 M, TAP-1, or TAP-2 is sufficient to eliminate antigen processing, it is surprising that the expression of six independent genes are affected simultaneously in six independently derived recurrent tumor cell lines. Whereas other interpretations cannot be ruled out at this stage, the simplest interpretation of this finding is that there exists at least one master regulator for the expression of multiple genes involved in MHC class I antigen presentation. Our recent studies demonstrate that the antigen presentation defect in one recurrent tumor ReB7 is restored by overexpression of the proto-oncogene PML and that the recurrent tumor has a dominant negative mutation in the PML gene (16). Given the significant difference in the levels of reduction between ReB7 and the other recurrent tumor cell lines studied here, it is unlikely that the same mutation of PML can be responsible for the antigen-presentation defects. Nevertheless, there can be more than one way to inactivate/attenuate PML function, and it would be of interest to test whether malfunction of PML is also responsible for down-regulation of MHC in all other recurrent tumor cell lines characterized in the current study.

An important observation reported in this study is that loss of B7-1 alone is sufficient to allow tumor evasion of CTL responses. Of the seven tumor cell lines from six independently recurred tumors, two retain normal levels of cell surface MHC class I but lack B7-1. Of particular interest is the BR-2 tumor that consists of two populations of cells: (a) B7-1⁻MHC^high cells; and (b) B7-1⁺MHC^low cells. The substantial representation of both populations and the coexistence of CTLs in the tumor reveal that loss of either B7-1 or MHC class I is sufficient to allow tumor evasion. Because TILs from this recurrent tumor contain effector CTLs, the simplest interpretation for the coexistence of the CTLs and the B7-1⁻MHC^high population is that it is not susceptible to cytolysis by ex vivo CTLs. Indeed, the recurrent B7-1⁻MHC^high cell line is resistant to lysis by CTLs freshly recovered from tumors, although it is efficiently recognized by the established CTL cell line that recognizes the same antigen. This finding corresponds to our previous observation that B7-1-transfected tumors are substantially more susceptible to lysis by ex vivo CTLs than J558 cells transfected with vector alone (5). The requirement of B7-1 for effector function suggests a blockade of CTL effector function that can be bypassed by expression of B7-1 on the tumor cells. Whereas the nature of the blockade and the mechanism by which B7-1 overrides it remain to be elucidated, two sets of observations may provide insights: (a) much like NK cells, CTLs can express MHC class I-specific inhibitory receptors (44, 45, 46); and (b) B7-1 expression on MHC class I⁺ target cells enhances NK cell recognition (47).

Since our initial report on the role of B7-1 at the effector phase of antitumor immunity (5), several studies have extended this observation (10, 48, 49). Nevertheless, it should be emphasized that for a large number of tumor models, expression of B7-1 on the tumor cells can induce antitumor immunity that is independent of B7 at the effector phase (3, 4, 7). Moreover, most somatic cells in vivo do not express B7; however, CTLs can eliminate viral infection by a perforin-dependent cytolytic function (50). It is therefore unlikely that B7-1 is required for the effector function of antimicrobial immunity. However, recent studies in several transgenic models support the notion that the expression of costimulatory molecule B7-1 on target tissue is required for severe autoimmune attack (51, 52, 53). Thus, antitumor immunity may be similar to autoimmune diseases at the level of T-cell effector function. In this regard, it is worth noting that in the tumor model reported here, the antigen is an unmutated antigen, P1A, that is expressed at low levels in normal tissue (48). In the tumor model in which the effector function is B7 independent, the tumor antigen is either unknown (3, 7) or known to be a viral protein (4). Whereas it remains to be demonstrated whether the endogenous expression of tumor antigen has imposed the special restriction on the CTL effector function, the fundamental requirement of B7 for the effector function needs to be taken into consideration in tumor immunotherapy.

We thank Dr. John Hirst for flow cytometry and Jennifer Kiel for editorial assistance.