Targeting Alpha-Fetoprotein (AFP)–MHC Complex with CAR T-Cell Therapy for Liver Cancer

The use of chimeric antigen receptor (CAR) or T-cell receptor (TCR) engineered T cells for cellular immunotherapy has shown significant promise in the fight against cancer, particularly for hematologic malignancies (1–4). The success of this method relies on autologous or allogeneic T cells genetically modified to express synthetic CARs/TCRs that redirect them to attack and kill tumor cells expressing a particular antigen. Despite recent successes in the field, the use of redirected T-cell therapy for the treatment of solid tumors remains challenging. Insufficient localization/persistence in solid tumors (5) and an immunosuppressive microenvironment (6, 7) have been implicated in the low efficacy observed to date. Furthermore, toxicity associated with off-target or on-target/off-tumor effects has proved to be a serious risk factor that has highlighted the lack of tumor-specific antigens for solid tumors (8, 9). Thus, identifying the precise target antigen and designing CARs/TCRs to be highly selective are critical for the clinical application of such therapies.

Hepatocellular carcinoma (HCC) is the predominant type of liver cancer, affecting over 700,000 people each year worldwide (10). Incidents of HCC are gradually rising, while survival rates remain stagnantly low (11, 12). Current treatments of HCC are limited to resection (for small localized tumors), radiation, ablation, chemoembolization, liver transplantation, and targeted therapy (namely sorafenib), which show limited efficacy and marginal survival benefits in the majority of treated patients (12, 13). For many patients the only treatment offered is palliative.

AFP is a secreted glycoprotein that is commonly overexpressed in tumors of endodermal origin including pediatric hepatoblastoma and HCC (14–17). AFP is also expressed in the fetal yolk sac, liver, and gastrointestinal tract, but rarely in adult tissues, making it an attractive candidate for T-cell immunotherapy. Elevated expression of AFP in tumors and serum is found in 60%–80% of HCC patients (14) and correlates with poor prognosis (18). Although the function of AFP remains unclear, it has been reported to promote cell proliferation, suppress apoptosis, and act as an immunosuppressive agent (19–23), suggesting a potential role in disease progression.

Despite the attractiveness of AFP as a cancer-specific antigen, AFP is expressed intracellularly and secreted, and therefore has been considered undruggable with conventional antibody-based therapies. However, all proteins in nucleated human cells are processed into peptides and presented by class I MHCs on the surface of cells. We have previously demonstrated that targeting such peptide–MHC complexes is feasible by engineering a “TCR-like” antibody against a peptide of Wilms Tumor 1 (WT1), an intracellularly expressed cancer-specific antigen, complexed with HLA-A*02:01 (24). AFP/MHC–specific TCRs have been identified in mice and humans, indicating that AFP–MHC complexes can be immunogenic. Furthermore, vaccines employing AFP plasmids or dendritic cells pulsed with AFP peptides elicit immune reactions in mice (25, 26), and several HLA-A*02:01–restricted AFP peptides have been found to be immunogenic in humans (27, 28). However, attempts to immunize HCC patients against AFP by activating endogenous TCRs have only shown modest therapeutic responses to date (29–31), potentially due to the inability of vaccine-related strategies to overcome tumor-related immunosuppression.

On the basis of the expression pattern of AFP in HCC and its demonstrated immunogenicity, we developed ET1402L1, a fully human antibody that selectively binds to AFP_158–166 (AFP₁₅₈) peptide presented by HLA-A*02:01, the most prevalent class I MHC in the human population (26, 32). We then engineered this “TCR-like” antibody into a second-generation CAR to combine antigen recognition with direct and potent immune effector activity. We show that T cells transduced with ET1402L1-CAR (herein referred to as AFP-CAR) selectively degranulated, released cytokines, and killed HLA-A*02:01⁺/AFP⁺ cells. Moreover, AFP-CAR T cells effectively induced tumor regression in multiple established xenograft models of human HLA-A*02:01⁺/AFP⁺ liver cancer, demonstrating the effectiveness of this immunotherapeutic approach in a solid tumor setting.

Cell lines were obtained from ATCC or Caliper Life Sciences. To generate SK-HEP-1-MG cells, parental SK-HEP-1 cells were stably transduced with a pLenti vector carrying a fragment of the AFP gene spanning the peptide region. Healthy human donor peripheral blood leukocytes were obtained from Blood Centers of the Pacific and T cells were isolated using an EasySep Human T Cell Isolation Kit (StemCell Technologies).

Peptides were synthesized by Elim Biopharmaceuticals. Biotinylated AFP₁₅₈/HLA-A*02:01 and control peptide-HLA-A*02:01 complexes were generated by refolding recombinant HLA-A*02:01 in the presence of peptide and β2-microglobulin (β2M). Streptavidin–PE–conjugated AFP₁₅₈/HLA-A*02:01 tetramer (PE-Tetramer) was made by Eureka Therapeutics.

Selection of human antibodies specific to AFP₁₅₈/HLA-A*02:01 by cell panning and generation of IgG1 antibody toward AFP₁₅₈/HLA-A*02:01 complexes were performed as described previously (24).

AFP₁₅₈ peptides with alanine substitutions were synthesized by Elim Biopharmaceuticals. Binding of phage clones A-D to T2 cells pulsed with mutated peptides at 50 μg/mL was detected by flow cytometry using mouse anti-M13 antibody (GE Healthcare) and R-PE–conjugated anti-mouse IgG(H+L) antibody (Vector Laboratories).

Anti-AFP₁₅₈/HLA-A*02:01 scFvs were grafted onto a second-generation CAR with CD28 and CD3ζ signaling domains engineered in cis to provide intracellular T-cell stimulation signals (33). The CAR sequence was cloned into a pCDH lentiviral vector (Systems Biosciences) for delivery into T cells. Human T cells were cultured in RPMI1640 supplemented with 10% FBS and activated with CD3/CD28 Dynabeads (Thermo Fisher Scientific). One day after activation, human T cells were transduced with concentrated lentivirus in RetroNectin (Takara) coated plates. Transduced T cells were then expanded in the presence of either 100 U/mL IL2 (Sigma) or 10 ng/mL IL7 and 5 ng/mL IL15 (Peprotech) for 8–12 days. Transduction efficiency was assessed by flow cytometry using Streptavidin–PE–conjugated AFP₁₅₈/HLA-A*02:01 tetramer.

Animal experiments were conducted at Murigenics. Female SCID-Beige or NSG mice aged 6–8 weeks were used. For subcutaneous tumor models, cells were mixed with 50% Matrigel prior to implantation and tumors were measured by calipers. For Hep G2 tumor models, 2.5 × 10⁶ cells were implanted per mouse; for SK-HEP-1-MG models, 5 × 10⁶ cells were implanted. When tumors reached approximately 100 mm³, mice were randomized to treatment groups. For intraperitoneal studies, 2.5 × 10⁶ Hep G2-luc2 cells were injected in PBS and tumors were measured by total luminescent flux using an IVIS Imaging System (Caliper Life Sciences). Serum AFP was measured by ELISA (Affymetrix). Human anti-CD3 IHC was performed by Ensigna Biosystems.

Statistical analyses were performed using Prism GraphPad software. For studies comparing two groups, we used a Student t test. For studies with multiple groups, we used a one-way ANOVA followed by Dunnett test which accounts for multiple comparisons. All analyses were two-tailed.

A schematic of our screening and selection process is shown in Fig. 1A. A phage library (E-ALPHA^TM Phage Display, Eureka Therapeutics) comprised of both naïve and semisynthetic human single-chain variable fragment (scFv) B-cell antibodies was screened for clones that selectively bound T2 cells (HLA-A*02:01⁺) pulsed with the AFP₁₅₈ peptide but not to cells pulsed with a mixture of 100 endogenous HLA-A2 peptides derived from various disease-related and house-keeping proteins. We validated the selective binding of 1,260 individual phage clones from three rounds of panning to AFP₁₅₈/HLA-A*02:01 complexes versus control complexes (HLA-A*02:01 bound to a mixture of 100 endogenous human peptides) by ELISA (representative clones are shown in Fig. 1B). We identified 605 positive clones by ELISA which corresponded to 82 unique clones by sequence, 44 of which demonstrated selectivity for AFP₁₅₈-loaded T2 cells over native T2 cells or T2 cells loaded with control peptides by flow cytometry (representative clones are shown in Fig. 1C).

We identified the epitope recognized by our top four antibodies (clones A–D) using alanine scanning mutagenesis. Human AFP₁₅₈ peptides with alanine substitutions at positions 1, 3, 4, 5, 6, 7, and 8 were synthesized and pulsed onto T2 cells. The anchor residues at positions 2 and 9 were not altered as mutations at these positions are known to ablate peptide–MHC binding (34). Antibody phage clones A–D were then tested for binding to these peptide-loaded T2 cells and analyzed by flow cytometry (Supplementary Fig. S1). Although all antibodies recognized the small conformational epitope formed by the AFP₁₅₈ peptide and its surrounding MHCα chain residues, the number and position of key peptide residues interacting with different antibodies varied among the clones (Fig. 1D and Supplementary Fig. S1). Clones B and D had the largest epitope, with 5 sensitive positions within the 9-mer peptide, indicating high selectivity for the AFP₁₅₈ peptide sequence.

To measure binding affinities to the target antigen (AFP₁₅₈ complexed with HLA-A*02:01), we engineered clones A–D into full-length antibodies with mouse IgG1 heavy- and light-chain constant regions as described previously (35). Binding affinities of the mouse chimeric antibodies against AFP₁₅₈/HLA-A*02:01 were measured by surface plasma resonance (SPR) using the BIACore X-100 system. K_d values were calculated to be in the subnanomolar range (Supplementary Table S1). A*02:01 is the most common subtype of HLA-A02 in the human population; however, other subtypes exist at relatively high frequencies in various ethnic groups. We therefore evaluated the binding affinities of our antibodies towards AFP₁₅₈ complexed with additional HLA-A02 subtypes including 02:02, 02:03, 02:05, 02:06, 02:07, and 02:11 using the ForteBio Octet system. Results indicate that clones A–D also bound other HLA-A02 subtypes to varying degrees, which might broaden the eligible patient population (data not shown).

We next tested the selectivity of clones A–D to antigen-positive cells in bispecific antibody (BsAb) format in a T-cell–mediated killing assay. Hep G2 cells, a human liver cancer cell line (hepatoblastoma; ref. 36), expressing both HLA-A*02:01 and AFP (A2⁺/AFP⁺) served as a positive control, along with 6 negative control tumor cell lines: the liver-derived cell line SK-HEP-1 which is HLA-A*02:01 positive but does not express AFP (A2⁺/AFP⁻), Hep 3B, an AFP⁺ HCC cell line that does not express HLA-A*02:01 (A2⁻/AFP⁺), and A2⁻/AFP⁻ cell lines: HeLa (cervical), Jurkat (T-cell), Raji and Daudi (B-cell), and K562 (myeloid). At low antibody concentrations (0.04 μg/mL), all four BsAbs showed selectivity towards Hep G2 cells, although at higher concentrations clone B demonstrated superior specificity for lysing Hep G2 cells (Fig. 2A). Ectopic expression of the AFP₁₅₈ peptide region (AFP mini-gene, MG) in AFP-negative, HLA-A*02:01–positive SK-HEP-1 cells (SK-HEP-1-MG) conferred sensitivity to killing mediated by the BsAb clones, while it had no effect on HLA-A*02:01-negative HeLa cells transduced with the same AFP mini-gene (HeLa-MG) (Fig. 2B). Similar to the results in Fig. 2A, clone B showed the highest degree of specific killing activity.

On average, each nucleated cell in the human body expresses about half a million different peptide–MHC class I complexes (37). To develop anti-peptide–MHC complex antibodies into anticancer drugs with high specificity and therapeutic index, it is essential for the antibodies to specifically recognize the target peptide–MHC complex, but not the MHC molecule itself or MHC molecules bound to similar or unrelated peptides. Therefore, we tested the binding of our top performing clone (clone B, named ET1402L1), in mouse chimeric IgG1 format against 100 known HLA-A*02:01 peptides derived from diabetes-, autoimmune disease-, cancer- and virus-related proteins. T2 cells were loaded with individual peptides and binding of ET1402L1 or an isotype control was measured by flow cytometry. While the isotype control did not bind any of the peptide-pulsed T2 cells, ET1402L1 bound in an exquisitely selective manner to T2 cells pulsed with AFP₁₅₈ (Supplementary Fig. S2).

We next engineered ET1402L1 into a second-generation CAR containing the CD28/CD3ζ costimulatory domains as described previously (33). The CAR construct was then cloned into a lentiviral vector for transduction of primary human T cells. T cells expressing ET1402L1-CAR (AFP-CAR) and untransduced donor-matched T cells (mock) were first tested against HLA-A*02:01–positive liver cancer cell lines. After 16 hours of incubation with AFP-CAR T cells at an effector:target (E:T) cell ratio of 5:1, specific lysis of A2⁺/AFP⁺ cells was approximately 60% for Hep G2 cells and approximately 75% for SK-HEP-1-MG cells, with minor lysis (<5%) of A2⁺/AFP⁻ SK-HEP-1 cells (Fig. 3A, left). Mock T cells showed mild activity (<10% specific lysis) against all three cell lines in a non-antigen–dependent manner. AFP-CAR and mock T cells were also tested against a panel of HLA-A02–positive and -negative human cell lines from various tissues of origin including skin, cervix, ovaries, breast, pancreas, hematologic, prostate, colon, and kidney. We observed little or no lysis of antigen-negative cells (lacking either HLA-A2 or AFP) by AFP-CAR or mock T cells (Fig. 3A, right). Furthermore, to address the possibility (although unlikely) that circulating AFP can be taken up by normal cells, crosspresented on class I MHC and therefore, targetable by AFP-CAR T cells, we incubated AFP-CAR T cells with A2⁺/AFP⁻ COLO-205 (colon), A204 (muscle), or HUVEC (endothelial) cells cultured with a high concentration of recombinant human AFP (0.4 μg/mL). However, we observed no specific killing of these cell lines (Supplementary Fig. S3A and S3B). Overall, these data demonstrate that AFP-CAR T cells exhibit superb specificity for the target antigen complex.

To further characterize the biological activity of AFP-CAR T cells, we used flow cytometry to detect CD107a surface expression as a measurement of degranulation activity. Consistent with the cell killing assay results, degranulation only occurred when AFP-CAR T cells were cocultured with the antigen-positive Hep G2 or SK-HEP-1-MG cells (Fig. 3B). Cytokine measurements by multiplex ELISA showed that TNFα, IFNγ, GM-CSF, IL2, IL4, IL6, IL8, and IL10 were elevated in the supernatants from AFP-CAR T cells incubated with Hep G2 or SK-HEP-1-MG, but not from cocultures with the antigen-negative SK-HEP-1 cells (Fig. 3C and Supplementary Fig. S4A). Flow cytometry analysis of AFP-CAR T cells stained for intracellular IL2, IL6, TNFα, and IFNγ after 4 hours of incubation with antigen-positive Hep G2 or SK-HEP-1-MG cells showed that IL2, TNFα, and IFNγ, but not IL6, are produced by the CAR T cells themselves (Fig. 3D and Supplementary Fig. S4B). The source of IL6 measured in the killing assays is presumably the target cells. No cytokines were produced when AFP-CAR T cells were cocultured with the antigen-negative SK-HEP-1 cells (Fig. 3C and D and Supplementary Fig. S4A and S4B). To test whether AFP-CAR T cells can penetrate the extracellular matrix in solid tumors to kill cancer cells, we embedded Hep G2, SK-HEP-1, and SK-HEP-1-MG cells in Matrigel where they grew into clusters encapsulated by a thick layer of laminin-rich matrix. AFP-CAR and mock T cells were laid on top of the Matrigel and incubated for 2 days. While both mock and AFP-CAR T cells effectively penetrated the matrix and reached the cell clusters after 48 hours of incubation, only AFP-CAR T cells specifically killed both Hep G2 and SK-HEP-1-MG cells with high efficiency (Supplementary Fig. S5).

We tested the in vivo antitumor activity of AFP-CAR T cells in several established human liver cancer xenograft models. For all in vivo experiments, injected T cells were approximately 100% CD3⁺, 60% CD4⁺, and 40% CD8⁺ (Supplementary Fig. S6A and S6B). To confirm specific lysis of cells expressing the target antigen, we tested each batch in an in vitro killing assay before administration to mice (Supplementary Fig. S6C). Analysis of T-cell differentiation markers in AFP-CAR T cells using anti-CCR7 and CD45RA showed that CD4⁺ CAR⁺ T cells were approximately 26% naïve(T_N), 21% central memory (T_CM), 41% effector memory (T_EM), and 13% effector (T_E) T cells while CD8⁺ CAR⁺ T cells were enriched for T_N (∼68%) and TE (∼24%) with approximately 3% of T_EM and 5% of T_CM (Supplementary Fig. S7). These ratios were similar in mock T cells.

First, HLA-A*02:01⁺/AFP⁺ Hep G2 cells were implanted subcutaneously (s.c.) over the right flank of SCID-Beige mice and then randomized into five groups: (i) no treatment, (ii) intravenous (i.v.) injection and (iii) intratumoral (i.t.) injection of Mock T cells, (iv) intravenous injection, and (v) intraperitoneal injection of AFP-CAR T cells. Tumors in untreated or mock-treated mice grew at a steady rate until reaching a size that required euthanasia. Intravenous administration of AFP-CAR T cells in Hep G2 tumor-bearing mice resulted in a delayed inhibition of tumor growth starting 28 days after the first dose (Fig. 4A). Approximately 17% reduction in tumor size relative to controls was observed by the time control mice had to be euthanized (Fig. 4A). This tumor growth inhibition was reproduced in a consistent manner in a second independent study with similarly treated mice (data not shown). In contrast to the delayed and modest effects of intravenous administration, intratumoral injections of AFP-CAR T cells caused rapid, profound, and lasting tumor regression. Remarkably, 75% of mice treated with AFP-CAR T cells experienced complete regression with no palpable tumor remaining. In mice with palpable tumors, approximately 97% reduction in tumor size relative to controls was observed by the time control mice had to be euthanized (Fig. 4A).

We further assessed the efficacy of AFP-CAR T cells against large/advanced Hep G2 tumors through intratumoral delivery. In this study, intratumoral injection of mock or AFP-CAR T cells was given when the average tumor volume reached 600 mm³. Even in this setting, AFP-CAR T cells significantly inhibited the growth of tumors, causing a 53% reduction in tumor volume by 1 week after dosing (Fig. 4B).

We next evaluated the in vivo activity of AFP-CAR T cells in a SK-HEP-1-MG subcutaneous xenograft model in SCID-Beige mice. For these experiments, mock or AFP-CAR T cells were given via intratumoral or intravenous delivery. Similar to the results obtained with Hep G2 tumors, intratumoral injection of AFP-CAR T cells resulted in robust tumor growth inhibition (Fig. 4C). Approximately 72% reduction in tumor size relative to mock-treated mice was observed by the time control mice had to be euthanized (Fig. 4C). Interestingly, in this model, intravenous administration of AFP-CAR T cells resulted in a more immediate and pronounced tumor growth inhibition than seen in the Hep G2 model (Fig. 4D). Approximately 28% reduction in tumor size was observed by the time control mice had to be euthanized (Fig. 4D). These data clearly show that while intratumorally delivered AFP-CAR T cells significant inhibit tumor growth in vivo, the efficacy of AFP-CAR T cells administered intravenously is relatively weak in SCID-Beige mice.

Previous studies have demonstrated better efficacy with intravenous administration in NSG mice (38–40), which have defective macrophages and dendritic cells and no natural killer cells, thus increasing the likelihood for in vivo CAR T-cell survival. To test this, NSG mice were implanted subcutaneously with Hep G2 cells and treated intravenously with either mock or AFP-CAR T cells. Here, intravenous injection of AFP-CAR T cells rapidly and significantly inhibited tumor growth (Fig. 4E). Approximately 64% reduction in tumor size relative to mock-treated mice was observed by the time control mice had to be euthanized (Fig. 4E). Importantly, we were able to detect CD3⁺ T cells in Hep G2 tumors in these mice 4 weeks after T-cell dosing. Moreover, T cells were enriched in AFP-CAR–treated mice relative to controls (Supplementary Fig. S8A). This is in contrast to the low number of detectable CD3⁺ T cells observed in intravenously treated Hep G2 tumors from SCID-Beige mice (Supplementary Fig. S8B). Overall, these data show a marked improvement in CAR T-cell survival and antitumor activity following intravenous administration in NSG mice compared with SCID-Beige.

Finally, we tested whether serum AFP levels in tumor-bearing mice might serve as a pharmacodynamic marker for response to treatment. Blood was collected from Hep G2 tumor-bearing mice 4 weeks after a single intratumor injection of either mock, 10⁶, or 10⁷ AFP-CAR T cells, and AFP levels were analyzed by ELISA. While serum AFP levels were very high (∼2 mg/mL) in mock-treated mice, they were substantially lower (∼1 mg/mL) in mice treated with the low dose of AFP-CAR T cells and even further reduced (∼100 μg/mL) in mice treated with the high dose of AFP-CAR T cells, which experienced near complete tumor regression. Thus, serum AFP levels correlate well with tumor size in these models (R = 0.9391) and can potentially be used as a marker for tumor response (Fig. 4F).

Although rare, peritoneal dissemination of liver cancer occurs in a subset of patients who are left with very limited treatment options. Therefore, we tested the antitumor activity of AFP-CAR T cells in an established intraperitoneal liver cancer xenograft model. Luciferase-tagged Hep G2 cells (Hep G2-luc2) were implanted intraperitoneally (i.p.) and tumor burden was assessed weekly by measuring tumor-derived bioluminescence. One week after tumor implantation, animals were randomized on the basis of total bioluminescent flux into four groups: (i) untreated, (ii) intraperitoneal injection of 10⁷ mock T cells, (iii) intraperitoneal injections of 10⁶ AFP-CAR T cells, and (iv) intraperitoneal injection of 10⁷ AFP-CAR T cells per mouse.

Tumor burden measured via luciferase photon emission in mock-treated animals showed no difference from that observed in the untreated control group (Fig. 5A and B). In contrast, mice treated with AFP-CAR T cells at either 10⁶ or 10⁷ cells per mouse showed robust tumor regression as early as one week after the first injection, which persisted for the duration of the study (11 weeks). No dose-dependent antitumor activity was observed; indicating that 10⁶ total T cells per mouse exceeds the maximum efficacious dose in this model. Collectively, these data demonstrate that AFP-CAR T cells can successfully inhibit the growth of peritoneal disseminated tumors as well.

The recent progress with redirected T-cell therapies against hematologic malignancies has revitalized the field of cancer immunotherapy. Despite the excitement surrounding this technology, many aspects of CAR/TCR–based therapies remain to be optimized to harness the full curative potential of tumor-specific immune responses. This is particularly true for solid tumors, where T-cell therapies have yet to demonstrate a significant therapeutic impact.

One of the major factors hindering the success of T-cell therapies against solid tumors is the lack of targetable tumor-specific antigens. First, unlike the blood cell lineages, which express unique cell surface markers, antigens expressed in solid tumors are often found in a variety of normal tissues. While patients can survive ablation of the hematologic cell compartment due to on-target/off-tumor effects, this is not the case with solid tumor antigens. As seen with T-cell therapies targeting MAGE-A3, MART-1, CEA, or ERBB2, even trace amounts or transient expression of a target antigen in cells of vital tissues can have adverse consequences (41). Second, the majority of proteins (including mutated and viral oncogenes) identified as highly specific markers for solid tumors are intracellularly localized, and therefore inaccessible by conventional antibodies or CARs. To circumvent this, one alternative strategy is to select intracellular antigenic targets that are processed and presented as peptide products by class I MHCs on the surface of tumor cells. Indeed, in prior work, we developed a human “TCR-like” mAb against WT1, an intracellular oncoprotein that is overexpressed in a myriad of leukemias and solid cancers. This “TCR-like” mAb (ESK1) is specific for a 9-mer peptide of WT1 in complex with HLA-A*02:01 (24). Importantly, we demonstrated that ESK1, either as a naked antibody or incorporated into a bispecific antibody, shows potent in vivo antitumor efficacy against several human cancer xenografts (24, 42). However, whether such tumor-specific antigen products can be targets for CAR T-cell therapy against solid tumors has not yet been reported.

To address this, we employed our previously developed antibody discovery platform to identify and optimize a highly specific antibody against the AFP₁₅₈ peptide in complex with HLA-A*02:01 for the treatment of liver cancer, one of the most pressing unmet medical needs. Notably, we show that redirected T-cell therapy employing a second-generation CAR constructed with this antibody displayed potent antitumor activity in multiple preclinical models of AFP/HLA-A*02:01–expressing human liver cancer. Thus, by expanding the repertoire of targetable antigens to include cancer-specific intracellular or secreted protein products, we have overcome a major obstacle to successful targeting of solid tumors by CAR T-cell therapy. These findings will be applicable to many other intracellular antigens as solid tumor-specific targets.

Furthermore, by engineering an scFv specific for a peptide–MHC complex into a second-generation CAR platform, we circumvented some of the disadvantages associated with conventional CARs (which cannot target intracellular antigens) and TCRs. For one, the mispairing of engineered TCRs with endogenous TCRs is avoided, which greatly reduces the chances for off-target effects. In addition, unlike conventional engineered TCRs, which compete with endogenous TCRs for surface expression and require external costimulatory signals to mount a full immune response, second-generation CARs can effectively modulate T-cell expansion and persistence through their dual-signaling receptors (i.e., CD28/CD3ζ). Thus, this scFv-CAR design allows us to target intracellular and secreted antigens with exquisite selectivity while providing the potency typically achieved with CARs.

One of the major potential side effects of conventional CAR T -cell therapy, however, is cytokine-release syndrome, commonly referred to as “cytokine storm.” This is due, in part, to target selection, in that the majority (if not all) of current CARs target tumor antigens expressed at high levels on the surface of tumor cells, thus increasing the likelihood for overactivation of T cells and the release of toxic levels of cytokines. As AFP peptide–MHC complexes are present at low numbers on the surface of tumor cells (43), we anticipate that CAR T cells designed to target AFP peptide–MHC complexes may effectively kill tumor cells without the associated “cytokine storm”. In support of this, we found that while AFP-CAR T cells killed Hep G2 cells (low level of HLA-restricted AFP₁₅₈ complexes on their surface; Supplementary Fig. S9) just as potently as SK-HEP-1-MG cells (high level of HLA-restricted AFP₁₅₈ complexes on their surface; Supplementary Fig. S9), AFP-CAR T cells produced substantially lower levels of cytokines when incubated with Hep G2 cells compared with SK-HEP-1-MG cells. Thus, these data underscore a few key points: (i) CAR T cells designed with sufficiently high binding affinity to their target peptide–MHC complex can effectively kill tumor cells with very low expression of the antigen on their surface; (ii) the ability of CAR T cells to be a strong cytokine producer upon antigen encounter may not always be a reliable predictor of CAR T-cell efficacy; and (iii) CAR T-cell therapy–induced “cytokine storm” may be mitigated or prevented by targeting tumor-specific peptide–MHC complexes.

Finally, for redirected T-cell therapy to be successful in solid malignances such as HCC, T cells must be able to reach the tumor and overcome its immunosuppressive microenvironment. CAR T cells that are administrated intravenously are required to survive in circulation long enough to penetrate the tumor, engage their target antigen, and become activated. Despite these hurdles, we show that intravenously administered AFP-CAR T cells are able to traffic to subcutaneous liver tumors in NSG mice and inhibit their growth. However, intratumoral injection of AFP-CAR T cells caused a much more profound, rapid, and lasting antitumor response. Thus, our data make a strong case that local delivery of T-cell therapy may be a more attractive clinical path as it not only shortens the time between CAR T injection and antigen encounter, but is potentially safer than systematic delivery. The analogous routes of administration in humans would be direct injection into liver tumors or injection through the hepatic artery, both of which are commonly used in the clinic to administer chemotherapeutic agents to liver cancer patients (44). Intrahepatic artery is a particularly attractive delivery route as it infuses the entire liver vascular system with T cells which can reach multiple lesions. In support of this, Katz and colleagues used the intrahepatic artery for delivery of anti-CEA CAR T cells in a recent phase I trial and found that the engineered T cells were able to penetrate metastatic tumors in the liver (45). However, the low efficacy of anti-CEA CAR T cells observed in that trial was attributed to the presence of myeloid-derived suppressor cells (MDSCs; ref. 46). Thus, combining intrahepatic artery delivery with inhibitors of MDSCs and/or checkpoint inhibitor might be required to alleviate the immunosuppressive effects of the tumor microenvironment.

Given that the first criteria for a successful CAR T-cell–based therapy is specificity, our ability to generate fully-human antibodies specific for peptide–MHC complexes significantly increases the pool of targets for identifying safe and effective anticancer therapies. Our work demonstrates that targeting intracellular and secreted tumor antigen products with CAR T cells is feasible and efficacious, potentially safer, and has promising applications for future therapies against currently incurable cancers.

H. Liu, Y. Xu, J. Xiang, L. Long, Z. Yang, J. Lu, N. Cheng, B. Liu, S. Yan, P. Wang, L. Jin, V.W. Chan, and C. Liu hold ownership interest (including patents) in Eureka Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: H. Liu, Y. Xu, J. Xiang, L. Long, S. Yan, S.J. Priceman, C. Liu

Development of methodology: H. Liu, Y. Xu, J. Xiang, L. Long, Z. Yang, J. Lu, L.H. Horan, B. Liu, P. Wang, C. Liu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Xu, J. Xiang, L. Long, S. Green, Z. Yang, B. Zimdahl, J. Lu, N. Cheng, L.H. Horan, B. Liu, L. Jin, Y. Nakano, J.F. Morales, P. Zhang, L.-X. Liu, B.K. Staley, S.J. Forman

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Liu, Y. Xu, J. Xiang, L. Long, S. Green, Z. Yang, B. Zimdahl, J. Lu, N. Cheng, L.H. Horan, Y. Nakano, P. Zhang, L.-X. Liu, S.J. Forman, V.W. Chan, C. Liu

Writing, review, and/or revision of the manuscript: H. Liu, Y. Xu, J. Xiang, L. Long, S. Green, B. Zimdahl, S. Yan, J. Diaz, S.J. Priceman, C.E. Brown, S.J. Forman, V.W. Chan, C. Liu

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L.H. Horan, P. Wang, L. Jin, L.-X. Liu, V.W. Chan

Study supervision: H. Liu, L. Long, P. Wang

Other (advisement on study design): C.E. Brown

We acknowledge Steve Noonan (Murigenics) for the management of preclinical studies.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.