T cells engineered against a native antigen can surmount immunologic and physical barriers to treat pancreatic ductal adenocarcinoma

SUMMARY

Pancreatic ductal adenocarcinomas (PDA) erect physical barriers to chemotherapy and induce multiple mechanisms of immune suppression, creating a sanctuary for unimpeded growth. We tested the ability of T cells engineered to express an affinity-enhanced T cell receptor (TCR) against a native antigen to overcome these barriers in a genetically engineered model of autochthonous PDA. Engineered T cells preferentially accumulate in PDA and induce tumor cell death and stromal remodeling. However, tumor-infiltrating T cells become progressively dysfunctional, a limitation successfully overcome by serial T cell infusions that resulted in a near-doubling of survival without overt toxicities. Similarly engineered human T cells lyse PDA cells in vitro, further supporting clinical advancement of this TCR-based strategy for the treatment of PDA.

Graphical Abstract

INTRODUCTION

The rising incidence and unparalleled lethality of pancreatic ductal adenocarcinoma (PDA) continue to outstrip advances in treatment (Siegel et al., 2012). Despite significant genomic instability, PDA have relatively few coding mutations compared to more immunogenic cancers (Lawrence et al., 2013). The robust desmoplastic response characteristic of PDA includes multiple mechanisms of immunosuppression that impede endogenous CD8 T cell infiltration and activity (Bayne et al., 2012; Clark et al., 2007; Feig et al., 2013; Stromnes et al., 2014a; Stromnes et al., 2014b). The limited neo-antigenic landscape together with the suppressive tumor environment likely contribute to the failure to engender effective immunity with checkpoint blockade (Brahmer et al., 2012; Le et al., 2013; Royal et al., 2010).

Pancreas cancers impose multiple constraints against immunity, including physical barriers that compromise perfusion (Jacobetz et al., 2013; Olive et al., 2009; Provenzano et al., 2012) and may also impede T cell infiltration of the tumor bed. As a result, many conventional immune strategies have had limited effect on patient survival, suggesting that targeting multiple forms of immune suppression concurrently may be necessary. Nevertheless, some clinical and preclinical data suggest it is possible to engender endogenous T cell activity, however transient. For example, vaccines designed to induce Mesothelin (MSLN)-specific T cells prolonged survival in patients with advanced PDA (Le et al., 2015), and targeted depletion of myeloid derived suppressor cells (MDSC) evoked endogenous T cell activity against autochthonous PDA (Stromnes et al., 2014a). Modulating other stromal elements in PDA may increase the therapeutic activity of more conventional immunotherapies such as checkpoint blockade (Feig et al., 2013; Winograd et al., 2015). These studies highlight both the potential and the challenges to effectively engage cancer immunity. Moreover, should an immunotherapy be sufficiently robust to alter the clinical course of a carcinoma, safety becomes a counterbalancing concern, lest autoimmunity and organ dysfunction ensue.

Mutated proteins that are obligate for tumor survival represent ideal antigens for immune targeting. Attempts to generate T cells specific for common cancer mutations in PDA, such as KRAS^G12D, have rarely proven successful, presumably due to limited processing or weak binding to HLA alleles (Chaft et al., 2014). The most reliably present and targetable cancer antigens are aberrantly expressed, non-mutated self-antigens (Argani et al., 2001). However, effectively targeting these antigens is compromised by the fact that high affinity T cells specific to self/tumor antigens undergo deletion during development, sculpting a T cell repertoire with low affinity (Hogquist et al., 2005). To overcome this otherwise adaptive limitation in the natural repertoire, T cells can be genetically engineered with extremely high affinity against selected antigens (June et al., 2012; Stromnes et al., 2014c). Current strategies incorporate either synthetic chimeric antigen receptors (CAR) or cloned T cell receptors (TCR) and it remains to be determined which approach will be safest and best suited to treat solid malignancies (June et al., 2012; Kalos et al., 2011; Schmitt et al., 2013; Stromnes et al., 2014c).

The rigorous evaluation of the potency, safety and limitations of T cells engineered to target naturally occurring tumor antigens can be accomplished in models that faithfully recapitulate the human disease from inception to invasion. Among a panel of candidate tumor antigens that are aberrantly expressed in both human PDA and the genetically engineered Kras^LSL-G12D/+;Trp53^LSL-R172H/+;p48^Cre/+ (KPC) mouse model of the disease (Hingorani et al., 2003; Hingorani et al., 2005), MSLN emerged as the most attractive target. We engineered lymphocytes to express a MSLN-specific TCR with an affinity beyond that obtainable in the normal repertoire and tested their safety and ability to overcome considerable physiologic barriers to treat invasive PDA.

RESULTS

Candidate tumor antigen expression in normal tissues and murine and human PDA

To generate pure inbred KPC mice, each of the three requisite alleles, Kras^LSL-G12D, Trp53^LSL-R172H and p48^Cre, were sequentially backcrossed to B6 and the progress informed by detailed SNP allelotyping. The engineered loci were the only distinguishing genetic differences in the final strains, otherwise representing 100% purity. Pure B6 KPC mice stochastically developed pancreatic intraepithelial neoplasms (PanIN) that spontaneously progressed to invasive and metastatic PDA, as seen with the original model on a mixed 129Sv/Bl/6 background (Hingorani et al., 2003; Hingorani et al., 2005). The histopathology of the primary tumors revealed the glandular architecture expected of an adenocarcinoma, together with an abundant inflammatory infiltrate, dense ECM, and scattered compressed blood vessels, hallmarks of human PDA and the original KPC model (Figure 1A and see below). B6 KPC mice also developed liver, lung and diaphragm metastases characterized by a complex stromal response (Figure S1A).

Figure 1. Tumor antigen expression in murine and human PDA.

(A) Immunohistochemical analyses of target antigens in murine and human (hPDA) tissues. d, duct; ac, acini; i, islets; arrows, high expression; arrowheads, low-moderate expression; *, tumor epithelial cells. Scale bar, 25 μm.

(B) Immunofluorescence for MSLN and cytokeratin (CK) in indicated KPC tissues. Arrows, MSLN⁺CK⁺ cells. Scale bar, 25 μm.

(C) FACs plot of MSLN, MHC I and p53 in early passage (<3X) primary KPC PDA cells.

(D) Relative expression of indicated mRNAs in primary KPC PDA cells. Each point represents an independent cell preparation. Mean ± SEM.

(E) Immunoblot analyses of primary murine pancreatic ductal cells, primary preinvasive KC cells (Pre) and two independent invasive KPC PDA primary cell preparations.

See also Figure S1.

To inform the rational design of a T cell therapy, we first performed specific immunohistochemistry to assess a variety of potential antigenic targets overexpressed by both murine and human tumor epithelial cells in preinvasive, invasive and metastatic PDA (Figure 1A). Marked intra- and inter-tumoral heterogeneity in both preinvasive and invasive disease was observed across a number of antigens that are immunological targets in PDA including the Wilms’ tumor antigen (WT1) (Koido et al., 2014; Oji et al., 2004), MUC1 (Shindo et al., 2014) and Annexin A2 (ANXA2) (Zheng and Jaffee, 2012). WT1 was absent from normal pancreas and expressed primarily in stromal cells of preinvasive and invasive PDA. MUC1 was expressed in normal pancreas and PanIN and was heterogeneous in PDA and metastases. ANXA2 was expressed at low levels in normal pancreatic ducts and to a higher degree in PanIN, PDA and metastases. Other antigens, such as COX2, are either highly expressed in tumor cells as well as normal tissues, precluding safe immunological targeting; or, as with EGFR and Her2/Neu, are expressed more heterogeneously (Hingorani et al., 2005). In comparison, despite some variation in intensity, MSLN was reliably expressed in all preinvasive, invasive and metastatic PDA specimens examined (Figure 1A). MSLN staining also identified micrometastases (Figure S1B). MSLN-expressing cells were positive for cytokeratin (CK) (Figure 1B), consistent with a ductal phenotype, and negative for the activated fibroblast marker, αSMA (Figure S1C). MSLN⁺ cells in PDA were positive for Trp53 (Figures 1C and S1C), reflecting expression of stabilized point-mutant Trp53, and also expressed low levels of MHC class I which was induced by IFNγ (not shown), as seen in studies of human PDA (Pandha et al., 2007). MSLN mRNA isolated from primary tumor epithelial cells was highly expressed (comparable to the ductal marker CK19) and significantly higher than WT1 (Figure 1D), consistent with the immunohistochemical analyses.

Despite being expressed in some normal tissues (Chang and Pastan, 1996), mice that lack MSLN have no discernable phenotype (Bera and Pastan, 2000) revealing that it is not essential for life. We detected MSLN in the normal pleura and pericardium, as well as in rare cells in the thymus, raising the possibility of central and peripheral tolerance to this self-antigen (Figures S1D-S1F). Human and murine MSLN share 55% amino acid identity and have similar expression profiles in normal tissues (Bera and Pastan, 2000; Chang and Pastan, 1996). There are also no known species differences in the regulation or function of MSLN. In summary, the absence of detectable MSLN in normal pancreas, increase in expression with disease progression (Figure 1E), high expression in invasive PDA and metastases and the suggestion that MSLN may promote tumor invasion (Chen et al., 2013) all contribute to its attractiveness as an immune target (see also Pastan and Hassan, 2014).

Generation and safety of affinity-enhanced TCR cell therapy

We immunized B6 MSLN^−/− and wild-type (WT) mice with a recombinant adenovirus expressing murine MSLN (Ad-MSLN) to elicit reactive T cells. T cells specific for epitopes MSLN_343-351, MSLN_484-492, MSLN_544-552 and MSLN_583-591 were isolated from MSLN^−/− mice but not WT mice, consistent with central tolerance (Figure 2A). However, both MSLN^−/− and WT mice generated responses to MSLN_406-414, previously shown to be processed and presented by a B6 ovarian cancer cell line (Hung et al., 2007). MSLN_406-414-specific T cells isolated from WT mice uniformly expressed the Vβ9 TCR chain as did the majority of MSLN_406-414-specific T cells from MSLN^−/− mice (Figures S2A). Despite expressing similar levels of Vβ9, MSLN_406-414-specific T cell lines from MSLN^−/− mice stained brighter with tetramer, consistent with higher affinity (Figures S2A and S2B). MSLN^−/− MSLN_406-414-specific T cell clones also responded to lower antigen concentrations than the corresponding WT clones (Figure S2C). We therefore focused on the isolation and study of these MSLN_406-414 T cells as the murine surrogate for the potential isolation and subsequent genetic modification of a naturally occurring human TCR for clinical applications (June et al., 2012; Schmitt et al., 2009; Stromnes et al., 2014c).

Figure 2. Cloning and expression of the enhanced-affinity TCR₁₀₄₅ MSLN_406-414-specific TCR.

(A) Epitope mapping of MSLN-specific T cells derived from WT and MSLN^−/− mice.

(B) CDR3 sequences of Vα4 and Vβ9 chains cloned from the highest avidity MSLN_406-414-specific T cell clones isolated from WT and MSLN^−/− mice.

(C) Expression of MSLN_406-414-specific TCR derived from WT (TCR₇₄₃₁) or MSLN^−/− (TCR₁₀₄₅) mice in P14 T cells after 2 in vitro stimulations.

(D) Functional avidity of engineered T cells assessed by intracellular IFNγ (normalized to maximum response).

(E) Dissociation kinetics of tetramer binding.

(F) Tetramer binding by 58α⁻β⁻ cells expressing TCR₇₄₃₁ or TCR₁₀₄₅ with or without CD8 co-receptor.

(G) Apoptosis of MHC class I⁺ KPC tumor cells following incubation with engineered T cells.

(H) Residual adherent tumor cells following incubation (5 hr) with specified T cells.

(I) Number of live adherent tumor cells in (H) (assessed by trypan blue exclusion).

Data are shown as mean ± SEM. See also Figure S2.

Most T cell clones isolated from WT and MSLN^−/− mice used the same germline Vα4 and Vβ9 TCR chains, restricting any sequence differences between the highest affinity from the respective strains to CDR3 (Figure 2B), the region that directly contacts peptides (Jorgensen et al., 1992; Kelly et al., 1993). These results suggest a similar preferential docking geometry of TCR chains for recognition of this epitope. We inserted codon-optimized TCR chains into retroviral vectors for expression in transgenic P14 T cells that endogenously express a TCR specific for the LCMV gp33 epitope (Pircher et al., 1989). As expected, T cells expressing the highest affinity TCR isolated from MSLN^−/− mice (TCR₁₀₄₅) stained brighter with tetramer and responded to 10-fold lower antigen concentration than the highest affinity cells (TCR₇₄₃₁) from WT mice (Figures 2C and 2D). Tetramer decay kinetics confirmed the higher affinity of TCR₁₀₄₅ (Figure 2E).

The contribution of CD8 binding to MHC class I on target cells or tetramers can minimize differences in TCR affinities for peptide in MHC complexes (Daniels and Jameson, 2000; Denkberg et al., 2001; Garcia et al., 1996). To better assess affinity differences, we transduced the CD8⁻ 58 TCRαβ^−/− cell line (Letourneur and Malissen, 1989) with CD8α and β and expressed TCR₇₄₃₁ or TCR₁₀₄₅ in both CD8⁺ and CD8⁻ cells. TCR₁₀₄₅ bound tetramer independently of CD8, whereas TCR₇₄₃₁ binding required CD8 (Figure 2F), suggesting TCR₁₀₄₅ might also be useful in CD4⁺ T cells. TCR₁₀₄₅ T cells lysed KPC MHC class I⁺ tumor cells more effectively than TCR₇₄₃₁ T cells (Figures 2G-2I). TCR₁₀₄₅ therefore represents an “affinity-enhanced” TCR and models what may be achieved by engineering CDR3 to optimize human TCR (Schmitt et al., 2009).

The value of an affinity-enhanced TCR depends upon improved function without prohibitive toxicity. Extensive experiments were performed to verify the safety and activity of engineered T cells in mice. Adoptively transferred TCR₁₀₄₅ T cells were detected in normal tissues at low frequency (≤1% of CD45⁺ cells) and did not express activation markers (Figures S2D and S2E). Conditioning with cyclophosphamide, but not gemcitabine, increased the expansion and memory formation of engineered T cells (Figures S2F-S2I), perhaps reflecting a more complete depletion of endogenous immune cells (data not shown). Persisting donor T cells expanded and downregulated CD62L following vaccination (Figure S2J). Basal levels of MSLN expression in normal organs did not elicit self-reactivity even in the context of vaccine- and lymphopenia-induced activation/expansion and IL-2 administration (Figure S2K), as reflected by lack of infiltration, accumulation or tissue injury.

Tumor-specific accumulation, activity and suppression of TCR₁₀₄₅ engineered T cells

We transduced P14 T cells with either TCR₁₀₄₅ or a control TCR (TCR_gag) specific to a retrovirus gag epitope (Dossett et al., 2009) to formally assess activity. After two in vitro stimulations, transduced T cells uniformly exhibited an effector phenotype (Figure S3A). In vitro-expanded T cells also transiently expressed inhibitory receptors PD1, Tim3, and 2B4, which reflect activation but can be associated with dysfunction. However, the transduced T cells secreted IFNγ after antigen encounter (Figure 2D) and were therefore clearly functional, suggesting that the post-stimulation receptor profile reflected a transient response from TCR signaling rather than T cell exhaustion.

A series of pilot studies was performed to evaluate the in vivo efficacy of this therapy in KPC mice with a defined pancreatic tumor. Donor TCR₁₀₄₅ cells were detected in the lungs at 2 hr post transfusion (presumably in vascular capillary beds) and redistributed to the pancreas by 4 days (Figure 3A). Preferential accumulation in pancreatic tumors of TCR₁₀₄₅- versus TCR_gag-transduced T cells was observed (Figure 3B). Similar percentages and numbers of donor T cells were detected in the spleen irrespective of TCR specificity, whereas the frequency and number of intratumoral TCR₁₀₄₅ T cells were significantly higher than that of TCR_gag T cells (Figures 3C and 3D). TCR₁₀₄₅ T cells were distributed throughout the tumor bed, interspersed within the stroma as well as adjacent to epithelial cells (Figure 3E). Trp53⁺CK⁻ cells in the stroma could be found in close contact with TCR₁₀₄₅ cells (Figure 3F) and may reflect tumor cells that had undergone EMT: CK⁺ cells in glandular structures co-expressed the prototypic epithelial marker E-cadherin but Trp53⁺CK⁻ cells in the stroma did not (Figure S3B). As expected, Trp53⁺CK⁻ cells also did not express the myeloid marker, CD11b, or the endothelial marker, CD31 (Figures S3C and S3D).

Figure 3. Biodistribution and in vivo effects of genetically engineered TCR₁₀₄₅ T cells.

(A) Biodistribution of TCR₁₀₄₅ T cells 2 hr and 4 days post transfer into KPC mouse.

(B) Distribution of donor TCR₁₀₄₅ or TCR_gag cells in tissues ex vivo 8 days post transfer.

(C) Donor T cell frequency 8 days post transfer. Plots are gated on CD45⁺CD8⁺ cells.

(D) Number of donor T cells isolated from spleen (top) and tumor (bottom) 8 days post transfer.

(E) Immunofluorescence for CD8 and CK in primary tumors 8 days post transfer. Scale bar, 50 μm; arrows, CD8 T cells adjacent to epithelium.

(F) Immunofluorescence for indicated molecules in primary tumors 8 days post TCR₁₀₄₅ cell transfer. Arrows, CD8 T cells adjacent to p53⁺CK19⁺ tumor cells; arrowheads, p53⁺CK19⁻ cells in the stroma. Scale bar, 10 μm.

(G) IHC for cleaved-caspase 3 (CC3) in PDA at day 8. Arrows, CC3⁺ cells. Scale bar, 10 μm.

(H) Tumor apoptosis 8 and 28 days post T cell transfer.

(I) Histology of pancreatic tumors 8 days post T cell transfer. *, absence of interstitial pink (H&E), blue (Masson's trichrome) and red (Picosirius) stain reflects a loss in ECM collagen content. Arrowhead, infiltrating mononuclear cells. Scale bar, 50 μm.

(J) Quantification of collagen content in tumors from (I).

Data are shown as mean ± SEM. See also Figure S3.

Increased tumor cell apoptosis was observed 8 days following TCR₁₀₄₅ T cell infusion, but not at day 28, indicative of a specific, albeit transient anti-tumor effect (Figures 3G and 3H). TCR₁₀₄₅ T cells also caused marked stromal involution (Figures 3I and 3J), as seen after anti-CD40 administration (Beatty et al., 2011) or targeted depletion of myeloid cells (Stromnes et al., 2014a). The failure to sustain target cell apoptosis suggested that the donor T cells either did not persist in the tumor and/or lost function. At day 28, TCR₁₀₄₅ cells were rare in the spleen, thymus, bone marrow, blood, salivary gland and draining lymph node (dLN), yet remained enriched in PDA (Figures 4A and 4B, and not shown). Nevertheless, TCR₁₀₄₅ T cell numbers decreased ~6-fold in spleen but ~18-fold in PDA between 8 and 28 days (Figure 4C), revealing a selective disadvantage for T cell survival in the tumor and consistent with the short-lived fate of differentiated effector T cells (Berger et al., 2008; Kaech et al., 2003). Intratumoral and splenic TCR₁₀₄₅ T cells bound tetramer with similar affinity, indicating the tumor does not modulate TCR expression (Figure 4D). In contrast to either splenic TCR₁₀₄₅ or intratumoral TCR_gag cells, intratumoral TCR₁₀₄₅ cells expressed Ki67, CD25, 41BB and CD69, revealing TCR signaling from specific antigen recognition (Figures 4E and 4F). In contrast, down-regulation of CD27, CXCR3 and CD44 on intratumoral donor T cells was independent of antigen specificity (Figures 4E, 4F, S4A and S4B). Intratumoral TCR₁₀₄₅ cells also progressively and selectively upregulated PD1, Tim3, Lag3, and 2B4, which were not expressed on splenic TCR₁₀₄₅ cells (Figures 4G and 4H) or intratumoral TCR_gag cells (Figure S4B), and such PD1^high cells can be refractory to PD-L1 blockade (Blackburn et al., 2009).

Figure 4. Tumor-infiltrating TCR₁₀₄₅ T cells have a phenotypic signature consistent with antigen recognition and are dysfunctional.

(A) Donor TCR₁₀₄₅ cell frequency 28 days post T cell transfer. Plots are gated on CD45⁺CD8⁺ T cells.

(B) Donor TCR₁₀₄₅ cell frequency in spleens (Spl) and tumors 8 (D8) or 28 (D28) days post T cell transfer.

(C) Number of donor TCR₁₀₄₅ cells 8 or 28 days post T cell transfer..

(D) Tetramer staining of splenic and intratumoral donor TCR₁₀₄₅ cells 28 days post transfer.

(E) Phenotype of donor (CD8⁺Thy1.1⁺) TCR₁₀₄₅ cells compared to concurrently isolated splenic host (CD8⁺Thy1.1⁻) T cells 28 days post transfer.

(F) Frequency or MFI of donor T cells positive for indicated antigens in spleen and PDA. Mean ± SEM, n=3 each.

(G) Inhibitory receptor expression by donor TCR₁₀₄₅ cells and endogenous T cells.

(H) Frequency of donor TCR₁₀₄₅ cells expressing indicated molecules 28 days post T cell transfer.

(I) Ex vivo cytokine production by T cells in presence (+) or absence (−) of antigen 8 days post T cell transfer.

(J) Donor T cell frequencies producing both IFNγ and TNFα after 5 hr restimulation with antigen.

Data are shown as mean ± SEM. See also Figure S4.

A significantly lower fraction of intratumoral versus splenic TCR₁₀₄₅ T cells secreted IFNγ and TNFα in response to antigen (Figures 4I and 4J), consistent with chronic TCR signaling and subsequent inhibitory receptor expression. However, non-specific TCR_gag T cells isolated from tumors also had decreased function compared to their splenic counterparts (Figures 4I and 4J) revealing that T cell dysfunction in PDA is also, in part, independent of these inhibitory receptors and persistent TCR signaling. CD4⁺Foxp3⁺ regulatory T cells (Treg), myeloid-derived suppressor cells (MDSC) and numerous suppressive factors are enriched in PDA (Figures S4C and S4D) and may contribute to T cell suppression in the tumor microenvironment independent of TCR signaling.

A second infusion of TCR₁₀₄₅ engineered cells readily infiltrates PDA

We next examined whether PDA remain susceptible to a second infusion of TCR₁₀₄₅ cells, providing a ready clinical strategy to circumvent the observed loss of function over time. The serial infusions were distinguished by injecting Thy1.2⁺/1.2⁺ KPC hosts first with Thy1.1⁺/1.2⁺ TCR₁₀₄₅ T cells and infusing Thy1.1⁺/1.1⁺ TCR₁₀₄₅ T cells 20 days later (Figures 5A and 5B). The 2^nd infusion also preferentially accumulated in PDA compared to other tissues (Figures 5C and 5D). At 8 days after transfer, the 2^nd infusion of cells expressed less PD1, Tim3, and Lag3 in PDA and dLN compared to cells persisting from the 1^st infusion (Figure 5C and data not shown). The 2^nd infusion increased the total number of PDA-localized donor T cells 10-fold (data not shown) and they were also more functional after 8 days than cells persisting from the 1^st infusion in the same tumor (day 28) (Figure 5E).

Figure 5. A second infusion of engineered TCR₁₀₄₅ cells also preferentially accumulates in PDA.

(A) Schematic of congenically distinct T cell infusions into same KPC recipients. Cyclophosphamide (Cy) was administered prior to the 1^st infusion only.

(B) Circulating CD8 T cells prior to transfer (D0), 20 days following first infusion of TCR₁₀₄₅ Thy1.1^+/− cells (D20), and 6 days following second infusion of TCR₁₀₄₅ Thy1.1^+/+ cells (D26).

(C) Preferential accumulation and phenotype of second TCR₁₀₄₅ infusion in PDA (gated on CD8⁺Thy1.1⁺ cells). Concurrent PD1 expression on donor T cells from first (blue lines) and second (red lines) infusions. Grey histograms represent endogenous CD8⁺Thy1.1⁻ T cells.

(D) Ratio of persisting cells 8 days after second infusion. dLN, draining lymph node; Thy, thymus; SG, salivary gland.

(E) Cytokine production by donor T cells at day 28 (see (A)). Cells were stimulated together and plots gated on CD8⁺ Thy1.1^+/− (1st infusion) or CD8⁺ Thy1.1^+/+ cells (2^nd infusion).

Data are shown as mean ± SEM.

Serial T cell infusions promote the survival of mice with advanced PDA

Encouraged by the pilot studies, we conducted a randomized, blinded, placebo-controlled trial with overall survival (OS) as the primary endpoint in KPC mice with invasive PDA (Table S1). Secondary endpoints included objective response and tumor cell apoptosis. Mice were enrolled based on a defined tumor burden and randomized to receive either engineered TCR₁₀₄₅ or TCR_gag T cell infusions every 2 weeks. The majority (63%) of KPC mice receiving TCR₁₀₄₅ cells showed objective responses. In contrast, all of the TCR_gag cell recipients imaged serially showed progressive disease (Figures 6A and 6B). At necropsy, TCR₁₀₄₅-recipient mice had noticeably hemorrhagic tumors (Figure 6C); an influx of mononuclear cells and loss of collagen were also observed, even in regions deep within the tumor bed (Figures 6C and S5A). Intratumoral vascular density was similar between control and TCR₁₀₄₅ cohorts, indicating that T cell therapy did not induce angiogenesis (Figures 6C and S5B). However, TCR₁₀₄₅ cell therapy did significantly increase vessel patency (Figure 6D). Prolonged T cell therapy did not induce detectable pleural or pericardial pathology (Figure S5C) and the number of donor T cells, endogenous CD8⁺ T cells and myeloid cells in these locations were also similar between the two cohorts (Figure S5D-F), underscoring the safety profile of this approach.

Figure 6. Serial infusions of engineered TCR₁₀₄₅ T cells significantly prolongs survival of KPC mice with established PDA.

(A) High-resolution ultrasound images of pancreatic head mass before (−) and after (+) TCR_gag or TCR₁₀₄₅ cell therapy. D, duodenum; pv, portal vein; K, kidney; ivc, inferior vena cava.

(B) Waterfall plots of best observed response by serial imaging (results confirmed by two investigators).

(C) Gross and immunohistochemical analyses of PDA following T cell therapy. Arrows, blood flow and patent blood vessels; *, mononuclear cell infiltrate; scale bar, 25 μm.

(D) Mean vessel diameter (MVD) in PDA.

(E) MSLN expression in PDA following serial T cell infusions. Scale bar, 25 μm.

(F) MSLN staining intensity in primary tumors and metastases following T cell therapy. Each dot represents a primary tumor or metastasis.

(G) Dual immunofluorescence for apoptosis (CC3) in PDA epithelial cells (CK) following T cell therapy. Arrows, CK⁺CC3⁺ cells. Scale bar, 50 μm.

(H) Quantification of data in (G).

(I) Expression of 41BB by donor TCR_gag and TCR₁₀₄₅ cells. Points represent individual animals.

(J) Survival of KPC mice with invasive disease that received serial TCR_gag (n=16) or TCR₁₀₄₅ (n=15) T cell therapy (54 days vs 96 days, respectively; p<0.0001).

Data are shown as mean ± SEM. See also Figure S5 and Table S1.

The intensity of MSLN expression in primary PDA decreased in the majority (6/10) of TCR₁₀₄₅ cell recipients evaluated but remained high in all (6/6) TCR_gag recipients examined, consistent with selection for tumor cell variants expressing lower levels of target antigen (Figures 6E and 6F). Regions of marked tumor epithelial cell apoptosis in TCR₁₀₄₅ T cell recipients were also observed (Figures 6G and 6H) and a higher frequency of intratumoral TCR₁₀₄₅ cells expressed activation markers 41BB, CD69 and Ki67 (Figure 6I and data not shown). In comparison, we did not detect significant differences in myeloid (CD11b⁺) or endothelial (CD31⁺) cell apoptosis between the cohorts (Figure S5G). Endothelial cells were rare, however, in areas of high tumor cell apoptosis in TCR₁₀₄₅ recipients (Figure S5H) and their loss may have been secondary to tumor destruction. A modest but significant increase in apoptosis of αSMA⁺ fibroblasts was detected in TCR₁₀₄₅ cell recipients (Figures S5I and S5J), which may contribute to the decreased collagen content. Since αSMA⁺ fibroblasts do not appear to express MSLN (Figure S1C), the stromal cell apoptosis could reflect ligation of death receptors, loss of paracrine signaling from tumor epithelial cells, or MSLN uptake and cross-presentation.

A significant and preferential increase in intratumoral TCR₁₀₄₅ vs. TCR_gag cell frequency was observed compared to the spleen (Figure S5K). Fewer tumor-infiltrating TCR₁₀₄₅ cells persisted after the final infusion than after the first infusion (data not shown), but only the first infusion was preceded by lymphodepletion and this difference could reflect the absence of induced cytokines that promote T cell proliferation and survival.

Trends toward decreased numbers of animals with metastatic disease (64% in TCR_gag vs. 46% in TCR₁₀₄₅ cohort) and malignant ascites (43% vs. 15%) were also observed (Table S1), consistent with the observed specific TCR₁₀₄₅ cell infiltration into metastases, expression of markers induced by antigen recognition and increased apoptosis (Figures S5L-S5N). Median OS from time of enrollment nearly doubled from 54 days for TCR_gag recipients to 96 days for the TCR₁₀₄₅ T cell group (Figure 6J), on par with or exceeding previous therapies tested in KPC mice, all of which depended on cytotoxic chemotherapy (Olive et al., 2009; Provenzano et al., 2012; Winograd et al., 2015). Thus, we demonstrate an effective T cell based therapy for invasive PDA that is safe and also circumvents the need for non-specific and toxic chemotherapy or radiotherapy.

Cloning and evaluation of human MSLN-reactive TCR

Appropriate target epitope selection is critical for the successful translation of T cell therapy (Chapuis et al., 2013; Stromnes et al., 2014c). To isolate receptors for patient treatment, we attempted to expand human MSLN-reactive T cells from normal donors to several peptides previously identified as HLA-A1 and HLA-A2 restricted epitopes (Table S2). We were able to generate T cell lines from two previously described, dominant HLA-A2-restricted epitopes, MSLN_20-28 and MSLN_530-538 (Thomas et al., 2004). MSLN_20-28 and MSLN_530-538-specific T cells are associated with improved outcomes in vaccinated PDA patients (Le et al., 2015; Thomas et al., 2004) suggesting that these epitopes are endogenously processed and presented and that the corresponding T cells have therapeutic activity. We cloned these T cells and measured tetramer binding (Figure 7A). The highest affinity MSLN_20-28-specific T cell clone had ~10-fold increased sensitivity to antigen compared to the highest affinity MSLN_530-538-specific clone (Figure 7B). The TCR that stained brightest with tetramer and had the highest functional avidity were cloned, sequenced, codon-optimized, inserted into lentiviral vectors and expressed in human T cells. The MSLN_20-28-specific TCR bound tetramer independently of CD8 whereas MSLN_530-538-specific TCR required CD8 (Figure 7C). The MSLN_20-28 and the MSLN_530-538 TCR were also expressed in primary human CD8 T cells, indicating successful competition with the endogenous TCR for expression (Figure 7D). The human Panc-1 PDA cell line expresses both MSLN (Figure 7E) and HLA-A2, and the latter could be increased by IFNγ (Figure 7F). Human CD8 T cells transduced to express either the MSLN_20-28- or MSLN_530-538-specific TCR specifically lysed HLA-A2⁺ MSLN⁺ Panc-1 cells, confirming their ability to recognize endogenously processed and presented antigen and kill tumor cells (Figure 7G). These receptors provide the essential substrate for genetic modifications of the CDR3 domains to produce an affinity enhanced TCR and translate our findings to patients with MSLN⁺ malignancies.

Figure 7. Isolation, expression and characterization of cloned human TCR specific to MSLN epitopes.

(A) Tetramer staining intensities of independently derived HLA-A2-restricted human T cell clones with the indicated specificities. Representative of 10-12 clones analyzed per epitope.

(B) MSLN_20-28 and MSLN_530-538-specific T cell clone lysis of HLA-A2⁺ T2 target cells loaded with titrating concentrations of peptide.

(C) Tetramer staining of CD8⁻ and CD8⁺ Jurkat cells transduced with MSLN_20-28- or MSLN_530-538-specific TCR.

(D) Tetramer staining of primary human CD8 T cells transduced with highest affinity MSLN_20-28 or MSLN_530-538-specific TCR.

(E) MSLN expression in human pancreatic ductal epithelial (HPDE) and Panc-1 cell lines (normalized to GAPDH).

(F) HLA-A2 expression by Panc-1 cells in the presence (+) or absence (−) of IFNγ.

(G) Lysis of HLA-A2⁺MSLN⁺ Panc-1 cell line by human CD8 effector T cell clones at indicated effector:target (E:T) ratios.

Data are shown as mean ± SD. See also Table S2.

DISCUSSION

The relative paucity of endogenous CD8 T cells in PDA has been attributed to the presence of immunosuppressive cells (Clark et al., 2007; Feig et al., 2013; Stromnes et al., 2014a) and elevated interstitial pressures. We show here that affinity-enhanced TCR T cells can nevertheless effectively infiltrate PDA, modify the ECM, induce tumor cell death and significantly prolong the survival of KPC mice with established invasive disease. The preferential accumulation of engineered TCR₁₀₄₅ cells in PDA indicates that energy-dependent T cell trafficking can overcome biophysical barriers to the passive transport of small molecules (Jacobetz et al., 2013; Provenzano et al., 2012) and overcome immunosuppression sufficiently to kill tumor cells. T cell-mediated tumor rejection has largely been studied with transplantable tumor models or artificially overexpressed antigens; however, such systems may not accurately reflect the breadth of obstacles to clinical translation, as evidenced by SV40-driven models of gastric and prostate cancer in which adoptively transferred T cells induced regression of engrafted tumors but not of the equivalent autochthonous disease (Bourquin et al., 2010; Chou et al., 2012). In a retrovirally-induced, Kras-driven lung cancer model engineered to overexpress ovalbumin, endogenous (naïve) ovalbumin-specific T cells delayed cancer progression but resistant tumors emerged that had lost the model antigen, highlighting a problem with artificially introduced antigens (DuPage et al., 2011). Thus, previous studies in transplantable or autochthonous solid tumor models have not adequately illuminated how to incorporate effector T cells specific to naturally expressed self/tumor antigens.

The adoptive T cell therapy described here can provide sustained clinical benefit and animals were treated for months without apparent on- or off-target toxicities. Unlike targeted ablation of stromal fibroblasts in autochthonous PDA which appeared to unleash a more aggressive disease (Ozdemir et al., 2014; Rhim et al., 2014), the T cell therapy presented here remodeled the stroma, including death of fibroblasts, and still conferred a survival benefit. Enzymatic degradation of hyaluronan in combination with gemcitabine also significantly increased survival in KPC mice and was accompanied by the loss of both tumor epithelial cells and activated fibroblasts (Provenzano et al., 2012). Thus, stromal remodeling in certain contexts can be beneficial.

T cell-mediated remodeling of the matrix was antigen-dependent, raising the question of which cell type(s) present MSLN to donor T cells in PDA. The increased apoptosis of fibroblasts with TCR₁₀₄₅ T cells suggests that MSLN may be cross-presented by stromal cells (Qin and Blankenstein, 2000; Spiotto et al., 2004; Zhang, 2008). Alternatively, direct lysis of tumor epithelial cells could have resulted in the loss of critical paracrine factors that promote stromal cell survival and/or function, indirectly decreasing collagen content. Understanding the precise mechanisms of antigen recognition and cell destruction in PDA has clinical implications, particularly as MSLN is abundantly shed from the cell surface (Pastan and Hassan, 2014) and could be taken up by neighboring cells. In this regard, a TCR-based therapy may have an advantage over MSLN-targeting CAR currently in clinical trials (Beatty et al., 2014), as the latter would not recognize cross-presented antigen and might also be inhibited by shed soluble antigen. A greater understanding of stromal contributions to anti-tumor T cell activity will undoubtedly inform future studies and refinements of this strategy.

Engineered T cells became progressively dysfunctional in PDA and upregulated the inhibitory receptors PD1, Lag3, Tim3 and 2B4, reflecting chronic TCR signaling. They did not express detectable levels of Foxp3, a transcriptional factor expressed in Treg, either prior to transfer or after PDA infiltration (data not shown). Since the majority of the PD1⁺ T cells co-expressed additional inhibitory receptors, determining if blockade of any of these pathways alone or in concert increases intratumoral T cell function is a cogent next step. We note that control TCR_gag T cells recovered from tumors also exhibited dysfunction. Thus, a component of the observed diminished T cell activity was independent of antigen recognition and the inhibitory receptor profile, highlighting a separate contributing role of the immunosuppressive TME. Together with the limited neo-antigen landscape in PDA (Lawrence et al., 2013), these results may help explain why PD1/PDL1 pathway blockade has not been effective in PDA patients, despite showing marked results against some malignancies (Brahmer et al., 2012; Topalian et al., 2012). Maximizing engineered T cell activity may require overcoming inhibitory pathways downstream of persistent TCR signaling as well as relieving the generalized immunosuppression that is independent of TCR specificity.

That affinity-enhanced TCR therapy selects for tumor cells expressing lower levels of MSLN highlights both the selective pressure that can be achieved with such an approach and a potential limitation of targeting only a single antigen. That T cells nevertheless continued to be activated in tumors despite decreased MSLN expression also emphasizes the potential value of a high affinity TCR. Our murine data reveal that many epitopes within MSLN are poorly immunogenic, emphasizing the importance of epitope selection when targeting antigens. The MSLN_20-28 and MSLN_530-538 epitopes are both immunogenic in humans and can elicit T cells that are associated with increased survival in PDA patients. The human MSLN_20-28-specific TCR we describe here also binds tetramer avidly and independently of CD8 and therefore represents an excellent starting point for initial trials. Understanding how to further refine TCR engineering to be more effective and how best to combine engineered T cells with agents that modulate the TME, including strategies to increase antigen presentation, may greatly expand their therapeutic potential. Targeting additional tumor antigens may also prove valuable for achieving more complete and durable responses. Nevertheless, the results reported here, including objective response and increased survival with T cells alone – and in an autochthonous model that may be incurable due to inherently multifocal disease and continual regeneration of new tumors – strongly support efforts to begin translating genetically engineered TCR-based cell therapy targeting known human tumor antigens to patients with PDA and other solid tumors.

EXPERIMENTAL PROCEDURES

Mouse Strains

The Institutional Animal Care and Use Committees of the Fred Hutchinson Cancer Research Center and the University of Washington approved all animal studies. Kras ^LSL-G12D/+;Trp53^LSL-R172H/+;p48^Cre (KPC) mice have been previously described (Hingorani et al., 2005). To place the alleles on a pure defined background, each strain was independently serially back-crossed to C57Bl/6 H-2^b genetic background informed by SNP allelotyping (RADIL, IDEXX-Bioresearch). Final enrichment of the B6 genetic backgrounds was assessed at 1449 genomic SNPs using GoldenGate Genotyping Assays (Illumina, Inc.) at the DartMouse Speed Congenic Core Facility (Dartmouth Medical School). Raw SNP data were analyzed with SNaP-MAP™ and Map-Synth™ software. Both female and male mice were included. MSLN^−/− mice were generously provided by Ira Pastan (NCI, Bethesda, MD) and previously described (Bera and Pastan, 2000). P14 mice have been previously described (Pircher et al., 1989).

Human specimens

All studies using human specimens were approved by the Fred Hutchinson Cancer Research Center Institutional Review Board and conducted according to the principles expressed in the Declaration of Helsinki. Tumor tissues were obtained from patients who provided written informed consent by NWBioTrust (Department of Pathology, University of Washington, Seattle, WA).

Cloning mouse MSLN_406-414-specific TCR and generation of plasmids

The Vα4 and Vβ9 TCR chains from the highest affinity wild-type and MSLN^−/− T cell clones were cloned using RACE PCR, codon optimized (Invitrogen), linked by a porcine teschovirus-1 2A element and inserted into the Mig-R1 retroviral vector, as previously described (Schmitt et al., 2013).

Statistics

The Kolmogorov-Smirnov test was used to test whether data met the assumption of normality. The Students’ t test was used to compare normally distributed two-group data. Multigroup data were analyzed using 2-way ANOVA followed by a Tukey post test to correct for multiple comparisons. All data are shown as mean ± SEM unless otherwise indicated. A Fisher's exact test was used to compare the difference in the frequency of events (e.g. metastases, ascites). Kaplan-Meier survival data were analyzed using a Log-rank (Mantel-Cox) Test. Unless otherwise indicated, symbols indicate statistical significance as follows: *, p<0.05; **, p<0.005; ***, p<0.0005.

SIGNIFICANCE.

Immune therapies hold considerable promise for the treatment of cancer. Although notable successes have been achieved in hematologic malignancies, progress in solid tumors has been more elusive. Pancreatic ductal adenocarcinomas present an especially formidable challenge given the robust desmoplasia that accompanies disease progression, creating barriers to both drug perfusion and anti-tumor immunity. We show here that T cells engineered to express affinity-enhanced T cell receptors reactive to Mesothelin, a native tumor antigen, can transiently overcome both the inordinately elevated interstitial pressures and the multiple modes of immune suppression to specifically infiltrate PDA and induce tumor cell death. Serial adoptive transfers of such engineered T cells can be given safely and significantly increase overall survival.