Enhancing the Potency and Specificity of Engineered T Cells for Cancer Treatment

Recent advances in T-cell engineering, particularly with chimeric antigen receptors (CAR), have demonstrated the therapeutic potential of adoptively transferred T cells, which are able to recognize and kill tumor targets. However, few antigens are absolutely tumor-specific, resulting in “on-target, off-tumor” toxicities. This phenomenon is particularly problematic when using later-generation CARs whose costimulatory endodomains induce local T-cell proliferation and persistence. These undesirable effects may be tolerable, for example, when targeting a restricted antigen such as CD19, which is expressed on malignant and normal B cells and results in B-cell aplasia (1, 2). However, with other targets (e.g., CAIX, ref. 3; and HER2, ref. 4), the side effects can be life-threatening. Hence, CAR T-cell therapy would be more broadly applicable if the infused cells could more reliably discriminate between normal and malignant tissue.

A number of strategies have been explored to enhance the tumor selectivity of transgenic T cells. For example, Kloss and colleagues developed a cooperative model whereby signals for T-cell activation and costimulation were split between two different CARs coexpressed on the same cell to promote antitumor effects only upon dual target engagement (5). Roybal and colleagues developed an inducible system based on a synthetic Notch (synNotch) receptor circuit, whereby engagement with one tumor antigen induced expression of a second CAR, resulting in potent T-cell activation only in the presence of both targets (6). Whereas both of these strategies seek to regulate when transgenic T cells get switched “on,” Fedorov and colleagues have explored an approach to turn T cells “off” at sites other than the tumor by pairing a stimulatory (tumor-directed) CAR with an inhibitory CAR (iCAR) directed to normal tissue in order to limit T-cell activation outside of the tumor (7).

We have extended the concept of pattern recognition by conferring engineered T cells with the ability to recognize not just tumor-expressed antigen(s) but rather an expression pattern that is unique to the tumor site. We have used Boolean “AND” operator logic and modified our cells with 3 individual receptors capable of recognizing independent signals [prostate stem cell antigen (PSCA), TGFβ, and IL4] present at the pancreatic tumor site and transmitting signals for activation (signal 1), costimulation (signal 2), and cytokine support (signal 3). We now show the enhanced potency and increased tumor selectivity and safety of these tumor-specific molecular-pattern activated and regulated T (SmarT) cells in vitro and in vivo.

To selectively target pancreatic cancer, we first identified a genetic pattern exclusive to the tumor site. This included the tumor-associated antigen PSCA (8) and the immunoinhibitory cytokines TGFβ (9) and IL4 (ref. 10; Supplementary Fig. S1), all of which have been independently correlated with disease progression (11–13). To harness these ligands in a manner that would maximize tumor selectivity, we generated 3 retroviral vectors, each specific for one of the targets, and whose endodomains delivered independent intracellular signals [signal 1, activation (T-cell receptor [TCR] ζ chain); signal 2, costimulation (41BB); signal 3, cytokine (IL7)] to transgenic T cells (Fig. 1A–C, top).

To deliver signal 1 to T cells, we utilized a humanized, codon-optimized, first-generation CAR (1G.CAR) targeting PSCA that could be stably expressed on the surface of activated T cells (Fig. 1A, bottom, representative donor; mean 97% ± 1%). Interaction of CAR T cells with their cognate antigen, PSCA, induced a gene-expression profile consistent with TCRζ chain signaling in transgenic T cells with upregulation of genes (IRAK1, DUSP6, TANK, CHUK/IKKα, IL2Rα, and IL21) synonymous with pathways associated with lymphocyte and immune response activation as well as cell adhesion [CAR vs. control (ΔCAR) T cells; Fig. 1D; Supplementary Table S1; refs. 14–17]. Next, to provide signal 2 (costimulation), we designed a hybrid cytokine receptor, “TBBR,” by fusing the TGFβ receptor II (TGFβRII) exodomain with the endodomain of 41BB, a member of the tumor necrosis factor (TNF) receptor superfamily whose signaling in T cells prevents apoptosis, ameliorates exhaustion, and enhances persistence. Consistent with this profile, transgenic expression of TBBR on T cells (Fig. 1B, bottom, representative donor; mean 87% ± 5%) and exposure to TGFβ resulted in upregulation of prototypic genes associated with 41BB signaling including BCL2, NLRP3, and DUSP4, which regulate the NF-κB and TNF superfamilies, and downregulation of proapoptotic and glycolytic pathway-associated genes including TP53, CCL4, and G6PD (TBBR vs. ΔTGFβRII; Fig. 1E; Supplementary Table S2; refs. 18–21). Finally, to deliver signal 3 to T cells, we generated an inverted cytokine receptor (ICR) containing the IL4R exodomain fused to the IL7R endodomain (4/7 ICR; Fig. 1C, bottom, representative donor; mean 67% ± 14%). As expected, exposure to IL4 increased expression of STAT5A/B target genes, including SOCS1, BCL2L1, CXCR4, CCL2, and CDKN1A, which are characteristic of the IL7 receptor signaling pathway, and downregulation of STAT1, STAT6, and NOTCH1 in 4/7 ICR transgenic T cells (Fig. 1F; Supplementary Table S3; refs. 22–28).

Next, to determine whether independent T-cell inputs could stimulate their cognate receptors and deliver unique benefits based on their signaling endodomains, we evaluated the cytolytic function (chromium release assay), viability (Annexin/7AAD staining), and expansion (cell counting by Trypan blue exclusion) of CAR-, TBBR-, and 4/7 ICR-modified T cells (Fig. 1G–I). As expected, only T cells expressing the CAR could kill CAPAN-1, a PSCA-expressing pancreatic tumor cell line (71% ± 2% specific lysis—CAR vs. 2% ± 2%—TBBR vs. 2% ± 0.5%—4/7 ICR, 20:1 E:T ratio; Fig. 1G). Similarly, the transgenic TBBR switch receptor was able to harness TGFβ to provide a costimulatory signal, resulting in increased cell viability (14% ± 3% annexin⁻/7AAD⁻ cells—CAR, 35% ± 7%—TBBR, and 11% ± 6%—4/7 ICR, day 5; Fig. 1H), while forced expression of 4/7 ICR improved the proliferative capacity of transgenic cells exposed to IL4 (0.5 ± 0.3 fold change in cell numbers—CAR, 0.5 ± 0.3—TBBR, and 9 ± 3—4/7 ICR, day 14 vs. day 0; Fig. 1I). Figure 1G–I contains representative donor and summary data, top and bottom, respectively. In summary, therefore, independent expression of CAR, TBBR, and 4/7 ICR allows T cells to engage with a tumor molecular pattern (PSCA, TGFβ, and IL4) and activate signals for tumor recognition (signal 1), T-cell survival (signal 2), and T-cell proliferation (signal 3).

To discover whether coexpressing the 3 transgenic receptors could be additive and result in superior performance at the tumor site (Fig. 2A), we transduced T cells with CAR, TBBR, and 4/7 ICR to generate SmarT cells (Fig. 2B; 72% triple-positive cells, representative donor; mean 61% ± 5%). To confirm that each of the receptors signaled appropriately, we evaluated the genetic profile of SmarT cells in conditions that recapitulated the tumor microenvironment (PSCA, TGFβ, and IL4). As shown in Fig. 2C, prototypic signatures associated with CAR (IL2Rα, TANK, CHUK/IKKα, and IL21), TBBR (BCL2, DUSP4, and G6PD), and 4/7 ICR (CCL2, SOCS1, CDKN1A, and BCL2L1) signaling were conserved in the SmarT cells when compared with control T cells modified to express delta constructs (Supplementary Table S4). In addition, the functionality of the individual receptors was retained, enabling SmarT cells to recognize and kill PSCA⁺ targets (68% ± 2% specific lysis, 20:1 E:T ratio), as well as survive and expand in a TGFβ- and IL4-rich milieu, respectively (29% ± 1% annexin⁻/7AAD⁻ cells, day 5, and 55 ± 5 fold change in cell numbers between days 0 and 14, respectively; representative donors, left; summary data, right; Fig. 2D).

To assess the consequences of exposing traditional CAR (expressing CD3ζ endodomain) and SmarT cells to tumor-mimicking conditions, we next compared the RNA expression profile of transgenic cells following exposure to antigen, TGFβ, and IL4 (Supplementary Fig. S2). CAR T cells exposed to the suppressive tumor milieu upregulated genes associated with the native TGFβ (SMAD2, RUNX1, EGR2, and PLAU) and IL4R pathways (BCL6, CCR4, PRKCD, and CCL1; refs. 26, 29–33). In contrast, SmarT cells displayed a reciprocal gene-expression pattern with decreased expression of the same native TGFβ and IL4R target genes, and upregulation of genes that are characteristic of 41BB and IL7R signaling pathways, including BCL2, CCL2, and CDKN1A (Supplementary Fig. S2 and Supplementary Table S5; refs. 18, 25, 26). Consistent with this profile, CAR T cells failed to expand in the presence of PSCA, TGFβ, and IL4, unlike SmarT cells that were able to thrive in tumor-mimicking conditions (fold change 9 ± 4 vs. 1,005 ± 101—CAR vs. SmarT cells, day 28, respectively; Fig. 2E), and although no differences were observed in the activation, exhaustion, and memory profile between CAR T cells versus SmarT cells [representative forward scatter (FS) vs. side scatter (SS) plot, top, Fig. 2E; CD25 (31% ± 7% vs. 45% ± 6%), CD69 (83% ± 2% vs. 91% ± 4%), PD-1 (34% ± 6% vs. 32% ± 13%), TIM3 (17% ± 4% vs. 25% ± 6%), CD45RO⁺/CCR7⁻ (84% ± 4% vs. 96% ± 1%); CAR vs. SmarT cells, day 28; Fig. 2F], the triple-modified T cells secreted higher levels of effector cytokines as shown in Fig. 2G (Granzyme A: 46,224 ± 5,619 vs. 136,075 ± 50,492 pg/mL; Granzyme B: 5,732 ± 1,489 vs. 14,121 ± 1,450 pg/mL; perforin: 1,399 ± 440 vs. 6,968 ± 1,297 pg/mL; IFNγ: 3,945 ± 1,984 vs. 7,393 ± 2,738 pg/mL, CAR vs. SmarT cells, day 21). Subsequently, when cocultured with target cells producing TGFβ and IL4 (79 ± 4 ng/mL and 32 ± 3 ng/mL, respectively, Supplementary Fig. S6) at a 1:25 ratio (E:T), SmarT cells expanded significantly (17 ± 4 × 10⁴ on day 6) when compared with CAR T cells alone (0.1 ± 0.004 × 10⁴ on day 6). This improved SmarT-cell expansion resulted in an enhanced antitumor effect (57 ± 7 × 10⁴ vs. 3 ± 0.4 × 10⁴ tumor cells in the CAR versus SmarT-cell condition, respectively, on day 6; Fig. 2H and I, representative donor and summary results).

Having confirmed the short-term (6-day) effector activity of SmarT cells, we next evaluated the behavior of each transgenic subpopulation [single (CAR only), dual (CAR.TBBR and CAR.4/7 ICR), and triple transgenic T cells; Fig. 3A] after long-term (35 days) exposure to tumor milieu conditions. As shown for a representative donor in Fig. 3B and detailed for 3 donors in Fig. 3C, exposure to the tumor expression pattern positively selected for triple-transgenic SmarT cells whereas cells that lacked expression of any one of the transgenes did not persist. This positive selection occurred rapidly, with a substantial enrichment of SmarT cells by day 14 [CAR only, 2% ± 2% transgenic cells, 0.1 ± 0.01 × 10⁶ absolute cell numbers; CAR.TBBR, 4% ± 1%, 0.4 ± 0.1 × 10⁶; CAR.4/7 ICR, 6% ± 2%, 0.5 ± 0.2 × 10⁶; SmarT cells, 88% ± 4%, 8 ± 3 × 10⁶; Fig. 3D). Next, to assess whether this enrichment produced a T-cell population capable of antigen- or cytokine-independent growth, we exposed the selected SmarT cells to each of the independent input signals (signal 1, PSCA antigen; signal 2, TGFβ; signal 3, IL4). As shown in Fig. 3E, the presence of antigen alone or either cytokine alone was insufficient to promote the growth of these selected transgenic T cells, highlighting their dependence on all 3 input signals [fold change during week 4 of T-cell expansion: 0 ± 0 (PSCA), 0 ± 0 (TGFβ), 0.7 ± 0.2 (IL4), and 7.3 ± 0.4 (PSCA + TGFβ + IL4)]. Of note, transgenic subsets expressing (i) TBBR only, (ii) 4/7 ICR only, and (iii) TBBR.4/7 ICR that lacked CAR expression (i.e., absence of signal 1) were also unable to expand, as shown in Supplementary Fig. S3.

Having evaluated the pattern recognition, function, and safety profile of SmarT cells, we next analyzed the basis for the cells’ superior antitumor activity. We compared the expansion profile of SmarT cells maintained under standard Th1 culture conditions (weekly antigen + IL2) with those in which all 3 transgenic receptors were activated (weekly antigen + TGFβ + IL4). As shown in Fig. 4A and B, activation of all 3 transgenic receptors resulted in a growth pattern and activation/memory profile that was similar to the cells maintained in standard IL2 culture conditions [238 ± 32 vs. 363 ± 95 fold change, day 35 vs. day 0; CD25 (33% ± 12% vs. 38% ± 11%), CD69 (73% ± 2% vs. 86% ± 2%), PD-1 (20% ± 2% vs. 27% ± 11%), TIM3 (32% ± 17% vs. 56% ± 16%), CD45RO⁺/CCR7⁻ (94% ± 1% vs. 95% ± 2%); tumor milieu vs. IL2 conditions]. However, transgenic receptor signaling in SmarT cells did produce an increase in CD4⁺ T-cell content (40% ± 14% vs. 1% ± 1%, PSCA + TGFβ + IL4 vs. PSCA + IL2, respectively, P < 0.05), which were negative for regulatory T-cell markers, assessed by CD4/CD25/FOXP3 staining (Fig. 4C–F).

To understand the contribution of both CD4⁺ and CD8⁺ SmarT cells in mediating tumor control, we positively selected each subpopulation using magnetic selection and separately evaluated their cytokine profile using luminex array. Figure 4G shows that both cell subsets produced effector cytokines, including IL2, IL6, and GM-CSF, but CD4⁺ cells produced >50% more than their CD8 counterparts, resulting in superior T-cell expansion when cultured in tumor milieu conditions (day 0, 1 × 10⁶ cells; day 35, 9.5 ± 2 × 10⁸ vs. 4.1 ± 3 × 10⁷, CD4⁺ vs. CD8⁺ T cells; Fig. 4H). However, we did not observe any difference in IFNγ and TNFα levels, as shown in Supplementary Fig. S4. In addition, CD4⁺ SmarT cells also exhibited increased mitochondrial function (Fig. 4I) with elevated levels of Krebs cycle and glycolytic pathway metabolites compared with their CD8 counterparts (Fig. 4J). In contrast, CD8⁺ SmarT cells produced higher levels of effector molecules associated with cell killing (Granzyme A, Granzyme B, and perforin) and exhibited superior cytotoxic activity in vitro (Fig. 4K and L). However, the combination of both cell subsets was required for optimal antitumor effects, as either alone was insufficient to mediate tumor control [CD4⁺ (5 ± 2 × 10⁴), CD8⁺ (4 ± 0.4 × 10⁴), and SmarT cells (14 ± 4 × 10⁴), T-cell numbers on day 6, 1:40 E:T ratio; Fig. 4M–O].

To assess the in vivo potential and tumor selectivity of SmarT cells, we generated a dual tumor animal model. On the left, flank NSG mice were engrafted subcutaneously (s.c.) with cells that just expressed PSCA to mimic normal tissue (5 × 10⁶ CAPAN-1 PSCA) whereas on the right flank the same animals were engrafted with tumor cells expressing PSCA, TGFβ, and IL4 to recapitulate the signature present at the pancreatic tumor (5 × 10⁶ CAPAN-1 PSCA/TGFβ/IL4; Fig. 5A). Upon tumor engraftment, mice were injected with 5 × 10⁶ SmarT cells labeled with firefly luciferase (FFluc⁺). By day 18, we observed a significant expansion (1.5 ± 0.4E+09 photons/s, T-cell bioluminescence signal on day 18) of SmarT cells (both CD4⁺ and CD8⁺; Supplementary Fig. S5) on the right flank where the cells expressed PSCA, TGFβ, and IL4. In contrast, on the left side, where only PSCA was expressed, SmarT cells failed to expand (0.04 ± 0.01E+09 photons/s, day 18; P = 0.0002; Fig. 5B–D). Consequently, the preferential expansion of SmarT cells resulted in selective elimination of PSCA + TGFβ + IL4 expressing tumors (tumor volume 0 ± 0 vs. 616 ± 34 mm³, PSCA + TGFβ + IL4 vs. PSCA only by day 33; Fig. 5E and F). Importantly, upon tumor elimination, the T-cell numbers rapidly contracted, demonstrating the requirement for antigen and both cytokines to sustain T-cell expansion—an important safety feature of this approach (Fig. 5G and H).

To evaluate the in vivo recall response of SmarT cells, we established a tumor rechallenge model in which NSG mice were first engrafted s.c. with 5 × 10⁶ tumor cells expressing PSCA, TGFβ, and IL4 and treated intravenously with 2 × 10⁶ SmarT cells (FFLuc⁺). As expected, by day 30 all animals treated with SmarT cells had experienced a complete response, coinciding with the expansion and subsequent contraction of transgenic T cells (Fig. 6A). To next determine if we could recall this T-cell expansion in a tumor-selective manner, we took these tumor-free animals and reengrafted them with 5 × 10⁶ PSCA only (left shoulder) or tumor milieu cells (PSCA + IL4 + TGFβ, right shoulder). Rechallenge induced SmarT cell reexpansion but only at the site engrafted with tumor cells expressing PSCA, IL4, TGFβ, thereby effectively controlling tumor growth (Fig. 6B). Taken together, these data demonstrate the persistence, proliferative capacity, potency, and selectivity of SmarT cells.

In the current study, we demonstrate how T cells can be engineered to recognize a pattern present exclusively at the tumor using independent chimeric molecules that coalesce to be fully functional in the tumor microenvironment. Using pancreatic cancer as a model system, we successfully engineered T cells with receptors that (i) recognized three tumor-derived input signals (PSCA, TGFβ, and IL4) present in combination at the pancreatic tumor site and (ii) were equipped with intracellular signaling domains responsible for T-cell activation (signal 1), costimulation (signal 2), and cytokine support (signal 3). This proof-of-concept study demonstrates how receptor engineering can enhance not only the function but also the safety of transgenic T cells at the immunosuppressive tumor microenvironment by limiting their maximal activity exclusively to the tumor site, thereby reducing concerns related to on-target off-tumor effects.

The adoptive transfer of CAR-modified T cells has effectively treated CD19⁺ hematologic malignancies including multiple subtypes of B-cell lymphoma, B-cell chronic lymphocytic leukemia (B-CLL), and acute lymphoblastic leukemia (ALL; refs. 2, 34, 35), resulting in the recent FDA approval of CAR-CD19 therapy: Kymriah (tisagenlecleucel) and Yescarta (axicabtagene ciloleucel) for the treatment of pediatric ALL and adult relapsed/refractory large B-cell lymphoma, respectively. However, successful extension of CAR therapy to solid tumors has proven challenging, mainly due to the immunosuppressive tumor microenvironment that impairs the effector function, proliferative capacity, and in vivo persistence of the infused T cells. To overcome this barrier, several strategies have been used to allow transgenic T cells to better withstand the inhibitory milieu. These include the incorporation of costimulatory endodomains within CARs (e.g., CD28, 4-1BB, and OX40; refs. 1, 2, 34–36) or by the combination of CARs with transgenic immune stimulatory cytokines (e.g., IL7, IL15, and IL12; refs. 37–39) or novel transgenic molecules designed to blunt (e.g., dominant-negative TGFβ receptors; ref. 40) or invert (e.g., PD-L1 or IL4; refs. 41–43) immunosuppressive tumor-derived signals. Our approach extends beyond this latter strategy by programming T cells to recognize a pattern exclusive to the tumor using a trio of receptors that bind to an antigen on the tumor but also invert the effects of immunologically suppressive molecules into signals that are immunostimulatory and recapitulate physiologic T-cell signaling.

One of the major challenges in the field of engineered T cells has been balancing potency with safety, given that most tumor-expressed antigens are not exclusively present on malignant cells. Thus, the interaction of transgenic T cells with normal tissue expressing the target antigen has led to “on-target, off-tumor” toxicities ranging from tolerable (e.g., lifelong B-cell aplasia in patients treated with T cells expressing anti-CD19 CAR; refs. 1, 2) to severe (e.g., fatalities following the infusion of HER2 CAR T cells attributed to cytokine release syndrome; ref. 4). These toxicities are currently managed either by nonspecific strategies that suppress all T cells (e.g., administration of high-dose corticosteroids), with side effects including enhanced susceptibility to infections, to the activation of suicide genes (e.g., HSV-Tk, iCaspase-9, CD20 monoclonal antibody, and EGFR antibody), which selectively target and kill transgenic T cells (44). However, in both cases, the therapeutic potential of the infused cells is lost, which has prompted the development of next-generation engineering approaches designed to promote T-cell activation selectively at the tumor site while sparing normal tissues.

To achieve this goal, T cells have been programmed to become active only after two independent receptors, whose signaling is complementary, have interacted with distinct antigens expressed by tumor cells. Roybal and colleagues demonstrated the feasibility of this approach by harnessing the Notch signaling pathway using either mesothelin and CD19 or GFP and CD19 antigens as model targets (6), whereas other groups have split CD3 and CD28/41BB signals between tumor-directed CARs to target breast (MUC1 and HER2 CARs; ref. 45) or prostate (prostate-specific membrane antigen and PSCA; ref. 5) cancer. Finally, Fedorov and colleagues sought to restrict the activity of transgenic cells to the tumor site using healthy-tissue antigen-specific inhibitory CARs (iCAR) expressing PD-1 or CTLA4 endodomains in order to render transgenic T cells inactive outside of the tumor site (7). In general, though, the risk of tumor immune escape due to antigen loss when using a dual antigen-targeted approach limits the potential effectiveness of this strategy. Thus, rather than incorporate additional antigenic targeting, we have chosen to render our T cells responsive to nominally immunosuppressive soluble factors produced by both the tumor and stroma (IL4 and TGFβ) that contribute to tumor growth and survival, thereby minimizing the risk of a mutational event leading to immune evasion. In addition, our inverted cytokine and costimulatory receptors not only restrict transgenic T-cell activity to the tumor site, but also render our cells resistant to immunosuppressive molecules that would otherwise adversely affect transgenic T-cell effector function and persistence. Although in this work we have focused our efforts on cytokines that are present at elevated levels in pancreatic cancer, this strategy can be customized to harness other components of the tumor milieu, including PD-L1, LAG3, IL10, IL13, and VEGF, depending on the immunosuppressive profile of the target tumor (46). Furthermore, these customized ICRs can be coupled with other CARs to prevent tumor escape by negative antigen selection.

T-cell activation, proliferation, and persistence requires the presence of a synchronized combination of 3 signals (antigen recognition, signal 1; costimulation, signal 2; and cytokine, signal 3) delivered upon engagement with independent ligands, which results in target recognition, expansion of both CD4⁺ and CD8⁺ T-cell subsets, and the establishment of long-term memory. However, absence of any one of these signals substantially impairs long-term native T-cell function, a phenomenon that was recapitulated in the initial clinical trials of first-generation CAR T cells (containing only the CD3ζ endodomain), which displayed limited expansion and persistence (35, 47, 48). A number of groups subsequently explored the incorporation of additional costimulatory moieties in tandem with the CAR to produce later-generation iterations. Although effective, this approach has produced variable therapeutic efficacy in patients, a feature some have attributed to the inconsistent nature of the product infused, which is often dominated by CD8⁺ T cells that display limited persistence (absent CD4⁺ “help”), resulting in diminished long-term potency (39, 49, 50). Indeed, to standardize product composition, Turtle and colleagues established a protocol to separately propagate both CD4⁺ and CD8⁺ subsets in vitro, which were subsequently mixed at a 1:1 ratio prior to clinical use and reported superior outcomes in patients with non-Hodgkin lymphoma and B-ALL (49, 50). Consistent with this clinical finding, we also demonstrate that a mix of both CD4⁺ and CD8⁺ SmarT cells is required for tumor elimination in vitro and in vivo. However, in our studies, we “naturally” achieve the optimal cell balance by recapitulating physiologic T-cell signaling in our transgenic cells using a trio of receptors, whose engagement with their respective ligands results in an orchestrated signaling cascade producing immediate T-cell activation and long-term persistence.

In the current study, we have used genetic reprogramming to enable T cells to discriminate tumor from healthy tissue, thereby increasing both the specificity and potency of these transgenic cells. Furthermore, expression of the inverted cytokine and costimulatory receptors both protected the T cells from the immunosuppressive tumor milieu and ensured that they received all 3 signals (activation, costimulation, and cytokine support) required for in vivo amplification, antitumor activity, and long-term persistence of polyclonal effector T cells. Importantly, transgenic cells remained dependent on the presence of tumor antigen and rapidly contracted upon tumor elimination, confirming the safety of this approach. Although this proof-of-concept study will require refinements prior to clinical application, this strategy can be readily extended to other immunotherapeutic modalities, such as αβ TCRs and tumor-specific T cells with native receptor specificity.

Peripheral blood mononuclear cells (PBMC) were derived from healthy donors and patients after informed and written consent on protocols approved by the Institutional Review Board at the Baylor College of Medicine (H-15152 and H-28601, respectively) conducted in accordance with the Declaration of Helsinki, the Belmont Report, and U.S. Common Rule. All cell lines (obtained between 2012 and 2013)—K562 (chronic erythroid leukemia cell line), 293T (human embryonic kidney cell line), and CAPAN-1 (pancreatic cancer cell line)—were obtained from the ATCC and cultured at 37°C in a humidified incubator containing 5% carbon dioxide (CO₂). K562 cells were maintained in RPMI-1640 media (Hyclone Laboratories), whereas 293T and CAPAN-1 were cultured in Iscove's Modified Dulbecco's Medium (IMDM; Gibco by Life Technologies Corporation). Media for maintaining the cell lines were supplemented with 10% heat-inactivated FBS (Hyclone Laboratories) and 2 mmol/L glutaMAX (Gibco by Life Technologies Corporation). All cell lines were authenticated by the University of Arizona Genetics Core using short tandem repeat profiling (last tested and confirmed in November 2017). The cell lines were also routinely tested for Mycoplasma every 6 months using the MycoAlert Mycoplasma Detection Kit (Lonza), and all results were negative.

The first-generation CAR was generated by synthesizing cDNA of a codon-optimized single-chain variable fragment of PSCA followed by a spacer (IgG2-derived hinge and CH3 domain), transmembrane (CD28 TM), and CD3ζ chain of the TCR complex (42). The TBBR construct was generated by synthesizing a DNA construct containing the signal peptide and extracellular domain of TGFβRII linked to the 41BB endodomain (sequence obtained from Uniprot, Q07011). The TBBR construct was cloned into a vector expressing IRES-GFP to serve as a surrogate marker. The 4/7 ICR construct was synthesized by fusing the signal peptide and extracellular domain of the IL4R with the transmembrane and intracellular component of the IL7R followed by an IRES-mOrange tag as described previously (26, 42). All constructs were cloned into SFG retroviral vectors and transfected using 293T cells to generate retroviral supernatant as previously described (26, 42).

To generate genetically modified T cells, 1 × 10⁶ PBMCs isolated from donor blood using Lymphoprep (Axis-Shield PoC AS) were plated in 24-well non-tissue culture-treated plates coated with OKT3 (1 mg/mL; Ortho Biotech) and CD28 antibodies (1 mg/mL; Becton Dickinson). Cells were cultured in complete media: RPMI containing 45% Clicks medium (Irvine Scientific), 10% FBS, and 2 mmol/L glutaMAX supplemented with 50 U/mL of recombinant human IL2 (NIH) on day 1. For transduction, 1 mL of retroviral supernatant was plated in a 24-well non-tissue culture-treated plate precoated with recombinant fibronectin fragment (FN CH-296; Retronectin; Takara Bio Inc.). The plates were centrifuged at 2,000 × g for 90 minutes. Next, OKT3/CD28-activated T cells (0.2 × 10⁶/mL) resuspended in 2 mL of complete media supplemented with IL2 (100 U/mL) were added to the wells and centrifuged at 400 × g for 5 minutes. The transduced cells were transferred to a 37°C, 5% CO₂ incubator and were subsequently split and fed with fresh media containing IL2 (50 U/mL) every 2 to 3 days. To generate SmarT cells, T cells were transduced with 1 mL of CAR, TBBR, and 4/7 ICR supernatant combined on day 3. Transduction efficiency was measured on day 3 by flow cytometry and all functional experiments were set up within 7 to 10 days after transduction.

The analysis and visualization of the gene expression pattern of PSCA (8), TGFβ (9), and IL4 (10) were confirmed using the Oncomine microarray database (http://www.oncomine.org).

T cells were sorted 5 days after transduction using an SH800S cell sorter (Sony Biotechnology) to normalize for the mean fluorescence intensity (MFI) between transgenic T cells and respective controls. RNA was isolated at 24 hours after stimulation from (i) 1G.CAR cells exposed to PSCA antigen (0.5 μg/mL); (ii) TBBR and 4/7 ICR T cells first stimulated with OKT3 (0.5 μg/mL) and then exposed to TGFβ (5 ng/mL) and IL4 (400 U/mL), respectively, after 3 days; and (iii) SmarT cells stimulated with PSCA (0.5 μg/mL), TGFβ (5 ng/mL), and IL4 (400 U/mL). As controls, we generated T cells modified to express ΔCAR, ΔTGFβRII, and ΔIL4R—truncated versions of the transgenic modifications that lack endodomains. RNA isolation was performed using the RNeasy mini plus kit (QIAGEN) and hybridized to the nCounter PanCancer immune profiling panel (human codeset). RCC files containing raw counts for 770 genes provided from NanoString were loaded into nSolver Analysis Software 3.0 and normalized for housekeeping genes and positive controls. Normalized data were exported and read into Java code (run in Eclipse IDE) to extract lists of differentially expressed genes (DEG) that were significant (P < 0.05), which was determined by performing a heteroscedastic two-tailed t test that assumes unequal variance on the log-transformed normalized data. The top 50 DEGs were mapped onto STRING v.10.5 Homo sapiens, which clusters genes into networks based on scored interactions, associations, and pathway knowledge drawn from databases such as the Kyoto Encyclopedia of Genes and Genomes, Gene Ontology (GO), and text mined from the literature. Three enriched GO functions were selected and uploaded to STRING to produce the functional association networks where the network edges were based on a minimum interaction confidence of 0.4, a maximum of 10 interactors shown first shell, and the selected interaction sources were text mining, experiments, databases, coexpression, neighborhood, gene fusion, and co-occurrence. The genes within each functional group were assigned a specific color and further ranked based on the magnitude of fold change from highest to lowest (fold change and P value listed in Supplementary Tables S1–S4). Established structured repositories for the NanoString data are available through the Gene Expression Omnibus. The accession numbers for the datasets are GSE113387, GSE113388, GSE113389, GSE113390, and GSE113391.

PSCA-GFP retroviral supernatant (1 mL) was plated in 24-well non-tissue culture-treated plates precoated with retronectin and centrifuged at 2,000 × g for 90 minutes. CAPAN-1 cells (0.2 × 10⁶/mL in IMDM) were added to each well and transferred to a 37°C, 5% CO₂ incubator. Transduction efficiency was measured by flow cytometry within 1 week, and the cells were sorted after 2 weeks. The sorted CAPAN-1 PSCA cells were further transduced with IL4 cytokine-IRES-mOrange and TGFβ-IRES-ΔCD19 retroviral constructs and sorted to generate CAPAN-1 PSCA/TGFβ/IL4-producing cell line.

One million transgenic T cells were stimulated weekly with 1 × 10⁶ irradiated CAPAN-1 PSCA supplemented with IL4 (400 U/mL, 3× weekly) and TGFβ (5 ng/mL, 1× weekly), or IL2 (50 U/mL, 3× weekly). T-cell expansion was quantified weekly by Trypan blue exclusion using a hemocytometer.

As described previously, the specificity and killing capacity of the modified T cells were assessed using CAPAN-1, a PSCA-positive cell line in a standard 4- to 6-hour ⁵¹Cr-release assay (42).

T cells were cocultured with 0.5 × 10⁶ CAPAN-1 PSCA/TGFβ/IL4-producing cell line at the specified effector:target ratio in 4 mL complete media in a 6-well plate. The cells were harvested every 3 days, labeled with CD3 APC, CD4 Krome Orange, and CD8 Pacific blue antibodies (Beckman Coulter) and quantified by flow cytometer using CountBrightTM Absolute Counting Beads (approximately 0.2 × 10⁵ beads/20 μL added to each condition; Invitrogen) and 5 μL of 7-AAD (BD Biosciences) to exclude dead cells. Total tumor and T-cell numbers were back calculated from the viable cell numbers obtained by terminating acquisition at 2,000 beads.

For flow-cytometry analysis, cells were collected, washed, and stained with antibodies for 30 minutes at 4°C in the dark. The cells were incubated with monoclonal antibodies against CD4 Krome Orange (Beckman Coulter), CD8 Pacific Blue (Beckman Coulter), CD25 PE Cy5 (BD Biosciences), CD69 ECD (Beckman Coulter), CD27 PE Cy7 (BD Biosciences), CD28 PE Cy5 (BD Biosciences), CD45RO PE Cy7 (BD Biosciences), CCR7 Alexa Fluor 700 (BD Biosciences), PD1 PE Cy7 (BD Biosciences), TIM3 PerCP Cy 5.5 (BioLegend), and LAG3 APC (R&D Systems). To detect CAR expression, cells were labeled with F(ab′)2 fragment goat anti-human IgG (H + L) antibody conjugated with AlexaFluor647 (Jackson ImmunoResearch Laboratories, Inc.). TBBR transgene expression was detected using antihuman TGFβRII antibody (Abcam) further labeled with rat anti-mouse APC (BD Biosciences) and was corroborated with the GFP surrogate marker. Similarly, 4/7 ICR expression was confirmed by labeling cells with anti-human IL4Rα–APC antibody (R&D Systems) to detect IL4R expression and correlated with the mOrange surrogate marker. Routine analysis of TBBR and 4/7 ICR-modified T cells, however, was performed by monitoring expression levels of the GFP and mOrange markers, respectively. All cells were washed prior to data acquisition on a Gallios flow cytometer and analyzed using Kaluza software (Beckman Coulter). To assess cell viability, the transgenic cells were stained as per the manufacturer's protocol using the Annexin V Apoptosis Detection Kit (BD Biosciences). For measuring FOXP3 expression, cells were first stained with cell-surface markers (CD4⁺/CD25⁺) followed by intracellular cytokine staining using the anti-human FOXP3 staining set (eBioscience).

The CD4⁺ T-cell fraction was isolated by labeling cells with CD4 microbeads that were passed through an LD column as per the manufacturer's instructions (Miltenyi Biotech). CD4⁺ fraction was further selected using an LS column to increase the purity. To obtain pure CD8⁺ T cells, the negative fraction collected from the previous step was labeled with CD8 microbeads and passed through an LS column.

To measure mitochondrial function using the Seahorse assay, sorted CD4⁺ and CD8⁺ SmarT cells stimulated with irradiated CAPAN-1 target cells (1:1 ratio) in the presence of TGFβ and IL4 for 10 days were plated on XF24 cell culture microplates (XF24 FluxPak mini) precoated with Cell Tak (Corning) and calibrated as per the manufacturer's instructions. During instrument calibration, 5 × 10⁵ T cells resuspended in 100 μL of XF assay medium containing 5.5 mmol/L glucose, 2 mmol/L l-glutamine, and 1 mmol/L sodium pyruvate were added to each well and centrifuged at 200 × g for 1 minute. After 20 minutes of incubation in a 37°C non-CO₂ incubator, 400 μL supplemented media were added to each well. Oxygen consumption rate was measured using an extracellular flux analyzer (Seahorse Bioscience) at basal levels and following treatment with 1 μmol/L oligomycin, 1 μmol/L FCCP, and 1 μmol/L rotenone/antimycin A (XF Cell Mito Stress Kit). The metabolomic profile of sorted CD4⁺ and CD8⁺ SmarT cells was assessed by submitting pelleted samples (5 × 10⁶ cells) to the Metabolomics core at the Baylor College of Medicine for analysis of 16 fatty acid and 12 tricarboxylic acid cycle metabolites.

Cytokine levels were measured from sorted CD4⁺ and CD8⁺ SmarT cells cultured with irradiated CAPAN-1 target cells (1:1 ratio) in the presence of TGFβ and IL4 for 10 days. Luminex assay was performed on supernatant collected 24 hours after stimulation with PSCA, TGFβ, and IL4 using the human CD8⁺ T-cell magnetic bead panel as per the manufacturer's instructions (HCD8MAG15K17PMX, EMD Millipore).

Four-to-five-week-old female NSG mice (NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ; stock number 005557; The Jackson Laboratory) were engrafted s.c. with 5 × 10⁶ CAPAN-1 PSCA (left flank) or CAPAN-1 PSCA/IL4/TGFβ (right flank) cells for the dual tumor animal model. When the tumor volume reached approximately 80 mm³, the mice were administered with 5 × 10⁶ SmarT cells labeled with GFP-firefly luciferase (FFluc) intravenously. Tumor volume was measured using calipers (tumor volume (mm³) = length × width × width/2), and T-cell bioluminescence signal was monitored by injecting mice intraperitoneally with 100 μL of luciferin (15 mg/mL) followed by imaging using the IVIS Lumina In Vivo Imaging system (Caliper Life Sciences). For tumor rechallenge, mice were first engrafted s.c. with 5 × 10⁶ CAPAN-1 PSCA/TGFβ/IL4 cells and injected with 2 × 10⁶ SmarT cells (FFluc⁺). Upon tumor elimination, the mice were rechallenged with 5 × 10⁶ CAPAN-1 PSCA cells on the left shoulder and CAPAN-1 PSCA/IL4/TGFβ cells on the right shoulder. Data analysis was performed using Living Image software. All in vivo experiments were performed according to the Baylor College of Medicine Animal Husbandry guidelines.

Statistical analysis was performed using GraphPad Prism 5 software (GraphPad Software, Inc.). For comparison between two groups, significance was determined using a Student two-tailed t test. Data comparing three or more groups were analyzed using one-way ANOVA with Bonferroni multiple comparisons test. For microarray data, nSolver analysis software was used to obtain DEGs that were statistically significant (P < 0.05) by performing a two-tailed t test that assumes unequal variance (heteroscedastic test) on the log-transformed normalized data.

M.K. Brenner is a cofounder and member of Tessa Therapeutics; has ownership interest (including stock, patents, etc.) in Viracyte, Tessa Therapeutics, and Markers; is a consultant/advisory board member for Tessa Therapeutics, Markers, and Viracyte; and has received other remuneration from Tessa Therapeutics, Viracyte, and Markers. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. Sukumaran, W.E. Fisher, A.M. Leen, J.F. Vera

Development of methodology: S. Sukumaran, N. Watanabe, A.M. Leen, J.F. Vera

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Sukumaran, N. Watanabe, S. Mohammed, W.E. Fisher, A.M. Leen, J.F. Vera

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Sukumaran, N. Watanabe, K. Raja

Writing, review, and/or revision of the manuscript: S. Sukumaran, P. Bajgain, K. Raja, S. Mohammed, W.E. Fisher, M.K. Brenner, A.M. Leen, J.F. Vera

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Sukumaran, P. Bajgain

Study supervision: W.E. Fisher, A.M. Leen, J.F. Vera

This work was supported by grants from the NIH-NCI (P01 CA094237, P50 CA126752, P50 CA186784), a Pancreatic Cancer Action Network Translational Research Grant (16-65-LEEN), the V Foundation for Cancer Research (T2016-006), the Elsa U. Pardee Foundation, and the National Pancreas Foundation, as well as the Adrienne Helis Malvin Medical Research Foundation in collaboration with Baylor College of Medicine. J.F. Vera is supported by a Mentored Research Scholars Grant in Applied and Clinical Research (MRSG-14-197-01–LIB) from the American Cancer Society. The authors acknowledge the editorial assistance provided by Catherine Gillespie as well as Walter Mejia for helping with the artwork and formatting of figures and tables for the manuscript. The authors would also like to thank Texas Children's Hospital for the use of the Small Animal Imaging Facility, the Mouse Metabolism and Phenotyping core (NIH UM1HG006348 and NIH 1R01DK114356), the Metabolomics core supported by the CPRIT Core Facility Award (RP120092) and P30 Cancer Center Support Grant (NCI-CA125123), and the support of the Flow Cytometry and Cell and Vector Production shared resources in the Dan L. Duncan Comprehensive Cancer Center. M.K. Brenner, A.M. Leen, and J.F. Vera are supported by P50 CA126752 and P50 CA186784. W.E. Fisher, M.K. Brenner, A.M. Leen, and J.F. Vera are supported by P01 CA094237. W.E. Fisher and A.M. Leen's research is supported by the 2016 Pancreatic Cancer Action Network Translational Research Grant (16-65-LEEN) and The V Foundation for Cancer Research (T2016-006). J.F. Vera is supported by a Mentored Research Scholars Grant in Applied and Clinical Research (MRSG-14-197-01–LIB) from the American Cancer Society. N. Watanabe and J.F. Vera were supported by the Elsa U. Pardee Foundation. The National Pancreas Foundation provided support to S. Sukumaran, N. Watanabe, and A.M. Leen.

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.