4-1BB Agonist Focuses CD8⁺ Tumor-Infiltrating T-Cell Growth into a Distinct Repertoire Capable of Tumor Recognition in Pancreatic Cancer

Pancreatic cancer is the third-highest cause of cancer-related death for men and women in the United States and is expected to become the second-leading cause of cancer mortality by 2030 (1, 2). The majority (85%) of pancreatic cancer diagnoses are classified as pancreatic ductal adenocarcinoma (PDAC; ref. 3). Patients afflicted with PDAC often present with late-stage cancer and face the poorest prognosis of all cancer types with a 5-year survival rate of around 6% (4). Despite efforts to improve treatment, surgery, chemotherapy, and chemoradiation remain the only options. These treatment strategies have shown limited effectiveness as most patients will recur within a year of treatment even after successful tumor resection (4, 5). Therefore, there is a great need to broaden treatment options.

Immunotherapy has made a tremendous mark in the treatment of cancer, especially in the past decade. Its success was first observed in the treatment of metastatic melanoma with high-dose IL2 and then more recently with agents that block CTLA-4 and PD-1 (checkpoint blockade; refs. 6–8). These treatments were later transposed to non–small cell lung cancer and renal cell carcinoma (9–11). In PDAC, however, they have been ineffective with no objective response seen for treatment with anti-CTLA-4 and anti-PD-L1 (12, 13). The lack of efficacy could be a result of the paucity of CD3⁺ T-cell infiltrate (14). An alternative approach to overcome the limitation posed by the modest immune infiltrate in PDAC is the ex vivo amplification of TILs for reinfusion through autologous adoptive cell therapy (ACT).

TIL ACT expands T cells up to several hundred-fold from surgically resected tumor and reinfuses them into the patient, providing a large influx of antitumor T cells. Our group and others have demonstrated its effectiveness in melanoma (15–18). With an average objective response rate (ORR) of 50%, TIL ACT is among the best treatment options for metastatic disease. The MDACC experience also demonstrated a positive correlation between CD8⁺ TILs infused and response (17). These results have already spurred efforts to translate ACT to other cancer types such as cervical (33% ORR), and gastrointestinal (25% ORR; refs. 19, 20). PDAC could also potentially benefit from TIL ACT as the presence of CD8⁺ TILs is associated with greater 5-year survival (21, 22). These suggest that endogenous PDAC TIL can exert some degree of tumor control, supporting the potential of TIL ACT.

One of the major challenges faced in growing TILs from gastrointestinal cancer types for ACT trials is the difficulty of expanding CD8⁺ T cells from the tumor tissue (23, 24). PDAC has a well-characterized immunosuppressive tumor microenvironment that might contribute to the difficulty of triggering the proliferation of cytotoxic CD8⁺ T cells from this tumor tissue and account for their decreased numbers (14, 25). A method to resolve this barrier is by manipulating 4-1BB/CD137, a member of the tumor necrosis factor receptor family, which provides a strong costimulatory signal for increased activation, proliferation, and survival. This receptor is predominantly expressed on recently activated CD8⁺ T cells with peak expression at 24 hours (26). In fact, our group demonstrated that inclusion of an agonistic 4-1BB mAb [Urelumab, Bristol-Myers Squib (BMS)] in TIL cultures was able to increase melanoma and triple-negative breast cancer CD8⁺ TIL proliferation (27, 28). Based on this previous work, we posited that use of an agonistic 4-1BB mAb in PDAC TIL culture would provide the same benefits of increased CD8⁺ TIL yield.

Here, we demonstrate that the addition of an agonistic 4-1BB mAb increases the ability to grow TILs from PDAC, improves the total yield, and stimulates the proliferation of more CD8⁺ T cells without overly differentiating them. In addition, these CD8⁺ TILs have a distinct repertoire compared to IL2 only grown TILs and display MHC class I-restricted autologous tumor recognition. These results support the use of 4-1BB-expanded TILs in ACT strategies for patients with PDAC.

After obtaining informed consent, 26 patients with primary or metastatic pancreatic ductal adenocarcinoma underwent surgical resection. Two patients underwent resection on two sites, therefore a total of 28 samples were analyzed from 26 patients. Further patient characteristics are summarized in Supplementary Table S1. Patients are referred to by their deidentified “MP” number. In 23 patients, prior chemotherapy and/or chemoradiation was administered. Tissue from surgical resections was used to expand TILs under protocols (PA15-0176, LAB00-396, PA15-0014 for PDAC samples and LAB06-0755 for melanoma samples) approved by the Institutional Review Board of The University of Texas MD Anderson Cancer Center. This study was carried out in compliance with Good Clinical Practice concerning medical research in humans, as described in the Declaration of Helsinki.

A fully human and purified IgG4 mAb against human CD137/4-1BB, Urelumab (663513), was kindly provided by BMS. Human recombinant IL2 (Proleukin) was generously provided by Prometheus Therapeutics and Diagnostics. MHC class I blocking antibody (clone W6/32) and isotype control (mouse IgG2a, clone eBM2a) were purchased from Invitrogen and eBioscience, respectively. CAPAN-1 cell line was purchased from ATCC. Autologous tumor targets were found to match the patients using STR DNA fingerprinting performed at MDACC and the tumoroid was confirmed mycoplasma-free.

The tumor samples were cut into 1 to 3 mm² fragments and placed in TIL culture media [TIL-CM: RPMI1640 with GlutaMax (Gibco/Invitrogen), 1 × Pen–Strep (Gibco/Invitrogen), 50 μmol/L 2-mercaptoethanol (Gibco/Invitrogen), 20 μg/mL Gentamicin (Gibco/Invitrogen), and 1 mmol/L pyruvate (Gibco/Invitrogen)] with 6000 IU/mL IL2 in 24-well plates for a period of 5 weeks, as previously described (17, 29). The same method was applied for the metastatic melanoma samples. For the 4-1BB condition, both 6000 IU/mL IL2 and 10 μg/mL 4-1BB mAb were added in the culture plates on day 0 and day 4 or 5. TILs were expanded for up to 35 days prior to performing the described assays or the rapid expansion protocol (REP). The REP was performed in the G-Rex 10 device (Wilson Wolf Manufacturing) following a scaled-down version of the previously described protocol (29). Briefly, TILs were put in culture with pooled allogeneic irradiated PBMC feeder cells at a ratio of one TIL to 200 feeders in combination with 6000 IU/mL IL2 and 30 ng/mL of anti-CD3 (OKT3 clone) on day 0 of the REP. The REP process lasted for 14 days, with REP-CM [half TIL-CM and half AIM-V (Invitrogen)] used for the first 7 days and only AIM-V for the last 7 days of expansion.

Four-micrometer-thick serial sections were obtained from representative formalin-fixed, paraffin-embedded (FFPE) blocks for IHC, as well as hematoxylin and eosin (H&E) staining. H&E slides were examined by a pathologist to confirm the presence of tumor. IHC was performed using a Leica Bond Max automated staining system (Leica Microsystems) with antibodies against CD3 (dilution 1:100; Dako). The expression of the marker was detected using a Leica Bond Polymer Refine Detection Kit (Leica Microsystems) with diaminobenzidine reaction to detect antibody labeling. Counterstaining was done using hematoxylin. Human tonsil FFPE tissues with and without CD3 primary antibody were used as positive and negative controls, respectively. For quantification of CD3 expression, the slides were digitally scanned at 200 magnification using the Aperio AT2 scanner (Leica Microsystems). The images were visualized using the ImageScope software (Leica Microsystems) and analyzed using the Aperio Image Toolbox (Leica Microsystems). Five regions of interests were randomly selected within the tumor area of each slide. The number of CD3 positive cells per mm² (cell density) was evaluated, and the final score was expressed as the average density of the five areas.

Fresh tumor samples were manually disaggregated between frosted-glass slides to obtain a single-cell suspension for analysis. Both the disaggregated tissue samples and expanded TILs were stained in FACS Wash Buffer (Dulbecco's phosphate buffered saline 1× with 1% BSA) for 30 minutes using fluorochrome-conjugated monoclonal antibodies for CD3, CD4, CD8, CD16, CD56, CD57, Granzyme B, T-bet, γδ T-cell receptor (TCR), CD27, CD28, CD45, CD45RA, CD45RO, CCR7, BTLA (clone J168; BD Bioscience), PD-1, KLRG1 (Biolegend), and Eomes (eBioscience/ThermoFisher). Stained cells were fixed in 1% paraformaldehyde solution for 20 minutes. Intracellular staining was performed using eBioscience transcription factor staining kit according to the manufacturer's instructions. Samples were acquired using the BD FACSCanto II or BD LSRFortessa X-20 and analyzed using FlowJo Software v10.2 (TreeStar). Dead cells were excluded using an AQUA or Yellow live/dead staining (Invitrogen).

Genomic DNA was extracted from samples using DNeasy Blood & Tissue Kit (Qiagen) as per manufacturer's instructions. TCRβ CDR3 regions were amplified from between 0.2 and 3 μg of DNA. All samples had the ImmunoSEQ assay performed at Adaptive Biotechnologies, with deep sequencing for PBMC DNA and survey-level sequencing for all others. Data analysis was performed at MDACC.

To provide greater cell numbers for functional assays, sorted CD8⁺ TILs underwent the REP process. In triplicate, T cells were then put at a 10:1 ratio with their autologous tumor target, CAPAN-1 (HLA mismatch control), or media (TIL alone). For some experiments, the autologous tumor targets were incubated with 80 μg/mL of anti-MHC class I antibody (clone W6/32) for 3 hours prior to addition of T cells, as previously described (30). After 24-hour incubation, the supernatants and T cells were collected. TILs were analyzed for 4-1BB/CD137 expression via flow cytometry. Detection of secreted cytokines in the corresponding supernatants were detected using a V-PLEX Plus Proinflammatory Panel 1 (human) Kit and analyzed on a QuickPlex SQ 120, both available from Meso Scale Discovery. Reported values have coefficient of variation <20%.

(GraphPad) Prism v6.0 (GraphPad Software) was used for graphing and statistical analysis. Differences between groups or experimental conditions were determined using either parametric or non-parametric, two-tailed t tests (paired or unpaired) as appropriate. Linear regression and Spearman correlation analyses were also used as indicated. Two-sided P-values <0.05 were considered statistically significant and in the figures are indicated as *, P < 0.05; **, P < 0.01; ***, P < 0.001, and ****, P < 0.0001.

To determine if the immune component of PDAC was sufficient for TIL ACT, we assessed the immune infiltrate by performing flow cytometry on manually disaggregated samples (n = 28). The amount of CD3⁺ TILs observed was less than 1% of all cells in the tumor sample on average as compared to metastatic melanoma with a CD3⁺ TIL infiltrate >2% (Fig. 1A). Quantitative IHC analysis found that the mean density of CD3⁺ TILs was 314 cells/mm² (Fig. 1B). This is fewer than what the literature reports for an immunogenic cancer like melanoma (422 cells/mm²; ref. 31). In addition, metastatic (closed circle) and primary (open circle) PDAC samples did not appear to stratify. Further IHC analysis showed that all samples displayed ≥50% MHC Class I expression that was homogenous throughout the tumor tissue, suggesting that lack of Class I was not the reason for low CD3⁺ infiltration (Fig. 1C). We also evaluated the proportion of CD8⁺ and CD4⁺ T cells. With an average CD8:CD4 ratio of 0.75, CD4⁺ TILs comprised the majority of the T-cell infiltrate (Fig. 1D). As a point of comparison, metastatic melanoma showed a CD8:CD4 ratio of 1.5 (Fig. 1D). This ratio is similar to IHC data also from Erdag and colleagues that exhibited a CD8:CD4 ratio of 1.6 (31). Metastatic and primary sites did not seem to show a difference in CD8:CD4 ratio. Phenotypic analysis on the CD8⁺ T cells determined their activation and differentiation state by assessing their expression of CD28 (50% ± 20%), CD45RA (10% ± 6%), PD-1 (45% ± 12%), and BTLA (20% ± 12%; Fig. 1E). There were not enough primary samples that could be compared with metastatic samples to discern a difference in the phenotype of their TILs. However, the low frequency of CD45RA expression combined with expression levels of the other three markers suggests a relatively activated and not terminally-differentiated immune infiltrate.

Enrichment of T-cell clones at the tumor site in comparison to the blood would suggest that the patient is mounting an immunogenic response to its tumor and that specific T-cell clones are migrating to the tumor and proliferating in the tissue. To determine the tissue-specific distribution of the T-cell repertoire, we sequenced the T-cell receptor beta-chain (TCRβ) CDR3 region of the T cells present in the blood, tumor, and normal tissue from seven patients when available (Fig. 2). This analysis, presented as productive clonality, revealed that the T-cell repertoire in the tumor is generally more clonal than in the blood (Fig. 2A). Productive clonality is a measure of the degree to which one or several unique clones dominate the repertoire (32). Linear regression analysis compared the relative frequencies of individual TCRβ clones present in both the blood and the tumor (Fig. 2B). All autologous blood–tumor pairs displayed slopes (m) <1, demonstrating a higher frequency of shared clones in the tumor than the blood. However, it is possible that high frequency clones present in the blood may correlate with high frequency clones present in the tumor. To determine the strength of correlation between clones in these sites, clones were partitioned and compared as follows using Spearman correlation: top clone frequency defined as ≥0.24% in the tumor (red circles), remaining clones partitioned in half with mid-frequency shared clones (green circles) and low-frequency shared clones (blue circles). Interestingly, Spearman correlation analysis showed weak or undefined correlation between the frequency of T-cell clones in the blood and their frequency in the tumor tissue (r_s < 0.5), even among the top ranking clones in the tumor (red circles). In addition, there were a few instances where the middle-ranking clones in the tumor (green circles) had greater correlation with their frequencies in the blood. Only MP31 showed good correlation (r_s = 0.61) of the frequencies of these top shared clones (Fig. 2B). Linear regression analysis was also used to compare the relative frequencies of individual TCRβ clones present in both the normal and tumor tissue (Fig. 2C). Similar to the tumor–blood pairs, the slopes for all the tumor-normal tissue pairs were <1, indicating that individual shared clones were found at higher frequencies in the tumor than the normal. In addition, clones were partitioned and correlation calculated in the same manner as in the tumor–blood pairs described above. In contrast to the tumor–blood pairs, the frequencies of the T cells in autologous tumor-normal tissue pairs showed stronger correlation among the top and mid-frequency clones. Four of the five pairs (MP64B, MP75, MP81, and MP84B) had r_s > 0.44 whereas two of them (MP64B and M75) had r_s > 0.5, showing that the repertoires were very similar between the tumor and the normal tissue in these cases. Also different from the tumor versus blood comparison was the observation that the top ranking clones in the tumor (red circles) were often high ranking in the normal tissue as shown by clustering along the hashed line. MP31 shared no top clones and very few clones overall, so it was not partitioned.

Prior work by our group detailed how infusion of melanoma patients with a higher proportion of CD8⁺ T cells and larger amount of TILs in general correlated with better clinical response (17). This result coupled with our observations of a predominance of CD4⁺ TILs and relative scarcity of CD3⁺ infiltration in general prompted us to consider ways to generate greater TIL growth that was rich in CD8⁺ T cells. Previous work showed that recently antigen-activated CD3⁺CD8⁺ TILs upregulate expression of the costimulatory molecule 4-1BB (26). Furthermore, additional work by our group and others demonstrated that stimulation of this pathway through the use of an agonistic anti-4-1BB antibody could decrease time of TIL culture while increasing total TIL growth, particularly that of CD3⁺CD8⁺ T cells (24, 27). Thus, we set up samples for TIL culture where one set of fragments received only the conventional high-dose IL2 (n = 28) and the other received high-dose IL2 plus the 4-1BB agonistic mAb (a4-1BB; n = 27; Supplementary Table S2). The addition of a4-1BB increased the average total TIL growth from 40 × 10⁶ cells for IL2 alone to 100 × 10⁶ for IL2 + a4-1BB (Fig. 3A). Only cultures that grew in at least one of the two conditions are represented in Fig. 3A. In addition, a4-1BB doubled the success rate (14/27; 52%) for TIL growth from fragments compared to IL2 only (8/28; 29%; data summarized in Supplementary Table S3). The benchmark for a successful TIL culture, 12 × 10⁶ total cells, was established from scaling down the MDACC Clinical Melanoma TIL Lab's criterion for success, where 20 fragments are set up for TIL expansion and 40 × 10⁶ cells is considered the minimum to treat a patient. Some IL2 only cultures did not reach the benchmark and several produced no discernible TIL growth while their companion a4-1BB culture produced ≥12 × 10⁶ cells (Supplementary Table S2). To that effect, our work demonstrates that use of a 4-1BB mAb could rescue cultures that would not have grown under the conventional methods.

Next, we determined whether the cells that grew out of the cultures treated with a4-1BB were enriched for CD3⁺ TILs and if CD8⁺ TILs now comprised the majority of CD3⁺ T cells. In cultures treated with a4-1BB, the total number of CD3⁺ TILs was significantly increased over IL2 only cultures on average from 30 × 10⁶ cells to 75 × 10⁶ cells, respectively (Fig. 3B). Only cultures that grew in at least one of the two conditions are represented in Fig. 3B. Because NK cells and γδ TCR⁺ T cells can also express 4-1BB, we stained for their presence in primary cultures to determine if the a4-1BB antibody being used was stimulating their growth (33, 34). Indeed, we found that three cultures with a4-1BB were enriched with a CD3⁻CD56⁺ population that was a greater proportion of the culture than CD3⁺ TILs (Fig. 3C). Likewise, some cultures also showed an increase in γδ TCR⁺ T cells in a4-1BB cultures versus IL2 alone (Fig. 3D, top graph). Only cultures that grew in both conditions are represented in Fig. 3C and D. Neither the increase in NK cell growth (P = 0.203) or γδ TCR⁺ T-cell growth (P = 0.078) due to 4-1BB costimulation was found to be significant overall. However, costimulation with a4-1BB produced a primary TIL culture that was on average 55% CD3⁺CD8⁺ TILs and 5% CD4⁺ TILs. We observed the opposite situation in IL2 alone cultures where 25% of CD3⁺ TILs were CD8⁺ and 60% were CD4⁺. Overall, the 4-1BB mAb caused a dramatic switch in the composition of CD3⁺ TILs toward the more favorable CD8⁺ TILs (Fig. 3D).

To better understand what effect the augmented growth via 4-1BB mAb stimulation had on CD8⁺ TIL differentiation, we performed phenotypic analysis of CD28, CD45RA, PD-1, and BTLA expression. Between IL2 only and a4-1BB cultures, the only significant change was a decrease in CD45RA expression (P = 0.031) (Fig. 4A). In fact, the level of expression of these markers was comparable to that seen on CD8⁺ TILs in the fresh tumors (Fig. 1E). Further phenotyping was done using the established memory markers CD45RA, CD27, CD28, and CCR7 (Fig. 4B; ref. 35). The vast majority of TILs, regardless of culture condition, were CD45RA⁻CCR7⁻, indicating that they are effector memory (EM) cells. Further characterization of their EM status was done by analyzing differential expression of CD27 and CD28, which has been shown to subdivide EM cells into four subsets termed EM1, EM2, EM3, and EM4 (Fig. 4C; ref. 36).The majority of TILs, again regardless of culture condition, fell in the EM3 (CD27⁻CD28⁻) subset which Romero and colleagues have shown to display stronger cytolytic activity (36). We also further explored the degree of differentiation with the expression of KLRG1, CD57, Eomes, T-bet and Granzyme B. As shown in Supplementary Fig. S1, KLRG1 was absent from both culture conditions. Combined with the other markers, this supplementary analysis further testifies that the stimulation of 4-1BB does not overly differentiate the cells and leads to the proliferation of effector/memory. Overall, this shows that even though a4-1BB stimulates aggressive expansion of activated CD8⁺ TILs, they do not become overly or terminally differentiated.

Next, we questioned how the culture conditions affected the repertoire of the TILs that grew out by sequencing sorted CD8⁺ T cells and comparing their relative frequencies to each other and to their starting frequency in the tumor (Fig. 5). Many clones are shared (red lines) between the tumor and both TIL culture conditions. However, these clones are present at different frequencies between the conditions, as demonstrated by the position of the lines where further away from center denotes higher frequency. This shows that the addition of 4-1BB mAb favors expansion of unique clones as compared to IL2 alone. This is further suggested by the presence of several T-cell clones that are shared between the tumor and only one of the culture conditions (blue lines). Finally, sequencing detected some clones that were not present in the tumor but were either present in only one culture (black lines) or shared only between the two culture conditions (green lines). Overall, in four out of five patients, the addition of 4-1BB mAb in the culture expands a greater number of CD8⁺ TILs but focuses their repertoire as evidenced by the smaller number of TIL clones in the a4-1BB cultures than the IL2 alone cultures.

Given the distinct repertoire generated by each culture condition, we assessed if the antitumor potential of both IL2 only and IL2 + a4-1BB cultured PDAC CD8⁺ TILs differed. To this end, sorted CD8⁺ TIL lines from both conditions were rapidly expanded (REP) and tested using a 24 hours co-culture with autologous tumor targets derived from patients MP81 and MP64B (Fig. 6 and Supplementary Fig. S2). Bulk TILs initially expanded using a4-1BB in combination with IL2 were also put through the REP process and achieved the expected fold expansion (between 500 and 1,500) after 14 days, confirming that both sorted CD8⁺ and bulk populations could exponentially grow after being propagated with a4-1BB (Supplementary Fig. S2). Prior to co-culture setup, autologous tumor cells and CAPAN-1, a HLA-mismatched pancreatic tumor line for MP81, were stained for MHC class I expression (HLA-ABC; Supplementary Fig. S3). Autologous tumor cells were found to express low, but detectable levels of MHC class I as compared to the CAPAN-1 cells (MFI: MP81 467, MP64B 606 vs. 5387 for CAPAN-1). In spite of this low level of MHC class I expression, both IL2 grown and a4-1BB grown MP81 TILs secreted more IFNγ in the presence of the autologous target than with CAPAN-1(Fig. 6A). Upregulation of 4-1BB on CD8⁺ T cells has been incorporated in tumor recognition assays previously (19, 20). As such, both MP81 TIL lines significantly upregulated 4-1BB expression after exposure to the tumor target as compared with TILs cocultured with CAPAN-1 (Fig. 6B and C). The upregulation of 4-1BB was particularly high on the CD56⁺ T-cell subset, a cytotoxic T-cell subset (37). Given that the 4-1BB expression was higher in the CD56⁺ CD8⁺ MP81 TIL subset, we assessed MHC class I-restricted recognition by blocking MHC class I on the autologous tumor target for MP81 (Fig. 6D and E, left). We observed that most of the recognition (4-1BB upregulation) in the total CD8⁺ population was MHC class I-restricted for TILs grown in both culture conditions. This experimental setup was repeated with an additional TILs and paired autologous tumor target (MP64B) with similar observations (Fig. 6D and E, right). CAPAN-1 was not used as a negative control for MP64B as it is partially HLA matched. Taken together, the IFNγ secretion and elevated 4-1BB expression indicate there are MHC class I-restricted tumor-reactive CD8⁺ T cells in the PDAC TIL repertoire.

In this study, we show that PDAC has a scarce, yet activated CD8⁺ TILs infiltrate that is preferentially expanded with the addition of an agonistic 4-1BB mAb to the TIL culture. The 4-1BB mAb consistently augmented total TIL numbers and doubled the success rate of TIL growth without overly differentiating them in spite of the aggressive proliferation spurred by 4-1BB costimulation. Finally, despite the fact that the anti-4-1BB mAb favored expansion of distinct CD8⁺ T-cell clones from the tumor in comparison to IL2 alone, TILs derived with either culture condition showed tumor recognition via IFNγ secretion and 4-1BB upregulation. Although only two TIL lines and paired autologous tumor targets were able to be tested, we observed a higher frequency of antitumor reactive CD8⁺ TILs in the a4-1BB grown cultures compared with TILs grown in IL2 alone. These results suggest that the 4-1BB mAb can facilitate TIL ACT for PDAC by increasing the final yield of the desirable, antitumor CD8⁺ T cells clones present in PDAC.

PDAC has been characterized as an immunologically “cold” tumor due to the low presence of T cells, an immunosuppressive tumor infiltrate, the existence of dense stroma that supports tumor growth, and the lack of response to checkpoint blockade (12–14, 25). However, this merits reconsideration as more recent work demonstrates that the presence of tertiary-lymphoid structures in pancreatic tumors and presence of TILs confer a survival advantage, suggesting the TILs are exerting a degree of tumor control (21, 38–40). Our work adds to these findings by showing that these CD8⁺ cells present in the fresh tumors have a desirable activation status and thus would be the type of TILs to exercise this tumor control. The low percentage of CD45RA (10% ± 6%) in general combined with PD-1 expression (45% ± 12%) suggests the population is mainly antigen-experienced but not terminally differentiated (41). In addition, other work in melanoma has shown that PD-1 can identify the patient-specific CD8⁺ tumor-reactive TILs, but further work is needed to confirm if this is also the case for PDAC (42).

Moreover, the lack of response to immune checkpoint blockade that contributes to the immunologically “cold” designation could be attributed to the paucity of CD8⁺ TILs as shown by our work and others (21, 38, 43). Although the desmoplasia characteristic of PDAC is indicated as a physical barrier to TIL infiltration, it is curious that CD8⁺ TIL are disproportionately affected. Pancreatic stellate cells (PSC) in the surrounding stroma have already been implicated in the lack of CD8⁺ TIL by sequestering them in the stroma, but we observed the same trend in liver metastases (44). Even though hepatic stellate cells (HSC) reside in the liver and are very similar to PSCs, the question remains if HSCs have a similar effect (45). Could additional immunosuppressive mechanisms be in place that would result in the dearth of CD8⁺ TIL in metastatic PDAC? The loss of MHC class I can be a mechanism of immune evasion by various tumors and that its loss is correlated with a decrease in CD8⁺ TIL infiltration (46–48). However, we detected high MHC class I expression in all our tested samples, but cannot rule out the possible loss of a specific allele which this global assessment would not detect. Another possibility for the lack of CD8⁺ TIL infiltration could be due to the relatively low amount of somatic mutations in PDAC (49). The fewer mutations could result in a less immunogenic tumor that does not stimulate as robust an immune response from effector T cells as has been suggested for multiple other tumor types (50). However, recent analysis of data from The Cancer Genome Atlas revealed that high T-cell cytolytic activity is not linked with increased mutational burden in PDAC (51). Further analysis is needed to better understand the reasons behind the low CD8-to-CD4 ratio in PDAC, although our proposed use of a 4-1BB mAb would address this issue for in vitro expansion.

TCR sequencing of the T cells found in the blood, tumor, and TIL cultures provides further insight into the TIL population we are propagating from the tumor. Comparison between autologous tumor–blood and tumor–normal tissue pairs showed enrichment of shared T-cell clones in the tumor, again suggesting an ongoing immune response that would be atypical of a “cold” tumor. These data corroborate recent work which showed the same enriched T-cell repertoire in primary PDAC tumors versus the blood (38). Although the T-cell clones found in the tumor were not enriched in the normal tissue for most patients, three normal tissue–tumor pairs (MP64B, MP75 and MP81) exhibited similarity between the frequencies of shared TIL clonotypes (m > 0.5), particularly MP75. Furthermore, in all samples assessed by partitioning, the top clones shared between both sites exhibited a good correlation (r_s > 0.44) as well as appeared to cluster along the hashed line in all but MP31. These suggest that high-frequency clones found in the tumor are also found at high frequency in adjacent normal tissue. This observation is not true of the blood (i.e., high frequency clones in the tumor are not necessarily circulating at high frequency in the blood), nor are the lower frequency tumor-associated clones enriched in the blood. Altogether, these suggest that PDAC TILs are being pulled from a local, tissue-resident immune response as opposed to a systemic one.

Although a4-1BB triggered massive expansion of CD8⁺ TILs from tumor fragments, we did not observe any signs of further differentiation as the majority of the expanded TILs were effector memory cells lacking CD27 and CD28 (EM3). This subset has been described as highly cytotoxic and we do observe high levels of Granzyme B expressed in the TIL product (36). Although we do detect CD57 expression on a4-1BB grown TILs, we do not detect any KLRG1 expression either, indicating that these cells are likely not senescent.

Our data show reactivity to autologous tumor targets, which builds upon the evidence for tumor-reactive TILs in PDAC as recently shown by other groups (24, 38, 52). The fact that many of the top clones detected in the expanded cultures were not top clones in the tumor and often were low frequency initially is notable. Although Pasetto and colleagues demonstrated in melanoma that the clones that are enriched in the tumor tend to be highly tumor-reactive, our data indicate that low-frequency TILs have antitumor potential in PDAC as well (53). This suggestion is not without precedent since Tran and colleagues detected very low-frequency mutation-reactive TILs in gastrointestinal tumors (19). Further analysis of the T-cell repertoires in each culture condition revealed that they favored expansion of distinct repertoires. These pre-REP repertoires were previously shown in melanoma to be conserved during the rapid expansion protocol (54). This result is consistent with the fact that the 4-1BB mAb selects recently activated TIL whereas IL2 only provides a general T-cell growth signal. However, direct comparison of the clonal composition of the repertoires expanded in both culture conditions is not possible because of the unavoidable bias of tissue sampling introduced at the onset of the culture. Although the repertoires are different, TILs cultured under both conditions displayed tumor recognition.

In conclusion, the data display that metastatic PDAC harbors tumor-reactive TILs and the use of a 4-1BB mAb (Urelumab) can potentiate their expansion while preserving antitumor function. Although the 4-1BB costimulation doubles the growth success, it remains to be answered what factors prevent TIL outgrowth for 50% of samples despite their presence. Regardless, 4-1BB mAb use can enable ACT with bulk TILs by generating a focused, yet oligoclonal T-cell repertoire retaining diverse antigenic specificity mitigating the chances of tumor escape through antigen loss. In addition, it is worth commenting that the effectiveness of TIL ACT in PDAC could be enhanced by combination with therapies that would target the immunosuppressive environment in the tumor and surrounding stroma. Ultimately, implementation of a 4-1BB mAb in TIL production would effectively double the number of patients eligible for therapy and provide a promising treatment option for metastatic disease.

M. Forget and C. Bernatchez report receiving commercial research grants from Iovance Biotherapeutics. J. Rodriguez-Canales is an employee of Medimmune. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M.-A. Forget, P. Hwu, A. Maitra, C. Haymaker, C. Bernatchez

Development of methodology: D. Sakellariou-Thompson, M.-A. Forget, V. Bernard, Y. Kang, J. Rodriguez-Canales, J.B. Fleming, H.A. Alvarez

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. Sakellariou-Thompson, C. Creasy, V. Bernard, Y.U. Kim, M.W. Hurd, C.A. Bristow, J. Rodriguez-Canales, J.B. Fleming, G. Varadhachary, M.J. Overman, H.A. Alvarez, P. Hwu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Sakellariou-Thompson, L. Zhao, N. Uraoka, E.R. Parra, M. Javle, M.J. Overman, J. Zhang

Writing, review, and/or revision of the manuscript: D. Sakellariou-Thompson, M.-A. Forget, V. Bernard, M.W. Hurd, N. Uraoka, E.R. Parra, M.J. Overman, P. Hwu, A. Maitra, C. Haymaker, C. Bernatchez

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T.P. Heffernan

Study supervision: M. Javle, C. Haymaker, C. Bernatchez

The authors would like to thank Alexandre Reuben for help with the hive plots, Feven Malu for help maintaining the tumoroid cell line, and Barbara Mino for help with IHC. We want to thank Bristol-Myers Squibb for their generous contribution with the agonistic anti-4-1BB antibody, Urelumab (BMS-663513). Human recombinant IL2 (Proleukin) was generously provided by Prometheus Therapeutics and Diagnostics. Finally, the authors would like to thank the past and present MDACC TIL lab members: Orenthial J. Fulbright, Arely Wahl, Esteban Flores, Shawne T. Thorsen, René J. Tavera, Renjith Ramachandran, and Audrey M. Gonzalez.

The study was supported by generous philanthropic contributions to The University of Texas MD Anderson Moon Shots Program. Additional support was provided by The University of Texas MD Anderson Cancer Center Support Grant (P30-CA16672) to the flow cytometry, research histology, and characterized cell line cores, the MDACC Institutional Tissue Bank, Multidisciplinary Research Proposal, and a supplement to this grant for the T-cell therapy program. Partial funding for maintenance of tumoroid cell line provided by the Cancer Prevention Research Institute of Texas (RP140106 to V. Bernard).

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.