Neoantigen-specific tumor infiltrating lymphocytes in gastrointestinal cancers: a phase 2 trial

Abstract

Adoptive transfer of unselected autologous tumor infiltrating lymphocytes (TIL) has mediated meaningful clinical responses in patients with metastatic melanoma, but not in cancers of gastrointestinal epithelial origin. In an evolving single-arm phase 2 trial design, TIL were derived from and administered to 91 patients with treatment-refractory mismatch repair-proficient metastatic gastrointestinal cancers in a schema with lymphodepleting chemotherapy and high-dose interleukin-2 (three cohorts of an ongoing trial). The primary endpoint of the study was the objective response rate as measured using RECIST 1.0; safety was a descriptive secondary endpoint. In the pilot phase, there were no clinical responses in 18 patients to bulk, unselected TIL; however, when TIL were screened and selected for neoantigen recognition (SEL-TIL), three responses were seen in 39 patients (7.7% [95%CI 2.7–20.3]). Based on the high levels of PD-1 in the infused TIL, pembrolizumab was added to the regimen (SEL-TIL + P), and eight objective responses were seen in 34 patients (23.5% [95% CI 12.4–40.0]). All patients experienced transient severe hematologic toxicities from chemotherapy. Seven (10%) patients required critical care support. Exploratory analyses for laboratory and clinical correlates of response were performed for the SEL-TIL and SEL-TIL + P treatment arms. Response was associated with recognition of an increased number of targeted neoantigens and an increased number of administered CD4+ neoantigen-reactive TIL. The current strategy (SEL-TIL + P) exceeded the parameters of the trial design for patients with colorectal cancer and an expansion phase is accruing. These results could potentially provide a cell-based treatment in a population not traditionally expected to respond to immunotherapy. ClinicalTrials.gov identifier: NCT01174121.

Introduction

Adoptive cell transfer (ACT) of ex vivo-expanded unselected tumor infiltrating lymphocytes (TIL) plus interleukin-2 (IL-2) has demonstrated efficacy in the treatment of advanced melanoma including durable complete response rates higher than 20%^1–5. Melanoma TIL can recognize melanoma differentiation antigens as well as neoantigens - the products encoded by somatic mutations expressed in the autologous cancer⁶. High tumor mutational burden (TMB) has been directly associated with clinical response to other forms of immunotherapy, specifically the widely used immune checkpoint inhibitors (ICIs) pembrolizumab and nivolumab^7–10. Perhaps unsurprisingly, the two cancer histologies with the highest mutational burden, cutaneous melanoma and non-small cell lung cancer, have been the only cancers reproducibly shown to respond to unselected TIL^{5 11–13}. Thus improved methods of cell-based immunotherapies are needed to treat patients with common solid metastatic epithelial cancers, a group of cancers that are responsible for most cancer deaths^14–17 Objective responses to TIL in clinical trials in patients with metastatic gastrointestinal (GI) cancers have not been reported^{18 19}.

Because TIL recognize their targets via human leukocyte antigen (HLA)-mediated presentation of target peptides, their anticancer function can be highly specific while also broadly polyclonal, representing an appealing approach for the personalized treatment of solid cancers. Recently, it has been shown that 1.6 to 2.7% of mutations in the cancer can give rise to neoantigen-reactive T cells^{20 21}, and that the majority of patients with GI tumors are able to generate neoantigen-reactive TIL²⁰. However, when unselected TIL, manufactured according to principles developed in the study of metastatic melanoma, were administered to the first 18 patients with GI cancers in the pilot arm of the current study, there were no objective clinical responses. Although T cells targeting mutation-encoded cancer neoantigens in epithelial tumors were present within infusion products, a high percentage of T cells appeared to be bystanders with unknown reactivity²² or represented tumor-reactive T cells that were present in a terminally exhausted state²³. Post-hoc analysis of TIL from one patient who received unselected TIL and did not respond allowed for re-treatment of the same patient with a single selected neoantigen-reactive TIL culture (~95% reactive) and resulted in a partial response of liver and lung metastases lasting 34 months¹⁴. This promising anecdote informed the current study design, with prospective identification of mutation reactivity of TIL cultures prior to clinical expansion for treatment. The question remained whether TIL selected for tumor reactivity may elicit clinical responses in a larger cohort of heavily pretreated patients with metastatic GI cancers, which led to the “selected TIL” strategy that was employed for the 73 patients reported herein (Table 1).

Table 1.

Characteristics of Patients at Baseline prior to Adoptive Transfer

Characteristic	Bulk TIL (n=18)	Selected TIL (n=39)	Selected TIL + Pembrolizumab (n=34)
Age
Median (range) - yr	49 (37–63)	52 (37–69)	50 (25–71)
Distribution - no. (%)     ≤50 yr	10(56)	17 (44)	18 (53)
51–60 yr	7 (39)	16 (41)	6 (18)
>60 yr	1 (6)	6 (15)	10(29)
Sex - no. (%)        Female	9 (50)	24 (62)	16(47)
Male	9 (50)	15 (38)	18 (53)
ECOG performance-status score - no. (%)         0	12(67)	32 (82)	27(79)
1	6 (33)	7 (18)	7 (21)
Diagnosis - no (%)
Upper Gastrointestinal	3 (17)	2 (5)	2 (6)
Esophageal Adenocarcinoma	2	0	1
GEJ Adenocarcinoma	0	0	1
Gastric Adenocarcinoma	1	2	0
Hepatopancreaticobiliary	4 (22)	8 (21)	4 (12)
Cholangiocarcinoma	4	5	1
Pancreatic Adenocarcinoma	0	3	3
Lower Gastrointestinal	11 (61)	29 (74)	28(82)
Colon Adenocarcinoma	9	22	22
Rectal Adenocarcinoma	2	7	6
Early Onset Colorectal (% of Lower GI)	10 (91)	17(59)	17 (61)
Primary Tumor In Situ - no. (%)	4 (22)	6 (15)	11 (23)
Metastatic Disease
Synchronous - no. (%)	7 (38)	15 (39)	22 (65)
Metachronous - no. (%)	11 (62)	24 (51)	12 (35)
Median Metastasis Free Interval (range) - mo	16.2 (4.2–113)	19.1 (4.2–100)	18.2 (12.6–36.2)
Prior Therapy - no. (%)
Median (range) - no.	4(2–10)	4(1–9)	5 (1–11)
Distribution - no. (%)     1–3	6 (33)	13(33)	10(29)
4–6	7 (39)	19(49)	20(59)
>6	5 (28)	7 (18)	4 (12)
Chemotherapy - no. (%)
5-Fluorouracil	17 (94)	37(95)	34 (100)
Oxaliplatin	15 (83)	33 (85)	33 (97)
Irinotecan	15 (83)	29 (74)	30 (88)
Bevacizumab	13 (72)	25 (64)	26(76)
Cetuximab	4 (22)	6 (15)	4 (12)
Regorafenib	1 (6)	5 (13)	6 (18)
Gemcitabine (% of HPB)	4 (100)	8(100)	3 (75)
Paclitaxel (% of HPB)	2 (50)	3 (38)	3 (75)
Cisplatin (% of HPB)	2 (50)	3 (38)	1 (25)
Taxotere (% of HPB)	2 (50)	1 (13)	0 (0)
Immune Checkpoint Inhibitor - no. (%)	0 (0)	1 (3)	4 (12)
Surgery (of Metastatic Disease) - no. (%)	7 (39)	19(49)	19(56)
Ablation (of Metastatic Disease) - no. (%)	3 (17)	4 (10)	1 (3)
Radiation - no. (%)	8 (44)	14 (36)	12 (35)
Time to treatment - Median (range) months	1.3 (1.2–6)	3.5 (1.6–28.7)	4.4 (3.0–23.9)

Results

Clinical trial overview

Between March 8, 2010 and August 31, 2023, 227 patients with metastatic gastrointestinal cancer had tumors resected for the growth of TIL (CONSORT diagram in Fig. 1). Key eligibility criteria included measurable treatment-refractory metastatic cancer, good performance status, cardiopulmonary fitness, absence of multiple or symptomatic brain metastases, and absence of major organ autoimmune disease. Please see Methods for more information on trial design.. The earliest 18 patients were treated with bulk unselected TIL, with no objective clinical responses, while TIL from 168 subsequent patients was prospectively screened for neoantigen reactivity (Extended Data Fig. 1A). A representative TIL screen is shown in Extended Data Fig. 1B, wherein TIL fragment cultures from pancreatic cancer #4493 demonstrated both CD4 and CD8 reactivity. Testing of the corresponding final TIL infusion product demonstrated >60% reactivity against the three target neoantigens (Extended Data Fig. 1C–D).

Figure 1. CONSORT diagram.

Diagram shows patients with metastatic epithelial cancer who underwent surgical resection of metastatic tumors for the purpose of treatment with TIL during the period reported. Laboratory and treatment details provided for patients with metastatic gastrointestinal cancer (including esophageal, gastric, pancreatic, cholangiocarcinoma, and colorectal cancer).

Of patients whose TIL were prospectively screened for neoantigen recognition, 95/168 (57%) showed at least one fragment culture with strong reactivity. Median time from resection to completion of screen was 2.6 months (1.3–8.2), and most were completed within 3 months (79%, 132/168). In total, 73 patients received selected TIL (schema in Extended Data Fig. 1E). 39 patients were treated with selected TIL alone (SEL-TIL); two patients in the SEL-TIL group have been the subjects of prior case reports^{14 15}. Due to the high levels of PD-1 in the infused TIL, an additional 34 patients received pembrolizumab immediately prior to TIL (SEL-TIL + P). Time to treatment was variable based on interval oncologic care, and most patients received cells within six months of resection (56/73, 77%). Patients receiving selected TIL were heavily pre-treated, having received a median of 4 lines of prior therapy. While this report described patients with metastatic upper GI (cohort A), hepatopancreaticobiliary (Cohort B), and lower GI (Cohort C) cancers, cancers of primary colon origin represented 44/73 (60%) of those receiving selected TIL (Table 1). To allow for follow-up analysis, the data cutoff for treatment was August 2023 and survival analysis was May 2024. Arms accrued consecutively during translational development (August 26, 2010 to July 19, 2023), and the final primary objective was to evaluate SEL-TIL + P (Arm 3). Clinical and laboratory correlates of response, including TIL product characteristics, were exploratory endpoints. Safety was a descriptive secondary endpoint with an expected toxicity profile similar to published trials of adoptive cell transfer. All data were analyzed in an as-treated rather than per-protocol basis, as described in the methods.

Efficacy

Following treatment with selected TIL, 11/73 (15.1% [95% CI 8.6–25.0%]) patients experienced confirmed partial responses including 3/39 patients (7.7% [95% CI 2.7–20.3%]) in the SEL-TIL group and 8/34 (23.5% [95%CI 12.4–40.0%]) in the SEL-TIL + P group (Figure 2). Objective responses by RECIST criteria were seen in patients with cholangiocarcinoma, pancreas, colon, and rectal cancer. The durations of response were 8, 24, and 70+ months in the SEL-TIL group and 4, 7, 7, 8, 10, 11, 17+, and 42 in the SEL-TIL + P group. The depth and duration of responses of individual patients are shown in Figure 2.

Figure 2. Clinical activity of treatment.

A. Percent changes in RECIST 1.0-defined target lesion measurements following SEL-TIL administration. B. Percent changes in RECIST 1.0-defined target lesion measurements following selected TIL + pembrolizumab (SEL-TIL + P) administration. In A. and B, baseline imaging was performed a median of 9 days prior to cell infusion (IQR 8–11d), and lesions were first assessed at 4–6 weeks post-TIL administration, then at 8–12 weeks and every 2–3 months thereafter until progression of disease. NED: No evidence of disease. C. Swimmer’s plot showing overall survival following treatment with SEL-TIL or SEL-TIL + P. Surviving patients are indicated by black arrows. Time of partial response (PR) is denoted by diamonds, with open diamonds the unofficial imaging-based PR start and solid diamonds the confirmed PR time point. “×” refers to time of disease progression, and “+” refers to timing of a post-TIL surgical intervention. Data cutoff was May 2024. The four patients with >48 month survival, including two previously reported cases, extend to 126¹⁴, 106¹⁵, 86, and 76 months, respectively.

Objective responses occurred in metastases to the liver, lungs, lymph nodes, subcutaneous tissue, and bone. The patient treated with the TIL characterized in Extended Data Fig. 1B–D, who had pancreatic ductal adenocarcinoma metastatic to the liver, lymph nodes, and peritoneum, demonstrated a partial response. This included complete resolution of one liver metastasis and overall tumor shrinkage of 44.1% post-TIL (Extended Data Fig. 1F) before progression of non-target disease at 7 months. A liver tumor biopsied at progression demonstrated loss of the HLA locus required for CD8 TIL recognition. Other examples of patients who experienced objective responses in bone, liver, and lung are shown in Figure 3.

Figure 3. Evidence of tumor regression at different metastatic sites.

A. Baseline (left) and post-treatment (right) cross-sectional imaging of a patient with cholangiocarcinoma in the SEL-TIL + P arm. Arrows indicate tumors in bone (top) and paraaortic lymph nodes (bottom). Resolution in both sites at 9 months. The site of the bone metastasis displayed evidence of healing. Additional sites of disease included liver, lung, and adrenal gland (not shown). B. Baseline (left) and post-treatment cross-sectional imaging of a patient with rectal cancer in the SEL-TIL + P arm. Circles highlight tumors in liver. Reduction in both sites at 10 months (right). The patient had additional disease present in lungs and lymph nodes (not shown) and progressed with a new site of disease at 11 months. C. Baseline (left) and post-treatment (right) cross-sectional imaging of a patient with colon cancer in the SEL-TIL + P arm. Arrows indicate tumors in right (top) and left (bottom) lung. As shown, two nodules completely resolved and others were smaller at 7 months. Other sites of disease included liver and a colonic recurrence. Eventual progression of disease at 10 months.

In addition to the 11 objective responses that met RECIST criteria, eight additional patients experienced target tumor reduction greater than 30% at one follow-up assessment (3 in the SEL-TIL group and 5 in the SEL-TIL + P group) (Extended Data Fig. 2A). Tumor regression in patients without RECIST response are exemplified by a patient with pancreatic cancer who exhibited complete regression of dozens of liver metastases at the 6-week follow-up (Extended Data Fig. 2B) but recurred at three months with three tumors, one of which when biopsied did not harbor the targeted TP53 mutation. Dozens of lung metastases regressed in another patient with pancreatic cancer at the first follow-up (Extended Data Fig. 2c) but symptomatic brain metastases developed soon thereafter. Similarly, a patient with metastatic colon cancer underwent near-complete regression of multiple lung (Extended Data Fig. 2D) and liver metastases lasting 10 months, however developed new disease (brain metastasis) prior to achieving a partial response. These and similar patients suggest that the sole use of RECIST criteria likely underestimates the true impact of transferred TIL on metastatic epithelial cancers as it cannot capture mixed responses that may be of immunologic importance.

Safety

Because the treatment regimen is designed to induce severe, but temporary, lymphopenia prior to the transfer of TIL, all patients experienced grade ≥3 adverse events (AE) (Table 2, Table S1). Serious AE occurred in 22 (30%), with one treatment-related death attributed to adenoviral hepatitis 49 days after TIL infusion. Seven patients required escalation of care for critical support: three for mechanical ventilation, one for continuous renal replacement therapy, and three required both modalities (including one patient with rapidly progressive disease). In patients that received pembrolizumab prior to cells, 8 (24%) experienced a grade ≥3 AE, with two serious AE related to pembrolizumab. One patient developed treatment-refractory inflammatory colitis, while another developed steroid-responsive pneumonitis. The majority of lower grade AEs (Table S1) were consistent with chemotherapy (e.g. nausea, vomiting, fatigue) and/or interleukin-2 (e.g. diarrhea, hypokalemia, increased creatinine).

Table 2.

Patients experiencing Treatment-related Adverse Events related to Research (Grade ≥3)^*

	Selected TIL (n=39)	Selected TIL + P (n=34)
All Grade ≥3	39 (100)	34 (100)
Attributed to Cells	10 (26)	8 (24)
Attributed to Pembrolizumab	n/a	8 (24)
Serious Grade ≥3 Events	11 (28)	11 (32)
Attributed to Cells	6 (15)	4 (12)
Attributed to Pembrolizumab	n/a	2 (6)
Treatment-related Death	0 (0)	1 (3)
Expected Grade ≥3 Cytopenia	39 (100)	34 (100)

^*highest grade for each, e.g. a resolving Grade 4 is not also counted as a Grade 3, etc., a patient may experience more than one event

Clinical correlates of response

Pretreatment patient characteristics and correlations with RECIST responses are presented in Extended Data Table 1. There was a trend towards higher age among partial responders to SEL-TIL + P (p=0.037). Additionally, in the SEL-TIL + P group, patients with metachronous metastatic disease exhibited a response rate of 50% (6/12), compared with 9% (2/22) in patients with synchronous metastases (p=0.013). Median time to treatment was not significantly different. Finally, response to TIL treatment was negatively associated with the number of lines of prior therapy, as responders had received a median of 3 (range 1 – 6) prior lines of therapy, while non-responders received a median of 5 (range 2 – 11) (p=0.027).

TIL infusion product characteristics

Patients received a median of 7.3×10¹⁰ (2.1 – 15.2×10¹⁰) cells in the SEL-TIL group and 8.5×10¹⁰ (2.5 – 14.7×10¹⁰) cells in the SEL-TIL + P group. There was no significant difference in likelihood of response based on the total number, CD4+, or CD8+ cells infused (Extended Data Table 2), although all patients received greater than 2×10¹⁰ cells. Because the response rate to the SEL-TIL + P infusion products appeared most promising, the functional reactivity of all 34 infusion products within this group toward their intended targets was determined. This analysis showed that these infusion products had a median of 34% (IQR 9.9–52%) reactivity against their intended targets, for a total estimate of 2.0×10¹⁰ (0.02 – 8.2×10¹⁰) reactive cells infused. When looking only at tumor antigen-reactive TIL, neither the total number of cells nor CD8+ cells correlated with response (Figure 4A); however, the number of infused antigen-reactive CD4+ cells was higher among responders than non-responders (median 2.4×10¹⁰ vs. 0.4×10¹⁰, p=0.041).

Figure 4. Characteristics of treatment of SEL-TIL + P arm.

Data (n=34) are divided by clinical response (PR: partial response; NR: no response). Shapes denote diagnosis (diamond: Cohort A/Upper GI, square: Cohort B/HPB, circle: Cohort C/Lower GI). Median values (A-C) are indicated by red bars. Comparisons calculated by Mann-Whitney 2-sided test (A-D); NS denotes p-values > 0.05. A. Numbers of reactive cells infused among all TIL (left), CD4 TIL (center, * p=0.041), and CD8 TIL (right). B. PD-1 expression was assessed on cryopreserved TIL samples. Cells were thawed and rested overnight without cytokines prior to staining with anti-PD-1 clone EH12.2H7 (BioLegend). C. Numbers of CD39-CD69-cells in CD3+, CD4+, and CD8+ TIL (left to right). D. Total selected targets recognized by infusion products in SEL-TIL + P cohort based on experimental testing (* p=0.013), with box plots showing 25^th-75^th percentile, whiskers denoting range, and black bars indicating median. E. Increasing likelihood of partial response (%) with an increasing number of targets (p=0.021, chi-square test for trend).

The immune checkpoint PD-1 was highly expressed in all the infusion products (Figure 4B) but did not differ significantly between responding and non-responding patients. Stem-like (CD39-/CD69-)²⁴ CD8+ TIL were present in the infusion products (Figure 4C), but at lower frequencies than in melanoma infusion products, and were not correlated with response. The median number of tumor-encoded targets in the TIL differed between responders and non-responders as well, with responders receiving TIL with a median of 3 (2 – 5) targets and non-responders only targeting 2 (1 – 4) antigens (p=0.013, Figure 4D). Patients had a higher likelihood of response with increasing targets (p=0.021, Figure 4E). The individual patient and treatment characteristics of the 34 patients in the SEL-TIL + P group including the target antigens are shown in Extended Data Table 3 and Supplementary Table S2. The majority of identified tumor targets were unique somatic mutations of unknown significance, although 22 of the total 81 defined mutation-encoded TIL targets were found in Cancer Gene Census (CGC) Tier 1 genes, suggesting they may provide functional benefit to cancer cells. Only two mutant proteins were recognized by TIL from more than a single patient, derived from hotspot mutations in KRAS and TP53. All other antigens recognized were unique to the autologous patient, thus emphasizing the highly personalized nature of TIL immunotherapy.

Tumor genomic features

Whole-exome sequencing data was obtained from all tumors resected for selected TIL screening. Tumors resected for the generation of TIL had a median tumor mutation burden (TMB) of 5.29 mutations/Mb, which falls below typical thresholds for TMB-high tumors and clinical indications for ICI therapy^{8 25}. TMB did not distinguish responders and non-responders within either the SEL-TIL or SEL-TIL + P groups (Extended Data Table 2). In terms of tumor mutational clonality, tumors showed a median of 98% (67–100%) of somatic mutations as clonal²⁶, and again no differences were observed according to clinical response. Among the total of 81 defined mutation-encoded TIL targets in the SEL-TIL + P arm, 77 were clonal in the corresponding tumor resected for TIL growth (95%, Table S2).

In order to determine whether features associated with ultimate clinical response could be detected within the tumors resected for TIL growth, differential gene expression analysis was performed between bulk RNA-seq libraries from tumors of responding (n=8) and non-responding patients (n=25) within the SEL-TIL + P group. This comparison showed 940 differentially expressed genes (DEGs) (adjusted p < 0.05, log₂FC > 2, Extended Data Fig. 3A, Table S3). Ingenuity Pathway Analysis revealed the top responder-enriched pathways to be associated with inflammation and wound healing, while non-responder enriched pathways were largely related to transcriptional and translational processes (Extended Data Fig. 4A). Hallmark gene-set enrichment analysis (GSEA) of these DEGs similarly highlighted immune-related features (such as inflammatory response, allograft rejection, and IL-2-STAT5 signaling) in responders and cancer cell-associated processes (including glycolysis, oxidative phosphorylation, and fatty acid metabolism) in non-responders (Extended Data Fig. 3B). Clustering of tumor RNA-seq libraries (SEL-TIL + P) according to the top and bottom DEGs showed four clusters that appeared to assemble according to ultimate clinical TIL response, with cluster 1 and 2 comprised of tumors whose TIL elicited 7/16 responses, while TIL from tumors in clusters 3 and 4 only led to responses in 1/17 patients (Extended Data Fig. 3c). DEGs derived from the SEL-TIL + P analysis were applied to the tumors from the SEL-TIL group (n=36 tumors, three from responders and 33 from non-responders). In a cluster analysis, four similar clusters appeared, despite the lower response rate in this treatment group (Extended Data Fig. 4B). Together, these data suggest that distinct transcriptional states of tumor microenvironments of the resected tumors might be informative in understanding SEL-TIL response in GI patients.

Discussion

Immunotherapy has revolutionized the care of patients with cancer and can provide long-lasting durable responses for some patients. For patients with rare mismatch repair-deficient tumors, those benefits are substantial, but have not yet extended to patients with more common gastrointestinal cancers, a cohort expected to suffer over 174,000 deaths in the U.S. this year, and increasingly common in younger people²⁷. The failure of immune checkpoint inhibition to eliminate metastatic disease in these patients suggests the need for a more highly personalized method of amplifying and activating immune cells to elicit an anti-cancer response. Genetically-engineered approaches targeting cancer differentiation antigens (e.g., gp100, carcinoembryonic antigen [CEA], or claudin-18.2) have met with limited success, demonstrating on-target, off-tumor toxicity^{28 29} or limited durations of response when cells were administered in conjunction with cytotoxic chemotherapy³⁰. Here we report interim results of a non-randomized, phase 2 trial comparing the use of tumor reactivity-selected TIL with or without pembrolizumab for the treatment of metastatic gastrointestinal cancer. Using RECIST 1.0 criteria, clinical responses were seen in 3/39 patients (7.7%) receiving SEL-TIL compared to 8/34 patients (23.5%) receiving SEL-TIL + P. Additional patients demonstrated clear tumor reduction, usually in a mixed fashion not adequately captured by RECIST metrics, highlighting the ongoing challenge of addressing tumor heterogeneity in patients with advanced disease and complicating the interpretation of factors that may distinguish clinical responsiveness to TIL. The safety profile of the regimen was similar to other studies of TIL.^{11 31 32}

The patients included in this study were heavily pre-treated with a variety of systemic therapies (median 5 [1–11] in SEL-TIL + P), with 100% of patients’ tumors refractory to a 5-fluorouracil-based regimen (e.g. FOLFOX, FOLFIRI, etc.) and most also having previously progressed through regimens including oxaliplatin, irinotecan, and bevacizumab (Table S1). Only a small percentage of patients previously received ICIs, but there is limited evidence of their efficacy (0–2%) when applied to the gastrointestinal cancers treated in this study, as no tumors had evidence of mismatch repair deficiency or high microsatellite instability, and all patients in this study underwent pretreatment lymphodepletion^33–35. Thus, it is highly unlikely that pembrolizumab alone is responsible for the 24% response rate of the SEL-TIL + P group. oreover, two patients whose disease had previously progressed through ICI were among the responders to TIL therapy. While patients had not been previously exposed to cyclophosphamide, fludarabine, or IL-2, these drugs would not be expected to have a clinical impact on gastrointestinal cancers.

The tumor regressions observed following TIL administration may be somewhat unexpected given the relatively low TMB seen in these patients’ tumors and the association of TMB with response to TIL seen in patients with melanoma^{6 36}. The lack of clinical responses among the 18 patients treated with bulk, unselected TIL, as used for patients with melanoma, underscores this point and is consistent with a lack of responses of GI and ovarian cancer to TIL seen elsewhere¹⁸. The ability to enrich for tumor-reactive cells through the TIL selection process may partially correct for the low overall mutational burden relative to patients with melanoma and non-small cell lung cancer that are the predominant tumor types responsive to immunotherapy and re-emphasizes the concept that while relatively few somatic tumor mutations may be immunogenic (1–2%)^{20 21}, the targeting of these relatively rare neoantigens can lead to clinically meaningful tumor regressions.

As TIL represents a living drug, the patient-specific nature of each infusion product makes them unique; assays that would functionally characterize an infusion product by assessing its reactivity towards the intended tumor mutation-encoded neoantigens require the generation of bespoke reagents including autologous antigen-presenting cells, peptides, mini-genes, and in some cases tumor organoid lines. By testing the SEL-TIL + P infusion products against their intended targets, we showed that apart from KRAS and TP53 every other target neoantigen was patient-specific and not shared. While some responses in the SEL-TIL arm occurred in patients receiving cells targeting one known neoantigen^{14 15}, in the SEL-TIL + P group the number of targeted antigens, while low, appeared to impact response. Developing strategies that result in the generation of treatment products that target a more diverse set of antigens to counter tumor heterogeneity and preempt resistance represents an approach that may enhance response rates to ACT in tumor types with relatively low TMBs.

While it is not always clinically possible to investigate progressing or recurring tumors after TIL therapy, anecdotal evidence has suggested mechanisms of resistance including loss of heterozygosity (LOH) of targeted HLA class I alleles^{15 37} as well as absence of mutated targets themselves, both examples of active immune editing and tumor heterogeneity that could be circumvented by targeting more class I- and II-restricted antigens. Further, while the number of CD8+ neoantigen-reactive TIL did not correlate with the likelihood of response, higher numbers of reactive CD4+ TIL were given to responding patients, suggesting a heightened importance of these cells in eliciting clinical responses^{14 38 39}. While many efforts to understand the mechanisms of successful TIL-based immunotherapy in melanoma have focused on CD8 TIL^{24 36 40}, it is noteworthy that in the initial report of the efficacy of TIL co-administered with IL-2 to patients with melanoma¹, 7/11 responding patients received a higher number of CD4+ than CD8+ TIL, including a 6-month PR following a cell infusion of 98% CD4+ TIL. While tumor-specific CD4 and CD8 TIL may act in concert to eliminate tumors³⁹, patients receiving both CD4 and CD8 tumor-reactive TIL did not achieve statistically better outcomes in our small SEL-TIL+ P cohort (6/19 responses with both CD4 and CD8, 2/15 with either CD4 or CD8). These exploratory correlatives may be underpowered by the small number of patients involved. Highlighting the living nature of the product, T cell proliferation during the final manufacturing phase is non-specific with considerable patient-to-patient heterogeneity, and maintenance of reactivity seen in pre-treatment screening assays remains an area of ongoing study. Future efforts that involve the selection of TIL on a cellular rather than a population level, such as cell surface marker-based sorting of bulk TIL, or in vitro sensitization (IVS) to specifically stimulate the proliferation of neoantigen-reactive TIL⁴¹, may enhance the ability to further enrich TIL for tumor reactivity, and such efforts are in progress. These efforts may enable the “on-demand” selection of both CD4+ and CD8+ reactive TIL and combination of such cells at desired doses. While TIL growth is a necessarily personalized process, selection of TIL for tumor relevance based on generalizable phenotypic markers may simplify the process and allow for selection without personalized screening reagents.

In the setting of metastatic melanoma, we previously identified an association between the number of stem-like (CD39-/CD69-) CD8+ TIL administered and response²⁴, yet in this cohort the median number of CD39-/CD69-CD8+ TIL did not distinguish between clinical response and was 10-fold lower overall. The clinical addition of pre-infusion pembrolizumab was supported by the presence of PD-1 on most cells in the final infusion products. Future studies will be necessary to specifically analyze the phenotypes of the mutation-reactive TIL within these infusion products, as bulk analyses that disregard known reactivity may be confounded by bystander cells of unknown reactivity.

The concept that there are transcriptomic features within the tumors harvested for generation of a TIL product that ultimately associate with the patients’ clinical response to TIL is compelling. While each of these tumors harbored lymphocytes with in vitro capacity to recognize neoantigens, features related to specific tumor biology and sensitivity to immunotherapy may be responsible for the range of clinical responses in vivo after adoptive transfer. While TIL are usually derived from a single source in patients with multiple sites of metastatic disease, the transcriptome-based co-clustering of tumors from patients with clinical response suggests relative intrapatient homogeneity. While a bulk RNA-seq analysis does not point directly to a specific causative cell type or network underlying the later response, numerous responder-associated genes were related to chemoattraction (CXCL9/CXCL10/CXCL11), costimulation (CD80/CD86), monocyte/myeloid lineages (CD14, S100A8/S100A9), and MHC class II antigen presentation (HLA-DPA1, HLA-DQB1, HLA-DRA), indicative of a antitumor immune microenvironment^{42 43} and reinforcing the likelihood that tumor-reactive CD4 cells play a key role in the response of GI tumors to TIL therapy. Conversely, the high enrichment of translation factors and cancer-related genes in the non-response-associated gene expression profiles are suggestive of a specific cancer cell-driven microenvironment in these patients. Future pathology, single-cell, and spatial studies will investigate what exact cellular networks are driving these transcriptomic signals, and whether features identified earlier in a patient’s disease course (primary tumor, biopsy of metastasis, ctDNA) may inform whether or not the patient may be a good candidate for TIL or other immunotherapies.

The cellular immunotherapy strategy described herein required bespoke tumor-informed reagents to guide the selection of cryopreserved TIL cultures for treatment. The highly, personalized nature and level of immunologic expertise required for implementation limits its current applicability outside the scope of experimental trials. However, the insights that have been achieved by deeper evaluation of the treatment products may yield ways to curate an effective lymphocyte population without the need for bespoke analysis. With improved selection techniques, the delay between resection and infusion should decrease for each potential recipient and further improve the feasibility of the process.

The interim results of this single-institution trial demonstrate that TIL selected for neoantigen reactivity can mediate tumor regression even in treatment-refractory metastatic GI cancers that are not thought to be sensitive to immunotherapy. Other advanced epithelial cancers not included in this report may also merit investigation for the use of selected TIL, as even TMB-low tumors not thought to be ICI-responsive demonstrated clinical responses. Future efforts to target cancer-specific mutations with T cells, either in vitro-expanded TIL, engineered T cells expressing tumor-targeting TCRs⁴⁴, or in vivo tumor-specific activation of endogenous immune cells with therapeutic agents can build upon these results.

Extended Data

Extended Data Figure 1. TIL selection process.

A. Schematic of tumor resection, TIL growth, and TIL screening pipeline. Tumors were surgically removed and dissected into small fragments, which were grown in IL-2 for TIL fragment culture expansion. Additional tumor fragments were sequenced by whole exome and RNA-seq. Based on tumor mutation calling, candidate neoepitopes were generated in vitro (25-amino acid peptides or minigene constructs with mutation-encoded amino acid at the center [13^th] position). Candidate neoepitopes are expressed by autologous dendritic cells in pools (peptide pools [PP] or tandem minigenes [TMG]). TIL fragment cultures are then co-cultured with these candidate neoepitope-expressing dendritic cells or PDTO if available and TIL demonstrating specific TCR-mediated activation (IFNγ release or induction of cell surface 4–1BB (CD137) or OX40 (CD134) following co-culture were selected for potential treatment. B. Example of TIL screening for tumor 4493. From tumor 4493, 12/24 TIL cultures expanded to numbers sufficient for testing. Based on the corresponding tumor sequencing, 48 candidate neoepitopes were screened in 3 PPs and 3 TMGs. Reactivity was observed against TMG3 (CD8+ TIL exhibiting 4–1BB) and PPs 1 and 2 (CD4+ TIL showing 4–1BB/OX40 induction). TIL fragments selected for treatment are indicated with arrows. Cultures not selected for treatment that appear reactive (e.g. F6, with ~15% CD8 reactivity vs. TMG3) were of inappropriate phenotype (e.g. F6 was <20% CD8 or <3% reactive in total ). PDTO was not available for this patient. C. Example of TIL infusion product retrospective testing for tumor 4493. Cryopreserved TIL were separated into CD8+ and CD4+ fractions and co-cultured with autologous dendritic cells expressing multiple concentrations of neoantigenic peptides within their “selected” target TMGs and PPs. The peak activation value (4–1BB for CD8, 4–1BB and/or OX40 for CD4) subtracting out vehicle control (DMSO) was considered the specific reactivity value against a neoantigen. Left, TMG3 reactivity was mediated by CD8+ TIL reactive to mutant DOP1A. Center, PP2 reactivity was mediated by CD4+ TIL reactive to mutant ZFP36L1. Right, PP1 reactivity was mediated by CD4+ TIL reactive to mutant PANK4. D. Reactivity calculations for example infusion product 4493 from C. Peak CD8 reactivity value against mutant DOP1A and CD4 reactivity against mutant ZFP36L1 and PANK4 was used to calculate numbers of reactive CD8 (left), CD4 (center), and all TIL (right). E. Overall clinical schema illustrating timing of cyclophosphamide (Cy), fludarabine (F), TIL, interleukin-2 (IL-2), and pembrolizumab (P) when added. F. Partial response of pancreatic ductal adenocarcinoma liver metastases following treatment with SEL-TIL + P. Magnetic resonance imaging (MRI) of the pre-treatment (left) and post-treatment (right) liver. Post-treatment images were obtained 5 months after 4493 TIL infusion.

Extended Data Figure 2. Regression of target tumors in patients receiving selected TIL.

A. Waterfall plot of maximal change from baseline of target tumors per RECIST 1.0 post-TIL infusion for SEL (left, n=39) and SEL + P (right, n=34) arms. Bars are colored according to primary tumor histology (Lower GI in blue, upper GI in red, HPB in green). Bars labeled with numbers indicate duration of confirmed partial responses, asterisks indicate clinical non-responders with >30% reduction, hexagons indicate the patients with further imaging in panels B-D, and the caret indicates a non-evaluable patient whose disease progressed prior to first follow-up visit. Underlined values represent previously published case reports^14,15 B. Regression of diffuse hepatic metastases in a patient with pancreatic ductal adenocarcinoma. Stable hemangioma noted (Hemang). Baseline (left) and six-week follow-up (right) shown. C. Regression of multiple pulmonary nodules and resolution of pleural effusion in a patient with pancreatic ductal adenocarcinoma. Baseline (left) and six-week follow-up (right) shown. D. Regression of pulmonary tumors in a patient with colon cancer. Baseline (left) and 10-month evaluation (right) shown. Patient is a non-responder for development of a new brain metastasis at 6 months (not shown).

Extended Data Figure 3. Transcriptomic analysis of TIL harvest tumors from SEL-TIL + P arm.

A. Volcano plot of DEGs between TIL harvest lesions of responders (n=8) and non-responders (n=25). Dotted lines indicate adjusted p-values < 0.05 and absolute log₂FC > 2. Highlighted genes include selected immune-related genes and those from IPA-indicated pathways (Extended Data Fig. 4A). B. Normalized enrichment scores (NES) of significantly enriched hallmark gene sets (nominal p-value < 0.05, GSEA) in TIL harvest lesions from responders (gold) or non-responders (blue). C. Clustering of bulk tumor RNA-seq data by patient (SEL-TIL + P) according to top and bottom 100 response-associated DEGs. Z-scaled gene expression is indicated red to blue, and clinical response to TIL is shown below cluster plots (orange for RECIST response, blue for non-response). Responder-enriched clusters 1 and 2 show enhanced expression of response-associated genes, while non-responder-enriched clusters 3 and 4 show heightened expression of non-response-associated genes.

Extended Data Figure 4. Ingenuity Pathway Analysis (IPA) of response-associated DEGs (SEL-TIL + P) and clustering of SEL-TIL tumor RNA.

A. Top 10 response-associated and non-response associated pathways of DEGs within TIL harvest lesions of SEL-TIL + P arm according to IPA. Pathways with highest significant z-scores are shown and ranked by -log₁₀(p-value), with responder-enriched pathways in orange and non-responder-enriched pathways in blue. B. TIL harvest tumor RNA samples from SEL-TIL group (n=36 samples) were clustered according to top and bottom 100 response-associated DEGs of SEL-TIL + P group. Patient response to TIL is shown below, with orange indicating RECIST response and blue indicating non-response to TIL.

Extended Data Table 1.

Clinical correlates of response to TIL selected for neoantigen reactivity (exploratory)

	Selected TIL(n=39)			Selected TIL + Pembrolizumab (n=34)
Characteristic	Partial Response (n=3,7.7%)	Non-Response (n=36)	p-value*	Partial Response (n=8,23.5%)	Non-Response (n=26)	p-value*
Age   Median (range) - yr	49 (45–54)	52 (37–69)	0.38	64 (40–68)	49.5 (25–71)	0.025
Distribution - no. (%of row)
≤50 yr	2(12)	15 (88)	0.331	2(11)	16 (89)	0.0371
51–60 yr	1(6)	15 (94)		1(17)	5 (83)
>60 yr	0(0)	6(100)		5 (50)	5 (50)
Sex - no. (% of row)   Female	3(13)	21 (88)	0.27	3(19)	13 (81)	0.69
Male	0(0)	15 (100)		5(28)	13 (72)
ECOG PS score - no. (% of row)           0	3(9)	29 (91)	>0.99	8(30)	19 (70)	0.16
1	0(0)	15 (100)		0(0)	7(100)
Diagnosis - no (% of row)
Upper Gastrointestinal	0(0)	2(100)	>0.992	0(0)	2(100)	0.612
Esophageal Adenocarcinoma	0	0		0	1
GEJ Adenocarcinoma	0	0		0	1
Gastric Adenocarcinoma	0	2		0	0
Hepatopancreaticobiliary	1(13)	7(88)		2(50)	2(50)
Cholangiocarcinoma	1	4		1	0
Pancreatic Ductal Adenocarcinoma	0	3		1	2
Lower Gastrointestinal	2(7)	27 (93)		6 (21)	22 (79)
Colon Adenocarcinoma	2	20		3	19
Rectal Adenocarcinoma	0	7		3	3
Early Onset Colorectal3	1(6)	16 (94)	>0.99	2(12)	15 (88)	0.17
Non-early onset Colorectal	1(8)	11(92)		4(36)	7(64)
Primary Tumor in Situ-no. (%of row)	0(0)	6(100)	>0.99	3(27)	8(73)	>0.99
Primary Resected	3(9)	30(91)		5(22)	18 (78)
Metastatic Disease
Synchronous - no. (% of row)	2(13)	13 (87)	0.54	2(9)	20 (91)	0.013
Metachronous - no. (% of row)	1(4)	23 (21)		6(50)	6(50)
Median DFI (range) - mo	4.2(n/a)	20.6 (4.9–100)	-	25.5 (12.8–36.2)	17.6(12.6–21.6)	0.24
Prior Therapy
Median (range) - no.	4(2–5)	4(1–9)	0.55	3 (1–6)	5 (2–11)	0.027
Distribution - no. (%of row)          1–3	1(8)	12 (92)	0.641	5(50)	5(50)	0.0191
4–6	2(11)	17 (89)		3(15)	17 (85)
>6	0(0)	7(100)		0(0)	4(100)
Prior Checkpoint Inhibitor	0	1(100)	>0.99	2(50)	2(50)	0.23
Time from resection to treatment median (range) - mo	3.2(3–19.4)	3.6 (1.6–28.7)	0.81	3.8 (3.2–6.3)	5.0(3.0–23.9)	0.051

^*univariate analysis and unless otherwise noted: Fisher’s exact for categorical variables, Mann-Whitney for discrete values

¹chi-square test for trend,

²Fisher’s exact Lower GI vs Other,

³<50 years old at time of diagnosis; DFI: Disease-free interval

Extended Data Table 2.

Laboratory correlates of response to TIL selected for neoantigen reactivity (exploratory)

	Selected TIL(n=39)			Selected TIL + Pembrolizumab (n=34)
Characteristic	Partial Response (n=3,7.7%)	Non-Response (n=36)	p-value*	Partial Response (n=8,23.5%)	Non-Response (n=26)	p-value*
Source of TIL - no. (%of row)
Lung metastasis	3(10)	27 (90)	>0.991	6(21)	22 (79)	0.611
Liver metastasis	0(0)	3(100)		1(50)	1(50)
Lymph node metastasis	0(0)	3(100)		0(0)	1(100)
Omental or peritoneal metastasis	0(0)	2(100)		1(50)	1(50)
Soft tissue metastasis	0(0)	1(100)		0(0)	1(100)
Tumor-based sequencing†
TMB-Median (range)	4.6(1.44.8)	4.9 (1.3–16.1)	0.92	4.4 (1.7–5.8)	4.4 (2.1–6.5)	0.59
1–3	0	6	0.842	3	2	0.442
>3–4	0	4		0	8
>4–5	2	10		2	7
>5–6	0	5		3	6
>6	0	9		0	3
Tumor mutation clonality	0.93 (0.88–0.98)	0.99 (0.67–1)	0.33	0.98 (0.82–1)	0.95 (0.71–1)	0.098
Infusion Characteristics
Total Cells, Median (range)	12.7×1010	8.2×1010	0.08	9.8×1010	8.2×1010	0.37
	(7.3–14.8)	(3.3–14.7)		(2.5–10.4)	(3.3–14.7)
Distribution - no. (% of row)
<5×1010	0(0)	9(100)	0.172	2(29)	5(71)	0.882
5–10×1010	1(6)	16 (94)		4(19)	17 (81)
≥ 1×1011	2(15)	11(85)		2(33)	4(67)
CD8+ cells, median (range)	2.9×1010	3.7×1010	0.94	2.4×1010	3.6×1010	0.56
	(0.5–14.1)	(0.1–15.2)		(0.9–5.8)	(0.1–14.0)
CD4+ cells, median (range)	4.4×1010	2.9×1010	0.57	5.1×11010	3.5×1010	0.35
	(0.7–12.4)	(0–12.4)		(0.3–9.6)	(0.14–8.1)
IL-2 doses (720,000 IU/kg)
Median (range) - no.	5 (4–6)	4.5 (0–8)	0.71	4(1–6)	3.5 (0–8)	0.42
Distribution - no. (% of row)
0	0(0)	1(100)	0.552	0(0)	3(100)	0.582
1–3	0(0)	12 (100)		3 (23)	10 (77)
4–6	3(14)	18 (86)		5(31)	11 (69)
>6	0(0)	5(100)		0(0)	2(100)

TMB: tumor mutation burden (mutations/Mb)

^*unless otherwise noted: Fisher’s exact for categorical variables, Mann-Whitney for discrete values

¹Fisher’s exact Lung vs Other,

²chi-square test for trend

^†n=70, excludes external commercial sequencing results

Extended Data Table 3.

Patient and Treatment Characteristics of SEL-TIL+P treatment group

ID	Age	Sex	Diagnosis	# Prior Treatments*	Source of TIL (Resected Metastasis)	Sites of Disease at Treatment	# Immunologic Targets	# Cells ×109	# IL-2 Doses	OR	Duration (months)
1	51	M	Colon	7	Lung	Liver, Lung, Lymph nodes, Adrenal, Peritoneum	2	69.1	4	NR	-
2	34	F	Colon	5	Lymph Node	Liver, Lymph nodes, Adrenal	2	75.9	4	NR	-
3	36	F	Colon	3	Lung	Liver, Lung, Lymph nodes, Colon°	2	91.8	0	NR	-
4	64	M	Colon	5	Lung	Lung	2	33.2	6	PR	4
5	49	M	Colon	11	Lung	Lung	4	91.2	7	NR	-
6	43	F	Colon	4	Lung	Lung	5	98.6	4	PR	42
7	37	F	Colon	4	Lung	Liver, Lung, Lymph nodes. Mesentery	1	84.9	3	NR	-
8	50	M	Colon	5	Lung	Liver, Lung, Lymph nodes, Mediastinum	2	32.6	4	NR	-
9	42	F	Colon	4	Lung	Liver, Lung	3	36.1	3	NR	-
10	35	M	Colon	7	Lung	Liver, Lung, Lymph nodes, Mediastinum	2	46.81	2	NR	-
11	60	F	Colon	2	Lung	Liver, Lung	2	89.7	4	NR	-
12	40	M	Colon	5	Lung	Lung, Bone	4	101	4	NR	-
13	25	M	Esophageal	3	Lung	Liver, Lung, Esophagus°	1	93.3	3	NR	-
14	52	M	Colon	6	Lung	Liver, Lung, Adrenal, Mediastinum	2	63.56	1	NR	-
15	62	F	Colon	3	Lung	Lung	4	66.9	5	NR	-
16	54	F	Colon	3	Lung	Liver, Lung, Lymph nodes, Colon°	2	32.9	4	NR	-
17	64	F	Colon	5	Lung	Liver, Lung	2	65.5	4	NR	-
18	50	F	Colon	5	Lung	Liver, Lung, Bone, Spine	1	79.2	0	NR	-
19	64	M	Colon	5	Lung	Liver, Lung, Lymph nodes, Colon°	3	74.1	5	NR	-
20	50	F	Colon	11	Lung	Liver, Lung, Lymph nodes	3	147	2	NR	-
21	33	F	GE Junction	5	Liver	Liver, Lymph Nodes, GE Junction°	2	87.9	8	NR	-
22	71	M	Pancreas	4	Peritoneum	Liver, Lymph nodes, Abdominal wall, Pancreas°, Peritoneum	3	90.2	2	NR	-
23	64	F	Cholangiocarcinoma	1	Lung	Liver, Lung, Lymph nodes, Adrenal, Bone	3	101.4	6	PR	17+
24	46	M	Rectal	6	Lung	Liver, Lung, Lymph nodes, Adrenal, Rectum°	3	92.7	0	NR	-
25	54	M	Rectal	3	Lung	Lung, Adrenal, Rectum°	5	81.1	3	PR	8
26	35	M	Rectal	6	Subcutaneous	Liver, Lung	1	41.4	1	NR	-
27	40	M	Colon	3	Liver	Liver, Lung, Colon recurrence	3	104	4	PR	10
28	40	M	Rectal	4	Lung	Liver, Lung, Rectum°	4	140.8	5	NR	-
29	64	M	Rectal	6	Lung	Liver, Lung, Lymph nodes	4	25.3	3	PR	11
30	64	F	Pancreas	2	Omentum	Liver, Abdominal wall, Pancreas°, Peritoneum	3	97.9	4	PR	7
31	68	M	Rectal	3	Lung	Lung, Lymph nodes, Rectum°	3	98.7	1	PR	6
32	52	F	Colon	5	Lung	Liver, Lung, Lymph nodes	3	53.3	5	NR	-
33	64	F	Pancreas	5	Lung	Liver, Lung	1	84.3	1	NR	-
34	34	M	Colon	5	Lung	Liver, Lung	1	140	3	NR	-

^*includes systemic and locoregional (i.e. surgery, radiation, ablation);

⁺indicates ongoing response; OR: objective response (PR: partial, NR: none);

^°Primary tumor