Claudin18.2-specific CAR T cells in gastrointestinal cancers: phase 1 trial interim results

Despite success in hematologic malignancies, the treatment landscape of chimeric antigen receptor (CAR) T cell therapy for solid tumors remains limited. Claudin18.2 (CLDN18.2)-redirected CAR T cells showed promising efficacy against gastric cancer (GC) in a preclinical study. Here we report the interim analysis results of an ongoing, open-label, single-arm, phase 1 clinical trial of CLDN18.2-targeted CAR T cells (CT041) in patients with previously treated, CLDN18.2-positive digestive system cancers (NCT03874897). The primary objective was safety after CT041 infusion; secondary objectives included CT041 efficacy, pharmacokinetics and immunogenicity. We treated 37 patients with one of three CT041 doses: 2.5 × 108, 3.75 × 108 or 5.0 × 108 cells. All patients experienced a grade 3 or higher hematologic toxicity. Grade 1 or 2 cytokine release syndrome (CRS) occurred in 94.6% of patients. No grade 3 or higher CRS or neurotoxicities, treatment-related deaths or dose-limiting toxicities were reported. The overall response rate (ORR) and disease control rate (DCR) reached 48.6% and 73.0%, respectively. The 6-month duration of response rate was 44.8%. In patients with GC, the ORR and DCR reached 57.1% and 75.0%, respectively, and the 6-month overall survival rate was 81.2%. These initial results suggest that CT041 has promising efficacy with an acceptable safety profile in patients with heavily pretreated, CLDN18.2-positive digestive system cancers, particularly in those with GC.

effects on GC, leading to significant tumor regression without weight loss in immunodeficient mice 12 . GC has a poor prognosis with limited treatment options, including first-line monotherapies or combinations of oxaliplatin, fluoropyrimidine, irinotecan, paclitaxel, anti-PD-1 antibody, docetaxel and carboplatin with or without trastuzumab depending on HER2 expression status, as well as ramucirumab (VEGFR-2 monoclonal antibody) plus paclitaxel as second-line treatment 13 . Similarly, limited treatments exist for metastatic pancreatic cancer (PC), including gemcitabine with or without nab-paclitaxel, FOLFIRINOX and nanoliposomal irinotecan with fluorouracil 14 . A previous single-center clinical trial (NCT03159819) indicated that CT041 was well-tolerated and could be a promising therapy for patients with gastric and pancreatic cancers 15 . Therefore, we initiated an open-label, multicenter, single-arm, dose-escalation/ de-escalation and dose-expansion phase 1 trial to investigate the safety, efficacy, pharmacokinetics and immunogenicity of CT041 in patients with cancers of the digestive system (Extended Data Fig. 1). Here we report the outcomes from a non-prespecified interim analysis of this ongoing trial (NCT03874897). Patients were excluded from the study if they had a target lesion larger than 4 cm at the time of apheresis; active infection including, but not limited to, hepatitis B virus (HBV), hepatitis C virus (HCV) and tuberculosis; brain metastases; unstable or active gastric ulcer or gastroenterological bleeding; central or widespread metastatic lung lesion(s); or massive liver metastases. A complete list of eligibility criteria can be found in the Methods. The trial consists of a modified '3 + 3' pattern doseescalation/de-escalation phase and a dose-expansion phase. In the dose-escalation/de-escalation phase, patients with CLDN18.2-positive tumors received one of three CT041 dose levels: 2.5 × 10 8 , 3.75 × 10 8 or 5.0 × 10 8 cells. Based on discussions with the Data Safety Monitoring Committee (DSMC), the 2.5 ×10 8 CAR T cell dose was recommended for the dose-expansion phase (Extended Data Fig. 1).
From 26 March 2019 through 8 April 2021, 74 participants signed informed consent; 59 patients were apheresed; and 49 patients were infused. The first 37 patients were included in this interim analysis, including 28 with GC/GEJ cancer, five with PC and four with other digestive system tumor types (Table 1 and Extended Data Table 1). These patients received CT041 infusion and had completed at least 12 weeks of safety, efficacy and pharmacokinetics assessments after the first infusion (Table 1, Extended Data Table 1 and Extended Data Fig. 2). The median age was 53.0 (25-74) years, and 64.9% (24/37) of patients were men. All enrolled patients were diagnosed with metastatic disease, and 50% of them had at least three organ sites involved. Among the 28 patients with GC/GEJ cancer, 25 (89.3%) had received at least two lines of therapy or at least a triple combination of fluoropyrimidine, oxaliplatin and paclitaxel, and 12 (42.9%) had received prior anti-PD-1/PD-L1 antibody therapy. No HER2-positive patients were enrolled in this study. The cell doses of 2.5 × 10 8 , 3.75 × 10 8 and 5.0 × 10 8 were administered to 28, six and three patients, respectively, with a median follow-up of 8.5 months (range, 2.9-19.4 months) (Supplementary Table 1) since apheresis. As of the data cutoff date (8 April 2021) for this interim analysis, patient enrollment continues for the study, and the results will be reported when available.
A total of 28 (75.7%) patients received at least one cycle of the bridging therapy regimen (Table 1), including FOLFIRI (39.3% (11/28)), nab-paclitaxel (32.1% (9/28)) or irinotecan (25.0% (7/28)) during the CT041 manufacturing period at the investigators' discretion based on patients' conditions (Extended Data Table 1). The median duration from apheresis to infusion was 27 days (range, 22-187 days) (Supplementary Table 2). Two patients received more than one cycle of bridging therapy because constraints related to the Coronavirus Disease 2019 pandemic delayed their prescheduled on-site infusion visits. One patient received two cycles of bridging therapy due to prolonged CT041 manufacturing. All patients underwent radiographic imaging at least 14 days after bridging therapy and before preconditioning treatment. Only three patients showed a limited decrease in tumor size (5.5−9.7%) compared to the screening visit imaging.
Thirty-five patients received fludarabine/cyclophosphamide plus nab-paclitaxel (FNC) preconditioning. Two patients with PC were preconditioned with gemcitabine instead of nab-paclitaxel (Extended Data Table 1) owing to either disease progression or persistent peripheral neurotoxicity after nab-paclitaxel as the last systemic treatment. Additional cycles of preconditioning and CT041 were administered to 18 of 37 patients (48.6%), including three patients with GC who received three cycles of preconditioning/CT041 and 15 patients (nine with GC, four with PC and two with other cancers) who received two cycles (Extended Data Table  1). The median interval from the first infusion to the second infusion, and the median interval from the second infusion to the third infusion, were 72 days  and 101 days (81-140), respectively (Supplementary Table 2).
Safety and tolerability of CT041. The primary objective was to evaluate CT041 safety and tolerability through 28 days after the first infusion. No predefined dose-limiting toxicities (DLTs) within 28 days after the first infusion were observed in the dose-escalation/ de-escalation phase. However, one patient (Pt02) at the 5.0 × 10 8 dose level suffered a grade 4 GI hemorrhage due to rapid tumor regression after the second infusion ( Table 2). As a result, the dose was decreased to 2.5 × 10 8 and 3.75 × 10 8 CAR T cells after discussion with the DSMC (Extended Data Fig. 1).
As shown in Table 2 and Supplementary Table 3, the most frequently reported adverse events (AEs) of grade 3 or higher were preconditioning-related hematologic toxicities in 37 patients (100%), leukopenia in 31 of 37 (83.8%), neutropenia in 25 of 37 (67.6%), anemia in 15 of 37 (40.5%) and thrombocytopenia in six of 37 (16.2%), which occurred within 28 days after the first infusion and generally recovered within a median of 4-9 days (Supplementary Table 4).
No grade 3 or higher CRS was observed. Per the American Society for Transplantation and Cellular Therapy (ASTCT) criteria 16  WHO, World Health Organization. a CLDn18.2 expression level by immunohistochemical staining intensity was graded as 1+, 2+, 3+ or 0 and multiplied by the percentage of tumor cells that were positive. Low expression was defined as any intensity with a percentage of <40% or intensity 1+ with any percentage; medium expression was defined as intensity 2+ or 3+ with a percentage of 40% (inclusive) to 69%; and high expression was defined as intensity 2+ or 3+ with a percentage of ≥70%. b Polykinase inhibitor: multi-target tyrosine kinase inhibitor, including apatinib, anlotinib, etc. Twenty-seven patients were treated with tocilizumab and four with glucocorticoids for CRS in the first cycle (Supplementary Table  5). Per ASTCT criteria 16 , no immune effector cell-associated neurotoxicity syndrome (ICANS) was observed. No treatment-related death was reported. Four patients (4/37, 10.8%) reported reversible  or others   15  14  13  12  11  10  9  8 Months after CAR T infusion grade 3/4 GI AEs, including the grade 4 GI hemorrhage mentioned above and three grade 3 GI AEs of abdominal pain, obstructive pancreatitis and gastric mucosal injury ( Table 2). Grade 1/2 gastric mucosal injuries were also identified in five patients who recovered after treatment with prokinetic agents and proton pump inhibitors, with a median recovery time of 19 days (10-80) (Supplementary Fig. 3 and Supplementary Table 6).
One patient developed anaphylactic shock immediately after the second infusion and recovered in 30 minutes after treatment with adrenaline and glucocorticoids. This patient had a history of multiple allergies, including allergies to previous anti-PD-1 antibody and platinum treatment. Notably, this patient's anti-CLDN18.2 CAR antibody (anti-drug antibody (ADA)) levels remained negative at all time points, and no specific cause of the anaphylactic shock could be determined.
For patients with GC, the ORR and DCR reached 57.1% (95% CI, 37.2, 75.5) and 75% (95% CI, 55.1, 89.3), respectively (Table 3 and  Supplementary Table 7). The mPFS was 4.2 months (95% CI, 3.7, 9.2), and the OS rate at 6 months was 81.2% (95% CI, 60.3, 91.8). Among the 16 patients with GC who responded to CT041, 14 patients achieved partial response (PR) at week 4 after the first CT041 infusion; one patient responded at week 8; and one patient responded after the second infusion (Fig. 1b). Responders were observed to have higher peak CAR copies as well as C-reactive protein after the first infusion than those of non-responders ( Fig. 2 and Extended Data Fig.  3f). As shown in Supplementary Table 7, 18 patients with GC who had failed at least two prior lines of therapy and were treated with 2.5 × 10 8 CAR T cells achieved an ORR of 61.1%, a DCR of 83.3%, an mPFS of 5.6 months (95% CI, 2.6, 9.2) and a DOR rate at 6 months of 57.1%.
Notably, Pt08 (GC, 57 years old), who had received three prior lines of therapy, including anti-PD-1 antibody, responded to CT041 treatment as seen by the shrinkage of an umbilicus tumor lesion. The lesion achieved remission after infusion, evaluated via visual inspection and computed tomography (CT) scans from week 12 to week 40, and the response was still ongoing as of the cutoff date (Fig. 1c).
Subgroup analyses in patients with GC indicated that CT041 had an ORR of at least 40% in most subgroups (Extended Data Fig. 5), especially in patients who had failed anti-PD-1/PD-L1 antibody treatment (12/28) and/or taxane treatment (21/28). The ORR reached as high as 70% in patients with Lauren intestinal type and 63% in patients with high CLDN18.2 expression level defined as intensity 2+ or 3+ with a percentage of ≥70%. Interestingly, the ORR was numerically higher in patients who did not receive bridging therapy, which was also observed in a CD19-targeted CAR T cell trial for large B cell lymphoma 17 . The basis for this observation could be that patients who required bridging therapy could have more rapidly progressive disease (PD) than those with less aggressive disease during the CAR T cell manufacturing period.
For the second infusion, the C max significantly decreased to a median value of 1,151 copies per microgram of gDNA (334-9,877), aligning with a lower toxicity profile after the second dose, such as decreased CRS incidence (72.2%, 13/18) ( Fig. 2b and Supplementary Table 5).
As shown in Fig. 2c, the C max of CAR copies in the blood was higher in the responders than in the non-responders (10,553 copies per microgram of gDNA versus 4,980 copies per microgram of gDNA, P = 0.059). The PFS positively correlated with increased AUC last , which represented CAR T cell expansion and persistence (Fig. 2d).
We explored the distribution of CAR T cells in other body fluids in four patients with GC, including bile (2/4), ascites (1/4) and pleural effusions (2/4). Compared to those in serum, the C max value increased significantly, and the T max value was similar (Extended Data Fig. 6).
In our study, repeated biopsies were conducted in eight patients after the first infusion, and no substantial changes in antigen expression level were observed after CT041 treatment ( Supplementary Fig. 4).
Correlation between ADAs and response. Twenty-eight of 37 patients had at least one sample with detectable ADAs after infusion, and the remaining nine patients were ADA-negative in all their visits until the data cutoff date. The ORRs were 53.6% and 33.3% in ADA-positive patients and ADA-negative patients, respectively, without statistical differences identified (P = 0.45) (Extended Data Table 2 and Supplementary Table 8). The sensitivities of the screening and the confirmatory ADA assay were 27.5 ng ml −1 and 52.6 ng ml −1 , respectively, which were more sensitive than the recommendation of the US Food and Drug Administration (100 ng ml −1 ) 18 .

CT041 T cell subset frequencies and clinical activity.
The correlation between T cell subset frequencies and clinical activity was an exploratory endpoint. We evaluated the T cell subsets, including CD3 + CD4 + CD8 − (Th) cells, CD3 + CD4 − CD8 + (cytotoxic T lymphocytes (CTLs)), CD45RA + /CCR7 + (naive T) cells, CD45RA − /CCR7 + (central memory T) cells, CD45RA − /CCR7 − (effector memory T) cells and CD45RA + /CCR7 − (terminally differentiated effector T) cells for all the patients before infusion and analyzed the correlation between T cell subset frequencies and CT041 anti-tumor activity in 28 patients with GC. Patients infused with CT041 that contained a lower frequency of CD45RA + /CCR7 − T cells were more likely to achieve longer PFS (P = 0.0058) (Fig. 3a). No statistically significant correlation was observed between any of the T cell subsets and OS (Fig. 3b). Frequencies of T cell subsets in CT041 products showed no statistically significant association with clinical response (Fig. 3c-h).
We also analyzed the correlation between frequencies of T cell subsets and C max or AUC last of CAR copies after the first infusion. The results revealed that CT041 products containing lower frequencies of CD45RA + /CCR7 − cells achieved higher C max and AUC last (P = 0.009 and P = 0.033; Extended Data Fig. 7a). The frequency of CD45RA − /CCR7 + cells was positively correlated with C max (P = 0.037; Extended Data Fig. 7b).

Discussion
The results of the present study revealed that CT041 was well-tolerated in patients with CLDN18.2-positive cancers of the digestive system. No DLTs (within 28 days after the first infusion), treatment-related deaths or AEs leading to study withdrawal were reported. All patients experienced expected transient hematologic toxicities mainly related to the preconditioning regimens. No difference was found in AE incidence between different dose levels. In terms of on-target off-tumor toxicity, only one grade 3 mucosal erosion was reported in one patient with PC. Months after CAR T infusion CAR copies (copies/µg gDNA)  Weeks after CAR T infusion CAR copies (copies/g gDNA) 35 40 45 64,000     Although CLDN18.2 is highly expressed in short-lived differentiated gastric mucosal cells 8 , manageable mucosal injury AEs were reported in only a minority (6/37) of the patients, of which five were grade 1 or grade 2. Our preclinical study indicated that NOD/SCID mice treated with CLDN18.2 CAR T cells displayed no obvious toxicities in normal stomach tissues 12 , which might be ascribed to the limited exposure of CAR-binding epitope in normal tissues, the normal tissue microenvironment or the rapid tissue repair capacity of CLDN18.2-negative stomach normal stem cells.
In prior clinical reports, nivolumab (ATTRACTION 2 study), pembrolizumab (KEYNOTE-059 study, cohort 1), trifluridine/tipiracil (TAGS study) and apatinib (a small-molecule VEGFR inhibitor developed mainly in China), which had been administered in the third-line setting for GC, achieved ORRs of 11.2%, 13.3%, 4% and 1.7%, respectively [19][20][21][22] . The mPFSs for these treatments were limited at 1.6-2.6 months, and OSs were no more than 6 months [19][20][21][22] . Compared to the aforementioned therapies, CT041 achieved promising results in the third-line setting for GC. Moreover, the patients with GC in this study had tumors pathologically characterized as 42.9% signet-ring cell carcinoma and 57.1% Lauren diffused/mixed type, which are regarded as poor prognostic subtypes of advanced GC 23,24 . In contrast, the KEYNOTE-059 study treated only 2.7% patients with signet-ring cell carcinoma. In addition, patients with GC and peritoneal metastases have an overall survival ranging from 2 months to 9 months 25 , although advances in combination intraperitoneal and systemic chemotherapy administration may offer future improvement 26 . This may, in part, be due to limited penetration of systemic administration of anticancer drugs to peritoneal carcinomas 27 . Conversely, despite the intravenous CT041 infusion, we surprisingly observed much higher CAR T cell expansion in ascites than in peripheral blood of patients with GC with multiple peritoneal metastases. In our study, 67.9% of patients with GC were diagnosed with peritoneal metastasis on baseline CT imaging, in contrast to the markedly lower percentages of 1.5%, 26% and 24.4% in the third-line setting studies of KEYNOTE-059, TAGS and apatinib, respectively [20][21][22] . In short, it is encouraging for CT041 to achieve such promising efficacy in the subgroup of patients with refractory advanced GC compared to currently available treatment options. In addition, CAR T cell expansion and/or accumulation in ascites might be advantageous for eliminating diffuse peritoneal metastatic lesions, which warrants further investigation.
Previous studies indicated that the response rate of anti-PD-1/ PD-L1 antibody might be affected by the tumor immune microenvironment [28][29][30] . Interestingly, we observed nearly 70% ORR in patients without PD-L1 expression, and six of 12 patients who had failed anti-PD-1/PD-L1 antibody treatment responded to CT041 (Extended Data Fig. 5). These data suggest that the efficacy of CAR T cell therapy might not be limited by PD-L1 expression status or impaired by previous use of anti-PD-1/PD-L1 blockade. Therefore, CT041 might potentially bring clinical benefits to patients who have no effective treatments.
In this study, 75% of the patients with GC received bridging therapy (Table 1), and only three patients had a modest (<10%) tumor reduction before preconditioning. The results suggest that the objective response after CT041 treatment should mainly be ascribed to CT041, although the potential contribution from bridging therapy and preconditioning regimen could not be fully excluded. Preconditioning chemotherapy with fludarabine and cyclophosphamide is commonly used for CD19 CAR T cell therapy; however, this regimen has low efficacy for cancers of the digestive system 13 . Thus, this regimen may not effectively improve the tumor microenvironment and help the infiltration and function of CAR T cells in solid tumors. Preclinical and clinical studies have shown that nab-paclitaxel can be transported across the endothelial cells, accumulate in the tumor stroma and disrupt cancer-stromal interactions [31][32][33] . Administration of low-dose nab-paclitaxel significantly decreased the accumulation and immunosuppressive activities of tumor-associated myeloid-derived suppressor cells 34 . Therefore, a non-cytotoxic dose of nab-paclitaxel may improve the infiltration and function of CAR T cells. In this study, we used a single 100-mg fixed dose of nab-paclitaxel, which was much lower than the standard pharmacological dose 35,36 . Despite the low dose of nab-paclitaxel, the subgroup analyses showed that 21 of 28 patients with GC who had failed taxane treatment, including six who had received nab-paclitaxel, achieved a high response rate. Because we observed preliminary good safety and promising efficacy results for CT041 with the FNC regimen, the FNC regimen was recommended in the protocol as the preconditioning regimen. Although it will be important to explore if/how the FNC regimen facilitates the high response of CT041, the requisite repeated biopsies before and after preconditioning would be challenging to perform owing to ethical considerations.
The expansion and persistence of CT041 was generally consistent with the response rate and PFS. Because no considerable difference in cell expansion was observed among the three dose levels, and a similar response rate was observed for the dose of 2.5 × 10 8 cells compared to other dose levels, this dose was recommended for further development in the expansion cohort after consulting with the DSMC. The median persistence of CT041 in peripheral blood after the first infusion was 28 days, shorter than that of CAR T cells in hematologic malignancies. CAR T cell persistence in peripheral blood is potentially associated with duration of tumor response in hematologic malignancies. However, the expansion and persistence of CAR T cells within solid tumor tissues are difficult to evaluate. We did observe durable responses in a few patients after CT041 was undetectable by qPCR in the peripheral blood. Future studies are needed for a better understanding of the mechanisms underlying response durability.
According to previous reports, CD19 epitope loss is a relatively common tumor escape mechanism after treatment with CD19-redirected CAR T cell therapy 37,38 . In contrast, BCMA antigen loss is an uncommon mechanism for relapsed or refractory multiple myeloma 4 . In this study, the CLDN18.2 expression levels were not downregulated after CT041 infusion ( Supplementary Fig.  4), suggesting that antigen escape may not be the common mechanism of acquired resistance against CT041, which deserves further investigation.
Interestingly, our data showed that a lower frequency of CD45RA + /CCR7 − terminally differentiated effector T cells in CT041 was more likely to achieve longer PFS and higher C max and AUC last . Additionally, we observed a positive correlation between the frequency of CD45RA − /CCR7 + central memory T cells and C max . These data suggest that the T cell phenotype composition of CT041 may be critical for its efficacy. Furthermore, previously reported data indicated that naive T cells and central memory T cells could drive anti-tumor activity 39 . More data are needed to investigate the association between CAR T cell subsets and treatment efficacy.
Taken together, these interim results indicate that CT041 has the potential to become an important treatment modality for patients with advanced GC, suggesting that CAR T cell therapy could benefit not only patients with hematologic malignancies but also those with solid tumors. CT041 showed a trend of more benefit in patients with higher intratumoral CLDN18.2 expression. These initial observations remain to be verified in the complete trial. In addition to GC, further investigation is needed to fully understand the characteristics of patients with sustained responses in other GI tumor types.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of tissue. Automated immunostaining was performed using the Leica BOND-III (Leica Biosystems) in the bio-sample analysis laboratory of CARsgen Therapeutics with the validated staining method. Slides were stained with antibodies against CLDN18.2 (clone 14F8, prediluted mouse monoclonal antibody, CARsgen made in-house) with Bond Polymer Refine Detection (Leica Biosystems) and counterstained with hematoxylin according to the manufacturer's instructions 40,41 . CLDN18.2 protein expression was determined using both the staining intensity and the percentage of stained tumor cells (the number of CLDN18.2 staining cells / the total number of viable tumor cells × 100%). Low expression was defined as any intensity with percentage <40% or intensity 1+ with any percentage; medium expression was defined as intensity 2+ and 3+ with percentage from 40% (inclusive) to 69%; and high expression was defined as intensity 2+ and 3+ with percentage ≥70%.
Flow cytometry of CAR T cell products. Standard staining and flow cytometry techniques were used to perform immunophenotyping of surface markers on all CT041 cell products at the CARsgen Therapeutics manufacturing site ( Supplementary Fig. 5). Th and CTL cells were stained with CD3-PerCP, CD4-FITC-and CD8-PE. The frequencies of naive T cells, central memory T cells, effector memory T cells and terminally differentiated effector T cells were defined by CD197-BV421 and CD45RA-FITC 39 . All antibodies for analysis were purchased from BD Biosciences. Samples  Table 9) 42 . Patient samples of 10 ml of peripheral blood were collected in K2EDTA tubes at baseline and after infusion. gDNA was extracted using the QIAamp DNA Midi Kit (Qiagen). The standard curve for the transcript copy number was established by the amplification of a ten-fold serially diluted linearized plasmid PSD001 between 2 × 10 6 and 200 copies. The number of transgene copies per microgram of gDNA was determined on a 7500 Fast (Thermo Fisher Scientific) triplicated for each sample. The limit of detection of this assay was 40 copies per microgram of gDNA.
ADA assay. ADAs to CAR-CLDN18.2 cells were detected at the central laboratory of CARsgen Therapeutics using the human K2EDTA plasma sample with a validated electrochemiluminescence (ECL) assay on MESO QuickPlex SQ 120 (Meso Scale Discovery (MSD)). This assay detects anti-CT041 antibody by measuring the light emission by SULFO-TAG label. All the samples were duplicated with this three-tier assay, which includes screening, confirmatory and titer assays 43 .
A 1-ml aliquot of the patient plasma sample was transferred into a K2EDTA tube at baseline and days 28, 56, 84, 126, 168, 280 and 392 after infusion. Standard MSD 96-well plates were coated with the solution of 100 μl per well of 0.5 ug ml −1 of capture drug (hu8E5-2I scFv) in PBS at room temperature for 120 minutes. The samples and positive controls (goat anti-hu8E5-2I scFv polyclone antibody) were diluted by adding following components into a Nunc 96-well conical-bottom polypropylene plate (Thermo Fisher Scientific) and mixed: 50 μl per well of the samples and 200 μl per well of Super Block Blocking Buffer (Thermo Fisher Scientific). After blocking and washing the standard MSD 96-well plates, 100 μl of the samples and controls were transferred from the Nunc plate to the standard MSD 96-well plate and incubated at room temperature for 60 minutes. After plate washing, 100 μl per well of detection drug (SULFO-TAG-hu8E5-2I scFv) was added and incubated at room temperature for 60 minutes. After plate washing, 150 μl per well of 2× MSD Read Buffer was added, and the ECL signal responses were quantified in a Meso Scale QuickPlex 120 MSD reader. The sensitivity of the screening and confirmatory assays were 27.5 ng ml −1 and 52.6 ng ml −1 , respectively.
Outcomes. The primary endpoints were safety and tolerability during the 4 weeks after the first CT041 infusion. Secondary endpoints were pharmacokinetics of CT041, 12-month safety and tolerability and anti-tumor efficacy. Pharmacokinetics was presented with the data of CAR-CLDN18.2 DNA in peripheral blood, which was detected by qPCR at each visit since infusion. CRS and ICANS events were graded per the ASTCT criteria 18 . The number of participants with treatment-related AEs (including the AEs related to preconditioning or CT041) was assessed by CTCAE version 5.0. The anti-tumor efficacy was measured according to RECIST version 1.1. The long-term survival follow-up was a maximum of 15 years after the first infusion. Exploratory endpoints were the effect of disease characteristics, such as expression of CLDN18.2, metastases site and Lauren classification, on the efficacy of CT041 and expansion of CT041 cells after infusion.

DSMC.
A DSMC consisting of two clinical experts and one statistical expert was organized for this study. DSMC meetings were held regularly to perform unblinded data review to identify potential risks for patients. Besides the regular meetings, a DSMC meeting must be held before the study proceeds with administering the next dose level.
Statistical analysis. The sample size was based on clinical considerations and a modified '3 + 3' dose-escalation/de-escalation design. Descriptive statistics include the number of cases, mean, median, standard deviation, minimum and maximum values for continuous variables and frequency distributions for categorical variables. All AEs were categorized according to the ICH MedDRA codes version 23.1, graded according to CTCAE version 5.0 and analyzed via frequency distributions, tables or other descriptive indicators. The number and percentage of patients experiencing a treatment-emergent AE were calculated based on system organ classification, preferred term and different groups. Tables containing CAR-CLDN18.2 copy numbers at specified time points were generated. Exact methods contain the Clopper-Pearson method with 95% CIs for ORR and DCR and the Kaplan-Meier method for mPFS, DOR rate and OS rate. Logistic regression was used to explore the relationship between the covariate and tumor response. The predictive curve based on the univariate logistic regression calculation visualized the trend. The Cox regression model was used to analyze the potential correlation of covariate and time-to-event endpoints and to estimate the hazard ratios and 95% CIs.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
All data used in the interim analyses supporting the findings of the present study are available within the manuscript and its supplementary information files. All requests for further data sharing will be reviewed by the leading clinical center, by the Department of Gastrointestinal Oncology, Peking University Cancer Hospital and Institute, and by the study collaborator, CARsgen Therapeutics Co., Ltd., to verify whether the request is subject to any intellectual property or confidentiality obligations. Figures 1a-cand 2a-c, Extended Data Table 1 and Extended Data Figs. 3, 4, 6 and 7 present individual participant-level data with privacy information de-identified to support understanding of the study results. Further requests for access to the individual participant-level data from this study can be submitted via email to the corresponding author with detailed proposals. Each participant's rights and privacy are key subjects to take into consideration when sharing information. A signed data access agreement with the collaborator is required before accessing shared data.