TL1A inhibitors in inflammatory bowel disease: A systematic review of phase 2 clinical trials

Abstract

Background:

Tumor necrosis factor-like ligand 1A (TL1A) (TNFSF15) signals through DR3 to drive both mucosal inflammation and intestinal fibrosis. Three anti-TL1A antibodies (tulisokibart, afimkibart, and duvakitug) have completed Phase 2 trials in inflammatory bowel disease (IBD). We performed a systematic review (SR) of all Phase 2 data, with the meta-analysis focused on the three randomized controlled trials (RCTs) in ulcerative colitis (UC).

Methods:

PubMed, Embase, CENTRAL, and ClinicalTrials.gov were searched through January 2026. Risk of bias was assessed using RoB 2.0 and ROBINS-I. Risk ratios (RR), risk differences (RD), and number needed to treat (NNT) were pooled using random-effects models.

Results:

Six studies from five trials (three RCTs and two single-arm studies; 804 patients) were included. In UC, pooled RR versus placebo was 2.56 (95% CI 1.60–4.09; I² = 0.0%) for clinical remission and 2.13 (1.44–3.16; I² = 0.0%) for endoscopic improvement (both P < 0.001). Corresponding RDs were +22.2% (NNT 4.5) and +22.6% (NNT 4.4). Duvakitug showed a dose-response, with 900 mg approximately doubling the effect of 450 mg. In Crohn’s disease (CD), duvakitug 900 mg achieved endoscopic response RR 3.67 versus placebo (95% CI 1.64–8.20; RD +34.8%, NNT 2.9). Serious adverse event rates were numerically low and comparable to or lower than placebo across trials. GRADE certainty of evidence was moderate for UC efficacy outcomes.

Conclusions:

Three structurally distinct anti-TL1A antibodies showed consistent Phase 2 efficacy signals in UC with zero between-study heterogeneity (I² = 0.0%), though, given only three UC RCTs and one CD RCT, this apparent class-level signal is underpowered for heterogeneity and remains tentative. Phase 3 trials are needed to confirm these preliminary findings.

INTRODUCTION

Inflammatory bowel disease (IBD) remains a chronic, relapsing condition affecting over 6.8 million people worldwide, with rising global incidence.[1] Despite an expanding therapeutic armamentarium, substantial unmet needs persist. Only 30–40% of patients achieve clinical remission with any given agent, secondary loss of response affects up to 45% of initial responders, and no approved therapy directly addresses intestinal fibrosis, the principal driver of stricture formation and surgical intervention in Crohn’s disease (CD).[2]

TL1A (tumor necrosis factor-like ligand 1A, encoded by TNFSF15) has emerged as a promising target that may address several of these gaps. First identified in 2002,[3] TL1A is a TNF superfamily cytokine that signals through death domain receptor 3 (DR3/TNFRSF25). Genome-wide association studies have consistently linked TNFSF15 variants to IBD susceptibility,[4,5] and TL1A expression is markedly elevated in inflamed intestinal mucosa in murine colitis models.[6] The TL1A-DR3 axis is pleiotropic in mucosal immunity: it co-stimulates Th1 and Th17 responses[3]; activates group 2 innate lymphoid cells (ILC2) to produce IL-13[7]; and directly activates intestinal fibroblasts and myofibroblasts to promote collagen deposition.[8,9] A soluble decoy receptor, DcR3, provides endogenous counter-regulation by competitively binding TL1A.[3,8]

Preclinical studies have demonstrated anti-fibrotic potential: in murine colitis models, neutralising TL1A antibody reduced colonic fibrosis.[9] Fibroblast-specific deletion of DR3 reduced collagen deposition despite persistent inflammation, confirming a direct, inflammation-independent pro-fibrotic pathway.[9] These findings prompted the development of three anti-TL1A monoclonal antibodies, namely tulisokibart (MK-7240, formerly PRA023; Merck/Prometheus), afimkibart (RG6631, formerly PF-06480605/RVT-3101; Roche/Genentech), and duvakitug (TEV-48574/SAR447189; Teva/Sanofi), all of which have completed Phase 2 evaluation and entered Phase 3 programs.[10,11,12,13,14,15]

This systematic review (SR) synthesizes efficacy, safety, biomarker, and translational findings from all Phase 2 clinical trials of TL1A inhibitors in UC and CD, providing an integrated class-level summary of the current evidence base. The meta-analysis was focused on the three randomized controlled trials (RCTs) in UC, the only studies permitting controlled quantitative pooling, whereas the single CD RCT and the two open-label single-arm studies were summarized narratively. The objective of this review is to summarize the emerging Phase 2 evidence for TL1A inhibition in IBD and to characterize whether a consistent class-level efficacy signal is discernible across structurally distinct agents, rather than to establish definitive efficacy that awaits the ongoing Phase 3 programs.

MATERIALS AND METHODS

Search strategy and information sources

This SR was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 checklist [Supplementary File 1].[16] The review protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO; CRD420261303403). A comprehensive literature search was performed across PubMed/MEDLINE, Embase-OVID, the Cochrane Central Register of Controlled Trials (CENTRAL), and ClinicalTrials.gov from database inception through January 31, 2026. The search strategy combined terms for the intervention (“TL1A,” “TNFSF15,” “anti-TL1A,” “tulisokibart,” “PRA023,” “MK-7240,” “afimkibart,” “PF-06480605,” “RVT-3101,” “RG6631,” “duvakitug,” “TEV-48574,” “SAR447189”) with terms for the condition (“inflammatory bowel disease,” “ulcerative colitis,” “Crohn’s disease”) and study design (“clinical trial,” “Phase 2,” “randomised”) [Supplementary File 1]. Reference lists of included studies and relevant reviews were hand-searched. Conference abstracts from digestive disease week (DDW), European Crohn’s and Colitis Organisation (ECCO) Congress, and the American College of Gastroenterology (ACG) annual meeting (2020–2025) were searched for unpublished trial data. Manufacturer press releases from Merck, Roche/Genentech, Teva, and Sanofi were reviewed for supplementary efficacy and safety information. The search was subsequently updated on July 21, 2026 during revision; this update identified full peer-reviewed publications of the RELIEVE UCCD ulcerative colitis and CD cohorts that had previously been available only as conference abstracts and manufacturer press releases and that were incorporated by updating the corresponding references, and identified no additional Phase 2 induction randomized trials of TL1A inhibitors.

Eligibility criteria

Studies were included if they: (1) enrolled adults (≥18 years) with moderately to severely active UC or CD; (2) evaluated a monoclonal antibody targeting TL1A; (3) were Phase 2 (including Phase 2a or 2b) clinical trials; and (4) reported at least one efficacy or safety outcome. Both RCTs and single-arm studies were eligible. Case reports, preclinical studies, Phase 1 pharmacokinetic studies, narrative reviews, and editorials were excluded. Studies reporting exclusively long-term extension data without primary induction-period results were excluded.

Data extraction and outcomes

Study selection and data extraction were performed independently by two reviewers (E.J.G. and C.S.B.), and any discrepancies were resolved by discussion or, when necessary, by referral to a third reviewer (J.J.L.). Data were extracted on study design, patient population, dosing regimen, primary and key secondary endpoints, efficacy outcomes, safety events, and biomarker findings. The primary outcomes of interest were clinical remission (defined by modified Mayo Score [mMS] for UC and CDAI <150 for CD), endoscopic improvement (Mayo Endoscopic Subscore [MES] 0 or 1 for UC) or endoscopic response (≥50% reduction in Simple Endoscopic Score for Crohn’s Disease [SES-CD] for CD), and histologic-endoscopic mucosal improvement (HEMI). Safety outcomes included rates of any adverse event (AE), serious AE (SAE), infections, AEs leading to treatment discontinuation, and deaths.

Risk of bias assessment

Risk of bias was assessed using the Cochrane Risk of Bias 2.0 (RoB 2.0) tool for RCTs (ARTEMIS-UC, TUSCANY-2, RELIEVE UCCD).[17] The Risk of Bias in Non-Randomised Studies of Interventions (ROBINS-I) tool was applied to APOLLO-CD and TUSCANY-1 as single-arm studies employing historical placebo comparators.[18] Each RCT was evaluated across five domains: randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result. Non-RCTs were evaluated across seven ROBINS-I domains including confounding, participant selection, classification of interventions, deviations, missing data, outcome measurement, and selection of reported results.

Data synthesis

Random-effects meta-analysis using the Der Simonian–Laird method was performed for outcomes reported by two or more randomized, placebo-controlled trials with comparable endpoint definitions.[19] Risk ratios (RR) with 95% confidence intervals (CI) were calculated from dichotomous event data. Risk differences (RD) were pooled using the same Der Simonian–Laird random-effects framework, and number needed to treat (NNT) was derived as 1/RD. Dose-arm and analysis-population selection decisions were prespecified before data extraction to minimize post hoc bias, guided by the following a priori rationale. For ARTEMIS-UC, the prespecified CDx-positive combined analysis (Cohorts 1 + 2 pooled, N = 75) was used as the primary analysis population for meta-analysis, with Cohort 1 all-comers (N = 135) additionally reported in the main analysis alongside the CDx-positive estimate. The CDx-positive population was selected because the trialists designated this as a key prespecified analysis and because its placebo remission rate (11%) was more representative of contemporary UC trials than the atypically low 1% observed in the Cohort 1 all-comers population. For TUSCANY-2, the 450 mg arm was selected as the expected Phase 3 dose based on the plateau in dose–response observed across efficacy endpoints in the original publication; mMS–based clinical remission (secondary endpoint) was used to harmonize with the mMS-based primary endpoints of ARTEMIS-UC and RELIEVE UCCD, as the mMS aligns with the current FDA draft guidance for UC drug development. The three TUSCANY-2 doses produced similar efficacy (mMS-based clinical remission 30%, 35%, and 32% for the 50, 150, and 450 mg arms, respectively), indicating a dose–response plateau; the highest tested dose (450 mg) was therefore selected as the arm most likely to be advanced to Phase 3, and excluding TUSCANY-2 altogether in a sensitivity analysis did not alter the pooled findings. For RELIEVE UCCD, the 900 mg arm was selected based on the prespecified Bayesian dose–response analysis reported in the original publication, which identified 900 mg as the maximally efficacious dose. We acknowledge that selecting optimally performing dose arms and, in the case of ARTEMIS-UC, a biomarker-enriched subpopulation, may bias pooled estimates toward larger treatment effects relative to what would be observed in unselected populations at non-optimized doses; this is addressed as a limitation. Between-study heterogeneity was assessed using Cochran’s Q test and the I² statistic, with I² values of 25%, 50%, and 75% representing low, moderate, and high heterogeneity, respectively.[20] Publication bias was assessed by funnel plot inspection, Egger’s weighted regression test,[21] Begg-Mazumdar rank correlation test,[22] and the Duval-Tweedie trim-and-fill method (R0 estimator).[23] All formal tests have severely limited statistical power with k = 3 studies,[24] and results are reported for completeness rather than definitive inference. Sensitivity analyses were conducted by excluding TUSCANY-2 given its use of a secondary rather than primary endpoint. Single-arm studies (APOLLO-CD, TUSCANY-1) were excluded from quantitative synthesis due to absence of concurrent placebo comparators. For CD, only one RCT (RELIEVE UCCD CD cohort) was available; results are reported as individual study estimates without pooling. The single CD RCT was therefore deliberately not meta-analyzed and is presented narratively as an individual study estimate; it is retained to provide a complete Phase 2 class-level picture across UC and CD, whereas all quantitative pooling and the class-level inference derive solely from the three UC RCTs.

Dose-response was assessed descriptively within the RELIEVE UCCD trial that randomized patients to two active dose levels (450 mg and 900 mg) and placebo, by comparing RDs across doses for UC clinical remission, UC endoscopic improvement, and CD endoscopic response.

All analyses were performed using Python 3.12 (NumPy, SciPy) and verified with R 4.3 (meta package, metabin function with Mantel–Haenszel method).

RESULTS

Study selection

The systematic search identified 869 records from databases and registers (PubMed n = 223, EMBASE n = 535, Cochrane Library n = 86, trial registers n = 20, and conference proceedings n = 5) and four additional records identified through other methods (manufacturer websites; no records were identified through organizations or citation searching). After removal of 437 duplicates, 432 records were screened by title and abstract, of which 153 full-text articles were assessed for eligibility. One hundred and forty three records were excluded (2 narrative reviews, 1 meta-analysis, and 140 studies with insufficient data). Ten records reporting results from five distinct Phase 2 clinical trials (six studies) met final inclusion criteria [Figure 1]: ARTEMIS-UC,[10] TUSCANY-1,[11] TUSCANY-2,[13] RELIEVE UCCD UC cohort,[14] APOLLO-CD,[25] and RELIEVE UCCD CD cohort.[26] Multiple publications reporting different aspects of the same trial (e.g. primary results and biomarker analyses for TUSCANY-1) were linked and counted as one study. In addition, four web-site based records including companion biomarker data,[12] manufacturer press releases,[15,27] and a supplementary conference abstract[28] provided data not reported in the primary publications and were used to extract additional efficacy, safety, and mechanistic information.

Figure 1.

PRISMA flow diagram. PRISMA, preferred reporting items for systematic reviews and meta-analyses

Study characteristics

Six Phase 2 studies evaluating three distinct anti-TL1A antibodies enrolled a total of 804 patients [Table 1]. Four studies enrolled patients with UC (ARTEMIS-UC,[10] TUSCANY-1,[11,12] TUSCANY-2,[13] RELIEVE UCCD UC cohort[14,15]) and two enrolled patients with CD (APOLLO-CD,[25] RELIEVE UCCD CD cohort[26,28]). Study designs ranged from open-label single-arm (TUSCANY-1, APOLLO-CD) to randomized, double-blind, placebo-controlled (ARTEMIS-UC, TUSCANY-2, RELIEVE UCCD). Publication status varied from peer-reviewed journal articles[10,13,25] to conference abstracts,[14,26,28] with supplementary data reported in manufacturer press releases.[15,27] All studies enrolled adults with moderately to severely active disease and permitted prior advanced therapy exposure, though the proportion of treatment-experienced patients varied across trials.

Table 1.

Characteristics of Phase 2 clinical trials of TL1A inhibitors in IBD

Study, Author	Year	Country	Drug	Phase	Design	Disease	Number randomized (treated)	Dose regimen	Duration	Primary endpoint	NCT number
ARTEMIS-UC[10]	2024	14 (Australia, Belgium, Bulgaria, Canada, Czechia, France, Georgia, Hungary, Israel, Italy, Poland, Ukraine, UK, USA)	Tulisokibart (MK-7240)	2	R, DB, PC	UC	178 (C1: 135; C2: 43)	IV 1000 mg D1, then 500 mg week 2, 6, 10	12 weeks	Clinical remission (mMS), week 12	NCT04996797
APOLLO-CD[25]	2025	7 (Australia, Belgium, Canada, Czechia, France, Poland, US)	Tulisokibart (MK-7240)	2a	OL, SA	CD	55	IV 1000 mg D1, then 500 mg week 2, 6, 10	12 weeks	Endoscopic response (≥50% SES-CD ↓) + Safety	NCT05013905
TUSCANY-1[11,12]	2022	6 (Belgium, Italy, Netherlands, Poland, South Korea, USA)	Afimkibart (PF-06480605)	2a	OL, SA (Simon 2-stage)	UC	50	IV 500 mg Q2W×7 doses	14 weeks	Endoscopic improvement (MES 0/1) + Safety	NCT02840721
TUSCANY-2[13]	2025	23 (Australia, Belgium, Bulgaria, Colombia, France, Germany, Hungary, India, Italy, Japan, Mexico, Poland, Romania, Russia, Serbia, Slovakia, South Africa, Spain, Thailand, Turkey, Ukraine, UK, USA)	Afimkibart (RG6631)	2b	R, DB, PC	UC	246	SC 50, 150, or 450 mg q 4 weeks	14 weeks	tMS clinical remission, week 14	NCT04090411
RELIEVE UCCD (UC)[14] and Press release data[15]	2025	22 (Austria, Belgium, Bulgaria, Canada, Czechia, France, Georgia, Germany, Hungary, Israel, Italy, Japan, Moldova, Norway, Poland, Serbia, Slovakia, South Africa, Spain, Ukraine, UK, USA)	Duvakitug (TEV-48574)	2b	R, DB, PC	UC	137	SC 2250 mg load, then 450 or 900 mg q 2 weeks	14 weeks	Clinical remission (mMS), week 14	NCT05499130
RELIEVE UCCD (CD)[26] and Oral presentation abstract[28]	2025	22 (Austria, Belgium, Bulgaria, Canada, Czechia, France, Georgia, Germany, Hungary, Israel, Italy, Japan, Moldova, Norway, Poland, Serbia, Slovakia, South Africa, Spain, Ukraine, UK, USA)	Duvakitug (TEV-48574)	2b	R, DB, PC	CD	138	SC 2250 mg load, then 450 or 900 mg q 2 weeks	14 weeks	Endoscopic response (≥50% SES-CD ↓), week 14	NCT05499130

IBD, inflammatory bowel disease; R, randomized; DB, double-blind; PC, placebo-controlled; OL, open-label; SA, single-arm; C1/C2, Cohort 1/2; mMS, modified Mayo Score; tMS, total Mayo Score; SES-CD, Simple Endoscopic Score for Crohn’s Disease; MES, Mayo Endoscopic Sub score. APOLLO-CD: 50 patients comprised the per-protocol analysis population. In TUSCANY-2, 245 of 246 randomized patients received treatment

Risk of bias assessment

Risk of bias assessment results are summarized in Table 2. Among the three RCTs assessed with Cochrane RoB 2.0, ARTEMIS-UC[10] was judged to have low risk of bias across all five domains: adequate computer-generated randomization with stratification by Companion Diagnostic (CDx) status and prior advanced therapy exposure, double-blind design with matched placebo, low attrition (<5%), centrally read endoscopic endpoints, and a preregistered statistical analysis plan. RELIEVE UCCD[14,26] was similarly judged low risk of bias, with adequate randomization, double-blinding, centrally adjudicated endpoints, and preregistered Bayesian analysis. TUSCANY-2[13] was judged to have some concerns: while randomization and blinding were adequate, the complex 9-arm randomization scheme (2:2:2:2:2:3:1:1:1) resulted in small per-arm sample sizes, the primary endpoint (total Mayo Score [tMS]) differed from FDA-recommended mMS, and the higher-than-expected placebo rate (12% vs assumed 6%) raised concerns about the appropriateness of the statistical assumptions.

Table 2.

Risk of bias assessment (ROB 2.0 for RCTs and ROBINS-I for non-RCTs)

Study	D1: Randomization		D2: Deviations		D3: Missing data	D4: Measurement	D5: Reporting	Overall
ARTEMIS-UC[10]	Low (computer-generated, stratified by CDx/prior Tx)		Low (DB, matched placebo, no crossover)		Low (attrition <5%, balanced across arms)	Low (central endoscopy reading, validated mMS)	Low (preregistered SAP, hierarchical testing)	Low
TUSCANY-2[13]	Low (adequate randomization, IVRS)		Low (DB, matched placebo)		Low (attrition balanced)	Some concern (primary endpoint used tMS including subjective PGA, not FDA-preferred mMS)	Some concern (complex 9-arm design with small per-arm N; higher-than-assumed PBO response rate reduced power)	Some concern
RELIEVE UCCD[14,26]	Low (centralized 1:1:1, stratified by disease/Tx history)		Low (DB, matched placebo)		Low (completion >90%)	Low (central endoscopy, validated mMS/SES-CD)	Low (preregistered Bayesian analysis with predefined PP threshold)	Low
Study	D1: Confounding	D2: Selection	D3: Classification	D4: Deviations	D5: Missing data	D6: Measurement	D7: Selection of results	Overall
APOLLO-CD[25]	Serious (no concurrent control; historical PBO from heterogeneous published trials with different populations/timepoints)	Moderate (single-arm enrollment may select motivated patients; 101 screened/55 enrolled)	Low (standardized IV dosing protocol)	Low (open-label but objective SES-CD endpoint)	Low (50/55 completed per-protocol)	Low (centrally read SES-CD)	Low (preregistered NCT; co-primary endpoints reported)	Serious
TUSCANY-1[11]	Serious (no concurrent control; historical null hypothesis without matched comparator)	Moderate (Simon’s two-stage design may enrich for responders in stage 2)	Low (standardized IV dosing)	Low (open-label but objective MES endpoint)	Low (46/50 completed)	Low (centrally read MES)	Low (preregistered NCT; primary endpoint reported)	Serious

RCT, randomized controlled trial; RoB 2, Cochrane Risk of Bias 2.0 tool; ROBINS-I, Risk of Bias in Non-Randomized Studies of Interventions; CDx, Companion Diagnostic; Tx, treatment; DB, double-blind; mMS, modified Mayo Score; SAP, Statistical Analysis Plan; IVRS, Interactive Voice Response System; tMS, total Mayo Score; PGA, Physician’s Global Assessment; PBO, placebo; SES-CD, Simple Endoscopic Score for Crohn’s Disease; PP, Posterior Probability. MES, Mayo Endoscopic Sub score. For ARTEMIS-UC, the CDx-positive combined analysis (Cohorts 1+2) was assessed as the prespecified primary population. TUSCANY-2 and RELIEVE UCCD were assessed based on their preregistered primary endpoints and SAPs

For the two single-arm studies assessed with ROBINS-I, APOLLO-CD[25] was judged to have serious risk of bias, primarily due to the open-label design, absence of a concurrent control group, and reliance on a historical placebo comparator derived from a meta-analysis of placebo response rates from six prior CD trials[25] [Table S1], with inherent confounding from cross-study population differences. TUSCANY-1[11] was similarly judged to have serious risk of bias as an open-label, single-arm study using a historical null hypothesis without a concurrent comparator.

Tulisokibart in ulcerative colitis: ARTEMIS-UC

The ARTEMIS-UC trial was a Phase 2, multicenter, randomized, double-blind, placebo-controlled study of tulisokibart in moderately to severely active UC, conducted across 14 countries.[10] Tulisokibart, a humanized IgG1κ monoclonal antibody binding both soluble and membrane-bound TL1A, was given intravenously (1000 mg Day 1, then 500 mg at Weeks 2, 6, 10) to patients with moderately-to-severely active UC and prior failure of conventional or advanced therapies.[10]

The trial comprised two cohorts: Cohort 1 (N = 135) enrolled all-comers regardless of CDx status, and Cohort 2 (N = 43) enrolled only CDx-positive patients identified by a proprietary genetic-based assay.[10]

The primary endpoint was clinical remission at Week 12 in Cohort 1, defined as a modified MES of 0 or 1. Tulisokibart achieved 26% clinical remission versus 1% for placebo (difference 25 percentage points [pp]; 95% CI 14–37; P < 0.001).[10] All ranked secondary endpoints in Cohort 1 were also met [Table 3].[10] The unusually low placebo remission rate of 1% in Cohort 1 (compared with typical rates of 6–15% in contemporary UC trials) amplified the observed treatment differences.

Table 3.

Efficacy outcomes of Phase 2 clinical trials of TL1A inhibitors in IBD

Study (UC)	Endpoint	Active Drug (Dose)	Placebo	Difference (pp)	P/PP
ARTEMIS-UC (C1)[10]	Clinical remission (mMS), week 12	Tulisokibart: 26%	1%	25 (95% CI 14–37)	P<0.001
ARTEMIS-UC (C1)[10]	Endoscopic improvement, week 12	37%	6%	31 (95% CI 17–43)	P<0.001
ARTEMIS-UC (C1)[10]	HEMI, week 12	31%	4%	27 (95% CI 14–40)	P<0.001
ARTEMIS-UC (C1)[10]	Histologic improvement, week 12	46%	18%	29 (95% CI 12–43)	P<0.001
ARTEMIS-UC (C1)[10]	Clinical response (mMS), week 12	66%	22%	44 (95% CI 27–57)	P<0.001
ARTEMIS-UC (CDx+)[10]	Clinical remission (mMS), week 12	32%	11%	21 (95% CI 2–38)	P=0.02
ARTEMIS-UC (CDx+)[10]	Endoscopic improvement, week 12	37%	19%	18 (95% CI−2 to 36)	P=0.06 (NS)
TUSCANY-1[11]	Endoscopic improvement, week 14	Afimkibart: 38.2% (UMVUE)	— (SA)	—	P<0.001 vs null
TUSCANY-1[11]	Clinical remission (tMS), week 14	24%	—	—	—
TUSCANY-2 (1°)[13]	tMS clinical remission, week 14	50 mg: 26%; 150 mg: 23%; 450 mg: 24%	12%	13.9/11.7/12.2	P=0.054/0.082/0.064 (NS)
TUSCANY-2 (2°)[13]	mMS clinical remission, week 14	50 mg: 30%; 150 mg: 35%; 450 mg: 32%	12%	18.2/23.4/20.2	NR (hierarchy stopped)
TUSCANY-2 (2°)[13]	Endoscopic improvement, week 14	38–41% (across doses)	19%	~19–22	NR
TUSCANY-2 (2°)[13]	HEMI, week 14	50 mg: 30%; 150 mg: 32%; 450 mg: 26%	5%	25/27/21	NR
RELIEVE UC[14,15]	Clinical remission (mMS), week 14	450 mg: 36%; 900 mg: 48%	20%	16/27	P=0.050/P=0.003; PP>0.90
RELIEVE UC[14,15]	Endoscopic improvement, week 14	450 mg: 45%; 900 mg: 50%	23%	22/27	NR
RELIEVE UC[14,15]	HEMI, week 14	450 mg: 30%; 900 mg: 33%	16%	14/17	NR
Study (CD)	Endpoint	Active Drug (Dose)	Placebo/Historical	Difference (pp)	P/PP
APOLLO-CD[25]	Endoscopic response (≥50% SES-CD), week 12	Tulisokibart: 26.0% (13/50)	12% (historical)	14	P=0.002
APOLLO-CD[25]	Clinical remission (CDAI<150), week 12	49.1%	16% (historical)	33	P<0.001
RELIEVE CD[26,28]	Endoscopic response, week 14	450 mg: 26%; 900 mg: 48%	13%	13/35	P=0.058/P<0.001; PP>0.90
RELIEVE CD[26,28]	Endoscopic remission, week 14	450 mg: 17%; 900 mg: 26%	9%	8/17	NR
RELIEVE CD[26,28]	Clinical remission (CDAI<150), week 14	450 mg: 50%; 900 mg: 54%	41%	9/13	NR
RELIEVE CD[26,28]	Clinical response (PRO-2), week 14	450 mg: 50%; 900 mg: 53%	29%	21/24	NR

IBD, inflammatory bowel disease; C1, Cohort 1; CDx+, companion diagnostic positive; SA, single-arm; UMVUE, uniformly minimum variance unbiased estimate; 1°, primary endpoint; 2°, secondary endpoint; PP, Bayesian posterior probability; NS=not significant; NR, not reported; pp, percentage points; UC, ulcerative colitis; CD, Crohn’s disease

In the prespecified combined CDx-positive analysis (N = 75), clinical remission was 32% versus 11% (difference 21 pp; P = 0.02), but endoscopic improvement did not reach significance (37% vs 19%; P = 0.06), terminating the hierarchical testing procedure.[10]

Safety across all 178 patients was favorable [Table S2]. SAE were less frequent with tulisokibart (1% vs 8%), infections were balanced, and no deaths or infusion reactions occurred.[10]

Tulisokibart in Crohn’s disease: APOLLO-CD

APOLLO-CD was a Phase 2a, open-label, single-arm study of tulisokibart in moderately-to-severely active CD.[25] Eligible patients had moderately-to-severely active CD with prior insufficient response or intolerance to conventional and/or biologic therapies.[25]

Of 55 treated patients, 50 comprised the per-protocol set.[25] The co-primary endpoint of Week-12 endoscopic response (≥50% SES-CD decrease) was achieved by 26.0% (13/50; 95% CI 15.9–39.6) versus a 12% historical placebo rate (P = 0.002).[25] Clinical remission (CDAI <150) was 49.1% versus 16% historical (P < 0.001) [Table 3].[25]

Tissue RNA-sequencing showed that tulisokibart modulated Th1, Th17, inflammatory-monocyte, and fibrosis pathways in endoscopic responders, supporting a pleiotropic mechanism.[25]

AEs occurred in 78% of patients [Table S2]. SAE (15%) were none drug-related, infections were mild-to-moderate, and no deaths occurred.[25] A subsequent erratum corrected a figure label without changing numerical data.[29]

Afimkibart in ulcerative colitis: TUSCANY-1 and TUSCANY-2

TUSCANY-1 was a Phase 2a, open-label, single-arm study of intravenous PF-06480605 (now afimkibart) in moderately-to-severely active UC (50 patients).[11] Week-14 endoscopic improvement was 38.2% (90% CI 23.8–53.7; P < 0.001 vs historical null).[11] Clinical remission (tMS ≤ 2) was 24% [Table 3].[11] The companion biomarker showed downregulation of tissue inflammation and fibrosis pathways and reduced activated-fibroblast gene expression.[12] A high 82% anti-drug antibody rate with the IV formulation prompted reformulation to subcutaneous delivery.[11]

TUSCANY-2 was a Phase 2b, randomized, double-blind, placebo-controlled, dose-ranging trial of subcutaneous afimkibart in moderately-to-severely active UC.[13] A total of 246 patients were randomized to afimkibart 50, 150, or 450 mg SC every 4 weeks versus placebo, with efficacy assessed at Week 14.[13]

The primary endpoint of Week-14 tMS clinical remission was not met for any dose (24–26% across doses vs 12% placebo; all one-sided P > 0.05).[13] This partly reflected a higher-than-expected placebo response rate that reduced statistical power.[13]

However, mMS-based secondary endpoints showed consistent efficacy across all doses (clinical remission 30–35% vs 12% placebo).[13] Endoscopic improvement was 38–41% versus 19% placebo, and HEMI 26–32% versus 5% [Table 3].[13] A prespecified biomarker-positive subgroup showed enhanced efficacy (mMS remission 37% and endoscopic improvement 51%, both vs 10% placebo).[13]

Safety was reassuring (treatment-emergent AE ~48%) [Table S2]. No deaths occurred. No neutralising antibodies were detected at Week 56 with the subcutaneous formulation.[13]

Duvakitug in ulcerative colitis and Crohn’s disease: RELIEVE UCCD

The RELIEVE UCCD trial was a Phase 2b, randomized, double-blind, placebo-controlled, dose-ranging basket trial of duvakitug in both UC and CD.[14,15,26,28] Duvakitug is a human IgG1-λ2 antibody preferentially inhibiting TL1A–DR3 signalling with reduced DcR3 blockade.[15] Patients received a 2250 mg SC load, then 450 or 900 mg SC Q2W or placebo for 14 weeks.[14,26] This trial is notable as the first randomized, placebo-controlled study of a TL1A inhibitor in CD

UC Cohort (N = 137): the primary endpoint, Week-14 mMS clinical remission, showed dose-dependent efficacy, 36% (450 mg) and 48% (900 mg) versus 20% placebo (placebo-adjusted 16 and 27 pp; one-sided P = 0.050 and 0.003; posterior probability > 0.90).[14,15] Endoscopic improvement was 45–50% versus 23% placebo [Table 3].[14,15]

CD Cohort (N = 138): the primary endpoint, Week-14 endoscopic response (≥50% SES-CD reduction), showed a dose-response, 26% (450 mg) and 48% (900 mg) versus 13% placebo (P = 0.058 and <0.001; posterior probability >0.90).[26,28] Clinical remission (CDAI <150) was 50–54% versus 41% placebo [Table 3].[26,28] Among advanced therapy–experienced patients, the 900 mg dose achieved 48% endoscopic response versus 4% placebo-adjusted.[26]

Safety across both cohorts was favourable [Table S2]. No deaths were reported in either cohort.[14,15,26,28]

Meta-analysis of ulcerative colitis outcomes

Three RCTs (ARTEMIS-UC,[10] TUSCANY-2,[13] RELIEVE UCCD UC cohort[14]) contributed to the meta-analysis of UC outcomes. For each trial, a single prespecified active-treatment arm was selected for pooling rather than including multiple dose arms against a shared placebo group, which would have double-counted placebo patients and introduced a unit-of-analysis error: ARTEMIS-UC CDx-positive combined (N = 38 vs 37 placebo), TUSCANY-2 450 mg (N = 88 vs 43 placebo), and RELIEVE UCCD 900 mg (N = 46 vs 44 placebo), yielding 172 active, and 124 placebo patients [Figure 2].

Figure 2.

Forest plot of meta-analysis of UC outcomes. Each row represents a randomized controlled trial, with squares indicating point estimates of RR and horizontal lines representing 95% CIs. Diamond at the bottom shows the pooled effect estimate. Panels a and b show the pooled UC analyses (clinical remission and endoscopic improvement, respectively) across the three RCTs, and Panel c shows CD endoscopic response from the single RELIEVE UCCD Crohn’s-disease-cohort RCT, presented without pooling; squares are colored by agent (tulisokibart, afimkibart, duvakitug) and sized by study weight. For each trial, the single prespecified active-treatment arm was used (ARTEMIS-UC CDx-positive combined, n = 38 vs 37 placebo; TUSCANY-2 450 mg, n = 88 vs 43 placebo; RELIEVE UCCD 900 mg, n = 46 vs 44 placebo), yielding 172 active and 124 placebo patients. RR = risk ratio; CI = confidence interval; UC = ulcerative colitis; RCTs = randomized controlled trials

For clinical remission, the pooled random-effects RR was 2.56 (95% CI 1.60–4.09; P < 0.001), with no evidence of between-study heterogeneity (I² =0.0%, Q = 0.16, P-heterogeneity = 0.92) [Figure 2a], with individual study RRs of 2.34–2.92.

For endoscopic improvement, the pooled random-effects RR was 2.13 (95% CI 1.44–3.16; P < 0.001), again with negligible heterogeneity (I² = 0.0%, Q = 0.07, P-heterogeneity = 0.97) [Figure 2b], with individual RRs of 1.95–2.20.

The unselected ARTEMIS-UC Cohort 1 all-comers analysis 25/68 vs 4/67 for endoscopic improvement) yielded higher pooled RRs, 3.21 (95% CI 1.48–6.97) for clinical remission and 2.77 (95% CI 1.59–4.82) for endoscopic improvement, but with moderate heterogeneity (I² = 44.7% and 40.7%, respectively), driven by the exceptionally low 1% placebo remission rate in Cohort 1.[10] Exclusion of TUSCANY-2 in the CDx+ analysis did not materially alter the findings: clinical remission RR 2.49 (95% CI 1.43–4.34, I² = 0.0%) and endoscopic improvement RR 2.10 (95% CI 1.29–3.41, I² = 0.0%). Substituting the alternative randomized dose arm of RELIEVE UCCD (450 mg instead of 900 mg) likewise did not alter the conclusions, attenuating the pooled estimates only modestly while preserving statistical significance and homogeneity: clinical remission RR 2.26 (95% CI 1.37–3.71, I² = 0.0%) and endoscopic improvement RR 2.02 (95% CI 1.33–3.06, I² = 0.0%); this modest reduction is consistent with the within-trial dose–response and indicates that the choice of dose arm did not drive the pooled result.

Pooled RDs, less sensitive to placebo-rate variability than RR, were +22.2% for clinical remission (95% CI 12.8–31.5%; I² = 0.0%), yielding a NNT of 4.5 (95% CI 3.2–7.8). For endoscopic improvement, the pooled RD was 22.6% (95% CI 12.3–32.9%; I² = 0.0%), with NNT 4.4 (95% CI 3.0–8.1), and consistent individual study RDs.

Dose-response analysis

Within the RELIEVE UCCD trial, which randomized patients to two active dose levels, a clear dose-response relationship was observed. In UC, clinical remission rates were 36% (450 mg) versus 48% (900 mg) versus 20% (placebo), with RDs of +15.6 and +27.3 pp, respectively. In CD, endoscopic response rates were 26% (450 mg, 12/46) versus 48% (900 mg, 22/46) versus 13% (placebo, 6/46), yielding RR 2.00 and 3.67, respectively.[14,15,26,28] The 900 mg dose approximately doubled the treatment effect of the 450 mg dose across both diseases and endpoints.

Publication bias

Funnel plots for both UC outcomes showed no visual asymmetry [Figure 3]. For clinical remission, Egger’s regression test yielded an intercept of 1.20 (P = 0.08, df = 1) and Begg-Mazumdar rank correlation was τ = 1.00 (P = 0.33). Duval-Tweedie trim and fill estimated two missing studies on the left; the adjusted pooled RR was 2.48 (95% CI 1.69–3.64), minimally attenuated from the unadjusted estimate of 2.56. For endoscopic improvement, Egger’s test was non-significant (intercept − 1.44, P = 0.24), Begg-Mazumdar rank correlation was τ = −1.00 (P = 0.33), and trim and fill yielded an adjusted RR of 2.11 (95% CI 1.56–2.84), virtually unchanged from 2.13. All formal tests are severely underpowered with only three studies,[24] and these results should be interpreted as exploratory rather than definitive. With only three pooled studies the funnel plot is unreliable for detecting small-study effects, and formal meta-regression to explore sources of heterogeneity was not feasible; these publication-bias analyses are therefore reported for completeness and interpreted with caution.

Figure 3.

Funnel plots for assessment of publication bias in the ulcerative colitis meta-analysis. (a) UC clinical remission; (b) UC endoscopic improvement. Each point represents one study, plotted by effect size (log RR, x-axis) against precision (standard error, y-axis). Dashed diagonal lines represent pseudo-95% confidence limits. RR = risk ratio; UC = ulcerative colitis

Crohn’s disease outcomes (single randomized controlled trial)

Only one RCT (RELIEVE UCCD CD cohort) was available for quantitative analysis of CD outcomes. At the 900 mg dose, duvakitug achieved an endoscopic response RR of 3.67 (95% CI 1.64–8.20; P = 0.002) compared with placebo, with a RDs of 34.8% (95% CI 17.4–52.2%) and NNT of 2.9.[26,28] APOLLO-CD, as an open-label single-arm study with historical comparator and serious risk of bias [Table 2], was not included in the quantitative synthesis but reported a 26% endoscopic response rate (vs 12% historical placebo, P = 0.002).[25]

Safety profile across the TL1A inhibitor class

Across all six Phase 2 studies, TL1A inhibitors demonstrated no unexpected safety signals during the induction phase [Table S2]. All safety data reported here reflect induction-period exposure only (12–14 weeks); long-term maintenance safety data were not available for any agent at the time of this analysis. No deaths occurred in any trial. SAE rates were low and generally comparable to or lower than placebo in controlled studies. In ARTEMIS-UC, the SAE rate was lower with tulisokibart (1%) than placebo (8%).[10] Infection rates were balanced between active treatment and placebo across all RCTs, with no serious or opportunistic infections attributed to TL1A inhibition.[10,13,14,26] The single notable safety signal was the high anti-drug antibody rate (82%) with IV afimkibart in TUSCANY-1 that was resolved by switching to subcutaneous administration in TUSCANY-2 (0% neutralizing antibodies at Week 56).[11,13]

Supplementary sensitivity and certainty analyses

Leave-one-out sensitivity analysis confirmed the stability of UC pooled estimates: sequential exclusion of each study yielded clinical remission RRs ranging from 2.47 to 2.81, all remaining statistically significant, with I² remaining 0.0% in all permutations [Table S1]. The 95% prediction interval could not be meaningfully calculated with only three studies (df = 1), yielding an uninformatively wide range [Table S3]; this limitation indicates the need for Phase 3 data to narrow the expected range of the true treatment effect.[24] Fixed-effect and random-effects models produced identical results, as expected given zero observed heterogeneity [Table S4]. As a methodological sensitivity analysis for small meta-analyses, Hartung–Knapp–Sidik–Jonkman (HKSJ) correction was applied: all pooled estimates remained statistically significant (clinical remission RR 2.59, HKSJ 95% CI 1.85–3.61, P = 0.007; endoscopic improvement RR 2.10, HKSJ 95% CI 1.81–2.44, P = 0.002), confirming robustness to small-sample CI adjustment [Table S4]. An additional sensitivity analysis substituting TUSCANY-2’s primary tMS-based clinical remission (24% vs 12% placebo) for the mMS-based secondary endpoint yielded a pooled RR of 2.38 (95% CI 1.46–3.89; I² =0.0%), remaining statistically significant, but with an attenuated RD of +19.5% (95% CI 9.1–29.8%; NNT 5.1), consistent with the known lower sensitivity of tMS relative to mMS.

A UC-only pooled safety analysis of SAE was performed, restricted to the two UC RCTs with arm-level SAE data (ARTEMIS-UC and the RELIEVE UCCD UC cohort), comparing tulisokibart (1/90) and duvakitug 900 mg (1/46) with placebo (7/88 and 1/44, respectively). Because SAE were rare, the pooled estimate was derived with the Mantel–Haenszel method: the SAE RR was 0.24 (95% CI 0.05–1.13) that did not reach statistical significance (a Peto odds ratio of 0.28 [95% CI 0.08–0.99] gave a similar, borderline estimate) [Table S5]. The numerically lower SAE rate with TL1A inhibitors is therefore imprecise and should be regarded as exploratory, likely reflecting effective disease control rather than a direct protective drug effect. Within the RELIEVE UCCD UC cohort, SAE were infrequent in all arms (duvakitug 450 mg 0/47, 900 mg 1/46, placebo 1/44); the previously reported higher SAE rate in the duvakitug 450 mg arm (6/46, 13%) occurred in the CD cohort, not in UC. This analysis was restricted to UC to align it with the UC-only efficacy pool, replacing an earlier cross-indication pool that had combined ARTEMIS-UC with the CD cohort. When HEMI was pooled across the three UC RCTs, the pooled RR was consistent with primary efficacy outcomes [Table S6].

GRADE assessment rated the certainty of evidence as moderate for UC clinical remission and endoscopic improvement (downgraded one level for indirectness due to use of a secondary endpoint in TUSCANY-2, CDx-positive subgroup analysis in ARTEMIS-UC, and the Phase 2 setting with limited generalizability), low for CD outcomes (single RCT, serious imprecision, and reliance on non-peer-reviewed conference abstracts), and moderate for the overall safety profile [Table S7]. For outcomes informed predominantly by non-peer-reviewed data sources (conference abstracts, press releases), certainty was further tempered given the potential for selective reporting and incomplete data availability.

DISCUSSION

Principal findings

This SR and meta-analysis provide the first quantitative synthesis of all Phase 2 TL1A inhibitor data in IBD, pooling three RCTs of three structurally distinct anti-TL1A antibodies (tulisokibart, afimkibart, duvakitug) in UC. The pooled RRs were 2.56 (95% CI 1.60–4.09) for clinical remission and 2.13 (95% CI 1.44–3.16) for endoscopic improvement, both with I² = 0.0%, corresponding to pooled RDs of +22.2 percentage points (NNT 4.5) and +22.6 percentage points (NNT 4.4), respectively.[10,13,14,15,25,26,28] While this negligible statistical heterogeneity is encouraging, it must be interpreted with caution given that the Q-test has severely limited power to detect true heterogeneity when k = 3, and I² = 0.0% may reflect insufficient power rather than genuinely absent variance. Nevertheless, the consistency across trials differing in sponsor, route (IV vs SC), dosing, and endpoint definition (mMS vs tMS) is notable. The all-comers analysis raised I² from 0.0% to 44.7%, but this reflected placebo-rate variability (1% in Cohort 1 vs 11–20% elsewhere) rather than differential efficacy, confirming the CDx-positive population yielded the most homogeneous pooling. The endpoint distinction also proved consequential: TUSCANY-2 failed its tMS-based primary endpoint,[30] yet demonstrated consistent efficacy on the mMS-based secondary endpoint endorsed by FDA 2022 guidance,[31] suggesting that the primary endpoint failure reflects endpoint selection rather than inadequate drug efficacy. For CD, only one RCT (RELIEVE UCCD) provided controlled data, with an endoscopic response RR of 3.67 (95% CI 1.64–8.20) at the 900 mg dose[26,28]; APOLLO-CD was open-label and excluded from pooling.[25] These findings collectively suggest that TL1A–DR3 axis blockade may exert a class-level therapeutic effect across structurally and pharmacologically distinct agents; however, because this inference rests on only three UC RCTs and a single CD RCT, with heterogeneity tests that are correspondingly underpowered, it remains tentative and requires confirmation in Phase 3 trials. These three antibodies are not pharmacologically equivalent, and “class-level” is used here descriptively rather than as an established pharmacologic claim: they differ in binding affinity and target epitope, mechanism of action (for example, duvakitug preferentially inhibits TL1A–DR3 signaling with reduced DcR3 blockade), Fc engineering, route and dosing schedule, and pharmacokinetics, and should not be regarded as a single interchangeable agent. The observed cross-trial consistency is therefore best interpreted as a hypothesis-generating signal to be confirmed in head-to-head and Phase 3 studies rather than as evidence of a uniform class effect. Although Phase 3 programs are ongoing, this Phase 2 class-level synthesis complements rather than substitutes for them: it tests whether efficacy is a reproducible property of TL1A–DR3 blockade rather than of any single molecule, and provides a prospectively updatable baseline to inform the design and interpretation of the confirmatory trials.

Preclinical data show that TL1A directly activates intestinal fibroblasts via DR3, and TL1A-neutralizing antibodies reversed established colonic fibrosis in murine models.[2,9] TUSCANY-112 and APOLLO-CD32 indicated that TL1A inhibition can modulate fibrosis-associated gene-expression pathways and activated-fibroblast signatures in human intestinal tissue, molecular signals rather than measures of clinical fibrosis. However, no Phase 2 trial included clinical fibrosis endpoints, and dedicated studies with fibrosis-specific outcomes are needed before anti-fibrotic claims can be substantiated. The anti-fibrotic potential of TL1A inhibition therefore remains speculative at the clinical level and should be regarded as a hypothesis to be tested rather than an established clinical effect.

Clinical implications and future perspectives

Tulisokibart’s development is paired with a CDx strategy, though ARTEMIS-UC results were equivocal: the CDx-positive combined analysis showed a smaller treatment effect (21 pp) than the all-comers Cohort 1 (25 pp), and endoscopic improvement in CDx-positive patients did not reach statistical significance.[10] TUSCANY-2’s prospective biomarker was more encouraging, with enhanced efficacy in biomarker-positive patients[13]; if validated, this would represent the first biomarker-paired therapy in IBD. Among the three agents, duvakitug demonstrated the clearest dose-response relationship, with 900 mg consistently outperforming 450 mg across UC and CD endpoints,[14,26] and is further differentiated by preferential inhibition of TL1A–DR3 signaling with reduced DcR3 blockade.[15] All three TL1A inhibitors have initiated Phase 3 programs: tulisokibart in ATLAS-UC and ARES-CD,[32] afimkibart in AMETRINE and SIBERITE,[33] and duvakitug in SUNSCAPE and STARSCAPE.[27] These Phase 3 results, expected from 2027 onwards, will determine whether the apparent Phase 2 class-level effect is confirmed in larger populations and whether CDx can refine patient selection. Biomarker-enriched populations and selected dose arms may overestimate effects relative to unselected practice.

Beyond efficacy confirmation, Phase 3 programs should address durability of response during maintenance, head-to-head positioning, whether anti-fibrotic effects translate to clinical benefit, and long-term safety including malignancy risk. Combination therapy with existing agents also warrants investigation given TL1A’s non-overlapping mechanism. Finally, real-world studies will be essential to confirm broader effectiveness.

Strengths and limitations

This SR and meta-analysis are the first quantitative synthesis of all Phase 2 TL1A inhibitor data in IBD, pooling three structurally distinct agents across three RCTs with preregistered protocol to evaluate whether TL1A–DR3 blockade exerts a possible class-wide therapeutic effect beyond individual molecule-level observations. The resulting I² = 0.0% across agents with different structures, routes, doses, and endpoint definitions is compatible with a possible mechanism-level therapeutic signal rather than molecule-specific efficacy, although the small number of trials limits the strength of this inference. Synthesizing early-phase trials to inform subsequent confirmatory trials is an established approach whose value is complementary to the ongoing Phase 3 programs; we intend this work as the baseline for a living review to be updated as the Phase 3 data emerge. Multiple sensitivity analyses (leave-one-out, prediction intervals, fixed- vs random-effects, and population substitution; Tables S1–S4) confirmed robustness, with both RRs and RDs reported.

This review also has important limitations. First, only three RCTs contributed to the UC meta-analysis: with k = 3, the Q-test has limited power, the observed I² = 0.0% may not reflect genuinely absent heterogeneity, I² is downward-biased,[24] the prediction interval was uninformative, and publication-bias tests are underpowered.[24] Despite this, the consistency of individual estimates (RRs 2.34–2.92; RDs 17.9–27.4 pp) across three independent programs provides qualitative confidence. Second, despite substantial clinical heterogeneity (enrichment, endpoints, duration, route, dosing), statistical homogeneity held; substituting ARTEMIS-UC Cohort 1 data raised I² to 44.7% via an atypically low 1% placebo rate, confirming that heterogeneity detection is population-dependent. The included trials also differed across several clinically important dimensions that warrant explicit consideration: disease type (UC versus CD), the proportion of patients with prior biologic or advanced-therapy exposure, endpoint definitions (mMS-based versus tMS-based clinical remission), induction duration (12 versus 14 weeks), route of administration (intravenous versus subcutaneous), and patient selection (biomarker-enriched CDx-positive enrollment in ARTEMIS-UC versus all-comers in the other trials). Such clinical heterogeneity limits direct cross-trial comparison and the interpretation of any pooled estimate. To mitigate this, quantitative pooling was deliberately restricted to the three UC RCTs that shared a placebo-controlled design and an mMS-based remission outcome, so that disease type was not mixed within the pooled analysis, whereas the single CD RCT and the two single-arm studies were summarized narratively; harmonized outcome definitions and prespecified analysis populations and dose arms were used to reduce residual heterogeneity. Nonetheless, given the small number of trials and these persisting clinical differences, the pooled estimates should be regarded as hypothesis-generating summaries of emerging evidence rather than definitive class-level effect sizes. Third, CD evidence is limited to a single RCT, and the class-effect argument is weaker for CD than UC. Fourth, the pooled safety meta-analysis is constrained by arm-level SAE data from only two trials; the UC-only pooled SAE estimate (Mantel–Haenszel RR 0.24, 95% CI 0.05–1.13) did not reach statistical significance, so any apparent safety advantage is imprecise and requires Phase 3 confirmation. Fifth, a substantial proportion of data derives from non-peer-reviewed sources, including conference abstracts and industry press releases. Specifically, RELIEVE UCCD efficacy data (both UC and CD cohorts) were available only as conference abstracts and manufacturer press releases at the time of analysis; the GRADE assessment was accordingly downgraded for these data. During revision, the RELIEVE UCCD ulcerative colitis and CD cohorts were published in full in a peer-reviewed journal, and the corresponding citations have been updated accordingly; because the reported values are unchanged, the pooled estimates are unaffected, while the certainty of this evidence is strengthened. Event counts were back-calculated from reported percentages in these non-peer-reviewed reports,[14,15,26,28] introducing potential rounding error and possible outcome misclassification. To mitigate this, reported percentages and denominators were cross-checked for internal consistency across all available sources (conference abstracts, oral presentations, and manufacturer press releases), the affected outcomes were downgraded in the GRADE assessment, and their influence on the pooled estimates was examined in the leave-one-out and dose-substitution sensitivity analyses; residual error from these unpublished sources cannot, however, be fully excluded until the corresponding peer-reviewed publications become available. Sixth, endpoint harmonization required using TUSCANY-2’s secondary mMS-based clinical remission rather than its primary tMS endpoint, representing a departure from the trial’s prespecified analysis. Seventh, the meta-analytic pooling relied on selected dose arms (TUSCANY-2 450 mg, RELIEVE UCCD 900 mg) and, for ARTEMIS-UC, a biomarker-enriched CDx-positive subpopulation, which may overestimate treatment effects relative to unselected, real-world IBD populations; the generalizability of these findings to broader clinical practice, particularly patients who would not meet biomarker-selection criteria, remains uncertain. Additional limitations include exclusion of two open-label single-arm studies from quantitative synthesis, evaluation of short-term induction endpoints only (12–14 weeks), absence of formal fibrosis endpoints despite promising translational data,[12,32] and the need for caution in interpreting subgroup analyses given limited sample sizes.

Notwithstanding these limitations, this study identifies key methodological considerations (including endpoint selection, placebo rate variability, and CDx strategy) that should inform the design and interpretation of ongoing Phase 3 programs.

Two reviews of TL1A inhibition in IBD were published in 2026: a PRISMA-guided, PROSPERO-registered SR with narrative synthesis,[34] and a narrative review.[35] Both summarized the same UC trials individually and concluded qualitatively that anti-TL1A agents are a promising class, but neither performed quantitative pooling. The present study adds the first random-effects meta-analysis of the three Phase 2 UC RCTs, providing pooled relative risks, RD, and NNTs with consistent point estimates across structurally distinct antibodies, with leave-one-out, dose-arm, and population-substitution sensitivity analyses and GRADE certainty rating. With only three trials, the observed I² = 0.0% reflects insufficient power to formally exclude between-agent heterogeneity rather than an established class-level effect, given differences in route of administration (IV vs SC), induction period (12 vs 14 weeks), and ligand-binding profile.

CONCLUSION

TL1A inhibitors represent a new mechanistic class in IBD therapeutics. Phase 2 data collectively suggest that blocking the TL1A-DR3 axis produces clinically meaningful improvements in clinical remission, endoscopic improvement, and HEMI in both UC and CD, with no major safety concerns identified across more than 800 treated patients. Whether preclinical anti-fibrotic signals translate to clinical benefit, and whether these efficacy signals are confirmed in larger, more diverse populations with longer follow-up, await Phase 3 determination. Since this analysis was completed, the first Phase 3 induction trial of a TL1A inhibitor (tulisokibart, ATLAS-UC) met its primary endpoint of clinical remission at week 12, providing early Phase 3 support for these Phase 2 findings.

Conflicts of interest

There are no conflicts of interest.

Supplementary file 1.

Section and Topic	Item #	Checklist item	Location where item is reported
TITLE
Title	1	Identify the report as a systematic review.	p1 (Title page)
ABSTRACT
Abstract	2	See the PRISMA 2020 for Abstracts checklist.	p4–5
INTRODUCTION
Rationale	3	Describe the rationale for the review in the context of existing knowledge.	p5–6 (Introduction, ¶1–3)
Objectives	4	Provide an explicit statement of the objective(s) or question(s) the review addresses.	p6 (Introduction, last ¶)
METHODS
Eligibility criteria	5	Specify the inclusion and exclusion criteria for the review and how studies were grouped for the syntheses.	p8 (Eligibility Criteria)
Information sources	6	Specify all databases, registers, websites, organisations, reference lists and other sources searched or consulted to identify studies. Specify the date when each source was last searched or consulted.	p7–8 (Search Strategy and Information Sources)
Search strategy	7	Present the full search strategies for all databases, registers and websites, including any filters and limits used.	p7–8; Supplementary file 2 (full search strategy)
Selection process	8	Specify the methods used to decide whether a study met the inclusion criteria of the review, including how many reviewers screened each record and each report retrieved, whether they worked independently, and if applicable, details of automation tools used in the process.	p8 (Eligibility Criteria, Selection process described)
Data collection process	9	Specify the methods used to collect data from reports, including how many reviewers collected data from each report, whether they worked independently, any processes for obtaining or confirming data from study investigators, and if applicable, details of automation tools used in the process.	p8 (Data Extraction and Outcomes)
Data items	10a	List and define all outcomes for which data were sought. Specify whether all results that were compatible with each outcome domain in each study were sought (e.g. for all measures, time points, analyses), and if not, the methods used to decide which results to collect.	p8 (Data Extraction and Outcomes)
	10b	List and define all other variables for which data were sought (e.g. participant and intervention characteristics, funding sources). Describe any assumptions made about any missing or unclear information.	p8 (Data Extraction and Outcomes); Table 1 (p29–30)
Study risk of bias assessment	11	Specify the methods used to assess risk of bias in the included studies, including details of the tool(s) used, how many reviewers assessed each study and whether they worked independently, and if applicable, details of automation tools used in the process.	p8–9 (Risk of Bias Assessment)
Effect measures	12	Specify for each outcome the effect measure(s) (e.g. risk ratio, mean difference) used in the synthesis or presentation of results.	p9 (Data Synthesis)
Synthesis methods	13a	Describe the processes used to decide which studies were eligible for each synthesis (e.g. tabulating the study intervention characteristics and comparing against the planned groups for each synthesis (item #5)).	p9 (Data Synthesis, ¶1)
	13b	Describe any methods required to prepare the data for presentation or synthesis, such as handling of missing summary statistics, or data conversions.	p9 (Data Synthesis, ¶1)
	13c	Describe any methods used to tabulate or visually display results of individual studies and syntheses.	p9 (Data Synthesis, ¶1); Figures 2–3
	13d	Describe any methods used to synthesize results and provide a rationale for the choice(s). If meta-analysis was performed, describe the model(s), method(s) to identify the presence and extent of statistical heterogeneity, and software package(s) used.	p9–10 (Data Synthesis, ¶1)
	13e	Describe any methods used to explore possible causes of heterogeneity among study results (e.g. subgroup analysis, meta-regression).	p9–10 (Data Synthesis, ¶1; sensitivity analysis described)
	13f	Describe any sensitivity analyses conducted to assess robustness of the synthesized results.	p9–10 (Data Synthesis, ¶1)
Reporting bias assessment	14	Describe any methods used to assess risk of bias due to missing results in a synthesis (arising from reporting biases).	p10 (Data Synthesis, ¶1; Egger’s, Begg, trim-and-fill)
Certainty assessment	15	Describe any methods used to assess certainty (or confidence) in the body of evidence for an outcome.	p9 (Risk of Bias Assessment; GRADE mentioned)
RESULTS
Study selection	16a	Describe the results of the search and selection process, from the number of records identified in the search to the number of studies included in the review, ideally using a flow diagram.	p10–11 (Study Selection); Figure 1 (p25)
	16b	Cite studies that might appear to meet the inclusion criteria, but which were excluded, and explain why they were excluded.	p10–11 (Study Selection, ¶1)
Study characteristics	17	Cite each included study and present its characteristics.	p11 (Study Characteristics); Table 1 (p29–30)
Risk of bias in studies	18	Present assessments of risk of bias for each included study.	p11–12 (Risk of Bias Assessment); Table 2 (p30–32)
Results of individual studies	19	For all outcomes, present, for each study: (a) summary statistics for each group (where appropriate) and (b) an effect estimate and its precision (e.g. confidence/credible interval), ideally using structured tables or plots.	p12–16 (individual trial results); Table 3 (p32–33)
Results of syntheses	20a	For each synthesis, briefly summarise the characteristics and risk of bias among contributing studies.	p16 (Meta-analysis of UC Outcomes, ¶1)
	20b	Present results of all statistical syntheses conducted. If meta-analysis was done, present for each the summary estimate and its precision (e.g. confidence/credible interval) and measures of statistical heterogeneity. If comparing groups, describe the direction of the effect.	p16–17 (Meta-analysis of UC Outcomes); Figure 2 (p26–27)
	20c	Present results of all investigations of possible causes of heterogeneity among study results.	p16–17 (sensitivity analysis with Cohort 1 substitution)
	20d	Present results of all sensitivity analyses conducted to assess the robustness of the synthesized results.	p19 (Supplementary sensitivity analyses); Supplementary Tables S1–S3
Reporting biases	21	Present assessments of risk of bias due to missing results (arising from reporting biases) for each synthesis assessed.	p18 (Publication Bias); Figure 3 (p27)
Certainty of evidence	22	Present assessments of certainty (or confidence) in the body of evidence for each outcome assessed.	p19 (GRADE assessment); Supplementary Table S6
DISCUSSION
Discussion	23a	Provide a general interpretation of the results in the context of other evidence.	p20–22 (Principal Findings; Comparison with Previous Literature)
	23b	Discuss any limitations of the evidence included in the review.	p23–24 (Strengths and limitations)
	23c	Discuss any limitations of the review processes used.	p23–24 (Strengths and limitations)
	23d	Discuss implications of the results for practice, policy, and future research.	p22–23 (Clinical implications and future perspectives)
OTHER INFORMATION
Registration and protocol	24a	Provide registration information for the review, including register name and registration number, or state that the review was not registered.	p3 (PROSPERO: CRD420261303403)
	24b	Indicate where the review protocol can be accessed, or state that a protocol was not prepared.	p7 (Methods, ¶1); PROSPERO
	24c	Describe and explain any amendments to information provided at registration or in the protocol.	N/A (no amendments)
Support	25	Describe sources of financial or non-financial support for the review, and the role of the funders or sponsors in the review.	p2 (Funding)
Competing interests	26	Declare any competing interests of review authors.	p1 (Conflict of interest)
Availability of data, code and other materials	27	Report which of the following are publicly available and where they can be found: template data collection forms; data extracted from included studies; data used for all analyses; analytic code; any other materials used in the review.	p3 (Data availability statement)

From: Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71. doi: 10.1136/bmj.n71. This work is licensed under CC BY 4.0. To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/

Supplementary file 2: Search Strategies

The following search strategies were used to identify relevant clinical trials of TL1A inhibitors in inflammatory bowel disease. Searches were conducted through January 2026 with no date restrictions. The search combined two key concepts: (1) TL1A inhibitors and (2) inflammatory bowel disease.

Table.

Summary of Database Searches

Database	Platform/Interface	Search Date
PubMed	NLM	January 2026
EMBASE	OVID	January 2026
Cochrane CENTRAL	Cochrane Library	January 2026

1. PubMed

Search query: #1 AND #2: 223

#1 TL1A inhibitor terms:

(“Tumor Necrosis Factor Ligand Superfamily Member 15”[MeSH] OR “TNFSF15 protein, human”[Supplementary Concept] OR TNFSF15[tiab] OR TL1A[tiab] OR “TNF-like ligand 1A”[tiab] OR “TNF-like cytokine 1A”[tiab] OR tulisokibart[tiab] OR “PRA023”[tiab] OR “PRA-023”[tiab] OR “MK-7240”[tiab] OR “MK7240”[tiab] OR afimkibart[tiab] OR “PF-06480605”[tiab] OR “PF06480605”[tiab] OR “RVT-3101”[tiab] OR “RVT3101”[tiab] OR “RG6631”[tiab] OR “RG-6631”[tiab] OR duvakitug[tiab] OR “TEV-48574”[tiab] OR “TEV48574”[tiab] OR “SAR447189”[tiab] OR “SAR-447189”[tiab] OR “anti-TL1A”[tiab]): 739

#2 IBD terms:

(“Inflammatory Bowel Diseases”[MeSH] OR “Colitis, Ulcerative”[MeSH] OR “Crohn Disease”[MeSH] OR “inflammatory bowel disease”[tiab] OR “inflammatory bowel diseases”[tiab] OR “ulcerative colitis”[tiab] OR “Crohn’s disease”[tiab] OR “Crohn disease”[tiab] OR IBD[tiab]): 163116

#1 AND #2: 223

2. EMBASE (OVID)

Search query: 535

1. exp inflammatory bowel disease/ : 246484

2. exp ulcerative colitis/ : 115614

3. exp Crohn disease/ : 133556

4. (inflammatory bowel disease* or IBD or ulcerative colitis or Crohn*).tw. : 250177

5. 1 or 2 or 3 or 4 : 298771

6. (TNFSF15 or TL1A or “TNF-like ligand 1A” or “TNF-like cytokine 1A” or “anti-TL1A”).tw. : 1168

7. (tulisokibart or PRA023 or “PRA-023” or “MK-7240” or MK7240).tw. : 72

8. (afimkibart or “PF-06480605” or PF06480605 or “RVT-3101” or RVT3101 or RG6631 or “RG-6631”).tw. : 49

9. (duvakitug or “TEV-48574” or TEV48574 or SAR447189 or “SAR-447189”).tw. : 26

10. 6 or 7 or 8 or 9 : 1285

11. 5 and 10 : 535

3. Cochrane CENTRAL

Search query: [N]

#1 [mh “Inflammatory Bowel Diseases”] : 5226

#2 [mh “Colitis, Ulcerative”] : 2428

#3 [mh “Crohn Disease”] : 2414

#4 (“inflammatory bowel disease” OR IBD OR “ulcerative colitis” OR Crohn):ti,ab,kw : 14939

#5 #1 OR #2 OR #3 OR #4 : 15135

#6 (TNFSF15 OR TL1A OR “TNF-like ligand 1A” OR “TNF-like cytokine 1A” OR “anti-TL1A”):ti,ab,kw : 74

#7 (tulisokibart OR PRA023 OR “PRA-023” OR “MK-7240” OR MK7240):ti,ab,kw : 44

#8 (afimkibart OR “PF-06480605” OR PF06480605 OR “RVT-3101” OR RVT3101 OR RG6631 OR “RG-6631”):ti,ab,kw : 12

#9 (duvakitug OR “TEV-48574” OR TEV48574 OR SAR447189 OR “SAR-447189”):ti,ab,kw : 12

#10 #6 OR #7 OR #8 OR #9 : 105

#11 #5 AND #10 : 86

4. Additional Sources

ClinicalTrials.gov: Condition: inflammatory bowel disease OR ulcerative colitis OR Crohn’s disease; Intervention: tulisokibart OR PRA023 OR MK-7240 OR afimkibart OR PF-06480605 OR RVT-3101 OR RG6631 OR duvakitug OR TEV-48574 OR SAR447189; Phase: Phase 2; Age: Adult, Older adult ->20

Conference abstracts: Digestive Disease Week (DDW), European Crohn’s and Colitis Organisation (ECCO) Congress, American College of Gastroenterology (ACG) Annual Meeting, United European Gastroenterology Week (UEGW) (2020–2025) were searched for TL1A inhibitor clinical trial data. ->5

Manufacturer press releases: Press releases and pipeline updates from Merck & Co., Roche/Genentech, Teva Pharmaceutical Industries, and Sanofi were reviewed.->4

Reference list searching: Reference lists of all included studies and relevant systematic reviews were manually searched for additional eligible studies.

Forward citation tracking: Google Scholar was used to identify studies citing the included articles.

Supplementary file 3

Supplementary Tables

Supplementary Table S1. Leave-One-Out Sensitivity Analysis

Each study was sequentially omitted from the pooled analysis, and the meta-analysis was re-run on the remaining two studies using DerSimonian–Laird random-effects models. The primary analysis used ARTEMIS-UC CDx-positive combined data, TUSCANY-2 450 mg mMS clinical remission (secondary endpoint), and RELIEVE UCCD 900 mg.

Panel A. Risk Ratios

Analysis	Study Omitted	Pooled RR	95% CI	I² (%)	P (het)
Clinical Remission	None (full)	2.56	1.60–4.09	0.0	0.92
	ARTEMIS-UC	2.47	1.46–4.18	0.0	0.78
	TUSCANY-2	2.49	1.43–4.34	0.0	0.72
	RELIEVE UCCD	2.81	1.44–5.50	0.0	0.92
Endoscopic Improvement	None (full)	2.13	1.44–3.16	0.0	0.97
	ARTEMIS-UC	2.20	1.40–3.47	0.0	0.99
	TUSCANY-2	2.10	1.29–3.41	0.0	0.81
	RELIEVE UCCD	2.09	1.25–3.48	0.0	0.82

RR, risk ratio; CI, confidence interval; I², inconsistency index; P (het), P-value for Cochran Q heterogeneity test. Random-effects (DerSimonian–Laird) model was used. For ARTEMIS-UC, the prespecified CDx-positive combined analysis population was used; for TUSCANY-2, the 450 mg arm with modified Mayo Score–based clinical remission; for RELIEVE UCCD, the 900 mg arm.

Panel B. Risk Differences

Analysis	Study Omitted	Pooled RD (%)	95% CI	I² (%)	NNT
Clinical Remission	None (full)	+22.2	12.8–31.5	0.0	4.5
	ARTEMIS-UC	+22.7	11.7–33.7	0.0	4.4
	TUSCANY-2	+23.9	11.0–36.8	0.0	4.2
	RELIEVE UCCD	+20.4	9.6–31.2	0.0	4.9
Endoscopic Improvement	None (full)	+22.6	12.3–32.9	0.0	4.4
	ARTEMIS-UC	+24.3	12.3–36.3	0.0	4.1
	TUSCANY-2	+22.8	9.1–36.5	0.0	4.4
	RELIEVE UCCD	+20.6	8.4–32.9	0.0	4.8

RD, risk difference; CI, confidence interval; I², inconsistency index; NNT, number needed to treat (1/RD). Random-effects (DerSimonian–Laird) model was used. Study populations as in Panel A.

The pooled estimate remained stable regardless of which study was omitted. For clinical remission, RR ranged from 2.47 to 2.81 and RD from +20.4% to +23.9%. For endoscopic improvement, RR ranged from 2.09 to 2.20 and RD from +20.6% to +24.3%. I² remained 0.0% in all leave-one-out analyses, indicating that no single study drives the pooled estimate; with only three trials, however, this consistency is underpowered to exclude between-agent heterogeneity.

Table S2.

Safety outcomes of Phase 2 clinical trials of TL1A inhibitors in IBD

Study	Arms	Any AE	SAE	Infections	AE → D/C	Deaths
ARTEMIS-UC10	Tulisokibart (N=90)	41 (46%)	1 (1%)	16 (18%)	1 (1%)	0
	Placebo (N=88)	38 (43%)	7 (8%)	16 (18%)	3 (3%)	0
APOLLO-CD25	Tulisokibart (N=55)	43 (78%)	8 (15%)*	25 (46%)†	NR	0
TUSCANY-111	Afimkibart (N=50)	33 (66%)	~3 (~6%)‡	NR	NR	0
TUSCANY-213	Afimkibart (all doses; N=200)	~48%	6 events	NR	6 pts	0
	Placebo (N=45)	~48%	4 events	NR	NR	0
RELIEVE UC14,15	Duvakitug (pooled; N=93)	~50%	1 (1%)‖	NR	NR	0
	Placebo (N=44)	~50%	1 (2%)‖	NR	NR	0
RELIEVE CD26,28	Duvakitug 450 mg (N=46)	31 (67%)	~6 (~13%)§	NR	4 (9%)	0
	Duvakitug 900 mg (N=46)	20 (43%)	~1 (~2%)§	NR	1 (2%)	0
	Placebo (N=46)	22 (48%)	~5 (~11%)§	NR	1 (2%)	0

*No SAEs deemed drug-related. †All infections mild to moderate. ‡Estimated from published description. §Back-calculated from percentages reported in ECCO 2025 conference report²⁸; absolute event counts not directly reported. AE, adverse event; SAE, serious adverse event; D/C, discontinuation; NR, not reported from primary source. ‖Per-arm SAE counts for the RELIEVE UCCD UC cohort (duvakitug 450 mg 0/47, 900 mg 1/46, placebo 1/44) became available with the full publication (Reinisch et al., Lancet Gastroenterol Hepatol 2026; NCT05499130).

Supplementary Table S3.

Prediction Intervals (IntHout et al. Method)

Prediction intervals estimate the range of true treatment effects expected in a future study, calculated as pooled estimate ± t(k−2, 0.975) × √(SE² + τ²) per IntHout et al. (IntHout J, Ioannidis JPA, Rovers MM, Goeman JJ. Plea for routinely presenting prediction intervals in meta-analysis. BMJ Open 2016;6:e010247).

Outcome	Pooled RR	95% CI	95% Prediction Interval	I² (%)	τ²
Clinical Remission	2.56	1.60–4.09	0.11–58.70	0.0	0.000
Endoscopic Improvement	2.13	1.44–3.16	0.16–28.39	0.0	0.000
HEMI	3.80	1.64–8.84	0.00–15018	42.5	0.240

RR, risk ratio; CI, confidence interval; I², inconsistency index; τ², between-study variance. Prediction intervals were calculated using the IntHout method: pooled estimate ± t(k−2, 0.975) × √(SE² + τ²), where k = number of studies. With k = 3 studies (df = 1), the t-distribution multiplier is 12.71, resulting in extremely wide intervals regardless of I² value.

Despite I²=0.0% for clinical remission and endoscopic improvement, the prediction intervals are extremely wide (e.g., RR 0.11–58.70 for clinical remission). This arises because with k=3 studies, the t-distribution has df=1 and t(1, 0.975)=12.71, inflating the interval far beyond the 95% CI. This demonstrates that even with zero observed heterogeneity, three studies provide limited certainty about the expected effect in a future (fourth) trial. The HEMI prediction interval is additionally widened by moderate heterogeneity (τ²=0.240). These findings emphasize that Phase 3 trials are essential for narrowing prediction intervals and determining whether a class-level effect exists.

Supplementary Table S4.

Fixed-Effect Versus Random-Effects Model Comparison

Both fixed-effect (inverse-variance weighted) and DerSimonian–Laird random-effects models are presented. When τ²=0, the models yield identical results. For HEMI (I²=42.5%), the RE model gives a slightly larger point estimate (3.80 vs 3.36) but a wider CI.

Outcome	Model	Estimate	95% CI	P	I² (%)	τ²
Clinical Remission (RR)	RE	2.56	1.60–4.09	<0.001	0.0	0.000
	FE	2.56	1.60–4.09	<0.001	—	—
	HKSJ	2.59	1.85–3.61	0.007	—	—
Endoscopic Improvement (RR)	RE	2.13	1.44–3.16	<0.001	0.0	0.000
	FE	2.13	1.44–3.16	<0.001	—	—
	HKSJ	2.10	1.81–2.44	0.002	—	—
Clinical Remission (RD)	RE	+22.2%	12.8–31.5%	<0.001	0.0	0.000
	FE	+22.2%	12.8–31.5%	<0.001	—	—
	HKSJ	+23.0%	13.9–32.2%	0.008	—	—
Endoscopic Improvement (RD)	RE	+22.6%	12.3–32.9%	<0.001	0.0	0.000
	FE	+22.6%	12.3–32.9%	<0.001	—	—
	HKSJ	+22.5%	10.6–34.5%	0.015	—	—
HEMI (RR)	RE	3.80	1.64–8.84	0.002	42.5	0.240
	FE	3.36	1.85–6.13	<0.001	—	—

RE, random-effects (DerSimonian–Laird); FE, fixed-effect (inverse-variance weighted); RR, risk ratio; RD, risk difference; CI, confidence interval; I², inconsistency index; τ², between-study variance. When τ² = 0, RE and FE models yield identical results.

For clinical remission and endoscopic improvement, RE and FE models are numerically identical because τ²=0, eliminating model choice as a source of analytic uncertainty. For HEMI, both models remain statistically significant (P≤0.002), suggesting heterogeneity does not materially alter the conclusion.

Supplementary Table S5.

Safety Meta-Analysis

Quantitative pooling of safety outcomes was restricted to studies with complete arm-level data. TUSCANY-2 reported aggregate event counts without arm-specific data; Per-arm SAE counts for the RELIEVE UCCD UC cohort became available with its full publication (Reinisch et al., Lancet Gastroenterol Hepatol 2026; NCT05499130) and are used for the UC-only SAE analysis; ARTEMIS-UC (Sands, NEJM 2024) and the RELIEVE UCCD CD cohort (Medicom/HCPLive reports of ECCO 2025 OP40 and ACG Oral 66) provided arm-level data for the remaining safety outcomes. Data sources for each cell are explicitly documented below.

Safety Outcome	Studies (k)	Pooled RR	95% CI	P	I² (%)	Data Sources
Any AE	2	1.00	0.77–1.31	0.993	0.0	ARTEMIS-UC (NEJM), RELIEVE CD 900mg (HCPLive/Sanofi)
SAE	2	0.24	0.05–1.13	0.072	0.0	ARTEMIS-UC (NEJM: 1/90 vs 7/88), RELIEVE UC 900mg (NCT05499130: 1/46 vs 1/44)
AE → D/C	2	0.51	0.09–2.90	0.449	0.0	ARTEMIS-UC (NEJM: 1/90 vs 3/88), RELIEVE CD 900mg (HCPLive: 1/46 vs 1/46)
Infections	1	0.98	0.53–1.82	0.947	—	ARTEMIS-UC only (NEJM: 16/90 vs 16/88); other trials NR

RR, risk ratio; CI, confidence interval; I², inconsistency index; AE, adverse event; SAE, serious adverse event; D/C, discontinuation; NR, not reported. Pooling was restricted to studies with complete arm-level event data. TUSCANY-2 reported aggregate safety data without arm-specific counts; TUSCANY-2 was therefore excluded from quantitative safety pooling; per-arm SAE counts for the RELIEVE UCCD UC cohort are available from its full publication and are used for the UC-only SAE analysis.

Any Adverse Event: Pooled RR 1.00 (95% CI 0.77–1.31; P=0.993), indicating no difference between TL1A inhibitors and placebo. Pooled RD +0.1% (95% CI −11.8 to +11.9%).

Serious Adverse Events: Pooled RR 0.24 (95% CI 0.05–1.13; P=0.072), not statistically significant (Peto OR 0.28, 95% CI 0.08–0.99). Data: ARTEMIS-UC 1/90 (1%) vs 7/88 (8%) from Sands NEJM 2024; RELIEVE UC 900mg 1/46 (2%) vs placebo 1/44 (2%) from the RELIEVE UCCD registry (NCT05499130) and Reinisch et al., Lancet Gastroenterol Hepatol 2026. However, interpretation requires extreme caution: (1) only two studies, (2) very low total SAE counts (2 active vs 8 placebo), (3) ARTEMIS-UC placebo SAE rate of 8% was atypically high, and (4) lower SAE with active treatment likely reflects effective disease control rather than a direct protective drug effect. Pooled RD −3.5% (95% CI −10.1 to +3.1%).

AE Leading to Discontinuation: Pooled RR 0.51 (95% CI 0.09–2.90; P=0.449), non-significant. ARTEMIS-UC 1/90 vs 3/88; RELIEVE CD 900mg 1/46 vs 1/46.

Infections: Only ARTEMIS-UC reported arm-level infection data: 16/90 (17.8%) vs 16/88 (18.2%), RR 0.98. Pooling not feasible due to non-reporting in other trials. No serious or opportunistic infections were attributed to TL1A inhibition in any study.

Supplementary Table S6.

Histologic-Endoscopic Mucosal Improvement (HEMI) Pooling

HEMI was reported in all three RCTs. For ARTEMIS-UC, Cohort 1 all-comers data were used because CDx-positive HEMI was not separately reported (Sands NEJM 2024). TUSCANY-2 450 mg HEMI and RELIEVE UCCD 900 mg HEMI were included from the manuscript Table 3 (sourced from Danese LGH 2025 and Sanofi press release, respectively).

Study	Active	Placebo	RR (95% CI)	RD (pp)	Wt (RR)	Wt (RD)	NNT
ARTEMIS-UC (C1)	21/68	3/67	6.90 (2.16–22.04)	+26.6	31.3%	33.6%	3.8
TUSCANY-2 (450mg)	8/31	2/45	5.81 (1.32–25.52)	+21.4	22.9%	23.3%	4.7
RELIEVE UCCD (900mg)	15/46	7/44	2.05 (0.92–4.54)	+16.7	45.8%	43.1%	6.0
Pooled (RE)	44/145	12/156	3.80 (1.64–8.84)	+22.7	100%	100%	4.4

HEMI, histologic-endoscopic mucosal improvement; RR, risk ratio; CI, confidence interval; RD, risk difference; pp, percentage points; Wt, study weight; NNT, number needed to treat; RE, random-effects (DerSimonian–Laird); C1, Cohort 1. For ARTEMIS-UC, Cohort 1 all-comers data were used because CDx-positive HEMI was not separately reported. For TUSCANY-2, the 450 mg arm was used. For RELIEVE UCCD, the 900 mg arm was used.

Results: Pooled RR for HEMI was 3.80 (95% CI 1.64–8.84; P=0.002), with moderate heterogeneity (I²=42.5%, Q=3.48, P=0.18, τ²=0.240). Fixed-effect RR was 3.36 (95% CI 1.85–6.13). 95% prediction interval: 0.00–15018 (extremely wide due to df=1 and non-zero τ²). Pooled RD was +22.7% (95% CI 14.3–31.2%; I²=0.0%), yielding NNT=4.4.

The heterogeneity in HEMI RR was driven by placebo rate disparity: ARTEMIS-UC C1 4% (3/67), TUSCANY-2 4% (2/45), and RELIEVE UCCD 16% (7/44). When measured as RD, I² dropped to 0.0%, confirming that the absolute HEMI benefit is consistent across TL1A inhibitors at approximately +17–27 percentage points.

Supplementary Table S7.

GRADE Evidence Profile

Certainty of evidence was assessed using the GRADE framework (Guyatt GH, Oxman AD, Vist GE, et al. GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. BMJ 2008;336:924–6). Evidence from RCTs starts at HIGH and is downgraded for risk of bias, inconsistency, indirectness, imprecision, and publication bias.

Outcome	Risk of Bias	Inconsistency	Indirectness	Imprecision	Pub. Bias	Overall	Certainty
UC Clinical Remission (k=3)	Not serious	Not serious (I²=0%)	Serious¹	Not serious	Undetected	−1	⊕⊕⊕⊖ MODERATE
UC Endoscopic Improvement (k=3)	Not serious	Not serious (I²=0%)	Serious¹	Not serious	Undetected	−1	⊕⊕⊕⊖ MODERATE
UC HEMI (k=3)	Not serious	Not serious²	Serious¹	Not serious	Undetected	−1	⊕⊕⊕⊖ MODERATE
CD Endoscopic Response (k=1)	Not serious	N/A	Serious³	Serious⁴	Not assessed	−2	⊕⊕⊖⊖ LOW
Any AE (k=2)	Not serious	Not serious	Not serious	Serious⁵	Not assessed	−1	⊕⊕⊕⊖ MODERATE
SAE (k=2)	Not serious	Not serious	Serious⁶	Serious⁷	Not assessed	−2	⊕⊕⊖⊖ LOW

GRADE, Grading of Recommendations, Assessment, Development and Evaluations; k, number of studies; I², inconsistency index; RR, risk ratio; RD, risk difference; AE, adverse event; SAE, serious adverse event. Certainty ratings: HIGH (⊕⊕⊕⊕), MODERATE (⊕⊕⊕⊖), LOW (⊕⊕⊖⊖), VERY LOW (⊕⊖⊖⊖). Superscript numbers refer to footnotes below the table.

Footnotes:

¹ Downgraded for indirectness: TUSCANY-2 mMS clinical remission was a secondary endpoint (primary tMS endpoint failed); ARTEMIS-UC used CDx-positive subgroup rather than full cohort; all studies are Phase 2 with smaller populations that may not generalize to Phase 3.

² I²=42.5% for HEMI RR, but 0.0% for HEMI RD; heterogeneity driven by placebo rate variation rather than treatment effect inconsistency. Not downgraded per GRADE guidance when direction of effect is consistent across studies.

³ Single Phase 2b RCT; 14-week induction only; conference abstract data not yet peer-reviewed.

⁴ Single study, N=92; wide relative CI (RR 1.64–8.20).

⁵ Only 2 studies pooled (total N=270); adequate events but limited generalizability.

⁶ SAE pooling limited to 2 studies; RELIEVE UC and TUSCANY-2 SAE data not separately reported by arm.

⁷ Very low event counts (2 vs 12 total SAEs); RR estimate unstable despite statistical significance.

Supplementary Note: Interpretation of Supplementary Analyses

1. Robustness: Leave-one-out analysis confirms no single study drives the pooled estimate. RR for clinical remission remained 2.47–2.81 and for endoscopic improvement 2.09–2.20, with I²=0.0% in all scenarios.

2. Model invariance: Because τ²=0 for primary outcomes, fixed-effect and random-effects models produce identical results, eliminating model choice as a source of analytic uncertainty.

3. Wide prediction intervals: Despite zero heterogeneity, the 95% PI for clinical remission (RR 0.11–58.7) is extremely wide due to df=1, indicating that three studies provide limited certainty about the expected effect in a future trial. This is the key statistical limitation of the current evidence base.

4. HEMI as a composite: HEMI pooling shows moderate heterogeneity in RR (I²=42.5%) but zero heterogeneity in RD (I²=0.0%), with a consistent absolute benefit of +22.7 percentage points (NNT 4.4). The RR heterogeneity is attributable to placebo rate variation, not differential drug efficacy.

5. Safety reassurance: Quantitative pooling shows no excess of adverse events (RR 1.00), numerically fewer SAEs in a UC-only pool (RR 0.24, 95% CI 0.05–1.13; P=0.072), which did not reach statistical significance and rests on very low event counts, and balanced infection rates. These findings are consistent with the favorable safety narrative across all six Phase 2 studies but are limited by incomplete arm-level reporting in TUSCANY-2.

6. Evidence certainty: GRADE rates UC efficacy outcomes as MODERATE (downgraded for indirectness) and CD efficacy as LOW (downgraded for indirectness and imprecision). Phase 3 trials are expected to upgrade the certainty of evidence.

Data verification note: All numerical values in this supplementary material are derived from published sources as cited. RELIEVE UCCD CD placebo SAE rate (11%, 5/46) was sourced from the Medicom Medical Publishers report of the ECCO 2025 OP40 presentation by Jairath et al. Infection rates were poolable only from ARTEMIS-UC; other trials did not report arm-level infection data.

METHOD

Search Strategy and Information Sources

This SR was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 checklist.¹⁶ A comprehensive literature search was performed across PubMed/MEDLINE, Embase-OVID, the Cochrane Central Register of Controlled Trials (CENTRAL), and ClinicalTrials.gov from database inception through January 31, 2026. The search strategy combined terms for the intervention (“TL1A,” “TNFSF15,” “anti-TL1A,” “tulisokibart,” “PRA023,” “MK-7240,” “afimkibart,” “PF-06480605,” “RVT-3101,” “RG6631,” “duvakitug,” “TEV-48574,” “SAR447189”) with terms for the condition (“inflammatory bowel disease,” “ulcerative colitis,” “Crohn’s disease”) and study design (“clinical trial,” “Phase 2,” “randomised”) (Supplementary File 1). Reference lists of included studies and relevant reviews were hand-searched. Conference abstracts from Digestive Disease Week (DDW), European Crohn’s and Colitis Organisation (ECCO) Congress, and the American College of Gastroenterology (ACG) annual meeting (2020–2025) were searched for unpublished trial data. Manufacturer press releases from Merck, Roche/Genentech, Teva, and Sanofi were reviewed for supplementary efficacy and safety information.

Eligibility Criteria

Studies were included if they: (1) enrolled adults (≥18 years) with moderately to severely active UC or CD; (2) evaluated a monoclonal antibody targeting TL1A; (3) were Phase 2 (including Phase 2a or 2b) clinical trials; and (4) reported at least one efficacy or safety outcome. Both RCTs and single-arm studies were eligible. Case reports, preclinical studies, Phase 1 pharmacokinetic studies, narrative reviews, and editorials were excluded. Studies reporting exclusively long-term extension data without primary induction-period results were excluded.

Data Extraction and Outcomes

Data were extracted on study design, patient population, dosing regimen, primary and key secondary endpoints, efficacy outcomes, safety events, and biomarker findings. The primary outcomes of interest were clinical remission (defined by modified Mayo Score [mMS] for UC and CDAI <150 for CD), endoscopic improvement (Mayo Endoscopic Subscore [MES] 0 or 1 for UC) or endoscopic response (≥50% reduction in Simple Endoscopic Score for Crohn’s Disease [SES-CD] for CD), and histologic-endoscopic mucosal improvement (HEMI). Safety outcomes included rates of any adverse event (AE), serious AE (SAE), infections, AEs leading to treatment discontinuation, and deaths.

Risk of Bias Assessment

Risk of bias was assessed using the Cochrane Risk of Bias 2.0 (RoB 2.0) tool for RCTs (ARTEMIS-UC, TUSCANY-2, RELIEVE UCCD).¹⁷ The Risk of Bias in Non-Randomised Studies of Interventions (ROBINS-I) tool was applied to APOLLO-CD and TUSCANY-1 as single-arm studies employing historical placebo comparators.¹⁸ Each RCT was evaluated across five domains: randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result. Non-RCTs were evaluated across seven ROBINS-I domains including confounding, participant selection, classification of interventions, deviations, missing data, outcome measurement, and selection of reported results.

Data Synthesis

Random-effects meta-analysis using the DerSimonian–Laird method was performed for outcomes reported by two or more randomized, placebo-controlled trials with comparable endpoint definitions.¹⁹ Risk ratios (RR) with 95% confidence intervals (CI) were calculated from dichotomous event data. Risk differences (RD) were pooled using the same DerSimonian–Laird random-effects framework, and number needed to treat (NNT) was derived as 1/RD. Dose-arm and analysis-population selection decisions were prespecified before data extraction to minimize post hoc bias, guided by the following a priori rationale. For ARTEMIS-UC, the prespecified CDx-positive combined analysis (Cohorts 1+2 pooled, N=75) was used as the primary analysis population for meta-analysis, with Cohort 1 all-comers (N=135) evaluated in sensitivity analysis. The CDx-positive population was selected because the trialists designated this as a key prespecified analysis and because its placebo remission rate (11%) was more representative of contemporary UC trials than the atypically low 1% observed in the Cohort 1 all-comers population. For TUSCANY-2, the 450 mg arm was selected as the expected Phase 3 dose based on the plateau in dose–response observed across efficacy endpoints in the original publication; mMS–based clinical remission (secondary endpoint) was used to harmonize with the mMS-based primary endpoints of ARTEMIS-UC and RELIEVE UCCD, as the mMS aligns with the current FDA draft guidance for UC drug development. For RELIEVE UCCD, the 900 mg arm was selected based on the prespecified Bayesian dose–response analysis reported in the original publication, which identified 900 mg as the maximally efficacious dose. We acknowledge that selecting optimally performing dose arms and, in the case of ARTEMIS-UC, a biomarker-enriched subpopulation, may bias pooled estimates toward larger treatment effects relative to what would be observed in unselected populations at non-optimized doses; this is addressed as a limitation. Between-study heterogeneity was assessed using Cochran’s Q test and the I² statistic, with I² values of 25%, 50%, and 75% representing low, moderate, and high heterogeneity, respectively.²⁰ Publication bias was assessed by funnel plot inspection, Egger’s weighted regression test,²¹ Begg-Mazumdar rank correlation test,²² and the Duval-Tweedie trim-and-fill method (R0 estimator).²³ All formal tests have severely limited statistical power with k=3 studies,²⁴ and results are reported for completeness rather than definitive inference. Sensitivity analyses were conducted by (1) substituting ARTEMIS-UC Cohort 1 data for CDx-positive data, and (2) excluding TUSCANY-2 given its use of a secondary rather than primary endpoint. Single-arm studies (APOLLO-CD, TUSCANY-1) were excluded from quantitative synthesis due to absence of concurrent placebo comparators. For CD, only one RCT (RELIEVE UCCD CD cohort) was available; results are reported as individual study estimates without pooling.

Dose-response was assessed descriptively within the RELIEVE UCCD trial, which randomized patients to two active dose levels (450 mg and 900 mg) and placebo, by comparing RDs across doses for UC clinical remission, UC endoscopic improvement, and CD endoscopic response.

All analyses were performed using Python 3.12 (NumPy, SciPy) and verified with R 4.3 (meta package, metabin function with Mantel–Haenszel method).

Informed consent was not applicable due to only database analysis.

This study did not use AI and/or LLM tools during the preparation of this manuscript

Results

Dose-response analysis

Within the RELIEVE UCCD trial, which randomized patients to two active dose levels, a clear dose-response relationship was observed. In UC, clinical remission rates were 36% (450 mg) versus 48% (900 mg) versus 20% (placebo), with RDs of +15.6 and +27.3 pp, respectively. In CD, endoscopic response rates were 26% (450 mg, 12/46) versus 48% (900 mg, 22/46) versus 13% (placebo, 6/46), yielding RR 2.00 and 3.67, respectively.^14,15,26,28 The 900 mg dose approximately doubled the treatment effect of the 450 mg dose across both diseases and endpoints.

Publication Bias

Funnel plots for both UC outcomes showed no visual asymmetry (Figure 3). For clinical remission, Egger’s regression test yielded an intercept of 1.20 (P=0.08, df=1) and Begg-Mazumdar rank correlation was τ=1.00 (P=0.33). Duval-Tweedie trim and fill estimated two missing studies on the left; the adjusted pooled RR was 2.48 (95% CI 1.69–3.64), minimally attenuated from the unadjusted estimate of 2.56. For endoscopic improvement, Egger’s test was non-significant (intercept −1.44, P=0.24), Begg-Mazumdar rank correlation was τ=−1.00 (P=0.33), and trim and fill yielded an adjusted RR of 2.11 (95% CI 1.56–2.84), virtually unchanged from 2.13. All formal tests are severely underpowered with only three studies,²⁴ and these results should be interpreted as exploratory rather than definitive.

CD outcomes (single RCT)

Only one RCT (RELIEVE UCCD CD cohort) was available for quantitative analysis of CD outcomes. At the 900 mg dose, duvakitug achieved an endoscopic response RR of 3.67 (95% CI 1.64–8.20; P=0.002) compared with placebo, with a RDs of 34.8% (95% CI 17.4–52.2%) and NNT of 2.9.^26,28 APOLLO-CD, as an open-label single-arm study with historical comparator and serious risk of bias (Table 2), was not included in the quantitative synthesis but reported a 26% endoscopic response rate (vs 12% historical placebo, P=0.002).²⁵

Safety profile across the TL1A inhibitor class

Across all six Phase 2 studies, TL1A inhibitors demonstrated no unexpected safety signals during the induction phase (Table S2). All safety data reported here reflect induction-period exposure only (12–14 weeks); long-term maintenance safety data were not available for any agent at the time of this analysis. No deaths occurred in any trial. SAE rates were low and generally comparable to or lower than placebo in controlled studies. In ARTEMIS-UC, the SAE rate was lower with tulisokibart (1%) than placebo (8%).¹⁰ Infection rates were balanced between active treatment and placebo across all RCTs, with no serious or opportunistic infections attributed to TL1A inhibition.^10,13,14,26 The single notable safety signal was the high anti-drug antibody rate (82%) with IV afimkibart in TUSCANY-1, which was resolved by switching to subcutaneous administration in TUSCANY-2 (0% neutralizing antibodies at Week 56).^11,13

Supplementary sensitivity and certainty analyses

Leave-one-out sensitivity analysis confirmed the stability of UC pooled estimates: sequential exclusion of each study yielded clinical remission RRs ranging from 2.47 to 2.81, all remaining statistically significant, with I² remaining 0.0% in all permutations (Table S1). The 95% prediction interval could not be meaningfully calculated with only three studies (df=1), yielding an uninformatively wide range (Table S3); this limitation indicates the need for Phase 3 data to narrow the expected range of the true treatment effect.²⁴ Fixed-effect and random-effects models produced identical results, as expected given zero observed heterogeneity (Table S4). As a methodological sensitivity analysis for small meta-analyses, Hartung–Knapp–Sidik–Jonkman (HKSJ) correction was applied: all pooled estimates remained statistically significant (clinical remission RR 2.59, HKSJ 95% CI 1.85–3.61, P=0.007; endoscopic improvement RR 2.10, HKSJ 95% CI 1.81–2.44, P=0.002), confirming robustness to small-sample CI adjustment (Table S4). An additional sensitivity analysis substituting TUSCANY-2’s primary tMS-based clinical remission (24% vs 12% placebo) for the mMS-based secondary endpoint yielded a pooled RR of 2.38 (95% CI 1.46–3.89; I²=0.0%), remaining statistically significant but with an attenuated RD of +19.5% (95% CI 9.1–29.8%; NNT 5.1), consistent with the known lower sensitivity of tMS relative to mMS.

A UC-only pooled safety meta-analysis of serious adverse events (SAE) was performed, restricted to the two UC RCTs with arm-level SAE data (ARTEMIS-UC and the RELIEVE UCCD UC cohort), comparing tulisokibart (1/90) and duvakitug 900 mg (1/46) with placebo (7/88 and 1/44, respectively). Because SAE were rare, pooling used the Mantel–Haenszel method: the pooled SAE risk ratio was 0.24 (95% CI 0.05–1.13; P=0.072; I²=0.0%), which did not reach statistical significance (Peto odds ratio 0.28, 95% CI 0.08–0.99; pooled risk difference −3.5%, 95% CI −10.1 to +3.1%) (Table S5). The numerically lower SAE rate with TL1A inhibitors is therefore imprecise and should be regarded as exploratory, likely reflecting effective disease control rather than a direct protective drug effect. This UC-only analysis replaces an earlier cross-indication pool that combined ARTEMIS-UC with the RELIEVE UCCD CD cohort (previously reported as RR 0.17). Within the RELIEVE UCCD UC cohort, SAE were infrequent in all arms (duvakitug 450 mg 0/47, 900 mg 1/46, placebo 1/44); the higher SAE rate previously reported for the duvakitug 450 mg arm (6/46, 13%) occurred in the CD cohort, not in UC. When HEMI (histologic–endoscopic mucosal improvement) was pooled across the three UC RCTs, the pooled RR was consistent with primary efficacy outcomes (Table S6).

GRADE assessment rated the certainty of evidence as moderate for UC clinical remission and endoscopic improvement (downgraded one level for indirectness due to use of a secondary endpoint in TUSCANY-2, CDx-positive subgroup analysis in ARTEMIS-UC, and the Phase 2 setting with limited generalizability), low for CD outcomes (single RCT, serious imprecision, and reliance on non-peer-reviewed conference abstracts), and moderate for the overall safety profile (Table S7). For outcomes informed predominantly by non-peer-reviewed data sources (conference abstracts, press releases), certainty was further tempered given the potential for selective reporting and incomplete data availability.

Funding Statement

This research was supported by the Bio and Medical Technology Development Program of the National Research Foundation (NRF) funded by the Korean government (MSIT) (No. RS-2023-00223501) and Hallym University Research Fund.