Is the conformation-selective PI3Kδ inhibitor roginolisib safe and active in advanced cancer?

A safety-first dose study across tumour types

The trial asked whether roginolisib, a PI3Kδ inhibitor, could be given continuously every day at a dose that actually blocks the target, and what that dose does to patients with advanced solid or haematological malignancies. Safety was the primary endpoint; pharmacokinetics, pharmacodynamics and anti-tumour response were secondary, and immune phenotyping, proteomics and genomics were exploratory.

The design had two parts. Part A escalated doses in 24 patients, starting at 10 mg and moving through 20 mg, 40 mg and 80 mg once daily in 28-day cycles, with a final cohort in follicular lymphoma. Part B then treated 20 metastatic uveal melanoma (mUM) patients at the dose chosen from Part A. Forty-four patients were enrolled in total between February 2020 and December 2024, with a median age of 59 years and 32 of 44 (72.7%) at ECOG performance status 0.

Locking PI3Kδ inactive instead of competing with ATP

PI3Kδ is an enzyme that supports tumour growth both directly and by activating immune-suppressive cells. Earlier PI3Kδ inhibitors competed with ATP at the enzyme's catalytic site; the paper states that this ATP-competitive approach could not establish a safe profile at an optimal dose and led to withdrawal for most indications, with a particularly difficult toxicity profile in solid tumours.

Roginolisib works differently: it is a selective, non-ATP-competitive inhibitor that locks PI3Kδ into an inactive state, an allosteric modulation. That mechanistic difference is the reason the safety question matters here rather than being a formality — the trial was testing whether selective conformational inhibition could be tolerated during continuous daily dosing, which the earlier class could not.

Dose selection tied to a functional PD readout

Dose escalation was guided by a non-clinical pharmacokinetic/pharmacodynamic model that was updated with each cohort's PK data and every other cohort's pharmacodynamic data. The 80 mg dose was declared the recommended phase 2 dose because steady-state free-fraction trough concentrations reached the IC50 range of about 235 nM (134 ng/mL) measured in SU-DHL06 cells, Treg proliferation and B cell cytokine release, and coincided with roughly 10-fold target engagement over the IC50 in the Jurkat cell line.

Target engagement in patients was measured functionally, not by PI3Kδ expression or downstream phospho-AKT: the team tracked CD63 expression on basophils, a marker tightly connected to PIK3CD activation. At 80 mg, steady-state trough levels (about 3000 ng/mL) sat above the 730 ng/mL concentration associated with basophil CD63 activation, and the basophil activation test confirmed consistent PI3Kδ inhibition above 90%. The authors note that exceeding the maximum concentration seen at 80 mg once daily would lose delta-isoform selectivity, so higher doses were not pursued.

No dose-limiting toxicity and stable immune cell counts

No dose-limiting toxicity was observed, and no drug-related toxicity required dose modification, including in patients treated longer than 6 months. Grade 3 or higher toxicities occurred in 3 of 44 patients (6.8%). At 80 mg, most treatment-emergent adverse events related to roginolisib were Grade 1 or 2.

The paper highlights the absence of serious infections and severe immune-related adverse events, which were concerns with first-generation PI3Kδ inhibitors, and reports that no anti-infective prophylaxis was needed. Total monthly CD4, CD8, NK and B lymphocyte populations were stable, and plasma immunoglobulin concentrations were unchanged over time — a contrast with the immune dysfunction seen in the earlier class. Four patients who started at lower doses escalated to 80 mg without additional toxicities.

Median survival of 18.4 months in a pre-treated mUM cohort

In the 20 mUM patients in Part B, median overall survival was 20.8 months at the December 2023 data cut-off and 18.4 months at the February 2026 cut-off. These are two cut-offs of the same cohort, not two independent results, and the later figure is the more mature estimate.

The most decision-relevant split in the paper is by radiological status at Week 16 (Cycle 5). Patients with stable disease at that point had a median overall survival of 28.5 months, whereas patients who had progressed had a median of 12.3 months. This is a comparison of outcomes after the fact, not evidence that stable disease causes longer survival; the authors describe it as a trend suggesting that radiological stability at Week 16 was associated with slower tumour growth kinetics and improved overall survival.

The cohort was heavily pre-treated and high-risk: except for 3 patients, all mUM patients had received at least one prior systemic therapy, 17 of 29 (59%) had baseline LDH above the upper limit of normal, and 16 of 29 (55%) had liver lesions larger than 3 cm and up to 8 cm. The paper also states that the single-arm design cannot establish single-agent activity in mUM or compare it with an immunotherapy such as tebentafusp.

Tregs fall while activated CD8+ T cells rise

The pharmacodynamic story is immune modulation rather than direct tumour-cell killing. T regulatory cell abundance decreased during treatment while activated CD8+ T cells increased, together with soluble plasma IL-15. In paired tumour biopsies from 9 patients in Part A, Foxp3+ Treg cells trended down in 6 of 9 (67%), CD8+ T cells were unchanged, and the CD8+/Treg ratio therefore rose.

In blood, immunosuppressive populations including Treg cells, CD5+ B cells, CD8+CD39− cells and CD56+CD16− NK cells were reduced, while activated CD8+CD39+ T cells increased and their increase was positively correlated with improved overall survival. Plasma proteins shifted toward lymphocyte activation, with increases in IL-15, IL-17D and IFNGR2, and soluble PD-1 fell — which the authors interpret as consistent with Treg depletion dominating over PD-1 upregulation on activated effector cells. They describe these findings as consistent with PI3Kδ inhibition resetting immune suppression rather than directly co-stimulating T cells, and state that validation in larger cohorts is required.

GPCR pathways drop in patients with stable disease

Whole transcriptome sequencing of tumour biopsies from 14 mUM patients found 326 significantly modulated genes in the 7 patients with stable disease at Week 16: 325 were downregulated and one, alpha-2-macroglobulin (A2M), was upregulated. Gene Ontology analysis of the downregulated genes pointed to G-protein coupled receptor pathways, particularly olfactory receptors.

The authors connect this to mUM biology: constitutive GPCR activation is mostly found in patients with stable disease at Week 16, and the enriched GPCR set includes driver mutations such as GNAQ and GNA11. They therefore suggest that the decreased GPCR signature seen in on-treatment biopsies may indicate a favourable outcome. This is a hypothesis-generating interpretation from a small biopsy set, not a validated predictive biomarker.

Single-arm evidence and the randomised test ahead

Several exploratory analyses rest on small numbers: paired tumour biopsies came from 9 of 16 eligible Part A patients, the transcriptomic comparison involved 7 stable-disease and 5 progressing patients at baseline, and cfDNA analysis covered 18 of 29 mUM patients (62%). The authors also note that a tumour-tissue-agnostic cfDNA approach may miss genetically distinct metastatic clones, and that the origin of circulating PI3K-mTOR pathway proteins is unclear.

The paper's own next step is the randomised Phase 2 OCCULE-01 study (NCT06717126), which randomised patients who had progressed on prior immunotherapy to roginolisib or investigator's choice regardless of HLA status, has closed recruitment, and is following survival events as the primary endpoint. That design is what the current data cannot substitute for: the 28.5 versus 12.3 month split by Week 16 status is a prognostic grouping within a single-arm cohort, and only a randomised comparison can show whether roginolisib changes survival rather than selecting patients who were going to do better.

Sources used in this answer