Jean J. Zhao

Genomic landscape of CSF and plasma ctDNA by WES (n = 10 samples/5 patients). Oncoprint summarizing genomic alterations detected by WES in CSF and ctDNA samples from five patients included in the correlative analyses (10 total samples: 7 CSF and 3 plasma samples). Columns represent individual samples, grouped by patient, and rows depict selected clinicopathologic features and recurrent genomic alterations relevant to HER2+ breast cancer. Genomic alterations include single-nucleotide variants and copy-number gains or losses, as indicated in the legend. Clinical annotations include hormone receptor (HR) status, HER2 status, best overall response assessed by RANO-BM and RECIST (v1.1), CBR, PFS, and OS status. Sample timing [baseline and on-treatment at cycle 2 day 1 (C2D1)] is indicated. BOR, best objective response; CNV, copy-number variation; CR, complete response.

Purpose:Human epidermal growth factor receptor 2 (HER2)–positive breast cancer brain metastases (BCBM) frequently harbor PI3K/AKT/mTOR alterations. Preclinical HER2+ BCBM models demonstrated durable tumor regression with dual PI3K and mTOR inhibition. Our phase II trial evaluated paxalisib, a brain-penetrant dual PI3K/mTOR inhibitor, combined with trastuzumab. Patients and Methods:Eligible patients had active/progressive HER2+ BCBM and could have received prior therapies; leptomeningeal disease was excluded. The trial included a safety run-in to determine the recommended phase II dose (RP2D), followed by expansion cohorts. The primary endpoint was confirmed central nervous system (CNS) objective response rate (ORR) per Response Assessment in Neuro-Oncology Brain Metastases (RANO-BM) criteria at the RP2D. Paired plasma and cerebrospinal fluid (CSF) circulating tumor DNA (ctDNA) was analyzed using ultralow-pass whole-genome sequencing (ULP-WGS) and whole-exome sequencing (WES). Results:Seventeen patients were treated (median age, 44 years; median 5 prior therapy lines). Paxalisib 30 mg daily plus trastuzumab was established as the RP2D. Among 12 patients treated at the RP2D, no confirmed CNS responses occurred [CNS ORR, 0%; 95% confidence interval (CI), 0%–26.5%], and the study closed for futility. The median overall survival (OS) was 16.5 months. Treatment was feasible, with a safety profile consistent with PI3K/mTOR inhibition. Exploratory ctDNA analyses showed higher tumor fraction (TF) in CSF than plasma, with longitudinal changes in CSF TF paralleling CNS tumor burden. Conclusions:Paxalisib combined with trastuzumab was feasible but did not meet the primary endpoint of CNS ORR in heavily pretreated HER2+ BCBM. Exploratory findings suggest that CSF-derived ctDNA may better capture tumor genomics and treatment response than plasma ctDNA in HER2+ BCBM. Significance:The brain-penetrant dual PI3K/mTOR inhibitor paxalisib plus trastuzumab showed minimal activity in heavily pretreated HER2+ BCBM. Paired liquid biopsy analyses demonstrated that CSF-derived ctDNA may better reflect tumor genomics and treatment response than plasma, supporting CSF for translational studies in this setting. Trial Registration:ClinicalTrials.gov identifier: NCT03765983.

Study flow, antitumor activity, and survival outcomes. A, Safety run-in, RP2D determination, and study flow. The study incorporated an initial safety run-in to evaluate DLTs and establish the RP2D of paxalisib in combination with trastuzumab. Five patients were treated at dose level 1 (paxalisib 45 mg daily), with two DLTs observed, leading to discontinuation of this dose level. A second safety run-in was conducted at dose level −1 (paxalisib 30 mg daily), in which six patients were evaluable for DLT assessment, and one DLT was observed. Paxalisib 30 mg daily was declared the RP2D and selected for expansion in cohort A. Twelve patients were treated at the RP2D in the first stage of cohort A, with no confirmed CNS responses per RANO-BM criteria. The trial was subsequently closed for futility. DLTs were assessed during the first 21-day treatment cycle. One patient in the second safety run-in was unevaluable for DLT assessment; therefore, one additional patient was enrolled to replace this patient. B, Best change in CNS target lesions. Waterfall plot showing the best percentage change from baseline in CNS target lesions assessed by RANO-BM among patients in the primary analysis population treated at the RP2D (n = 12). Case 7 was classified as progressive disease (PD) per RANO-BM due to clinical deterioration. C, Treatment duration and response. Swimmer plot depicting treatment duration (weeks) among all treated patients (n = 17). Each bar represents an individual patient and is colored according to best overall response per RANO-BM. Symbols indicate key clinical events, including CNS progression per RANO-BM, extracranial progression per RECIST v1.1, clinical deterioration, unconfirmed partial response (PR) per RECIST v1.1, and death, as indicated in the legend. Circles denote the paxalisib 45 mg cohort and triangles the 30 mg cohort. Case 2 met criteria for RANO-BM progressive disease; however, treatment discontinuation was due to toxicity. Case 7 was classified as progressive disease based on clinical deterioration, although treatment was discontinued because of toxicity. One additional patient discontinued treatment for clinical progression (case 16). D, Bicompartmental PFS. Kaplan–Meier estimate of bicompartmental PFS among patients treated at the RP2D in the primary analysis population (n = 12). E, OS. Kaplan–Meier estimate of OS among patients treated at the RP2D in the primary analysis population (n = 12).

Best percentage change in CNS target lesions from baseline per RANO-BM criteria among all treated patients (n=17).

Longitudinal plasma tumor fraction (TF) dynamics in individual patients relative to changes in CNS target lesions assessed by RANO-BM criteria.

TF dynamics, clonal evolution, and phylogenetic analyses from plasma and CSF and plasma ctDNA. A, Median TF estimated by ULP-WGS in paired plasma and CSF samples at baseline and cycle 2 day 1 (C2D1). TF was consistently higher in CSF than plasma at both timepoints (n = 10 samples/5 patients). B, Longitudinal CSF TF dynamics in individual patients relative to changes in CNS target lesions assessed by RANO-BM criteria (n = 10 samples/5 patients). Patients with declining CSF TF exhibited reductions in CNS tumor burden, whereas increasing TF was associated with lesion growth or nonevaluable disease. The asterisk (*) identifies the patient who derived clinical benefit. C, Clonal evolution in CSF ctDNA from patient #5, showing changes in cancer cell fraction (|ΔCCF| >0.20) between timepoints. D, Phylogenetic reconstruction for patient #4 based on WES of plasma and CSF ctDNA, illustrating shared truncal mutations and CSF subclones present at low clonality in plasma. CB, clinical benefit.

REMARK diagram for CSF and plasma ctDNA analyses.

Treatment-related adverse events (TRAEs) among patients treated at 45 mg safety run-in (n=5)

Baseline patient and treatment characteristics (n = 17).