Advances in molecular pathobiology of PCNSL – towards clinical implementation of novel diagnostics and therapeutics

Abstract

Primary central nervous system lymphoma (PCNSL) is a rare aggressive extranodal non-Hodgkin’s lymphoma which outcomes remain poor despite therapeutic advances. Recent improvements in our understanding of the molecular diversity of aggressive diffuse large B cell lymphomas suggest that the biological refinement of PCNSL diagnostics may be enabled by an integrated multi-modal approach, augmenting clinical data with modern minimally-invasive liquid biopsy-based assays and multi-omic molecular profiling. Single cell sequencing and spatial imaging technologies dissecting the immune microenvironmental architecture of PCNSL are anticipated to further refine future diagnostics. Concurrently, several innovative strategies featuring immunomodulatory agents, oncogenic kinase inhibitors and immunotherapies including chimeric antigen receptor (CAR) T-cell therapy have emerged. In this review, we provide an update of recent advances in the understanding of the pathobiology of PCNSL focusing on clinically relevant molecular subclassifications, highlight emerging therapeutics, and discuss the translational roadmap to clinical implementation of next-generation diagnostic platforms based on novel technologies.

Background

Primary central nervous system lymphoma (PCNSL) is a rare aggressive subtype of extranodal non-Hodgkin’s lymphoma (NHL) that accounts for 1–2% of all NHL, with median age of onset in the 6th -7th decade of life and an increasing incidence worldwide [1–4]. Commonly affected sites in the CNS include the brain, meninges, eye (vitreoretinal) and spinal cord. Given its unique anatomical location, PCNSL has been classified under the umbrella term of large B-cell lymphomas of immune-privileged sites under the 5th edition of WHO classification of Haematolymphoid Tumors, although this posited subcategorization remains highly contentious [5, 6].

Over 90% of PCNSL are diffuse large B cell lymphomas (DLBCL) and the majority are of the Activated B-Cell (ABC) subtype based on cell-of-origin signatures [7, 8]. Earlier molecular studies revealed common associated mutations in B-cell receptors (BCR) and toll-like receptors (TLR) cell signalling pathways that drive nuclear factor kappa-b (NFkB) and JAK-STAT activity, such as MYD88 and CD79B [9–11]. Additional genetic alterations facilitating proliferation include B-cell maturation interference such as PRDM1 and TBL1XR1, cell cycle regulation failure through CDKN2A gene deletion, and immune evasion pathways through deletions in human leukocyte antigen (HLA) gene loci and copy number gains in 9p24.1 resulting in increased programmed death-ligand 1/2 (PD-L1/PD-L2) expression [11–15]. On the other hand, primary CNS T-cell lymphomas make up a minority of cases, harboring molecular alterations akin to systemic peripheral T-cell lymphomas [16, 17], along with indolent low-grade B-cell lymphomas, NK/T cell lymphomas and other rare subtypes [18, 19]. The recognition and eventual integration of molecular underpinnings of PCNSL is expected to eventually lead to a biologically refined and clinically meaningful diagnostic entity, with inherent implications on therapeutic selection and patient prognosis [20].

Survival outcomes have modestly improved over decades of evolution in treatment options including high dose methotrexate (HD-MTX)-based chemoimmunotherapy combinations, consolidation strategies with autologous stem cell transplant (ASCT) or whole brain radiotherapy (WBRT), as well as emerging novel agents [21]. However, significant toxicities frequently preclude frail patients with comorbidities from such intensive therapies, as such 5-year overall survival generally remains poor at 30–40% [2–4]. Recently, better understanding of the underlying pathogenesis of PCNSL has brought about much needed avenues for novel diagnostic biomarkers and innovative therapies including targeted agents and immunotherapy that are anticipated to further improve patient outcomes.

Given these recent advances, this review aims to:

1. Provide a summarised update of recent advances in the understanding of the pathobiology of PCNSL focusing on clinically relevant molecular subclassifications.

2. Discuss technologies that enable deeper understanding of the disease and formulate a roadmap to clinical implementation of novel diagnostics.

3. Examine the landscape of known and emerging predictive biomarkers, highlighting targets for therapeutic selection and development.

Towards a molecular classification of PCNSL

Clinical presentation and workup

PCNSL can present with a varying constellation of neurological and neuropsychiatric symptoms depending on the anatomical location and extent of CNS involvement, with up to 40% of patients presenting with multifocal disease [22]. The brain is the most common CNS site affected, with lesions occurring predominantly in the periventricular white matter, corpus callosum, frontal lobes and basal ganglia [22]. Ocular involvement occurs in approximately 15–20% of patients and is bilateral in most cases, while fewer than 5% present with isolated primary vitreoretinal lymphoma (PVRL) [23–25]. As such, all patients should receive an ophthalmologic evaluation including slit-lamp examination and dilated fundus examination, with retinal angiography and optical coherence tomography performed where indicated [25]. Primary spinal cord lymphomas are rare and represent < 1% of all PCNSL [26]. Cerebrospinal fluid (CSF) evaluation should include biochemical profile, conventional cytology and flow cytometry, with increasing interest in incorporating molecular assays such as circulating tumor DNA (ctDNA) analysis and cytokine profiling to improve diagnostic sensitivity [27, 28].

Imaging with contrast-enhanced magnetic resonance imaging (MRI) of the brain in keeping with recommendations of the International PCNSL Collaborative Group (IPCG) is preferred [29]. Whole body scans with either positron emission tomography (PET) scan or a computed tomography (CT) with a bone marrow aspirate and trephine is typically performed if PET imaging is not available [25]. MRI of the spine may be indicated in patients with symptoms suggestive of spinal cord involvement. Histopathological confirmation by stereotactic brain biopsy remains the diagnostic gold standard, with surgical resection as an option only in cases involving exceptional patients with rapidly increasing intracranial pressure who may benefit from surgical debulking at the time of tumour biopsy [25]. If PVRL is suspected, vitrectomy is required. Corticosteroids should be avoided unless necessary prior to biopsy due to altered morphology and transient tumor regression, which can delay diagnosis [30, 31]. Despite advances in neuroimaging and neurosurgical practices, brain biopsies may be technically challenging, particularly for deep brain structures that can increase risks of haemorrhage, neurological deficits, and non-diagnostic yields [32, 33]. Furthermore, stereotactic biopsies typically sample only a small portion of the tumour and may not fully capture the spatial molecular heterogeneity of PCNSL. These limitations have driven increasing interest in molecular diagnostic approaches, including liquid biopsy assays and comprehensive molecular profiling of tissue, CSF, and plasma, which have the potential to complement conventional pathology, improve diagnostic accuracy, and facilitate disease monitoring (Fig. 1). In a recent prospective cohort study of patients with suspected CNS neoplasms, the incorporation of targeted rapid CSF genotyping into routine clinical workflows has been suggested to potentially accelerate the diagnosis of PCNSL to a median of 2 days, though such an approach would be applicable to limited cases harbouring MYD88 L265P mutations [34]. In patients already diagnosed with PCNSL, an ultrasensitive tumor-informed targeted sequencing approach (Cancer Personalized Profiling by Deep Sequencing, CAPP-Seq) for ctDNA profiling in plasma and CSF may also aid prognostication and guide disease monitoring by measurable residual disease detection during treatment [35]. However, as we shall discuss in the subsequent sections, even such novel approaches present their own constraints, and may not necessarily overcome the problem of sampling biases and intratumoral heterogeneity.

Fig. 1.

Current and emerging diagnostic paradigm for PCNSL

Molecular diversity of PCNSL

Modern advances in the molecular understanding of DLBCL have led to new ways of classification based on their genetic features. Based on their somatic mutations, copy number alterations and structural variants, molecular subtypes of DLBCL were consistently identified in several studies, capturing true biological heterogeneity beyond conventional subdivision by cell-of-origin [36–38]. Using a probabilistic classification tool, the LymphGen Classifier revealed seven genetic subtypes of DLBCL with distinct genetic alterations, gene expression signatures, and differential outcomes to modern immunochemotherapy and targeted agents [39]. These include “MCD” (characterized by MYD88 and CD79B mutations), “BN2” (characterized by BCL6 translocation and NOTCH2 mutation), “N1” (driven by NOTCH1 mutation), “EZB” (defined by EZH2 mutations and BCL2 translocations), “A53” (TP53 alterations), “ST2” (associated with SGK1 and TET2 mutations), and other unclassified tumors.

While earlier studies have suggested a proximity of PCNSL to the “MCD” subtype of DLBCL, harbouring frequent mutations in MYD88, PIM1, CD79B, and CDKN2A deletions, subsequent multi-omic analyses suggest that PCNSL comprises several biologically distinct molecular subgroups beyond the canonical MCD-like phenotype [7, 11]. In PCNSL, comprehensive multi-omic studies incorporating whole exome sequencing, RNA sequencing and DNA methylation sequencing revealed four distinct molecular patterns with prognostic implications. Furthermore, based on their molecular hallmarks, these individual subtypes have also been proposed to potentially benefit from specific targeted therapies such as BTK inhibitors, JAK1 inhibitors, immune checkpoint inhibitors and EZH2 inhibitors [12]. More recently, a study in a Chinese cohort identified three genetic subtypes of PCNSL based on whole exome sequencing data, including a favorable subgroup with co-occurring mutations in at least two genes among BTG1, MYD88 and IRF4, an unfavorable subgroup with mutation in EP300, and an intermediate subgroup without characteristic mutations [40]. Taken together, while these findings have improved our understanding of the molecular and immune diversity of PCNSL, differences in study design and heterogeneity across the patient cohorts preclude direct comparison. A unified standard classification system will require further investigation and validation in independent cohorts, before the possibility of clinical implementation to guide patient stratification, treatment selection and future therapeutic development.

Tumor immune architecture

The tumor immune microenvironment of PCNSL has attracted increasing interest because of its prognostic significance and therapeutic relevance, particularly in the era of immune checkpoint blockade [41, 42]. Alame et al. evaluated the immune milieu of PCNSL patients investigating its association with underlying molecular signalling and prognostic outcomes [43]. Using bulk RNA transcriptomic analysis, three subtypes were identified based on immune gene expression of a cohort of newly diagnosed PCNSL enriched with additional PCNSL and DLBCL microarray data from Gene Expression Omnibus (GEO): Immune-poor (Cluster 1) which lacked activated CD4 + and CD8 + T cells, Immune-intermediate (Cluster 3) with no specific pattern and a variable immune milieu and immune-rich (Cluster 2) which were enriched with CD8 + T cells, tumor-associated macrophages, and T-reg cells. Only one third of patients had an immune-rich phenotype compared to half in the DLBCL cohort, with one fifth of tumors of patients being immune-poor and the rest immune-intermediate. Immune-rich PCNSL subtypes were significantly associated with prolonged time to relapse, with a trend towards superior overall survival, a finding recapitulated in subsequent studies [12]. More recently, single cell transcriptomics and chromatin accessibility sequencing provided a more detailed profile of the different immune cell types in PCNSL, demonstrating predominantly exhaustion signatures in tumor-infiltrating CD8 + T cells. A plasmablast-like program (exhibiting a phenotype of XBP1⁺MZB1⁺MS4A1⁻) reflecting aggressive tumor biology within PCNSL was also associated with worse outcomes [44].

Immunodeficiency-associated PCNSL

Immunodeficiency-associated PCNSL (ID-PCNSL) represents a unique subtype of PCNSL, arising in the context of iatrogenic immunosuppression for transplantation or autoimmune disease, or HIV infection. Clinically, amongst ID-PCNSL, Epstein-Barr virus (EBV)-positivity confers a significantly worse prognosis, along with older age and poorer performance status [45]. In EBV-positive PCNSL, the genomic alterations and tumor microenvironment have also been reported and are biologically distinct from conventional PCNSL. Mutations in MYD88, CD79B, and PIM1 were not featured, while CDKN2A and HLA class I/II antigen-presenting/processing genes were often retained; rather, JAK/STAT and NOTCH pathway variants were more frequently observed [46]. Notably, EBV-positive PCNSL were mainly germinal center B cell (GCB) in terms of cell-of-origin by the Lymph2Cx assay. There was greater expression of genes for macrophage markers and immune checkpoints compared with other PCNSL subtypes, suggestive of a tolerogenic tumor microenvironment [47]. A more recent study using RNA-sequencing and EPIC DNA methylation arrays showed distinct transcriptional and epigenetic profiles in EBV-positive and EBV-negative PCNSL. Upregulation of IL-10, NOTCH and viral life cycle transcriptomic signatures was observed in EBV-positive cases, along with reduced global promoter methylation [48]. These findings support ID-PCNSL as a separate clinicopathological entity along the diverse biological spectrum of PCNSL.

The evolving landscape of PCNSL treatment

Current treatment paradigm

The treatment landscape of PCNSL has evolved substantially over the past several decades, transforming a disease once managed predominantly with WBRT into one treated with multimodal, HD-MTX-based strategies [49, 50]. Despite these advances, overall prognosis remains significantly worse compared to systemic DLBCL, reflecting its unique biological and clinical challenges [2]. Contemporary treatment strategies consist of induction therapy with HD-MTX as the backbone of combination induction systemic therapy. Subsequent consolidation aims to eradicate residual disease with either ASCT or WBRT, with transplant fitness and treatment-related neurotoxicity guiding the respective choice [25]. Given the high relapse rates, maintenance strategies have also been explored, particularly for older patients or those unsuitable for intensive consolidation. Treatment-related toxicity and morbidity remain as significant limitations especially in the elderly and frail populations [51, 52]. Hence, although current treatment paradigms guided by patient fitness and transplant eligibility have improved outcomes over the decades, several key issues remain in the field. Accordingly, major priorities include integrating targeted therapies and immunotherapeutics into frontline treatment, developing chemotherapy-light or chemotherapy-free strategies, and incorporating molecular biomarkers to personalize therapeutic selection and optimize outcomes.

Major clinical trials in PCNSL

The treatment landscape of PCNSL has been reviewed extensively in recent literature and will only be briefly summarized here [21, 25, 53, 54].

A series of earlier studies demonstrated efficacy for HD-MTX and combination chemo-immunotherapy induction strategies and these formed the foundation of modern-day trials [55–60]. Although the reported optimal dose of HD MTX varies from 3 to 8 g/m², doses < 3 g/ m² generally led to poorer PFS and OS [61, 62]. IELSG32 was a randomised phase 2 trial with 3 arms, comparing induction HD-MTX/Ara-C (arm A) alone as control arm with the addition of rituximab (arm B) and the addition of rituximab and thiotepa (MATRIX, arm C). At a median follow up of 88 months, it demonstrated that the addition of rituximab and thiotepa to induction HD-MTX /Ara-C had higher complete response rates (23% vs. 30% vs. 49%) and 7-year OS (21% vs. 37% vs. 56%) respectively [63]. The long term benefit of rituximab remains uncertain with a meta-analysis showing significantly improved 3-year OS but not 5-year OS [64]. Other regimens for fit patients used in Europe include HD-MTX, carmustine, teniposide and prednisolone (MBVP) with or without rituximab (R-MBVP) [65]. Alternative induction regimens especially in fit elderly patients include HD-MTX and temozolomide (MT) with or without rituximab (RMT) or rituximab, HD-MTX and procarbazine [66–68].

Established consolidation strategies currently consist of ASCT or WBRT, with different toxicity profiles and careful assessment of patient fitness determining preferred consolidation strategy [25]. Two prospective trials compared consolidation ASCT and WBRT after induction chemo-immunotherapy [63, 69]. The phase 2 PRECIS trial showed significant 8-year event-free survival (EFS) benefit of consolidation ASCT compared to WBRT (67% vs. 39%) but not OS (69% vs. 65%), likely due to salvage ASCT in more than half of those who relapsed post WBRT. Conversely the IELSG32 trial showed similar 7-year progression-free survival (PFS) and OS between both consolidation strategies with similar relapse rates post consolidation (32% vs. 31%). Treatment related mortality was overall low but higher with ASCT in both the PRECIS (7.5% vs. 11.4%) and IELSG32 (0% vs. 3.4%). Long term neurotoxicity of WBRT were significantly higher in both trials, including balance, attention and quality of life (QOL) scores, while only 1 patient out of both trials developed acute leukemia post ASCT. Further studies on reduced-dose WBRT may provide an alternative strategy balancing toxicity concerns particularly in vulnerable patient populations [70].

ASCT in the elderly is feasible and active as highlighted in the MARTA trial, however there is still significant haematological toxicity with 6% of treatment related deaths [71]. Careful patient selection and comprehensive geriatric assessment would likely expand the role of ASCT in the elderly [72]. PRIMA-CNS is an ongoing randomised phase 3 trial that will clarify the safety and efficacy of such an approach compared to maintenance chemotherapy after induction in the elderly [73]. Non-myeloablative or maintenance chemotherapy have been explored as possible less intensive consolidative approaches especially in the elderly. However, outcomes have generally been poorer compared to ASCT or WBRT and toxicities, especially haematological and infections, remain significant [74–77]. Recently, the IELSG43 study evaluated patients in at least partial remission following four cycles of MATRIX induction therapy, being randomized to two cycles of R-DeVIC (rituximab plus dexamethasone, etoposide, ifosfamide, and carboplatin) or ASCT with carmustine and thiotepa conditioning. ASCT significantly improved both PFS and OS compared with R-DeVIC consolidation chemotherapy, thus supporting its continued role as the preferred strategy in transplant-fit patients [78]. Ongoing trials including BLOCAGE-01 (NCT02313389) and IELSG45 (NCT03495960) would clarify these approaches further, though the latter suffered high drop-out rates, reflecting the challenges treating patients with borderline fitness and the need for novel therapeutics apart from cytotoxics [79].

Novel therapeutics

Oncogenic kinase and other small molecule targeted agents

Bruton tyrosine kinase inhibitors (BTKi)

BTK is integral to BCR signalling and activation of NFkB involved in B-cell proliferation [80]. Earlier generations of BTKi covalently bind to C481, while newer non-covalent BTKi have been developed to overcome acquired resistance [81]. Due to hallmark mutations present in up to 90% of PCNSL cases in the Toll/interleukin-1 receptor (TIR) domain of MYD88 and the tyrosine-based activation motif (ITAM) domain of CD79A/B which are involved in NFkB signaling (i.e. akin to the LymphGen MCD subtype of DLBCL), BTKi have been investigated in emerging regimens for PCNSL [10, 82, 83]. The co-occurrence of MYD88 and CD79B mutations has been associated with high response rates (80%) to ibrutinib monotherapy in DLBCL [84, 85].

In the relapsed/refractory (R/R) setting, BTKi monotherapies can achieve objective response rates (ORR) of approximately 50–75%, with second generation agents better tolerated, although responses are not durable [86–88]. Combination therapies have been tested with immune checkpoint inhibitors (ICI) and/or chemotherapy, especially due to the synergistic effect of BTK inhibition increasing T cell diversity, and these have achieved higher ORR and CR rates of up to 84% and 79% respectively although longer term follow up is required for durability of responses [89–94]. BTKi is also increasingly being evaluated in the first-line setting in view of its good tolerability, especially in novel maintenance strategies in patients who may not be fit for conventional consolidation with transplant or WBRT. These have generally achieved higher ORR (74–100%) and CR rates (35–97%) compared to historical induction chemotherapy trials, as well as promising efficacy with 2-year PFS and OS of up to 84% and 95% respectively, but these remain to be evaluated in large randomised trials with longer duration follow up [95–102]. A meta-analysis evaluating BTKi efficacy in PCNSL across all lines either as monotherapy or in combination with chemotherapy or chemo-immunotherapy revealed a pooled ORR of 73% and CR rates of 49% [103]. Hence BTKi play an emerging role in achieving higher and more durable responses in the first line setting for frailer patients who may otherwise not tolerate consolidation therapies well or in the R/R setting where options are limited.

Immunomodulatory drugs (IMiD)

Lenalidomide exerts anti-neoplastic effects on NHL by disrupting kinase signalling pathways involved in B cell proliferation and have demonstrated efficacy in multiple B cell lymphomas [104–106]. Lenalidomide combinations and newer generation IMiDs have demonstrated clinical activity in R/R PCNSL [107, 108]. Incorporation of IMiDs into induction and maintenance therapies in the first line setting, similar to BTKi, has also been assessed in several trials, achieving ORR of up to 88.2% and CR rate of 78% [109, 110]. The LOC-R01 trial assessed non-comparatively the addition of either lenalidomide or ibrutinib to R-MPV induction with responding patients undergoing ASCT consolidation, with primary endpoint being CR rate at end of induction [111]. 86% and 82% of patients in the lenalidomide and ibrutinib arm respectively achieved CR with similar 18-month PFS and OS rates, as well as consistent safety profiles seen previously for these regimens. Further trials are ongoing in evaluating its role in first line therapy (NCT06554561).

mTOR/PIK3CA pathway inhibitors

PI3K/AKT/mTOR signalling plays a significant role in the pathogenesis of various cancers and can be activated by BCR signalling [112]. Aberrant activation of the pathway was found to be significantly increased in PCNSL and associated with shorter PFS [113, 114]. However earlier studies in PI3K/AKT/mTOR pathway inhibition produced only short-lived responses, attributed partly due to poor CNS penetration and its role in PCNSL remains uncertain [115–117]. Ongoing trials with newer generation, CNS penetrant, selective PI3K/mTOR inhibitors (NCT04906096) may help reignite interest for this therapeutic class.

BTK degraders

BTK inhibition although promising is generally limited by acquired resistance mutations that bypass C481 binding and can also overcome non-covalent BTKi [118, 119]. Some of these mutations surprisingly downregulate kinase activity, and instead facilitate interaction with downstream signalling molecules via a scaffold function that sustain BCR signalling [120]. As such BTK degradation overcomes such resistance mutations by eliminating BTK entirely, and early phase trials mainly in CLL have been promising [120, 121]. The catalytic efficiency of BTK degraders also helps overcome the pharmacokinetic limitations of the blood-brain-barrier since low drug concentrations are sufficient to achieve degradation activity [122]. A proof of concept study with a novel oral CNS penetrant BTK degrader NX-5948 in murine models implanted with a patient-derived refractory PCNSL demonstrated significantly longer survival compared to treatment with ibrutinib [123]. Ongoing early phase clinical trials (NCT05131022) may shed light on its future potential.

PIM inhibitors

PIM1 mutations are one of the most common mutations seen in the genomic landscape of PCNSL [40]. Studies in DLBCL have previously shown PIM1 expression correlating with increased JAK/STAT signalling, migratory potential facilitating extranodal dissemination and inferior prognosis [124–126]. Despite its therapeutic potential due to its prevalence in PCNSL, PIM1 inhibition has been disappointing thus far in early phase trials across multiple cancers due to limited efficacy [127–129]. DLBCL cell lines have elucidated possible synergism of combined PIM and BCL2 inhibition with increased apoptosis [130]. Further clinical trials are needed to assess whether these strategies will translate into clinical efficacy.

Selective inhibitor of nuclear export (SINE)

Nucleo-cytoplasmic shuttling plays a key role in cell homeostasis and disruptions to this traffic can result in tumorigenesis through decreased tumor suppressor, apoptosis and differentiation [131]. The exportin XPO1 has been previously shown to be associated poorer outcomes in B cell malignancies and has additional functions in drug resistance [132]. Selinexor was previously approved for R/R DLBCL based on the SADAL trial but this was subsequently withdrawn in April 2026 as further phase 3 confirmatory trials were not conducted [133, 134]. It has been previously shown to have good CNS penetration in preclinical studies with glioblastoma and clinical activity in a case of secondary CNS lymphoma (SCNSL) [135, 136]. Li et al. demonstrated in DLBCL cell lines that selinexor with BTKi had synergistic inhibition NF-κB, P-STAT3, and NFATc1 signaling pathways, with increased T cell infiltration in mouse models [137]. Building upon this work, a single arm phase 2 trial was conducted in first line CNS lymphoma and treated with HD-MTX, zanubrutinib and selinexor. CSF sampling done during the trial showed measurable CNS penetration. The trial achieved a CR rate of 95.2%, 1-year PFS and OS of 94.1% and 100% respectively [137]. Longer term follow up and larger cohorts are required to validate these promising results.

Current strategies incorporating these targeted molecular agents (summarised in Table 1) still lack precise biomarkers contributing to heterogenous efficacy signals. Incorporating multi-omic profiling would likely in future allow more precise therapeutic targeting, better predict response and avoid excessive toxicity in patients.

Table 1.

Clinical studies on oncogenic kinase and other small molecule inhibitors in PCNSL

Setting	Drug class	Study [Ref]	Design	N	Strategy	Population (PCNSL n)	ORR	CR (CR/CRu)	PFS	OS
First-line	BTKi + chemotherapy	Gao 2025 [95]	Single-arm phase 2b	35 (33)	HD-MTX + ibrutinib + TMZ → ibrutinib maintenance	PCNSL	93.9%	72.7%	2-yr PFS 57.6%	2-yr OS 84.8%
First-line	BTKi + chemotherapy	Schaff 2026 [96]	Single-arm phase 2b	30	Ibrutinib + R-MPV → cytarabine or ASCT	PCNSL	100%	97%	2-yr PFS 84.2%	mOS NR
First-line	BTKi + chemotherapy	Zhao 2025 [97]	Single-arm phase 2b	65 (61)	R-HD-MTX + orelabrutinib → ASCT, WBRT or maintenance	PCNSL	92.3% ITTg	35.4% ITTg	2-yr PFS 74.9%	2-yr OS 85.4%
First-line	BTKi + chemotherapy	Zhao 2025 [98]	Single-arm phase 2b	22	R-HD-MTX + orelabrutinib → ASCT, WBRT or maintenance	PCNSL	*91%	*41%	2-yr PFS 59.2%	2-yr OS 66.3%
First-line	BTKi + IMiD + chemotherapy	Song 2026 [99]	Single-armb	48 (45)	Zanubrutinib + R² + TMZ ± HD-MTX → ASCT or maintenance	PCNSL	86.6%	44.4%	mPFS 26 m; 2-yr PFS 51.7%	2-yr OS 70.8%
First-line	BTKi	Bairey 2023 [101]	Single-arm phase 2b	20	Ibrutinib maintenance after first-line response	PCNSL	--	--	2-yr PFS 72.6%	2-yr OS 89%
First-line	BTKi + chemotherapy	Xia 2025 [102]‡	Single-armb	29	Zanubrutinib + R-HD-MTX → zanubrutinib maintenance	PCNSL	*89.7%	*89.7%	2-yr PFS 82.8%h	2-yr OS 95.2%h
First-line	IMiD + chemotherapy	Yuan 2024 [109]	Single-arm phase 2b	17	R² + HD-MTX → R² maintenance	PCNSL	*88.2%	*76.5%	2-yr PFS 58.8%	3-yr OS 76.0%
First-line	IMiD + BTKi	Zhang 2025 [110]‡	Single-arm phase 2b (interim analysis)	33 (32)	Pomalidomide + orelabrutinib + rituximab → R-HD-MTX + orelabrutinib	PCNSL	*84.3%	*78%	2-yr PFS 56.2%	2-yr OS 81%
First-line	IMiD vs. BTKi (+ chemotherapy)	Soussain 2025 [111]‡	Randomised, non-comparative phase 2a	92 (89)	R-MPV + lenalidomide or ibrutinib → ASCT	PCNSL	*91% vs. 93%i	*86% vs. 82%	18-m PFS 72% vs. 82%	18-m OS 87% vs. 88%
First-line	BTKi + SINE/XPO1 + chemotherapy	Li 2025 [137]‡	Single-arm phase 2b	21	HD-MTX + zanubrutinib + selinexor → ASCT, radiotherapy or maintenance	Bothj	--	95.2%	1-yr PFS 94.1%j	1-yr OS 100%
R/R	BTKi	Grommes 2024 [86]	Single-arm phase 1/2b	46	Ibrutinib monotherapy	Both (PCNSL 31)k	74%	38.7%	mPFS 4.5 m; 3-yr PFS 10.1%	mOS 25.4 m
R/R	BTKi	Nayak 2025 [87]‡	Single-arm phase 2b	48	Tirabrutinib monotherapy	PCNSL	66.7%	43.8%	mPFS 6.0 m	mOS NR
R/R	BTKi	Fox 2025 [88]‡	Single-arm phase 2b	23 (20)	Zanubrutinib monotherapy	PCNSL	55%l	35%l	mPFS 6.9 m; 1-yr PFS 37%	mOS 13 m; 1-yr OS 53%
R/R	BTKi + chemotherapy	Lin 2024 [89]	Single-arm phase 2b	34	Zanubrutinib + high-dose cytarabine	PCNSL	64.7%	47.1%	mPFS 4.5 m	mOS 18 m
R/R	IMiD + anti-CD20	Ghesquieres 2019, REVRI [107]	Single-arm phase 2b	50 (45)	R² → lenalidomide maintenance	PCNSL + PVRL (34 + 11)	*35.6%m	*29%m	mPFS 7.8 m	mOS 17.7 m
R/R	IMiD + dexamethasone	Tun 2018 [108]	Single-arm phase 1b	29 (25)	Pomalidomide + dexamethasone	PCNSL + PVRL	48%	32%	mPFS 5.3 m	--
R/R	mTOR inhibitor	Korfel 2016 [115]	Single-arm phase 2b	37	Temsirolimus monotherapy	PCNSL	54%	21.6%	mPFS 2.1 m; 1-yr PFS 5.4%	mOS 3.7 m; 1-yr OS 19%
R/R	PI3K inhibitor	Grommes 2023 [116]	Single-arm phase 2b (terminated after 4 patients)	4	Buparlisib monotherapy	Both (PCNSL 2)	25%	0%	mPFS 39 days	mOS 196 days
R/R	PI3K inhibitor + BTKi	Schaff 2025 [117]‡	Single-arm phase 1bb (terminated early)	18	Copanlisib + ibrutinib (concurrent or sequential)	PCNSL	57% concurrent; 50% sequential	--	mPFS 3 mn	--
All lines	BTKi	Tan 2025 [103]	Meta-analysis (40 studies)e	935o	Monotherapy or combination (any line)	Both (PCNSL 674)o	73%	49%	--	--

Abbreviations: ASCT, autologous stem-cell transplantation; BTKi, Bruton tyrosine kinase inhibitor; CNS, central nervous system; CR, complete response; CRu, complete response, unconfirmed; HD-MTX, high-dose methotrexate; IMiD, immunomodulatory drug; ITT, intention-to-treat; m, months; mOS, median overall survival; mPFS, median progression-free survival; mTOR, mammalian target of rapamycin; NR, not reached; ORR, overall response rate; OS, overall survival; PCNSL, primary central nervous system lymphoma; PFS, progression-free survival; PI3K, phosphoinositide 3-kinase; PVRL, primary vitreoretinal lymphoma; R², rituximab plus lenalidomide; R-HD-MTX, rituximab plus high-dose methotrexate; R-MPV, rituximab, methotrexate, procarbazine and vincristine; R/R, relapsed or refractory; SCNSL, secondary central nervous system lymphoma; SINE/XPO1, selective inhibitor of nuclear export / exportin 1; TMZ, temozolomide; WBRT, whole-brain radiotherapy; yr, year

Study design: ^aprospective randomised trial; ^bprospective single-arm trial (phase 1, 1b, 1/2 or 2, or pilot); ^emeta-analysis. ^‡Conference abstract only

^*End of induction. ^gAfter four cycles (98.4%/37.7% in 61 evaluable); CR 72.2% after six. ^hTwo-year estimates for the 25 patients not consolidated with ASCT. ⁱORR derived from reported response counts (lenalidomide 40/44; ibrutinib 42/45). ^jNewly diagnosed primary or secondary CNS lymphoma (breakdown and ORR not reported); 1-year PFS 94.1% (results) vs. 95.2% (conclusions) in the abstract. ^kPCNSL subgroup shown. ^lCentral review; investigator-assessed ORR 50% and CR/CRu 15%. ^mResponses in the 45 assessable patients; by intention to treat (n = 50) ORR 32.0% and CR/uCR 13/50 (26%). ⁿ2.93 m concurrent vs. 3.65 m sequential, as presented; terminated early when copanlisib was withdrawn. ^oPrimary and secondary CNS lymphoma, 40 studies; pooled estimates for the PCNSL subgroup (n = 674 ORR, 626 CR)

N, patients enrolled (evaluable for response in parentheses, where different)

Trials combining agents from more than one class are listed once only (immune checkpoint inhibitor combinations in Table 2; cellular, bispecific-antibody or antibody–drug conjugate combinations in Table 3)

Immune checkpoint inhibitors and other novel immunotherapeutics

The PD1/PDL1 pathway plays a significant role in immune evasion of PCNSL contributed by increased frequency of copy number amplification at the 9p24.1 locus increasing PDL1/2 expression, providing a biological rationale for immune checkpoint inhibitors (ICI) [15]. TOX mutations, present in up to 50% of PCNSL, may also predict ICI response due to its role promoting CD 8 T cell exhaustion [11, 138]. Furuse et al. evaluated 70 patients retrospectively for PDL1/PDL2 expression which highlighted minimal to no expression in the tumor cells but was present in both intratumoral and peritumoral macrophages, with higher PDL1 expression in these macrophages associated with better survival [139]. Therapeutic studies evaluating ICI in PCNSL are summarised in Table 2. An early case series showed promising responses with nivolumab monotherapy but subsequent phase 2 studies were disappointing [140–142]. Combination ICI with other systemic agents were further assessed for potential improvement in outcomes. Ambady et al. highlighted the promising efficacy of rituximab with ICI in a small series of 6 patients with R/R CNS lymphoma to leverage T cell augmentation together with antibody dependent cellular cytotoxicity of rituximab, achieving 50% CR [143]. As discussed earlier, several trials have also evaluated synergistic combination of BTKi and ICI [90, 91, 93]. Combining these approaches, Chukwueke et al. evaluated in a phase 1b/2 study the combination of rituximab, ibrutinib and pembrolizumab in R/R PCNSL and achieved an ORR of 80% [94]. Although median duration of response was only 177 days, treatment was well tolerated with only 2 patients discontinuing due to toxicity, with the phase 2 expansion cohort ongoing (NCT04421560). Similarly, Chang et al. tested rituximab, acalabrutinib and durvalumab in another phase 1b study of 17 patients with predominantly R/R PCNSL, with more modest results: an ORR of 62.5% and a CR rate of 12.5%, and a median PFS and OS of 4.3 and 11.9 months, respectively [144]. Other ICI combination trials in the R/R setting are ongoing (NCT06556199). Hence, better selection of patients with predictive biomarkers, particularly for immune rich subsets as detailed earlier, as well as combination therapies may help overcome some of the challenges with ICI monotherapy [43].

Table 2.

Studies on immune checkpoint inhibitors in PCNSL

Setting	Drug class	Study [Ref]	Design	N	Strategy	Population (PCNSL n)	ORR	CR (CR/CRu)	PFS	OS
First-line	BTKi + ICI + chemotherapy	Zhao 2026 [100]	Single-arm phase 1/2b	31 (27)	Orelabrutinib + anti-PD-1 + fotemustine	PCNSL	85.2%	66.7%	mPFS 9.4 m	mOS 22.8 m; 2-yr OS 48%
First-line	ICI + chemotherapy	Zeng 2024 [145]	Single-arm phase 2b	28 (27)	Sintilimab + R-HD-MTX + TMZ; no consolidation	PCNSL	96.3%	92.6%	2-yr PFS 57.2%	2-yr OS 91.5%
First-line	ICI	Park 2025 [148]‡	Single-arm phase 1/1bb	14	Nivolumab consolidation after HD-MTX induction (age ≥ 65 year)	PCNSL	86%	79%	2-yr PFS 63%	2-yr OS 79%
R/R	BTKi + ICI	Chihara 2025 [90]	Single-arm phase 2b	18	Ibrutinib + nivolumab	Both (PCNSL 16)	78%	50%	mPFS 6.5 m; 1-yr PFS 42%	mOS 21.0 m; 1-yr OS 63%
R/R	BTKi + ICI + chemotherapy	Yuan 2025, OST [91]‡	Single-arm, phase not statedb	19	Orelabrutinib + sintilimab + TMZ → sintilimab maintenance	PCNSL	84.2%	78.9%	mPFS 9.0 m; 2-yr PFS 17.5%	2-yr OS 53.1%
R/R	BTKi + ICI	Nizamuddin 2024 [93]‡	Single-arm phase 1b	13 (10)	Acalabrutinib + durvalumab	PCNSLg	40%	40%	--	--
R/R	BTKi + ICI + anti-CD20	Chukwueke 2023 [94]‡	Single-arm phase 1b/2b (phase 1b results)	10	Pembrolizumab + ibrutinib + rituximab	PCNSL	80%	60%	6-m PFS 50%h	--
R/R	BTKi + ICI + anti-CD20	Chang 2026 [144]	Single-arm phase 1bb	17 (16)	Rituximab + acalabrutinib + durvalumab	PCNSL (15 R/R)i	62.5%	12.5%	mPFS 4.3 m	mOS 11.9 m
R/R	ICI	Nayak 2017 [140]	Retrospective case seriesd	5	Nivolumab monotherapy	PCNSL (4); PTL with CNS relapse (1)	100%	80%	--	--
R/R	ICI	CheckMate 647 (NCT02857426) [141]†	Single-arm phase 2b	47	Nivolumab monotherapy	PCNSL cohort	6.4%	2.1%	mPFS 1.41 m	mOS 6.77 m
R/R	ICI	Hoang-Xuan 2020, AcSé [142]‡	Single-arm phase 2b	50	Pembrolizumab monotherapy	PCNSL (incl. 9 PVRL)	26%	16%	mPFS 2.6 m; 6-m PFS 29.8%	6-m OS 60.4%
R/R	ICI + anti-CD20	Ambady 2019 [143]	Retrospective case seriesd	6	Pembrolizumab or nivolumab + rituximab	Both (PCNSL 3)	50%	50%	--	--

Abbreviations: anti-PD-1, anti–programmed cell death protein 1; BTKi, Bruton tyrosine kinase inhibitor; CNS, central nervous system; CR, complete response; CRu, complete response, unconfirmed; HD-MTX, high-dose methotrexate; ICI, immune checkpoint inhibitor; m, months; mOS, median overall survival; mPFS, median progression-free survival; MTX, methotrexate; ORR, overall response rate; OS, overall survival; PCNSL, primary central nervous system lymphoma; PFS, progression-free survival; PTL, primary testicular lymphoma; PVRL, primary vitreoretinal lymphoma; R-HD-MTX, rituximab plus high-dose methotrexate; R/R, relapsed or refractory; SCNSL, secondary central nervous system lymphoma; TMZ, temozolomide; yr, year

Study design: ^bprospective single-arm trial (phase 1, 1b, 1/2 or 2, or pilot); ^dretrospective case series (< 10 patients). ^‡Conference abstract only. ^†Efficacy data derived from the ClinicalTrials.gov results record and not reported in a peer-reviewed publication

^gThree patients with SCNSL enrolled before a protocol amendment restricting enrolment to PCNSL were excluded from the efficacy analysis. ^hMedian duration of response 177 days; phase 2 expansion cohort ongoing (NCT04421560). ⁱ15 of the 17 patients had R/R disease

N, patients enrolled (evaluable for response in parentheses, where different)

Trials combining agents from more than one class are listed once only (immune checkpoint inhibitor combinations in Table 2; cellular, bispecific-antibody or antibody–drug conjugate combinations in Table 3)

ICI have also been tested in the frontline setting with combination chemo-immunotherapy. Zeng et al. evaluated RMT with sintilimab with no planned consolidation [145]. Among 27 evaluable patients, ORR was 96.3% including 93% CR. At a median follow up of 24.4 months, 2-year PFS and OS rates were 57.2% and 91.5%, respectively. The regimen was generally well tolerated with mainly reversible grade 3–4 adverse events of raised serum transaminases and hematologic toxicities attributed to HD MTX and chemotherapy. 35.7% experienced grade 1–2 immune-related adverse events (IRAE), mainly pruritus and rash, although 2 patients developed grade 1 pneumonitis. No patients experienced dose reduction or discontinuation. PDL1 expression was not associated with PFS, but baseline CSF cytokine profiles did, with higher IL10 level or IL10/IL6 ratio, as well as lower IL4 and IL6 levels associated with poorer PFS. Tumor whole-exome sequencing done on 26 samples also identified patients with durable responses had higher frequency of RTK-RAS and PI3K pathway mutations. NOTCH and Hippo pathway mutations were more common in those with no durable response, which have been shown to affect ICI efficacy in other cancers [146, 147]. Similarly, a phase 1/1B study was conducted in elderly patients ≥ 65 with newly diagnosed PCNSL evaluating consolidation nivolumab up to 6 cycles post HD-MTX induction combinations. Median age was 72 years and 36% had ECOG 2–3 performance status. In patients who received at least 1 cycle of nivolumab, ORR and CR rate was 86% and 79% respectively, with median PFS and OS not reached at a median follow up of 33 months. 2-year PFS and OS was 63% and 79% respectively, with no dose limiting toxicities seen in stage 1 of the trial although 1 patient developed Stevens-Johnson syndrome after cycle 1 [148]. Other trials looking at ICI consolidation are ongoing [149]. This may represent a novel strategy especially for elderly patients who may not be fit for consolidation ASCT and want to avoid the neurotoxicity of WBRT.

Chimeric antigen receptor (CAR)-T cell therapy

CAR-T cell therapy has revolutionised the treatment landscape in NHL, with 4 different products approved across multiple different NHLs and lines of therapy [150]. Recently, updated ten-year outcomes of patients with R/R DLBCL and follicular lymphomas (FL) treated with tisagenlecleucel revealed approximately one third of DLBCL patients remained in lymphoma-free remission at ten years in heavily pre-treated settings who traditionally had dismal outcomes [151]. Patients with CNS involved lymphomas were specifically excluded from earlier landmark CAR-T trials in view of theoretical concerns of immune effector cell-associated neurotoxicity syndrome (ICANS) [152]. However, multiple real-world studies including SCNSL, PCNSL or both were reassuring and encouraging for activity [153–157]. A 2023 meta-analysis by Cook et al. included 128 patients with R/R PCNSL and SCNSL showed that the PCNSL subgroup had CR rate of 56% with 37% 6-month CR rate, with any-grade ICANS in 53% and grade 3–4 ICANS in 18%, showing similar neurotoxicity to registration trials in DLBCL [158].

Another CAR-T toxicity in the setting of CNS lymphoma and other brain tumors, is a pseudoprogression syndrome termed tumor inflammation-associated neurotoxicity (TIAN), which is related to focal peri-tumoral inflammation elicited by antitumor T-cell activity [159]. This is distinct from ICANS, in which a more global, off-target inflammatory mechanism takes place, typically in the setting of cytokine release syndrome (CRS). In patients with CNS lymphomas treated with CD19-directed CAR-T cells, TIAN occurred in 17.9% of cases and was typically transient, though a single fatal event was reported, with higher risks associated with larger baseline tumor volume. The occurrence of TIAN was correlated with improved treatment response rates and PFS outcomes [160]. In both ICANS and TIAN, recognition of at-risk patients and close surveillance are warranted to ensure early recognition and appropriate management (including consideration for corticosteroids, interleukin-1 blockade, hyperosmolar agents, and CSF diversion when indicated) [161].

Building on the promising safety and efficacy data from retrospective studies, several early phase trials of CAR-T therapy in CNSL have been reported (summarised in Table 3) [162–164]. Nayak et al. conducted a pilot study in 18 patients with R/R CNSL who were treated with axicabtagene ciloleucel and achieved an ORR of 94%, with a median PFS and OS of 14.3 months and 26.4 months, respectively. 89% and 44% developed CRS and ICANS respectively with grade 3 or higher ICANS in 28% of patients [162]. Li et al. conducted a phase 1/2 trial in R/R PCNSL and SCNSL with 69 patients using CD19 CAR-T cell therapy, of which 45 patients (23/45 PCNSL) received cell infusion [163]. The ORR was 75.6% and CR rate was 66.7% with median PFS of 20.1 months. 13.3% of patients developed ICANS with only 1 patient who had grade 4 ICANS. As a result of the pilot study by Nayak et al., the FDA updated the prescribing information for axicabtagene ciloleucel (February 2026), removing R/R PCNSL from its prior limitations of use [165].

Table 3.

Studies on CAR-T cells, BITEs and ADCs in PCNSL

Setting	Drug class	Study [Ref]	Design	N	Strategy	Population (PCNSL n)	ORR	CR (CR/CRu)	PFS	OS
R/R	CAR-T (CD19)	Choquet 2024 [154]	Retrospective multicenter cohortc	27 (25 infused)	CD19 CAR-Tg	PCNSL only	80%	64%	mPFS 8.4 m; 1-yr PFS 46%h	mOS 21.2 m; 1-yr OS 55%h
R/R	CAR-T (CD19)	Mercadal 2025 [155]	Registry analysis (CIBMTR)c	24 (23)	CD19 CAR-Tg	PCNSL only	61%i	48%i	2-yr PFS 28%	2-yr OS 50%
R/R	CAR-T (CD19)	Ossami Saidy 2025 [156]	Registry analysis (EBMT)c	100	CD19 CAR-Tg	Both (PCNSL 16)	60%i, j	36%i, j	2-yr PFS 28%	2-yr OS 37%
R/R	CAR-T (CD19)	Cassanello 2025 [157]	Retrospective cohort (multicenter)c	54 (52)	CD19 CAR-Tg	Both (PCNSL 8)	73%i	65%i	1-yr PFS 35%; mPFS 7.5 m	1-yr OS 63%; mOS 19 m
R/R	CAR-T (CD19)	Kaulen 2026 [166]	Retrospective cohort (single center)c	60	CD19 CAR-T	Both (PCNSL 20)	60%	45%	Median CNS-PFS 4 mk	--
R/R	CAR-T (CD19)	Cook 2023 [158]	Meta-analysis (15 studies)e	128	Mainly CD19 CAR-T	Both (PCNSL 30)l	64%	56%	--	--
R/R	CAR-T (CD19)	Nayak 2024 [162]‡	Pilot studyb	18	Axicabtagene ciloleucel	Both (PCNSL 13)	94%	67%	mPFS 14.3 m	mOS 26.4 m
R/R	CAR-T (CD19)	Li 2025 [163]‡	Single-arm phase 1/2b (single center)	69 (45 infused)	CD19 CAR-T	Both (PCNSL 23)	75.6%	66.7%	mPFS 20.1 m	mOS NR
R/R	CAR-T (CD19)	Frigault 2022 [164]	Single-arm phase 1/2b	13 (12 infused)	Tisagenlecleucel	PCNSL only	58.3%	50%	--m	--m
R/R	CAR-T (CD19)	Yang 2026 [168]	Retrospective cohort (single center)c	60 (ASCT 33; CAR-T 27)	ASCT vs. CAR-T consolidation in CR after salvagen	Both (PCNSL 42)	NA (all in CR at entry)	NA (all in CR at entry)	2-yr PFS 92.9% vs. 65.7% (ASCT vs. CAR-T), p = 0.026n	2-yr OS 91.7% vs. 74.1% (ASCT vs. CAR-T), p = 0.101n
R/R	Bispecific (CD20×CD3)	Gilbert 2025 [178]‡	Retrospective cohort (multicenter)c	55 (41)	Epcoritamab or glofitamab ± other agents	Both (PCNSL 4)	65.9%	36.6%	mPFS 8.5 m	mOS NR
R/R	Bispecific (CD20×CD3)	Xing 2026 [181]	Retrospective cohort (multicenter)c	84	Glofitamab ± other agentso	Both (PCNSL 53)	92%o	68%o	mPFS 19.5 m; 1-yr PFS 70.5%o	mOS NR; 1-yr OS 84.7%o
R/R	Bispecific (CD20×CD3)	Chohan 2026 [180]	Retrospective cohort (multicenter)c	28p	CD20×CD3 bispecific ± other agents	Both (PCNSL 3)	56%p	33%p	6-m PFS 43%p	6-m OS 77%p
R/R	ADC (anti-CD79b)	Gao 2025 [184]	Preclinicalf + retrospective case seriesd	3	Polatuzumab vedotin-based combinations	Both (PCNSL 1)	66.7%	66.7%q	--	--

Abbreviations: ADC, antibody–drug conjugate; ASCT, autologous stem-cell transplantation; BITE, bispecific T-cell engager; CAR-T, chimeric antigen receptor T-cell; CIBMTR, Center for International Blood and Marrow Transplant Research; CNS, central nervous system; CR, complete response; CRu, complete response, unconfirmed; EBMT, European Society for Blood and Marrow Transplantation; m, months; mOS, median overall survival; mPFS, median progression-free survival; NA, not applicable; NR, not reached; ORR, overall response rate; OS, overall survival; PCNSL, primary central nervous system lymphoma; PFS, progression-free survival; R/R, relapsed or refractory; SCNSL, secondary central nervous system lymphoma; yr, year

Study design: ^bprospective single-arm trial (phase 1, 1b, 1/2 or 2, or pilot); ^cretrospective cohort or registry analysis; ^dretrospective case series (< 10 patients); ^emeta-analysis; ^fpreclinical study. ^‡Conference abstract only

^gBridging therapy in 96%, 20%, 82% and 70% of patients in [154–156] and [157]. ^hMedians from leukapheresis; 1-year rates from infusion. ⁱBest response by day 100 after infusion. ^jAt day 180, ORR was 51% (bridged) vs. 44% (non-bridged), with CR 41% in both. ^kCNS progression-free survival from infusion. ^lEstimates shown are for the PCNSL subgroup (n = 30). ^mMedian PFS and OS not reported; 3/12 progression-free and 7/12 alive at median follow-up 12.2 m. ⁿPCNSL subgroup (n = 42); non-randomised, ASCT patients younger (median age 53 vs. 65 years). ^oEfficacy for the PCNSL subgroup (n = 53): ORR/CR 88%/59% with monotherapy (n = 32), 100%/81% with combination (n = 21). ^pActive CNS involvement (n = 18); previously treated CNS involvement (n = 10): ORR 60%, CR 50%, 6-m PFS 46%, 6-m OS 89%. ^qBoth CRs were in SCNSL patients (one consolidated with CAR-T); the PCNSL patient had stable disease

N, patients enrolled (evaluable for response in parentheses, where different)

Trials combining agents from more than one class are listed once only (immune checkpoint inhibitor combinations in Table 2; CAR-T, BITE or ADC combinations in Table 3)

Emerging predictors of treatment response and toxicity have also been reported, including the presence of leptomeningeal disease, peripherally contrast-enhancing lesions, type of CAR-T product, response to bridging therapy, performance status and tumor burden [154, 156, 157, 166]. Lacan et al. analysed the blood and CSF profiles of 21 patients with R/R CNSL treated with tisagenlecleucel or axicabtagene ciloleucel. 16 R/R CNSL patients with CSF analysis confirmed CAR-T cell trafficking in all patients, with responders having higher peaks. Among 20 patients who had phenotypic characterisation of blood CAR-T cells at the expansion peak, responders had stronger CD8 + cytotoxic T cell and IFN-γ response, with weaker PD1 expression. CD4 + T cells with higher IL7R and IL3 expression were also associated with better clinical responses [167]. Hence future CAR-T products incorporating IL7 activation may offer a potential approach for PCNSL.

Recently, Yang et al. retrospectively compared ASCT and CAR-T therapy as consolidation after CR to salvage second line therapy in R/R CNSL patients [168]. In the PCNSL subgroup, 2-year PFS was significantly higher (92.9% vs. 65.7%) in the ASCT group than in the CAR-T group, whilst 2-year OS was similarly higher (91.7% vs. 74.1%) although it did not achieve statistical significance. Hence, although ASCT is likely still preferred in chemotherapy-sensitive and fit patients, CAR-T therapy may serve as an effective consolidation strategy especially in transplant ineligible patients. First line CAR-T therapy consolidation in transplant ineligible patients is now being investigated (CAROLYN trial NCT07015242).

Bispecific T-cell engager (BITEs)

BITE therapies consist of engineered bispecific antibodies that bind to B cell surface antigens and immune effector T-cells, mainly CD20 and CD3 respectively for agents used in NHL, with monotherapy approvals granted across different NHLs mainly in the R/R setting based on pivotal phase 2 trials [169–172]. Recently, multiple phase 3 trials have also demonstrated safety and superiority of BITE treatment combinations compared to salvage chemotherapy [173, 174]. Although long term outcomes currently are limited compared to CAR-T therapies, their off-the-shelf characteristic and different toxicity profile especially with lower neurological complications serve as an attractive alternative for patients who require immediate therapies [175]. However, infective concerns remain especially due to the longer duration of treatment required and is the biggest cause of non-relapse mortality [176]. CNSL patients have generally been excluded from these prospective trials as there were concerns on BBB penetration due to the large sizes of these antibodies. However, multiple real-world studies evaluating CNSL with small numbers of PCNSL have been published that suggest these concerns are unfounded (summarised in Table 3) [177–181]. Godfrey et al. first demonstrated CSF penetration with glofitamab in R/R SCNSL and although concentrations were much lower than blood concentration, they were sufficient to drive T-cell activation and cytotoxicity [177]. In the largest series to date with a multicenter retrospective study of 84 patients with R/R CNSL, Xing et al. showed that 53 patients with R/R PCNSL treated with glofitamab, either as monotherapy or in combination with other systemic therapies, achieved an ORR of 92% and a CR rate of 68%, with a median PFS of 19.5 months and median OS not reached at a median follow-up of 14.7 months [181]. Notably, this was a heavily pretreated population: 60% of PCNSL patients had an ECOG performance status of ≥ 2, the median number of prior lines of therapy in the overall cohort was 2.5, and 81% had previously received a BTKi. Across the whole cohort, CRS occurred in 40% of patients (all grade 1–2) and ICANS in 8% (grade ≥ 3 in 4%), with no treatment-related deaths. Although 27% of patients proceeded to consolidation with CAR-T, ASCT or WBRT, which may have contributed to the durability of response, these data suggest that glofitamab is active and tolerable in R/R PCNSL. Prospective trials would be required to validate these findings and are ongoing (NCT06931652, NCT06922604 and NCT07082868).

Antibody drug conjugates (ADC)

ADCs have been incorporated mainly in DLBCL with polatuzumab vedotin, which contains an anti CD79b targeting antibody linked with monomethyl auristatin E as its payload, and loncastuximab tesirine which targets CD19 linked to its payload pyrrolobenzodiazepine [182, 183]. Similar to CAR-T and BITEs, CNS lymphomas were excluded from these registration trials. In a proof-of-concept study, Gao et al. demonstrated significant reduction in tumor burden and survival prolongation in PCNSL and SCNSL mouse models when treated with polatuzumab [184]. Encouraged by these findings, polatuzumab was subsequently tested in a case series of 3 patients with R/R CNSL at a dose of 1.8 mg/kg every 3 weeks in combinations including IT MTX/Ara-C, mosunetuzumab and ibrutinib separately. CSF samples were collected during treatment and polatuzumab was detected at clinically active levels. 2 patients achieved CR (SCNSL) and 1 had stable disease (SD) (summarised in Table 3). The brain penetrating abilities of ADCs mirror that seen in solid cancers and hence future trials incorporating ADCs in PCNSL should be considered [185]. The SOLAR (NCT06607549) trial [186] testing sequential stereotactic radiosurgery (SRS) followed by loncastuximab and rituximab is currently ongoing.

Oncolytic viruses and cancer vaccines

Oncolytic virus therapy is a rapidly evolving field of growing interest in solid cancers as well as haemolymphoid malignancies. Their mechanism of action is dependent on the specific viral vector and genetic engineering that increases specificity for tumor cell kill, immune stimulation and tumor microenvironment modification, often in combination with other systemic therapies [187]. Several oncolytic virus therapies have received approvals, including teserpaturev/G47delta (DELYTACT), approved in Japan for malignant glioma and delivered by intratumoral injection [188]. Currently, there have been no studies published specifically in PCNSL but preclinical studies in lymphoma models have been attempted. Li et al. demonstrated in implanted subcutaneous lymphoma mouse models that an engineered oncolytic vaccinia virus (OVV) expressing anti-CD47 nanobody targeting macrophage immune-checkpoint generated greater tumor reduction through higher immune activation with increased NK cell infiltration and M1 polarisation, compared to placebo and OVV without the CD47 gene fragment incorporated [189]. Greater macrophage phagocytosis of tumor cells was seen in those which secreted anti CD47 nanobody. Synergism with CAR-T was also tested with sequential administration of CD19 CAR-T cells demonstrating greater CAR-T cell infiltration into the tumor and higher reduction in tumor size compared to the CAR-T monotherapy group. Translation into clinical practice would require optimising virus delivery due to the challenges of both intratumoral and intraperitoneal delivery models currently.

Similarly, non-viral therapeutic cancer vaccine is another active area of investigation in lymphomas, with studies mainly conducted in non-CNS involved lymphomas thus far. Various vaccination strategies have been attempted. Autologous vaccination strategies include irradiated inactive lymphoma cells with vaccine adjuvant for immune stimulation, in-situ vaccination with radiotherapy and vaccine adjuvant and autologous dendritic cell vaccines. Allogeneic approaches include tumor-associated antigen-based, neoantigen based and idiotype therapeutic vaccines [190]. Overall these studies have failed to translate clinically in trials, and their role in PCNSL treatment remains to be elucidated [191].

The future of diagnostics in PCNSL

Cell-free DNA and cytokines

Brain tissue biopsy is currently the gold standard for PCNSL diagnosis, though technical challenges are common particularly for deep-seated lesions or in patients with poor performance status. Minimally-invasive liquid biopsy approaches using CSF have thus been evaluated as a promising adjunctive tool. Traditional CSF cytology may only have up to approximately 30% sensitivity and although flow cytometry increases this two- to three-fold, both are limited by disease burden, sample quantity and processing, prior corticosteroids and cytopathologist experience [28]. The MYD88 L265P mutation occurs in up to 70% of patients with PCNSL, including PVRL [11, 14, 192, 193]. IL10 and CXCL13 levels are often elevated in the CSF [194–196]. The PAMINA study evaluated combined CSF testing for MYD88 mutation and IL10 via circulating tumor DNA (ctDNA) PCR and ELISA, respectively, in 36 patients with newly diagnosed PCNSL [197]. MYD88 mutation testing showed a sensitivity of 72%, with an 82% concordance in paired tissue-CSF samples. IL10 was also significantly higher compared to controls with a sensitivity and specificity of 88% and 99% respectively. The combined MYD88/IL10 analysis achieved 94% sensitivity and 98% specificity for newly diagnosed PCNSL. Yamagishi et al. using digital PCR to detect MYD88 L265P mutation in newly diagnosed PCNSL yielded a sensitivity and specificity of 92.2% and 100% respectively with CSF cellular DNA [198]. Although plasma sampling of MYD88 mutations has also been evaluated, they generally demonstrate lower sensitivity with comparable specificity compared to CSF evaluation [28]. Similarly, for PVRL, MYD88 mutation testing can be complementary with a study by Miserocchi et al. showing 6 out of 8 cytologically proven PVRL had detectable mutation in the aqueous humor, with good consistency between aqueous and vitreous humor sampling [199]. In a recent prospective multicenter cohort study (PRICELUS), a triplet combination of MYD88, IL10 and CXCL13 was proposed as a potential diagnostic for PCNSL, though its utility real-world implementation remains to be validated [200]. This multiparametric approach illustrates a broader trend towards combining genomic and soluble biomarkers to improve diagnostic accuracy over single-marker assays.

Beyond diagnosis, minimal residual disease (MRD) monitoring by liquid biopsy has emerged as a useful tool to prognosticate and guide treatment decisions in systemic lymphomas [201, 202]. Sheng et al. prospectively evaluated the role of CSF ctDNA monitoring compared with routine PET/CT imaging in 22 patients with newly diagnosed PCNSL treated with orelabrutinib, rituximab and HD-MTX [203], demonstrating that clearance of ctDNA was associated with higher CR rates, longer PFS and OS compared to patients with positive ctDNA. Improved prognostic risk models incorporating ctDNA into traditional clinical models have also been developed into a molecular prognostic index for CNSL (MOP-C), incorporating the IELSG risk score with new factors such as presence of CR on MRI, baseline ctDNA levels and the log₁₀ change of ‘peripheral residual disease’ from baseline as assessed solely from plasma DNA [204].

DNA methylation profiling

DNA methylation via CpG nucleotides is an epigenetic mechanism to regulate gene expression, and abnormal methylation profiles can promote carcinogenesis through tumor suppressor gene silencing while hypomethylation can activate oncogenes and promote chromosomal instability [205]. Tumor specific DNA methylation profiles serve as a fingerprint that include both lineage and cancer specific signatures that can help to identify and classify cancer behaviour and prognosis better than morphology, including for PCNSL [206]. Previously, a comprehensive genome-wide DNA methylation-based CNS tumor classifier across all known entities was proposed, with potential application in routine diagnostic settings [207]. When compared with systemic DLBCL, PCNSL generally possesses a distinct DNA methylation profile characterized by hypermethylation of genes involved in cell adhesion and hypomethylation of genes involved in glycosylation [208]. Further distinction may be made based on the presence of EBV, with EBV-positive PCNSL characterized by global DNA promoter hypomethylation compared to its EBV-negative counterparts [48].

Downs et al. evaluated a pilot liquid biopsy study by identifying 8 methylation markers from the TCGA database unique to DLBCL, which were subsequently validated in GEO submissions containing PCNSL that showed good discrimination compared to 11 other CNS tumor types as well as with normal B cells, brain and blood components [209]. A tailed amplicon multiplexed-methylation-specific PCR was then developed using methylation markers SCG3 and cg054, selected for best amplification efficiency whilst retaining 100% accuracy in distinguishing PCNSL from other CNS tumor types. A pilot test done in the plasma of 15 newly diagnosed PCNSL patients showed 3/15 (20%) had higher methylation markers than normal individuals or other CNS tumors. Despite the low overall sensitivity using plasma samples, it illustrates the potential of liquid methylation profile testing in aiding diagnosis. Further studies in CSF testing would likely help improve its sensitivity in view of higher ctDNA levels compared to plasma [204]. Challenges remain that would need to be overcome for widespread adoption, including individual variability in methylation profiles even within the same cancer type, reproducibility across platforms due to the lack of standardised laboratory detection methods and clinical validation [205]. The feasibility of a minimally invasive CSF testing approach has been suggested by combining both proteomic and DNA methylation profiles in the CSF for distinguishing PCNSL from glioblastoma and brain metastases [210].

Exosomes

Exosomes are a form of extracellular vesicles with a lipid bilayer membrane produced via the endosomal pathway that are involved in cell-to-cell communication, containing metabolites, proteins and nucleic acid depending on their cellular origin [211]. Lu et al. investigated plasma exosomal miRNAs and found the expression levels of hsa-miR-148a-3p, hsa-let-7f-5p, hsa-miR-345-5p and hsa-miR-4433b-5p in patients with PCNSL were significantly higher compared to patients with glioblastoma (GBM), while immunohistochemistry (IHC) analysis of EGFR was significantly higher in GBM patients [212]. Combining 3 plasma exosomal miRNA (hsa-miR-148a-3p, hsa-miR-345-5p and hsa-miR-4433b-5p) into a panel achieved an area under receiver operating characteristic (ROC) curve of 0.791, demonstrating a possible use of plasma exosomal miRNA as a novel method for liquid biopsy. These exosomes may also serve a role in prognosticating and predicting high risk subgroups at risk of relapse that may benefit from individualised treatment in addition to standard treatment protocols. Lv et al. collected CSF samples from newly diagnosed PCNSL patients and categorised them into 3 separate groups: pretreatment, treatment (after 2–4 treatment cycles) and CR [213]. Higher exosomal protein concentration and exosomal miR-3681-3p expression at diagnosis led to significantly shorter PFS, while the former together with exosomal miR-101-3p expression were also significantly associated with earlier recurrence in patients who achieved CR. These studies require larger cohort validation and reproducibility across different testing platforms may face challenges; however they represent a potential molecular mechanism in diagnosing and prognosticating PCNSL better.

Single cell sequencing and spatial transcriptomics

Single cell RNA-seq (scRNA-seq) platforms enable transcriptomic profiling of individual tumor cells, allowing the profiling tumor cell heterogeneity either in a high throughput manner, or captured in isolation, to facilitate diagnosis and classification [214]. In addition, information on the spatial organization of the various cellular constituents may be delineated through one of several spatial transcriptomic technologies, to better understand the tumor stromal context and immune microenvironment [215]. Earlier studies have demonstrated the feasibility of utilizing a single cell sequencing approach for the diagnosis of PVRL by detecting MYD88 mutations and also suggested its potential in identifying PCNSL through CSF studies [216, 217]. Heming et al. combined flow cytometry, scRNA-seq and spatial transcriptomics – revealing distinct spatial patterns of malignant B cells. Interesting, hyperexpanded B cell clones were shared between tissue and CSF, but not in the blood. An immunosuppressive tumor immune microenvironment dominated by expression of several immune checkpoint molecules was further demonstrated, supporting the incorporation novel therapies such as antibody-drug conjugates, bispecific T-cell engagers, and CAR-T therapies [218]. Subsequent work further suggests that the tumor microenvironment may be remodelled by specific tumor cells – including a FKBP5-expressing tumor subgroup presumptive contributing to a barrier or immune-cold environment [219]. In other recent studies, scRNA-seq and spatial profiling identified subpopulations of plasmablast-like tumor cells distributed within PCNSL contributing to worse prognosis, extending our biological understanding of PCNSL further [44, 220]. Practical challenges in applying single cell and spatial transcriptomic technologies in the clinical setting include sampling constraints in obtaining adequate tumor cells and tissue material (e.g. from stereotactic brain biopsy). A validated framework, similar to one used for diffuse glioma, to integrate these molecular findings with clinical imaging and standardized tissue-sampling approaches may be necessary in order to apply such technologies in the clinic [221].

Conclusions

Today, invasive tissue-based biopsy remains the gold standard for PCNSL diagnosis – however, as we have discussed in our current review, the future diagnostic pathway will likely rely more upon minimally-invasive liquid biopsy-based approaches that incorporate several advanced technologies. However, as would be expected, the search for a reproducible biomarker for clinical application would require further investigation and validation. In addition, routine clinical implementation would be difficult unless technological barriers and cost limitations can be circumvented. Together with improved understanding of the molecular diversity and immune microenvironmental architecture of PCNSL, we anticipate an integrated approach towards multi-modal information capture for a comprehensive strategy for tumor classification and therapeutic selection. Future research directions may yet include multi-dimensional investigations of PCNSL consisting of clinical and radiological features, as well as tissue- and liquid-based molecular biomarkers, augmented by artificial intelligence and machine learning tools.

Abbreviations

ABC

Activated B-Cell

ADC

Antibody Drug Conjugates

ASCT

Autologous Stem Cell Transplant

BCR

B-cell Receptors

BITE

Bispecific T-cell Engager

BTKi

Bruton Tyrosine Kinase Inhibitors

CAR

Chimeric Antigen Receptor

CR

Complete Response

CRS

Cytokine Release Syndrome

CSF

Cerebrospinal Fluid

ctDNA

Circulating Tumor DNA

DLBCL

Diffuse Large B Cell Lymphomas

EBV

Epstein-Barr virus

EFS

Event-Free Survival

FL

Follicular Lymphomas

GCB

Germinal center B cell

GEO

Gene Expression Omnibus

HD-MTX

High Dose Methotrexate

HLA

Human Leukocyte Antigen

ICANS

Immune Effector Cell-Associated Neurotoxicity Syndrome

ICI

Immune Checkpoint Inhibitors

ID-PCNSL

Immunodeficiency-associated PCNSL

IHC

Immunohistochemistry

IMiD

Immunomodulatory Drugs

IPCG

International PCNSL Collaboration Group

IRAE

Immune-Related Adverse Events

ITAM

Tyrosine-Based Activation Motif

MBVP

Methotrexate, Carmustine, Teniposide, Methylprednisolone

MOP-C

Molecular Prognostic Index for CNSL

MRD

Minimal Residual Disease

NHL

Non-Hodgkin’s Lymphoma

ORR

Objective Response Rates

OVV

Oncolytic Vaccinia Virus

PCNSL

Primary Central Nervous System Lymphoma

PD-1

Programmed Cell Death Protein 1

PD-L1/PD-L2

Programmed Death-Ligand 1/2

PFS

Progression-Free Survival

PVRL

Primary Vitreoretinal Lymphoma

QOL

Quality of Life

R-DeVIC

Rituximab, Dexamethasone, Etoposide, Ifosfamide, Carboplatin

R-MBVP

Rituximab, Methotrexate, Carmustine, Teniposide, Methylprednisolone

R-MPV

Rituximab, Methotrexate, Procarbazine, Vincristine

RMT

Rituximab, Methotrexate, Temozolomide

R/R

Relapsed/Refractory

ROC

Receiver Operating Characteristic

SCNSL

Secondary CNSL

scRNA-seq

Single Cell RNA-seq

SINE

Selective Inhibitor of Nuclear Export

SRS

Stereotactic Radiosurgery

TIAN

Tumor Inflammation-associated Neurotoxicity

TIR

Toll/Interleukin-1 Receptor

TLR

Toll-Like Receptors

WBRT

Whole Brain Radiotherapy

WHO

World Health Organization

Author contributions

Y.C.L. prepared the first draft of the manuscript; Y.C.L. and J.Y.C. conceptualized, revised and approved the final manuscript.

Funding

This work was supported by the Singapore Ministry of Health’s National Medical Research Council Research Transition Award (TA21jun-0005), Clinician Scientist Individual Research Grant (CIRG25jan-0007), Large Collaborative Grant (OFLCG23May-0039), TETRAD II Collaborative Centre Grant (CG21APR2002), SingHealth Duke-NUS AM/ACP-Designated Philanthropic Fund Grant Award (08/FY2023/EX/27-A65, 2025/AMS/000033/A, 2026/EX/000063, 2026/EX/000160/08, 2025/AMS/000046), Tanoto Foundation, NCCS Cancer fund, as well as the Khoo Bridge Funding Award provided by Duke-NUS Medical School and the “Estate of Tan Sri Khoo Teck Puat” (Duke-NUS-KBrFA/2025/0090).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.