Brain‐Targeting siRNA Delivery to Tackle Alzheimer's Disease

ABSTRACT

Alzheimer's disease (AD) is a progressive neurodegenerative disorder that places an increasing burden on patients, caregivers, and healthcare systems worldwide. Current disease‐modifying therapies (DMTs) are limited by high costs, complex administration, and reliance on advanced biomarker infrastructure, highlighting the shortcomings of existing treatment paradigms. These limitations have sparked growing interest in gene‐ and nucleic acid–based interventions as upstream strategies to modify AD pathogenesis. Among these, small interfering RNA (siRNA) is especially compelling because it can be rationally programmed, directed at multiple molecular pathways, and paired with rapidly evolving delivery technologies. However, the clinical translation of siRNA therapies for AD is still constrained by challenges in brain‐targeted delivery, safety, and sustained efficacy. In this review, we summarize current concepts in AD pathology, highlight recent clinical and translational advances, and critically assess emerging brain‐targeted siRNA delivery platforms and their key bottlenecks. Within a precision‐medicine framework, brain‐targeted siRNA offers the possibility of aligning patient selection, molecular targets, and delivery strategies with biomarker‐defined AD endotypes. We discuss both the therapeutic promise and the realistic limitations of siRNA‐based approaches for AD, outline priorities for future development, and identify key gaps that must be addressed to enable meaningful clinical implementation.

To advance precision siRNA therapy for Alzheimer's disease, the integration of nanomedicine with artificial intelligence accelerate to safely and efficiently traverse the blood–brain barrier and enabling smart, patient‐stratified delivery systems that accelerate clinical translation.

Abbreviations

5 × FAD

transgenic mouse model carrying five familial Alzheimer's disease mutations

AAIC

Alzheimer's Association International Conference

AAL

Aleuria aurantia lectin

AD

Alzheimer's disease

ADNI

Alzheimer's disease neuroimaging initiative

AGO2

Argonaute 2

AI

artificial intelligence

AICD

APP intracellular domain

ALN‐APP

Alnylam pharmaceuticals APP‐targeting RNA interference

AMT

adsorption‐mediated transcytosis

Ang

Angiopep‐2

ApoE

apolipoprotein E (protein)

APOE

apolipoprotein E gene

APOE4

apolipoprotein E4

APOEε4

apolipoprotein E epsilon 4

APP

amyloid precursor protein

APP/PS1

double‐transgenic Alzheimer's disease mouse model expressing human APP and PS1 mutations

AQP4

aquaporin‐4

ARIA

amyloid‐related imaging abnormalities

ARIA‐E

amyloid‐related imaging abnormalities—edema

ASO

antisense oligonucleotides

AT(N)

amyloid/tau/neurodegeneration biomarker framework

AuNPs

gold nanoparticles

Aβ

amyloid‐β peptide

Aβo

Aβ oligomers

BACE1

β‐site APP‐cleaving enzyme 1 (β‐secretase 1)

BBB

blood–brain barrier

BBBO

BBB opening

BBR

berberine

BIN1

bridging integrator 1

BMSCs

bone marrow‐derived mesenchymal stem cells

C16‐siRNA

C16 (hexadecyl) conjugated small interfering RNA

C1q

complement component 1q

C3

complement component 3

CAA

cerebral amyloid angiopathy

CARPA

complement activation‐related pseudoallergy

CD33

cluster of differentiation 33

CD98hc

cluster of differentiation 98 heavy chain

CMB

cerebral microbleeds

CMC

chemistry, manufacturing and controls

CNS

central nervous system

CPP

cell‐penetrating peptides

CR1

complement receptor 1

CRISPR

clustered regularly interspaced short palindromic repeats

CSF

cerebrospinal fluid

cSS

cortical superficial siderosis

CTF

C‐terminal fragment

DAM

disease‐associated microglial state

DMTs

disease‐modifying therapies

DoE

experimental design

DOPE

1, 2‐Dioleoyl‐sn‐glycero‐3‐phosphoethanolamine

DSPC

distearoylphosphatidylcholine

ED50

effective dose for 50% of the population

ELS

endocytic‐lysosomal system

EV

extracellular vesicle

FUS

focused ultrasound

GBM

glioblastoma

GFAP

glial fibrillary acidic protein

GLUT1

glucose transporter 1

GSDMD

gasdermin D

HDAC3

histone deacetylase 3

H&E

Hematoxylin and Eosin

hMSCs

human bone marrow mesenchymal stem cells

ICV

intracerebroventricular

IL‐1α

interleukin‐1 alpha

IT

intrathecal

IVT

intravitreal

kDa

kilodalton

KEAP1

Kelch‐like ECH‐associated protein 1

KLVFF

Lys‐Leu‐Val‐Phe‐Phe peptide (Aβ‐binding pentapeptide)

LDLR

low‐density lipoprotein receptor

LLPS

liquid–liquid phase separation

LNPs

lipid nanoparticles

LRP1

low‐density lipoprotein receptor‐related protein 1

LTP

long‐term potentiation

MAPT

microtubule‐associated protein tau (gene)

MC3

DLin‐MC3‐DMA ionizable cationic lipid

miR‐124

microRNA‐124

miR‐193b‐3p

microRNA‐193b‐3p

MOR

μ‐opioid receptor

MRgFUS

magnetic resonance‐guided focused ultrasound

MRI

magnetic resonance imaging

mRNA

messenger RNA

MS4A

membrane spanning 4 domains A

NBDD

nose‐to‐brain drug delivery

NfL

neurofilament light chain

NFTs

neurofibrillary tangles

NF‐κB

nuclear factor‐kappa B

NHP

non‐human primates

NIA‐AA

National Institute on Aging–Alzheimer's Association

NLRP3

NLR family pyrin domain containing 3

NREM

non‐rapid eye movement

NtB

nose‐to‐brain

NVU

neurovascular unit

OTV

oligonucleotide transport vehicle

P301S

tau transgenic mouse line carrying the P301S mutation

PAMAM

poly(amidoamine) dendrimer

PBAE

poly(β‐amino ester)

PCD

passive cavitation detection

PD

pharmacodynamics

PEG

polyethylene glycol

PEI

polyethylenimine

PHFs

paired helical filaments

PiB‐PET

Pittsburgh compound b positron emission tomography

PICALM

phosphatidylinositol binding clathrin assembly protein

PLCG2

phospholipase C gamma 2

PLGA

poly(lactic‐co‐glycolic acid)

PSEN1

presenilin 1

PSEN2

presenilin 2

PTMs

post‐translational modifications

RAGE

receptor for advanced glycation end‐products

RISC

RNA‐induced silencing complex

RMT

receptor‐mediated transcytosis

RNAi

RNA interference

ROS

reactive oxygen species

RVG

rabies virus glycoprotein

SAH

subarachnoid hemorrhage

sAPPα

soluble amyloid precursor protein alpha

sAPPβ

soluble amyloid precursor protein beta

SFs

straight filaments

SIR

sterile inflammation

siRNA

small interfering RNA

SLN

solid lipid nanoparticles

SORT

selective organ targeting

SPIONs

superparamagnetic iron oxide nanoparticles

STING

stimulator of interferon genes

SWI

susceptibility‐weighted imaging

Tau‐PET

tau positron emission tomography

TFEB

transcription factor EB

TfR

transferrin receptor

TMD

transmembrane domain

TNF

tumor necrosis factor

TNF‐α

tumor necrosis factor‐alpha

TREM2

triggering receptor expressed on myeloid cells 2

UPR

unfolded protein response

WHO

world health organization

ZnO

zinc oxide nanoparticles

α‐Syn

alpha‐synuclein

1. Introduction

Alzheimer’s disease (AD) is the most common cause of dementia and imposes an increasing burden on patients, caregivers, and healthcare systems as populations age (Alzheimer’s Association ²⁰²⁵). Clinically, AD is characterized by its insidious onset, with a progressive decline in episodic memory and executive function, eventually eroding the ability to live independently (Jack et al. 2018; McKhann et al. ²⁰¹¹). The economic burden is increasing; in the United States, total payments for health care, long‐term care, and hospice services for people aged 65 years and older with Alzheimer’s or other dementias were estimated at $384 billion in 2025 (Alzheimer’s Association ²⁰²⁵).

Since the formulation of the amyloid cascade hypothesis in 1992, research on AD has placed major emphasis on amyloid‐β (Aβ) and its precursor, the amyloid precursor protein (APP) (Hardy and Higgins 1992). This hypothesis proposes that the accumulation of Aβ is an upstream event that triggers or amplifies downstream pathologies, including synaptic dysfunction, neuroinflammation, tau aggregation, and neurodegeneration (Selkoe 2001). However, it is now clear that AD pathophysiology is multifactorial. In addition to abnormalities in Aβ and tau, chronic innate immune activation, disruption of protein homeostasis, and failure of cellular and systemic clearance mechanisms all contribute to disease onset and progression (Congdon and Sigurdsson 2018).

In parallel with these mechanistic insights, gene‐ and nucleic acid–based therapies have emerged as promising upstream interventions for AD. Among them, small interfering RNA (siRNA) has attracted particular interest because it can selectively silence key genes such as APP/BACE1 and MAPT (tau), thereby potentially interrupting pathogenic cascades at their molecular origin (Ghaffari et al. 2020; Liang et al. ²⁰²⁴; Yang, Liu, et al. ²⁰²⁴). The ability to rationally program siRNA sequences and to modulate multiple molecular pathways positions siRNA as a flexible therapeutic modality (Jan et al. 2025). Nonetheless, efficient and safe delivery of siRNA to the brain—especially crossing the blood–brain barrier (BBB)—remains a principal bottleneck for clinical translation (Liu, Xia, et al. ²⁰²⁴). Recent progress in receptor‐mediated transport technologies (such as transferrin receptor–mediated delivery) and lipid nanoparticles (LNPs) has provided new options for brain‐directed siRNA delivery (El Moukhtari et al. 2023; Wei et al. ²⁰¹⁶).

Recent WIREs reviews provide useful context for the design of AD nanomedicines. Zhang et al. described functional nanoassemblies that integrate diagnostic and therapeutic components, multivalent targeting, and endogenous or exogenous stimulus responsiveness to address the multifactorial Aβ, tau, and inflammatory microenvironment of AD (Zhang, Sun, et al. ²⁰²¹). Complementing this theranostic perspective, Parekh et al. reviewed molecular MRI nanoprobes in which high payloads of contrast agents and targeting ligands can amplify signals from AD‐relevant targets, including Aβ, tau, and microglial activation; their analysis also underscores the importance of BBB transport, specificity, and safety for translation (Parekh et al. 2024). Building on these insights, the present review specifically focuses on brain‐targeted siRNA therapeutics, emphasizing BBB delivery barriers, CNS transport mechanisms, and translational design principles. We also examine the intersection between siRNA technology, AD biology, and precision medicine. We first outline the biological heterogeneity of AD using the AT(N) framework and discuss major therapeutic targets—including APP, BACE1, MAPT (tau), APOE, and regulators of neuroinflammation—in the context of precision intervention, rather than as isolated pathways. We then summarize emerging brain‐targeted delivery systems, focusing on their mechanisms, efficacy, and suitability for long‐term, personalized therapy. Finally, we explore how artificial intelligence (AI) can accelerate the design and optimization of these complex therapeutics and propose a strategic roadmap for their clinical translation.

2. Milestones in AD Research Development

The modern history of AD research has progressed from pathological descriptions to genetic discoveries, biomarker developments, disease‐modifying therapies, and, most recently, upstream molecular interventions. In 1906, Alois Alzheimer first described the characteristic amyloid plaques and neurofibrillary tangles in the brain of a patient with progressive cognitive decline. The term “Alzheimer’s disease” was subsequently coined by Emil Kraepelin in the 8th edition (1910) of his psychiatry textbook (Hippius and Neundörfer 2003). By the mid‐1980s, it had been established that amyloid plaques are composed of amyloid‐β (Aβ), whereas neurofibrillary tangles consist mainly of abnormally phosphorylated tau (Glenner and Wong 1984; Goedert ¹⁹⁹³). Between 1991 and 1995, pathogenic mutations in genes encoding amyloid precursor protein (APP) and presenilins 1 and 2 (PSEN1, PSEN2) were identified in familial AD, providing strong genetic support for an Aβ‐centered mechanism (Goate et al. 1991; Sherrington et al. ¹⁹⁹⁵). In 1993, the apolipoprotein E ε4 (APOE ε4) allele was recognized as the strongest genetic risk factor for late‐onset AD, further reinforcing a conceptual framework focused on abnormal Aβ generation and clearance (Corder et al. 1993).

In the early 21st century, therapeutic strategies for AD began to shift from purely symptomatic treatment toward disease‐modifying approaches. Between 1993 and 2003, the approval of cholinesterase inhibitors and memantine established the first generation of symptomatic pharmacotherapies (Birks et al. 1996; Robinson and Keating ²⁰⁰⁶; Tariot et al. ²⁰⁰⁴). An early attempt at active Aβ immunization (AN1792) was halted in 2002 because of safety concerns, redirecting efforts toward passive immunotherapy (Pasinetti et al. 2002). In 2004, the introduction of Pittsburgh compound B positron emission tomography (PiB‐PET) enabled in vivo visualization of Aβ deposition (Klunk et al. 2004), and the launch of the Alzheimer's Disease Neuroimaging Initiative (ADNI) provided a framework for standardizing imaging and fluid biomarkers (Mueller et al. 2005). In 2011, the NIA–AA incorporated biomarkers into the diagnostic criteria for AD (Albert et al. 2011). In 2018, the AT(N) research framework formally defined AD as a biological construct based on amyloid, tau, and neurodegeneration (Knopman et al. 2018). By 2020, approval of Tauvid marked the first clinical application of tau PET imaging (Jie et al. 2021), while the rapid progress in blood‐based biomarkers, such as plasma p‐tau181/217, ushered in a new era of minimally invasive AD diagnostics (Triana‐Baltzer et al. ^{2020, 2021}).

In parallel, disease‐modifying therapies have advanced from experimental concepts to clinical reality. The anti‐Aβ monoclonal antibody aducanumab received accelerated approval in 2021 (U.S. Food and Drug Administration [FDA] ²⁰²¹) but was withdrawn in 2024; lecanemab received full approval in 2023, based on clear clinical benefits. In 2024, donanemab was also approved, further consolidating anti‐Aβ antibodies as a standard option for early AD (DeMattos et al. ²⁰¹²; Sims, Zimmer, Evans, et al. ²⁰²³). At the same time, upstream molecular interventions and delivery engineering have begun to emerge. The APP‐targeting siRNA ALN‐APP, delivered intrathecally, produced robust reductions in soluble APPα/β and Aβ40/42 in cerebrospinal fluid (CSF) after a single dose in humans (Cohen et al. 2023). Receptor‐mediated delivery strategies using targets such as transferrin receptor (TfR) and CD98hc have successfully transported oligonucleotides into the brains of non‐human primates, and early human studies are underway (Barker et al. 2024; Chew et al. ²⁰²³; Wells et al. ²⁰²⁵). Together, these advances outline a plausible path toward intravenous administration of siRNA and rational multi‐target combinations (e.g., APP/BACE1 and MAPT) for AD.

Taken together, AD research has evolved from foundational pathological and genetic discoveries to sophisticated biomarker systems and clinically approved disease‐modifying antibodies. The field is now moving toward the integration of upstream molecular interventions with engineered BBB‐crossing delivery platforms. Key milestones are summarized in Figure 1.

FIGURE 1.

Milestones in Alzheimer's disease research and development. Timeline of representative advances across pathology, genetics, biomarkers/imaging, and therapeutic/delivery development. AD, Alzheimer's disease; APP, amyloid precursor protein; AT(N), amyloid/tau/neurodegeneration biomarker framework; Aβ, amyloid‐β; BBB, blood–brain barrier; ChEIs, cholinesterase inhibitors; fAD, familial Alzheimer's disease; LOAD, late‐onset Alzheimer's disease; NIA‐AA, National Institute on Aging–Alzheimer's Association; PiB‐PET, Pittsburgh compound B positron emission tomography; PSEN1/2, presenilin 1/2; RMT, receptor‐mediated transcytosis; tau‐PET, tau positron emission tomography.

3. Pathology of AD

AD is a neurodegenerative disorder caused by the interplay of multiple pathological mechanisms, primarily characterized by memory loss and cognitive decline. Key pathological features of AD include the accumulation of Aβ, hyperphosphorylation of tau protein, and neuroinflammation. Aβ forms amyloid plaques, which directly damage neuronal structures and function. Abnormal tau aggregation leads to neurofibrillary tangles within neurons, disrupting normal intracellular signaling. At the same time, the activation of innate immune responses and microglia exacerbates neuronal damage, further driving disease progression. BBB dysfunction is also increasingly recognized as an important component of AD pathology. Disruption of neurovascular homeostasis can impair vascular exchange and Aβ clearance, while endothelial injury and increased barrier permeability promote neuroinflammation, thereby contributing to disease progression. An overview of these interacting pathological processes is summarized in Figure 2.

FIGURE 2.

Pathology of AD. Schematic overview of interconnected processes driving AD pathology across neurons, glia, the neurovascular unit, and CNS clearance routes. APP, amyloid precursor protein; Aβ, amyloid‐β; BACE1, β‐site APP‐cleaving enzyme 1; BBB, blood–brain barrier; CSF, cerebrospinal fluid; CSF–ISF, cerebrospinal fluid–interstitial fluid; IL‐1β, interleukin‐1 beta; ISF, interstitial fluid; NFTs, neurofibrillary tangles; NVU, neurovascular unit; PHFs, paired helical filaments; PrPᶜ, cellular prion protein; ROS, reactive oxygen species; sAPPα, soluble amyloid precursor protein alpha; sAPPβ, soluble amyloid precursor protein beta; SFs, straight filaments; TNF‐α, tumor necrosis factor alpha.

3.1. Aβ/APP

The amyloid‐beta (Aβ) hypothesis has long been central to AD research. Aβ is generated from its precursor protein, amyloid precursor protein (APP), through the cleavage by β‐secretase (BACE1) and γ‐secretase. This cleavage pathway leads to the formation of Aβ peptides, which aggregate into plaques that disrupt neuronal signaling. Although Aβ accumulation is one of the earliest pathological events, it is still debated whether Aβ alone can fully explain AD. Genetic studies and clinical data from anti‐amyloid therapies support the role of Aβ in initiating disease progression (Bellenguez et al. 2022; Kunkle et al. ²⁰¹⁹). APP processing occurs via two primary pathways: the non‐amyloidogenic pathway, which produces neuroprotective soluble APPα (sAPPα), and the amyloidogenic pathway, APP is first cleaved by BACE1 to generate soluble APPβ (sAPPβ) and the membrane‐bound fragment, which is subsequently cleaved by γ‐secretase to produce Aβ peptides. The balance between these pathways influences Aβ production, and dysregulation can lead to amyloid plaque formation, contributing to AD pathology (Chen et al. 2017; Kepp et al. ²⁰²³; O'brien and Wong ²⁰¹¹). Aβ accumulation begins with the formation of soluble oligomers, followed by fibrillation and plaque deposition (Long and Holtzman 2019; Walker ²⁰²⁰). Aβ oligomers (Aβo) bind to receptors like the cellular prion protein (PrP^C), initiating signaling pathways that disrupt synaptic plasticity, enhance tau pathology, and impair neuronal function (Lacor et al. 2004; Larson and Lesné ²⁰¹²; Tu et al. ²⁰¹⁴). Targeting specific Aβ species, such as N‐terminally pyroglutamated Aβ, has shown promise in clinical trials, demonstrating significant plaque clearance and slowing of cognitive decline in early AD (Bellenguez et al. 2022).

3.2. Tau

Tau is a microtubule‐associated protein whose hyperphosphorylation and aggregation into neurofibrillary tangles are hallmark features of AD. Tau's primary function is to stabilize microtubules, but in AD, tau becomes hyperphosphorylated, leading to its aggregation and disruption of microtubule stability. Tau aggregates form paired helical filaments (PHFs) and straight filaments (SFs), contributing to neuronal dysfunction and cognitive decline (Wang and Mandelkow 2016). In AD, tau undergoes various post‐translational modifications (PTMs) such as acetylation, truncation, and ubiquitination, which promote its aggregation and impair cellular function (Amadoro et al. 2020; Liu et al. ²⁰¹⁶). The relationship between Aβ and tau is complex, with Aβ‐induced tau pathology contributing to the neurodegenerative process. Recent studies suggest that tau pathology correlates more strongly with cognitive decline than Aβ plaque burden (Harris et al. 2025). Tau pathology and Aβ accumulation interact in a positive feedback loop, amplifying neuronal damage. The disruption of tau's normal function further exacerbates Aβ‐induced toxicity (Canepa and Fossati 2021; Sanchez‐Rodriguez et al. ²⁰²⁴; Zhang, Wei, et al. ²⁰²¹). Therefore, therapeutic strategies targeting both tau and Aβ may offer a more comprehensive approach to AD treatment.

3.3. Innate Immunity and Glia

Recent studies have revealed that innate immunity, particularly microglial and astrocytic activation, plays a crucial role in AD pathology. Microglia, the brain's primary immune cells, exhibit heterogeneous and context‐dependent reactive states in response to Aβ and tau, contributing to neuroinflammation and exacerbating neuronal damage (Cardinali et al. 2025; Kim et al. ²⁰²⁴; Leng and Edison ²⁰²¹). These reactive states can have either protective or detrimental effects depending on the disease stage and local microenvironment. The NLRP3 inflammasome is one of the most studied inflammatory complexes in AD. Aβ, tau, and oxidative stress activate NLRP3, leading to the release of pro‐inflammatory cytokines such as IL‐1β and IL‐18, which further enhance neuroinflammation and tau pathology (Paesmans et al. 2024; Szabo et al. ²⁰²⁵). Recent studies suggest that inhibiting NLRP3 activation can reduce microglial activation, improve cognitive function, and attenuate tau pathology in animal models (Barczuk et al. 2022; Ising et al. ²⁰¹⁹; Zhang et al. ²⁰²⁴). Astrocytes also exhibit heterogeneous and context‐dependent reactive states in AD rather than undergoing a uniform transition to a neurotoxic phenotype. Microglial cytokines, including IL‐1α and TNF‐α, can modulate astrocytic reactivity and contribute to synaptic damage (Liu et al. 2020; Santiago‐Balmaseda et al. ²⁰²⁴). Targeting glial dysfunction and restoring normal immune responses may represent a novel therapeutic strategy for AD.

3.4. Neurovascular Unit and Cerebral Amyloid Angiopathy

The neurovascular unit (NVU) is composed of brain endothelial cells, tight junctions, pericytes, the basement membrane, AQP4‐enriched astrocytic endfeet, as well as neurons and microglial cells (Kerkhofs et al. 2021). It serves as the fundamental unit responsible for maintaining brain homeostasis, metabolic exchange, and barrier function (Yue et al. 2026).

In AD, the disruption of the BBB and imbalance of the neurovascular unit have been repeatedly observed, interacting with various pathological processes such as impaired Aβ clearance, amplified inflammatory responses, and reduced cerebral blood flow (Kelly et al. 2019). This phenomenon may begin even in the preclinical stages of dementia. Cerebral amyloid angiopathy (CAA) refers to the deposition of Aβ, primarily Aβ40, particularly in the walls of cortical arterioles and capillaries (Zhang et al. 2023). CAA is one of the leading causes of age‐related brain hemorrhages, cerebral microbleeds (CMB), and superficial cortical siderosis (cSS), and it can independently contribute to cognitive impairment (Sharma et al. 2022). Studies have shown that BBB dysfunction in AD leads to the accumulation of Aβ plaques and other toxic substances in the brain. Clinical trials targeting Aβ with monoclonal antibodies, such as lecanemab and donanemab, have demonstrated plaque clearance and slowed cognitive decline in early AD, offering hope for future treatments (Rashad et al. 2022; Yang ²⁰²⁵).

3.5. Endocytosis to Lysosome and Protein Homeostasis

The endocytic‐lysosomal system (ELS) plays a vital role in clearing cellular waste, including Aβ and tau. Dysfunction of this system is a key factor in AD progression, as it impairs the degradation of toxic proteins and promotes their accumulation (Kim et al. 2025). In AD, lysosomal dysfunction leads to the misdelivery of cargo such as APP and Aβ, which further exacerbates the pathological process (Colacurcio et al. 2018). Lysosomal acidification and hydrolytic imbalance contribute to protein aggregation and cellular degeneration (Peric and Annaert 2015). Targeting lysosomal function and enhancing protein degradation pathways could provide new therapeutic avenues for AD treatment (Mustaly‐Kalimi et al. ²⁰²²; Zhang et al. ²⁰²²).

3.6. Clearance Systems and Metabolism

AD progression is not only driven by the increased generation of toxic proteins but also by the failure of clearance systems. The BBB plays a crucial role in the efflux of Aβ from the brain. Dysfunction of this pathway, combined with aging and APOE genotype, increases the risk of AD (Tarasoff‐Conway et al. ²⁰¹⁵). The glymphatic system and meningeal lymphatic pathways have been implicated in CNS fluid and waste clearance, and their dysfunction has been associated with AD (Yamada and Iwatsubo 2024). Impaired CSF turnover in aging and neurodegenerative diseases contributes to the accumulation of Aβ plaques and tau tangles (Uchida 2022). Enhancing CSF flow and improving clearance mechanisms may provide novel therapeutic strategies for AD (Lv et al. 2021).

Although the pathological mechanisms of AD have been extensively studied, and our understanding of processes such as Aβ accumulation, Tau hyperphosphorylation, and neuroinflammation continues to evolve, current treatment options remain limited. Traditional symptomatic therapies, including acetylcholinesterase inhibitors and NMDA receptor antagonists, mainly provide symptomatic relief and do not substantially modify the underlying disease process (Mendiola‐Precoma et al. ²⁰¹⁶). However, these methods only provide temporary relief and do not significantly impede the core pathological progression of AD (Reddi Sree et al. 2025). As our understanding of AD's pathological mechanisms deepens, researchers have increasingly recognized that symptom alleviation alone is insufficient to combat this complex neurodegenerative disease. Consequently, there has been a growing emphasis on developing novel therapeutic strategies aimed at directly addressing the underlying causes of AD.

Against this backdrop, gene‐ and RNA‐based therapies have emerged as promising upstream interventions. siRNA, in particular, offers a programmable and pathway‐selective modality that can directly reduce the production of disease‐driving factors such as APP/BACE1 and MMAPT, or modulate key nodes in immune, proteostasis, and clearance networks. siRNA therefore represents a precision molecular tool that is, in principle, well aligned with the multi‐target, pathway‐level nature of AD pathophysiology. However, the same clearance and barrier systems that shape AD progression—especially the BBB, NVU, and glymphatic/CSF pathways—also pose major challenges for clinical translation of siRNA‐based therapies by limiting brain access and influencing distribution and safety profiles. In the following sections, we build on the pathophysiological framework outlined above to discuss candidate siRNA targets in AD and examine emerging brain‐targeted delivery strategies designed to overcome these barriers and enable effective, safe siRNA interventions in the human brain.

4. Working Mechanism of siRNA Therapeutics

siRNAs are short, double‐stranded, non‐coding RNAs of approximately 21–23 nucleotides in length that mediate sequence‐specific post‐transcriptional gene silencing. In the cytoplasm, mature siRNA duplexes are loaded into Argonaute‐containing RNA‐induced silencing complexes (RISC), where the passenger strand is removed and the antisense (guide) strand is retained for target recognition. (Wang et al. 2009). Guided by base pairing, the siRNA–RISC complex binds to complementary sequences in target mRNAs and AGO2 cleaves the target mRNA at the phosphodiester bond between nucleotides paired with guide positions 10 and 11, followed by exonucleolytic degradation of the mRNA. Because AGO2 acts catalytically, a limited number of RISC complexes can degrade multiple copies of the same transcript. Appropriately designed and modified siRNAs can sustain gene silencing for weeks to months in vivo, depending on tissue context and chemistry (Bernard et al. 2015; Meister ²⁰¹³). The core RNAi mechanism and its conceptual contrast with small‐molecule inhibition are illustrated in Figure 3.

FIGURE 3.

siRNA‐mediated gene silencing and the therapeutic rationale versus small‐molecule drugs. This diagram illustrates (A) the process of gene silencing mediated by siRNA through the RISC and (B) a comparison with traditional small‐molecule drugs. AGO2, Argonaute 2; Ds‐siRNA, double‐stranded small interfering RNA; RISC, RNA‐induced silencing complex.

The nucleotide sequence of an siRNA determines its target specificity, conferring a high degree of programmability and design flexibility. Once a suitable target site is identified on the mRNA, siRNAs can be designed to achieve predictable, cleavage‐based silencing via the canonical RISC/AGO2 pathway. This is particularly attractive for proteins such as APP and tau that engage in extensive protein–protein interactions and lack well‐defined, stable binding pockets for small molecules. Conventional small‐molecule inhibitors or antibodies may only partially modulate their function (Chandradoss et al. 2014). By acting at the mRNA level, siRNAs can bypass the requirement for “druggable” protein domains, extending the therapeutic target space to genes and pathways that are otherwise difficult to manipulate pharmacologically.

The enzymatic cycle of RISC and the slow‐release property of tissue accumulation allow many siRNA drugs to be administered infrequently, typically over a period of several months to half a year. This dosing rhythm has significant practical implications for long‐term treatment of chronic diseases, especially in enhancing patient compliance and treatment accessibility (Nair et al. 2017). With the development of standardized oligonucleotide chemistry and modular delivery platforms, the manufacturing pathways for siRNA have become clearer and more reproducible. In summary, the core value of siRNA lies in its programmable sequence, verifiable mechanism, reusable platform, and long‐lasting pharmacological effect. It opens up new therapeutic avenues for many traditionally difficult‐to‐target proteins while offering a low‐frequency dosing strategy that meets the long‐term treatment needs in chronic disease management.

5. Applications of siRNA Delivery Systems in AD Treatment

While a diverse array of delivery vectors has been explored for AD therapy, their translational maturity varies significantly. Current delivery platforms can be broadly categorized into inorganic nanomaterials, lipid‐based systems (LNPs), polymeric/hybrid nanoparticles, and biological vesicles (e.g., exosomes). Critically, there is a dichotomy between academic exploration and clinical implementation. Inorganic materials offer superior stability and tunability but face significant hurdles in biodegradability and long‐term toxicity, limiting their use primarily to preclinical proof‐of‐concept studies. In contrast, lipid‐based systems (e.g., LNPs) and GalNAc conjugates have achieved clinical validation primarily outside the CNS, supported by established safety profiles and scalable manufacturing. CNS‐directed conjugates such as C16, however, remain investigational and currently rely largely on local or intrathecal administration. In this section, we critically examine these platforms, focusing not only on their silencing efficiency but also on their translational potential regarding BBB penetration, safety, and manufacturability. Representative delivery platforms discussed in this section are summarized in Table 1, and major carrier classes and intracellular barriers are illustrated in Figure 4.

TABLE 1.

siRNA delivery strategies in AD.

Delivery systems category	Specific delivery platform	Strategy for CNS access	BBB targeting ligand	AD therapeutic target	References
Inorganic nanomaterials	PEGylated/dendronized AuNPs conjugated with siRNA (siAPOE4)	In vitro BBB‐cell uptake only. CNS access not established in vivo	None	APOE4	(Okła et al. 2023, 2025)
	PEGylated magnetite nanoparticles	CNS access not demonstrated (cellular delivery only)	None	BACE1	(Lopez‐Barbosa et al. 2020)
	Prussian Blue nanoparticles (PRM‐siRNA)	BBB modulation: NIR photothermal‐assisted transient opening	None	BACE1	(Ding et al. 2025)
Lipid‐based delivery systems	ALN‐APP (Mivelsiran)	BBB bypass: intrathecal (IT)	None	APP	(Cohen et al. 2025; Sorets et al. 2024)
	Microglia‐targeted LNPs (MG‐LNP)	BBB bypass: intracisternal (cisterna magna)	None	PU.1 (transcription factor SPI1)	(Ralvenius et al. 2024)
	Rapa@DAK/siRNA (Lipid‐Polymer)	Nose‐to‐brain: intranasal (olfactory/trigeminal pathways)	None	BACE1	(Yang et al. 2022)
Polymer/hybrid nanoparticles	Glycosylated polymeric NPs (Gal‐NP)	Transcytosis: RMT (GLUT1)	Galactose (Glucose analog)	BACE1	(Zhou et al. 2020)
	RVG‐modified PAMAM (PPR@siBACE1)	Transcytosis: RMT (RVG–nAChR)	RVG29 Peptide	BACE1	(Li et al. 2025)
	PEG‐PDMAEMA nanocomplexes (CT/siRNA)	Transcytosis: ligand‐mediated (CGN peptide)	CGN (BBB penetration)	BACE1	(Wang et al. 2018)
	PLGA nanoparticles	BBB bypass: intracisternal injection (cisterna magna)	None	p16Ink4a (Senescence marker)	(Shin et al. 2024)
	LPEI‐g‐PEG micellar NPs (shape‐dependent)	BBB bypass: intracerebroventricular (ICV) infusion	None	BACE1/APP	(Shyam et al. 2015)
Biological and biomimetic systems	Engineered Exosomes (RVG‐Exo)	Transcytosis: RMT (RVG–nAChR)	RVG peptide	BACE1	(Alvarez‐Erviti et al. 2011)
	Stem Cell Exosomes (MsEVB)	Nose‐to‐brain: intranasal (reported)	None	BACE1 (co‐delivered with Berberine)	(Sun, Sha, et al. 2025)
	Hybrid Cell Membrane Nanoparticles	Inflammation‐homing BBB crossing after i.v. administration (CCR2/platelet membrane‐guided)	Hybrid membrane	Neuroinflammation targets	(Lin et al. 2024)

Note: This table summarizes representative siRNA delivery platforms explored for AD, including lipid‐based nanoparticles/conjugates, polymeric nanoparticles, inorganic nanomaterials, and exosome‐based carriers. For each system, the proposed route/mechanism of CNS access, the carrier’s targeting ligand/agent, the intended AD molecular target, and the corresponding primary references are provided.

Abbreviations: AD, Alzheimer's disease; AuNP, gold nanoparticle; BBB, blood–brain barrier; CSF, cerebrospinal fluid; LNP, lipid nanoparticle; NIR, near‐infrared; PAMAM, poly(amidoamine); PLGA, poly(lactic‐co‐glycolic acid).

FIGURE 4.

siRNA delivery systems and mechanism of action. This diagram illustrates various siRNA delivery systems, including inorganic nanomaterials, lipid nanoparticles, lipid conjugates, polymer nanoparticles, biomimetic vesicles, and exosomes. BBB, blood–brain barrier; EVs, extracellular vesicles; LNPs, lipid nanoparticles; PEG, polyethylene glycol; RISC, RNA‐induced silencing complex; siRNA, small interfering RNA; C16, 16‐carbon lipid moiety used for lipid–siRNA conjugation.

5.1. Inorganic Nanomaterials

The application of inorganic nanomaterials for siRNA delivery in the treatment of AD holds great promise. With the advancement of nanotechnology, the potential of inorganic nanomaterials such as gold nanoparticles (AuNPs), superparamagnetic iron oxide nanoparticles (SPIONs), and quantum dots (QDs) as siRNA carriers in AD is gradually being recognized. These materials provide structural stability and tunable physicochemical properties (e.g., size, shape, and surface chemistry) that can be leveraged for siRNA loading and theranostic integration. However, brain exposure after systemic administration is generally limited and highly context‐dependent, and both BBB transport and biosafety vary substantially with formulation parameters and disease state (Wu et al. 2023). Surface functionalization (e.g., stealth coatings and receptor‐targeting ligands) may improve circulation time and endothelial interactions, yet rigorous biodistribution/clearance studies and long‐term safety assessments are still required to confirm meaningful parenchymal delivery and to exclude chronic neurotoxicity or off‐target accumulation (Saker et al. 2024).

Moreover, quantum dots can also be used for early diagnosis, combining therapeutic functions to provide new possibilities for precise treatment of AD. Notably, Ding et al. reported a multifunctional Prussian blue (PB) nanocomplex (PRM‐siRNA) that co‐loads BACE1 siRNA, metallothionein, and a Ru‐based tracker. Under near‐infrared (NIR) irradiation, the PB core produced a photothermal effect that transiently increased BBB permeability, thereby enhancing brain delivery. In APP/PS1 mice, PRM‐siRNA treatment reduced BACE1‐associated amyloidogenic burden, and mitigated neuroinflammation and tau‐related pathology, leading to improved learning and memory. Given that Prussian blue has prior clinical use (e.g., as an FDA‐approved agent for treatment of known or suspected internal contamination with radioactive cesium and/or radioactive or non‐radioactive thallium), this work also highlights a potentially more translation‐friendly inorganic platform when paired with controlled, reversible BBB modulation (Ding et al. 2025). Despite their imaging advantages, quantum dots raise well‐recognized concerns regarding neurotoxicity (e.g., oxidative stress and potential heavy‐metal associated effects), underscoring the need for careful material selection and systematic safety profiling (Hu et al. 2024). For example, Okła et al. discussed the strategy of using PEG‐modified AuNPs as delivery carriers to transport siRNA to AD‐affected brain areas (Okła et al. 2025). AuNPs are widely explored due to their readily functionalized surfaces; nevertheless, “biocompatibility” is formulation‐dependent (size/coating/dose), and long‐term biodistribution and clearance remain key considerations for chronic CNS indications. Although this study emphasizes the potential of gold nanoparticles, it still primarily focuses on in vitro experiments and hypothetical research. In addition, Naimi et al. utilized SPIONs to deliver siRNA, particularly targeting BACE1, for the treatment of AD (Naimi et al. 2024). Magnetic nanoparticles offer good biocompatibility and can be guided to the target area by an external magnetic field, which is crucial for regulating Aβ‐related proteins through siRNA. Song et al. reviewed the role of various metallic nanomaterials (such as gold, silver, and iron oxide) in the diagnosis and treatment of AD (Song et al. 2024). Although a detailed discussion on siRNA delivery was not provided, the multifunctionality of these inorganic materials makes them potential carriers for AD therapy. They are not only used for drug delivery but also as biomarkers. Similarly, Jin et al. explored the application of various nanocarriers, including inorganic nanomaterials, in traumatic brain injury (TBI)‐related AD, particularly in the delivery of siRNA to target AD‐related genes. While the focus was primarily on TBI‐related AD, this study provides valuable perspectives and experimental evidence for siRNA delivery using inorganic nanomaterials (Jin et al. 2025).

Despite these promising preclinical results, the clinical translation of inorganic nanomaterials for AD remains stagnant compared to organic counterparts. The translational limitations of inorganic nanomaterials are material‐dependent and may include limited biodegradability, tissue retention, and potential toxicity during chronic dosing. Consequently, the suitability of individual inorganic nanomaterials for chronic siRNA delivery should be evaluated on a material‐specific basis, particularly with respect to biodegradation, long‐term retention, and safety.

5.2. Lipid Delivery System

Lipid‐based delivery systems are among the most promising approaches for siRNA therapeutics, offering efficient encapsulation, stability, and targeted delivery. This section will focus on key lipid‐based platforms, including C16 lipid conjugates, LNPs, and liposomes, which have emerged as versatile and scalable systems for effective siRNA delivery.

5.2.1. C16 Lipid Conjugates

The delivery of siRNA remains a critical challenge in therapeutic applications, particularly when targeting the CNS. C16 lipid conjugates have shown promising results in overcoming these challenges, providing an efficient, safe, and reproducible method for siRNA delivery (Brown et al. 2022). This section will focus on the role of C16 lipid conjugates in improving the pharmacokinetics of siRNA, with a special emphasis on their application in the CNS.

As discussed above, the critical bottleneck in siRNA therapy lies not only in sequence design, but also in how to safely, effectively, and reproducibly deliver siRNA to target cells. For non‐viral RNA delivery, three major pillars are organ or cell selectivity, endosomal escape, and immune safety (Guo et al. 2024). By introducing lipid chains of varying lengths at the 2′‐O position of the sense strand, studies have shown that longer lipid chains enhance siRNA activity. Among these, C16 lipid conjugates demonstrated the best overall performance in the CNS of rats. C16‐conjugated siRNAs retaining 2′‐F/2′‐OMe modifications and incorporating GNA show enhanced efficacy and specificity while preserving activity in the CNS. C16 (2′‐O‐hexadecyl) conjugation expands siRNA activity beyond the liver to extrahepatic tissues including the CNS, eye, and lung (Brown et al. 2022).

Research by Brown et al. demonstrated that C16‐siRNA achieved potent and sustained gene silencing across multiple cell types in the CNS, eyes, and lungs of both rodents and non‐human primates (NHP), with good tolerability (Brown et al. 2022). This study laid the foundation for the application of local/intrathecal administration combined with long‐acting chemistry. Subsequently, L2 (albumin‐binding) lipid‐siRNA, utilizing perivascular transport pathways following CSF‐directed administration, further enabled widespread distribution across the entire CNS, with a single administration maintaining effect for approximately 5 months, significantly enhancing drug distribution uniformity and therapeutic duration.

The advantages of the C16/L2 system include its applicability to multiple species, functional pharmacodynamics (PD) studies, and complete time dimension, with good manufacturability. However, its limitation lies in the need for receptor‐mediated transport mechanisms or other carrier systems for intravenous administration to cross the BBB (Brown et al. 2022). For chronic diseases such as AD, the low‐frequency, long‐acting dosing regimen of the C16/L2 technology directly translates into therapeutic benefits and accessibility for patients, offering significant clinical potential.

5.2.2. Liposomes/LNP Delivery of siRNA

For liposomes and LNPs, the most relevant AD studies are those that explicitly target the brain or exploit local CNS administration. A notable example is the microglia‐targeted LNP platform reported by Ralvenius et al., in which anti‐PU.1 siRNA was delivered via localized intracisternal administration. Because PU.1 is a microglial transcription factor linked to AD risk and neuroinflammatory signaling, this study is especially important conceptually: it showed that LNP‐mediated RNA delivery can reduce PU.1 levels and attenuate neuroinflammation in mouse models including CK‐p25 mice, which recapitulate aspects of chronic neuroinflammation relevant to AD. This makes the platform more directly relevant to AD biology than generic liver‐directed LNP studies (Ralvenius et al. 2024).

A second lipid‐based direction is intranasal nose‐to‐brain delivery. In APP/PS1 mice, liposomal formulations co‐delivering donepezil, memantine, and BACE1 siRNA improved short‐term memory, reduced Aβ40/42 levels, lowered BACE1 mRNA, and downregulated inflammatory cytokines after intranasal administration. Although this is a combination‐therapy design rather than a pure siRNA monotherapy platform, it provides direct in vivo evidence that liposomal systems can be adapted for AD‐relevant NtB delivery (Lee et al. 2024).

More recently, lactoferrin‐functionalized LNPs co‐encapsulating α‐mangostin and BACE1 siRNA were reported as a targeted nose‐to‐brain system in APP/PS1 mice (Xu, Ye, et al. ²⁰²⁵). In that study, lactoferrin functionalization improved brain targeting after intranasal administration, while the combination of BACE1 silencing and microenvironment modulation led to cognitive improvement, plaque reduction, and attenuation of neuroinflammation and oxidative stress. This type of multifunctional LNP is attractive for AD because it addresses both amyloid production and downstream inflammatory dysfunction, although it should still be regarded as a preclinical proof‐of‐concept rather than a mature translational solution.

Taken together, lipid‐based siRNA delivery for AD currently appears strongest in three translationally relevant formats: long‐acting intrathecal lipid conjugates, localized CSF‐delivered LNPs for microglial targeting, and intranasal liposomal/LNP formulations for nose‐to‐brain transport. By contrast, many canonical LNP design studies, while foundational for the field, remain focused on hepatic or non‐CNS delivery and are therefore better treated as general formulation background rather than as central AD examples.

5.3. Polymer/Hybrid Nanoparticle Delivery of siRNA

The use of polymer or hybrid nanoparticles for siRNA delivery represents an exciting frontier in therapeutic strategies, offering significant improvements in both drug stability and targeted delivery. By combining biodegradable polymers with lipid shells, these systems can effectively prolong siRNA release while ensuring efficient gene silencing. This section explores polymeric and hybrid nanoparticle formulations with potential relevance to CNS delivery, while emphasizing examples that are more directly related to brain delivery rather than non‐CNS applications.

Shi et al. proposed a hybrid lipid nanoparticle (LPNP) delivery system with a polymer core and lipid shell, which balances sustained release with efficient gene silencing (Shi et al. 2014). The biodegradable polymer core (e.g., PLGA) provides mechanical stability and controlled release, while the lipid shell enhances membrane interactions and facilitates endosomal escape (Natarajan et al. 2014). Cell and animal experiments have shown that this structure prolongs the duration of action and optimizes drug delivery efficiency. Research has shown that the core‐to‐shell ratio, surface charge, and PEGylation are three key factors in hybrid nanoparticle‐mediated siRNA delivery, particularly for local or intrathecal delivery, which effectively reduces systemic exposure (Mandal et al. 2013). Meanwhile, the PBAE (poly (β‐amino ester)) family also has been explored for siRNA delivery, with PBAE synthesized via a one‐step Michael addition, making easy screening of terminal and side‐chain libraries to optimize drug delivery systems (Karlsson et al. 2020). The degradable cationic polymers improve in vivo tolerance (Chen et al. 2024). Systematic reviews have highlighted how subtle changes in pKa, hydrophobicity, and amino structure of the polymer can significantly influence its drug‐loading capacity, release efficiency, and cytotoxicity (Hussain et al. 2024; Routkevitch et al. ²⁰²⁰). PBAE has shown in vivo activity both for local and systemic delivery. However, much of the PBAE optimization literature remains focused on general formulation development or non‐CNS settings rather than direct AD or brain‐targeted siRNA delivery.

Polyethylenimine (PEI), especially the 25 kDa branched form, is a popular delivery material with excellent loading capacity and endosomal escape performance, but its cytotoxicity limits its widespread application. To address this, PEI has been engineered with modifications such as reduced molecular weight, fluorination, or amphoteric ionization, or controlled oxidation or degradation (Helmschrodt et al. 2017; Saqafi and Rahbarizadeh ²⁰¹⁸). Introducing fluorocarbon segments into 1.8–2 kDa PEI side chains successfully formed self‐assembled nanocomplexes, significantly reducing toxicity, improving serum tolerance, and enhancing delivery efficiency. The fluorine element imparts hydrophobic and hydrophilic dual properties to the polymer, improving its interactions with membranes and proteins (Deng et al. 2017; Yuan et al. ²⁰²⁰). Moreover, modifying PEI with amphoteric monomers, such as sulfonic or carboxyl betaine, significantly reduces protein adsorption, increases siRNA protection and serum stability, and maintains strong binding capabilities. Surface modification may reduce nonspecific interactions, although repeated‐dose safety requires further validation (Badihi et al. 2021). By adjusting the oxidation and degradation degrees of linear PEI, researchers have found the optimal balance between delivery efficiency and cellular compatibility. For formulations requiring large‐scale production, linear PEI and controlled oxidation or degradation offer a more process‐friendly route (Li et al. 2024; Yang, Hu, et al. ²⁰²⁴).

In 2011, Patil et al. used dendritic macromolecules (PAMAM) to construct a PAMAM‐PEG‐PLL tri‐block multifunctional shell (Patil et al. 2011). PAMAM provided tertiary amine buffering to promote endosomal escape, PEG reduced enzymatic degradation and nonspecific interactions, and PLL enhanced electrostatic loading. This system demonstrated efficient delivery and significant silencing effects in both in vitro and in vivo experiments. In 2009, Agrawal et al. developed a dendrimer‐nano‐worm platform with a magnetic fluorescent core for in vivo functional siRNA delivery (Agrawal et al. 2009). Notably, systemic administration of specific carbosilane dendrimer formulations has been reported to achieve measurable siRNA delivery to the brain in rodents. However, BBB transport and parenchymal distribution are highly dependent on dendrimer generation, surface chemistry/charge, dosing regimen, and experimental context, making it non‐universal. While dendrimers' single‐molecule programmability makes them suitable for complex functional integration (e.g., imaging and delivery), scaling production and addressing toxicity remain challenges (Serramía et al. 2015).

The lipid‐polymer hybrid (LPNP/LPHNP) structure leverages the strengths of both lipids and polymers. In this context, Shi et al. constructed a siRNA carrier with an aqueous core, where G0‐C14 served as a stabilizer, PLGA as the core, and the outer layer consisted of a neutral or near‐neutral lipid‐PEG composite (Shi et al. 2014). This design enabled continuous siRNA release for over a month and illustrated how surface‐charge tuning can improve loading and cellular uptake. These features are conceptually relevant for CNS‐directed siRNA delivery, where prolonged exposure and controlled release may help reduce dosing frequency. However, the original efficacy demonstration was performed mainly in non‐CNS tumor models, so this study is better interpreted here as a formulation‐design reference rather than direct evidence for AD therapy.

Overall, polymeric and hybrid nanoparticles provide valuable engineering principles for siRNA stabilization, controlled release, and intracellular delivery. However, compared with lipid conjugates, CNS‐directed LNPs, or EV‐based approaches, direct evidence for polymer/hybrid nanoparticle‐mediated siRNA delivery in AD remains relatively limited. Therefore, in the context of AD, these systems should currently be viewed as promising but still early‐stage platforms whose relevance depends on future demonstrations of robust brain distribution, cellular specificity, and acceptable safety after clinically realistic routes of administration.

5.4. Biological Delivery Systems

In recent years, biological delivery systems have emerged as an important complement to synthetic nanocarriers for siRNA transport. By leveraging native membrane components and naturally occurring cellular communication mechanisms, these platforms may provide unique advantages in terms of biocompatibility, circulation behavior, and cellular uptake. In particular, extracellular vesicles (EVs) and biomimetic vesicles have shown considerable potential for improving siRNA delivery efficiency and expanding tissue‐targeting capabilities. This section summarizes the main features of these biological carriers, with a focus on their design principles, therapeutic applications, and remaining translational challenges.

5.4.1. Exosomes (EVs) and Biomimetic Vesicle Delivery of siRNA

Exosomes (EVs) are attractive RNA carriers because of their endogenous membrane composition and cellular uptake pathways. Several studies suggest that EVs can facilitate CNS delivery in preclinical models. However, the extent and mechanisms of BBB crossing remain incompletely resolved, and the literature is still relatively scarce with inconsistent methodologies. Therefore, brain delivery using EVs should be considered variable across EV sources, isolation methods, and disease contexts (Ramos‐Zaldívar et al. ²⁰²²; Xu, Jin, et al. ²⁰²⁵). Engineered EVs, especially those derived from dendritic cells and modified with targeting peptides like rabies virus glycoprotein (RVG), offer a unique solution for targeted delivery to the CNS. This section explores the use of EVs in siRNA delivery, highlighting their advantages, challenges, and promising applications in treating AD and other brain disorders.

In 2011, Alvarez et al. conjugated the neuron‐specific RVG peptide with Lamp2b and anchored it to the surface of exosomes derived from autologous dendritic cells (Alvarez‐Erviti et al. ²⁰¹¹). siRNA was loaded into these exosomes using electroporation. After intravenous injections, these engineered EVs were delivered into mice, achieving specific silencing of multiple cell types in the brain, including neurons, microglial cells, and oligodendrocytes. In a mouse BACE1 model, mRNA levels decreased by approximately 60%, and protein levels decreased by about 62%, with no significant nonspecific uptake observed. This study demonstrated the feasibility of engineered EVs and systemic administration crossing the BBB. However, large‐scale production and consistency remain bottlenecks for translational applications. Moreover, in 2014, Cooper et al. used RVG‐EVs to deliver α‐Syn siRNA to the whole brain, effectively downregulating mRNA and protein levels within 7 days, reducing the formation of inclusions in dopaminergic neurons of the substantia nigra, and inhibiting the associated inflammatory response (Cooper et al. 2014). This study showed promising functionality in the treatment of neurodegenerative diseases, providing solid animal model evidence for EV‐siRNA application in brain disorders. In a 2024 review, Ubanako highlighted that while electroporation is commonly used for siRNA loading, it can lead to aggregation and precipitation of siRNA, thus overestimating loading efficiency (Ubanako et al. 2024). To address this issue, hydrophobic modifications, such as the incorporation of 3′‐cholesterol siRNA, could improve stability and cellular uptake. Moreover, source differences are a critical factor in the efficiency of EV delivery. For instance, MSC‐derived EVs typically exhibit stronger tumor‐homing abilities, while some tumor cell‐derived EVs, despite having higher loading capacities, may not distribute as effectively in vivo (Kalluri and LeBleu ²⁰²⁰). This emphasizes the importance of the source‐function pairing.

RVG‐EVs have also been applied in the treatment of other neurodegenerative diseases. For example, RVG‐EVs were used to deliver miR‐193b‐3p to the brain of subarachnoid hemorrhage (SAH) mice (Lai et al. 2020). By transfecting bone marrow‐derived mesenchymal stem cells (BMSCs) to express Lamp2b and fuse with the targeting peptide RVG, followed by electroporation loading of miR‐193b‐3p, treated SAH mice showed reduced expression of histone deacetylase 3 (HDAC3), upregulated acetylation of NF‐κB p65, and a reduction in pro‐inflammatory cytokines, which improved neurological function and reduced brain edema and BBB damage (Kim and Kim 2017; Lai et al. ²⁰²⁰). Similarly, RVG‐EVs have been used to deliver miR‐124 to the infarcted area of a cortical ischemia mouse model. Yang et al. demonstrated that RVG‐EVs carrying miR‐124 promoted neurogenesis by converting neural progenitor cells into neurons and protected the mice from ischemic injury (Yang et al. 2017). Additionally, RVG‐EVs have also been used to deliver opioid receptor mu (MOR) siRNA to treat morphine addiction (Liu et al. 2015). Studies showed that RVG‐EV‐delivered MOR siRNA downregulated MOR expression in the mouse brain and inhibited morphine relapse. Furthermore, engineered EVs fused with RVG and Lamp2b have been used to coat gold nanoparticles (AuNP) for therapeutic applications (Khongkow et al. 2019). Bioluminescence imaging showed that, after intravenous injections, RVG‐Lamp2b‐EV‐coated AuNPs successfully accumulated in the mouse brain, providing a new approach for EV‐based targeted therapy.

5.4.2. Biomimetic Vesicles

Biomimetic vesicles, which are cell membrane‐coated nanoparticles, offer a novel approach to siRNA delivery by mimicking natural processes to improve targeting and reduce immune clearance. This strategy leverages the natural properties of cell membranes, such as self‐recognition and homologous recognition, which enhance the stability and efficiency of the delivery system. This section will explore the development of biomimetic vesicles, focusing on their production process, advantages, and potential relevance to CNS siRNA delivery, while acknowledging that direct evidence in AD remains limited.

The production process of cell membrane‐coated nanoparticles primarily involves three key stages: cell lysis and membrane disruption, membrane separation, and nanoparticle coating (Fernández‐Borbolla et al. ²⁰²⁴). The outer membrane coating provides self‐recognition markers (e.g., CD47) and homologous recognition, which can reduce immune clearance, extend circulation time, and potentially enhance uptake by target tissues or pathologically relevant cells (Chen et al. 2022). Additionally, the outer membrane and inner core physically protect nucleic acids, preventing their leakage and degradation by serum components, thereby improving the stability and efficiency of the delivery system. More broadly, Liu's review points out that erythrocyte, platelet, leukocyte, stem cell, and tumor membranes each have distinct characteristics (Liu et al. 2023). These membranes can be hybridized to simultaneously achieve immune evasion and recognition of inflammation or adhesion sites. In principle, these features may also be useful for brain‐directed siRNA delivery, particularly when immune evasion and selective interaction with inflamed neurovascular environments are required. However, differences in preparation steps remain substantial, and there is currently no unified gold standard, which poses major challenges for reproducibility, large‐scale manufacturing, and process standardization.

Overall, biomimetic vesicles offer an attractive conceptual framework for siRNA delivery because they combine membrane‐derived biocompatibility with the possibility of engineering targeting and immune‐evasive functions. However, in the context of AD, direct evidence for robust brain delivery and therapeutic efficacy remains limited, and further work is needed to address standardization, scalability, and clinically relevant validation.

6. Strategies of Improved siRNA Delivery Across the BBB

The BBB represents a significant challenge for effective drug delivery to the brain. As a highly selective permeability barrier, the BBB restricts the passage of therapeutic molecules, including siRNA, making it difficult to treat neurological diseases such as AD. In recent years, various strategies have been developed to overcome this barrier, enabling siRNA delivery to the brain. These strategies leverage different mechanisms, including RMT, adsorption‐mediated transcytosis (AMT), and non‐invasive approaches such as nose‐to‐brain (NtB) delivery. Understanding the underlying mechanisms that facilitate BBB crossing is critical to advancing siRNA‐based therapies for AD and other CNS disorders. In this section, we will discuss the main siRNA delivery pathways, focusing on receptor‐mediated transcytosis and its potential for crossing the BBB and delivering therapeutic siRNA to the brain. A schematic overview of these CNS delivery routes is provided in Figure 5.

FIGURE 5.

Strategies for CNS siRNA delivery. Approaches include nose‐to‐brain (NtB) delivery; transient BBB modulation; and transcytosis‐based shuttles, such as receptor‐mediated transcytosis (RMT) or adsorptive‐mediated transcytosis (AMT). AMT, adsorptive‐mediated transcytosis; BBB, blood–brain barrier; CNS, central nervous system; NtB, nose‐to‐brain; RMT, receptor‐mediated transcytosis; siRNA, small interfering RNA.

6.1. Receptor‐Mediated Transcytosis as Main Systemic Administration Pathway Across the BBB

Drug delivery to the brain is highly restricted by the BBB (Abbott et al. 2010). While the BBB strictly limits passive transport, receptor‐mediated transcytosis (RMT) serves as the most viable “Trojan horse” strategy for the systemic delivery vectors, as discussed in Section 5. Whether utilizing LNPs, polymeric nanomedicine, or biomimetic vesicles, the successful translation of these carriers largely depends on their ability to engage specific transporters (e.g., TfR, CD98hc, and LRP1) expressed on brain endothelial cells. RMT allows these engineered vectors to bind to receptors on the luminal side of the BBB, undergo endocytosis, and release their siRNA payload into the brain parenchyma (Haqqani et al. 2024; Tashima ²⁰²²).

Recent advances have shifted from using high‐affinity antibodies, which often get trapped in lysosomes, to monovalent, moderate‐affinity ligands that facilitate successful transcytosis and parenchymal release (Barker et al. 2024). This new approach has resulted in significant and sustained target RNA downregulation, outperforming high‐affinity bivalent TfR antibody conjugates and direct CSF delivery routes. These advancements validated the feasibility of delivering nucleic acid drugs to the brain parenchyma via RMT. Importantly, this mechanism is also the foundation for the lipid‐conjugate strategies and ligand‐modified LNPs, where integrating TfR‐targeting ligands (e.g., specific peptides or antibodies) enables these hydrophobic cores to bypass the BBB limit.

In addition to canonical RMT shuttles (e.g., TfR or CD98hc), a conceptually distinct non‐viral conjugate strategy has recently been introduced. Wang et al. developed a BBB‐crossing conjugate (BCC) platform that leverages γ‐secretase–mediated transcytosis to transport biomacromolecules into the CNS after intravenous administration. Importantly, a BCC10‐oligonucleotide conjugate achieved effective BBB transport and gene silencing in wild‐type mice, human brain tissues, and an ALS mouse model, suggesting an orthogonal, potentially scalable route for systemic oligonucleotide delivery that may be adaptable to siRNA therapeutics (Wang et al. 2025).

Pornnoppadol's team designed a 2 × 1 bispecific antibody brain shuttle that fuses a single‐chain antibody (scFv) at the C‐terminus of a heavy chain to bind the endothelial receptor CD98hc with monovalent, lower affinity, while retaining the bivalent ligand recognition ability of IgG (Pornnoppadol et al. 2024). Compared to the classic TfR‐1 shuttle, the CD98hc shuttle achieved longer brain retention (≥ 7 days), whereas the TfR‐1 shuttle was undetectable after 5 days. When carrying non‐targeting IgG, the CD98hc shuttle predominantly showed vascular‐associated distribution in the brain, while the TfR‐1 shuttle was more likely to be taken up by parenchymal cells and cleared more quickly. This difference creates conditions for subsequent selective binding to secondary targets in the brain parenchyma. By modifying IgG to recognize astrocytes (M2) or neurons (M6), the authors demonstrated cell‐type‐selective in situ targeting. Furthermore, the TrkB‐agonist IgG‐CD98hc shuttle activated the TrkB‐AKT/ERK pathway in vivo, sustaining activation for several days. Monovalent binding (2 × 1) at sufficient doses led to higher and longer brain retention. Compared to the bivalent IgG form of the same antibody, the 2 × 1 format showed superior retention kinetics on the CD98hc pathway. This delivery method allows for targeted delivery to astrocytes or neurons, enabling both extracellular sustained‐release drug delivery and intracellular siRNA delivery.

In the case of peptides‐decorated nano‐systems, Zheng et al. developed a triple interaction stable polymer‐siRNA nanomedicine (3I‐NM@siRNA) that stabilizes siRNA through electrostatic interactions, hydrogen bonding, and hydrophobic interactions, enhancing its resilience in physiological environments (Zheng et al. 2019). This system exhibits sequential destabilization/self‐destructive release in response to ROS in the tumor environment, efficiently releasing siRNA at the lesion site. Compared to a control system relying solely on electrostatic interactions, the 3I system demonstrated superior stability and controlled release. After functionalizing the carrier surface with Angiopep‐2 (Ang) (Ang‐3I‐NM@siRNA), the system successfully crossed the BBB via LRP1‐mediated transport. This serves as a prime example of the polymeric hybrid strategy described in Section 5.3 and demonstrates how surface functionalization with RMT ligands can empower a stable electrostatic complex to achieve targeted brain accumulation. Similarly, Zhou et al. developed galactose‐modified triple interaction polymer siRNA nanocarriers (Gal‐NP@siRNA), which significantly improved circulation stability and in vivo persistence through the combined electrostatic, hydrogen bonding, and hydrophobic interactions (Zhou et al. 2020). Using a glucose transporter 1 (GLUT1; encoded by SLC2A1)‐mediated mechanism, the nanocarriers crossed the BBB and delivered siRNA to downregulate BACE1. In the APP/PS1 Alzheimer's disease model, Gal‐NP@siRNA effectively reduced BACE1 expression, inhibited Aβ aggregation, and improved cognitive function. This design cleverly integrates the strategies of glucose/galactose‐dependent transport and high stability delivery, advancing the development of siRNA BBB delivery platforms and providing more precise control for future intracerebral nucleic acid therapies. This strategy highlights the potential of hijacking nutrient transporters, such as GLUT1, to enhance siRNA delivery, beyond traditional protein receptors like TfR.

Recent advancements also highlight the potential of engineered nanomedicines to actively recognize and penetrate the BBB under pathological conditions. For instance, strategies utilizing tannic acid‐based nanocomposites have demonstrated the ability to specifically target injured BBB sites and deliver therapeutic cargo to the brain parenchyma (Shi et al. 2025), offering valuable design principles for AD‐targeted delivery.

In conclusion, receptor‐mediated transcytosis (e.g., TfR, CD98hc, or LRP1) provides the primary pathway for delivering drugs to the brain parenchyma under systemic administration conditions. Ligand functionalization and monovalent moderate‐affinity designs exhibit kinetic advantages over traditional high‐affinity conjugates. However, RMT shuttles face three recurring constraints. First, BBB receptor sequence and trafficking differ across species, so efficacy observed in rodents may not translate unless the ligand is cross‐reactive or validated in humanized receptor models and non‐human primates (Wu et al. 2023). Second, RMT is saturable and exhibits a dose‐ and affinity‐dependent “sweet spot”. Overly high apparent affinity/avidity can promote endothelial sequestration and lysosomal routing, whereas too low affinity reduces engagement, producing a bell‐shaped relationship between TfR binding Kd and brain exposure, as predicted by translational PBPK modeling (Sato et al. 2023). Third, the abundant TfR/CD98hc expression in peripheral tissues creates a “peripheral sink” that shapes pharmacokinetics, limits brain bioavailability, and may introduce on‐target off‐tissue liabilities. Whole‐body and single‐cell mapping has further shown distinct peripheral biodistribution and parenchymal cell‐type distribution for TfR‐ versus CD98hc‐based transport vehicles, underscoring the need to co‐optimize brain exposure with peripheral distribution (Khoury et al. 2025).

6.2. Adsorption‐Mediated Transcytosis (AMT) via Cell‐Penetrating Peptides (CPPs)

AMT offers a promising method for delivering drugs to the brain across the BBB. The BBB is well‐suited for the AMT process, offering a pathway for cationic molecules to bind to the luminal surface of endothelial cells and be internalized, followed by exocytosis at the opposite surface. The transcytotic pathways and their morphological and enzymatic characteristics at the BBB provide a mechanism for molecular movement through endothelial cytoplasm. Cationic proteins and CPPs are utilized to deliver AMT‐based drugs to the brain.

For example, Stalmans et al. quantified the in vitro kinetics of CPPs and demonstrated significant differences in peptides such as pVEC, SynB3, and Tat. Approximately 80% of the brain signal was derived from brain parenchyma rather than from capillary remnants, indicating that CPP is not equivalent to BBB shuttle peptides and necessitating individual quantification (Stalmans et al. 2015). Similarly, Wang et al. designed and prepared a PEGylated poly(2‐(N,N‐dimethylamino)ethyl methacrylate) (PEG‐PDMAEMA)‐based siRNA nanocarrier, which was modified with CGN peptide for BBB penetration and Tet1 peptide for neuron‐specific binding (Wang et al. 2018). The nanocomplexes, CT/siRNA, composed of CGN‐PEG‐PDMAEMA and Tet1‐PEG‐PDMAEMA at a 1:1 weight ratio, demonstrated good stability in blood and did not induce hemolysis at an N/P ratio of 10. The internalization of the nanocomplex in neuronal cells relied on clathrin‐mediated endocytosis and micropinocytosis, with caveolae‐mediated endocytosis playing a major role in CT/siRNA entering the brain endothelial cells (bEnd). The nanocomplex successfully escaped the lysosome and entered the cytoplasm of neuronal cells, inducing effective gene silencing (BACE1 mRNA level decreased by approximately 50%) and reversing Aβ25–35 oligomer‐induced synaptic damage. After intravenous injection, CT/siRNA showed higher brain accumulation (brain targeting index = 2.62) compared to unmodified nanocomplexes, co‐localizing with neurons or being in close proximity. In APP/PS1 transgenic mice, the nanocomplex significantly reduced BACE1 mRNA and amyloid plaques, inhibited phosphorylated tau levels, and promoted hippocampal neurogenesis. Notably, administration of the nanocomplex restored cognitive function in AD transgenic mice to wild‐type control levels without significantly affecting myelination. Moreover, Zheng's team developed a hybrid system for targeting amyloid plaques using PEGylated poly(2‐(N,N‐dimethylamino)ethyl methacrylate) (PEG‐PDMAEMA) conjugated with two D peptides, one for brain penetration (CGN) and the other for amyloid‐β binding (QSH) (Zheng et al. 2017). The hybrid complex CQ/siRNA, composed of 25% MPEG‐PDMAEMA, 50% CGN‐PEG‐PDMAEMA, and 25% QSH‐PEG‐PDMAEMA, demonstrated negligible cytotoxicity and was able to protect siRNA from enzymatic degradation. After being internalized by neuronal cells, the complex escaped from the lysosome and released siRNA into the cytoplasm, leading to effective gene silencing (reducing protein levels by 18.5%). Following intravenous injection, CQ/siRNA was delivered intact to the brain and localized around the plaques in transgenic AD mice. Precise amyloid plaque targeting increased therapeutic activity, as reflected by a strong reduction in BACE1 mRNA (36.4%), lower levels of the enzymatic digestion product sAPPβ (−42.6%), and improved neuronal protection compared to single‐component complexes.

6.3. Physical Windowing: Focused Ultrasound and Microbubble‐Mediated Reversible BBB Opening

Magnetic resonance‐guided or navigation‐guided focused ultrasound (MRgFUS or NgFUS), combined with intravenous microbubbles, can transiently and reversibly increase BBB permeability in targeted brain regions, enhancing the brain influx of systemically administered molecules (Lamsam et al. 2018). The primary biophysical basis for this method is stable cavitation‐driven transient loosening of endothelial tight junctions and upregulation of endothelial transcytosis (caveolae/transcytosis). However, when the ultrasound parameters are too strong and enter inertial cavitation, risks such as microbleeds, edema, and inflammation increase. Clinically, BBB opening and closing are confirmed through T1 contrast‐enhanced MRI, while passive cavitation detection (PCD)/ultrasound spectroscopy and cavitation dose monitoring enable closed‐loop control (Chien, Xu, et al. ²⁰²²).

In a study by Lipsman et al., MRgFUS was used in five mild‐to‐moderate AD patients to open the BBB in targeted regions, demonstrating reversible and repeatable BBB opening (Lipsman et al. 2018). Three‐month follow‐up showed no serious clinical or imaging adverse effects, establishing a safety baseline for controllable windowing in the human brain. Rezai et al. conducted monthly injections of aducanumab while performing FUS windowing in selected brain regions on the day of infusion (Rezai, D'Haese, Finomore, et al. ²⁰²⁴). Over a period of 6 months, the opening window allowed for more complete plaque clearance in the target region (compared to the contralateral mirrored area), suggesting that combining windowing with systemic drug delivery could amplify local therapeutic signals. While the sample size was small, this provided the first human therapeutic insight for combined drug delivery. Bae et al. developed a portable navigation‐guided FUS system that achieved real‐time two‐dimensional cavitation mapping in vivo for the first time (Bae et al. 2024). In six subjects, five successfully underwent windowing, with an average window volume of ~983 ± 626 mm³. The cavitation dose and window volume showed a linear correlation (R ² > 0.9). Additionally, changes in plasma/EV biomarkers such as Aβ42, Tau, and pTau181 were correlated with window volume/cavitation dose, and MRI imaging confirmed closure within 2–3 days. This work demonstrates the engineering feasibility and quantifiable dosimetry of non‐MRI, outpatient‐based windowing. These clinical applications are all prospective, small‐sample studies, using imaging and fluid biomarkers as primary outcomes, while safety is assessed through neuroimaging, MMSE scales, and AE monitoring. NgFUS studies incorporate real‐time cavitation mapping as process control.

From a translational standpoint, clinical FUS‐BBB opening (FUS‐BBBO) requires strict controllability and reproducibility (Chien, Yang, et al. ²⁰²²). Opening volume and permeability can vary with skull attenuation, vascular anatomy, and microbubble dose, necessitating individualized acoustic parameter selection and real‐time closed‐loop monitoring (e.g., PCD/cavitation mapping) to remain within stable‐cavitation regimes (Bae et al. 2024). Although early studies demonstrate that BBB opening can be reversible and repeatable in AD patients, the long‐term safety of repeated treatments, particularly regarding microhemorrhage, edema, and neuroinflammatory responses, needs clarification in larger cohorts and in populations with cerebral amyloid angiopathy or high ARIA risk (Lipsman et al. 2018; Rezai, D'Haese, Finomore, et al. ^{2024, 2020}). Moreover, standardization of dosimetry (reporting MI, microbubble dose, cavitation dose and imaging‐defined window volume) will be essential for inter‐site reproducibility and for comparing delivery gains across clinical trials (Padilla and Ter Haar ²⁰²²).

Shumer et al. used ionizable lipid nanoparticles combined with low‐frequency aggregation ultrasound (~850 kHz) and low‐pressure levels (~125–180 kPa) to achieve safe windowing in healthy brain and glioblastoma models, increasing the brain uptake of LNP‐siRNA by approximately one order of magnitude (Shumer‐Elbaz et al. ²⁰²⁵). In vivo bioluminescence signals of mRNA‐LNP were enhanced by about 12 times, and tumor‐associated siRNA‐LNP fluorescence increased ~6–7 times. This demonstrates that non‐invasive FUS can significantly amplify the brain delivery of systemically administered nucleic acid nanoparticles. Martinez et al. integrated mechanical index (MI) and microbubble volume dosage (MVD, μL/kg) to evaluate the BBB opening (BBBO) in mice using MRI–PCD–transcriptomics (Martinez et al. 2024). Their results showed a linear positive correlation between BBB volume and MI, MVD. They also found that the threshold for sterile inflammation (SIR) increased with parameters, proposing an operable therapeutic window between significant BBBO and SIR onset. Harmonic or broadband cavitation doses were linearly related to BBBO volume or CE, and these could serve as process prediction parameters.

The suggestion is to perform systemic siRNA/LNP administration followed by synchronized and short‐duration FUS windowing. For multiple brain regions, sequential zonal windowing could control the total window volume and reduce risks. For diffuse lesions, such as those in AD, clinical practice could first identify high‐load, symptom‐related cortical regions using metabolic and amyloid imaging, prioritizing them for sequential windowing. In humans and large animals, sub‐MHz (~0.2–0.7 MHz) frequencies are typically used to reduce skull attenuation. The goal is to stabilize cavitation, setting MI and MVD axes together with PCD for real‐time closed‐loop control, avoiding inertial cavitation. The use of clinically approved contrast agents (e.g., Definity or SonoVue) in combination with saline flushing is preferred. Harmonic, superharmonic and broadband three‐spectrum cavitation dosing curves and two‐dimensional cavitation imaging monitor dose to target.

FUS‐BBBO provides a controllable, region‐ and timing‐specific channel for systemic siRNA/LNP delivery, useful for both single‐dose high exposure (induction phase) and regular enhancement during the maintenance phase. Parallel or additive use with biological shuttles, such as RMT (TfR1 or CD98hc), is expected to achieve a better balance between dose, frequency, and coverage. Recent in vivo evidence of LNP‐siRNA combined with FUS has advanced non‐invasive nucleic acid delivery to the brain from principle to parameterization and effect quantification (Guo et al. 2021).

6.4. Nose‐To‐Brain (NtB)

In recent years, the nose‐to‐brain drug delivery (NBDD) route has emerged as a promising strategy to bypass the BBB, offering a direct, non‐invasive, and efficient pathway for drug delivery to the brain. NtB delivers drugs from the nasal cavity directly to the brain and CSF via the olfactory epithelium and the anatomical channels provided by the trigeminal nerve, bypassing the endothelial tight junctions of the BBB (Misra and Pathak 2023; Xinchen et al. ²⁰²³). Anatomical and pharmacokinetic studies show that drugs can enter the olfactory bulb and more distant brain regions through both extracellular routes and intracellular pathways. However, challenges remain due to mucociliary clearance in the nasal cavity, local enzymatic degradation, limited dosing volume, and differential deposition, which constitute the main delivery bottlenecks. These challenges require the use of carriers and formulation engineering to enhance retention, penetration, and stability.

Rassu et al. developed a system that first complexes BACE1 siRNA with the RVG‐9R peptide and encapsulates it within solid lipid nanoparticles (SLN), which are further coated with chitosan to enhance adhesion (Rassu et al. 2017). After nasal administration in mice, this system achieved effective nose‐to‐brain delivery, with fluorescence distribution observed in multiple brain regions. They investigated the impact of the RVG and siRNA molar ratio and the surface chemistry of the carrier on transmembrane and brain delivery efficiency. Based on these findings, the authors proposed SLN‐CS as a feasible platform for BACE1 siRNA NtB delivery. Likewise, Yang et al. constructed dual‐targeted nanoparticles (Rapa@DAK/siRNA) modified with Aleuria aurantia lectin (AAL) and KLVFF peptides to co‐deliver BACE1 siRNA and rapamycin. AAL promotes olfactory epithelium uptake and brain entry, while KLVFF facilitates the accumulation and inhibition of Aβ aggregation near plaques. In the APP/PS1 transgenic AD mouse model, BACE1 downregulation, reduced Aβ burden, and cognitive improvement were observed, confirming that hierarchical delivery targeting both the brain and lesion sites can enhance the efficacy of a single dose. This study demonstrated, for the first time, the feasibility of NtB and dual‐targeted delivery combining siRNA and small molecules, although the dosage and long‐term safety still need further evaluation. Additionally, Lee et al. prepared liposomal formulations of donepezil, memantine, and BACE1 siRNA, proposing that the NtB combined delivery strategy could enhance brain exposure and synergistic pharmacological effects (Lee et al. 2024). Their study emphasized how formulation processes (e.g., rotary evaporation for film hydration), particle size, potential, and stability affect the reproducibility of NtB systems, providing a process paradigm for fixed formulations combining small molecules and siRNA for nasal delivery. Further in vivo brain delivery and pharmacodynamics, and long‐term local tolerance verification are required. Sun et al. used human mesenchymal stem cell‐derived EVs as carriers to load BACE1 siRNA and berberine (BBR) via ultrasound and chemical methods, followed by RVG‐29 peptide modification for neuron and glial targeting (Sun, Sha, et al. ²⁰²⁵). After nasal administration in 5xFAD mice, EVs distributed in the olfactory bulb, hippocampus, and cortex, resulting in BACE1 and Aβ reduction, suppressed neuroinflammation, neurorepair markers, and cognitive improvement. This study combined siRNA and anti‐inflammatory small molecules in EVs, achieving synergy in both the delivery pathway and pathological axes.

NtB provides a repeatable, outpatient, and anatomically direct route for siRNA brain delivery. From the early SLN × RVG paradigm to the dual‐targeted AAL and KLVFF collaboration, and now to engineered EV co‐delivery, recent studies continue to prove the development of an effective brain targeting and efficacy‐engineering loop. For translational applications, the key lies in stable formulations combined with devices, quantifiable brain region exposure‐pharmacodynamics relationships, and long‐term local and systemic safety evaluations. With advancements in standardized evaluation and manufacturing processes, NtB is expected to become an important delivery option for CNS diseases like AD, complementing intrathecal and systemic RMT approaches.

7. siRNA Targets and Clinical Progress for AD Treatment

Building on the pathological framework outlined in Section 3, this section focuses on the therapeutic tractability of major AD‐related siRNA targets rather than re‐describing their underlying biology. For each target, we emphasize three aspects: the rationale for transcript‐level intervention, the current preclinical or clinical evidence, and the main safety or translational constraints. For an overview of representative carrier designs, targeting ligands, routes of CNS access, and molecular targets, see Table 1.

7.1. APP

APP is therapeutically attractive because transcript‐level suppression can reduce the upstream substrate pool that gives rise to all Aβ isoforms and β‐CTF. In the siRNA setting, however, the key issue is not whether APP is biologically relevant to AD, but whether APP can be reduced to a degree that lowers amyloidogenic burden without excessively disrupting its physiological roles in synaptic function and plasticity (Luo et al. 2025). APP‐targeted siRNA strategies should therefore be viewed as a problem of controllable knockdown, regional exposure, and long‐term tolerability rather than maximal target suppression.

Roselli et al. characterized APP‐BACE1 interactions and their endosomal localization in iPSC‐derived cortical neurons, APP siRNA was used in SH‐SY5Y cells to validate the specificity of the assay (Roselli et al. 2023). Senechal et al. showed that 2 weeks of intracerebroventricular infusion of APP siRNA in adult mice reduced APP expression in the hippocampus but impaired hippocampus‐dependent spatial working memory, highlighting the risk of excessive or nonselective knockdown (Senechal et al. 2007). Alnylam presented data at ISC 2025 and AAIC 2025 showing that a single dose of APP‐lowering siRNA in Aβ‐overexpressing rodent models significantly reduced APP levels, along with brain and vascular amyloid burden. Additionally, reductions in the number and volume of brain microbleeds were reported in the rat model. In the 5xFAD early/late intervention framework, improvements in both pathological and behavioral outcomes were reported, with earlier intervention showing more pronounced effects. Mivelsiran (ALN‐APP), the world's first APP‐targeted siRNA, has entered human clinical trials. In 2023, Alnylam and Regeneron published Phase I clinical results of their APP‐targeting RNAi therapy (ALN‐APP) (Cohen et al. 2024). In a Phase I study with early‐onset AD patients, intrathecal administration of siRNA demonstrated a dose‐dependent effect on reducing APP, thus synchronously decreasing all Aβ isoforms. Single‐dose ALN‐APP administered intrathecally showed good tolerability, and a dose‐dependent, rapid, and sustained reduction in cerebrospinal fluid soluble APPα (sAPPα) and APPβ (sAPPβ), with maximum reductions of 84% and 90%, respectively. Aβ40/42 levels were also reduced, while suppression of sAPPα and sAPPβ persisted for up to 10 months. The clinical feasibility of this project has been demonstrated, and it has progressed to Phase II clinical trials in CAA patients to evaluate its impact on the progression of vascular amyloid disease.

While the robust reduction of APP demonstrated by ALN‐APP is therapeutically desirable, the simultaneous suppression of sAPPα (up to 84%) warrants careful consideration. sAPPα is widely recognized for its neuroprotective and neurotrophic functions, including the modulation of long‐term potentiation (LTP) and synaptic plasticity. Unlike BACE1 inhibitors, which selectively target the amyloidogenic pathway, siRNA strategies targeting the APP transcript inevitably reduce all cleavage products. Consequently, the long‐term safety profile of ALN‐APP will depend on whether the therapeutic benefit of removing toxic Aβ oligomers outweighs the potential physiological cost of substantial sAPPα depletion. Future clinical trials should therefore incorporate rigorous monitoring of synaptic biomarkers (e.g., neurogranin) and cognitive domains specifically linked to plasticity to rule out potential loss‐of‐function toxicity.

7.2. BACE1

BACE1 remains one of the most intervention‐ready amyloid‐related targets because it has already been extensively explored in AD drug development. In the siRNA context, the key issue is not to restate its role in APP processing but to test whether transcript‐level reduction can achieve a safer and more titratable form of inhibition than prior small‐molecule programs. Because BACE1 has multiple substrates, the main translational challenge is to identify a therapeutic window that preserves efficacy while avoiding excessive suppression and associated cognitive liabilities.

Representative preclinical studies have shown that BACE1 siRNA can be delivered to the brain through intravenously administered BBB‐crossing nanocarriers, including neuron‐targeted polymeric nanocomplexes and GLUT1‐mediated transcytotic nanoparticles, resulting in reduced BACE1 expression, decreased amyloid burden, and improved cognitive performance in APP/PS1 mice (Wang et al. 2018; Zhou et al. ²⁰²⁰). Yang et al. co‐delivered BACE1 siRNA and rapamycin via dual‐targeted nanoparticles through nasal administration. This strategy led to BACE1 downregulation and autophagy upregulation, resulting in decreased Aβ deposition and cognitive improvement (Yang et al. 2022). Sun et al. used stem cell‐derived engineered exosomes to co‐deliver BACE1 siRNA and berberine, reducing BACE1 and Aβ levels, inhibiting inflammation and glial responses, and improving cognition in an AD model. This provides empirical support for the delivery and inflammation modulation combination strategy (Sun, Sha, et al. ²⁰²⁵). Li et al. constructed an RVG29‐PAMAM‐PEG (PPR@siBACE1) nanocomplex, which effectively crossed the BBB and distributed into the brain parenchyma after intravenous injection, demonstrating neuron‐specific targeting with potential microglial targeting (Li et al. 2025). In vivo and in vitro studies confirmed BACE1 silencing, reduced Aβ production, enhanced microglial phagocytosis, reduced neuroinflammation, and cognitive improvement.

Through various models and pathways, these studies demonstrate that BACE1 siRNA effectively lowers BACE1 in vivo, reducing Aβ pathology and improving cognitive function. They also suggest that the combination of autophagy and anti‐inflammatory mechanisms provides a synergistic effect at the animal level.

7.3. Tau

Tau is an especially important siRNA target because it is more closely associated with neurodegeneration and cognitive decline than plaque burden alone, and transcript‐level reduction offers a direct route to lowering total tau output. In this section, the emphasis is therefore on CNS‐adapted siRNA chemistry, durability of knockdown, and early translational progress, rather than on re‐summarizing tau biology, which has already been covered in Section 3.2.

In 2019, Nature Biotechnology reported the use of bivalent siRNA (di‐siRNA), where two fully modified siRNA strands are dimerized using a linker arm (Alterman et al. 2019). A single intrathecal (IT) administration in mice and non‐human primates resulted in widespread, persistent silencing of brain/spinal cord targets, with controlled toxicity and off‐target risks associated with seed regions. This study laid the chemical foundation for CNS‐adapted siRNA. Following this, dual‐target di‐siRNAs capable of simultaneously knocking down two genes with a single molecule emerged. Hoglinger et al. used primary cortical neurons from transgenic mice expressing the P301S mutation of human tau and Lund human midbrain (LUHMES) cells to validate siRNA inhibition in vitro (Xu et al. 2014). Stereotaxic injections of siRNA into the brains of P301S mice showed effective tau suppression both in vitro and in vivo, without the need for specific delivery agents. Following a single injection, siRNA exhibited moderate distribution in the hippocampus of mice, with no adverse effects on safety indicators.

The C16‐siRNA (lipid conjugate) platform, which primarily supports IT administration, allows for sustained, monthly reductions in target levels in the CNS and has been extended to tau targets. Sloan et al. (AAIC 2024) applied MAPT C16‐siRNA, where a single IT injection in P301S mice resulted in sustained reductions in mRNA and protein levels, accompanied by behavioral benefits. C16‐siRNA targeting MAPT (in non‐human primates, single IT 60 mg) showed significant, lasting reductions in MAPT mRNA and tau protein in the prefrontal cortex, hippocampus, and spinal cord for over 4 months, with no abnormal findings in CSF NfL or tissue pathology.

A Phase I clinical study of LY3954068 in early symptomatic AD is currently recruiting participants to evaluate its safety and tolerability (ClinicalTrials.gov identifier: NCT06297590) (Harris et al. 2025).

7.4. APOE

APOE is a high‐value siRNA target because it links genetic risk with therapeutic tractability. In this section, the key issue is not to restate the broad biological influence of APOE across AD pathology, but to evaluate whether CNS‐selective APOE lowering can generate therapeutic benefit without unacceptable systemic consequences. This makes APOE particularly useful for discussing tissue selectivity, depth of knockdown, and the distinction between central and peripheral target engagement.

Belgrad et al. targeted APOE in the CNS using di‐siRNA, showing that when administered before or after the appearance of pathology, it significantly reduced Aβ plaque burden and insoluble Aβ40/42 levels (Belgrad et al. 2024). Meanwhile, silencing liver APOE alone did not improve Aβ in the brain and even increased blood cholesterol levels, indicating that the therapeutic effect is dependent on CNS‐targeting rather than peripheral APOE. The study also demonstrated that APOE‐siRNA activates immune clearance pathways and showed good tolerability in short‐term behavioral and safety assessments. A technical paper from the same team further illustrated how chemical modifications can switch silencing targets between the CNS and liver, supporting the biological conclusion that APOE's effects are centrally mediated and confirming its pharmacological feasibility (Ferguson et al. 2024).

Previous reports of ASOs reducing brain APOE by approximately 50% showed limited improvement in Aβ pathology, whereas siRNA, achieving deeper reductions in APOE in the adult mouse CNS, resulted in significant Aβ improvement (Li and Huang 2025). This suggests that the magnitude of reduction and the location of intervention are key to efficacy; for the tau pathway, a 50% reduction of APOE via ASO already shows an effect, reflecting differences in pathological pathways (Hinrich et al. 2016). The CNS‐adapted chemistry of siRNA (such as di‐siRNA/C16) and tissue‐selective conjugation offers advantages in long‐lasting, adjustable site‐specific targeting (Sorets et al. 2024). However, the choice of administration route—whether intrathecal or intraventricular—and dosing frequency remain key hurdles for clinical translation.

7.5. Microglial Aging, Inflammation, and Autophagy

Beyond canonical amyloid and tau targets, siRNA strategies are increasingly being extended to microglial senescence, inflammatory signaling, and autophagy‐related pathways. Here, we focus on intervention‐ready nodes with in vivo proof‐of‐concept, including p16INK4a, NF‐κB‐related signaling, pyroptosis‐associated mediators, and autophagy regulators.

Kim's team developed carriers such as poly (lactic‐co‐glycolic acid) (PLGA) nanoparticles that preferentially enter microglia in the brain, enabling cell‐type‐specific siRNA delivery (Shin et al. 2024). In patients and 5xFAD mice, p16^INK4a expression was upregulated around plaques in microglia. Using microglia‐targeted PLGA‐siRNA nanoparticles to silence p16^INK4a improved microglial phagocytic capacity, enhanced lysosomal activity (increased LAMP1), reduced Aβ plaques in 5xFAD, decreased the proportion of senescent microglia, and reversed learning and spatial memory deficits.

Furthermore, Wang et al. demonstrated that microglial NF‐κB drives tau propagation and toxicity, whereas its inactivation partially restored microglial homeostasis and rescued spatial learning and memory deficits (Wang et al. 2022). Pahan et al. used NF‐κB inhibitory peptides (NEMO‐binding) administered intranasally in AD models, reducing inflammation and glial responses, providing evidence that this upstream pathway is amenable to localized therapeutic intervention (Rangasamy et al. 2015). These observations support the broader intervention logic for inflammasome‐ and pyroptosis‐related pathways; however, in the context of this review, the key question is whether such nodes have progressed to actionable siRNA‐based modulation in vivo. For direct siRNA interventions, silencing caspase‐1 or GSDMD using siRNA inhibited Aβ1‐42‐induced pyroptosis in neuronal models; at the animal level, multiple reviews and recent studies continue to identify NLRP3/GSDMD as key intervention targets (Hong et al. 2023). siRNA can indirectly activate TFEB by downregulating its inhibitor, such as KEAP1 (which de‐represses Nrf2–antioxidant/lysosomal transcription programs) or the newly reported TFEB transcriptional repressor USF2 (Gu et al. 2018; Li et al. ²⁰²²). Given the pivotal role of autophagy in aggregate clearance, nanomedicine platforms designed to precisely modulate autophagic flux are emerging as powerful therapeutic tools. As reviewed recently, autophagy‐targeting nanomedicines can be engineered to intervene at specific stages of the autophagic process (Sun, Zuo, et al. ²⁰²⁵), a strategy that holds immense promise for restoring microglial homeostasis in AD. The former has shown in vivo evidence in Aβ toxicity models that KEAP1 inhibition leads to neuroprotection (Kim et al. 2012).

8. Artificial Intelligence: Accelerating the Precision of siRNA Therapy in AD

While the potential of siRNA therapeutics for AD is evident, the translation from bench to bedside is frequently stalled by the triad of precision, identifying the right target within a complex network, designing sequences with maximal potency and minimal off‐target effects, and engineering carriers that can efficiently breach the BBB (Paul et al. 2021). Recently, artificial intelligence (AI)—spanning machine learning (ML), deep learning (DL), and generative models—has shown great promise in addressing these challenges. However, the credibility of AI in this context hinges on the clear definition of its intended use, transparent reporting of data provenance, and independent validation in relevant experimental settings.

To avoid overstatement, it is important to distinguish between AI applications that are experimentally validated and those that are primarily hypothesis‐generating or proof‐of‐concept. For siRNA design, many ML models are trained on aggregated in vitro knockdown datasets, often of limited scale and heterogeneous protocols (e.g., differing cell‐type, dosing, readout methods, and time points). This variability can introduce domain shifts and label noise, reducing the generalizability of the models (Martinelli 2024). For delivery optimization, most large datasets and AI models have been built around liver‐ or vaccine‐focused LNPs. However, extrapolating these models to brain delivery is non‐trivial, as BBB transport and in vivo biodistribution depend on additional determinants, such as administration route, protein corona, and neurovascular context (Dorsey et al. 2024; Saber et al. ²⁰²⁴). Accordingly, AI outputs should be treated as design hypotheses until confirmed through orthogonal assays and in vivo models, with reproducibility benefiting from open reporting of datasets, code, and workflows that follow FAIR data principles (Wilkinson et al. 2016).

8.1. Beyond Amyloid: AI‐Driven Discovery of Novel Therapeutic Targets

Traditional AD drug discovery has largely focused on the amyloid cascade hypothesis. However, the high failure rate of clinical trials necessitates a broader search for upstream drivers of pathology. AI approaches, particularly network medicine and multi‐omics integration, excel at deconvoluting the high‐dimensional heterogeneity of AD biology.

Emerging platforms like PandaOmics utilize DL to analyze multi‐omics data, identifying targets that may be overlooked by conventional analysis. For instance, a recent study combining PandaOmics with the FuzDrop algorithm prioritized MARCKS and CAMKK2 as candidate AD‐associated proteins with altered phase‐separation behavior (Lim et al. 2023). Unlike traditional targets, these proteins drive the early biophysical condensation events that precede irreversible aggregation, offering a novel intervention point for siRNA therapy. Similarly, network‐based AI frameworks have re‐evaluated the human interactome, pinpointing kinases like GSK3β and CDK5 as central hubs in the AD molecular network (Fang et al. 2022). While small‐molecule inhibitors for these kinases often lack specificity, siRNA offers the precision required to silence specific isoforms or nodes within these pathogenic networks without widespread toxicity.

Furthermore, AI is also unlocking the potential of cell‐type‐specific interventions. By applying deep learning to single‐cell RNA sequencing (scRNA‐seq) data, researchers have identified transcription factors such as PU.1 (encoded by Spi1) as master regulators of microglial neuroinflammation. In a CK‐p25 mouse model of chronic neuroinflammation localized intracisternal delivery of LNP‐encapsulated anti‐PU.1 siRNA reduced PU.1 expression and neuroinflammatory responses (Ralvenius et al. 2024).

8.2. Generative AI for Rational siRNA Sequence Design

Once a target is identified, the next challenge is designing siRNA sequences that balance potency, specificity, and tolerability. Rule‐based heuristics (e.g., early positional/thermodynamic rules) remain useful for initial triage, but they do not fully capture the non‐linear sequence‐context effects. As a result, ML models, ranging from linear methods to deep neural networks, have been trained to predict siRNA knockdown efficacy based on sequence and engineered features. A key limitation is that available training labels are often aggregated across heterogeneous experimental protocols (e.g., cell line, dose, time point, and readout), which can introduce label noise and impair cross‐study generalization unless rich metadata and external validation sets are available (Martinelli 2024).

For chemically modified siRNAs, the Cm‐siRPred models use a multi‐view learning strategy that integrates double‐strand sequences, chemical modification patterns, and physicochemical descriptors, providing a more explicit framework for modification‐aware activity prediction (Liu, Huang, et al. ²⁰²⁴). Generative models, such as GenerRNA, based on the Transformer architecture, can propose de novo RNA sequences. While these models show great promise, most demonstrations remain computational, and thus require experimental validation to confirm functional activity, immunostimulation, and off‐target liability. In practice, credible AI‐guided siRNA design benefits from curated datasets with standardized efficacy readouts, explicit annotation of chemistry and experimental context, and prospective testing in new targets and delivery settings.

8.3. Breaking the Barrier, AI‐Optimized Delivery Systems

AI has also been applied to delivery design, where the combinatorial chemical space of ionizable lipids, helper lipids, and targeting ligands makes exhaustive experimental screening impractical. Most published AI/ML studies to date focus on LNP formulation and process optimization and on structure–activity relationships of ionizable lipids, often in liver‐ or vaccine‐oriented settings. For CNS delivery, however, training data are comparatively sparse, and the determinants of BBB transport and parenchymal cell‐type distribution introduce additional challenges. Therefore, brain‐specific endpoints (e.g., brain exposure, cell‐type expression/knockdown and safety readouts) are needed for model generalization (Haghighi et al. 2024; Saber et al. ²⁰²⁴). A landmark advance is the LiON (Lipid Optimization using Neural networks) platform, which uses large LNP activity datasets and deep learning to predict nucleic acid delivery from lipid structure and guide the discovery of new ionizable lipids. Although developed for pulmonary delivery, this approach illustrates how data‐driven lipid design may eventually be adapted to CNS delivery after validation in brain‐relevant models. Validating this AI‐driven approach, a 2025 study published in ACS Nano demonstrated the power of AI in predicting BBB permeability. The researchers used a neural network to screen small‐molecules and identified acetylcholine as a high‐probability ligand for BBB transport. Conjugating acetylcholine to LNPs resulted in superior brain accumulation and specific transfection of neurons and astrocytes in mice, validating the AI model's predictive power (Sela et al. 2025). Additionally, related computational efforts are underway to identify or design BBB‐penetrating peptides using tools like DeepB3Pred (stacked BiGRU) or Augur (data augmentation), although these tools remain primarily predictive/design frameworks (Arif et al. 2025; Gu et al. ²⁰²⁴). When used as delivery shuttles for oligonucleotides, these peptides still require rigorous in vivo validation and safety assessments (Saber et al. 2024).

Despite these advances, common pitfalls include overfitting to assay‐specific artifacts, data leakage between training and test sets, and poor generalization under protocol heterogeneity or domain shift, highlighting the need for external validation and prospective testing (Dorsey et al. 2024; Martinelli ²⁰²⁴). When AI‐generated analyses are intended to support regulatory decision‐making regarding drug quality, safety, or effectiveness, the FDA recommends a risk‐based credibility assessment, anchored to a well‐defined context of use, with documentation of data provenance, model development, performance, and limitations. Similarly, the EMA emphasizes transparency, governance, and lifecycle management of AI/ML applications used in the medicinal product lifecycle. Adopting FAIR data principles will further improve the traceability, reproducibility, and reuse of datasets and workflows.

8.4. Precision Medicine and AI‐Enabled Patient Stratification for Brain‐Targeted siRNA Therapies

AD is increasingly viewed as a biologically defined continuum rather than a single, homogeneous clinical entity. The 2018 NIA‐AA research framework formalized this concept by introducing the AT(N) system in which amyloid‐β (A), pathologic tau (T), and neurodegeneration (N) biomarkers jointly define AD biology, largely decoupled from syndromic stage (Jack et al. 2018). More recently, high‐throughput multi‐omics analyses using ML have revealed at least four molecularly distinct AD profiles that differ in cognitive trajectories, survival time, degree of neurodegeneration, astrogliosis, and dysregulation of synaptic, endocytic, and mTOR pathways, highlighting that AT(N) alone does not fully capture disease heterogeneity (Eteleeb et al. 2024). These data provide a biological foundation for precision‐medicine approaches that match specific therapeutic mechanisms to well‐defined patient subgroups.

Current anti‐amyloid monoclonal antibodies, such as lecanemab and donanemab, have already implemented elements of precision medicine by restricting enrolment to amyloid‐positive individuals with early symptomatic AD. For example, donanemab showed stronger clinical benefit in participants with low to medium baseline tau PET signals compared to those with high tau burden, illustrating how biomarker‐based stratification can modulate observed treatment effects (Sims, Zimmer, Evans, et al. ²⁰²³). At the same time, both efficacy and safety vary based on genotype. APOE ε4 is the strongest common genetic risk factor for late‐onset AD and ε4 carriers have an increased risk of amyloid‐related imaging abnormalities (ARIA) during anti‐amyloid treatment (Rajič Bumber et al. 2025). These observations support the integration of APOE genotyping, AT(N) status, and blood‐based biomarkers into precision medicine workflows for AD individualized benefit–risk assessment and dosing.

For brain‐targeted siRNA nanomedicines, the same logic can be extended beyond amyloid. Multi‐omics clustering demonstrates that some AD molecular profiles are dominated by synaptic dysfunction and endocytosis, whereas others feature pronounced neuroinflammation, astrogliosis, or metabolic dysregulation (Eteleeb et al. 2024). In principle, siRNA payloads could be designed to selectively silence tau kinases, microglial activation pathways, endocytic regulators, or autophagy‐related genes in those patients whose molecular signatures indicate that these processes are the main disease drivers, while sparing individuals whose pathology is driven by alternative mechanisms. Within such a framework, brain‐targeted delivery systems would be paired with patient selection based on AT(N) status, APOE and other risk genotypes, and assignment to multi‐omics–defined molecular subtypes, so that each siRNA “module” is deployed only in the biologically appropriate subgroup.

AI and machine learning tools are particularly well‐suited for enabling this level of patient stratification. Predictive prognostic models that integrate multimodal data—clinical scales, structural MRI, amyloid and tau PET, APOE status, and other biomarkers—have been used to derive individualized risk indices and categorize patients into slow‐ and fast‐progressing trajectories. A recent AI‐guided re‐analysis of the AMARANTH lanabecestat trial used such modeling to re‐stratify participants, showing that prognostic‐index–based enrichment can improve the ability to detect treatment effects and substantially reduce the required sample size, even in a trial originally deemed futile (Vaghari et al. 2025). More broadly, dedicated frameworks for AI‐driven drug discovery and trial optimization in dementia are being developed to refine inclusion/exclusion criteria, prioritize biomarker modalities, and inform go/no‐go decisions for candidate therapies.

In parallel, surveys of AI for personalized prediction of AD progression emphasize the role of digital‐twin‐like models that simulate individual trajectories under different hypothetical interventions using longitudinal, multimodal data (Koksalmis et al. 2025). As blood‐based AT(N) markers, genotyping, multi‐omics signatures, and digital biomarkers from wearable devices become more widely available, these AI systems will help estimate which patients are most likely to benefit from a given brain‐targeted siRNA construct, determine the optimal timing for intervention (e.g., during preclinical or prodromal stages), and update treatment plans as new data emerge. Ultimately, converging AT(N) classification, APOE and other genetic risk factors, multi‐omics–defined endotypes, and AI‐derived prognostic indices within a unified decision framework will be critical for translating brain‐targeted siRNA nanomedicines into truly precision‐guided therapies for AD.

9. Safety

A comprehensive review of the existing evidence from a safety perspective reveals that siRNA‐based therapeutic strategies for AD have gradually formed a relatively clear risk–benefit framework.

9.1. Delivery‐Associated Risks: Immunogenicity and Neurotoxicity

In terms of delivery routes, three main pathways are being explored, including intravenous delivery platforms that cross the BBB, IT or ICV administration, and NtB routes. For BBB‐crossing systems, preclinical studies have explored early‐stage gate‐controlled transport, and newer technologies such as receptor‐mediated transmembrane domain (TMD) allosteric targeting, brain microcirculation regulation (e.g., VIP‐derived ionizable lipids), and neuroreceptor targeting (e.g., D3R), showing their potential to improve systemic brain delivery. Corresponding safety evaluations are increasingly standardized, including routine monitoring of body weight, behavior, liver and kidney function, inflammatory markers, complement activation (CARPA), major organ histology (H&E staining), and brain distribution or pathology. In individual studies, tolerability findings should be interpreted in a platform‐, dose‐, route‐, and species‐specific manner (Mi et al. 2025). The IT or ICV delivery has demonstrated sufficient central CNS exposure and strong therapeutic effects in mice and non‐human primates, with single doses of siRNA resulting in sustained target inhibition over several months (Brown et al. 2022). CSF neurofilament light chain (NfL) may be monitored as a marker of neuroaxonal injury (Blandino et al. 2024). NtB delivery may reduce systemic exposure while providing direct access to the brain, and has shown feasibility in preclinical delivery and combination‐therapy studies (Rassu et al. 2017; Yang et al. ²⁰²²).

Class‐wide risks primarily arise from innate immune activation, miRNA‐like seed‐mediated off‐target effects, and delivery‐system–related systemic reactions (Flynn et al. 2022). The former can be significantly mitigated through chemical modifications such as 2′‐O‐Me, 2′‐F, or phosphorothioate modifications, and sequence engineering (e.g., reducing seed region pairing stability, site‐specific modifications, controlling exposure levels), complemented by in vitro transcriptome and cytokine screening. The latter requires systematic quantification of immune, coagulation, and complement responses during formulation development, with attention to antibody/IgM complex deposition and temperature responses during repeated dosing. New‐generation brain‐targeting LNPs have shown controllable reductions in immunogenicity and CARPA risks.

It is also important to note that local CNS delivery presents technical risks of rapid‐onset neurotoxicity. Animal studies show that the introduction of Ca²⁺/Mg²⁺ into formulation significantly reduces transient adverse reactions without compromising distribution and efficacy, suggesting operational protective strategies at the formulation level (Miller et al. 2024).

9.2. Target‐Specific On‐Target Toxicity

When evaluating target safety, BACE1 remains the most established, yet cautionary target. Clinical experience with small molecule inhibitors of BACE1 suggests that excessive or prolonged inhibition may be associated with mild, reversible cognitive decline (McDade et al. ²⁰²¹). This indicates the need for siRNA strategies to avoid excessive silencing, instead focusing on partial inhibition, intermittent dosing, and spatial and cell‐type‐specific targeting. Safety boundaries can be dynamically monitored using imaging (Aβ/Tau‐PET), fluid biomarkers (Aβ products, p‐tau, NfL/GFAP), and cognitive assessments.

Target‐specific safety considerations involve balancing pharmacological benefits and physiological function. MAPT‐targeting siRNA has shown durable target knockdown and acceptable tolerability in preclinical CNS studies, including non‐human primates. The first‐in‐human intrathecal study of LY3954068 in early symptomatic Alzheimer’s disease is ongoing; therefore, human pharmacodynamic and safety outcomes remain to be established (ClinicalTrials.gov identifier: NCT06297590). For the APP axis, over‐interference with physiological cleavage products should be avoided, with dose escalation and regional targeting strategies ensuring safety windows. In preclinical models, reducing ApoE in the brain alone provides therapeutic benefit without disturbing blood lipid levels, though transcriptional changes in central immune pathways suggest that glial‐related indicators should be monitored during follow‐up.

9.3. Monitoring Strategies for Clinical Translation

The monitoring system must align with the delivery pathway. For IT or ICV administration, it is recommended to focus on NfL/GFAP biomarkers, alongside monitoring injection rate and needle specifications. For intravenous‐to‐brain platforms, routine biochemical, inflammatory, complement panels, and multi‐organ H&E should be incorporated, with immune monitoring added for long‐term repeated dosing. Nasal‐to‐brain routes require evaluation of both mucosal tolerance and long‐term local safety.

A reusable safety workflow should be established during development. Initially, the focus should be on de‐immunization chemistry, sequence engineering, and in vitro off‐target and immune screening. During animal studies, path‐matching pharmacodynamics and safety readouts (including NfL/GFAP, complement/coagulation panels, and histology) can be used to iteratively refine formulations and dosages. At the target level, partial inhibition, individualized frequency, and spatially directed strategies can help to alleviate target risks. During the clinical transition, multimodal safety biomarkers, cognitive assessments, and imaging co‐endpoints should be employed for comprehensive monitoring. Overall, the current evidence indicates that with collaborative optimization of chemical modifications, sequence design, delivery formulations, and monitoring systems, siRNA in AD represents a manageable, verifiable, and iterative therapeutic strategy, with a developing framework for further clinical evaluation alongside efficacy.

10. Challenges and Prospects

Despite the growing feasibility of CNS gene silencing and the emergence of early human data, the translation of siRNA into a practical therapy for AD still faces several interconnected challenges. These challenges arise not only from delivery barriers, but also from the biological complexity of AD, the pharmacological properties of siRNA itself, and the difficulty of translating encouraging animal data into clinically meaningful benefit in humans. Accordingly, the future of siRNA therapy for AD will depend on whether these challenges can be addressed through target stratification, chemical optimization, delivery innovation, and more rigorous translational frameworks.

First, one major challenge lies in the biology of AD itself. AD is not a uniform disease but a heterogeneous syndrome with substantial temporal, spatial, and cell‐type variability (Eteleeb et al. 2024; Jack et al. ²⁰¹⁸). A target that is relevant in one disease stage or patient subgroup may be much less important in another. For example, early intervention may be better aligned with APP processing and Aβ generation, whereas later‐stage disease may require stronger emphasis on tau pathology, network degeneration, neuroinflammation, and lysosomal‐autophagy dysfunction. This heterogeneity makes it difficult to define a single universal siRNA strategy. A practical solution is to move toward biology‐informed stratification, in which siRNA targets are matched to disease stage, molecular phenotype, and risk background, including APOE genotype, CAA burden, and inflammatory status (Eteleeb et al. 2024; Rajič Bumber et al. ²⁰²⁵).

Second, siRNA itself presents several pharmacological challenges. Although siRNA offers highly selective post‐transcriptional gene silencing, its therapeutic window can still be limited by seed‐mediated off‐target effects, innate immune activation, and uncertainties surrounding long‐term or repeated dosing (Flynn et al. 2022). These issues are especially important in the CNS, where adverse effects may be subtle, cumulative, and difficult to detect in early studies. Chemical optimization provides an important route forward. Modifications such as 2′‐O‐Me, 2′‐F, phosphorothioate incorporation, and ligand conjugation have already improved stability, tolerability, and durability (Flynn et al. 2022). However, these advances must be accompanied by systematic transcriptomic, immunologic, and pharmacodynamic monitoring, particularly for chronic dosing regimens and combination‐target strategies.

Third, delivery remains the central bottleneck determining both efficacy and scalability. For AD, local CSF‐directed administration currently represents the most mature route, because intrathecal or intracerebroventricular delivery can bypass the BBB and achieve broad CNS exposure when combined with long‐acting chemistries such as bivalent di‐siRNA and C16‐conjugated siRNA (Brown et al. 2022; Cohen et al. ²⁰²³). These platforms have already shown sustained knockdown in rodents and non‐human primates, supporting the feasibility of low‐frequency dosing. However, local delivery is still invasive and may not be ideal for large patient populations. Therefore, systemic BBB‐crossing delivery remains a major unmet need. Here, several promising solution pathways are emerging, including receptor‐mediated transcytosis, oligonucleotide transport vehicles, γ‐secretase‐mediated BBB‐crossing conjugates, and brain‐targeted nanocarriers such as TfR‐guided, or other biomimetic lipid systems (Barker et al. 2024; Bian et al. ^{2025, 2024}; Wang et al. ²⁰²⁵; Wells et al. ²⁰²⁵; Zhou et al. ²⁰²⁰). These strategies suggest that systemic delivery may eventually become feasible, but their reproducibility, manufacturability, and robustness across species still require careful validation. At the same time, transport‐vehicle and nanocarrier paradigms solve different parts of the CNS delivery problem and therefore should be viewed as complementary rather than interchangeable. Chemically defined OTVs and BCCs offer tighter control of stoichiometry and pharmacokinetics and have shown relatively uniform CNS biodistribution after intravenous administration, but receptor engagement introduces potential peripheral sink effects, affinity windows, and saturation constraints. By contrast, nanocarriers can package higher siRNA payloads and incorporate endosomolytic or multifunctional components, yet they often face greater barriers to parenchymal access and remain strongly dependent on efficient endosomal escape for cytosolic RNAi. Together, these modalities provide transferable design lessons regarding affinity tuning, peripheral biodistribution mapping, cell‐type‐resolved exposure, and the matching of transport strategy to siRNA pharmacology (Barker et al. 2024; Wang et al. ²⁰²⁵).

Fourth, translation from animal models to humans remains a major challenge. Many delivery systems show strong effects in rodents, but rodent BBB physiology, receptor abundance, immune tone, and disease evolution differ substantially from those in humans. As a result, efficient brain uptake or therapeutic benefit in mice does not necessarily predict human performance. This is particularly relevant for receptor‐targeted systems, in which species‐dependent affinity, peripheral sink effects, receptor saturation, and dose scaling can strongly influence CNS exposure (Barker et al. 2024; Wang et al. ²⁰²⁵). To address this problem, future development should rely more heavily on large‐animal validation, quantitative PK/PD bridging, cell‐type–resolved biodistribution studies, and standardized comparisons across routes of administration. Without such translational discipline, apparently successful preclinical platforms may fail to generalize beyond proof‐of‐concept settings.

Fifth, safety evaluation and therapeutic index must be addressed more rigorously. Clinical experience with BACE1 inhibition has already shown that excessive target suppression can produce unwanted cognitive effects, underscoring the need for a controlled rather than maximal knockdown strategy. Similarly, high‐concentration oligonucleotides delivered into the CSF may cause acute neurotoxicity, even when the target itself is valid (Miller et al. 2024). Potential solutions include partial rather than complete target reduction, formulation optimization, improved ionic composition for CSF‐administered products, and longitudinal cognitive as well as functional monitoring (Blandino et al. 2024; McDade et al. ²⁰²¹; Miller et al. ²⁰²⁴). In AD, safety cannot be judged solely by short‐term tolerability; it must also include long‐term neural function, repeated‐dose effects, and the consequences of perturbing targets that retain physiological roles in the adult brain.

Sixth, biomarkers and clinical trial design are becoming essential determinants of success. Because AD progresses slowly and exhibits marked biological heterogeneity, conventional trial designs may be too insensitive or inefficient for siRNA therapeutics. The key challenge is to link siRNA sequence, chemistry, and delivery properties to measurable biological and clinical pharmacodynamic fingerprints. Potential solutions include integrating AT(N)‐based staging, blood and CSF biomarkers, digital imaging, and longitudinal multimodal monitoring into trial design (Albert et al. 2011; Jack et al. ²⁰¹⁸; Triana‐Baltzer et al. ^{2020, 2021}). Such approaches may help identify target engagement earlier, enrich for responsive populations, and improve decisions about dose, duration, and combination‐therapy. This will be especially important if siRNA is ultimately used alongside antibody‐based clearance strategies or in genetically defined subgroups such as APOE ε4 carriers (Rajič Bumber et al. 2025).

Finally, chemistry, manufacturing, and controls (CMC) remain decisive for real‐world translation. Even when therapeutic rationale and preclinical efficacy are strong, brain‐targeted RNA formulations often face difficulties in batch consistency, scale‐up, storage stability, and coupling of quality attributes to in vivo behavior. This is particularly relevant for complex BBB‐crossing LNPs and ligand‐functionalized nanocarriers. Standardized raw‐material characterization, closed and continuous manufacturing, and QbD‐guided integration of formulation parameters with PK/PD behavior will be essential if siRNA therapies are to move from experimental promise to accessible medicines.

In summary, the future of siRNA therapy for AD depends on solving four linked problems: matching targets to disease biology, controlling siRNA pharmacology and safety, achieving reliable CNS delivery, and building translational frameworks that bridge animal data to human benefit. Current progress in long‐acting intrathecal chemistries, systemic BBB‐crossing transport vehicles, and biomarker‐guided clinical evaluation provides genuine reason for optimism. However, the field will advance most effectively if it treats delivery, disease stratification, safety, and translation not as separate topics, but as parts of a single therapeutic design problem. A structured translational pathway and key decision points are summarized in Figure 6.

FIGURE 6.

Key translational considerations for siRNA therapeutics in AD. Overview of key development tracks and decision points, including dosing route selection, manufacturability/CMC, biodistribution and pharmacodynamic readouts, and safety/clinical trial considerations. CMC, chemistry, manufacturing and controls; IT, intrathecal; NHP, non‐human primate; POC, proof of concept.

11. Conclusion

The emergence of siRNA as a clinically validated drug modality has opened a new chapter in the treatment of genetically and molecularly defined diseases. However, for AD, the path to translation remains uniquely challenging because effective therapies must not only navigate the BBB but also address a widely distributed and slowly progressive pathology over many years, all while ensuring safety. In this review, we have outlined how advances in AD pathobiology, siRNA chemistry, and nanotechnology are beginning to converge on this complex problem. Human evidence is currently strongest for intrathecal delivery of long‐acting oligonucleotide and siRNA therapeutics. In contrast, systemic brain‐targeted platforms—including lipid conjugates, brain‐tropic LNPs, polymeric and hybrid carriers, and biological vectors such as exosomes—remain emerging and are largely at the preclinical or early translational stage.

At the same time, growing experience with RNA therapeutics in non‐CNS indications is clarifying class‐wide safety considerations and regulatory expectations. These lessons, coupled with mechanistic insights into AD heterogeneity, point toward a development paradigm in which siRNA therapies are designed as modular combinations of targets and delivery systems, embedded within precision‐medicine frameworks. AI can accelerate each step of this process, from target prioritization and siRNA sequence design to formulation optimization, patient stratification, and adaptive trial design.

Looking forward, the successful integration of brain‐targeted siRNA into clinical practice for AD will depend on coordinated progress in several domains: robust and scalable manufacturing of complex nanomedicines; harmonized safety monitoring and long‐term pharmacovigilance; equitable access and reimbursement models for chronic RNA‐based therapies; and continued investment in biomarker infrastructure to support individualized dosing and follow‐up. If these scientific, technological, and translational pieces can be effectively assembled, siRNA‐based interventions may ultimately evolve from experimental tools into a durable, programmable cornerstone of AD therapy—used not in isolation, but in rational combination with antibodies, small molecules, and lifestyle interventions as part of long‐term, precision disease management.

Author Contributions

Qingshan Yang, Zhouchun Chen, and Jianchao Zhu: contributed equally to this work. Fengjie Tang: visualization, writing – original draft. Jianchao Zhu: writing – original draft. Qingshan Yang: conceptualization, investigation, formal analysis, visualization, writing – original draft, writing – review and editing. Yang Liu: conceptualization, writing – review and editing, investigation, funding acquisition. Mengyu Fan: writing – original draft, visualization. Zhouchun Chen: investigation, writing – original draft. Tianran Chai: writing – original draft, resources. Shuangxi Chen: supervision, conceptualization, writing – review and editing. Meng Zheng: conceptualization, writing – review and editing, supervision, funding acquisition, project administration. Yundong Li: investigation, validation.

Funding

This work was supported by the Program of Technology Innovation Team of the Colleges and Universities of Henan Province (25IRTSTHN032), the Henan Provincial Health Commission Medical Science and Technology Research Program (Joint Provincial‐Ministerial Project SBGJ202503048), the China Postdoctoral Science Foundation (2025M780051), and the National Natural Science Foundation of China (32601935).

Conflicts of Interest

The authors declare no conflicts of interest.

Yang, Q. , Chen Z., Zhu J., et al. 2026. “Brain‐Targeting siRNA Delivery to Tackle Alzheimer's Disease.” Wiley Interdisciplinary Reviews: Nanomedicine and Nanobiotechnology 18, no. 5: e70076. 10.1002/wnan.70076.

Data Availability Statement

Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.