A computational model-powered platform to inform the development of GalNAc-conjugated siRNA therapeutics

Abstract

N-acetylgalactosamine-conjugated small interfering RNA (GalNAc-siRNA) therapeutics have emerged as a groundbreaking modality with unparalleled efficacy for battling previously “undruggable” diseases. The unique pharmacokinetic (PK) and pharmacodynamic (PD) characteristics of GalNAc-siRNA therapeutics provide an opportunity to leverage PK/PD modeling strategies for drug development. By utilizing the wealth of literature data, we developed and validated a mechanistic computational model-driven platform to guide the development of new GalNAc-siRNA therapeutics, optimizing their clinical translation. This platform integrates preclinical and clinical data from all seven FDA-approved GalNAc-siRNA drugs—fitusiran, givosiran, inclisiran, lumasiran, vutrisiran, nedosiran, and plozasiran—spanning multiple species (mouse, rat, monkey, and human). To enhance user accessibility, we further implemented a web-based Shiny application. The platform was used to inform the development of an investigational new angiotensinogen-silencing GalNAc-siRNA (SAL0132). Multiple PK/PD datasets from rats and monkeys were satisfactorily fitted, and extrapolated to humans. The platform successfully predicted the PK and simulated the PD profiles of SAL0132 in humans, which demonstrated model-informed strategies to support efficient drug development of this modality. In conclusion, this platform enables users to predict GalNAc-siRNA PK/PD profiles across species by inputting specific model parameters, providing a powerful resource to guide the development of next-generation GalNAc-siRNA therapeutics.

Graphical abstract

Yan and colleagues developed a computational model-powered, user-friendly platform for GalNAc-siRNA therapeutics. The platform was applied to facilitate the development of a novel GalNAc-siRNA, SAL0132, demonstrating the reliability and advantages of this platform in predicting drug behavior across species and accelerating clinical translation, especially for non-specialist users and clinicians.

Introduction

N-acetylgalactosamine-conjugated small interfering RNA (GalNAc-siRNA) therapeutics have revolutionized the landscape of modern medicine, particularly in the treatment of genetic and metabolic disorders. These therapeutics leverage the RNA interference (RNAi) mechanism to selectively silence the expression of disease-causing genes, offering a level of specificity and precision to target intracellular and “undruggable” targets that is unparalleled by conventional small-molecule drugs or biologics.^1,2 The conjugation of siRNA molecules with GalNAc has been a pivotal advancement, enabling targeted delivery to hepatocytes via asialoglycoprotein receptors (ASGPRs) predominantly expressed in the liver.³ This targeted delivery not only enhances the therapeutic index but also minimizes off-target effects, making GalNAc-siRNA a safer and more effective option for patients.

The clinical success of 7 FDA-approved GalNAc-siRNA drugs to date, including givosiran (2019), lumasiran (2020), inclisiran (2021), vutrisiran (2022), nedosiran (2023), fitusiran (2025), and plozasiran (2025), underscores their transformative potential (Figure 1A). These drugs have demonstrated remarkable efficacy in reducing disease biomarkers, with prolonged pharmacodynamic (PD) effects that allow for infrequent dosing schedules.⁴ This feature significantly improves patient compliance and quality of life, particularly in chronic conditions requiring long-term treatment. As the pipeline of GalNAc-siRNA therapeutics continues to expand, their significance in addressing unmet medical needs becomes increasingly evident.

Figure 1.

A workflow of this study

(A) Overview of the mechanism of action of N-acetylgalactosamine-small interfering RNA (GalNAc-siRNA) and the seven approved GalNAc-SiRNA drugs collected in this study. (B) Schematic structure of the developed computational model for a GalNAc-siRNA. (C) Schematic workflow of the model informed GalNAc-siRNA development powered by the established computational platform.

One of the defining features of GalNAc-siRNA therapeutics is their unique and highly predictable pharmacokinetic (PK) and PD profiles.^4,5 Unlike traditional small molecules, which often exhibit complex distribution and metabolism patterns, GalNAc-siRNA drugs benefit from a well-characterized mechanism of hepatic uptake via ASGPRs. This results in consistent and efficient delivery to the liver, where the RNAi machinery can effectively silence the expression of target genes. The predictability of their PK/PD behavior across different molecules within this class is a critical advantage, enabling streamlined drug development processes.⁶ Moreover, GalNAc-siRNA therapeutics exhibit prolonged PD effects due to their ability to sustain gene silencing over extended periods. This is attributed to the stability of the RNA-induced silencing complex (RISC) within hepatocytes, which allows for durable therapeutic effects even after the drug has been cleared from systemic circulation.⁷ Such characteristics not only reduce the frequency of dosing but also simplify clinical management, making these drugs particularly appealing for both patients and healthcare providers.

The unique attributes of GalNAc-siRNA therapeutics present a compelling case for the application of model-informed drug development (MIDD) strategies. MIDD integrates quantitative modeling and simulation approaches to inform decision-making throughout the drug development life cycle.⁸ For GalNAc-siRNA drugs, the availability of rich preclinical and clinical data, coupled with their predictable PK/PD profiles, provides an ideal foundation for the development of robust mechanistic models. Recently, several PK/PD models were established to assess the preclinical and clinical PK-PD behavior of GalNAc-siRNA therapeutics.^9,10,11,12 Ayyar et al. utilized literature data and established a minimal physiologically based-PK/PD (mPBPK/PD) model of fitusiran in preclinical species and humans.⁹ Tian et al. constructed a similar mPBPK/PD model for a newly developed GalNAc-siRNA to support its efficient development, evaluation, and clinical application.¹¹ Another mechanistic PK/PD model was developed to quantify ASGPR-mediated disposition and downstream pharmacology of givosiran.¹⁰ These models incorporate detailed physiological and molecular mechanisms, enabling the prediction of drug behavior across different species and patient populations. By leveraging such models, developers can optimize dosing regimens, identify potential safety concerns, and accelerate the clinical development of new GalNAc-siRNA candidates. Additionally, MIDD strategies can reduce the need for extensive empirical testing, thereby saving time and resources while maintaining scientific rigor. However, these models are limited to single molecules, and the complexity of these models has limited their accessibility, highlighting the need for user-friendly tools that can democratize their application in the broader scientific community. It would be valuable to develop an R Shiny application that leverages the common properties of GalNAc-siRNAs to provide further insight into their development.

The aim of this study is to develop a refined mechanistic computational-powered platform to inform the development of GalNAc-siRNA therapeutics (Figure 1B). By integrating preclinical and clinical data from all 7 FDA-approved GalNAc-siRNA drugs, this platform seeks to establish a comprehensive framework for translational drug development (Figure 1C). Furthermore, the platform will be implemented as a user-friendly web-based Shiny application, enabling researchers to predict PK/PD profiles across species by inputting specific model parameters.¹³ Finally, the platform was utilized to support the development of an investigational new angiotensinogen (AGT)-silencing GalNAc-siRNA, termed SAL0132 (developed by Shenzhen Salubris Pharmaceuticals Co., Ltd), from preclinic rats and monkey to clinical studies. This study facilitates the efficient and precise development of next-generation GalNAc-siRNA therapeutics while establishing a robust platform that offers a scientifically grounded approach to advancing GalNAc-siRNA-based medicine.

Results

Successful development of the computational model

The developed model (Figure 1B) was simultaneously fitted to multiple PK and PD datasets in mice, including plasma and liver fitusiran measurements, as well as target mRNA and protein change profiles. Input parameters for GalNAc-siRNA in mice are listed in Table 1. Key parameters related to ASGPR binding, turnover, and uptake were fixed, while others were estimated. Due to the lack of plasma IV disposition profiles, SC bioavailability (F) was fixed at 90% (0.9) across species based on prior studies.¹⁰ This value was a conservative estimate based on previous reports across other GalNAc-siRNAs reporting nearly complete SC bioavailability.¹⁴ All model parameters were well-estimated, with relative standard errors (RSE) below 50%. The observed vs. predicted goodness-of-fit (GOF) plots for each model fitting (Figure S1) showed a random normal scatter around the identity line. The general trend and the observed variability were well captured, confirming the model’s adequacy to describe the data.

Table 1.

Model estimates of the PK and PD parameters of GalNAc-SiRNA in mice, rats, monkeys, and humans together with their %RSE

Parameters (units)	Description	Micea	Ratsb	Monkeysc	Humansa
PK
ka (1/h)	absorption rate constant of SC	1.48 (21.16)	0.71 (9.45)	0.38	0.23 (14.58)
Fu (1/h)	elimination rate constant from plasma	0.0066(28.65)	0.21 (9.19)	0.39	39.32 (7.76)
PSLiver (L/h)	membrane permeability in the liver	0.96 (20.43)	211 (45.69)	56.73	144.7 (18.19)
KPLiver	liver to plasma partition coefficient	1.02 (41.36)	3.19 (26.7)	1.02	0.18 (35.57)
KPRem	remainder to plasma partition coefficient	2.04 (31.62)	1.85 (8.08)	2.04	4.24 (14.5)
KDEGD (1/h)	rate constant for free cytosolic siRNA degradation	0.012 (19.3)	0.040(34.3)	0.0012	0.014 (35.72)
RTOT0 (nM)	total RISC concentration in hepatocytes	33.1 (17.37)	33.1	33.1	14.34 (20.89)
Vm (nmol/h)	Michaelis-Menten maximum elimination rate	0.98 (32.64)	1.94(19.49)	5.81	4.37 (30.96)
Km (∗105 nM)	Michaelis constant	27.44 (25.88)	37.73(31.58)	27.44	97.75 (24.86)
koff (1/h)	dissociation rate constant for GalNAc to ASGPR	0.021 (42.28)	0.021	0.021	0.023 (22.93)
kon (nM−1·h−1)	association rate constant for GalNAc to ASGPR	Calculated as kon = koff/KDd
fesc	fraction of siRNA in endosome escape into cytosol	0.019 (31.49)	0.01948	0.019	0.037 (22.9)
kint (1/h)	internalization rate constant for bound ASGPR	7.84 (12.91)	1215 (14.57)	2.01	4.69 (15.56)
ASGPR0 (nM)	baseline ASGPR density in liver	633e	513e	339e	340e
kdeg, ASG (1/h)	degradation rate constant for ASGPR	0.04e	0.04e	0.04e	0.04e
Kcle (1/h)	rate constant for GalNAc cleavage from siRNA	1.32e	1.32e	1.32e	1.32e
Ksyn (nM/h)	synthesis rate for ASGPR	Calculated as ksyn = ASGPR0 · kdeg,ASG
Kass (∗10−6 h−1)	liver association rate constant for binding	0.013 (30.51)	0.0051(56.85)	0.013	0.18 (27.71)
Kdis (∗10−3 h−1)	liver dissociation rate constant for binding	0.14 (25.9)	0.073 (65.45)	0.14	0.0061 (45.37)
KDEGE (∗10−6 h−1)	degradation rate constant for endosomal siRNA	0.0083(49.81)	0.014 (55.27)	0.0083	0.0076 (59)
PD
kdeg, m (1/h)	degradation rate constant for target mRNA	0.0012(13.08)	0.0022(24.46)	0.0015(11.01)	0.00016 (10.1)
kdeg, p (1/h)	degradation rate constant for target protein	0.039 (10.71)	-f	0.083(16.34)	0.037 (31.45)
Smax	maximal stimulation of mRNA degradation	33.79 (33.73)	66.48 (53.88)	33.79	33.79
SC50 (nM)	bound RISC at half maximal effect	12.74 (47.98)	0.050 (56.8)	20.93(37.33)	0.0050 (16.1)
γ1	Hill factor for mRNA-protein translation	0.54 (42.43)	-f	1.54 (55.8)	0.074 (27)
γ2	Hill factor for protein translation	1.79 (8.01)	-f	0.39 (25.67)	0.72 (9.93)

^aThe parameters were estimated from fitusiran. RSE for ω and σ are reported on the approximate S.D. scale (standard error/variance estimate)/2.

^bThe parameters were estimated from givosiran.

^cThe PK parameters were scaled from mice due to the lack of PK data of fitusiran in monkeys. The PD parameters were estimated.

^dKD = 2.48 nM according to the literature.

^eFixed to the literature value.

^fNot available due to the lack of target protein data.

As illustrated in Figures 2A and 2B, the model effectively captured the initial steep, rapid decline in plasma siRNA concentrations post-dosing (Figure 2A), consistent with the short plasma half-life characteristic of GalNAc-siRNAs. The estimated endosomal siRNA degradation rate constant (KDEGE) was 0.0083 h⁻¹, indicating the prolonged escape of siRNA from endosomes into the cytoplasm. While previous studies fixed the endosome escape fraction of siRNA (fesc) at 0.01,^9,10,11 our model estimated fesc at 0.019 through fitting to the data. This estimated value falls within the range of 1%–2% reported in quantitative intracellular trafficking studies of siRNA delivery systems.¹⁵ This consistency further validates the accuracy of the established model. Figure 2B shows the observed and model-predicted PK of fitusiran in mouse liver after SC doses of 1–5 mg/kg. The proposed computational model accurately described fitusiran exposure in the liver over time, demonstrating its reliability in capturing liver PK profiles.

Figure 2.

Observed and fitted PK and PD profiles for fitusiran in mice, monkey, and human

The results are shown in mice (A–E), monkeys (F–H), and human (I–L), respectively. Solid lines represent model fitted results, and symbols indicate observed data points. Different colors represent different dosing regimens. Model-based characterization of (A) plasma concentration after single doses in mice. (B) Liver concentration after single doses in mice. (C) mRNA (%) after single doses in mice. (D) Protein (%) after single doses in mice. (E) Protein (%) after multiple doses in mice once weekly (QW) × 5 doses. (F) Protein (%) after single doses in monkeys. (G) Protein (%) after multiple doses in monkeys at 0.5 mg/kg QW × 8 doses + 0.125 mg/kg × 12 doses, 1 mg/kg Q2W × 4 doses + 0.25 mg/kg QW × 12 doses, and 1.5 mg/kg QW × 5 doses. (H) Protein (%) after multiple doses in monkeys QW × 6 doses. (I) Plasma concentration after single doses in humans. (J) Protein (%) after multiple doses in humans QW × 3 doses. (K) Protein (%) after multiple doses in humans once monthly (QM) × 3 doses. (L) Protein (%) after multiple doses in humans QM × 3 doses.

A sequential modeling approach was employed, where the typical PK parameters derived from the PK modeling were used to drive the PD effects. As depicted in Figures 2C–2E, the model effectively described the mRNA PD profiles (Figure 2C) and protein dynamics (Figures 2D and 2E) in mice under both single and multiple dosing regimens. Notably, hysteresis among target mRNA and protein was successfully described using a tandem, transit compartment, and precursor-like indirect response model. The model accounted for the turnover rates of target mRNA (0.0012 h⁻¹) and protein (0.039 h⁻¹), providing a robust representation of the observed dynamics.

The model structure developed in mice was physiologically scaled to capture PD datasets in monkeys and PKPD datasets in humans. The expansion across species retained identical compartment structures and parameters, with adjustments made to account for species-specific differences in parameter values. This approach offers significant advantages in PBPK model development, as it enables the use of shared parameters from data-rich species, such as rodents, to extrapolate predictions for data-sparse species, like primates, while accommodating updates based on newly available datasets.¹⁶ Such scalability is crucial for the rational translation of preclinical findings into human applications.

As depicted in Figures 2F–2H, the model accurately captured the protein profiles of fitusiran in monkeys following single-dose administration (Figure 2F) and multiple-dose regimens (Figures 2G and 2H). Similarly, the fitted curves effectively described the PK (Figure 2I) and PD protein profiles (Figures 2J–2L) of fitusiran across a wide dose range in humans. The model-estimated PK area under the curve (AUC) in the plasma and liver, along with the maximum changes in PD mRNA and protein levels across species and dose ranges, demonstrated a variation within a 2-fold error margin compared to observed data (Figure S2), indicating acceptable predictive accuracy.

In summary, we successfully developed a translational computational model for fitusiran, capable of predicting the PK and PD profiles of GalNAc-siRNA fitusiran across multiple species.

Sensitivity analysis identified key parameters related to the corresponding PK and PD profiles

To evaluate the uncertainty in how input parameters influence various PK and PD levels within the presented model—ranging from plasma and liver pharmacokinetics to target mRNA and protein reductions—a sensitivity analysis was conducted. By calculating the percentage change in the AUC, the sensitivity indices were derived, as shown in Figure 3. The results indicated that plasma PK was most sensitive to ka, PS_liver, and KP_rem, whereas liver PK was primarily influenced by KDEGD and FU. These findings align with the mechanism of GalNAc-siRNA, where key determinants of liver exposures, besides the plasma concentration-time profiles, are the intracellular liver degradation (KDEGD). Specifically, KDEGD represents the degradation rate of siRNA within hepatocytes, whereas the parameter FU regulates siRNA elimination in plasma, thereby impacting the amount of GalNAc-siRNA transported into hepatocytes via ASGPR.

Figure 3.

Parameter sensitivity analysis of the computational model

The warmer red color represents the more positive influential parameters corresponding to PK and PD, and the cooler blue color represents the more negative influence.

The PD responses for target mRNA and protein levels were influenced by several PK and PD parameters, notably γ₂, Smax, SC50, k_deg,m, and k_deg,p, which are directly related to the production and degradation of target mRNA and protein. These results are consistent with the mPBPK-PD modeling outcomes and highlight the critical parameters shaping the PK/PD curves, thereby supporting model development and enhancing mechanistic understanding.

The established model was validated by five approved GalNAc-SiRNA drugs

To confirm the applicability of the established computational model in guiding the development of GalNAc-siRNA drugs, it was externally validated using five other FDA-approved GalNAc-siRNA drugs. As illustrated in Figure 4, the model successfully captured the observed plasma and liver PK as well as mRNA and protein PD across rats, monkeys, and humans, spanning a broad range of scenarios. For instance, the model accurately simulated the time-concentration profiles of givosiran in plasma and liver PK and effectively characterized the full time course and extent of mRNA and protein reductions following single or multiple doses of givosiran in all three species (Figures 4A–4J). Furthermore, the model was validated with data from inclisiran (Figures 4K–4T), Lumasiran (Figures 4U–4W), vutrisiran (Figures 4X–4Z), nedosiran (Figures 4AA–4AB), and plozasiran (Figures S3A–S3F). The estimated parameters for GalNAc-siRNAs in humans are listed in Table S1. Notably, the estimated f_esc for nedosiran and plozasiran are 0.258 and 0.39, respectively, which are greater than the other five GalNAc-siRNAs, suggesting potential increased endosomal escape and subsequent RISC loading efficiency.

Figure 4.

Observed and fitted PK and PD profiles for givosiran, inclisiran, lumasiran, vutrisiran, and nedosiran in rats, monkeys, and humans, respectively

The model-based characterization is detailed as follows: givosiran (A) plasma concentration after single doses in rats. (B) Liver concentration after single doses in rats. (C) mRNA (%) after single doses in rats. (D) Plasma concentration after single doses in monkeys. (E) Liver concentration after single doses in monkeys at 1, 5, 10 mg/kg SC and 10 mg/kg IV. (F) Plasma concentration after single doses in humans. (G) mRNA (%) after single doses in humans. (H) mRNA (%) after multiple doses in humans QW × 2 doses. (I) Protein (%) after single doses in humans. (J) Protein (%) after multiple doses in humans QW × 2 doses. Inclisiran (K) plasma concentration after single doses in monkeys. (L) mRNA (%) after single doses in monkeys. (M) mRNA (%) after multiple doses in monkeys quarterly × 4 doses. (N) Protein (%) after multiple doses in monkeys. (O) Plasma concentration after single doses in humans.

(P) Plasma concentration after single doses in Chinese. (Q) mRNA (%) after single doses in humans. (R) mRNA (%) after multiple doses in humans quarterly × 2 doses. (S) Protein (%) after single doses in humans. (T) Protein (%) after multiple doses in humans quarterly × 2 doses. Lumasiran (U) plasma concentration after single doses in humans. (V) Protein (%) after multiple doses in humans at 1 and 3 mg/kg QM × 3 doses and 1 mg/kg quarterly × 2 doses. (W) Protein (%) after multiple doses in humans at 1 and 3 mg/kg QM × 3 doses and 1 mg/kg quarterly × 1 doses. Vutrisiran (X) plasma concentration after single doses in humans. (Y) Protein (%) after single doses in humans. (Z) Protein (%) after single doses in Japanese. Nedosiran (AA) plasma concentration after single doses in humans. (AB) Protein (%) after single doses in humans. Solid lines represent model fitted results, and dots indicate observed data points. Different colors represent different dosing regimens.

In brief, the validated model demonstrates robust predictive capability for describing the PK and PD profiles of diverse GalNAc-siRNA drugs across species. The simulated results were generally consistent with observed data, reinforcing the model’s potential to support the development of GalNAc-siRNA therapeutics.

Model simulations revealed that the prolonged effects of GalNAc-siRNA drugs can be attributed to their sustained presence in the liver

To quantitatively explore the interrelationships between GalNAc-siRNA dose, plasma exposure, liver exposure, and PD profiles of mRNA and protein, Monte Carlo simulations were conducted using the qualified mPBPK-PD model. These simulations predicted the plasma and liver PK as well as the PD effects of inclisiran, incorporating human-specific model parameters. As illustrated in Figure S3, following a 300 mg SC administration of inclisiran once every 3 months for two doses, plasma concentrations initially rose sharply but declined rapidly, which is consistent with GalNAc-siRNA’s short half-life in plasma (Figure S4A). In contrast, inclisiran accumulated in the liver, where it persisted for several months (Figure S4B), aligning with its durable ability to knockdown the target mRNA (Figure S4C) and protein (Figure S4D). This sustained liver exposure and associated PD effects mechanistically explain and support the long dosing interval of inclisiran (once every 3 months).

The web-based user-friendly Shiny app facilitates the development of GalNAc-SiRNA drugs

To enhance accessibility and streamline the development of GalNAc-siRNA drugs, we created a user-friendly web-based Shiny application built on the mPBPK-PD model. This intuitive platform features a clean interface: input parameters are displayed on the left, while simulation results—including plasma and liver PK, as well as mRNA and protein PD profiles—are shown on the right. Users can select variables, such as species, dosing schedule, and dose amount to conduct simulations. The application enables researchers to input specific model parameters and to predict PK/PD profiles across species, significantly simplifying the process of designing next-generation GalNAc-siRNA therapeutics. Accessible via the link: https://galnacsirnadevelop.shinyapps.io/shiny/, the tool provides valuable insights for drug development and precision dosing strategies.

The validated platform promotes the development of SAL0132

To accelerate the development of SAL0132, a novel GalNAc-siRNA targeting hepatic AGT mRNA to lower blood pressure, the validated platform was applied. As shown in Figure 5, SAL0132 rapidly distributed to plasma, reaching peak concentrations 30 min after dosing in rats. Drug levels declined swiftly, falling below the limit of quantitation 10 h after SC administration (Figure 5A) and 2 h after IV administration (Figure 5B). The platform effectively characterized the plasma concentration-time profiles of SAL0132 across the dose range examined (3 mg/kg to 12 mg/kg) for both IV and SC routes. The PK parameters are detailed in Table S2. The platform-predicted plasma AUC values across the dose range exhibited variations within a 2-fold error margin compared to observed data (Figure 5C), demonstrating acceptable predictive accuracy.

Figure 5.

Observed and fitted PK profiles for SAL0132 in rats

Solid lines represent model fitted results, and dots indicate observed data points. Different colors represent different dosing regimens. Model-based characterization of (A) plasma concentration after single SC doses in rats at 3, 6, and 12 mg/kg. (B) Plasma concentration after single IV dose in rats at 6 mg/kg. (C) The correlation between observed and predicted AUC of SAL0132 in plasma in rats. The solid line indicates unity, while dashed lines on either side of the unity line represent 2-fold deviations. Different color and shape denote different dosing regimens.

The model was then scaled from rats to monkeys and calibrated using monkey data. As shown in Figure 6, SAL0132 plasma levels decreased rapidly after IV administration, with the concentration were below the limit of quantitation after 6 h (Figure 6A). Following SC administration, SAL0132 rapidly distributed to plasma with maximum concentrations occurring at 1–4 h after dosing, while liver levels increased rapidly, persisted over time, and decreased slowly across dose groups (Figure 6B). Drug concentrations remained measurable in the liver on day 91, reflecting the short plasma half-life and prolonged liver retention typical of this compound class. Both plasma and liver PK profiles were generally dose proportional. The model accurately captured the steep initial drop in plasma siRNA concentrations after IV and SC dosing, as well as the extended retention and slow decline in liver concentrations after SC dosing. Furthermore, dose-dependent inhibition of serum Agt protein levels was observed following a single SC administration of SAL0132 (Figure 6C). Maximum Agt protein suppression occurred 42 days post-dose, with reductions of approximately 89%, 96%, and 96% at doses of 3, 6, and 12 mg/kg, respectively. Agt suppression was potent and durable, with a single 3 mg/kg dose maintaining ∼20% serum Agt protein levels for 42 days, followed by gradual recovery after 63 days. The platform accurately characterized the full time course and extent of Agt protein level reductions, with calibrated PD parameters listed in Table S2. Predicted plasma (Figure 6D) and liver (Figure 6E) AUC values, along with the maximum changes in PD protein levels (Figure 6F), demonstrated variations within a 2-fold error margin compared to observed data, indicating satisfactory predictive accuracy.

Figure 6.

Observed and fitted PK and PD profiles for SAL0132 in monkeys

Solid lines represent model fitted results, and dots indicate observed data points. Different colors represent different dosing regimens. Model-based characterization of (A) plasma concentration after single SC or IV doses at 3, 6, and 12 mg/kg. (B) Liver concentration after single SC dose at 3, 6, and 12 mg/kg. (C) Protein (%) after single SC doses in monkeys at 3, 6, and 12 mg/kg. (D and E) The correlation between observed and predicted AUC of SAL0132 in plasma (D) and in the liver (E). (F) The correlation between observed and predicted protein (%). The solid line indicates unity, while dashed lines on either side of the unity line represent 2-fold deviations. Different color and shape denote different dosing regimens.

Finally, the PK and PD parameters were scaled from monkeys to humans to support SAL0132’s clinical development. As depicted in Figure 7, the platform successfully characterized observed plasma PK profiles of SAL0132 across a wide dose range (Dose1–Dose1×12) following single SC administration in humans (Figure 7A), with predicted AUC values falling within a 2-fold error margin compared to observed data (Figure 7B). Tuned human PK parameters and scaled PD parameters were used to simulate PK profiles in human livers and PD profiles of mRNA and protein, which are not yet available. The platform predicted SAL0132 liver PK profiles in humans align with the prolonged liver retention typical of this compound class (Figure 7C). A single SC administration resulted in potent, dose-dependent, and durable suppression of AGT mRNA (Figure 7D) and protein (Figure 7E) levels, with reductions persisting for over six months at doses of Dose1×3, Dose1×6, and Dose1×12. The predicted overall AGT reductions were consistent with the reported trends for AGT-targeted RNAi therapeutics.¹⁷ These results will be validated and calibrated through ongoing clinical studies, guiding and accelerating SAL0132’s future clinical development.

Figure 7.

Observed, fitted, and simulated PK and PD profiles for SAL0132 in humans

(A) Plasma concentration after single SC doses at Dose1, Dose1×3, Dose1×6, and Dose1×12. Symbols expressed as mean ± SD represent observed data. (B) The correlation between observed and predicted AUC of SAL0132 in plasma. (C–E) Model predicted SAL0132 liver PK profiles (C); model predicted PD profiles for target mRNA (D); model predicted PD profiles for target protein (E); after Dose1, Dose1×3, Dose1×6, and Dose1×12 single doses in humans.

Discussion

GalNAc-siRNA therapeutics have redefined the treatment landscape for genetic and metabolic disorders by leveraging the RNAi mechanism to selectively silence the expression of disease-causing genes.¹⁸ This modality offers unparalleled specificity and precision, targeting intracellular pathways and “undruggable” targets that are beyond the reach of traditional small-molecule drugs or biologics. The conjugation of siRNA with GalNAc facilitates highly efficient delivery to hepatocytes via ASGPRs, ensuring targeted drug action while minimizing off-target effects.² The prolonged PD effects, allowing dosing intervals of 3–6 months, translate into improved patient compliance and reduced healthcare burdens. The clinical success of seven FDA-approved GalNAc-siRNA drugs (e.g., inclisiran’s >50% sustained LDL-C reduction with biannual dosing) underscores their transformative potential. Additionally, these therapeutics exhibit remarkably consistent PK/PD profiles across different molecules within the class, attributed to their shared delivery mechanism and RNAi-based action, providing an exceptional opportunity to streamline drug development by using the MIDD strategy.^4,19

MIDD strategies have emerged as an invaluable tool that leverages quantitative modeling to de-risk and accelerate drug development. For GalNAc-siRNAs, the predictability of their ASGPR-mediated uptake, hepatic disposition, and delayed RNAi effects (e.g., RISC complex stability) provides a unique opportunity to generalize modeling approaches to predict drug behavior across species and optimize dosing regimen, which can reduce reliance on extensive empirical testing, saving both time and resources while maintaining scientific rigor. Several mechanistic models have already been developed to characterize the PK/PD behavior of GalNAc-siRNA drugs, including fitusiran, givosiran, and others.^9,10,11,12 These models incorporate detailed physiological and molecular mechanisms, providing a robust framework for translational drug development. However, these models are limited to single molecules, and the complex mathematical equations may not be easy for application.

Our work advances the field by integrating MIDD principles into a unified platform, capturing shared mechanisms while accommodating molecule-specific parameters (e.g., siRNA stability, target mRNA turnover). We developed a refined mechanistic computational model-powered platform to inform the development of GalNAc-siRNA therapeutics. This platform integrates preclinical and clinical data from all seven FDA-approved GalNAc-siRNA drugs (Figure 1A), spanning multiple species (mouse, rat, monkey, and human). The platform was rigorously validated using these drugs. The model incorporates species-specific ASGPR expression and liver physiology, enabling accurate human dose prediction (validation error <2-fold), demonstrating its ability to accurately predict PK/PD profiles across species (Figure S2). This consistency highlights its utility in guiding first-in-human dosing and trial design for novel GalNAc-siRNAs. Key findings from the sensitivity analysis revealed that the parameter KDEGD (representing free cytosolic siRNA degradation) significantly influences hepatic drug exposure and PD effects, providing critical insights for drug design (Figure 3). Model simulations further highlighted the relationship between hepatic drug exposure and the sustained therapeutic effects of GalNAc-siRNA drugs, emphasizing the importance of liver-targeted delivery in achieving prolonged gene silencing (Figure S4). A time-delayed RNAi module quantifies gene silencing duration, critical for infrequent dosing.

To enhance accessibility, we implemented the platform as a user-friendly web-based Shiny application (https://galnacsirnadevelop.shinyapps.io/shiny/). This tool allows researchers to input specific model parameters and to predict PK/PD profiles across species, streamlining the development of next-generation GalNAc-siRNA therapeutics. By making advanced modeling capabilities accessible to a wider audience, this platform has the potential to accelerate drug development while maintaining high precision in predicting clinical outcomes (Figure S5).

We then applied this platform to facilitate the development of SAL0132, a novel GalNAc-siRNA targeting AGT mRNA to lower blood pressure. Despite significant progress in antihypertensive therapies, patient adherence remains a major challenge.²⁰ Innovative therapeutic strategies that offer convenient, safe, and long-lasting blood pressure-lowering solutions are urgently needed, holding significant clinical implications for improving hypertension management. AGT serves as the precursor of angiotensin II, a potent vasoconstrictive octapeptide in the renin-angiotensin-aldosterone system. Inhibiting AGT-converting enzymes has proven effective in lowering blood pressure.²¹ Recently, zilebesiran, an investigational GalNAc-siRNA targeting AGT, has progressed to phase 3 clinical trials and demonstrated promising potential for hypertension management.²² Given the prolonged PD effects of GalNAc-siRNAs, SAL0132 is anticipated to deliver sustained and stable antihypertensive efficacy through intermittent SC administration, similar to zilebesiran. This approach addresses poor medication adherence caused by the short half-life and frequent dosing of oral antihypertensive agents while mitigating blood pressure fluctuations, which are independently associated with renal impairment and cardiovascular risks. The platform successfully captured the PK profiles of SAL0132 in rat plasma (Figure 5) and PK/PD profiles in monkey plasma and liver (Figure 6) across diverse dose levels. After scaling to humans, the calibrated platform accurately characterized the plasma PK profiles of SAL0132 and simulated the liver PK profiles and PD profiles of mRNA and protein (Figure 7). These simulations, which can be validated through ongoing clinical studies, provide confidence in employing this platform to support and accelerate SAL0132’s clinical development through platform-based simulations.

An important consideration for the clinical application of our platform is the impact of organ dysfunction, particularly hepatic and renal impairment. Liver diseases can be associated with changes in functional hepatocyte mass, liver blood flow, and ASGPR expression/activity, which could alter the efficiency of hepatic uptake and thereby influence downstream intracellular steps (endosomal trafficking/escape, RISC loading, and effective mRNA knockdown).²³ In the context of our platform, these effects would be expected to manifest primarily through parameters governing hepatic uptake capacity (e.g., receptor abundance/availability) and liver distribution/processing terms, and could lead to reduced liver exposure and attenuated or delayed PD in more advanced hepatic impairment. Notably, available clinical data for approved GalNAc-siRNAs generally suggest no clinically meaningful PK/PD changes in mild to moderate hepatic impairment for certain compounds.²⁴ This is consistent with model sensitivity analysis, indicating that PK and PD were not sensitive to kint (Figure 3). However, data remain limited and are not available for all drugs and all severities, particularly severe hepatic impairment. Therefore, predictions for patients with advanced liver disease should be interpreted cautiously unless impairment-specific parameters are incorporated and supported by dedicated clinical or translational datasets. Renal impairment presents a different set of considerations. Given that GalNAc-siRNA therapeutics are primarily cleared through hepatic uptake and intracellular degradation, renal impairment has minimal impact on their PD.^4,25,26 In our platform, renal impairment would primarily be reflected in systemic clearance and distribution terms rather than in the core liver RNAi module. Clinical data for approved GalNAc-siRNAs often indicate minimal PD impact and no required dose adjustment across mild to severe renal impairment for some compounds, though the magnitude and clinical relevance of PK changes may be compound- and population dependent. Importantly, the framework established in our platform provides a foundation for addressing these clinical complexities. By modifying key system-specific parameters (e.g., ASGPR expression levels, endosomal degradation rates, clearance pathways), the model could theoretically be adapted to simulate GalNAc-siRNA behavior in special populations. However, empirical clinical data from patients with varying degrees of organ impairment will be essential to validate and refine these adaptations. As such data become available through ongoing clinical studies, our platform can evolve to incorporate population-specific parameters, ultimately supporting dose optimization strategies for patients with comorbidities, a critical step toward personalized dosing of GalNAc-siRNA therapeutics.

Another important consideration for the translational utility of our platform is its generalizability to GalNAc-siRNA designs that diverge from the current generation of approved drugs. A key demonstration of this generalizability is the platform’s capacity to capture the behavior of both nedosiran and plozasiran, two molecules developed using distinct proprietary platforms. nedosiran employs Dicerna’s proprietary GalXC platform, which features a unique tetraloop motif and tetra-antennary GalNAc presentation, while plozasiran utilizes Arrowhead’s TRiM platform, designed to enhance potency.^27,28,29 Importantly, our model not only captures the shared mechanistic features common to all GalNAc-siRNAs but also distinguishes platform-specific design elements that influence PK/PD behavior. For instance, structural features such as the tetraloop motif in the GalXC platform and the optimized configuration of the TRiM platform may impact intracellular trafficking, including endosomal escape efficiency and RISC loading. This is reflected in the estimated endosomal fesc for nedosiran and plozasiran, which are notably higher than those of the five Alnylam-developed GalNAc-siRNA drugs (Table S1), indicating potential advantages in endosomal escape and RISC loading compared to other GalNAc-siRNA drugs. For molecules with enhanced chemical modifications, the platform can accommodate altered stability through parameters governing endosomal and cytoplasmic degradation (KDEGE and KDEGD), provided that the fundamental mechanisms of ASGPR-mediated uptake and RISC loading remain intact. Similarly, modifications to GalNAc valency would primarily affect the receptor-binding submodule, which could be adapted by incorporating valency-dependent affinity constants derived from surface plasmon resonance or cellular uptake studies. The platform’s validation across approved drugs with diverse dosing intervals suggests robustness to regimen variations, though extreme regimens would benefit from prospective validation. For novel candidates, the platform enables virtual screening, simulating PK/PD profiles based on projected parameter changes, to prioritize molecules with the greatest therapeutic potential before committing to synthesis and extensive in vivo testing. Furthermore, by identifying which physiological parameters most critically influence translational scaling, the platform enables targeted experimental confirmation rather than exhaustive dose-ranging studies across species. For emerging applications targeting extra-hepatic tissues, the current liver-specific platform would require substantial restructuring. Nevertheless, the modular architecture we have established (disposition, intracellular trafficking, and RNAi) provides a conceptual foundation for developing analogous platforms for other delivery systems. The RNAi module, in particular, is tissue-agnostic and could be directly repurposed, whereas the disposition and uptake modules would need replacement with tissue-appropriate frameworks. As the field progresses toward extra-hepatic RNAi therapeutics, our platform may serve as a template for building specialized models that balance mechanistic fidelity with practical utility for drug development.

Future work will expand the platform’s scope to include rare disease populations and combination therapies, further establishing GalNAc-siRNAs as a cornerstone of precision medicine. Beyond GalNAc-conjugated siRNAs, the mechanistic framework underlying our platform holds significant potential for expansion to other classes of siRNA conjugates, particularly antibody-siRNA conjugates (Abs-siRNAs) and lipid nanoparticle (LNP)-formulated siRNAs. Such extensions would broaden the platform’s utility across the rapidly evolving landscape of RNA therapeutics while leveraging the shared principles of receptor-mediated delivery and intracellular processing. For example, Abs-siRNAs often rely on receptor-mediated endocytosis via target-specific surface antigens, which may differ significantly from ASGPR-mediated uptake in terms of receptor expression, recycling rates, and tissue specificity. Incorporating these parameters into the platform would require defining new receptor-ligand binding kinetics, endosomal trafficking, and intracellular degradation pathways. Similarly, lipid-based siRNA formulations, such as LNPs, exhibit unique biodistribution and clearance profiles due to their reliance on both passive and active targeting mechanisms. patisiran, the first FDA-approved LNP-delivered siRNA therapeutic, uses particle-based biodistribution to enable delivery to tissues beyond the liver, including the central nervous system.^2,5 Administered intravenously, patisiran demonstrates a plasma half-life of approximately 3.2 h, allowing for effective systemic distribution and therapeutic action.⁵ Therefore, for LNPs, the model could be expanded to include additional compartments representing key clearance organs (e.g., spleen, lungs) and processes such as opsonization and phagocytosis. Overall, the modular nature of our platform allows for such expansions, enabling the incorporation of new biological pathways and tissue-specific parameters as needed.

In conclusion, this study presents a mechanistic computational model-powered platform that serves as a comprehensive framework for the development of GalNAc-siRNA therapeutics. By integrating data from seven FDA-approved drugs and implementing sophisticated modeling approaches, the platform provides a reliable and versatile tool for predicting PK/PD profiles across species. The successful validation of the model with multiple drugs and species underscores its robustness and broad applicability. The implementation of the platform as an easy-to-use web-based Shiny application ensures accessibility to researchers without extensive modeling expertise, facilitating the efficient and precise development of next-generation GalNAc-siRNA therapeutics. Furthermore, the validated platform was applied to facilitate the development of a novel GalNAc-siRNA, SAL0132, demonstrating the reliability and advantages of this platform.

This work highlights the transformative potential of MIDD strategies in advancing RNAi-based medicine, offering a pathway to reduce reliance on empirical dose-finding trials and optimize dosing regimens (e.g., monthly vs. quarterly dosing), thereby accelerating the clinical translation of these groundbreaking therapies. As GalNAc-siRNA technology continues to evolve, this platform represents a significant step forward in enabling the rational design and optimization of GalNAc-siRNA drugs, ultimately improving patient outcomes and addressing unmet medical needs.

Materials and methods

Data collection and software

The datasets used for model development and qualification in this study were compiled through digitization (WebPlotDigitizer, v.4.6) of data from the literature and FDA review reports. These datasets encompass seven FDA-approved GalNAc-SiRAN drugs—givosiran,^10,30,31 lumasiran,^32,33 inclisiran,^{34,35,36,37,38} vutrisiran,³⁹ nedosiran,^40,41 fitusiran,^42,43,44 and plozasiran,^45,46,47,48 across four species—mice, rats, monkeys, and humans. Relevant experimental details, including dosage and species information, are summarized in Table S3.

The model was developed using NONMEM (v.7.6, Icon Development Solutions, USA), with the software supported by Perl-speaks-NONMEM (v.4.9.6, http://psn.sourceforge.net/docs.php). Ordinary differential equations were solved using the ADVAN14 subroutine. For parameter estimation, the expectation maximization (EM) algorithm (stochastic approximation EM followed by importance sampling) was applied to PK parameters, while the first-order conditional estimation method with interaction (FOCEI) algorithm was employed for PD parameters.⁴⁹ Model diagnostics and graphical visualizations were performed using the R programming language (v.4.3.3, www.r-project.org).⁵⁰

Model development

A multiscale PK-PD model was developed step by step based on the physiological and pharmacological mechanisms underlying GalNAc-siRNA disposition and action (Figure 1B). Fitusiran, which provided the most comprehensive data on the PK and PD profiles of GalNAc-siRNA across multiple species, was initially used to construct the mechanistic computational model. The model’s predictive performance for PK and PD was then validated using other GalNAc-siRNAs across different species through modeling and simulation techniques. First, the computational model for mice was refined from an existing framework and incorporated mechanisms specific to GalNAc-siRNA⁹ (Figure 1B). Briefly, three primary compartments, including the plasma, liver, and the remainder that lumped the other organs, were contained in the PK model. The kidney was lumped into the remainder compartment due to the predominant slow metabolism of GalNAc-siRNAs in the liver and the limited availability of kidney-specific data.⁴

Model parameters were categorized into two main types: physiological parameters (Table S4) and drug-specific parameters (Table 1). Physiological parameters, including organ volume (V) and plasma flow to organs (Q), were adjusted based on species and body weight.^51,52,53,54 The kinetics of GalNAc-siRNA in various compartments were described using the following ordinary differential equations.

• (1)Systemic PK:

When administered subcutaneously (SC) (A₁), GalNAc-siRNA enters the plasma compartment (A₂) following first-order absorption kinetics from the SC site. It is assumed to undergo clearance (FU) and distribution into the liver or other tissues.

$$\frac{d{A}_1}{dt}=-k_a·A_1$$

$$\frac{d{A}_2}{dt}=k_a·A_1-\left(Q_{liv}+Q_{rem}\right)·C_2-FU·C_2+Q_{liv}·C_3+\frac{Q_{rem}}{{KP}_{rem}}·C_{11}$$

where A₁ and A₂ are the amounts of GalNAc-SiRNA in the deposition and plasma compartments, respectively. FU and k_a represent the apparent clearance in plasma and the first-order absorption rate of GalNAc-SiRNA, respectively. Q_liv and Q_rem denote the blood flow rates to the liver and the remainder organs, respectively, while KP_rem is the tissue-to-plasma partition coefficient of GalNAc-SiRNA in the remainder compartment. The plasma concentration of GalNAc-SiRNA (C₂) is given by $C_2=\frac{A_2}{V_{pla}}$, where V_pla is the plasma volume.

The liver compartment is further divided into vascular (A3), extracellular (A4), and cellular (A10) spaces to account for ASGPR-mediated intracellular uptake of GalNAc-SiRNA.^9,10 GalNAc-SiRNA enters the liver vascular compartment from plasma, regulated by the liver blood flow (Q_liv). It then moves sequentially into the extracellular space and cellular compartment, controlled by liver permeability and ASGPR-mediated intracellular uptake.

$$\frac{d{A}_3}{dt}=Q_{liv}·C_2-Q_{liv}·C_3-{PS}_{liv}·C_3+\frac{{PS}_{liv}}{{KP}_{liv}}·C_4$$

$$\frac{d{A}_4}{dt}={PS}_{liv}·C_3-\frac{{PS}_{liv}}{{KP}_{liv}}·C_4-\frac{k_{on}·C_4·A_5}{MW·1000000}+k_{off}·A_6$$

$$\frac{d{A}_5}{dt}=k_{syn}-k_{\mathit{deg},ASG}·A_5-\frac{k_{on}·C_4·A_5}{MW·1000000}+k_{off}·A_6$$

$$\frac{d{A}_6}{dt}=\frac{k_{on}·C_4·A_5}{MW·1000000}-k_{off}·A_6-k_{\mathit{int}}·A_6$$

$$\frac{d{A}_7}{dt}=k_{\mathit{int}}·A_6-k_{cle}·A_7$$

where A₅, A_6, and A₇ are the free ASGPR, bound ASGPR, and the internalized siRNA-ASGPR complex, respectively. PS_liv represents the membrane permeability of GalNAc-SiRNA in the liver, while KP_liv is the liver-to-plasma partition coefficient. The concentrations in the liver vascular (C₃) and extracellular (C₄) spaces are defined as $C_3=\frac{A_3}{V_{liv,vas}}$ and $C_4=\frac{A_3}{V_{liv,ext}}$ , where V_liv,vas and V_liv,ext are the respective volumes. Molecular weight (MW) of GalNAc-SiRNA is used to convert its concentration to nM units. k_on and k_off are the association and dissociation rate constants for GalNAc binding to ASGPR, while k_int and k_cle represent the cellular internalization rate of bound ASGPR and the cleavage rate of GalNAc from siRNA.

Upon internalization into endosomes, GalNAc-SiRNA undergoes cleavage, releasing siRNA that can bind to a cytoplasmic pool, degrade, or escape into the cytoplasm. A target-mediated drug disposition model⁵⁵ was combined to characterize the binding of free siRNA to cytoplasmic RISC.

$$\frac{d{A}_8}{dt}=k_{cle}·A_7-KDEGE·A_8-f_{esc}·A_8-k_{ass}·A_8+k_{dis}·A_9$$

$$\frac{d{A}_9}{dt}=k_{ass}·A_8-k_{dis}·A_9$$

$$\frac{d{A}_{10}}{dt}=f_{esc}·A_8-KDEGD·C_{10}-\frac{V_M·C_{10}}{K_M+C_{10}}$$

where A₅ and A₆ are the free and bound ASGPR compartments, respectively. A7, A8, A9, and A10 represent the internalized siRNA-ASGPR complex, free endosomal siRNA, endosomal bound pool, and free cytoplasmic siRNA, respectively.

The free siRNA (C₁₀), total siRNA (C_TOT), and drug-target complex (RC) concentrations can be derived as

$$RC=R_{TOT0}·\frac{A_{10}}{V_{liv,cell}}/\left(KM+\frac{A_{10}}{V_{liv,cell}}\right)$$

$$C_{TOT}=\frac{A_{10}}{V_{liv,cell}}$$

$$C_{10}=\frac{1}{2}·\left[\left(C_{TOT}-R_{TOT0}-KM\right)+\sqrt{{\left(C_{TOT}-R_{TOT0}-KM\right)}^2+4\times KM\times C_{TOT}}\right]$$

For the remainder compartment A11,

$$\frac{d{A}_{11}}{dt}=Q_{rem}·C_2-\frac{Q_{rem}}{{KP}_{rem}}·C_{11}$$

• (2)PD:

For PD, transit compartments have been widely used to model delays in biological processes.^56,57 In this study, we employed two series of transit compartments to capture delays associated with the mRNA and protein production. The mRNA transduction process is represented by transit compartments (mRNAn, n = 10) with first-order transition rates k_trm. The model equations are as follows:

$$\frac{dm{RNA}_1}{dt}=k_{syn,m}-k_{trm}·{mRNA}_1$$

$$\frac{dm{RNA}_i}{dt}=k_{trm}·{mRNA}_{i-1}-k_{trm}·{mRNA}_i\mathrm{i}=2,\dots ,\mathrm{n}-1$$

$$\frac{dm{RNA}_n}{dt}=k_{trm}·{mRNA}_i-k_{\mathit{deg},m}·{mRNA}_n·\left(1+\frac{S_{\mathit{max}}·RC}{{SC}_{50}+RC}\right)$$

Similarly, Protn (n = 10) represents the target protein with the transition rates k_trp.

$$\frac{d{Prot}_1}{dt}=k_{syn,p}·{\left(\frac{{mRNA}_n}{{mRNA}_0}\right)}^{\gamma 1}-k_{trp}·{Prot}_1·{\left(\frac{{mRNA}_n}{{mRNA}_0}\right)}^{\gamma 2}$$

$$\frac{d{Prot}_i}{dt}=k_{trp}·{Prot}_{i-1}·{\left(\frac{{mRNA}_n}{{mRNA}_0}\right)}^{\gamma 2}-k_{trp}·{Prot}_i·{\left(\frac{{mRNA}_n}{{mRNA}_0}\right)}^{\gamma 2}\mathrm{i}=2,\dots ,\mathrm{n}-1$$

$$\frac{d{Prot}_n}{dt}=k_{trp}·{Prot}_i·{\left(\frac{{mRNA}_n}{{mRNA}_0}\right)}^{\gamma 2}-k_{\mathit{deg},p}·{Prot}_n$$

where k_deg,m and k_deg,p denote degradation rate constants for target mRNA and protein, while k_syn,m and k_syn,p are zero-order rate constants for their synthesis. k_deg,m and k_deg,p were assumed to be equal to k_trm and k_trp, respectively, to reduce the number of model parameters.

The secondary parameters and baseline equations defined by the steady-state value were used to reduce the number of model parameters as follows:

$$k_{syn,m}=k_{\mathit{deg},m}·{mRNA}_0$$

$$k_{syn,p}=k_{\mathit{deg},p}·{Prot}_0$$

Physiological parameters such as blood flow rates and tissue volumes were obtained from the literature (Table S4), while tissue-specific parameters (such as PSliv, KPliv, and KPrem) were estimated. Plasma and liver data were simultaneously fitted in NONMEM v.7.6, with residual errors modeled using either additive (Y_ij = Ŷ_ij+ε₁) or proportional (Y_ij = Ŷ_ij·(1+ε₂)) error models. Here, Y_ij represents the observed value for individual i at time t_j, Ŷ_ij is the model-predicted value, and ε₁ and ε₂ are independent, normally distributed random variables with zero mean and variances σ₁² and σ₂², respectively. Given that only mean data are available, a naive pooled data approach was used, treating all data as originating from a single individual. To enhance the efficiency of the EM algorithm in NONMEM, the variance of random effect parameters was constrained to a small value (0.0225, equivalent to a 15% coefficient of variation).

Model selection and evaluation were based on objective function values, parameter precision, and diagnostic plots (observed vs. predicted values). The final model was validated by simulating PK profiles and overlaying them with observed data. Considerations of drug mechanism, model complexity, and parameter precision ensured a balance between model fit and interpretability.

Model validation in different GalNAc-siRNAs and multiple species

To validate the capability of the established model in predicting the PK and PD profiles of GalNAc-siRNA in different species, as well as to reproduce in different GalNAc-siRNAs, allometric scaling, model-based characterization, and simulations were conducted.⁵⁸

Initially, to translate the mPBPK-PD model from mice to rats, monkeys, and humans, different allometric scaling exponents were employed.⁹ Specifically, physiological parameters like volume of organ (V) and plasma flow rate (Q) were adjusted based on species and body weight.⁵¹ Parameters such as ka, k_int, and KDEGD were scaled by using the allometric scaling exponent −0.25,^10,59 whereas FU, Psliver, and V_M employed an exponent of 0.75.^60,61,62 Other model parameters, including KP, k_ass, k_dis, KDEGE, KM, k_off, f_esc, and RTOT₀, remained consistent across species.⁶³ The scaled model was externally calibrated and validated using the species-specific data from the literature.

Sensitivity analysis

A sensitivity analysis was conducted to identify key input parameters influencing the simulated AUC for plasma and liver concentrations at 24 and 720 h, as well as mRNA and protein changes at 720 h, in mice and other species.

Each parameter was increased by 20%, and the corresponding relative change in AUC (ΔAUC), which served as a sensitivity index, was calculated using the formula ΔAUC = (AUC_change−AUC₀)/AUC₀ × 100%. The perturbed parameters that caused relatively higher changes in PK/PD exposure were deemed more sensitive.

Model-based simulations

To explore the relationship between PK exposure in plasma and liver and the PD response of target mRNA and protein, simulations were conducted using the mPBPK-PD model developed in this study. Inclisiran was selected as the simulation drug, and its dosing regimen followed the approved drug label: 300 mg administered once every 3 months for two doses. The simulation spanned 6 months, providing sufficient time to observe the PK and PD profiles and analyze their correlation.

Development of a user-friendly Shiny app

To make the model accessible to researchers without extensive modeling expertise and to facilitate the efficient and accurate development of next-generation GalNAc-siRNA therapeutics, we developed an online platform using the shiny package in R. This platform is hosted on Shinyapps.io (www.shinyapps.io), a cloud-based service that supports R-based web applications. This setup ensures broad accessibility and ease of use across various devices without requiring local installations. The application seamlessly integrates several R packages, including Shiny (v.1.10.0), ggplot (v.3.5.1), bslib (v.0.9.0), rsconnect (v.0.4.1), and deSolve (v.1.40).

By leveraging Shiny, we combined the computational power of R with modern web technologies, allowing us to create an interactive user interface with innovative features. To use the application, select the species and enter the necessary parameters listed in Table 1, along with the dosing interval and amount (separated by commas). Once the simulation is conducted, the PK profiles of GalNAc-siRNA in plasma and liver, as well as the PD profiles of mRNA and protein, will be displayed.

PK and PD studies of SAL0132 in rats, monkeys, and humans

SAL0132 (also known as GW906), initially developed by Gowell Pharmaceutical and subsequently transferred to Salubris Pharmaceuticals, is a first-in-class GalNAc-siRNA therapeutic targeting hepatic AGT mRNA to lower blood pressure.²¹ It is currently undergoing clinical development for the treatment of hypertension. To evaluate its PK and PD characteristics, a single-dose PK study in Sprague-Dawley (SD) rats, a single-dose PK/PD study in cynomolgus monkeys (Macaca fascicularis), and an ongoing clinical trial have been conducted. All animal experiments were approved by the local scientific ethics committee. The clinical trial complies with legal and regulatory requirements, adheres to the principles of the Declaration of Helsinki, and is registered with chinadrugtrials.org.cn (CTR20251035).

For rats PK studies, 24 male SD rats were divided into 4 groups of 6 rats each. Three groups were administered a single SC dose of SAL0132 at 3, 6, and 12 mg/kg, respectively, while the fourth group received a single intravenous (IV) dose of SAL0132 at 6 mg/kg. Blood samples were collected at various intervals, including pre-dose and at 0.25, 0.5, 1, 1.5, 2, 3, 4, 5, 6, 8, 10, and 24 h post-dose for the SC groups, with an additional time point at 0.033 h included for the IV group.

In the monkey PK and PD studies, 24 healthy cynomolgus monkeys were divided into 4 groups, each receiving either 3, 6, or 12 mg/kg SAL0132 via SC injection or 6 mg/kg SAL0132 via IV injection. Each dose group consisted of 6 monkeys. For the PK studies, blood samples were collected at 0.25, 0.5, 1, 2, 3, 4, 6, 8, 10, 14, 24, 32, and 48 h post-dose for each SC group, with an additional time point at 0.083 h for the IV group. Liver samples were collected at 48, 504, 1,008, 1,512, and 2,184 h post-dose for each SC group. For the PD studies, blood samples were collected at 504, 1,008, 1,512, and 2,184 h post-dose for each SC group. The PD marker, serum AGT protein, was measured using a human total angiotensinogen assay kit (IBL, Japan).

In an ongoing phase 1 clinical trial, the safety, tolerability, PK, and PD profiles of SAL0132 were assessed in healthy and mildly hypertensive subjects. Participants were randomized to receive single SC doses of SAL0132 at Dose1, Dose1×3, Dose1×6, or Dose1×12 (n = 6 per group, The specific dosing details are withheld due to ongoing intellectual property considerations. This information is the subject of a pending patent application and will be made available upon its publication). Blood samples were collected at multiple intervals post-dose, including 0.5, 1, 2, 3, 4, 6, 8, 12, 16, 24, and 48 h, to evaluate PK profiles.

Model-informed development of SAL0132

To facilitate the development of SAL0132, the validated platform was employed. First, PK data from rats were fitted into the platform, and the calibrated PK parameters were scaled to simulate the PK profiles of SAL0132 in monkeys. Next, monkey PK data were incorporated to refine the model for monkeys, and the calibrated PK model was used to drive the PD, which was further optimized using monkey PD data. Finally, the PK parameters from monkeys were scaled to humans and adjusted based on human PK data. The calibrated human PK parameters, along with the scaled PD parameters, were employed to simulate liver PK profiles and PD outcomes that are not yet available due to the ongoing clinical trials.

Data and code availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Acknowledgments

We would like to thank Sichuan Gowell Pharmaceutical Co., Ltd. for their significant contributions and pivotal role in the early development of the GalNAc-siRNA SAL0132 (GW906). The authors would like to acknowledge the support from Guangdong-Hong Kong-Macao Joint Laboratory for New Drug Screening (8326200).

Author contributions

X.F. collected the data, performed the analysis, and drafted the manuscript. X.F., Y.X., K.C., R.Z., and X.Y participated in the interpretation of the results and edited the manuscript. Y.X., K.C., R.Z., and X.Y participated in the conceptualization and interpretation of the analysis and critically reviewed the manuscript. The author(s) read and approved the final manuscript.

Declaration of interests

Y.X. and R.Z. are employees of Shenzhen Salubris Pharmaceuticals Co., Ltd.