Circulating Extracellular Vesicle MicroRNAs as Predictive Biomarkers for Kidney and Cardiovascular Events

Abstract

Background

Chronic kidney disease (CKD) leads to premature mortality from cardiovascular events before kidney replacement therapy. Despite recognition of syndromes like cardiorenal anemia and cardiovascular‐kidney‐metabolic, predictive models for kidney and cardiovascular outcomes remain inadequate. This study aimed to develop a minimally invasive, risk model using circulating small extracellular vesicle‐derived miRNAs among patients with CKD.

Methods

A derivation cohort (n=36) underwent microarray‐based miRNA profiling, and a least absolute shrinkage and selection operator‐penalized Cox proportional hazards model was constructed. Validation was performed using TaqMan quantitative polymerase chain reaction in a cohort of 234 patients with CKD without kidney replacement therapy. The primary outcome was a ≥30% reduction in estimated glomerular filtration rate or progression to kidney replacement therapy. The secondary outcome included all‐cause mortality, kidney replacement therapy initiation, and major adverse cardiovascular events.

Results

In the derivation cohort, 36% of patients had hypertensive glomerulosclerosis as the underlying CKD cause, increasing to 48% in the validation cohort. Twenty‐three miRNAs were significantly downregulated in advanced CKD, associated with cellular senescence, FOXO (forkhead box, class O) signaling, and cell cycle pathways. From these, 3 miRNAs—hsa‐let‐7d‐5p, hsa‐miR‐24‐3p, and hsa‐miR‐126‐3p—were selected and integrated into the final risk score with cystatin C and urinary protein levels, following optimization in the validation cohort. Lower miRNA levels were linked to cardiovascular comorbidities and cardiorenal anemia syndrome. Over a median follow‐up of 39 and 59 months, 108 kidney events and 70 composite outcomes occurred. The model effectively predicted adverse outcomes across CKD causes, further stratifying risk within cardiovascular‐kidney‐metabolic stage classifications.

Conclusions

Circulating small extracellular vesicle‐derived miRNA profiles enable a noninvasive, longitudinally predictive model for adverse kidney and cardiovascular outcomes in CKD. This approach may improve early risk identification and clinical decision‐making.

Nonstandard Abbreviations and Acronyms

cEVs

circulating extracellular vesicles

CKM

cardiovascular‐kidney‐metabolic

miRNA

microRNA

VSMCs

vascular smooth muscle cells

Clinical Perspective.

What Is New?

• Distinct miRNA profiles within circulating extracellular vesicles were identified in individuals with chronic kidney disease.

• A least absolute shrinkage and selection operator‐penalized Cox proportional hazard’s model was developed incorporating 3 miRNAs to predict chronic kidney disease progression, cardiovascular events, and mortality.

What Are the Clinical Implications?

• The derived risk score equation provides a minimally invasive approach to predict future kidney and cardiovascular outcomes, and the molecular pathways activated due to the reduction of these miRNAs may play a role in the pathophysiology of the underlying mechanism of chronic kidney disease and its cardiovascular associations.

Kidney disease is now acknowledged as a global noncommunicable epidemic and a major contributor to premature mortality. ¹ Chronic kidney disease (CKD), defined by a glomerular filtration rate (GFR) <60 mL/min per 1.73 m² or persistent renal impairment indicated by findings such as proteinuria for >3 months, affects >850 million adults globally and its prevalence continues to increase. ² By the year 2040, CKD is projected to become the fifth leading cause of years of life lost. ¹ In addition to impairing fluid and electrolyte balance, CKD contributes to multimorbidity, increased clinical complexity, ³ , ⁴ and accelerated deterioration of physical and cognitive function. ⁵ Cardiovascular events represent a major pathway through which CKD leads to premature mortality. ⁶ Kidney protection is increasingly regarded as a key strategy in preventing heart failure. ⁷ Efforts to elucidate the mechanisms underlying the coexistence of CKD and cardiovascular disease (CVD) have led to various syndrome classifications, including cardiorenal syndromes, ⁸ cardiorenal anemia syndrome, ⁹ and cardiorenal cachexia syndromes, ¹⁰ culminating in the more recently proposed cardiovascular‐kidney‐metabolic (CKM) syndrome. ¹¹ Despite the recognition of these syndromes, their underlying mechanisms remain poorly understood, and there is still a lack of validated biomarkers for effective risk stratification.

The pathogenesis of CKD is multifactorial, commonly involving diabetes, hypertension, and atherosclerosis. CKD progression varies significantly among individuals and does not follow a uniform pattern, which complicates both risk prediction and personalized treatment planning. ¹² These challenges are key barriers to accurately forecasting long‐term renal outcomes and formulating individualized therapeutic approaches. Recent large‐scale clinical studies have introduced additional treatment options aimed at slowing CKD progression and minimizing associated complications, including sodium‐glucose cotransporter 2 inhibitors, glucagon‐like peptide‐1 receptor agonists, and mineralocorticoid receptor antagonists. ¹³ , ¹⁴ Consequently, identifying patients with rapidly declining renal function has become increasingly important for initiating appropriate treatment in a timely manner. At the same time, avoiding unnecessary interventions in individuals at low risk of progression is crucial for minimizing global health care costs.

Various approaches have been explored to predict renal outcomes, with a notable example being a single risk equation developed from a Canadian cohort that incorporates 4 variables: age, sex, estimated GFR (eGFR), and albuminuria. ¹⁵ More recently, there has been a growing number of studies focused on developing prognostic tools for renal outcomes using advanced technologies such as omics analyses and artificial intelligence. ¹⁶ , ¹⁷ However, a universally accepted, noninvasive method for predicting long‐term CKD prognosis across diverse populations has yet to be established. ¹⁸

Our previous research demonstrated that circulating extracellular vesicles (cEVs) derived from patients with CKD carry biologically active components that may facilitate interorgan signaling and contribute to disease progression. ¹⁹ We showed that CKD‐associated cEVs can transmit atypical signals that play a key role in harmful interorgan interactions. cEVs are membrane‐bounded vesicles under 200 nm in size, secreted by all cell types, and contain nucleic acids, proteins, and lipids. ²⁰ Although cEV‐encapsulated miRNAs have been extensively investigated in oncology as diagnostic and prognostic markers, their relevance in kidney disease remains insufficiently explored. ²¹ We hypothesized that cEV miRNA profiles could support multiorgan network analyses and serve as predictive indicators for CKD progression.

Through comprehensive transcriptomic analysis of human miRNAs, followed by validation using quantitative polymerase chain reaction (qPCR), we aim to develop a minimally invasive, longitudinal prognostic model that integrates key clinical factors. This approach has the potential to advance individualized treatment planning and improve the precision of long‐term kidney outcome predictions.

METHODS

Data Availability

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

Data Source and Cohort Selection

Both the prospective derivation cohort and the prospective validation cohort studies were conducted at our university hospital, Institute of Science Tokyo Hospital. All patients gave written informed consent before enrollment.

The validation cohort was originally designed to investigate the epidemiological characteristics and risk factors for sarcopenia and rapid CKD progression in older patients. ²² , ²³ Patients with CKD aged ≥65 years without kidney replacement therapy (KRT) were enrolled between June 2016 and March 2017. As shown in Figure S1, the inclusion criteria were (1) age >65 years; (2) CKD stages G3 to G5, according to the Kidney Disease: Improving Global Outcomes (KDIGO) classification ²⁴ ; (3) no malignancies; (4) no active infection; (5) no corticosteroid treatment before recruitment; and (6) no major surgery within 6 months before enrollment. Out of 260 patients meeting these criteria, exclusions were made for missing data on variables of interest (n=4), missing follow‐up (n=4), initiation of KRT within 1 month after enrollment (n=1), and lack of availability for purification of cEVs and miRNA analysis within cEVs (n=17). The validation cohort ultimately included 234 participants aged 65 to 91 years for analysis of clinical data, cEVs, and encapsulated miRNAs. GFR was estimated using the 3‐variable Japanese GFR equation developed by the Japanese Society of Nephrology: eGFR (mL/min per 1.73 m²)=194×serum creatinine^−1.094×age^−0.287 (multiplied by 0.739 if female). ²⁵ This equation is commonly used for Japanese populations, who tend to have lower serum creatinine levels compared with White populations, as the Chronic Kidney Disease Epidemiology Collaboration equation tends to overestimate GFR in Japanese patients. ²⁵ Blood and urine samples were collected during follow‐up visits with patient consent. Patients were followed until death, transfer, initiation of KRT, or February 2022.

The derivation cohort, ¹⁹ established later, was specifically designed to evaluate the prognostic utility of cEVs and their miRNA content in patients with CKD. The derivation cohort included participants based on the following criteria: (1) inpatients or outpatients at our university hospital between October 2020 and December 2021, (2) age 18 years or older, and (3) CKD without KRT, according to the KDIGO classification. ²⁴ Patients were excluded if KRT was started within 1 month after enrollment (n=1). Notably, 16 participants overlapped with the validation cohort, as they underwent abdominal computed tomography scans required for a separate analysis of aortic calcification and were eligible for inclusion in both studies. The derivation cohort comprised a total of 36 adults with CKD, aged 22 to 86 years. eGFR was calculated using the same 3‐variable Japanese GFR equation developed by the Japanese Society of Nephrology. ²⁵ The baseline eGFR was defined as the average of 3 measurements taken before and after the enrollment date.

The protocols of both studies were approved by the ethics committee of the Institute of Science Tokyo (M2020‐134 and M2000‐2224), and the studies were conducted in accordance with the principles outlined in the Declaration of Helsinki.

Definitions of Diseases

A history of hypertension was defined by a systolic blood pressure of 140 mm Hg or higher, a diastolic blood pressure of 90 mm Hg or higher, or treatment with antihypertensive medications. ²⁶ Diabetes was defined as having hemoglobin A1c levels of ≥6.5% or the use of antidiabetic medications. ²⁷ A history of CVDs included any occurrence of stroke, ischemic heart disease, congestive heart failure, or peripheral arterial disease. Exercise habits were identified when patients regularly performed aerobic exercise, resistance training, or other sports activities. CKM staging, ¹¹ modified from the initially published definitions, was applied based on the available data set, as detailed in Table S3. Additionally, KDIGO criteria were used for CKD risk stratification via a heatmap based on urinary protein‐to‐creatinine ratio (UPCR) and eGFR (Figure S2). All participants in both cohorts underwent whole‐body dual‐energy X‐ray absorption to assess skeletal muscle mass and total body mineral density. Muscle mass was calculated using the skeletal muscle mass index, defined as the sum of lean mass in the arms and legs (kg) divided by height² (m²). ²⁸ Handgrip strength was measured by physicians using a handheld dynamometer. Patients gripped the device with maximum force while standing, keeping their arms straight at their sides. The better of 2 attempts for each upper limb was recorded, and the highest value was used for analysis. Slow gait speed was defined as walking at 0.8 m/s or slower. Sarcopenia was diagnosed according to the 2014 criteria of the Asian Working Group for Sarcopenia as follows: (1) age ≥65 years, (2) low handgrip strength (<26 kg for men and <18 kg for women) or low usual gait speed (<0.8 m/s), and (3) low skeletal muscle mass index (<7.0 kg/m² for men and <5.4 kg/m² for women). ²⁸

Outcome of the Derivation and Validation Prospective Cohorts

In both the derivation and validation cohorts, the primary outcome was a kidney outcome defined as a 30% decline in creatinine‐based eGFR or the initiation of KRT. For the validation cohort, the secondary outcome was a composite measure including all‐cause mortality, initiation of KRT, and hospital admissions due to major adverse cardiovascular events (MACEs), such as acute myocardial infarction, stroke, and heart failure. ²⁹ Follow‐up clinical data were collected from outpatient visits, hospital admissions, or phone contacts and recorded in electronic health records. Additionally, the annual decline in eGFR was evaluated using mixed‐effect models based on multiple eGFR measurements during the follow‐up period.

cEV Purification and Characterization

Venous blood was collected from the participants and allowed to remain at room temperature for ≈60 minutes. After centrifugation at 3000g for 10 minutes, the supernatant was collected and stored at −80 °C until further use. For this study, cEVs are defined as small EVs present in peripheral blood, isolated from stored serum samples. This definition is based on the source (serum) and size characteristics of the vesicles. cEVs were isolated from 200 or 250 μL of serum using a polymer‐based precipitation method with the Total Exosome Isolation kit (from serum, Cat. 4478360, Invitrogen, CA). ³⁰ , ³¹ This method was chosen due to its high recovery efficiency (>90% of cEVs) and suitability for clinical application. Following the manufacturer’s instructions, serum was centrifuged at 2000g for 30 minutes to remove cell debris (Figure 1B). One‐fifth volume of Total Exosome Isolation reagent was mixed with serum and incubated at 4 °C for 30 minutes. After centrifugation at 10000g for 10 minutes, the pellet was resuspended in PBS for further analysis. To validate the sizes and morphologies of the isolated cEVs, we performed transmission electron microscopy and Videodrop analysis, which uses interferometric light microscopy technology (Myriade, Paris, France). These analyses confirmed that the majority of vesicles were within the size range consistent with small EVs (typically 50–200 nm), supporting the validity of our purification approach and the classification of these vesicles as cEVs.

Figure 1. Unbiased signature analysis of cEV‐encapsulated miRNAs in patients with CKD not on dialysis and development of a Lasso‐Cox prediction model for longitudinal kidney outcomes.

A, Graphical summary of the study methods. Transcriptomic analysis of miRNAs carried by cEVs was conducted in a derivation cohort. A Lasso‐Cox prediction model was created using a subset of miRNAs, and the model equation was refined in the validation cohort by quantifying miRNAs through quantitative PCR. B, Polymer‐based purification was applied for subsequent cEV analyses. C, Representative size distributions of cEVs from a derivation cohort patient. D, Electron microscopy image of a single EV. E, Immunoblotting results for EV marker proteins. F, Heatmap displaying miRNA expression levels per case in the derivation cohort, clustered by k‐means. All 7 cases clustered on the right side of the heatmap, highlighted in light gray in the top row, correspond to CKD stage G5 (eGFR <15.0 mL/min per 1.73 m²). Twenty‐three miRNAs within “cluster F” were selected due to their reduced expression in patients with CKD stage G5. G, Using miRTargetLink 2.0 software, 50 target molecules shared by at least 3 of the 23 miRNAs, with strong sequence evidence supported by experimental validation, were identified (Table S2). H and I, The top 6 KEGG pathways unrelated to cancer were identified; these pathways include multiple shared hub genes illustrated by STRING version 12.0. J, Cross‐validation indicated that 3 variables maximized the concordance index. K, Kaplan–Meier curves for kidney outcomes in the derivation cohort, dividing patients into 2 groups based on the threshold of the Lasso‐Cox‐derived M3 risk score equation for maximizing the Youden index. Event‐free survival was compared by log‐rank test. P values <0.05 were considered statistically significant. AGE‐RAGE indicates advanced glycation end‐products‐receptor for advanced glycation end‐products; cEV, circulating extracellular vesicle; CKD, chronic kidney disease; eGFR, estimated glomerular filtration rate; FDR, false discovery rate; IB, immunoblotting; KEGG, Kyoto Encyclopedia of Genes and Genomes; KRT, kidney replacement therapy; Lasso, least absolute shrinkage and selection operator; MACE, major adverse cardiovascular event; miRNA, microRNA; and PCR, polymerase chain reaction.

Immunoblotting

The purified cEVs, as described previously, were resuspended in PBS and mixed with an equal volume of 2× SDS sample buffer (Cosmo Bio USA, Carlsbad, CA) and then denatured at 95 °C for 5 minutes. The protein extracts were separated by SDS‐PAGE, transferred electrically onto a nitrocellulose membrane, and incubated with the following primary antibodies: mouse anti‐CD9 (sc‐13 118, 1:1000; Santa Cruz Biotechnology, Dallas, TX), rabbit anti‐CD63 (ab134045, 1:1000; Abcam, Cambridge, UK), mouse anti‐CD81 (sc‐166 029, 1:1000, Santa Cruz Biotechnology), mouse anti‐Alix (sc‐53 540, 1:1000; Santa Cruz Biotechnology), rabbit anti‐Flotillin‐1 (18 634, 1:1000, Cell Signaling Technology, Topsfield, MA), rabbit anti‐Flotillin‐2 (3436, 1:1000; Cell Signaling Technology), and rabbit anti‐β‐actin (13E5, 1:1000; Cell Signaling Technology). Secondary antibodies were anti‐mouse IgG (H+L) AP conjugate (S3721, 1:5000; Promega Corporation, Madison, WI) and anti‐rabbit IgG (H+L) HRP conjugate (W4011, 1:5000; Promega Corporation).

miRNA Transcriptomic Analysis

Total RNAs within cEVs were purified using a Total Exosome RNA and Protein Isolation Kit (Cat.4478545, Invitrogen) according to the manufacturer’s instructions. ³⁰ , ³¹ The array‐based transcriptomic analysis of miRNAs was conducted using an Affymetrix GeneChip miRNA 4.0 (Affymetrix, Santa Clara, CA). After RNA quality was confirmed by spectrophotometry (Implen GmbH, Munich, Germany) and an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA), 3000 ng of each sample was biotin‐labeled with a Flash TagTM Biotin HSR RNA labeling kit for Affymetrix GeneChip miRNA arrays (Affymetrix) following the manufacturer’s protocol. The labeled RNA was hybridized onto a GeneChip miRNA 4.0 Array (Affymetrix) at 48 °C for 18 hours. The arrays were washed with a GeneChip Fluidics Station 450 (Affymetrix) and scanned using a GeneChip Scanner 3000 7G (Affymetrix).

Following hybridization and scanning, raw data were processed using Affymetrix Expression Console Software. To ensure high‐quality and biologically meaningful miRNA signals, we applied a multistep filtering strategy based on probe detection flags and expression levels. First, miRNAs detected in fewer than 3 samples (ie, with <3 “P” (pass) flags across all 36 cases) were excluded to eliminate low‐confidence signals. For the remaining miRNAs, detection consistency was assessed across 3 expression‐level groups (the low, median, and high groups), which were defined based on eGFR. The low group included samples with eGFR <15, the median group included those with 15≤eGFR<50, and the high group included those with 50 ≤ eGFR. A miRNA was retained for downstream analyses only if it met at least 1 of the following criteria:

1. The proportion of “P” flags was ≥0.4 in either the low‐expression group or the high‐expression group.

2. The proportion of “P” flags was ≥0.3 in the low or high group and ≥ 0.2 in the median group.

3. The proportion of “P” flags was ≥0.2 in all 3 groups.

This stringent filtering process was designed to prioritize miRNAs with consistent and reliable detection across a range of expression levels.

Quantitative Real‐Time PCR

Total RNA within cEVs was isolated using the Total Exosome RNA and Protein Isolation Kit (Cat. 4478545, Invitrogen), followed by reverse transcription with primers specified for each miRNA using the TaqMan MicroRNA Assays (Applied Biosystems, CA) and the TaqMan MicroRNA Reverse Transcription Kit (Applied Biosystems). real‐time qPCR was carried out on a LightCycler 480 System II (Roche, Basel, Switzerland) using THUNDERBIRD Next Probe qPCR Mix (TOYOBO CO., LTD, Osaka, Japan) with specific TaqMan qPCR primers. PCR amplification involved 50 cycles at 95 °C for 5 seconds and 60 °C for 30 seconds after an initial denaturation at 95 °C for 60 seconds. The 2^−ΔΔct method was used to quantify miRNA expression levels, normalizing the relative expression of each target gene to has‐miR‐139‐5p as an internal control. ¹⁹

Statistical Analysis

Baseline characteristics are presented as numbers and percentages for categorical variables or medians (25th–75th percentile) for continuous variables. Group comparisons between the derivation and validation cohorts were performed to assess differences in baseline characteristics. For continuous variables, nonparametric Mann–Whitney U tests were applied. Categorical variables were compared using the chi‐square test or Fisher’s exact test depending on cell counts. Statistical comparisons were made using 1‐way ANOVA followed by Turkey’s post hoc test for >2 groups, or an unpaired t test between 2 groups, when variables showed normal distribution and homogeneity of variance. The Shapiro–Wilk test was used to assess normality. For nonparametric variables or when the sample size was <6 and normality could not be reliably assessed, nonparametric tests were applied: Mann–Whitney U tests for 2‐group comparisons and Kruskal–Wallis test for multiple comparisons, followed by Dunn’s multiple comparisons test. The chi‐square test compared categorical variables. The Jonckheere–Terpstra test was applied to detect trends among nonparametric variables. Batch effect correction was performed using the removeBatchEffect function in the limma R package, and endogenous control miRNAs were identified using geNorm and NormFinder algorithms to ensure robust normalization. Differential expression analyses between CKD cases and controls were performed using the limma R package, with batch effects included as covariates. miRNAs with an absolute log₂ fold change >1 and a false discovery rate <0.1 were considered significant. Hierarchical and k‐means clustering, along with heatmap visualizations, were performed on normalized and quality‐filtered miRNA expression data. We applied mean and variance filters to exclude low‐expression and low‐variability miRNAs, resulting in 519 features used for clustering. The number of clusters (k) was treated as a hyperparameter and explored from 2 to 20. The optimal k=10 was selected based on standard clustering metrics and biological interpretability. Heatmaps were generated using scaled expression values, and CKD G5 was designated as the case group (highlighted in light gray in the top row) and was clearly distinguished from other stages (displayed in dark gray in the top row). The least absolute shrinkage and selection operator (Lasso)‐penalized Cox proportional hazard’s model (using the glmnet R package) was applied to optimize the prediction equation using multiple variables. The penalty parameter (λ) was optimized through five‐fold cross‐validation to maximize the concordance index (C‐index). In both the derivation and validation cohorts, we refitted unpenalized Cox proportional‐hazards models using the Lasso‐selected predictors and conducted formal diagnostic assessments. The proportional‐hazards assumption was tested with Schoenfeld residuals (global and covariate‐specific tests). Martingale and deviance residuals were examined for nonlinearity and overall fit, and Dfbeta statistics were inspected to identify influential observations. Kaplan–Meier survival curves and log‐rank tests were used to evaluate the association of miRNA expression level categories, risk score equations, or CKD and CKM stages with outcomes. Subgroup analyses examining the relationship between established risk score equations and outcomes were conducted using a Cox proportional hazards model adjusted for covariates including age (<75 years or ≥ 75 years), sex, body mass index (<25 kg/m² or ≥25 kg/m²), history of CVD, hypertension, diabetes, eGFR (<30 mL/min per 1.73 m² or ≥30 mL/min per 1.73 m²), and UPCR (<0.5 g/gCr or ≥0.5 g/gCr). The discriminatory ability of the established risk score and traditional biomarkers (serum creatinine, cystatin C, and urinary protein) for kidney outcomes at 1‐ and 2‐year follow‐up was evaluated using the C‐index. According to general guidelines, models with a C‐index >0.7 are considered good predictors, those >0.8 indicate very good performance, and a value of 1.0 represents a perfect model. To further assess the incremental value of the established risk score compared with individual biomarkers, we performed integrated discrimination improvement and continuous net reclassification improvement analyses. ³² These metrics quantify the improvement in risk prediction and classification when a new marker or model is added. Net reclassification improvement evaluates the correctness of reclassification into risk categories, and integrated discrimination improvement measures the improvement in average sensitivity without sacrificing specificity. Model calibration was assessed by estimating the linear association between predicted risk scores and observed event incidence using bootstrapped calibration analysis. ³³ Specifically, we constructed a program to repeatedly resample the data set (200 iterations) and calculate regression slopes and intercepts between the original and resampled prediction scores. This procedure provided bias‐corrected estimates of calibration slope and intercept, accounting for potential optimism in model fitting. Logistic regression and restricted cubic spline analyses assessed the association between miRNA expression levels within cEVs and the risk of rapid eGFR decline (defined as <−5 mL/min per 1.73 m²), ³⁴ adjusting for age, sex, body mass index, UPCR, eGFR, CVD, diabetes, and hypertension. Target genes of miRNAs were analyzed using miRTargetLink 2.0, a tool that constructs hierarchical miRNA–gene interaction networks. ³⁵ Additionally, STRING (version 12.0) was used to perform enrichment analysis of predicted target genes, which can identify pathways targeted by multiple miRNAs. ³⁶ Statistical analyses and graphical outputs were generated using Stata (version 18.0 software, Stata Corp., College Station, TX), GraphPad Prism (version 10, GraphPad Software, Inc., Boston, MA), or R (version 4.4.1; including the limma, glmnet, survival, and qvalue packages). All tests were 2 sided, and values <0.05 were considered statistically significant unless otherwise stated.

RESULTS

Patient Characteristics of Derivation and Validation Cohorts

Baseline demographics and clinical characteristics of the study participants are presented in Table 1 and Table S1. The derivation cohort included 36 patients with CKD who were not on KRT. The median age was 71 years (interquartile range [IQR], 61–79), with 25% female, and a median body mass index of 22.3 kg/m² (IQR, 20.2–26.4). CVD, hypertension, ²⁶ and diabetes ²⁷ were observed in 22%, 83%, and 25% of patients, respectively. The primary CKD causes were hypertensive nephrosclerosis (36%), diabetic kidney disease (11%), chronic glomerulonephritis (25%), and other causes (28%). Median values for creatinine‐based eGFR, ²⁵ UPCR, and hemoglobin A1c were 33.9 mL/min/1.73 m² (IQR, 20.7–51.2), 0.29 g/gCr (IQR, 0.12–2.62), and 5.9% (IQR, 5.7–6.2), respectively. CKD stages ranged from G1 to G5, with stages G3b and G4 representing the majority of patients in the derivation cohort.

Table 1.

Baseline Characteristics of the Derivation and the Validation Cohorts

Variable	Derivation cohort (n=36)	Validation cohort (n=234)	P value
Age, y	71 (61–79)	76 (69–81)	0.001
Female sex, %	9 (25)	82 (35)	<0.001
Body mass index, kg/m2	22.3 (20.2–26.4)	22.5 (20.8–24.9)	0.885
History of cardiovascular disease, %	8 (22)	68 (29)	0.396
Hypertension, %	30 (83)	179 (77)	0.361
Diabetes, %	9 (25)	67 (29)	0.652
Primary cause of CKD, %			0.403
Hypertensive nephrosclerosis	13 (36)	113 (48)
Diabetic kidney disease	4 (11)	31 (13)
Chronic glomerulonephritis	9 (25)	46 (20)
Others	10 (28)	44 (19)
Hemoglobin, g/dl	12.6 (10.8–14.3)	12.5 (11.4–13.9)	0.867
Creatinine, mg/dl	1.52 (1.05–2.26)	1.52 (1.20–2.33)	0.609
Creatinine‐based estimated glomerular filtration rate, mL/min per 1.73 m2	33.9 (20.7–51.2)	32.2 (20.5–41.7)	0.214
CKD stage, %			<0.001
G1	2 (5.6)	0 (0)
G2	4 (11)	0 (0)
G3a	5 (14)	39 (18)
G3b	9 (25)	84 (36)
G4	9 (25)	78 (33)
G5	7 (19)	33 (14)
Urinary protein‐to‐creatinine ratio, g/gCr	0.29 (0.12–2.62)	0.29 (0.14–0.96)	0.374
Corrected calcium, mg/dl	9.2 (9.1–9.6)	9.3 (9.0–9.5)	0.758
Phosphate, mg/dl	3.4 (3.0–3.8)	3.4 (3.0–3.7)	0.436
Glucose, mg/dl	111 (92.0–130)	105 (95.0–119)	0.692
Hemoglobin A1c, %	5.9 (5.7–6.2)	6.0 (5.7–6.4)	0.168
Total cholesterol, mg/dl	178 (157–213)	182 (164–213)	0.769
Low‐density lipoprotein cholesterol, mg/dl	117 (88.0–137)	105 (89.0–127)	0.314
High‐density lipoprotein cholesterol, mg/dl	52 (44–63)	52 (44–68)	0.787
Triglyceride, mg/dl	137 (104–191)	126 (96.0–179)	0.734

Data are presented as numbers (percentages) or medians (25th–75th percentile).

For group comparisons, nonparametric Mann–Whitney U tests were applied to continuous variables, and categorical data were analyzed using chi‐square or Fisher’s exact tests depending on cell counts. CKD indicates chronic kidney disease.

The external validation cohort ²² , ²³ consisted of 234 patients with CKD not undergoing KRT. Their baseline characteristics are also summarized in Table 1 and Table S1. Compared with the derivation cohort, the validation cohort was slightly older, had a significantly greater proportion of female patients, and displayed similar rates of CVD, hypertension, and diabetes. The most common causes of CKD in this group were hypertensive nephrosclerosis (48%) followed by chronic glomerulonephritis (20%). CKD stages G3 and G4 were most frequent, with no participants classified as stage G1 or G2. The majority of patients were receiving antihypertensive treatment, with renin–angiotensin–aldosterone system inhibitors being the most commonly prescribed drugs—72% in the derivation cohort and 64% in the validation cohort (Table S1). Approximately half of the patients were treated with urate‐lowering medications (44% and 50%, respectively). Erythropoiesis‐stimulating agents (ESAs) were given to 31% of the derivation cohort and 22% of the validation cohort. Serum cystatin C levels were lower in the validation cohort (median 1.77 [IQR, 1.32–2.49]) compared with the derivation cohort (median 2.23 [IQR, 1.62–2.60]), resulting in higher cystatin C‐based eGFR values in the validation group (median 33.8 [IQR, 20.7–47.6]) versus the derivation group (median 24.9 [IQR, 19.3–39.2]).

miRNA Transcriptomic Analysis Using Human cEVs From the Derivation Cohort

In previous research, we identified 4 miRNAs that were reduced in cEVs from rodent models of CKD. ¹⁹ Considering the greater diversity and complexity of miRNAs in humans compared with rodents, we aimed to profile human‐specific cEV‐encapsulated miRNAs with potential as biomarkers and signaling molecules in CKD.

Figure 1A presents an overview of the study protocol. miRNA transcriptomic analysis was conducted using cEVs isolated from the serum of the 36 participants in the prospective derivation cohort. Using Lasso‐Cox analysis, ³⁷ we developed an equation to predict kidney outcomes, defined as a ≥ 30% decline in eGFR or initiation of KRT. A total of 23 miRNAs and creatinine‐based eGFR were included in the Lasso‐penalized Cox proportional hazard’s model. Three specific miRNAs were selected during the initial Lasso analysis. In the validation cohort of 234 patients, these cEV‐encapsulated miRNAs were measured by TaqMan qPCR. The predictive equation was further refined by incorporating these miRNAs along with 41 covariates routinely collected in clinical practice. Additionally, we evaluated the ability of this risk score to predict not only kidney outcomes but also a composite end point including all‐cause mortality, KRT initiation, and hospital admissions due to MACEs. ²⁹

To comprehensively quantify human miRNAs within cEVs from patients with CKD, we conducted array‐based miRNA transcriptomic analysis using cEV samples from 36 patients with CKD not on dialysis. As illustrated in Figure 1B, cEVs were purified using a polymer‐based precipitation method. ³⁰ , ³¹ Size distribution histograms obtained through interferometric light microscopy technology (Videodrop, Myriade, Paris, France) and electron microscopy images of individual cEVs confirmed successful purification of cEVs from patient serum (Figure 1C and 1D). Immunoblotting verified the presence of cEV protein markers, including tetraspanins CD9, CD63, and CD81, as well as ALIX, Flotillin‐1, and Flotillin‐2, ²⁰ along with β‐actin as a housekeeping protein (Figure 1E).

Figure 1F presents a heatmap showing the expression profiles of 519 miRNAs that passed a stringent multistep quality control process out of 2578 detected miRNAs. These miRNAs were selected based on consistent detection across 36 patients with CKD not on KRT, using probe quality flags and group‐wise expression thresholds (see Methods for details). The filtered miRNAs were analyzed using the k‐means clustering method (k=10), ³⁸ revealing distinct expression patterns associated with CKD severity. The light gray bars at the top mark patients with CKD stage G5, defined by an eGFR <15 mL/min/1.73 m². The modified equation for the Japanese population was used to calculate creatinine‐based eGFR. ²⁵ The group with severe CKD, consisting of G5 patients located on the right side, was characterized by reduced expression of 23 miRNAs grouped in “Cluster F.” This cluster included hsa‐let‐7a‐5p, hsa‐let‐7b‐5p, hsa‐let‐7c‐5p, hsa‐let‐7d‐5p, hsa‐miR‐15b‐5p, hsa‐miR‐16‐5p, hsa‐miR‐17‐5p, hsa‐miR‐20a‐5p, hsa‐miR‐23a‐3p, hsa‐miR‐24‐3p, hsa‐miR‐26a‐5p, hsa‐miR‐92a‐3p, hsa‐miR‐103a‐3p, hsa‐miR‐106a‐5p, hsa‐miR‐107, hsa‐miR‐122‐5p, hsa‐miR‐126‐3p, hsa‐miR‐191‐5p, hsa‐miR‐320a, hsa‐miR‐320b, hsa‐miR‐320c, hsa‐miR‐486‐5p, and hsa‐miR‐4497. The expression levels of miRNAs in the other 9 k‐means clusters did not show any association with the proportion of patients in CKD stage G5. Notably, this cluster of 23 miRNAs included 3 miRNAs—miR‐16‐5p, miR‐17‐5p, and miR‐20a‐5p—that were previously identified in our rodent study. ¹⁹

To investigate the biological relevance of these 23 miRNAs, we used a database for miRNA‐target gene interactions (miRTargetLink2.0), ³⁵ identifying 50 target genes regulated by 2 or more of these miRNAs (Figure 1G and Table S2). Pathway analysis was conducted to determine enriched pathways among these 50 miRNA‐targeted genes. Figure 1H and 1I, and Figure S3A through S3F show the top 6 noncancer‐related KEGG pathways, including cellular senescence, FOXO (forkhead box, class O) signaling pathway, cell cycle, advanced glycation end‐products‐RAGE (receptor for advanced glycation end‐products) signaling pathway in diabetic complications, PI3K‐Akt signaling pathway, and endocrine resistance, as visualized with STRING version 12.0. ³⁶ These pathways (Figure 1I) share several hub genes that intersect with the 50 miRNA‐targeted genes (Figure 1H). These results indicate that under normal conditions, cEVs suppress these biological pathways. Conversely, in CKD, the reduction of this subset of miRNAs may facilitate harmful signaling cascades involved in disease progression.

Development of a Lasso‐Cox Model to Predict Longitudinal Kidney Outcomes Using the Derivation Cohort

To create a prediction model for CKD progression based on miRNAs within cEVs, we initially applied the Lasso‐penalized Cox proportional hazard’s model (using the glmnet R package) and developed an optimized risk score equation for a kidney outcome of ≥30% eGFR decline or initiation of KRT.

Over a median follow‐up of 23 months (IQR, 11–32), 22 (61%) kidney outcomes occurred, including 19 cases with ≥30% eGFR reduction and 7 with KRT initiation. KRT initiation occurred exclusively in the high‐risk group, affecting 7 of 23 patients (30.4%) during follow‐up. No events were observed in the low‐risk group. Fivefold cross‐validation was used to select the penalty parameter (λ) that maximized the concordance index. A Cox model was then fit using the optimal λ, incorporating 23 selected miRNAs, baseline creatinine‐based eGFR, and adjusting for batch effects from 3 separate transcriptomic experiments. Here, although the expression data were initially corrected for batch effects, the batch variable—reflecting the 3 distinct transcriptomic runs—was also included in the regression model to further control for any residual batch‐related confounding.

This process identified 3 miRNAs—hsa‐let‐7d‐5p, hsa‐miR‐24‐3p, and hsa‐miR‐126‐3p—as key contributors to the equation named the M3 equation (Figure 1J).

Higher values of the risk score were associated with greater risk.

These miRNAs showed stronger predictive ability for kidney outcomes than eGFR alone within the model. By optimizing the M3 risk score cutoff based on the Youden index and plotting Kaplan–Meier curves for high and low groups, the model significantly stratified renal prognostic risk (log‐rank test, P=0.002) (Figure 1K). KRT initiation occurred exclusively in the high‐risk group, affecting 46 of 117 patients (39.3%) during follow‐up. No events were observed in the low‐risk group.

Model diagnostics, including Schoenfeld residuals, Martingale and deviance residuals, and Dfbeta statistics, showed no violations of the proportional hazards assumption or influential observations, supporting the adequacy of the Lasso‐Cox model (Figure S4).

Validation of miRNAs Predictive Performance in 234 Patients With CKD Using cEV‐Encapsulated Expression Profiles

To determine if the 3 miRNAs identified in the derivation cohort maintained their prognostic significance, we measured their expression levels in the external validation cohort (n=234) using TaqMan qPCR.

cEVs were isolated by polymer‐based precipitation (Figure 1B), and their purity was confirmed through interferometric light microscopy, electron microscopy, and Western blotting (Figure 2A through 2C). As presented in Figures 2D through 2F, the expression of each miRNA declined progressively with advancing CKD stage and decreasing eGFR. The Jonckheere–Terpstra test showed significant linear trends between higher CKD stages and lower miRNA levels for hsa‐let‐7d‐5p, hsa‐miR‐24‐3p, and hsa‐miR‐126‐3p (p=2.0 × 10⁻⁴, 2.0 × 10⁻⁴, and 1.0 × 10⁻², respectively).

Figure 2. Validation of miRNAs predictive performance in 234 patients with CKD using cEV‐encapsulated expression profiles.

A, Size distribution histogram of cEVs purified from serum via polymer‐based precipitation (see also Figure 1B). B, Electron microscopy image of representative cEVs. C, Immunoblotting confirming cEV markers (CD9, CD63, CD81, Alix, Flotillin‐1, Flotillin‐2) and loading control (β‐Actin). D through F, Expression levels of hsa‐let‐7d‐5p, hsa‐miR‐24‐3p, and hsa‐miR‐126‐3p (relative to hsa‐miR‐139‐5p), stratified by CKD stage: G3a (eGFR 45.0–59.9, n=39), G3b (eGFR 30.0–44.9, n=84), G4 (eGFR 15.0–29.9, n=78), and G5 (eGFR <15.0, n=33). miRNA expression decreased with CKD severity. Kruskal–Wallis and Dunn’s multiple comparison test were applied for multiple group comparisons; the Jonckheere–Terpstra test assessed linear trends. P values <0.05 were considered statistically significant. G through I, Kaplan–Meier curves showing kidney event‐free survival (≥30% decline in eGFR or initiation of KRT) for high vs low miRNA expression (median cutoff). P values were calculated using the log‐rank test and P values <0.05 were considered statistically significant. Shaded areas indicate 95% CIs. J through L, Restricted cubic spline models evaluating rapid eGFR decline (≥5 mL/min per 1.73 m²/year), adjusted for age, sex, body mass index, history of cardiovascular disease, hypertension, diabetes, eGFR, and urinary protein‐to‐creatinine ratio. Solid line: Estimated hazard ratio from restricted cubic spline model; dotted lines: 95% CIs. Higher miRNA levels were consistently associated with lower risk for all 3 miRNAs. cEV indicates circulating extracellular vesicle; CKD, chronic kidney disease; eGFR, estimated glomerular filtration rate; IB, immunoblotting; KRT, kidney replacement therapy; and MiRNA, microRNA.

To assess the contribution of each miRNA to risk stratification, patients were divided into high and low expression groups based on the median, and Kaplan–Meier curves were generated (Figure 2G and 2I). The log‐rank test showed that lower expression groups of hsa‐let‐7d‐5p, hsa‐miR‐24‐3p, and hsa‐miR‐126‐3p were associated with a higher incidence of kidney outcomes (log‐rank test P < 0.0001, 0.033, and 0.077 compared with higher expression groups, respectively).

To adjust for underlying covariates and reduce the influence of baseline eGFR on kidney outcomes, we also analyzed the risk of rapid eGFR decline, defined as a decrease of −5 mL/min per 1.73 m²/year or more, ³⁴ using multivariable logistic regression models. Both logistic regression and restricted cubic spline analyses, using median miRNA expression as a reference, showed that higher miRNA levels were consistently linked to a lower risk of rapid eGFR decline, independent of baseline kidney function and proteinuria (Figure 2J through 2L). Specifically, expression levels below the median were strongly linked to an increased risk of rapid CKD progression in the analysis of each miRNA. These results confirm that lower levels of hsa‐let‐7d‐5p, hsa‐miR‐24‐3p, and hsa‐miR‐126‐3p in cEVs are independently associated with CKD progression, supporting their potential as minimally invasive biomarkers for long‐term kidney risk stratification.

Optimization of the Lasso‐Cox Model for Predicting Kidney Outcomes and Cardiovascular Events

To validate and optimize the Lasso‐Cox prediction model developed in the derivation cohort, we applied Lasso‐Cox analysis to the 234 participants in the validation cohort. This analysis included the expression levels of the 3 miRNAs in cEVs measured by qPCR along with 41 additional variables such as comorbidities, biochemical parameters, and medications.

In the validation cohort, over a median follow‐up period of 39 months (IQR, 22–61), 108 (46%) kidney outcomes occurred, consisting of 108 cases with ≥30% eGFR decline and 46 (20%) initiations of KRT. During a median follow‐up period of 59 months (IQR, 32–64), secondary outcomes were observed in 70 patients (30%), including 21 (9%) all‐cause deaths, 46 initiations of KRT, and 17 (7%) MACEs. We developed the M3V2 equation optimized for predicting kidney outcomes in the validation cohort. This equation includes the 3 miRNAs, each with negative coefficients consistent with the derivation cohort model, along with cystatin C and UPCR having positive coefficients, as follows:

$$\begin{array}{c}\mathrm{M}3\text{V2 risk score}=-0.1814\left[\mathit{hsa}-\mathit{let}-\mathit{\text{7d}}-\mathit{\text{5p}}\right]\hfill \\ -0.3113\left[\mathit{hsa}-\mathit{miR}-\mathit{24}-\mathit{\text{3p}}\right]\hfill \\ -0.2337\left[\mathit{hsa}-\mathit{miR}-\mathit{126}-\mathit{\text{3p}}\right]\hfill \\ +0.8096\left[\text{cystatin}\mathrm{C}\right]+0.6111\left[\text{UPCR}\right]\hfill \end{array}$$

Higher values of the risk score were associated with greater risk.

Diagnostics (Schoenfeld, Martingale and deviance residuals, and Dfbeta) identified a proportional‐hazards violation for cystatin C (Schoenfeld test P=0.0045); all other covariates were nonsignificant (P>0.05; Figure S5). As shown in Figure 3A, a risk score above the median was strongly associated with a higher cumulative incidence of kidney outcomes (log‐rank test P<0.0001). Performance of the M3V2 equation in predicting kidney outcomes at 1‐ and 2‐year follow‐up was assessed using the C‐index, which is appropriate for time‐to‐event data. As shown in Table 2, the C‐index values of the M3V2 equation consistently exceeded 0.9, indicating superior predictive performance compared with individual traditional biomarkers such as serum creatinine, cystatin C, or urinary protein. To further evaluate the incremental value of the M3V2 equation, we conducted integrated discrimination improvement and continuous net reclassification improvement analysis. ³² These analyses demonstrated that the M3V2 equation significantly improved risk classification at both the 1‐year and 2‐year follow‐up. Notably, although net reclassification improvement values were less pronounced at 1 year, they became clearly significant at 2 years, suggesting enhanced long‐term predictive utility of the M3V2 equation. Following the assessment of discriminatory performance, we additionally evaluated the calibration of the M3V2 equation by estimating the calibration slope and intercept based on bootstrap‐corrected regression of predicted risk scores. ³³ At both the 1‐year and 2‐year time points, the calibration slope approached unity (1.015 and 0.998, respectively), and the intercept was nearly 0, suggesting excellent agreement between predicted and observed event rates. These values indicate minimal systematic bias or overfitting and reinforce the reliability of the M3V2 equation in predicting individualized renal risk.

Figure 3. Kaplan–Meier curves illustrating predictive performance of the M3V2 risk score and CKM staging in the validation cohort (n=234).

A, Kaplan–Meier curve for kidney outcome (≥30% eGFR decline or initiation of KRT), stratified by CKM stages 2, 3, and 4. B, Kaplan–Meier curve for the composite outcome (all‐cause mortality, KRT initiation, or MACE), stratified by CKM stages 2, 3, and 4. P values were calculated using the log‐rank test and P values <0.05 were considered statistically significant. Shaded areas indicate 95% CIs. C through E, Expression levels of hsa‐let‐7d‐5p, hsa‐miR‐24‐3p, and hsa‐miR‐126‐3p (relative to hsa‐miR‐139‐5p), stratified by CKM stage: stage 2 (n=47), stage 3 (n=119), and stage 4 (n=68). miRNA expression decreased with CKM stage severity. Kruskal–Wallis and Dunn’s multiple comparison test were applied for multiple group comparisons; the Jonckheere–Terpstra test assessed linear trends. P values <0.05 were considered statistically significant. F through I, In CKM stages 3 (F and G) and 4 (H and I), the risk score further distinguished groups with favorable vs unfavorable kidney or composite outcomes. The M3V2 risk equation further stratified risk within each CKM stage. CKD indicates chronic kidney disease; CKM, cardiovascular‐kidney‐metabolic; eGFR, estimated glomerular filtration rate; KRT, kidney replacement therapy; and MACE, major adverse cardiovascular event.

Table 2.

Comparison of Predictive Performances of the M3V2 Risk Equation and Classical Risk Factors for Kidney Outcome

Covariates	C‐index (95% CI)	NRI (95% CI) gain by M3V2 vs base model	IDI (95% CI) gain by M3V2 vs base model	Calibration slope (95% CI)	Calibration Intercept (95% CI)
Kidney outcomes at 1‐y follow‐up
M3V2 equation	0.934 (0.898–0.969)	Reference	Reference	1.015 (0.195–1.835)	−0.28×10−9 (−3.36×10−9–2.80×10−9)
Serum creatinine	0.915 (0.863–0.966)	0.164 (−0.519–1.481)	0.110 (0.013–0.306)	1.457 (0.709–2.205)	2.215 (1.979–2.450)
Cystatin C	0.920 (0.884–0.957)	−0.336 (−0.749–1.027)	0.076 (0.009–0.240)	1.757 (0.928–2.587)	3.003 (2.722–3.283)
Urinary protein	0.899 (0.854–0.944)	0.071 (−0.404–1.279)	0.180 (0.053–0.326)	1.483 (0.588–2.377)	0.318 (0.200–0.436)
Kidney outcomes at 2‐y follow‐up
M3V2 equation	0.903 (0.870–0.937)	Reference	Reference	0.998 (0.154–1.843)	1.69×10−9 (−0.96×10−9‐4.33×10−9)
Serum creatinine	0.860 (0.808–0.912)	0.722 (0.067–1.199)	0.129 (0.048–0.235)	1.149 (0.517–1.781)	1.867 (1.668–2.065)
Cystatin C	0.879 (0.839–0.919)	0.266 (−0.106–0.840)	0.114 (0.049–0.205)	1.541 (0.767–2.314)	2.815 (2.552–3.078)
Urinary protein	0.863 (0.817–0.910)	1.154 (0.856–1.389)	0.151 (0.063–0.246)	1.514 (0.550–2.477)	0.347 (0.219–0.476)

The performance of the established risk equation, as well as individual biomarkers including serum creatinine, cystatin C, and urinary protein, in predicting kidney outcomes at 1‐y and 2‐ follow‐up was quantified using the C‐index. C‐index indicates concordance index; IDI, integrated discrimination improvement; and NRI, continuous net reclassification improvement.

Importantly, the high‐ and low‐risk score groups almost completely separated the occurrence of the composite end point of all‐cause mortality, KRT initiation, or cardiovascular events (Figure 3B). Over a follow‐up period >1000 days, individuals with a low‐risk score experienced no such events (log‐rank test P < 0.0001). Although initially aimed at predicting kidney outcomes, the model also unexpectedly generated equations capable of predicting more severe end points, including cardiovascular events. These results demonstrate the robustness and flexibility of the optimized M3V2 equation in predicting kidney outcomes as well as clinically important composite events in patients with CKD.

Close Association of miRNA‐Based Prediction Model With CKM Syndrome

To explore the biological significance of reduced miRNA expression, we analyzed patients with lower levels of cEV‐encapsulated miRNAs across various clinical and biochemical parameters. Figure S6 presents the relationships between high and low hsa‐let‐7d‐5p expression and clinical variables categorized by physiological systems: cardiovascular (orange), renal (blue), metabolic (yellow), hematopoietic (red), and musculoskeletal (green). Patients with lower hsa‐let‐7d‐5p expression in their cEVs had a higher prevalence of CVD history (P=0.0002), hypertension history (P=0.009), use of renin–angiotensin–aldosterone system inhibitors and other antihypertensive drugs (P=0.042 and 0.0002, respectively), use of loop diuretics (P=0.006), lower creatinine‐based eGFR (P=0.0002), lower cystatin C‐based eGFR (P=0.0006), higher UPCR (P=0.035), elevated intact parathyroid hormone (P=0.029), cholesterol (P=0.047), low‐density lipoprotein‐cholesterol (P=0.047), use of antihyperglycemic drugs (P=0.0007), use of antihyperuricemics (P=0.009), lower hemoglobin (P=0.015), and receiving ESA (P=0.0002). Comparable trends were found for hsa‐miR‐24‐3p (Figure S7) and hsa‐miR‐126‐3p (Figure S8), where lower expression correlated with higher prevalence of CVD and increased use of antihypertensives, loop diuretics, antidiabetic medications, and ESAs. These associations were stronger than those seen with eGFR decline alone, indicating a wider biological link between miRNA depletion and systemic CKD‐related complications such as ESA resistance and cardiorenal dysfunction. These results prompted further investigation into the relationship among miRNAs, CKM syndrome staging, and patient outcomes.

We aimed to examine the correlation between our new prediction model and the ability of CKM stages to predict outcomes. First, we analyzed the association between CKM staging and kidney or composite outcomes. Because complete biochemical data needed for the American Heart Association definition were not available, we used a modified version of the CKM staging criteria ¹¹ , ³⁹ (Table S3). As shown in Figure 3C and 3D, the risk of both kidney outcomes and composite outcomes increased with advancing CKM stage. The predictive accuracy of CKM staging surpassed that of the KDIGO CKD risk stratification heatmap ²⁴ (Figure S2) for both kidney and composite outcomes (Figures S9A and S9B). This confirmed the validity of risk stratification based on the CKM syndrome concept in Japanese patients. Furthermore, we found that the expression levels of the 3 miRNAs significantly decreased as CKM stage advanced (Figure 3E), showing that higher CKM stages were associated with lower miRNA expression in cEVs, suggesting a potential biological link between the 3 miRNAs and CKM syndrome.

Next, we assessed whether the M3V2 risk score could further stratify patients’ risk of adverse outcomes within each CKM stage. We also evaluated if the M3V2 risk score provided additional prognostic value within each CKM or KDIGO category. In CKM stage 2, all patients were classified into the low‐risk M3V2 group. For stages 3 and 4, the M3V2 risk score further divided patients according to risk for both kidney and composite outcomes (Figure 3F through 3I). A similar pattern of refinement was seen within the KDIGO red zone (Figure S9C through S9H). These results emphasize that miRNA depletion in cEVs reflects multisystem dysfunction in CKD and support combining molecular profiling with clinical CKM staging to improve risk prediction.

Stratified Analysis of the Association Between the Risk Score Equation and Clinical Subgroups

To assess whether the predictive accuracy of the M3V2 risk score differs across clinically relevant subgroups, especially by CKD cause, we performed stratified analyses.

As shown in Figure 4, a risk score above the median value was linked to a higher risk of kidney disease progression across all subgroups, including age, sex, body mass index (overweight or not), CVD, hypertension, diabetes, primary CKD causes, eGFR (<30 or ≥ 30), and UPCR (<0.5 or ≥ 0.5). These results indicate that the predictive ability of the risk score is broadly applicable across patients with CKD in clinical settings, irrespective of their backgrounds or CKM syndrome status. Finally, we compared intervention‐modifiable risk factor profiles between patients with low and high M3V2 risk scores (Figure S10). Those with higher risk scores were more reliant on antihypertensive drugs, loop diuretics, active vitamin D3, and ESA. Statin use prevalence did not differ between groups. Routine exercise habits were more common in the low‐risk score group, suggesting that exercise may be a modifiable factor that could lower the risk score.

Figure 4. Stratified analysis of the association between the risk score equation and clinical subgroups.

HRs with 95% CIs, calculated using Cox proportional hazards models, are presented in multivariable models adjusted for covariates including age (<75 years or ≥75 years), sex, BMI (<25 kg/m² or ≥25 kg/m²), history of CVD, hypertension, diabetes, eGFR (<30 mL/min per 1.73 m² or ≥30 mL/min per 1.73 m²), and UPCR (<0.5 g/gCr or ≥0.5 g/gCr). P values <0.05 were considered statistically significant. BMI indicates body mass index; CKD, chronic kidney disease; CVD, cardiovascular disease; eGFR, estimated glomerular filtration rate; HR, hazard ratio; and UPCR, urinary protein‐to‐creatinine ratio.

Overall, these findings show that the M3V2 risk score delivers consistent and clinically relevant prognostic information across various CKD subgroups, regardless of differences in age, comorbidities, or disease causes. The model’s capacity to further stratify risk within existing staging systems (CKM and KDIGO), along with its link to potentially modifiable lifestyle factors, highlights its potential usefulness for personalized management of CKD.

DISCUSSION

In this study, we identified miRNA transcriptomic signatures present in human peripheral cEVs in CKD, highlighting the depletion of 23 biologically important miRNAs. We developed a novel, minimally invasive CKD prediction model using cEV‐encapsulated miRNAs combined with the Lasso‐Cox model in the derivation cohort to predict longitudinal kidney outcomes. The risk score equation, based on 3 key miRNAs, was then validated and optimized in a larger external cohort. The model effectively predicted kidney outcomes as well as composite outcomes, including all‐cause mortality, KRT initiation, and MACEs, independently of eGFR, UPCR, and CKD causes. Depletion of each miRNA was linked to higher rates of overt CVD and increased use of antihypertensives, loop diuretics, urate‐lowering agents, and ESAs. These results indicate that the risk prediction model is ready for application as a reliable biomarker for kidney and cardiovascular events. Additionally, this subset of miRNAs offers biological insights into interorgan communication, especially between the kidneys and cardiovascular system, aiding in understanding CKM syndrome.

We have previously demonstrated that cEVs have greater potential than urinary EVs as carriers of biological molecules and as biomarkers for extrarenal organ dysfunctions. In earlier work, we profiled miRNAs in CKD rodent models and found that cEVs lacked 4 key miRNAs known to reduce vascular calcification by repressing phenotypic switching of vascular smooth muscle cells (VSMCs). ¹⁹ In this study’s omics analysis, we further identified that cEVs from patients with severe CKD were deficient in a total of 23 miRNAs that regulate biological signaling molecules involved in cellular senescence, advanced glycation end‐products‐RAGE signaling, and FOXO signaling pathways. Although many previous studies have focused on analyzing urinary EVs due to their potential as noninvasive biomarkers, the majority of urinary EVs originate from the urinary tract. ⁴⁰ cEVs rarely pass through a healthy glomerulus into the urinary tract. Svenningsen et al. showed that 5090 out of 5113 (99.96%) proteins detected in urinary EVs are expressed in the urinary tract epithelium, including kidneys and bladder. ⁴⁰ Studies conducting transcriptomic analysis of human cEVs remain limited. The progressive failure of kidney and cardiovascular functions is increasingly understood as a coordinated systemic disease, reflected in syndromes such as cardiorenal syndromes, cardiorenal anemia, and the emerging CKM syndrome. We therefore focused on the potential of cEV‐encapsulated molecules, rather than urinary EVs, as key mediators of interorgan communication in CKD and as predictors of kidney and cardiovascular outcomes.

We found that the baseline expression level of let‐7d‐5p encapsulated in cEVs was especially predictive of CKD progression. Several studies have reported the protective effects of this miRNA on kidney tubular epithelial cells and VSMCs. ⁴¹ , ⁴² Parietal epithelial cell‐derived EVs enriched with let‐7d‐5p attenuate renal fibrosis through transcriptional repression of TGFβR1 (transforming growth factor beta receptor 1) and ARID3a (AT‐rich interaction domain 3A) in tubular epithelial cells. ⁴¹ Vartak et al. ⁴² demonstrated that overexpression of let‐7d‐5p in VSMCs broadly inhibits TNF‐α (tumor necrosis factor‐alpha), IL‐1β (interleukin‐1beta), IFN‐γ (interferon‐gamma), and NF‐κB (nuclear factor kappa B) activity. Conversely, reduced let‐7d‐5p expression leads to dysfunction of smooth muscle cells, and notably, statins have been shown to upregulate let‐7d‐5p expression in VSMCs. Wang et al. ⁴³ reported that carotid plaques from patients with diabetes had lower let‐7d‐5p levels. In VSMCs under high glucose conditions, let‐7d‐5p expression was decreased, and this miRNA inhibited VSMC proliferation and migration by suppressing HMGA2 (high‐mobility group AT‐hook 2) mRNA. They also showed that the GLP‐1 (glucagon‐like peptide‐1) receptor agonist liraglutide increased let‐7d‐5p expression and prevented VSMC migration and proliferation. In another study related to hepatic steatosis, let‐7d‐5p expression was increased by liraglutide treatment in steatotic HepG2 cells. ⁴⁴

The cardioprotective effects of miR‐24‐3p on cardiomyocytes have been extensively documented. It has been shown that IL‐1β promotes the proliferation of hypoxic vascular endothelial cells through the miR‐24‐3p/NKAP/NF‐κB axis. In blood samples from patients with acute myocardial infarction, miR‐24‐3p expression was decreased, whereas IL‐1β and NKAP expression were increased, demonstrating a negative correlation between miR‐24‐3p and IL‐1β or NKAP. ⁴⁵ miR‐24‐3p derived from M2 macrophage EVs was reported to provide cardioprotection against myocardial injury following sepsis by downregulating Tnfsf10 expression. ⁴⁶ In the setting of ischemia/reperfusion injury, miR‐24‐3p has been found to reduce apoptosis and protect cardiomyocytes by modulating the Keap1‐Nrf2 pathway, ⁴⁷ as well as by suppressing RIPK1 (receptor‐interacting serine/threonine‐protein kinase 1) expression, ⁴⁸ which is known to worsen myocardial ischemia/reperfusion injury via the TNF signaling pathway. In ischemic limb muscles, inhibition of miR‐24‐3p increased the formation of dysfunctional microvessels and impaired perfusion, emphasizing its role in regulating postischemic microvascular responses by targeting Notch and other vascular morphogens. ⁴⁹ Another study ⁵⁰ suggests that miR‐24‐3p may reduce inflammatory response and cell apoptosis in hepatic ischemia/reperfusion injury by targeting STING (stimulator of interferon genes), which plays a key role in promoting kidney inflammation and fibrosis. ⁵¹

In addition to miR‐24‐3p, miR‐126‐3p has also been involved in vascular protection. The intercellular transfer of miR‐126‐3p via endothelial microparticles has been shown to reduce VSMC proliferation and limit neointima formation by inhibiting LRP6 (low‐density lipoprotein receptor‐related protein 6). ⁵² Furthermore, 17β‐estradiol increases vascular endothelial Ets‐1/miR‐126‐3p expression, which may help reduce atherosclerosis. ⁵³ These findings together highlight the diverse roles of miR‐24‐3p and miR‐126‐3p in cardiovascular and vascular diseases. As mentioned, multiple studies have reported functions of miRNAs contained within endothelium‐derived EVs. Our group previously demonstrated that 4 miRNAs depleted in cEVs from CKD rodents were also decreased in human umbilical vein endothelial cells exposed to CKD serum. ¹⁹ Therefore, we speculate that kidneys and endothelial cells are the main sources of pathogenic cEVs lacking subsets of functional miRNAs in CKD. The specific origins of each miRNA from kidney segments and endothelial cells need to be examined in further animal and human studies, which will deepen the understanding of interorgan pathogenic crosstalk in CKD.

When we analyzed the prognostic impact of the established risk score equation on kidney and composite outcomes, the survival curves for patients with risk scores below or above the median closely resembled those for CKM stages 2 and 4 within the stage 2 to 4 population in this study. Epidemiological research on the prevalence of CKM syndrome is increasing, and although the concept has started to gain attention in nephrology, most publications remain review articles. The biological significance of CKM syndrome is still not fully understood, and its prospective impact on prognosis, especially regarding renal and CVDs, has yet to be clarified. This study is the first to show the effect of CKM syndrome on kidney and cardiovascular events in Japanese patients with CKD not on KRT, as well as the potential biological connection between CKM syndrome and changes in miRNA expression within cEVs, providing both clinical and biological insights. Importantly, the ability to further stratify patients within the same CKM stage based on the high‐ or low‐risk scores found here is a notable finding. It suggests that, rather than relying only on traditional classifications of kidney and CVD based on overt symptoms, high‐risk groups can be identified at a subclinical level through biological mechanisms, allowing for earlier detection and intervention before disease becomes clinically apparent.

Cystatin C‐based eGFR better reflects kidney function than creatinine‐based eGFR in the validation cohort, which included a high proportion of older patients and patients with sarcopenia. However, in constructing the optimized M3V2 risk equation, the inclusion of cystatin C—rather than serum creatinine or eGFR—as a conventional risk factor represents a novel aspect of our model. Underlying biological links between sarcopenia and miRNA‐targeted signaling molecules may have influenced variable selection during the Lasso‐Cox optimization process. Nevertheless, the predictive ability of the optimized M3V2 equation was independent of eGFR subgroups (eGFR <30 or ≥ 30), UPCR subgroups (<0.5 or ≥ 0.5), and CKD origins. We also compared the risk prediction performance of the established model with the KDIGO CKD classification, which incorporates UPCR, demonstrating the superiority of our risk prediction model. Furthermore, the risk score could further stratify patients into high‐ or low‐risk groups within each KDIGO stage. The unbiased omics and Lasso‐Cox analyses identified the 3 miRNAs included in the equation. However, we speculate that 23 miRNAs depleted in cEVs work together to regulate interorgan communication in CKD and CKM syndrome.

Interestingly, the risk score demonstrated strong stratification ability, particularly among individuals with no documented history of hypertension, CVD, or diabetes. This might reflect the model’s capacity to detect subclinical or preclinical disease processes that conventional clinical parameters cannot capture. In contrast, patients with overt comorbidities are already at high risk for kidney events, and the incremental value of the risk score may be less pronounced in this group. Additionally, the model demonstrated enhanced discriminative power in individuals with lower eGFR values (<30), which is biologically plausible given that reduced nephron function may amplify the molecular signals captured by cEV miRNA profiling. These findings suggest that the risk score might be particularly useful for identifying high‐risk individuals in clinically ambiguous or early‐stage populations, complementing conventional clinical parameters such as the eGFR and UPCR.

The new EV‐based risk prediction model may advance conventional clinical practice to a higher level, potentially enabling personalized medicine. Future work should validate this risk score equation in larger, multiethnic cohorts and explore the integration of cEV profiling into clinical workflows for personalized risk stratification.

Limitations

First, this study was conducted prospectively at a single center and involved participants of a single race or ethnicity. Although the prediction equation was effective in both the derivation and validation cohorts for predicting CKD progression and cardiovascular event mortality over time, its generalizability needs to be tested in other cohorts, especially larger and ethnically diverse populations. As a specific example, some subgroups in the stratified analyses included a limited number of patients, which may render the results less reliable and generalizable. To obtain more robust and generalizable findings, further validation using larger and independent cohorts will be necessary. Second, hypertensive glomerulosclerosis was the primary cause of CKD in both cohorts, and the validation cohort included only patients with eGFR <60 mL/min per 1.73 m². Therefore, the applicability of our findings to the population without CKD cannot be fully established. Third, not all miRNAs were thoroughly investigated. In the initial comprehensive screening of the derivation cohort, 519 miRNAs were reliably detected in the array‐based transcriptomic analysis out of a total of 2578 miRNAs. From a heatmap based on that subset, 23 miRNAs were further selected. Additional experimental methods to screen other coding and noncoding RNAs not covered in this study may be beneficial to better understand the biological interactions between kidneys and the cardiovascular systems. Fourth, the Schoenfeld test indicated a modest violation of the proportional hazards assumption for cystatin C. Although the deviation was limited, caution is warranted when interpreting results involving this variable. Finally, the miRNA transcriptomic analysis and the development of the risk score equation were based on a single measurement of cEV‐encapsulated miRNAs. Future studies incorporating averaged repeated measurements may improve the predictive accuracy of the equation.

CONCLUSIONS

This study identified a specific miRNA signature within cEVs that is strongly associated with the risk of CKD progression and cardiovascular events. This study’s minimally invasive risk score demonstrated robust predictive ability, offering potential utility for early risk stratification in patients with CKD. Furthermore, the identified miRNAs provide valuable insights into the molecular pathways underlying CKD, highlighting their potential role in the multisystem pathogenesis of hypertension, cardiovascular disease, CKD, and CKM syndrome. These findings suggest that cEV‐derived miRNAs not only serve as biomarkers but may also contribute to the pathophysiology of organ crosstalk in CKD, paving the way for novel therapeutic strategies. Future studies in larger and more diverse populations are essential to validate these findings, confirm their generalizability, and support their clinical application.

Sources of Funding

This work was supported by JSPS KAKENHI Grants‐in‐Aid for Young Scientists (20K16514 to Shintaro Mandai), KAKENHI Scientific Research (B) (22H02966 to Shintaro Mandai), a Grant‐in‐Aid for Challenging Exploratory Research (24K22093 to Shintaro Mandai), and Scientific Research (B) (23K24346 and 22H03085 to Shinichi Uchida) from the Japan Society for the Promotion of Science (JSPS), a Health and Labour Sciences Research Grant from the Ministry of Health, Labour and Welfare, JST Strategic Basic Research Programs – ACT‐X Grant (JPMJAX191 to Shintaro Mandai) from the Japan Science and Technology Agency (JST), JST FOREST (Fusion Oriented REsearch for disruptive Science and Technology) Program (grant number JPMJFR225O), Japan Agency for Medical Research and Development (AMED) (JP24ek0310025 to Shintaro Mandai), a grant from the Japanese Association of Dialysis Physicians (to Shintaro Mandai), the Pharmacodynamics Research Foundation (to Shintaro Mandai), Takeda Science Foundation (to Shintaro Mandai), the Young Innovative Medical Scientist Unit at Tokyo Medical and Dental University (to Shintaro Mandai), and JST SPRING (Support for Pioneering Research Initiated by the Next Generation) (JPMJSP2180 to Shunsuke Inaba).

Disclosures

None.

Supporting information

Tables S1–S3

Figures S1–S10