Introduction

Globally, the World Health Organization estimates that more than a billion people have hypertension, a risk factor for adverse cardiovascular and cerebrovascular events, and kidney disease [1]. Hypertension is a major cause of premature death worldwide, with a disproportionate burden in low- and middle-income countries, largely due to a rising prevalence of risk factors in recent decades [1].

As an upper middle-income country, South Africa has a high prevalence of hypertension, and hypertensive kidney disease is the most common cause of kidney failure documented by the South African Renal Registry [2]. Hypertension and kidney disease are inextricably linked, with common pathophysiological denominators that begin with foetal programming in-utero [3, 4]. There is strong epidemiological evidence linking low birth weight, as a marker of adverse intrauterine circumstances, to adult hypertension and kidney disease [4, 5]. Potential mechanisms include a congenital deficit in nephron number which may arise from changes in DNA methylation, accelerated apoptosis in the developing kidney, changes in renal renin–angiotensin system activity, and an increase in foetal glucocorticoid exposure [4]. A decrease in nephron number is associated with compensatory glomerular hypertrophy and an increased susceptibility to the progression of kidney disease [4].

Angiotensin, an endocrine hormone peptide, is a vital part of the renin-angiotensin-aldosterone system, an inter-related endocrine system that plays a significant role in volume and blood pressure control [6]. In response to a drop in blood pressure, or sympathetic nerve activity in the kidney, renin is released and cleaves off two amino acids enzymatically to form angiotensin I (ATI), which is cleaved by the angiotensin converting enzyme (ACE) to form angiotensin II (ATII) [6]. ATII is the main effector molecule of this system, increasing blood pressure, enhancing renal tubular reabsorption of sodium and water, and stimulating aldosterone release from the adrenal gland [7]. In addition to being an effective vasoconstrictor, ATII has also been shown to activate proliferative, pro-inflammatory and pro-fibrotic pathways and stimulates the production of ET-1 resulting in an increase in oxidative stress [7,8,9]. These combined effects of ATII contribute to the development of kidney disease [10,11,12].

With sustained elevations of blood pressure, the afferent arterioles in the kidney undergo structural changes and hypertrophy, and intraglomerular pressure results in glomerular hypertrophy, hyperfiltration, and albuminuria [13,14,15]. Sustained glomerular hyperfiltration can result in glomerular scarring and irreversible kidney injury [13,14,15,16,17].

Kidney disease is diagnosed using estimated glomerular filtration rate (eGFR) and albuminuria, as each independently confer an increased risk of cardiovascular and all-cause mortality [18]. Chronic kidney disease (CKD) is defined as eGFR <60 mL/min/1.73 m2; and/ or albuminuria, defined as an albumin:creatinine ratio ≥3.0 mg/mmol [18]. Albuminuria may precede the onset of reduced eGFR, which makes it appealing as a potential for early detection and treatment strategies for kidney disease, but this is not a consistent, replicable finding [19, 20]. The search for more sensitive markers of early kidney disease has extended to exploration of urinary proteomics with significant progress in the proteomic analysis of biological fluids, in this case, urine.

The past decade has seen omics studies contributing to the diagnosis, therapeutic intervention and prognosis of kidney disease [21,22,23]. Samples derived from urine are non-invasive and easy to collect compared to other body fluids, and can be used to identify genomic, metabolomic, transcriptomic, and proteomic biomarkers that are strongly associated with pathophysiologic mechanisms of disease [24]. Proteomic strategies such as mass spectrometry combined with liquid chromatography and capillary electrophoresis are used to identify urine biomarkers involved in early detection of kidney disease and potential disease pathways that provide insight into the pathogenic mechanisms of disease [25,26,27]. One such example is a classifier based on 273 urinary peptides (CKD273) which reliably allows for early detection of kidney disease and is more sensitive than albuminuria in predicting a decline in eGFR [27].

To date, there is one study from South Africa that has investigated the urinary proteome in young adults with hypertension [28]. This study found that combining 20 peptides into a single classifier resulted in the separation of normotensive and hypertensive groups with an area under the curve of 0.85 (P < 0.001) [28]. There are no published studies from South Africa investigating the urinary proteome in the setting of hypertension and albuminuria – the latter being an early indicator of kidney injury. In this exploratory study, we hypothesise that urinary proteomic analysis will identify urinary markers of kidney disease in hypertensive individuals with albuminuria and well preserved eGFR.

Methods

Study participants

This case-control pilot study is a sub-study of the African Research Kidney Disease (ARK) study, a well characterised population-based cohort study of 2021 adults (20–80 years) of self-identified black ethnicity from Agincourt in the rural Mpumalanga Province, South Africa [29]. Demographic, health, and family history information were obtained for all participants, with collection of urine and blood samples. Participants did not disclose food or beverage consumption prior to urine collection. From the cohort, ninety participants with stored urine samples were selected for proteomic analysis (Fig. 1).

Fig. 1
figure 1

Flow diagram for this case-control study. ARK, African Research Kidney Disease; DBP, diastolic blood pressure; DM, diabetes mellitus; HIV; human immunodeficiency virus; POC, point of care; SBP, systolic blood pressure; UACR, urine albumin-to-creatinine ratio

Full size image

The inclusion criteria of this sub-study were participants (cases) with albuminuria (urine albumin-to-creatinine ratio [UACR] ≥3 mg/mmol) and hypertension (systolic blood pressure [SBP] ≥140 mm Hg and/or diastolic blood pressure [DBP] ≥90 mm Hg, JNC7 criteria). The control group included participants with UACR <3 mg/mmol, SBP <140 mm Hg, and DBP <90 mm Hg. Individuals with other potential causes or consequences of kidney disease, including diabetes mellitus and HIV infection were excluded as cases and controls. Diet was not considered in this study. The control group was matched by age and sex.

Urine collection and measurement of eGFR and UACR

Approximately 20–30 ml urine was collected from ARK study participants. Urine samples were processed and stored at −80 °C prior to shipping to the Council for Scientific and Industrial Research laboratory for proteomic testing [30]. Serum and urine creatinine was measured using an isotope dilution mass spectrometry traceable Jaffe method [30]. The 2009 Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) creatinine equation without adjustment for African American ethnicity was used to calculate eGFR (ml/min/1.73 m2) [31]. Immunoturbidimetry was used to measure urine albumin concentration [30]. The UACR was calculated (mg/mmol) based on this measurement [30].

Urinary protein extraction by precipitation

An in-house urinary proteome sample preparation method was used. For each urine sample, protein precipitation was achieved by adding 1,600 µL ice-cold acetone to 400 µL crude urine, stored at –32 °C for 1 h. The samples were centrifuged at 12,000 g at 4 °C for 1 h. Thereafter, the supernatant was discarded and the pellet was allowed to air dry. The precipitated protein pellet was resuspended in 100 µL 2% sodium dodecyl sulfate solution and sonicated for 5 min (Elmi, Riga, Latvia). Dithiothreitol was added to the solution (final 10 mM), and placed on a 70 °C heating block for 15 min, and then on a 40 °C heating block for 15 min. The tubes were cooled to room temperature (RT) and iodoacetamide was added for alkylation (final 30 mM) and kept in the dark at RT for 30 min. The protein sample was mixed with an equal volume HILIC binding buffer (30% acetonitrile [MeCN]/200mM Ammonium acetate [NH4Ac]) and kept at RT before it was added to a KingFisher™ deep well plate.

Automated on-bead digestion using MagReSyn® HILIC

Protein samples were digested on-bead using multimode magnetic microparticles (MagReSyn® HILIC, ReSyn Biosciences) in a KingFisher Duo™ system (Thermo Fisher Scientific), as previously described [32, 33], with minor modifications. Briefly, magnetic hydrophilic affinity microparticles (20 µl, 200 µg) were equilibrated in 200 µl 100 mM NH4Ac pH 4.5, 15% MeCN. The microparticles were transferred to a well containing the protein-binding buffer solution and mixed for 30 min at RT. The captured proteins were washed twice in 200 µl 95% MeCN and transferred to 200 µl 25 mM ammonium bicarbonate containing 1 µg sequencing grade modified trypsin (Promega, Madison, USA) and mixed for 4 h at 37 °C. Finally, beads were washed in 1% trifluoracetic acid to elute any remaining bound peptides. The resulting peptides (pool of digest and eluate) were vacuum dried, resuspended in 2% MeCN, 0.2% FA and quantified using the Pierce™ Peptide Quantification (Thermo Fisher Scientific) assay as per the manufacturer’s instructions.

A project specific system suitability-quality control (PQC) sample was prepared by pooling an equal volume of 16 urinary peptide samples. Additionally, a complex proteome digest was used as a general system suitability assessment. These PQC samples were injected at least once with each batch and analysed and processed together with the study samples.

Low pH reverse phase liquid chromatography with mass spectrometry (LCMS/MS) data acquisition

Individual participant peptide samples (500 ng, single shot) were analysed using a Dionex UltiMate™ 3000 UHPLC in nanoflow configuration. Samples were inline desalted on an Acclaim PepMap C18 trap column (75 μm × 2 cm; 2 min at 5 µl/min using 2% MeCN/0.2% FA). Trapped peptides were gradient eluted and separated on a nanoEase M/Z Peptide CSH C18 Column (130 Å, 1.7 μm, 75 μm X 250 mm) (Waters) at a flow-rate of 300 nl/min with a gradient of 6–35% over 30 min (A: 0.1% FA; B: 80% MeCN/0.1% FA).

Data was acquired using DIA - or Sequential Window Acquisition of all Theoretical Mass Spectra (SWATH) using a TripleTOF® 6600 mass spectrometer (SCIEX, Massachusetts, USA) [34]. Eluted peptides were delivered into the mass spectrometer via a Nanospray® III ion source equipped with a 20 μm Sharp Singularity emitter (Fossil Ion Technology, Madrid, Spain). Source settings were: Curtain gas − 20, Gas 1 –16, Gas 2 − 0, temperature – 0 (off) and ion spray voltage – 2,600 V.

Data was acquired using 64 MS/MS scans of overlapping sequential precursor isolation windows (variable m/z isolation width, 1 m/z overlap, high sensitivity mode), with a precursor MS scan for each cycle. The accumulation time was 100 ms for the MS1 scan (from 400 to 900 m/z) and 15 ms for each product ion scan (100 to 1,800 m/z) for a 1.06 sec cycle time.

DIA (SWATH) library generation and data extraction

A spectral library was built (from all patient DIA files) in Spectronaut™ 15 software using default settings with minor adjustments as follows: segmented regression was used to determine the normalised retention time (iRT) in each run; iRTs were calculated as median for all runs; the digestion rule was set as “Trypsin” and modified peptides were allowed; fragment ions between 300 and 1,800 m/z and ions with greater than 3 amino acids were considered; peptides with a minimum 3 and maximum 6 (most intense) fragment ions were accepted. This study specific spectral library was concatenated with an in-house generated urinary proteome spectral library (using Spectronaut™ 15’s “Search Archives” feature).

Raw SWATH (.wiff) data files were converted into Spectronaut™ HTRMS format and analysed using Spectronaut™ 15. The default settings that were used for targeted analysis were: dynamic iRT prediction with correction factor for window 1; mass calibration was set to local; decoy method set as scrambled; the FDR, based on the mProphet approach [35], set at 1% on the precursor peptide and protein group levels; protein inference set to “default” which is based on the ID picker algorithm [36] and global cross-run normalisation on median. The concatenated urinary proteome spectral library (peptides – 20,616, protein groups – 2,604) was used as a reference for targeted data extraction.

Following acquisition, data was curated and filtered for sample preparation and/or LCMS related failures.

Statistical and functional data annotation

For the table of descriptive data, t-tests and a Chi-squared analysis using STATA were used to compute associations between the case and control groups, as appropriate.

Protein abundances were imported into SRplot and log2 transformed. Principal component analysis (PCA) based on protein abundances was performed to assess stratification within the cohort. A non-parametric t-test was performed to assess the number of proteins that differed between the case and control groups. Statistically significant differentially abundant proteins between the cases and controls were calculated by a two-sided t-test, with a cut-off minimum of 2-fold difference and P values adjusted for multiple testing by FDR at 1%.

The differentially abundant proteins were constructed into networks and annotated using Cytoscape and STRING functional enrichment [37]. Enrichment was considered statistically significant when corrected for multiple testing by FDR with adjusted q-values <0.05. Additional networks were imported on Cytoscape using public databases for hypertension and CKD with a high confidence score of 0.8 (i.e., 80% confidence) and maximum 200 proteins.

Machine learning

OmicLearn (v1.4) was used for data analysis, model execution, and creation of plots and charts [38]. Machine learning was performed in Python (3.10.12). Spectronaut feature tables were imported via the Pandas package (1.5.3) and manipulated using the Numpy package (1.24.2). The machine learning pipeline was employed using the scikit-learn package (1.2.2). The Plotly (5.9.0) library was used to generate plots and charts. No normalisation on the data was performed. To impute missing values, a Zero-imputation strategy was used. Features were selected using a ExtraTrees (n_trees = 100) strategy with a maximum number of 20 features. During training, normalisation and feature selection was individually performed using the data of each split. A repeated (RepeatedStratifiedKFold, n_repeats = 10), stratified cross-validation (n_splits = 5) approach to classify the groups was used. This resulted in total 50 iterations for training the model, each time with 20 features. The XGBoost-Classifier (version: 1.7.4, random_state = 23, learning_rate = 0.3, min_split_loss = 0, max_depth = 6, min_child_weight = 1) was used for classification.

Results

Of the initial 90 participants selected for proteomic analyses, 9 were removed due to poor peptide/protein recoveries and 8 were removed due to poor liquid chromatography separation possibly due to incomplete removal of sample specific contamination during sample preparation. Baseline characteristics of the overall ARK study and the 24 case and 49 control participants whose data passed quality control are displayed in Table 1.

Table 1 Baseline characteristics of participants in the ARK study and this case-control study
Full size table

The age and sex distribution of participants was similar in the case and control groups. The case group had a higher mean body mass index compared to controls, and as expected, blood pressure and UACR differed between the groups. In addition, apolipoprotein L1 (APOL1) allele distribution was similar between cases and controls.

Project specific system suitability-quality control

The study specific and commercial system suitability quality controls are shown in (Supplementary Fig. 1A–F). The protein co-efficient of variation (CV) was 11.6% for the HeLa digest system suitability assessment and 18.0% for the urine peptide pool over the seven days of analysis (Supplementary Fig. 1A and B). Protein, peptide and precursor counts were stable throughout the data acquisition process (Supplementary Fig. 1C and D). The correlation plots indicate the system was stable with minimal drift over the course of data acquisition (Supplementary Fig. 1E and F).

Identification of proteins by LCMS analysis

On average, 915 (± 264) proteins were quantified in the case group and 1,225 (± 227) proteins were quantified in the control group (Fig. 2, Supplementary Fig. 2). The number of quantified proteins significantly differed between the case and control groups (P < 0.001).

Fig. 2
figure 2

Summary of quantified proteins. The number of proteins quantified post SWATH LCMS analysis averaged across individual samples and grouped per condition. LCMS, liquid chromatography with mass spectrometry. ****Denotes P < 0.001 performed using a non-parametric t-test

Full size image

Principal Component (PC) analysis

PC analysis using protein abundances highlighted differential clustering of the two groups with some overlap (Fig. 3).

Fig. 3
figure 3

PC (1 and 2) analysis based on protein abundance features. PC analysis using log-transformed protein abundance values indicates clustering of cases and controls. Red and blue circles represent cases and controls, respectively. PC, principal component

Full size image

Identification of proteins with significantly different abundances between cases and controls

Differentially abundant proteins between the cases and controls are shown in Fig. 4. Majority of the proteins deemed statistically different in abundances between cases and controls, were found to have lower abundance in the cases. The full list of differential protein abundances is shown in Supplementary Table 1.

Machine learning model

The performance of the machine learning algorithm was evaluated by receiver operating characteristics analysis with an area under the curve of 0.98 and confusion matrix analysis where predicted and actual classification closely matched (Supplementary Fig. 3). The top 20 features identified from the classifier are shown in Figs. 4 and 5. When comparing the top 20 features to the list of 332 differentially abundant proteins identified (using a student t-test), 16 were found to overlap and four were unique to the XgBoost algorithm (Supplementary Table 2).

Fig. 4
figure 4

Differential protein abundances between cases and controls illustrated in a volcano plot

Full size image

Proteins were graphed by fold change (difference) on the x-axis and significance (–log10P) on the y-axis using an FDR of 0.01 and a fold-change of 2. Statistical analysis was performed in Perseus. Orange dots indicate proteins with higher abundances in the cases. Blue dots indicate proteins with lower abundances in the cases. The top 20 features of the machine learning algorithm are annotated in red.

FDR, false discovery rate.

Fig. 5
figure 5

Top 20 features identified from the classifier using the XGBoost algorithm. Gene names for each protein are shown

Full size image

Functional annotation of proteins with significantly different abundances between cases and controls

Three hundred and thirty-two proteins with significantly different abundances between cases and controls were constructed into a network and annotated using Cytoscape. To gain more insight on the biological significance of these proteins, STRING enrichment was performed. The network was split into sub clusters based on relevant pathways related to CKD. We identified 112 proteins associated with the immune system (q-value [FDR] = 1.4 × 10− 45), and 89 proteins associated with the innate immune system (q = 1.1 × 10− 32) (Supplementary Figs. 4 and 5). Additionally, 17 and six proteins were associated with extracellular matrix (ECM) organisation (q = 0.03) and activation of matrix metalloproteinases (q = 0.04), respectively (Supplementary Fig. 6A and B).

Using Cytoscape public databases, 14 common proteins were identified between this pilot case-control study and other studies on CKD and hypertension, as imported from public databases (Fig. 6). The size of the circle represents the disease score, where a score of 0 indicates that the protein is not related to the disease and a score of 5 indicates that the protein is related to the disease. Proteins angiotensinogen (AGT), albumin (ALB), APOL1, and uromodulin (UMOD) were associated with CKD and hypertension with high disease scores of 5.0 (100% confidence), 4.0 (80%), 3.9 (78%), and 3.8 (76%), respectively. The proteins matrix metalloproteinase 9 (MMP9) and TIMP Metallopeptidase Inhibitor 1 (TIMP1) were associated with pathways involved in degradation of the ECM, interleukin (IL)-4 and IL-13 signalling, and activation of matrix metalloproteinases.

Fig. 6
figure 6

Common proteins between this case-control pilot study and other studies on CKD and hypertension, as imported from public databases

Full size image

The gene names for each protein in the network are shown. Red indicates proteins with higher abundance in cases while blue represents proteins with lower abundance.

Size of the circle represents the disease score (0 [not related to disease] – 5 [related to disease]). The closer to 5, the higher the confidence that the protein is associated with CKD and hypertension. AGT = 5.0; ALB = 4.0; APOL1 = 3.9; UMOD = 3.8; APOB = 3.2; APOA1 = 3.1; MMP9 = 2.9; FN1 = 2.9; PECAM1 = 2.7; LCN2 = 2.7; MPO = 2.4; PLG = 2.2; TIMP1 = 2.2; SLC12A1 = 2.2.

Edges represent protein-to-protein interactions that were selected by default in Cytoscape. Pathways are annotated on the network using a split donut ring. Orange indicates proteins associated with degradation of the extracellular matrix; pink indicates proteins associated with IL-4 and IL-13 signalling; green indicates proteins associated with activation of matrix metalloproteinases.

CKD, chronic kidney disease; IL, interleukin.

Discussion

This case-control pilot study aimed to identify potential proteins and pathways involved in hypertension-associated albuminuria by assessing urinary proteomic profiles in black South African participants with both hypertension and albuminuria compared to those who have neither condition.

Peptide and protein CVs showed that the methods were reproducible and that the system and workflow contributed low technical variability. The instrumental drift observed throughout the data acquisition period was minimal, as shown in the commercial HeLa digest system suitability data that were acquired from a fresh sample on each day. A larger drift was observed in the study-specific peptide pool system suitability data; however, this can be attributed to autosampler-associated degradation as the sample was kept in the autosampler throughout the data acquisition process. This suggests that most of the variation observed in the study can be attributed to biological variation. On average, more proteins (1,225) were quantified in the control group than in the case group (915). Possible explanations include that a minor difference in the missed cleavage rate (∼3%) was observed between the two groups and several high abundance proteins, such as ALB, APOL1, apolipoprotein A1 (APOA1), and apolipoprotein B (APOB) are significantly higher in abundance in the cases ‘masking’ identification of lower abundance proteins, resulting in fewer quantifications overall.

The majority of differentially abundant proteins were associated with pathways involving the immune system, innate immune system, extracellular matrix organisation and activation of matrix metalloproteinases. In the setting of sustained hypertension, high blood pressure can cause permanent damage to the nephrons through arterial injury and glomerular ischaemia [39]. Damage from renal ischaemia can lead to the production of various inflammatory cytokines that lead to immune cell infiltration and tubular atrophy [39]. The latter is supported by Mattson (2014), [40] who showed an infiltration of inflammatory mononuclear cells in the arterioles and small arteries in kidney tissue from patients with hypertension. Enrichment analysis has shown that pathways related to inflammation and immune response are associated with the progression of kidney disease [41].

We identified six proteins associated with the activation of matrix metalloproteinase (MMPs) pathways. Of these, five had a lower abundance in participants with hypertension-associated albuminuria. It has been suggested that decreased activity of matrix metalloproteases may cause an accumulation of proteins in the extracellular matrix leading to fibrosis, which is one of the features of hypertension-associated kidney disease [42,43,44]. Most of the proteins (71%; n = 12/17) associated with the ECM organisation pathway were also reduced in participants with hypertension-associated albuminuria. It has been suggested that the reduced abundance of ECM organisation proteins may reflect reduced ECM turnover (decreased degradation), subsequently causing an increased deposit in ECM, resulting in fibrosis [45]. Currently, kidney biopsy is the only validated approach to evaluate fibrosis. The identification of ECM organisation proteins may provide a non-invasive approach to detect early modifications in the ECM that lead to renal fibrosis.

Fourteen common proteins were identified between this pilot study and public databases on CKD and hypertension. Proteins including AGT, ALB, APOL1, and UMOD had the highest disease scores (76–100% confidence) for CKD and hypertension. The PLG protein has been shown to regulate macrophage migration in inflammation through activation of the MMP9 protein, which regulates the ability of the cells to migrate across the ECM [46]. Therefore, targeting the PLG/MMP9 pathways may lead to regulation of inflammatory responses in CKD.

In this pilot study, UMOD had a lower abundance in cases with hypertension-associated albuminuria. Similarly, Nqebelele et al. 2019 found reduced levels of UMOD in black African individuals with hypertensive-attributed CKD [47]. Navise et al. (2023) also found lower levels of UMOD in black African individuals with CKD [48]. In addition to its role in ECM remodelling and fibrosis, UMOD has been involved in regulating ion transport in the kidney [49]. Reduced levels of UMOD have been postulated to be a result of decreased secretion from damaged tubules [50]. Results from a mouse model with altered UMOD expression have shown that the induction of pro-inflammatory signalling is one of the first events that occur in the kidneys [41]. The animal study also showed that the initiation of inflammatory signals precedes fibrosis and kidney damage, and possibly plays a vital role in disease onset [41].

It is important to note that hypercholesterolaemia was present in 38% of the cases in this study. Several studies have linked apolipoproteins (APOA1, APOB and APOL1) with CKD [51,52,53] and hypertension [54]. Moreover, a Mendelian-randomization study identified a causal association between lipids (including low-density lipoprotein and triglycerides) and kidney function in individuals of African ancestry [55]. It has been postulated that the association between apolipoproteins and kidney disease may be mediated by the impact of these lipoprotein particles on the kidney. Apolipoproteins and their associated lipids may have toxic effects on glomerular cells leading to glomerulosclerosis, which is scarring of the glomerulus [56]. As a result, Kidney Disease: Improving Global Outcomes (KDIGO) developed guidelines for lipid management in patients with CKD [57].

Machine learning enabled us to train a classifier that on average correctly identified cases with an 88% true positive rate. Additionally, it identified controls with a 97% true negative rate. This analysis revealed that alpha-1-antitrypsin (P01009) and afamin (P43652) were the two most important features of the classifier. Studies have shown that alpha-1-antitrypsin and afamin are potential biomarkers for the diagnosis of early diabetic kidney disease (DKD) and can predict the decline in renal function [58, 59]. Interestingly, in this pilot study, we have identified DKD markers in normoglycemia participants who have hypertension and albuminuria, which suggests the generalisability of the findings.

This is the first study to characterise the urinary proteome in South African individuals with hypertension-associated albuminuria. This case-control study identified common proteins with previous studies on CKD and hypertension from other ethnicities, and proteins associated with DKD, indicating generalisability of the findings to other populations and diseases. This study contributes to the understanding of pathways involved in hypertension-associated albuminuria and suggest opportunities for translation into the clinical setting.

The results from this study must be interpreted in the context of its limitations, which included a relatively small sample size, consisting of only 73 individuals. All participants are from a single centre, therefore, results may not be generalisable. The machine learning models applied in the current study show preliminary data and will require validation in a larger dataset with blinded data to assess overfitting.

Further research is essential to validate these outcomes in larger cohorts. In addition, the investigation of other diseases such as DKD and acute kidney injury would help to further explore generalisability of findings.

Conclusions

In this pilot study, pathways associated with hypertension-associated albuminuria included the immune system, innate immune system, ECM organisation and activation of matrix metalloproteinases. These pathways contribute insights into the pathophysiology (e.g., immune cell infiltration, tubular atrophy, and ECM remodelling) of hypertension-related albuminuria. Additionally, proteins such as AGT, ALB, APOL1 and UMOD had the highest disease scores (76–100% confidence) for hypertension and CKD. The urinary proteomic data combined with the machine learning approach classified disease status and identified proteins and pathways linked to hypertension-associated albuminuria.