Research ArticleNephrology Open Access | 10.1172/jci.insight.186070

Fadhl Alakwaa,1 Vivek Das,2 Arindam Majumdar,3 Viji Nair,1 Damian Fermin,1 Asim B. Dey,3 Timothy Slidel,4 Dermot F. Reilly,5 Eugene Myshkin,5 Kevin L. Duffin,3 Yu Chen,3 Markus Bitzer,1 Subramaniam Pennathur,1 Frank C. Brosius,6 Matthias Kretzler,1 Wenjun Ju,1 Anil Karihaloo,7 and Sean Eddy1

1Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan, USA.

2Novo Nordisk A/S, Måløv, Denmark.

3Eli Lilly & Co., Indianapolis, Indiana, USA.

4Biopharmaceuticals R&D, AstraZeneca, Cambridge, United Kingdom.

5Johnson & Johnson, New Brunswick, New Jersey, USA.

6University of Arizona, Tucson, Arizona, USA.

7Novo Nordisk Research Center Seattle, Inc, Seattle, Washington, USA.

Address correspondence to: Sean Eddy, Associate Research Scientist, University of Michigan, Department of the Internal Medicine, Division of Nephrology, Ann Arbor, Michigan 48109, USA. Email: seaneddy@med.umich.edu.

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Find articles by Alakwaa, F. in: JCI | PubMed | Google Scholar |

1Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan, USA.

2Novo Nordisk A/S, Måløv, Denmark.

3Eli Lilly & Co., Indianapolis, Indiana, USA.

4Biopharmaceuticals R&D, AstraZeneca, Cambridge, United Kingdom.

5Johnson & Johnson, New Brunswick, New Jersey, USA.

6University of Arizona, Tucson, Arizona, USA.

7Novo Nordisk Research Center Seattle, Inc, Seattle, Washington, USA.

Address correspondence to: Sean Eddy, Associate Research Scientist, University of Michigan, Department of the Internal Medicine, Division of Nephrology, Ann Arbor, Michigan 48109, USA. Email: seaneddy@med.umich.edu.

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Find articles by Das, V. in: JCI | PubMed | Google Scholar

1Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan, USA.

2Novo Nordisk A/S, Måløv, Denmark.

3Eli Lilly & Co., Indianapolis, Indiana, USA.

4Biopharmaceuticals R&D, AstraZeneca, Cambridge, United Kingdom.

5Johnson & Johnson, New Brunswick, New Jersey, USA.

6University of Arizona, Tucson, Arizona, USA.

7Novo Nordisk Research Center Seattle, Inc, Seattle, Washington, USA.

Address correspondence to: Sean Eddy, Associate Research Scientist, University of Michigan, Department of the Internal Medicine, Division of Nephrology, Ann Arbor, Michigan 48109, USA. Email: seaneddy@med.umich.edu.

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Find articles by Majumdar, A. in: JCI | PubMed | Google Scholar

1Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan, USA.

2Novo Nordisk A/S, Måløv, Denmark.

3Eli Lilly & Co., Indianapolis, Indiana, USA.

4Biopharmaceuticals R&D, AstraZeneca, Cambridge, United Kingdom.

5Johnson & Johnson, New Brunswick, New Jersey, USA.

6University of Arizona, Tucson, Arizona, USA.

7Novo Nordisk Research Center Seattle, Inc, Seattle, Washington, USA.

Address correspondence to: Sean Eddy, Associate Research Scientist, University of Michigan, Department of the Internal Medicine, Division of Nephrology, Ann Arbor, Michigan 48109, USA. Email: seaneddy@med.umich.edu.

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Find articles by Nair, V. in: JCI | PubMed | Google Scholar

1Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan, USA.

2Novo Nordisk A/S, Måløv, Denmark.

3Eli Lilly & Co., Indianapolis, Indiana, USA.

4Biopharmaceuticals R&D, AstraZeneca, Cambridge, United Kingdom.

5Johnson & Johnson, New Brunswick, New Jersey, USA.

6University of Arizona, Tucson, Arizona, USA.

7Novo Nordisk Research Center Seattle, Inc, Seattle, Washington, USA.

Address correspondence to: Sean Eddy, Associate Research Scientist, University of Michigan, Department of the Internal Medicine, Division of Nephrology, Ann Arbor, Michigan 48109, USA. Email: seaneddy@med.umich.edu.

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Find articles by Fermin, D. in: JCI | PubMed | Google Scholar |

1Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan, USA.

2Novo Nordisk A/S, Måløv, Denmark.

3Eli Lilly & Co., Indianapolis, Indiana, USA.

4Biopharmaceuticals R&D, AstraZeneca, Cambridge, United Kingdom.

5Johnson & Johnson, New Brunswick, New Jersey, USA.

6University of Arizona, Tucson, Arizona, USA.

7Novo Nordisk Research Center Seattle, Inc, Seattle, Washington, USA.

Address correspondence to: Sean Eddy, Associate Research Scientist, University of Michigan, Department of the Internal Medicine, Division of Nephrology, Ann Arbor, Michigan 48109, USA. Email: seaneddy@med.umich.edu.

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Find articles by Dey, A. in: JCI | PubMed | Google Scholar

1Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan, USA.

2Novo Nordisk A/S, Måløv, Denmark.

3Eli Lilly & Co., Indianapolis, Indiana, USA.

4Biopharmaceuticals R&D, AstraZeneca, Cambridge, United Kingdom.

5Johnson & Johnson, New Brunswick, New Jersey, USA.

6University of Arizona, Tucson, Arizona, USA.

7Novo Nordisk Research Center Seattle, Inc, Seattle, Washington, USA.

Address correspondence to: Sean Eddy, Associate Research Scientist, University of Michigan, Department of the Internal Medicine, Division of Nephrology, Ann Arbor, Michigan 48109, USA. Email: seaneddy@med.umich.edu.

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Find articles by Slidel, T. in: JCI | PubMed | Google Scholar |

1Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan, USA.

2Novo Nordisk A/S, Måløv, Denmark.

3Eli Lilly & Co., Indianapolis, Indiana, USA.

4Biopharmaceuticals R&D, AstraZeneca, Cambridge, United Kingdom.

5Johnson & Johnson, New Brunswick, New Jersey, USA.

6University of Arizona, Tucson, Arizona, USA.

7Novo Nordisk Research Center Seattle, Inc, Seattle, Washington, USA.

Address correspondence to: Sean Eddy, Associate Research Scientist, University of Michigan, Department of the Internal Medicine, Division of Nephrology, Ann Arbor, Michigan 48109, USA. Email: seaneddy@med.umich.edu.

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Find articles by Reilly, D. in: JCI | PubMed | Google Scholar |

1Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan, USA.

2Novo Nordisk A/S, Måløv, Denmark.

3Eli Lilly & Co., Indianapolis, Indiana, USA.

4Biopharmaceuticals R&D, AstraZeneca, Cambridge, United Kingdom.

5Johnson & Johnson, New Brunswick, New Jersey, USA.

6University of Arizona, Tucson, Arizona, USA.

7Novo Nordisk Research Center Seattle, Inc, Seattle, Washington, USA.

Address correspondence to: Sean Eddy, Associate Research Scientist, University of Michigan, Department of the Internal Medicine, Division of Nephrology, Ann Arbor, Michigan 48109, USA. Email: seaneddy@med.umich.edu.

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Find articles by Myshkin, E. in: JCI | PubMed | Google Scholar

1Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan, USA.

2Novo Nordisk A/S, Måløv, Denmark.

3Eli Lilly & Co., Indianapolis, Indiana, USA.

4Biopharmaceuticals R&D, AstraZeneca, Cambridge, United Kingdom.

5Johnson & Johnson, New Brunswick, New Jersey, USA.

6University of Arizona, Tucson, Arizona, USA.

7Novo Nordisk Research Center Seattle, Inc, Seattle, Washington, USA.

Address correspondence to: Sean Eddy, Associate Research Scientist, University of Michigan, Department of the Internal Medicine, Division of Nephrology, Ann Arbor, Michigan 48109, USA. Email: seaneddy@med.umich.edu.

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Find articles by Duffin, K. in: JCI | PubMed | Google Scholar

1Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan, USA.

2Novo Nordisk A/S, Måløv, Denmark.

3Eli Lilly & Co., Indianapolis, Indiana, USA.

4Biopharmaceuticals R&D, AstraZeneca, Cambridge, United Kingdom.

5Johnson & Johnson, New Brunswick, New Jersey, USA.

6University of Arizona, Tucson, Arizona, USA.

7Novo Nordisk Research Center Seattle, Inc, Seattle, Washington, USA.

Address correspondence to: Sean Eddy, Associate Research Scientist, University of Michigan, Department of the Internal Medicine, Division of Nephrology, Ann Arbor, Michigan 48109, USA. Email: seaneddy@med.umich.edu.

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Find articles by Chen, Y. in: JCI | PubMed | Google Scholar

1Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan, USA.

2Novo Nordisk A/S, Måløv, Denmark.

3Eli Lilly & Co., Indianapolis, Indiana, USA.

4Biopharmaceuticals R&D, AstraZeneca, Cambridge, United Kingdom.

5Johnson & Johnson, New Brunswick, New Jersey, USA.

6University of Arizona, Tucson, Arizona, USA.

7Novo Nordisk Research Center Seattle, Inc, Seattle, Washington, USA.

Address correspondence to: Sean Eddy, Associate Research Scientist, University of Michigan, Department of the Internal Medicine, Division of Nephrology, Ann Arbor, Michigan 48109, USA. Email: seaneddy@med.umich.edu.

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Find articles by Bitzer, M. in: JCI | PubMed | Google Scholar |

1Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan, USA.

2Novo Nordisk A/S, Måløv, Denmark.

3Eli Lilly & Co., Indianapolis, Indiana, USA.

4Biopharmaceuticals R&D, AstraZeneca, Cambridge, United Kingdom.

5Johnson & Johnson, New Brunswick, New Jersey, USA.

6University of Arizona, Tucson, Arizona, USA.

7Novo Nordisk Research Center Seattle, Inc, Seattle, Washington, USA.

Address correspondence to: Sean Eddy, Associate Research Scientist, University of Michigan, Department of the Internal Medicine, Division of Nephrology, Ann Arbor, Michigan 48109, USA. Email: seaneddy@med.umich.edu.

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Find articles by Pennathur, S. in: JCI | PubMed | Google Scholar

1Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan, USA.

2Novo Nordisk A/S, Måløv, Denmark.

3Eli Lilly & Co., Indianapolis, Indiana, USA.

4Biopharmaceuticals R&D, AstraZeneca, Cambridge, United Kingdom.

5Johnson & Johnson, New Brunswick, New Jersey, USA.

6University of Arizona, Tucson, Arizona, USA.

7Novo Nordisk Research Center Seattle, Inc, Seattle, Washington, USA.

Address correspondence to: Sean Eddy, Associate Research Scientist, University of Michigan, Department of the Internal Medicine, Division of Nephrology, Ann Arbor, Michigan 48109, USA. Email: seaneddy@med.umich.edu.

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Find articles by Brosius, F. in: JCI | PubMed | Google Scholar

1Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan, USA.

2Novo Nordisk A/S, Måløv, Denmark.

3Eli Lilly & Co., Indianapolis, Indiana, USA.

4Biopharmaceuticals R&D, AstraZeneca, Cambridge, United Kingdom.

5Johnson & Johnson, New Brunswick, New Jersey, USA.

6University of Arizona, Tucson, Arizona, USA.

7Novo Nordisk Research Center Seattle, Inc, Seattle, Washington, USA.

Address correspondence to: Sean Eddy, Associate Research Scientist, University of Michigan, Department of the Internal Medicine, Division of Nephrology, Ann Arbor, Michigan 48109, USA. Email: seaneddy@med.umich.edu.

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Find articles by Kretzler, M. in: JCI | PubMed | Google Scholar |

1Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan, USA.

2Novo Nordisk A/S, Måløv, Denmark.

3Eli Lilly & Co., Indianapolis, Indiana, USA.

4Biopharmaceuticals R&D, AstraZeneca, Cambridge, United Kingdom.

5Johnson & Johnson, New Brunswick, New Jersey, USA.

6University of Arizona, Tucson, Arizona, USA.

7Novo Nordisk Research Center Seattle, Inc, Seattle, Washington, USA.

Address correspondence to: Sean Eddy, Associate Research Scientist, University of Michigan, Department of the Internal Medicine, Division of Nephrology, Ann Arbor, Michigan 48109, USA. Email: seaneddy@med.umich.edu.

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Find articles by Ju, W. in: JCI | PubMed | Google Scholar

1Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan, USA.

2Novo Nordisk A/S, Måløv, Denmark.

3Eli Lilly & Co., Indianapolis, Indiana, USA.

4Biopharmaceuticals R&D, AstraZeneca, Cambridge, United Kingdom.

5Johnson & Johnson, New Brunswick, New Jersey, USA.

6University of Arizona, Tucson, Arizona, USA.

7Novo Nordisk Research Center Seattle, Inc, Seattle, Washington, USA.

Address correspondence to: Sean Eddy, Associate Research Scientist, University of Michigan, Department of the Internal Medicine, Division of Nephrology, Ann Arbor, Michigan 48109, USA. Email: seaneddy@med.umich.edu.

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Find articles by Karihaloo, A. in: JCI | PubMed | Google Scholar

1Department of Internal Medicine, Division of Nephrology, University of Michigan, Ann Arbor, Michigan, USA.

2Novo Nordisk A/S, Måløv, Denmark.

3Eli Lilly & Co., Indianapolis, Indiana, USA.

4Biopharmaceuticals R&D, AstraZeneca, Cambridge, United Kingdom.

5Johnson & Johnson, New Brunswick, New Jersey, USA.

6University of Arizona, Tucson, Arizona, USA.

7Novo Nordisk Research Center Seattle, Inc, Seattle, Washington, USA.

Address correspondence to: Sean Eddy, Associate Research Scientist, University of Michigan, Department of the Internal Medicine, Division of Nephrology, Ann Arbor, Michigan 48109, USA. Email: seaneddy@med.umich.edu.

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Find articles by Eddy, S. in: JCI | PubMed | Google Scholar |

Authorship note: FA and VD are co–first authors. AK and SE are co–senior authors.

Published March 10, 2025 - More info

Published in Volume 10, Issue 5 on March 10, 2025
JCI Insight. 2025;10(5):e186070. https://doi.org/10.1172/jci.insight.186070.
© 2025 Alakwaa et al. This work is licensed under the Creative Commons Attribution 4.0 International License. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/. Published March 10, 2025 - Version history
Received: August 23, 2024; Accepted: January 22, 2025 View PDF Abstract

Chronic kidney diseases (CKDs) are a global health concern, necessitating a comprehensive understanding of their complex pathophysiology. This study explores the use of 2 complementary multidimensional -omics data integration methods to elucidate mechanisms of CKD progression as a proof of concept. Baseline biosamples from 37 participants with CKD in the Clinical Phenotyping and Resource Biobank Core (C-PROBE) cohort with prospective longitudinal outcome data ascertained over 5 years were used to generate molecular profiles. Tissue transcriptomic, urine and plasma proteomic, and targeted urine metabolomic profiling were integrated using 2 orthogonal multi-omics data integration approaches, one unsupervised and the other supervised. Both integration methods identified 8 urinary proteins significantly associated with long-term outcomes, which were replicated in an adjusted survival model using 94 samples from an independent validation group in the same cohort. The 2 methods also identified 3 shared enriched pathways: the complement and coagulation cascades, cytokine–cytokine receptor interaction pathway, and the JAK/STAT signaling pathway. Use of different multiscalar data integration strategies on the same data enabled identification and prioritization of disease mechanisms associated with CKD progression. Approaches like this will be invaluable with the expansion of high-dimension data in kidney diseases.

Introduction

Chronic kidney diseases (CKDs) are a global public health concern and a risk factor for adverse outcomes in many diseases, including cardiovascular disease. In its 2022 report, the United States Renal Data System estimates that 14% of US adults have CKD and 2 in every 1,000 living persons are at risk of kidney failure (1). To address the large medical need in CKD, renin-angiotensin-aldosterone system blockers were developed and approved as first-line treatments in diabetic kidney disease (DKD) and CKD more than 2 decades ago (2–4). Recently, SGLT2 inhibitors and nonsteroidal mineralocorticoid receptor antagonists have been added to the armamentarium. These 2 new options have been specifically approved for diabetic and nondiabetic CKD by the Food and Drug Administration after successful phase III studies (5–7). Despite these advances, the residual risk of CKD progression and kidney failure remain high due to several factors, including underlying pathomechanistic heterogeneity (8, 9). Furthermore, patients have differential responses to treatments, highlighting the clinical importance of understanding disease heterogeneity. One approach to understanding disease heterogeneity is by characterizing the underlying differences in molecular phenotypes (8, 10). This approach has focused on well-characterized molecular phenotypes along a single -omics data type (e.g., transcriptomics). Additionally, multiple -omics data types can be integrated, in aggregate, to characterize molecular clusters. Multi-omics data integration has many applications, including biomarker discovery (11), drug and target discovery (12), drug repurposing (13), and patient stratification (14). These integration and systems biology approaches have been used to explore molecular pathophysiology in other diseases, including cancer (15), type 2 diabetes (16), osteoarthritis (17), Alzheimer disease (18), systemic lupus erythematosus (19), inflammatory bowel disease (20), and in rare diseases (21). We can now apply these approaches to CKD.

In molecular profiling studies of CKD, a significant challenge lies in comprehensively exploring the interrelationships across diverse high-dimension -omics data types to molecularly define or classify patients. Advanced statistical and machine-learning techniques, particularly multivariate- and Bayesian-based methods, have demonstrated superior performance in low to moderate sample sizes (22, 23). Here, we leverage 2 validated computational algorithms for integrating multiple -omics datasets, multi-omics factor analysis (MOFA) and data integration analysis for biomarker discovery using latent components (DIABLO). MOFA, an unsupervised algorithm, identifies sources of disease-associated variation by generating computed factors (24, 25). DIABLO is a supervised method that focuses on uncovering disease-associated multi-omic patterns (26). Notably, both MOFA and DIABLO were specifically validated for chronic diseases and exhibited robust performance even with small sample sizes (26, 27). By harnessing both these validated computational algorithms, we hypothesized that the challenges posed by the complexity of CKD could be overcome to gain insights into disease-associated variation and multi-omic patterns. This study applied this nuanced and innovative approach to molecularly define and classify disease biology in the context of CKD.

Longitudinal and multi-omics data from the Clinical Phenotyping and Resource Biobank Core (C-PROBE) cohort were used for this study. C-PROBE is a multisite cohort designed to accelerate translational research in kidney disease (28). The availability of multi-omic profiles for participants in C-PROBE facilitated a comprehensive evaluation of the effectiveness of MOFA and DIABLO. The selection of urine and plasma proteomic, kidney transcriptomic, and metabolomic data, coupled with longitudinal information, was deliberately designed to strategically capture diverse facets of molecular information associated with CKD progression in this cohort. By encompassing this spectrum, our study was designed to capture a range of molecular events implicated in CKD progression.

Results

We developed a comprehensive framework to integrate diverse data points (henceforth denoted as features) from metabolomics, urine proteomics, plasma proteomics, and tubulointerstitial transcriptomics from a discovery cohort 37 participants enrolled in the CKD C-PROBE cohort, coupled with longitudinal data using the 2 independent integration approaches, MOFA and DIABLO (Figure 1 and Supplemental Figure 1; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.186070DS1). An additional 94 participants from C-PROBE that were independent of the discovery cohort were used to validate select findings from the MOFA and DIABLO integration.

Figure 1

MOFA and DIABLO integrative approaches applied to the C-PROBE cohort. First step is training of the MOFA and DIABLO models. Second is the visualization of the variation and analysis of top-ranked features by both algorithms. Pathway analysis for top-ranked features is the third step. Validation of shared features in C-PROBE is the fourth step in this integrative approach.

To determine whether this framework might be broadly applicable to CKD, we used discovery and validation cohorts of varying clinical and histopathological makeup. Baseline clinical characteristics are provided for participants included in the discovery and validation datasets, including those who reached the composite CKD endpoint, in Table 1. The discovery cohort consisted of patients diagnosed with nondiabetic glomerular diseases with an average age of 39 years, a 5-year follow-up period, an estimated glomerular filtration rate (eGFR) average of 74 mL/min/1.73 m2, and a urinary albumin to creatinine ratio (uACR) average of 1.89 mg/mg. Participants in the validation cohort were older, predominantly diagnosed with DKD, had lower average eGFR (58 mL/min/1.73 m2), and lower uACR (0.69 mg/mg) in comparison with the discovery cohort, with 33 participants reaching the composite endpoint (Table 1). The observed differences in eGFR and uACR between the cohorts may be attributed to both age and distinct disease etiologies.

Table 1

Demographic characteristics of participants who provided the discovery (n = 37) and validation (n = 94) samples within the C-PROBE cohort

The number of features in the transcriptomics data (16,840 features) was more than an order of magnitude larger than the number of features in the next largest platform (1,301 urine and plasma features) and 2 orders of magnitude greater than the smallest platform included (164 metabolomic features). To normalize the data space to comparable dimensionality across data types, the top 20% most variable individual gene expression profiles across the 37 samples were retained. This resulted in 3,368 gene expression profiles as input features. The analysis pipelines for the 2 distinct integration algorithms employed are summarized in Supplemental Figure 2. MOFA, depicted in Supplemental Figure 2A, integrates multiple -omics datasets, capturing their complex interrelationships. DIABLO, illustrated in Supplemental Figure 2B, identifies shared variation across datasets through multivariate analysis, providing a comprehensive view of common molecular patterns.

Unsupervised data integration using MOFA identified key disease-associated mechanisms. The first step in the unsupervised MOFA analysis was to reduce the dimensionality of the -omics data from 6,134 input features into uncorrelated and independent factors. Based on MOFA guidelines for factor selection with the given dataset dimensionality (24), we identified 7 independent factors (outlined by K = 7) from the 6,134 input features (Supplemental Figure 3A). With 7 factors the model explains 42% of the variation in the plasma proteomic data, 43.7% of the variation in the urine proteomic data, 26.4% of transcriptomic data, and 3.4% of metabolomic data (Figure 2A). MOFA Factors 1 and 5 explained most of the variance in plasma proteomics, Factors 2 and 6 explained most of the variance in urine proteomics, Factor 3 explained variance across all data types, while Factors 4 and 7 mostly explained variance in transcriptomic data (Figure 2B).

Figure 2

Factors from MOFA model. (A) Total percentage of variance explained by MOFA factors. (B) Data variance explained by each MOFA factor. (C) Kaplan-Meier (KM) survival curve using the value of MOFA Factor 2 reaching composite endpoint. (D) KM survival curve using the value of MOFA Factor 3 reaching composite endpoint. Log-rank test was used to determine significant differences in KM curves.

Because MOFA is a framework for unsupervised discovery of inherent variability in a given dataset, we next sought to prioritize factors for further investigation by asking whether any MOFA factors identified were associated with outcomes. Participants were grouped into either high or low factor expression categories based on optimal cut points derived from survival analysis, which assessed their risk of reaching the composite kidney endpoint for disease progression (40% loss of eGFR or kidney failure); significance was determined by log-rank test. Lower levels of Factor 2 and Factor 3 were significantly associated (P = 0.00001 and P = 0.00048, respectively) with CKD progression as shown in Kaplan-Meier (KM) curves (Figure 2, C and D, and Supplemental Figure 3, B–F). As noted above, Factor 2 was explained by variance in urine proteomic profiles, while Factor 3 was explained by variance across multiple -omics types (Figure 2B).

Given the associations with outcomes for Factors 2 and 3, we explored the contributing features to understand the underlying biology. Interestingly, many of the urine features, which explained the majority of variance in Factor 2, were assigned a negative weight by the MOFA model and were inversely correlated with Factor 2 expression (Figure 3A). As can be seen in the heatmap, urinary protein analytes F9, F10, APOL1, and AGT were among those contributing the most to Factor 2. Given that these proteins were inversely correlated with Factor 2, higher expression levels of these features are associated with worse outcomes. The top features from each -omics domain and weighting of all features contributing to Factor 2 are included in Supplemental Figure 4A and Supplemental Table 1, respectively. Similarly, many features contributing to Factor 3 were also inversely correlated with Factor 3 expression found in Supplemental Table 2 and depicted in Supplemental Figure 5A.

Figure 3

Expression levels and pathway enrichment of top 10 MOFA extracted features. (A) The expression levels of urine proteomics that are top ranked by MOFA Factor 2. (B) Enriched pathways of the top 100 features extracted from MOFA Factor 2. (C) Protein-protein interaction network between complement components from 3 -omics data types. Gray-colored nodes depict features identified in a minimum of any 2 of the -omics data types.

To characterize biology most associated with each factor, the top 100 features ranked by MOFA Factor 2 and Factor 3 were used for pathway enrichment analysis. The complement and coagulation cascades pathway was enriched across the 3 -omics data types (Figure 3B and Supplemental Figure 4, B–D) in Factor 2. Enriched complement components in urine proteomics, intrarenal transcript expression, and plasma proteomics were highly connected in a protein-protein interaction network derived from the STRING (29) database (Figure 3C). For MOFA Factor 3, the pathway enrichment analysis revealed the enrichment of the cytokine–cytokine receptor interaction pathway across 3 -omics data types (Supplemental Figure 5, B–E). Furthermore, the urine proteomics profiles derived from Factor 3 exhibited specific enrichment in the JAK/STAT signaling pathway (Supplemental Figure 5C). Notably, a number of additional shared pathways were identified from enrichment analysis of Factors 2 and 3, including the PI3K/Akt signaling pathway, MAPK signaling pathway, NF-κB signaling pathway, Rap1 signaling pathway, and axon guidance, suggesting the MOFA factors are independently capturing orthogonal features in CKD that are shared across multiple pathways. The primary features contributing to Factors 2 and 3 are included in Supplemental Tables 1 and 2, respectively.

Supervised data integration using DIABLO identified key disease-associated mechanisms. We used the same input data as MOFA (6,134 input features) in a supervised analysis of the data where the 37 participants were stratified into progressors versus nonprogressors. DIABLO identified 38 mRNA species, 24 plasma protein features, 34 urinary protein features, and 12 metabolites (overall 108 features) that discriminated progressors from nonprogressors. These features that are highly correlated (coexpressed) also captured the variance across each -omics block (Figure 4A). Transcriptomics-, plasma-, and urine proteomics–driven variance was 21.2%, 22.9%, and 19.4%, respectively, while metabolite variance was 12.4%. Supplemental Table 3 lists features from all 4 -omics data types that were selected by DIABLO. Figure 4B represents the overall expression distribution of the top 10 urinary proteins identified by DIABLO. The expression levels of the top 10 features from all 4 -omics data types, demonstrating the integrated nature of data selection in DIABLO, are depicted in Supplemental Figure 6. Figure 4C shows the canonical pathways enriched in these 108 features. It depicts complement and coagulation cascades and JAK/STAT signaling as the top enriched KEGG pathway terms from urinary proteins and cytokine–cytokine receptor interactions from transcripts and urinary proteins, while glycine, serine, and threonine metabolism are among the top enriched metabolic pathways.

Figure 4

Top 10 features, expression levels, and enriched pathways using DIABLO. (A) Sparse partial least-squares discriminant analysis (sPLS-DA) plot for features identified by DIABLO. (B) Normalized expression levels of top 10 features identified by DIABLO in progressors and nonprogressors. (C) Pathway enrichment analysis for features identified by DIABLO model. KEGG pathway was the top pathway identified from RNA and proteins by DIABLO, while the lower panel, created using DIABLO, identified metabolites using MetaboAnalyst.

Combining results from MOFA and DIABLO. We discovered that unsupervised MOFA Factor 2 (urine selective), MOFA Factor 3 (multi-omic driven), and supervised DIABLO results converged at the pathway level. Features contributing to MOFA Factor 2 and DIABLO resulted in enrichment for the complement and coagulation cascades pathway (Figure 3B and Figure 4C), while features contributing to MOFA Factor 3 and DIABLO resulted in enrichment for JAK/STAT and cytokine–cytokine receptor interactions pathways (Supplemental Figure 5B). This indicated that common features contributing to Factor 2 and Factor 3 from MOFA and features identified from DIABLO could be used to identify high-confidence findings in the urine. The intersect of 8 urinary protein probes, corresponding to 7 unique proteins, between the MOFA Factor 2 and DIABLO models (Figure 5A, details in Table 2) were significant (hypergeometric P < 0.003).

Figure 5

Consensus urinary proteins identified by MOFA and DIABLO. (A) Venn diagram of intersection between top 100 urinary proteins ranked by MOFA Factor 2 and 34 urinary proteins identified by DIABLO. (B) KM curve of 1 shared urinary protein, complement C9, identified by both methods depicting higher concentration is associated with increased risk of progression to composite endpoint in the validation cohort. Log-rank test was used to determine significant differences in KM curves.

Table 2

Association of shared urinary proteins between MOFA and DIABLO with CKD progression in the validation cohort

While MOFA selects factors independently of any outcome, DIABLO selects individual features based on their direct association with a desired endpoint. This targeted approach by DIABLO is evident in the significant associations observed between higher levels of specific urinary proteins and CKD progression. As shown in Table 2, higher levels of urinary proteins were significantly associated with CKD progression, as evidenced by the KM survival curves, with significant differences evaluated using the log-rank test (P < 0.05). An example KM curve is presented for complement C9 (Figure 5B). Additionally, the P values from the KM model of other shared urinary proteins are provided in Table 2.

MOFA Factor 3 (Supplemental Figure 5, C–E) and DIABLO (Figure 4C) jointly highlighted the JAK/STAT pathway and cytokine–cytokine receptor interaction pathway. In the cytokine pathway, MOFA Factor 3 and DIABLO shared enrichment of 2 transcripts (CXCL6, CXCL1) and 1 plasma protein (CCL23). In the JAK/STAT pathway, only 1 urinary protein (PIAS4) was enriched in both DIABLO and MOFA Factor 3 (Supplemental Table 4). A full intersection of pathways identified by MOFA Factor 2, MOFA Factor 3, and DIABLO are pres