A Holistic Approach to AKI: Integrating Clinical and Molecular Data in the Human Kidney

Summary

Despite intensive research efforts, acute kidney injury (AKI) is a common clinical syndrome that has limited treatment options apart from supportive care. The increasing availability of molecular interrogation data from patients with Acute Kidney Injury provides an unparalleled opportunity to leverage systems biology approaches. In this review, we discuss the challenges with AKI research, explain how systems biology approaches can link molecular data to clinical phenotypes, review available molecular interrogation tools and techniques, and provide examples where systems biology approaches have been successfully applied in nephrology.

THE CHALLENGES OF AKI

Acute kidney injury (AKI) is a costly syndrome associated with increased risk of chronic kidney disease (CKD) and death.^1,2 Current clinical definitions of AKI rely on a rising serum creatinine or decreasing urine output, which can encompass several underlying diagnoses including prerenal azotemia, acute tubular injury (ATI), acute interstitial nephritis (AIN), and urinary obstruction. Despite more than 100,000 research publications on AKI published to date, no treatment options apart from supportive care are available for most patients. In contrast, while there was a prolonged period of limited advances in clinical care for patients with CKD, another common kidney disease syndrome, there has recently been a welcome plethora of progress thanks to successful clinical trials with SGLT2 inhibitors,^3–5 GLP-1 receptor agonists,⁶ and mineralocorticoid antagonists.⁷

There are several facets of AKI that pose challenges, with the foremost issue being that it is a heterogeneous syndrome. Serum creatinine has not been supplanted by any other assays in clinical care, but it has multiple limitations. It increases many hours after the initial AKI insult and provides no information regarding the underlying etiology of the initial insult. Moreover, it is influenced by such factors as muscle mass and volume status, which are often deranged in ill, hospitalized patients. With serum creatinine remaining the mainstay for the detection of AKI, we have limited tools available to non-invasively identify subgroups, or endophenotypes, of AKI. Indeed, urine microscopy, which was developed over a thousand years ago,⁸ is still a mainstay of AKI diagnosis and evaluation.⁹ AKI is a dynamic process with molecular mechanisms that change over the course of hours to days, which complicates the interpretation of any data. All of these facets of disease heterogeneity have resulted in numerous negative clinical trials of AKI,^10–16 with some early successes recently reported.^17,18 Indeed, many positive trials focus on optimizing supportive care instead of providing a novel therapeutic option for patients.^19–21

In contrast to the disappointing clinical trials in the AKI field, there is a wealth of careful experimental work from animal models that has yielded insights into the underlying molecular mechanisms of AKI. Murine animal models, including ischemia-reperfusion, lipopoly-saccharide, and ureteral obstruction, have elucidated molecular mechanisms of AKI and identified molecular changes associated with the transition from AKI to CKD, sometimes termed maladaptive repair. While a detailed review of the findings from this work is outside the scope of this article, excellent reviews are available.^22,23 One of the most commonly cited success stories is the identification of the tubular injury biomarker, Kidney Injury Marker-1 (KIM-1), which was initially identified in an animal model system²⁴ and translated to human kidney tissue,²⁵ and is now commonly used in human research studies. Despite the depth and breadth of findings from experimental systems, many of these results have failed to translate into human research. Rodents have a lower susceptibility to AKI than humans do, and experimental systems rely on a single kidney insult in contrast to patients who commonly experience multiple potential kidney insults superimposed on preexisting subclinical chronic damage. While many of the molecular mechanisms identified in murine models are likely applicable in human disease, there is currently no clinically available strategy to link murine models responding to an experimental agent to the endophenotypes of human AKI.

Recognizing that further work is needed to translate molecular mechanisms identified in murine animal models into humans, the development of noninvasive biomarkers of AKI has been an actively pursued area of research. The rationale is that urinary biomarkers of AKI could potentially function as a “liquid biopsy” that would help clinicians more accurately identify the cause and histologic type of injury and help prognosticate patients.²⁶ Carefully done prospective cohorts of biomarkers of AKI have demonstrated that biomarkers do prognosticate patients and identify tubular injury earlier than serum creatinine.²⁷ Furthermore, follow-up trends in urinary biomarkers could help risk-stratify patients who are at high risk of progression to CKD.²⁸ However, urinary biomarker profiles highlight the heterogeneity of AKI in humans, and the diagnostic sensitivity and specificity of most urinary biomarkers are limited as of now. Some investigators have begun identifying subphenotypes of AKI by classifying patients based on urinary biomarker profiles. For example, investigators were able to identify two subphenotypes (SP1 and SP2) of AKI in critically ill patients, which could be accurately identified by measuring Ang-2/Ang-1 and sTNFR-1.²⁹ These two subphenotypes differed in their 30-day mortality rate and response to vasopressin in the VASST study (Vasopressin and Septic Shock Trial). Additionally, the NephroCheck platform urine levels of insulin-like growth factor binding protein 7 (IGFBP7) and tissue inhibitor of metalloproteinases (TIMP-2), markers of tissue damage and remodeling, are determined to assess the risk of moderate to severe AKI within 12 hours of insult.³⁰ The PrevAKI trial demonstrated that identifying patients at high risk of AKI after surgery by measuring their NephroCheck successfully increased rates of adherence to a KDIGO bundle (a combination of supportive measures such as optimizing volume status and avoiding nephrotoxic drugs), which was the primary outcome of the study.^19,20 The trial also found lower rates of moderate to severe AKI in those who were randomized to receive the KDIGO bundle. Moreover, recent work is applying innovative strategies to employ biomarkers for stratification of patients in clinical trials of AKI.³¹ With these advances, successful clinical trials based on sub-phenotyping patients by biomarker profiles will be possible and needed to increase the uptake of biomarkers into clinical practice.

SYSTEMS BIOLOGY APPROACHES IN ADVANCING UNDERSTANDING OF AKI

Given the limited progress to identify novel treatment options for patients with AKI through clinical trials, murine models, and translational biomarkers, investigators have considered alternate scientific approaches. Mechanistic, hypothesis-based research proceeds in a reductionist fashion where individual components of a complex system are examined in detail, which has yielded in-depth insight into specific molecular mechanisms altered in AKI. A systems biology approach strives to take knowledge from the definition of individual molecular changes to an evaluation of the multiple interacting elements of a disease process in a cell, an organ, or an entire organism. We have previously defined systems biology as “the use of computational modeling and mathematical techniques for developing an integrative picture of a biologic system derived from multiple types of data”³² (Figure 1). By integrating multiple layers of data in a patient-centric manner, including clinical phenotypes, pathology data, and molecular data, we can reconstruct what elements of the entire system are perturbed. The data integration approach should allow the identification of novel endotypes. We believe that applying systems biology approaches to pressing questions in AKI may lead us to a precipice of novel discoveries. However, this approach requires access to multiscalar datasets of the damaged organ, so access to human kidney tissue from individuals with AKI becomes critical to advance the molecular interrogation to succeed. With the prerequisites now established, novel opportunities for disease insight in patients with AKI emerge.

Figure 1.

Integrative systems approach to address challenges in kidney disease. A systems biology approach to the kidney adds to the reductionist approach by leveraging recent advances in computational technology and methods to integrate diverse sets of data. Reproduced with permission from Schaub et al.³²

ADVANCES IN MOLECULAR INTERROGATION TECHNOLOGIES

The earliest iterations of molecular interrogation technologies for kidney tissue were hampered by two main limitations. First, technologies would analyze all kidney tissue in bulk or from microdissected glomerular and tubular compartments, which created an average signal of all the cell types in the kidney. There are over 20 different cell types in the kidney, so this approach does not allow the identification of cell-specific responses and cellular cross-talk. Second, earlier versions of molecular interrogation technologies did not retain spatial information about the distribution of damage and recovery across cellular neighborhoods.

Despite the limitations in earlier versions of molecular interrogation technologies for human kidney tissue, the approach did yield important advances. For example, the identification of Anti-PLA2R antibodies in the serum and protein extracts from the glomeruli of patients with “idiopathic” membranous nephropathy could be considered one of the earliest systems biology approaches in patients with kidney disease.³³ This discovery reclassifies idiopathic membranous nephropathy into a disease identified by an underlying pathologic mechanism. Measurement of anti-PLA2R antibodies can now facilitate diagnosis and monitor response to treatment.³⁴ Moreover, despite the limitations of bulk RNAseq data from kidney tissue, urinary EGF was identified as a robust biomarker of CKD progression by associating the kidney tissue transcriptome with eGFR. By identifying candidate predictors of CKD progression from the transcriptome, investigators could interrogate the urine proteome to identify urinary EGF.³⁵ Further research shows that urinary EGF can be a predictor to response of SGLT2 inhibitors and a prognostic biomarker in AKI.^35–37

The past decade has seen a surge in molecular interrogation technologies that allow comprehensive interrogation of the epigenome, genome, transcriptome, proteome, and metabolome (Table 1). Many of these technologies are now at or close to single-cell resolution, which overcomes some of the earlier limitations, and several technologies retain spatial information of the structural context of the damaged kidney. Biologic information flows, with multiple feedback loops, from the genome and the associated epigenome to the transcriptome, proteome, and metabolome, and there are now molecular interrogation technologies at each stage of biologic information that can extract information from kidney tissue.

Table 1.

Overview of Molecular Interrogation Technologies

Data Type	Technology	Resolution	Spatial	Genome Scale
Epigenomic
	ATACseq	Bulk	No	Genome
	snATACseq	Cellular	No	Genome
	CUT&RUN	Bulk	No	Genome
	CUT&Tag	Bulk	No	Genome
	WGBS	Bulk	No	Genome
Genomic
	Genome-Wide Association Study	N/A	N/A	Genome
	Whole Exome Sequencing	N/A	N/A
Transcriptomic
	RNA Sequencing	Bulk	No	Genome
	scRNAseq	Cellular	No	Genome
	snRNAseq	Cellular	No	Genome
	Spatial Transcriptomic	Near single cell	Yes	Genome
Proteomic
	Imaging Mass Cytometry	Single cell	Yes	Panels
	CODEX	Single cell	Yes	Panels
Metabolomics
	Mass Spectrometry	Near single cell	Yes	High abundance

Epigenomic and Genomic

The epigenome consists of modifications to DNA and DNA-associated proteins that regulate if a gene will be transcribed or not. There are several layers of regulation, including changes in chromatin accessibility, DNA methylation, and histone modifications. Each of these regulatory elements can now be interrogated. For example, single-nucleus ATAC (Assay for Transposase Accessible Chromatin) seq (snATACseq) can detect regions of open chromatin at the single nucleus level. Technologies such as CUT&RUN can measure histone modifications for active chromatin (H3K27ac, H3K4me3, H3K4me1) and repressive chromatin (H3K27me3). Depending on the antibody used, CUT&RUN can also be used to identify if a particular transcription factor is binding to a particular binding site. Whole-genome bisulfite sequencing (WGBS) can measure the extent of DNA methylation. Combining this information allows a detailed overview of what areas of DNA are regulated epigenetically.

DNA sequencing technologies have become increasingly more successful and affordable, although AKI is a complex trait that may not have high heritability. Genome-wide association studies (GWAS), which can link single-nucleotide polymorphisms to patient traits, have become commonly used to evaluate if there are genetic underpinnings to diseases. Thus far, most of the GWAS studies to examine the genetic propensity for AKI have likely been underpowered.^38–41 Much larger sample sizes, such as those in the CKD-Gen Consortium,^42,43 will likely be needed before we can definitively identify any potential genetic underpinnings to AKI. Whole-exome sequencing has demonstrated that genetic underpinnings in CKD may impact up to 10% of patients, so it is possible that with sufficient samples we may identify potential genetic causes of AKI.⁴⁴

Transcriptomic

One of the most consequential technologies has been the development of single-cell and single-nuclear RNA sequencing (scRNAseq and snRNAseq). Both technologies provide a comprehensive readout of the transcriptome of each individual cell (or nucleus) in a sample of kidney tissue. The transcriptome of each cell is used to agnostically cluster the cells into similar types, and the expression profile of each cell cluster is then used to annotate what cell type each cluster is likely to represent. These technologies have been particularly impactful in the kidney because it contains over 20 cell types, and they allow detailed insight into the transcriptional profile at a single-cell level. Moreover, multiome sequencing now obtains both snATACseq and snRNAseq from the same nucleus, which links epigenetic information with transcriptional information. However, there are limitations to each technology. scRNAseq tends to capture immune cells better but under-capture differentiated epithelial cells and podocytes compared to snRNAseq.

Despite the substantial advancements from scRNAseq or snRNAseq, the cells/nuclei are disassociated during the sequencing process, so the spatial localization of each cell type is lost. Given the structural changes that occur with kidney disease, associating molecular changes with structural changes is particularly important. This challenge is addressed by the development of spatial transcriptomic technologies, which is a rapidly developing area. There are now multiple different technologies and platforms, and a detailed review of these technologies is available elsewhere.⁴⁵ The first generation of spatial transcriptomic technologies, such as Visium, applied beads to each piece of kidney tissue with hybridized probes that provided a comprehensive readout of the transcriptome of each piece of tissue. However, each bead covers about 10–30 cells, so the resolution is at a regional, not cellular, level, still requiring deconvolution strategies to define the individual elements of the profiled kidney tissue neighborhoods. Subsequent generations of the technology are now at or below single-cell resolution, which will allow integration of transcriptomic alterations within the spatial context of the kidney.

Proteomics

Transcript expression does not ensure changes in the proteins or metabolites of the cell that result in functional changes. There are now novel proteomic imaging techniques that can provide spatial localization for a range of proteins across a cell, such as CODEX and Imaging Mass Cytometry.⁴⁶ However, there is typically a limited number of proteins available for interrogation. Despite these limitations, proteomics technologies, such as CODEX, have facilitated the identification of novel collecting duct subcell types in patients who form kidney stones.⁴⁷

Metabolomics

Lastly, spatial metabolomics technology provides the opportunity to examine metabolites within a spatial context in kidney tissue. Spatial metabolomic technology provides spatial localization of metabolites within kidney tissue, such as MALDI-MS.⁴⁸ However, there are currently a limited number of metabolites that can be detected, and the resolution is not yet at the single-cell level,⁴⁹ with rapid progress expected to address these limitations in the near future. Currently, investigators have applied MALDI-MS in combination with multiplex immunofluorescence to identify metabolism differences in proximal tubular (PT) cells after ischemia-reperfusion injury (IRI), which demonstrated that there may be changes in lactate production between persistently injured PT cells and healthy PT cells after a renal insult.⁵⁰

NOVEL COMPUTATIONAL APPROACHES FOR OMICS DATA ANALYSIS

Integration of Multi-Omics Data With Kidney Tissue and Phenotypes

The complexity of kidney diseases demands multilevel molecular analysis to understand pathophysiology and identify therapeutic targets. The intricate structure of the kidney, with diverse cell types organized into functional units, also creates unique challenges for molecular profiling and data integration.⁵¹ Recent advancements in science and technology have enabled researchers to better characterize renal tissue at various molecular levels. Yet translating these disparate data modalities into cohesive biological insights remains a herculean task. Integrating transcriptomics, proteomics, and digital pathology with clinical phenotypes such as glomerular filtration rate and proteinuria has begun revealing regulatory networks and cellular crosstalk patterns specific to nephron segments.⁵² Despite the computational progress, each approach may face limitations when applied to kidney tissue given its structural and cellular complexity. Thus, each approach will also have to leverage spatial data to identify regional variations in functional responses.

Several frameworks have emerged for multi-omics data integration, each with distinct advantages for kidney research applications. These broadly fall into three categories:

• Early/pre-analysis integration methodologies combine raw or preprocessed datasets before analysis, creating unified high-dimensional matrices encompassing multiple molecular layers. Researchers typically employ dimensionality reduction techniques like principal component analysis to manage computational complexity. While conceptually straightforward, this approach struggles with heterogeneous data characteristics, including varying noise profiles and measurement scales across omics platforms.⁵³ Dominant signals from one molecular layer may overshadow subtle but biologically relevant patterns in others, potentially compromising cross-platform relationship discovery.

• Late/post-analysis integration frameworks analyze each omics dataset independently before synthesizing results at the interpretation level. This methodology preserves data type-specific characteristics and enables the application of specialized algorithms tailored for genomics, transcriptomics, proteomics, or metabolomics data. The approach maintains signal integrity and leverages domain-specific statistical methods effectively. However, this compartmentalized strategy may overlook crucial inter-platform molecular interactions fundamental to understanding complex kidney disease pathophysiology, as temporal separation of analyses can miss coordinated regulatory networks spanning multiple molecular layers.⁵⁴

• Intermediate integration approaches represent a balanced compromise, incorporating sophisticated frameworks like factor analysis and network-based methodologies. These techniques demonstrate efficacy in kidney applications by preserving platform-specific characteristics while capturing meaningful cross-omics relationships. Multi-Omics Factor Analysis (MOFA) exemplifies this approach by decomposing complex data into interpretable latent factors explaining coordinated variation patterns across molecular layers, proving valuable for disease subtyping and biomarker discovery.⁵⁴

Network-based integration strategies show promise for nephrology research, modeling intricate cellular and molecular interdependencies characteristic of kidney architecture. These approaches represent molecular entities as network nodes with functional relationships as connecting edges, creating computational models mirroring the nephron hierarchical organization. Such representations facilitate identifying regulatory hubs and pathway perturbations driving disease progression across multiple molecular scales.

Computational biologists are continuously innovating in the computational space to develop novel methods to analyze molecular interrogation data. Given that molecular interrogation data is highly multiscalar and there are hierarchical layers of variability (patient level, sample level, cell level), analyzing these data requires sophisticated analytical techniques, some of which are featured below (Table 2).

Table 2.

A Summary of Data Integration Techniques

Tool	Integration Approach	Example Use in Kidney Research	Key Features
MOFA	Intermediate	Integrated transcriptomic, proteomic, and clinical data in CKD	Captures shared latent factors
SNF	Early	Integrated genomic, transcriptomic, and clinical data in KIRC	Fuse similarity networks
WGCNA	Late	Correlated co-expression modules with clinical traits in diabetic kidney disease	Builds co-expression networks
Nephroseq	Late	Analyzed gene expression with clinical phenotypes in FSGS	Resource for meta-analysis, biomarker discovery
OmicsIntegrator	Intermediate	Integrated transcriptomic and proteomic data in KIRC for pathways	Network-based, pathway prioritization
DIABLO/mixOmics	Intermediate	Identified urinary proteins in CKD progression (C-PROBE)	Supervised, biomarker discovery via sPLS-DA

Multi-Omics Factor Analysis v2

Multi-Omics Factor Analysis (MOFA)^55,56 is a statistical framework for integrative analysis of multi-omics data from common samples. However, this version was lacking in scalability for multimodal analysis. MOFA+, which addresses critical limitations in MOFA by developing a stochastic variational inference framework for GPU computations, enables analysis of datasets with potentially millions of cells and incorporates priors for structure regularization to jointly model multiple groups and data modalities. MOFA+ takes multiple datasets with features aggregated into nonoverlapping modalities and cells aggregated into nonoverlapping groups. The model employs an extended group-wise prior hierarchy enabling simultaneous integration of multiple data modalities and sample groups. After training, MOFA + output supports variance decomposition, feature weight inspection, differentiation trajectory inference, and clustering, supporting biological explainability of the mapping. Both versions of MOFA have been applied to kidney research in many ways, including for understanding mechanistic insights into kidney diseases, such as identifying urinary proteins associated with CKD progression.^57,58 This approach could be adapted in AKI patients to identify markers of incomplete recovery from an AKI event.

Weighted Gene Co-expression Network Analysis (WGCNA)

While initially designed for single-omics applications, extended versions of WGCNA have been developed for multi-omics integration. WGCNA has successfully identified co-expression modules associated with diabetic kidney disease progression, revealing potential drug repurposing opportunities.⁵⁹ Its ability to detect coordinated gene programs across nephron segments provides biologically interpretable results. However, it is limited in nonoverlapping gene assignment and can be sensitive to outliers in heterogeneous kidney tissue samples. This approach would also be suitable for adoption in AKI to identify molecular pathways that are associated with patient outcomes, such as dialysis requirements, severity of AKI, or incomplete recovery.

DIABLO/mixOmics

This supervised framework identifies multi-omics signatures associated with outcomes of interest. Gupta and colleagues applied this approach to lupus nephritis, integrating transcriptomics, proteomics, and metabolomics to identify patient subgroups with distinct treatment responses,⁶⁰ which can also be adapted to identify disease subgroups in AKI. The method’s discriminative power makes it valuable for kidney disease stratification, though it requires complete datasets, which can be challenging in nephrology cohorts.

Similarity network fusion (SNF)

SNF constructs patient similarity networks using each data type before fusing them into an integrated network. This approach has shown utility in capturing complementary kidney disease signals from genomics and histopathology that would be missed by single-data approaches.⁶¹ Its emphasis on patient relationships rather than feature relationships aligns well with precision nephrology goals.

Omics Integrator

By leveraging protein interaction networks as a scaffold, Omics Integrator connects disparate omics signals through known biological pathways. Shved and colleagues applied this approach to identify compounds that could be repurposed for AKI by connecting transcriptomic changes to downstream pathway effects.⁶² The method’s incorporation of prior knowledge helps constrain biological plausibility but depends heavily on the completeness of reference interactomes.

POTENTIAL OF KIDNEY BIOPSY COHORT STUDIES

With the advent of molecular interrogation technologies with single-cell resolution and retention of spatial information and computational techniques to integrate the data, the remaining critical piece was access to human kidney tissue. In contrast to the oncology field, which requires tissue specimens in clinical care to initiate treatment, the field of nephrology often relies on clinician judgment to make a diagnosis. Most clinicians forgo obtaining a kidney biopsy in AKI because most patients will be managed with supportive care. This potentially perpetuated a vicious cycle where clinicians did not obtain biopsies in AKI patients because they were unlikely to alter clinical management, and the lack of access to human kidney tissue precluded the development of novel therapies.

In recognition of these challenges, the NIH in 2017 released a call for applications to develop the Kidney Precision Medicine Project (KPMP). The goals of the KPMP were to ethically obtain biopsies from individuals with AKI and CKD; to identify relevant cell types, pathways, and targets using molecular interrogation data; and to make the data publicly available for use in Atlas. The design of the KPMP has been described in detail previously, and protocols are publicly available on the KPMP website.⁶³ Potential participants with AKI or CKD are approached and consented to a biopsy, which is not necessarily needed for their clinical care. Detailed clinical information is collected regarding the participants, and the biopsy tissue undergoes detailed histopathologic analysis and cutting-edge molecular interrogation. All data are made publicly available in the KPMP atlas.

While research biopsies have been completed in other populations, such as in participants with diabetic kidney disease or glomerular disease,^64,65 we are unaware of any other cohort study to obtain protocol kidney biopsies in individuals with AKI. We anticipate that as KPMP recruitment continues, a robust sample size should enable identification linking clinical phenotypes in AKI with molecular interrogation data.

SYSTEMS BIOLOGY APPROACHES IN KIDNEY DISEASE

Reference Tissue

Careful examination of reference tissue is critical to understanding human disease. Using data from 56 adults, KPMP investigators developed a reference kidney tissue atlas.⁶⁶ The authors demonstrated that using a variety of transcriptomic, proteomic, and spatial metabolomic data provided complementary information regarding human kidney physiology, including sodium transport and energy metabolism. KPMP investigators have also comprehensively interrogated the epigenome from 25 reference samples using CUT&RUN, WGBS, and multiome.⁶⁷ Reassuringly, the various technologies provided overlapping information that validated each other. Moreover, the authors were then able to predict a proximal tubule transcription factor network that regulated the transition from a healthy proximal tubular cell state to an injured proximal tubular cell state, which they then validated with a knock-down model.

Investigators have also sampled other regions of the kidney, including the kidney medulla, papilla, renal artery, and ureter. By integrating snRNAseq, snATACseq, and spatial metabolomic data, investigators identified potentially novel cell types in the renal medulla and papilla, which they termed the thin limb loop of Henle.⁶⁸ Moreover, they demonstrated that cells that cross anatomic regions of the kidney, such as the thick ascending limb, demonstrate unique metabolic profiles, further highlighting the importance of spatial and anatomic localization.

Taken together, these studies demonstrate that molecular interrogation data from human kidney tissue can recapitulate known physiology. Moreover, the data can highlight previously underappreciated aspects of physiology, such as novel cell types in the medulla and papillae.

Chronic Kidney Disease

Further work has been done on CKD and provides insight into molecular mechanisms of CKD progression. The KPMP kidney atlas analyzed scRNAseq and snRNAseq data from 36 diseased individuals (both CKD and AKI) and 45 reference kidneys, which identified altered cell states, basically a cluster of cells from a particular anatomic region with a shared transcriptional signature that suggested they were responding to an insult. For example, adaptive proximal tubular cells (aPT) were found in both AKI and CKD participants and expressed high amounts of injury markers such as VCAM-1 and KIM-1. The transcriptional profile of aPT cells shared many features of its transcriptional signature with “failed to repair” cells, which were identified as cells that underwent maladaptive repair in a murine ischemia-reperfusion injury model.⁶⁹ A high abundance of aPT cells in RNAseq data from the kidney biopsies of individuals with nephrotic syndrome (NEPTUNE cohort) was associated with worse long-term outcomes. These results demonstrate that molecular signatures gleaned from human kidney tissue have prognostic relevance and further help translate findings from reproducible mouse models. Other investigators integrated snRNAseq, scRNAseq, spatial transcriptomic, and snATACseq data from 58 participants with minimal fibrosis and with CKD to identify a transcriptional signature of a fibrotic microenvironment.⁷⁰ In an independent cohort of individuals with CKD, the score of this transcriptional signature was associated with future kidney decline. Taken together, these studies suggest that molecular interrogation data can be used to predict patient outcomes in patients with CKD. With sufficient samples from patients with AKI, it will be possible to apply similar approaches to identify molecular signatures of patients who recover or fail to recover from an AKI event.

Effects of Pharmacologic Treatment

ScRNAseq from kidney biopsies has also identified potential treatment effects, such as SGLT2 inhibitors. While SGLT2 inhibitors have been resoundingly successful at slowing the progression of CKD,^3–5 their mechanism of nephroprotection is not entirely clear. In a cross-sectional study of scRNAseq data from kidney biopsies of youth with T2D with and without SGLT2 inhibitors, investigators were able to infer from the transcriptional data that decreases in mTORC1 activity were one potential impact of the treatment.⁷¹ This finding was further supported by decreased phospho-S6 staining in kidney tissue and by transcriptional data from a murine model of DKD treated with SGLT2 inhibitors. Reassuringly, this finding from human data has also been shown in experimental systems of SGLT2 inhibition,⁷² indicating that interrogation of human kidney tissue yields robust, reproducible findings.

Clinical trials are now recognizing the power of leveraging molecular interrogation data from human kidney tissue and using it to innovate design. For example, the REMODEL study (funded by Novo Nordisk) is obtaining kidney biopsies from individuals pre- and post-treatment with a GLP-1 receptor agonist to identify the molecular mechanisms of nephroprotection.⁷³ The paired study design will help minimize residual confounding. Should potential novel therapies become available for patients with AKI, this clinical trial framework could be adapted for AKI participants.

Focal and Segmental Glomerulosclerosis and Minimal Change Disease

Another area where systems biology approaches have been successfully employed is the identification of endotypes of individuals with Focal and Segmental Glomerulosclerosis (FSGS) and Minimal Change Disease—two causes of nephrotic syndrome where patients exhibit highly variable responses to treatment. Molecular interrogation data suggest alternate trial designs can be used to identify endotypes of patients that respond to treatments. Using RNAseq data from the NEPTUNE cohort of individuals with FSGS, investigators identified three subclusters of patients.⁷⁴ One subcluster, which was identified by increased expression of TNF-α, was at increased risk of progression. This subgroup of patients could be reliably identified by two urinary biomarkers: TNF-α and MCP-1. The results of this study have been a driving impetus behind the NEPTUNE Match program, which is a novel approach to recruit patients for randomized clinical trials.⁷⁵ In this design, patients are assigned a “molecular activity score” and provided with a report card suggesting which available clinical trials are most likely to investigate a pharmacologic agent that is potentially helpful for them. The approach demonstrates that linking molecular interrogation data with noninvasive urinary biomarkers may prove to be a viable approach to identifying endotypes within AKI.

Acute Interstitial Nephritis

One recent area of success in AKI has involved using systems biology approaches to identify novel potential biomarkers for AIN, which is a common cause of AKI in the hospital. In the case of drug-induced AIN, withdrawal of the offending agent and potentially treatment with steroids can improve patient outcomes.^76,77 While nephrologists previously relied on urinary eosinophils to diagnose patients with AIN, a cohort study that compared urine eosinophils to a biopsy diagnosis of AIN actually had limited sensitivity and specificity.⁷⁸ Investigators recently completed an unbiased screen of the urine proteome to identify putative markers of AIN, which they then linked to adjudicated biopsy diagnosis and transcriptional data. This work identified urinary CXCL9 as a potential novel biomarker for the diagnosis of AIN.⁷⁹ While further investigation is needed to ascertain whether this biomarker is suitable for clinical practice, this investigation is an excellent example of how a systems biology approach can help identify novel biomarkers.

Application of Systems Biology Approaches to AKI

Each of these studies, which have mostly been published in the last 5 years, demonstrate remarkable progress in linking molecular interrogation data to phenotypes. In different realms of nephrology, investigators have adopted systems biology approaches to use molecular interrogation data to identify molecular mechanisms that explain variability in patient outcomes, identify molecular changes associated with pharmacologic treatment, and identify novel biomarkers for diagnosis or prognosis. We hope that with the increasing availability of kidney tissue from patients with AKI, these approaches can be adapted for AKI. These efforts will hopefully help expedite the identification of novel drug targets or facilitate the development of novel clinical trial designs.

In the interim, researchers have been successfully leveraging the power of molecular interrogation technologies in murine models of AKI. One of the initial efforts mapped changes in PT cellular trajectories and states using sequential snRNAseq data from murine kidneys after an ischemia-reperfusion injury (IRI) model and identified that a subset of PT cells failed to revert back to a healthy state and instead reached a “fail-to-repair” state (FR-PTC).⁶⁹ FR-PTC expresses high levels of Vcam1, and similar cell states were observed in human CKD kidneys. By obtaining subsequent snATACseq data from the same experiment, investigators identified epigenetic alterations and transcription factors critical to maintain FR-PTC, including the NF-kB and CREB5.⁸⁰ Additionally, the same group subsequently obtained spatial transcriptomic data, which spatially demonstrated that FR-PTCs are near C3+ monocytes.⁸¹

Other investigators identified an inflammatory ICAM1 + PT cell state in nephrectomy specimens from individuals with obstruction, which seemed to play a critical role in the generation of fibrosis.⁸² This inflammatory PT cell state was spatially localized to areas of fibrosis, and the transcription factor family AP-1 may promote the inflammatory cell state. Moreover, in a unilateral IRI model of AKI, inhibition of AP-1 activity ameliorated subsequent interstitial fibrosis. Other investigators have leveraged spatial transcriptomic data from murine IRI models to identify neighborhoods of injured PT cells that interact with fibroblasts via Clcf1-Crlf1 interactions.⁸³

Apart from investigations that have elucidated the role of altered PT cellular states in perpetuating fibrosis, novel molecular interrogation technologies have also enabled insights into the role of other cell types. Using scRNAseq data, investigators compared a unilateral IRI model, which promotes tubular atrophy, and a unilateral IRI model with contralateral nephrectomy, which promotes tubular repair.⁸⁴ Both models showed immune infiltration of macrophages, dendritic cells, and T cells within 7 days of injury. However, after 7 days, the unilateral IRI model had a “second wave” of inflammatory T cells and neutrophils, which resulted in extensive tubular atrophy. Other investigators have used cell isolation techniques to capture the responses of lymphatic endothelial cells to AKI.⁸⁵

POTENTIAL PITFALLS

While systems biology approaches hold great promise to identify which molecular pathways are relevant in human disease, there are several limitations.

Limited Access and Sampling of Human Kidney Tissue from AKI Patients

While human kidney biopsies contain a treasure trove of information and we should consider obtaining kidney biopsies more frequently in standard clinical care, it is unlikely that we will ever rely on kidney tissue for clinical diagnosis and management of AKI like our colleagues in oncology do. Kidney biopsies have risks of bleeding requiring blood transfusions, hematuria, arterio-venous fistula, and even death,⁸⁶ and AKI patients are at an even higher risk for biopsy complications.⁸⁷ Studies such as KPMP have strict safety criteria for participants, so the KPMP AKI cohort may not be representative of the AKI population as a whole. Given the risks associated with kidney biopsy, it is challenging to obtain multiple sequential kidney biopsies over the course of an AKI event as one can with urinary biomarker measurements. Thus, it is still critically important to identify novel biomarkers or determine how to employ previously identified biomarkers effectively.

Typical kidney biopsies sample between 10 to 30 glomeruli, while the average kidney has over one million glomeruli. Thus, a kidney biopsy core may not be representative of the entire pathologic process in a kidney. Upon morphometric analysis of nephrectomy specimens, which obtain, on average, over 200 glomeruli, segmentally sclerosed glomeruli can be found in fewer than 1% of glomeruli.⁸⁸ Statistical modeling based on the frequency of segmental glomerulosclerosis demonstrates that if segmental glomerulosclerosis is present in 10% of glomeruli, a biopsy with 10 glomeruli would only have a sensitivity of 65% for capturing the segmentally sclerosed glomerulus.⁸⁸ Moreover, abnormal glomeruli are not randomly distributed throughout the kidney cortex,⁸⁹ and simulated kidney biopsies of 2D maps of nephrectomy specimens demonstrate that kidney biopsies that miss the capsule are less likely to include abnormal glomeruli.⁹⁰ While spatial heterogeneity of tubular cells has been studied less extensively than that of glomeruli, it is well established that the S3 segment of the proximal tubular cells, which crosses into the renal medulla, is more susceptible to AKI,⁹¹ suggesting that a single biopsy core may not completely capture the pathologic processes in human kidney tissue.

An alternate solution to these challenges is obtaining scRNAseq data from urinary pellets, which provides a transcriptional readout of the cells shed into the urine that are traditionally examined in urine microscopy. ScRNAseq of urinary pellets from individuals with AKI is technically feasible and reflects intrarenal signatures.^92,93 Thus, urine scRNAseq of urinary pellets may be one potential avenue to obtain sequential molecular interrogation data from individuals with AKI. This would also provide the possibility for sequential sampling from individuals with AKI to establish molecular trajectories of recovery.

Access to Reference Tissue

Many computational analyses require comparing diseased tissue to a reference group, but there is nuance regarding what constitutes the “correct” reference group to use. Even in healthy individuals, there are aging-associated changes in kidneys at the molecular, cellular, and structural levels.^89,94,95 Thus, when one is trying to identify the molecular underpinnings of AKI, which often occurs in older patients with preexisting CKD, should one use as a comparator group age-matched patients with similarly low baseline estimated glomerular filtration rates or tissue from a healthy 20-year-old? Additionally, the method of tissue procurement can have a substantial impact on the readout from the molecular interrogation data. For example, nephrectomy specimens, which are a common source of reference tissue, can have a robust ischemia and stress response signature.⁹⁶ While some investigators have obtained kidney biopsies from healthy young adults as reference tissue,⁷¹ this has not been common practice thus far, partially due to ethical concerns. However, given the high scientific value of appropriate control specimens, recent reconsideration of the ethical issues suggests that this is a scientifically feasible and ethical approach.⁹⁷

Limited Phenotypes of Patients With AKI

Currently, patients with AKI are often classified based on the severity of their AKI, which is typically based on their KDIGO stage, duration of AKI, and extent of recovery. All of these phenotypes rely on various definitions of changes in serum creatinine. More nuanced phenotypes would be extremely useful to help better parse molecular data. Thus, further work on identifying alternate phenotypes in AKI is needed, such as changes in tubular secretion.⁹⁸

FUTURE DIRECTIONS

The abundance of molecular interrogation data from human kidney tissue provides unprecedented opportunities for scientific progress in AKI. Kidney-specific platforms have emerged to address domain-unique challenges from individual laboratories⁹⁹—and consortia efforts have emerged to make these data more accessible to the scientific community. These resources integrate genome-scale datasets with nephron-segment resolution, providing specialized visualizations and analysis pipelines tailored to kidney anatomy.¹⁰⁰ Broadly accessible data on platforms with user-friendly interfaces should accelerate the rate of discovery.

Linking molecular interrogation data from human kidney tissue to clinical phenotypes has shown immense scientific value by identifying subgroups of patients who may benefit from a novel treatment, identifying novel biomarkers, and elucidating molecular-based therapies. As the amount of kidney tissue available from individuals with AKI continues to increase, we anticipate that a systems biology approach will help identify novel subgroups of patients, novel biomarkers, and potentially beneficial therapies. By effectively integrating molecular interrogation data with human clinical phenotypes, it should facilitate the identification of novel drug targets and biomarkers and can help revitalize clinical trial design for patients with AKI (Figure 2). While translational studies with human kidney tissue are likely to be a key component of providing novel management options, all components of the scientific pipeline from “bench to bedside” will be critical to ensure novel therapies become available to help patients with AKI.

Figure 2.

The systems biology process toward targeted treatments in nephrology. The emergence of large, detailed, multilevel biological and clinical data from national databases, cohort studies, and trials provides critical pieces needed to improve the pathway to targeted treatments in nephrology. Reproduced with permission from Schaub et al.³²