Spatiotemporal transcriptomic analysis during cold ischemic injury to the murine kidney reveals compartment-specific changes

Abstract

Kidney transplantation is the gold standard treatment strategy for end-stage renal disease. Deceased donor kidneys usually undergo cold storage until kidney transplantation, leading to cold ischemia injury that may contribute to poor graft outcomes. However, the molecular characterization of potential mechanisms of cold ischemia injury remains incomplete. To bridge this knowledge gap, we leveraged spatial transcriptomics technology to perform full transcriptome characterization of cold ischemia injury (0–48 hours) using a murine model. We developed a computational workflow to identify spatiotemporal transcriptomic changes that accompany the injury pathophysiology in a compartment-specific manner. We identified potential metabolic reprogramming preferentially within the kidney inner medulla displaying strong oxidative phosphorylation signature in an ischemic environment. We found commonalities between the spatiotemporal transcriptomic presentation of cold ischemia and warm ischemia‒reperfusion injury, including an induction of an anti-viral like immune response throughout the renal tissue. Altogether, these systems-level biological insights enabled by our full transcriptome temporal characterization unveil a molecular basis for how cold ischemia injury may negatively affect kidney outcomes. Moreover, our spatial analyses highlight pathological developments deep within the renal tissue, suggesting potential opportunities for new insights beyond biopsy-focused superficial tissue examinations. We also developed an interactive online browser at https://jef.works/vitessce-cold-ischemia/ to facilitate exploration of our results by the broader scientific and clinical community.

Introduction:

Kidney transplantation remains the gold standard treatment strategy for end-stage renal disease, relying on viable allografts from either living or deceased donors. Although live donor kidneys usually have better short and long-term graft outcomes, their limited availability has contributed to a major push to expand the deceased donor pool (1). A key challenge with deceased donor kidneys is the cold ischemia time (CIT) —the period during which the kidneys are preserved in cold storage after procurement and before transplantation (2–4). In this period, kidneys are flushed with cryopreserving solution and kept on ice to minimize cellular injury and preserve the organ viability during the transit between transplantation centers. The kidneys experience cold ischemia injury (5) during this phase, which can lead to delayed graft function and poor graft outcomes (6).

The pathophysiology of cold ischemia induced renal injury is incompletely understood. While prior studies (7–9) have sought to characterize processes that are believed to accompany cold ischemia and lead to cellular injury, they usually do not recapitulate the complex structure of kidney tissue. Moreover, these studies primarily investigate only a few targeted cold ischemia related pathological processes and their associated biomarkers, potentially overlooking other clinically relevant pathological developments. Additionally, findings from several large cohort, multi-variate retrospective studies (6, 10–13) lack consensus on a standard clinically acceptable CIT for kidney allografts. Collectively, this highlights the opportunity for a more thorough molecular characterization of cold ischemia injury of the entire kidney to elucidate the underlying biological processes that may contribute to poor prognosis with increasing CIT. High-throughput sequencing data-driven approaches can surveil the genome-wide molecular landscape during disease pathogenesis. In particular, advances in spatial transcriptomics (ST) technologies (14–16) now enable full transcriptome spatially resolved profiling of gene expression patterns. Such a data-driven approach can facilitate the generation of hypotheses towards the identification of potential biomarkers of prolonged cold ischemia injury and therapeutic targets.

We therefore leveraged 10x Visium ST technology to perform full transcriptome profiling of murine kidneys subject to varying durations of cold ischemia to characterize spatiotemporal molecular changes across typical durations of cold ischemia in a deceased donor renal transplant setting (Fig 1A). Our integrative spatiotemporal computational analyses delineated trends that facilitated the identification of differential spatiotemporal molecular dynamics in a compartment-specific manner, especially pertaining to aberrant metabolic reprogramming, to shed light on how prolonged cold ischemia may contribute to adverse transplant outcomes. An interactive browser with our data and results is available to facilitate exploration at https://jef.works/vitessce-cold-ischemia/

Figure 1. Computational workflow for analysis of multiple murine kidney spatial transcriptomics datasets.

(A) Spatial transcriptomics datasets were generated from murine kidneys subjected to cold ischemia injury of various durations: 0, 12, 24 and 48 hours. Thereafter, the datasets were integrated with previously published ST datasets and subjected to computational analysis. (B) tSNE plot demonstrating sample-specific effects within the integrated ST datasets. (C) Harmonized tSNE plot leading to a shared embedding across all datasets. (D) Clustering facilitated identification of shared transcriptionally distinct compartments across all the datasets, namely cortex, outer medulla (O.Medulla), inner medulla (I.Medulla) and other. (E) Spatial plots visualizing the anatomic location of the different transcriptionally-defined compartments within the renal tissue specimens belonging to different experimental groups like cold ischemia injury and ischemia-reperfusion injury. Created with BioRender.com

Results

1. Computational workflow identifies shared spatial compartment across multiple murine kidney spatial transcriptomics datasets.

To establish a quantitative molecular understanding of cold ischemia across the entire kidney, we applied 10x Visium to perform full transcriptome spatial transcriptomic (resolution: 55 μm pixels/spots) profiling of cold ischemia kidney injury in a murine model. The model introduced 0, 12, 24 and 48 hours of cold ischemia injury that represent typical durations of cold ischemia in a deceased-donor renal transplant setting (Fig. 1A, S1A–D). Quality control filtering resulted in 19454 unique gene species detected across 12530 pixels belonging to the cold ischemia kidney injury tissue specimens (4.18 ± 0.29 transcripts/spot and 2.6 ± 0.96 spots/gene (log₁₀ scale)) (Fig. S1E–L). To facilitate comparative analysis across the cold ischemia kidney specimens as well as with kidney specimens from previously published studies (17, 18)(especially, ischemia-reperfusion injury as a model for acute kidney injury (AKI)), we sought to computationally integrate all 17 ST datasets (44816 pixels) to identify transcriptionally similar tissue compartments via graph-based clustering (Fig. 1B). However, such unified clustering was driven by batch effects as well as unique biological processes resulting in condition and sample-specific clusters, thereby hindering the identification of shared features. We therefore performed batch correction using Harmony (17) (Fig. 1C) and re-clustered to identify shared compartments across all datasets (Fig. 1D). Due to lack of single cell resolution and concerns for variations in cell-type proportions confounding small clusters, we opted for coarser clustering. Ultimately, this computational workflow led to identification of distinct compartments which we labeled as cortex, outer medulla (O.Medulla), inner medulla (I.Medulla) based on their relative anatomical location within all the kidney tissue specimens (Fig. 1E, Fig. S1M,N). Review of the corresponding H&E images by an expert renal pathologist validated our labels, confirming congruence between the transcriptional and histological assessments. Remaining tissue which was not part of the previous three compartments was labeled as other, likely corresponding to the transition from medulla to pelvis based on pathologist’s evaluation, and was not used for any analyses henceforth. We were then able to perform differential expression analysis across compartments of all ST datasets (Fig. S2) to identify marker genes of the different compartments, which were found to be consistent with known annotations of associated functional units. For instance, markers for functional units like Slc14a2 for descending thin limb of Loop of Henle (LOH) and Slc5a3 for thick descending limb of LOH (20) were identified as one of the top gene markers for inner medulla compartment. Similarly, proximal tubule S3 segment marker Slc22a7 (21) was identified as one of the top gene markers for outer medulla while pan-proximal tubule marker Slc34a1 and proximal convoluted tubule marker Slc5a2 were associated with the cortex (18) (Fig. S3). This observation was consistent across all datasets. As such, our computational workflow identified spatially distinct compartments, thereby enabling us to further analyze transcriptomic temporal dynamics in the different compartments of kidney tissue.

2. Compartment-specific analysis characterizes spatially confined gene expression dynamics in cold ischemia kidney injury

To characterize the temporal transcriptional dynamics associated with cold ischemia injury, we leveraged linear regression modeling to identify genes with expression patterns that significantly change with duration of cold ischemia injury (i.e., CIT). Briefly, after compartment identification within the kidney samples, we applied linear regression to each compartment separately. Such linear regression modeling identified many genes that decreased gradually in expression magnitude with longer CIT i.e., exhibited a downward temporal trend (number of genes in inner medulla: 2413 (12.4%), outer medulla: 2227 (11.44%) and cortex: 1591 (8.17%))(Table S1). Surprisingly, given lack of blood flow, many other genes gradually increased in expression magnitude with longer cold ischemia time i.e., exhibited an upward temporal trend (inner medulla: 922 (4.7%), outer medulla: 1344 (6.91%) and cortex: 1683 (8.65%))(TableS2). Comparing across compartments, we found genes exhibiting temporal trends either in a compartment-agnostic or compartment-specific manner. For compartment-agnostic trends, 436 (2.24%) genes exhibited an upward temporal trend across all compartments whereas 471 (2.42%) exhibited a downward temporal trend across all compartments (Table S3). For instance, Fth1 (Fig. 2A), Ig1p7, and Umod were upregulated over time across all compartments whereas Kap (Fig. S4A), Gpx3, and Acsm2 were downregulated over time across all compartments. In comparison, there were comparatively more genes that exhibited a compartment-specific temporal trend (upward trend: inner medulla: 275 (1.41%), outer medulla: 394 (2.02%), cortex: 702 (3.61%)(Table S4); downward trend: inner medulla: 1391 (7.15%), outer medulla: 966 (4.97%), cortex: 579 (2.98%) (Table S5)). For example, Ranbp3l was uniquely upregulated over time in the inner medulla (Fig. 2B), whereas Mep1a was uniquely upregulated in the outer medulla (Fig. 2C), and Slc34a1 in the cortex (Fig. 2D). In contrast, Egf was uniquely downregulated over time in the inner medulla, whereas Fxyd2 was uniquely downregulated in the outer medulla, and Aqp1 in the cortex (Fig. S4B–D). Within each compartment, we observed a great variation in the magnitude of gene expression changes (i.e., linear regression slopes) with most genes demonstrating subtle changes (smaller slopes i.e., normalized scores<0.5 (Fig. 2E)) with increasing CIT while a certain number of genes changed remarkably (larger slopes) such as Fth1, Umod, and Kap. (Fig. 2E, F–H, Table S1,2). To investigate whether these trends covary between compartments, we compared the regression slopes of genes in the different compartments. Pearson correlation analysis suggested that the cortex and outer medulla (Pearson correlation coefficient, PCC= 0.94, p<0.05, R²= 0.88) (Fig. 2F) are more similar to each other in terms of temporal gene expression changes as compared to the inner medulla (cortex vs inner medulla: PCC= 0.63, p<0.05, R²= 0.4) (Fig. 2G); outer medulla vs inner medulla: PCC= 0.68, p<0.05, R²= 0.46) (Fig. 2H)). As such, our compartment-specific temporal analysis identified compartment-agnostic as well as compartment-specific gene expression trends.

Figure 2. Compartment-specific temporal transcriptional dynamics in cold ischemic kidneys.

(A) Spatial gene expression plots and corresponding line plots demonstrating spatiotemporal dynamics of representative genes that demonstrated an upward temporal trend during the cold ischemic phase (0–48 hours). For instance, (A) Fth1 was upregulated in all the compartments whereas (B) Ranbp3l, (C) Mep1a and (D) Slc34a1 were uniquely upregulated in the inner medulla, outer medulla and cortex compartments of cold ischemic injury renal tissue respectively. (E) Violin plots highlight the skewed distribution of magnitude of transcriptional changes (linear regression slopes) within the different compartments of CIS renal tissue. Scatterplots depicting the correlation analysis of the temporal transcriptional dynamics between the different compartments: (F) cortex vs outer medulla, (G) cortex vs inner medulla and (H) outer medulla vs inner medulla of the CIS kidney tissue. Genes exhibiting covarying (green) or divergent (purple) temporal trend within these compartments can be visualized along with their Pearson correlation coefficients (PCC) and linear regression slopes (blue line). Genes which exhibited strong temporal changes like Fth1, Spp1, Umod, Ig1p7, Kap within these compartments have been highlighted red.

3. Pathway characterization of identified spatiotemporal trends suggest compartment-specific metabolic reprogramming

To better understand the molecular pathways associated with genes that demonstrated significant time dependent expression with increasing CIT, we performed gene set enrichment analysis on the temporally upregulated and temporally downregulated genes with longer CIT obtained through our previous linear regression analysis ranked by slope in a spatial compartment specific manner (Methods). We identified significantly enriched molecular pathways and evaluated whether they exhibited covarying or divergent trends across different compartments based on their normalized enrichment scores (NES) (Fig. 3A–C, Table S6, S7). Within the renal inner medulla, we observed significant positive enrichment (NES>0) of energy metabolism pathways amongst several enriched biological pathways (Fig. 3A), suggesting that genes within these pathways were being temporally upregulated in this compartment during cold ischemia injury. For instance, oxidative phosphorylation (OXPHOS) pathway (ID: mmu00190) thermogenesis (ID: mmu04714), and citrate cycle (TCA cycle)(ID: mmu00020) were positively enriched primarily in the inner medulla. In contrast, these pathways demonstrated a weak positive or negative enrichment within the outer medulla and cortex (Fig. 3E–G), suggesting that genes within these pathways were not being consistently temporally upregulated and even temporally downregulated in these compartments during cold ischemia injury. Within the renal inner medulla, the leading-edge genes of the positively enriched energy metabolism pathways were related to various processes of mitochondrial machinery like Atp5b for ATP synthesis, Cox6a1 (OXPHOS pathway), Ppargc1a for mitochondrial biogenesis (Thermogenesis pathway) and Mdh1 for TCA cycle pathway (Fig. 3H–K). (The complete list of leading-edge genes associated with the corresponding enriched molecular pathways are summarized in Table. S6). We performed compartment-specific qPCR to validate these observations using 4–5 animals per time point (Methods). Although we observe a comparatively higher expression of the energy metabolism related genes like Atp5b, Cox6a1 and Ppargc1a (Fig. 3L–N) within the medulla (inner medulla and outer medulla combined) as compared to cortex throughout the cold ischemic phase, confirming compartmental differences, a consistent temporal trend was lacking. We speculate that this inconsistency might be resulting from the pooling of the outer and inner medulla tissue into a single group (called medulla) due to the limitations in manual surgical segmentation of the two compartments. Of note, in the normal kidney, the inner medulla, being in an oxygen lean environment, is known to primarily rely on anaerobic respiration while the cortical region (proximal tubules) is known to utilize OXPHOS owing to their high metabolic demand (22). However, the enrichment of OXPHOS and TCA cycle specifically within the inner medulla contrasts with this, suggestive of a potential metabolic reprogramming in response to the cold storage conditions.

Figure 3. Compartment-specific enrichment of molecular pathways within cold ischemic kidney tissue.

Bar plots highlighting all the enriched molecular pathways that demonstrated similar temporal dynamics (enrichment: positive (red), negative (cyan)) within the corresponding (A) inner medulla (B) outer medulla and (C) cortex of cold ischemic injury renal tissue. Running enrichment score plots highlighting differential enrichment of molecular pathways related to energy metabolism: oxidative phosphorylation, citrate cycle (TCA cycle) and thermogenesis within the (D) inner medulla, (E) outer medulla and (F) cortex of the CIS renal tissue. Spatial gene expression plots and corresponding line plots demonstrating spatiotemporal dynamics of representative genes associated with the above enriched energy metabolism related pathways: (H) Atp5b and (I) Cox6a1 (oxidative phosphorylation); (J) Ppargc1a (thermogenesis) and (K) Mdh1 (TCA Cycle). Quantitative comparison of qPCR derived mRNA transcript levels of (L) Atp5b, (M) Cox6a1 and (N) Ppargc1a between the cortex and medulla (inner medulla + outer medulla) of CIS renal tissue.

To understand how OXPHOS could be sustained within a nutrient limited cold ischemic environment, we scrutinized significantly enriched metabolic pathways that can potentially generate metabolic substrates to drive the process. While most metabolic pathways such as fatty acid degradation (ID: mmu00071; genes: Cpt1a, Acox1, etc. ), tryptophan metabolism (ID: mmu00380, genes: Haao, Kmo, Gcdh, etc. ), and valine, leucine and isoleucine metabolism (ID: mmu00280 ; genes: Dbt, Mccc1, Aldh6a1, etc. ) were negatively enriched within all the renal compartments, glycolysis/ gluconeogenesis (ID: mmu00010, genes: Hk1, Pdha1, Pdhab, Pkm, Dlat, etc.) was positively upregulated only within the inner medulla (Fig. S5A–E). Temporal upregulation of genes like Pdha1, Pdhb, Dlat, etc. within the inner medulla suggests potential coupling between glycolysis and TCA cycle via Acetyl-CoA intermediate (Fig. S5F–H). Consistently, we observed positive enrichment of HIF-1 signaling pathway (ID: mmu04066; genes: Mapk1, Akt1, Vegfa, etc.) within the inner medulla along with the temporal upregulation of glucose transporter gene Slc2a1 (Fig. S5I). While the heightened glycolytic signature is consistent with previous studies demonstrating association between HIF-1 pathway and glycolytic metabolism (23, 24), the co-enrichment of glycolytic and OXPHOS pathways within the cold ischemic renal inner medulla seems aberrant.

Unlike energy metabolism pathways which exhibited a compartment specific gene regulation, we observed several pathways that were regulated in a compartment agnostic manner indicative of pan-tissue effects. For instance, pathway related to reactive oxygen species (ID: mmu05208; genes: Sod2, etc.) was positively enriched across all compartments suggestive of oxidative stress (Fig. 3A–C, S6). Positive enrichment of pathways related to protein folding and quality control like Alzheimer disease (mmu05010), Huntington disease (mmu05016), prion disease (mmu05020), amyotrophic lateral sclerosis (mmu05014), and neurodegeneration (mmu05022), involving shared genes like Hspa5, Creb3, Prnp, Psmd2, Psmd3, Psmd14, etc (Fig. 3A–C). was observed. Finally, antigen processing and presentation pathway (ID: mmu04612, genes: Ctsb, B2m, Psme, Cd74, H2-Aa, etc.,) involving some genes related to major histocompatibility complex (MHC)-I and II was positively enriched across all compartments suggesting the buildup of an immunogenic environment in response to cold ischemia injury (Fig. S7A–H). Collectively, both compartment-specific metabolic reprogramming and tissue-side pathological molecular changes were identified within cold ischemic renal tissue.

4. Comparison of cold ischemia and warm ischemia-reperfusion-induced AKI identifies overlapping molecular dynamics

To better understand the pathogenesis of renal cold ischemic injury, we next sought to compare its spatiotemporal transcriptomic dynamics with another ischemic injury setting. We therefore repeated the spatiotemporal analysis with a set of previously published ST datasets from a well-characterized murine model of warm ischemia-reperfusion-induced acute kidney injury (AKI) at various timepoints postinjury covering the peak phases of injury response (sham to 48 hours)(17). Our temporal analysis recapitulated putative AKI injury markers like Havcr1, Lcn2 and Spp1 across all the compartments with Spp1 demonstrating the sharpest change (linear regression slope) during the acute injury phase (Fig. S8A–C). Cross-compartmental analysis highlighted similar temporal transcriptional dynamics between the AKI cortex and inner medulla (PCC=0.79, p<0.05, R²=0.63) while the outer medulla and inner medulla were comparatively less similar (PCC=0.43, p<0.05, R²=0.19) (Fig. S8D–F).

We next compared the temporal transcriptional dynamics between the corresponding compartments of the two injury models (i.e., CIS vs AKI). In contrast to CIS, in the warm ischemia-reperfusion-induced AKI model, while compartment-specific correlation analysis indicated a weak correlation between the inner medulla (PCC=0.04, p>0.05; R²~0 (0.0019)) (Fig. 4A), comparatively stronger correlations were observed between the outer medulla (PCC=0.8, p<0.05; R²=0.64) (Fig. 4B) and cortex (PCC=0.5, p<0.05; R²=0.25) (Fig. 4C) of both the injuries. Interestingly, we observed Spp1 upregulation across all compartments in both injury models with the inner medulla exhibiting the largest expression magnitude (Fig. 4A–C).

Figure 4. Compartment-specific comparison of transcriptional dynamics between the CIS and AKI renal tissue.

Scatter plots depicting the correlation analysis of the temporal transcriptional dynamics between the corresponding (A) inner medulla, (B) outer medulla and (C) cortex tissue compartments of the two renal injury models i.e., cold ischemic injury (CIS) and warm ischemia-reperfusion injury (AKI) kidney tissue. Genes exhibiting covarying (green) or divergent (purple) temporal trend within these compartments can be visualized along with their Pearson correlation coefficients (PCC) and linear regression slopes (blue line). Bar plots highlighting all the enriched molecular pathways that demonstrated similar temporal dynamics (enrichment: positive (red), negative (cyan)) within the corresponding (D) inner medulla, (E) outer medulla and (F) cortex of CIS and AKI renal tissue. Running enrichment score plots of positively enriched viral response like pathways like Epstein-Barr virus infection (red) and Hepatitis C (black) pathways within the corresponding (G) inner medulla, (H) outer medulla and (I) cortex of CIS and AKI renal tissue. (J) Running enrichment score plots of positively enriched complement pathway within the corresponding outer medulla of CIS and AKI renal tissue. Spatial gene expression plots and corresponding line plots demonstrating spatiotemporal dynamics of genes associated with the positively enriched Epstein-Barr virus infection pathway like (K) B2m, (L) H2-Aa, (M) N[b2; and complement pathway like (N) C1qa within both the CIS and AKI renal tissue.

Thereafter, we performed gene set enrichment analysis to characterize pathways impacted in AKI as we did with CIS (Table S8, S9). In contrast to CIS, we observed that most molecular pathways demonstrated similar temporal trends across all the compartments (i.e., covarying pathways) with no divergent trends observed (Fig. S9A–F). Within the covarying pathways, pathways associated with energy metabolism like oxidative phosphorylation, citrate cycle (TCA cycle), fatty acid degradation, etc., demonstrated negative enrichment. In contrast, positive enrichment of both reparative processes like DNA replication (ID: mmu03030; genes: Pold4, Pcna, Mcm2, Mcm7, Lig1, etc.), PI3K-Akt signaling pathway (ID: mmu04151; genes: Spp1, Hras, Cdk4, Cdk6, Pdgfra, Pdgfrb, etc.) as well as pathological processes like apoptosis (ID: APOPTOSIS; genes: Pea15a, Dap, Bax, Ifngr1, Anxa1, etc.), epithelial-to-mesenchymal transition (or EMT) (ID: EPITHELIAL_MESENCHYMAL_TRANSITION; genes: Spp1, Col1a1, Acta2, Fn1, Vim, etc.), endoplasmic reticulum stress (ID: UNFOLDED_PROTEIN_FOLDING; genes: Hsp90b1, Eif4ebp1, Calr, Eef2, Spcs3, etc.), immune response (ID: ALLOGRAFT_REJECTION; genes: Ikbkb, Ifngr1, Ifng2, Stat1, Ccr1, Ccr2, Tlr1, Ncf4, etc.), etc., was observed, suggesting a diseased environment undergoing tissue repair. Of note, positive enrichment of coagulation pathway (ID: COAGULATION; genes: Clu, Fgg, Rac1, Thbd, Plat, etc.) and complement pathway (ID: COMPLEMENT; genes: C1qa, C1qc, C3, etc.) hinted at potential endothelial dysfunction/vascular remodeling and immune-complex formation activating the complement (classical) pathway during reperfusion. Moreover, the enrichment of a multitude of viral response pathways such as hepatitis C (ID: mmu05160), Epstein-Barr virus infection (ID: mmu05169), and human papillomavirus (ID: mmu05165), involving shared genes like Irf3, Stat1, Ifnar1, Ifnar2, Bax, Bak1, etc., highlighted a heightened immunogenic environment. These findings suggest that AKI injury may induce similar pathological events across the whole renal tissue in a more compartment-agnostic manner compared to CIS. Lastly, processes like apoptosis, EMT, complement, vascular dysfunction, immune response are known to be associated with ischemia-reperfusion injury (25) further validating the ability of our computational workflow in identifying relevant temporal gene and pathway changes that accompany the pathogenesis of diseases like AKI.

Finally, we compared the enriched pathways between the two injury models in a compartment-specific manner (Fig. 4 D–F). For compartment-specific pathways exhibiting divergent trends between the two injury models, most were confined to the inner medulla. For instance, pathways related to glycolysis/gluconeogenesis (ID: mmu00010; genes: Hk1, Aldoa, Dlat, Pdha1, Pdhb, etc.), 2-oxocarboxylic acid metabolism (ID: mmu01210; genes: Got2, Ogdh, Idh2, Idh3b, etc.), and TCA cycle (ID: mmu00020; genes: Mdh1, Mdh2, Sdha, Suclg2, etc.) were positively enriched only within the CIS inner medulla while negatively enriched within the AKI inner medulla. This finding is consistent with the weak correlation observed between the inner medulla earlier (Fig. 4A).

Given that the AKI datasets belonged to tissue specimens from female C57BL/6 mice while the CIS datasets were from male C57BL/6 mice tissue, we next sought to evaluate whether these compartment-specific divergent transcriptional dynamics were driven by sexual dimorphism. To investigate this, we analyzed normal control kidneys (CTRL) and 24-hours warm ischemia reperfusion injury kidney tissues (AKI24) from male C57BL/6 mice (18). First, we observed that the presentation of AKI maker genes like Havcr1, Lcn2 and Spp1 within the 24-hour male (AKI24) kidney tissues was very similar to that of female kidney specimens (AKI) at 12 and 48 hours (Fig. S8G–I). Next, we performed differential expression analysis between inner medulla compartment belonging to the CTRL and AKI24 kidneys. We observed that the leading genes: Hk1, Dlat, Pdha1, Pdhb (glycolysis/gluconeogenesis); Got2, Ogdh, Idh2, Idh3b (2-oxocarboxylic acid metabolism); Mdh1, Mdh2, Sdha, Suclg2 (TCA cycle) associated with the divergent pathways identified earlier (CIS inner medulla vs AKI inner medulla), exhibited similar expression trend during AKI injury within the inner medulla of these male kidneys as what was previously identified in the female kidneys. In particular, genes previously identified to be downregulated over time within the inner medulla in AKI in the female kidneys were again identified as downregulated within the inner medulla of AKI24 compared to CTLR in the male kidneys. Likewise, genes previously identified to be upregulated in the female kidneys were such as AKI markers Havcr1, Lcn2 and Spp1, were again identified as upregulated in the male kidneys (Fig. S8J). The only exception was Aldoa demonstrating a slight potential divergence upregulation. The above observations highlight that, overall, the compartment-specific divergent transcriptional dynamics observed between CIS and AKI are likely not driven by sexual dimorphism.

In addition, we identified several pathways that displayed similar trend across all or specific compartments suggesting that the two injuries share some commonality at a transcriptomic level. For instance, metabolic pathways like fatty acid degradation (ID: mmu00071; genes: Cpt1a, Ehhadh, etc.,), tryptophan metabolism (ID: mmu00380; genes: Aadat, Gcdh, etc.), valine, leucine and isoleucine metabolism (ID: mmu00280; genes: Dbt, Aldh6a1, etc.) were negatively enriched (Fig. 4D–F) while viral response pathways like Hepatitis C (ID: mmu05160; genes: Tnfrsf1a, Mavs, Tbk1, Stat1, etc.) and Epstein-Barr virus infection (ID: mmu05169; genes: B2m, Mavs, Cxcl10, Stat1, H2-D1, H2-Aa, N[b2, etc.) were positively enriched in all the compartments (Fig.D–F, G–I, K–M) across both injury models. Intriguingly, we also observed positive enrichment of complement pathway within the outer medulla of both the injuries (Fig. 4J, N). Collectively, the transcriptomic presentation of both the injuries shared many molecular trends across evaluated compartments, highlighting potential similarities in injury response across these distinct ischemic contexts and presenting implication in the poor clinical outcome associated with longer CIT.

5. Metabolic reprogramming mediated mitochondrial dysfunction may be implicated in cold ischemia injury

To summarize, our temporally-resolved spatial transcriptomics data and analyses have identified potential compartment-specific metabolic changes that arise in murine kidneys with increasing durations of cold ischemia. Notably, we identified two striking trends: (1) a metabolic switch to an oxygen demanding OXPHOS pathway within a hypoxic environment particularly within the inner medulla of the cold ischemic renal tissue and (2) concomitant pan-tissue downregulation of most metabolic intermediate generating pathways (fatty acid degradation, tryptophan metabolism, valine, leucine and isoleucine metabolism, etc.) that can provide metabolites to drive OXPHOS (Fig. 5). These metabolic intermediates normally serve as substrates for the tricarboxylic acid (TCA) cycle, which in turn produces high-energy electron carriers (e.g., NADH and FADH₂) that drive the electron transport chain (ETC) and facilitate ATP generation via OXPHOS. The simultaneous upregulation of OXPHOS along with the downregulation of upstream metabolic intermediate pathways that supply its substrates thus presents a physiological paradox. We speculated that downregulation of metabolic intermediate pathways could reduce the production of high-energy electron carriers that fuel the ETC. Reduced ETC activity may lead to a loss of proton gradient across the mitochondrial membrane and may be a source of mitochondrial dysfunction. Mitochondrial dysfunction has been implicated in AKI due to the increased oxidative stress generated in the damaged mitochondria of proximal tubule cells and has been shown to trigger the onset of injury responses such as mitochondrial membrane depolarization-induced mitophagy (26–28).

Figure 5. Summary of prolonged cold ischemia injury induced compartment-specific metabolic changes in murine kidneys.

(A) Normal kidney (inner medulla): The inner medulla of normal kidney relies primarily on glycolysis pathway (black dashed lines) to (a) generate ATP to meet its low metabolic demand. (B) Cold ischemic kidney (inner medulla): During cold ischemia injury, molecular pathways related to energy metabolism like (a) OXPHOS and (b) TCA cycle were positively enriched (red arrows). Enrichment of (c) glycolysis pathway and (d) glucose transporter gene Slc2a1 were positively enriched and upregulation of genes like Dlat, Pdha1, Pdhb suggested (e) coupling of glycolysis and TCA cycle. Finally, (f) negative enrichment (blue arrows) of fatty acid oxidation (FAO) pathway (blue dashed lines) was observed. Simultaneous positive enrichment of OXPHOS and glycolysis with negative enrichment of fatty acid oxidation suggests aberrant metabolic reprogramming with the inner medulla of cold ischemic kidneys. (C) Normal kidney (cortex): The cortex of normal kidney relies primarily on OXPHOS to generate (a) ATP to meet its high energy demand with FAO (black dashed lines) providing the required metabolic intermediates to drive OXPHOS. (D) Cold ischemic kidney (cortex): Both (a) OXPHOS and (b) TCA cycle as well as (c) FAO were negatively enriched during cold ischemia injury. This trend was also observed within the cortex of AKI renal tissue wherein OXPHOS, TCA cycle and FAO were observed to be negatively enriched. (Different colors represent different compartments of the renal tissue: inner medulla (green), outer medulla (blue) and cortex (red)). Created with BioRender.com

To check for mitochondrial dysfunction during cold ischemia injury, we probed for one of the putative effector molecules involved in mitophagy, PTEN-induced kinase 1 (Pink1), which is known to be upregulated in response to mitochondrial distress and initiate the mitophagy faciliated clearance of damaged mitochondria (29). Using our ST data, we observe that Pink1 is temporally upregulated with increasing duration of cold ischemia across all compartments (Fig. S10A), though pathway analysis identified positive enrichment of mitophagy (ID: mmu04137, genes: Pink1, Becn1, Tbk1, Tfeb, etc.) as unique to the CIS inner medulla (Fig. S10B–E)). To further validate these transcriptional trends, we again performed qPCR. Consistent with our ST findings, qPCR revealed that Pink1 is upregulated with increasing CIT in the medulla (Fig. S10F). Consistent with our pathway analysis, Pink1 expression within the cortex did not exhibit the same upregulation with prolonged CIT. In general, this spatial divergence in Pink1 expression suggests that mitophagy may be differentially regulated across renal compartments during cold ischemia injury. Collectively, these finding point towards a potential dysfunction in the mitochondrial machinery which might be the result of a compartment-specific aberrant metabolic reprogramming within a cold ischemic environment.

Discussion

Long durations of cold preservation storage results in cold ischemia injury and contributes to worse graft outcomes (6). However, the molecular events during the pathogenesis of this injury are incompletely understood, particularly in the medulla of the kidney less amenable to safe percutaneous or wedge biopsy sampling. We therefore performed a full transcriptome spatial characterization of cold ischemia in murine kidneys (0–48 hours) using ST with 10x Visium. Using a data driven computational analysis approach, we identified temporal regulation of genes within transcriptionally distinct compartments of the kidney tissue, which allowed us to study regulation of genes and associated molecular pathways within the individual compartments of the cold ischemia tissue. We further extended this workflow to enable comparison with early injury response in warm ischemia-reperfusion-induced AKI.

Our analysis identified several genes that demonstrated significant time dependent regulation within the different compartments of cold ischemic kidneys. These genes were associated with a wide range of processes but importantly energy metabolism, oxidative stress and immune response. In general, all the compartments elevated their expression of genes related to antigen presentation protein complexes (i.e., MHC-I, MHC-II) and displayed heightened molecular signatures of an immunogenic environment reminiscent of a response to viral infection. While metabolic pathways were downregulated across all compartments, the inner medulla unexpectedly demonstrated elevated expression of OXPHOS pathway. The upregulation of OXPHOS pathway within the oxygen lean renal inner medulla, which relies on anaerobic respiration, suggests potential metabolic reprogramming. However, the limited availability of oxygen along with downregulation of essential metabolic substrates to drive OXPHOS contributes to oxidative stress mediated mitochondrial dysfunction and potentially the subsequent induction of an allo-immunogenic environment.

Comparative analysis between the cold ischemia injury and warm ischemia-reperfusion-induced AKI settings in a compartment specific manner identified many covarying as well as divergent trends. Both the injured tissue displayed elevated molecular signatures of viral response over time indicating the presence of an immunogenic environment, and a marked decrease in metabolism of many relevant substrates. Moreover, cross-compartment correlation analysis highlighted that warm ischemia-reperfusion-induced AKI evoked similar transcriptional response across all compartments while cold ischemia induced molecular changes were more compartment specific. This was particularly evident within the spatiotemporal molecular presentation of the OXPHOS pathway between the two injuries. These observations suggest that even though the two injuries have different etiologies and involve distinct molecular pathways within the different tissue compartments, they may still share some commonality in their spatiotemporal transcriptomic presentation.

Although our current study brings in new molecular insights into the pathogenesis of cold ischemic injury within the kidney, there are limitations which may be addressed in future studies. First, the lack of single cell resolution in our spatial transcriptomic profiling limits our understanding of the cell-type specific effects within the different compartments during cold ischemia injury. Single-cell resolution ST platforms like Xenium (30) or MERFISH (31) could be utilized to perform such cellular characterization, though they are limited in the number of genes they can profile simultaneously and therefore should be informed by full-transcriptome characterizations such as this study. Next, our findings rely on observations made from computational analysis of ST data derived from a single sample per timepoint (i.e., n=1 per cold ischemia duration) experimental design. We employed qPCR with a larger animal cohort (n=4–5 per timepoint) to check the validity of key findings from CIS data. Moreover, results from our AKI analyses were consistent with existing literature related to acute kidney injury despite also being limited to a single sample per timepoint. Although this underscores the validity of the computational workflow, the experimental rigor would benefit from a larger sample size. We also acknowledge that static cold storage injury likely has differences from newer storage approaches like hypothermic pulsatile perfusion and normothermic machine perfusion.

We acknowledge that our comparison of CIS (in male mice) and AKI (in female mice) datasets is confounded by potential sexual dimorphism. However, we note that known sexual dimorphism in renal tissue has been primarily confined to the cortical region (with strong differences observed in the S3 segment of proximal tubule) of the kidney (21), whereas the differences we observed were within the inner medulla. Still, the effect of sexual dimorphism on the response to cold ischemia injury presents an opportunity for future investigation.

Finally, there are logistical differences between the current experimental design and a clinical scenario wherein deceased donor kidneys undergo nephrectomy first and perfused ex vivo using renal arteries whereas the murine kidneys in the current study were perfused through the hepatic portal vein first and harvested thereafter. Given that murine arteries present size-related challenges during perfusion, we anticipate large animal models like porcine models can be strategized to recapitulate the clinical scenario more accurately in the future.

We believe that these findings will serve as a basis for future studies enhancing the clinical decision-making process related to selection and reduced discard of deceased donor kidneys. For instance, kidney biopsies are routinely examined for underlying pathologies primarily within the cortical region containing the proximal tubules as they are sensitive to oxidative stress and susceptible to cellular injury. As such, pathologies developing deep within the tissue such as in the renal inner medullary region remain undetected. This might not be an issue in an AKI setting where we found the cortex and inner medulla to exhibit strong positive correlation, mirroring each other’s temporal molecular dynamics. However, in a cold ischemia setting where the cortex and inner medulla demonstrated very weak correlation it might be worth evaluating the potential benefit of examining the inner medulla as part of the clinical routine. Furthermore, comparison of static cold ischemia vs. cold perfused vs. normothermic perfusion is also needed in the future in this evolving field. In order to make our data more accessible and understandable beyond those with expertise in spatial transcriptomic analysis, we also developed an interactive online browser at https://jef.works/vitessce-cold-ischemia/. Overall, we anticipate such spatiotemporally resolved molecular comparative analysis will enhance our understanding of biological processes underlying disease pathogenesis to help elucidate novel therapeutics in the future.

Materials and Methods

Cold Ischemia Model and Kidney Collection:

7–8 weeks old C57BL/6 male mice were anesthetized via isoflurane inhalation and perfused with ice-cold University of Wisconsin (UW) solution through the hepatic portal vein using a 10 cc syringe with a 25G needle. To facilitate circulation, the inferior vena cava below the liver was incised, enabling whole-body perfusion with UW solution. Following perfusion, the kidneys were harvested and immediately stored in UW solution at 4°C for 0, 12, 24, or 48 hours to induce cold ischemia. After the designated cold ischemia period, the kidneys were removed from the UW solution and fixed in 10% phosphate-buffered formalin for a minimum of 48 hours.

Sample preparation for spatial transcriptomics characterization:

To profile the global spatial transcriptome after cold ischemia at 0, 12, 24, and 48 hours, formalin fixed kidneys were cut sagittally, embedded in paraffin and sectioned under RNase free conditions. Kidney section (5mm) were placed on a positively charged glass slide, H&E stained and scanned. One kidney section from each timepoint was placed on 10x Visium FFPE (formalin fixed paraffin embedded) slide using CytAssist instrument. Adjacent sections (20 mm) were used for RNA quality assessment and samples with RQN value 7 were used for spatial transcriptomic analysis. Tissue processing and downstream analyses were achieved by adopting the following 10x Genomics protocols without any alteration: (1) Visium CytAssist Spatial Gene Expression for FFPE (CG000520, Revision B) for deparaffinization, staining, decrosslinking and imaging of slides. (2) “Sequencing” sections of the Visium CytAssist Spatial Gene Expression Reagent Kits User Guide (CG000495) for hybridization. The CytAssist libraries were sequenced using a Novaseq 6000 sequencer. Gene expression counts and spot positions were obtained using the SpaceRanger pipeline (v2.0.1) for downstream analysis.

Public data access:

Previously published 10x Visium spatial transcriptomics datasets from coronal sections of female murine kidneys with bilateral ischemia reperfusion injury (ID: AKI (n=5), timepoints: sham, 4, 12, 48 hours and 6 weeks) (17) were obtained from the Gene Expression Omnibus (GEO) under accession GSE182939. Likewise, previously published 10x Visium spatial transcriptomics (ST) datasets sections of sagittal sections of normal control male murine kidneys (ID: CTRL(n=4)) and of male murine kidneys with bilateral ischemia reperfusion injury (ID: AKI24 (n=4), timepoint: 24 hours (all specimens)) were obtained from the authors (18). All datasets were accessed as gene counts matrices and spatial positions in preprocessed h5 files.

Quality Control and Processing:

Any 10x Visium spots with less than 100 counts were removed. This resulted in a total of 44,816 spots belonging to all the 17 different spatial transcriptomics datasets. R statistical software (version 4.3.1 (2023–06-16)) was used for this purpose and all the computational analyses hereafter.

Data Integration, Batch Correction and Graph Based Clustering:

For data integration, for each of the individual ST dataset (prior to CPM normalization), we removed genes with less than 10 reads across all spots. Thereafter, we feature selected for over-dispersed genes in each dataset using the getOverdispersedGenes() function from MERINGUE (32) To mitigate sample-specific effects, we identified genes that were overdispersed in at least 2 datasets resulting in 1,378 final shared overdispsersed genes. Using these over-dispersed genes, the top 30 principal components (PCs) were identified. Batch correction was performed on the PCs to obtain harmonized PCs using the HarmonyMatrix() function (parameters: theta = 10, lambda = 0.05) from Harmony (19). Low-dimensional 2D embeddings of the initial and harmonized PCs were obtained using tSNE (33). Graph-based Louvain clustering was performed on the harmonized PCs to identify transcriptionally similar spots using getClusters() function (parameters: k=100) from MERINGUE (32). Thereafter, for normalization purposes and subsequent linear regression analysis (covered in next section), the gene list was restricted to 19,454 unique genes that were shared across all 17 ST datasets.

Linear regression modeling:

Gene count matrices of CIS tissue specimens containing 19,454 unique shared genes (rows) at each timepoint 0, 12, 24 and 48 hours were subset into smaller compartment-specific matrices based on the annotations obtained from harmonized clustering. This resulted in generation of 3 matrices per timepoint capturing all the spots within the inner medulla, outer medulla, and cortex compartments respectively. Thereafter, we performed CPM (counts per million) normalization on the gene counts matrices to obtain count-normalized matrices using normalizeCounts() function from MERINGUE (32). To avoid over-representation from any particular compartment (especially the cortex), 190 random spots were sampled for each of these matrices belonging to the 3 compartments (per timepoint) to perform compartment-specific linear regression at each timepoint. lmfit() function from limma (34) package was used to perform linear regression with time (encoded as log_e(hours+1)) as the independent variable (x) and count-normalized gene counts as the dependent variable (y). Adjusted p-value (adj. p< 0.05) was used as a criterion to select the genes whose expression had significant temporal association (reflected by the magnitude of linear regression slope with unit as: CPM normalized read/log_e(hours+1)). Similar procedure was implemented on AKI tissue specimens at timepoints (0 (sham), 4, 12 and 48 hours).

Gene Set Enrichment Analysis:

These genes with significant temporal association identified through linear regression were ordered based on the strength of their temporal association (linear regression slopes). These genes were assigned a value which reflected their rank in the ordered list such that temporally upregulated genes received a positive score while temporally downregulated genes received a negative score. Thereafter, gene set enrichment analysis was performed on the ordered list of genes along with their assigned values using gseKEGG() function from ClusterProfiler (35) package and GSEA() function from Msigdbr (36) package to identify enriched KEGG and Hallmark (gene set of Msigdbr) molecular pathways respectively. Adjusted p-value (adj. p<0.05) was used as a cut-off criterion to select significantly enriched molecular pathways. Running enrichment score plots were plotted using gseaplot() and gseaplot2() functions from enrichplot (37) package.

Compartmental correlation analysis and linear regression:

Correlation analysis of the temporal transcriptomic dynamics between the different compartments was performed using cor.test() function in R to obtain Pearson correlation coefficient (PCC) and p-value. Linear regression was performed using lm() function in R and summary function was used to obtain the R² value.

Differential Expression Analysis:

Differential expression analysis was performed to identify top marker genes of inner medulla, outer medulla, and cortex compartments. Briefly, count matrices (raw counts) belonging to all the 17 ST datasets were split into three groups based on the labels (from earlier assigned compartmental annotations) of the spots. Thereafter, differential expression analysis was performed between the interest group and the other two groups pooled together into one group using DESeq2 (38). Top 30 genes based on their log2 fold change (and adjusted p-value<0.05) from each of the compartments were identified and plotted as heatmaps. Differential expression was also performed between the inner medulla compartment of all 4 kidney tissue specimens belonging to normal control kidneys (CTRL) and 4 warm ischemia reperfusion injury (AKI24) to evaluate consistency with linear regression modeling.

RNA isolation and qPCR:

Kidney cortex and medulla (inner + outer medulla) were collected as previously described (3S).)Total RNA was isolated from samples using a QIAshredder (Qiagen, Valencia, CA, Cat.No.79654) and a RNeasy Mini kit (Qiagen, Cat.No.74104) and reverse transcribed using High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Foster City, CA, Cat.No. 4368814). Real-time PCR was performed in QuantStudio 12K Flex (Applied Biosystems, Foster City, CA) using the SYBR Green PCR master mix (Applied biosystems, Cat.No. 4309155). β-actin served as the reference gene and the relative fold expression values were calculated using a DD cycle threshold method. The primer sequences for genes are listed below:

Genes	Forward primer	Reverse primer
Actb	GTGACGTTGACATCCGTAAAGA	GCCGGACTCATCGTACTCC
Atp5b	GCAAGGCAGGGACAGCAGA	CCCAAGGTCTCAGGACCAACA
Cox6a1	CAACGTGTTCCTCAAGTCGCGG	GCCAGGTTCTCTTTACTCATC
Pink1	TTCTTCCGCCAGTCGGTAG	CTGCTTCTCCTCGATCAGCC
Ppargc1a	CCCTGCCATTGTTAAGACC	TGCTGCTGTTCCTGTTTTC

Data availability:

Count matrices, spatial positions, and H&E images for the cold ischemia Visium datasets generated in this study have been deposited in a Zenodo repository at https://doi.org/10.5281/zenodo.15359609