Clustering-based spatial analysis (CluSA) framework through graph neural network for chronic kidney disease prediction using histopathology images.

Machine learning applied to digital pathology has been increasingly used to assess kidney function and diagnose the underlying cause of chronic kidney disease (CKD). We developed a novel computational framework, clustering-based spatial analysis (CluSA), that leverages unsupervised learning to learn spatial relationships between local visual patterns in kidney tissue. This framework minimizes the need for time-consuming and impractical expert annotations. 107,471 histopathology images obtained from 172 biopsy cores were used in the clustering and in the deep learning model. To incorporate spatial information over the clustered image patterns on the biopsy sample, we spatially encoded clustered patterns with colors and performed spatial analysis through graph neural network. A random forest classifier with various groups of features were used to predict CKD. For predicting eGFR at the biopsy, we achieved a sensitivity of 0.97, specificity of 0.90, and accuracy of 0.95. AUC was 0.96. For predicting eGFR changes in one-year, we achieved a sensitivity of 0.83, specificity of 0.85, and accuracy of 0.84. AUC was 0.85. This study presents the first spatial analysis based on unsupervised machine learning algorithms. Without expert annotation, CluSA framework can not only accurately classify and predict the degree of kidney function at the biopsy and in one year, but also identify novel predictors of kidney function and renal prognosis.

Unsupervised machine learning for identifying important visual features through bag-of-words using histopathology data from chronic kidney disease.

Pathologists use visual classification to assess patient kidney biopsy samples when diagnosing the underlying cause of kidney disease. However, the assessment is qualitative, or semi-quantitative at best, and reproducibility is challenging. To discover previously unknown features which predict patient outcomes and overcome substantial interobserver variability, we developed an unsupervised bag-of-words model. Our study applied to the C-PROBE cohort of patients with chronic kidney disease (CKD). 107,471 histopathology images were obtained from 161 biopsy cores and identified important morphological features in biopsy tissue that are highly predictive of the presence of CKD both at the time of biopsy and in one year. To evaluate the performance of our model, we estimated the AUC and its 95% confidence interval. We show that this method is reliable and reproducible and can achieve 0.93 AUC at predicting glomerular filtration rate at the time of biopsy as well as predicting a loss of function at one year. Additionally, with this method, we ranked the identified morphological features according to their importance as diagnostic markers for chronic kidney disease. In this study, we have demonstrated the feasibility of using an unsupervised machine learning method without human input in order to predict the level of kidney function in CKD. The results from our study indicate that the visual dictionary, or visual image pattern, obtained from unsupervised machine learning can predict outcomes using machine-derived values that correspond to both known and unknown clinically relevant features.

Renin-angiotensin system inhibition reverses the altered triacylglycerol metabolic network in diabetic kidney disease.

OBJECTIVE: Dyslipidemia is a significant risk factor for progression of diabetic kidney disease (DKD). Determining the changes in individual lipids and lipid networks across a spectrum of DKD severity may identify lipids that are pathogenic to DKD progression. METHODS: We performed untargeted lipidomic analysis of kidney cortex tissue from diabetic db/db and db/db eNOS-/- mice along with non-diabetic littermate controls. A subset of mice were treated with the renin-angiotensin system (RAS) inhibitors, lisinopril and losartan, which improves the DKD phenotype in the db/db eNOS-/- mouse model. RESULTS: Of the three independent variables in this study, diabetes had the largest impact on overall lipid levels in the kidney cortex, while eNOS expression and RAS inhibition had smaller impacts on kidney lipid levels. Kidney lipid network architecture, particularly of networks involving glycerolipids such as triacylglycerols, was substantially disrupted by worsening kidney disease in the db/db eNOS-/- mice compared to the db/db mice, a feature that was reversed with RAS inhibition. This was associated with decreased expression of the stearoyl-CoA desaturases, Scd1 and Scd2, with RAS inhibition. CONCLUSIONS: In addition to the known salutary effect of RAS inhibition on DKD progression, our results suggest a previously unrecognized role for RAS inhibition on the kidney triacylglycerol lipid metabolic network.

Gene expression profiles of diabetic kidney disease and neuropathy in eNOS knockout mice: Predictors of pathology and RAS blockade effects.

Diabetic kidney disease (DKD) and diabetic peripheral neuropathy (DPN) are two common diabetic complications. However, their pathogenesis remains elusive and current therapies are only modestly effective. We evaluated genome-wide expression to identify pathways involved in DKD and DPN progression in db/db eNOS-/- mice receiving renin-angiotensin-aldosterone system (RAS)-blocking drugs to mimic the current standard of care for DKD patients. Diabetes and eNOS deletion worsened DKD, which improved with RAS treatment. Diabetes also induced DPN, which was not affected by eNOS deletion or RAS blockade. Given the multiple factors affecting DKD and the graded differences in disease severity across mouse groups, an automatic data analysis method, SOM, or self-organizing map was used to elucidate glomerular transcriptional changes associated with DKD, whereas pairwise bioinformatic analysis was used for DPN. These analyses revealed that enhanced gene expression in several pro-inflammatory networks and reduced expression of development genes correlated with worsening DKD. Although RAS treatment ameliorated the nephropathy phenotype, it did not alter the more abnormal gene expression changes in kidney. Moreover, RAS exacerbated expression of genes related to inflammation and oxidant generation in peripheral nerves. The graded increase in inflammatory gene expression and decrease in development gene expression with DKD progression underline the potentially important role of these pathways in DKD pathogenesis. Since RAS blockers worsened this gene expression pattern in both DKD and DPN, it may partly explain the inadequate therapeutic efficacy of such blockers.

Podocyte-specific JAK2 overexpression worsens diabetic kidney disease in mice.

Activation of JAK-STAT signaling has been implicated in the pathogenesis of diabetic kidney disease. An increased expression of JAK-STAT genes was found in kidney glomerular cells, including podocytes, in patients with early diabetic kidney disease. However, it is not known whether increased expression of JAK or STAT isoforms in glomerular cells can lead to worsening nephropathy in the setting of diabetes. Therefore, we overexpressed JAK2 mRNA specifically in glomerular podocytes of 129S6 mice to determine whether this change alone could worsen diabetic kidney disease. A 2-3 fold increase in glomerular JAK2 expression, an increase similar to that found in humans with early diabetic kidney disease, led to substantial and statistically significant increases in albuminuria, mesangial expansion, glomerulosclerosis, glomerular fibronectin accumulation, and glomerular basement membrane thickening, and a significant reduction in podocyte density in diabetic mice. Treatment with a specific JAK1/2 inhibitor for 2 weeks partly reversed the major phenotypic changes of diabetic kidney disease and specifically normalized expression of a number of downstream STAT3-dependent genes implicated in diabetic kidney disease progression. Thus, moderate increases in podocyte JAK2 expression at levels similar to those in patients with early diabetic kidney disease can lead directly to phenotypic and other alterations of progressive diabetic glomerulopathy. Hence, inhibition of these changes by treatment with a JAK1/2 inhibitor suggests that such treatment may help retard progression of early diabetic kidney disease in patients.

Maintenance of GLUT4 expression in smooth muscle prevents hypertension-induced changes in vascular reactivity.

Previous studies have shown that expression of GLUT4 is decreased in arterial smooth muscle of hypertensive rats and mice and that total body overexpression of GLUT4 in mice prevents enhanced arterial reactivity in hypertension. To demonstrate that the effect of GLUT4 overexpression on vascular responses is dependent on vascular smooth muscle GLUT4 rather than on some systemic effect we developed and tested smooth-muscle-specific GLUT4 transgenic mice (SMG4). When made hypertensive with angiotensin II, both wild-type and SMG4 mice exhibited similarly increased systolic blood pressure. Responsiveness to phenylephrine, serotonin, and prostaglandin F2α was significantly increased in endothelium-intact aortic rings from hypertensive wild-type mice but not in aortae of SMG4 mice. Inhibition of Rho-kinase equally reduced serotonin-stimulated contractility in aortae of hypertensive wild-type and SMG4-mice. In addition, acetylcholine-stimulated relaxation was significantly decreased in aortic rings of hypertensive wild-type mice, but not in rings of SMG4 mice. Inhibition of either prostacylin receptors or cyclooxygenase-2 reduced relaxation in rings of hypertensive SMG4 mice. Inhibition of cyclooxygenase-2 had no effect on relaxation in rings of hypertensive wild-type mice. Cyclooxygenase-2 protein expression was decreased in hypertensive wild-type aortae but not in hypertensive SMG4 aortae compared to nonhypertensive controls. Our results demonstrate that smooth muscle expression of GLUT4 exerts a major effect on smooth muscle contractile responses and endothelium-dependent vasorelaxation and that normal expression of GLUT4 in vascular smooth muscle is required for appropriate smooth muscle and endothelial responses.

Targeted Lipidomic and Transcriptomic Analysis Identifies Dysregulated Renal Ceramide Metabolism in a Mouse Model of Diabetic Kidney Disease.

Both type 1 and type 2 diabetes are associated with altered lipid metabolism, which might in part contribute to debilitating complications such as diabetic kidney disease (DKD). Ceramides are bioactive sphingolipids that have been implicated in a variety of diseases as they can regulate cellular responses to stress and invoke a myriad of downstream signaling responses. To investigate a potential role of altered ceramide metabolism in DKD, we utilized a highly sensitive and specific mass spectrometry (MS) method to quantitatively measure species in plasma and kidney cortex from the C57BLKS db/db mouse model of DKD and littermate controls. Long-chain ceramides (C14:0, C16:0, C18:0, C20:0) and a glucosylceramide (Glu-Cer C18:0) were increased in diabetic mouse plasma, while long-chain (C14:0, C16:0, C18:0) and very-long-chain (C24:0, C24:1) ceramides and a glucosylceramide (Glu-Cer C16:0) were decreased in diabetic mouse kidney tissue. Kidney and plasma ceramide levels correlated to functional and histopathological features of DKD. Transcriptomic analysis of mouse kidney tissue revealed expression changes indicative of decreased ceramide synthesis (Degs2, Smpd2) and increased conversion to sphingosine (Acer2) and downstream sphingosine-1-phosphate signaling. Correlation analysis identified a negative relationship between plasma and kidney tissue levels of ceramide C16:0 and ceramide C24:1. Overall, the findings suggest a previously unrecognized role for ceramide metabolism in DKD.

Podocyte-specific GLUT4-deficient mice have fewer and larger podocytes and are protected from diabetic nephropathy.

Podocytes are a major component of the glomerular filtration barrier, and their ability to sense insulin is essential to prevent proteinuria. Here we identify the insulin downstream effector GLUT4 as a key modulator of podocyte function in diabetic nephropathy (DN). Mice with a podocyte-specific deletion of GLUT4 (G4 KO) did not develop albuminuria despite having larger and fewer podocytes than wild-type (WT) mice. Glomeruli from G4 KO mice were protected from diabetes-induced hypertrophy, mesangial expansion, and albuminuria and failed to activate the mammalian target of rapamycin (mTOR) pathway. In order to investigate whether the protection observed in G4 KO mice was due to the failure to activate mTOR, we used three independent in vivo experiments. G4 KO mice did not develop lipopolysaccharide-induced albuminuria, which requires mTOR activation. On the contrary, G4 KO mice as well as WT mice treated with the mTOR inhibitor rapamycin developed worse adriamycin-induced nephropathy than WT mice, consistent with the fact that adriamycin toxicity is augmented by mTOR inhibition. In summary, GLUT4 deficiency in podocytes affects podocyte nutrient sensing, results in fewer and larger cells, and protects mice from the development of DN. This is the first evidence that podocyte hypertrophy concomitant with podocytopenia may be associated with protection from proteinuria.

Divergent functions of the Rho GTPases Rac1 and Cdc42 in podocyte injury.

Podocytes are highly specialized epithelial cells with complex actin cytoskeletal architecture crucial for maintenance of the glomerular filtration barrier. The mammalian Rho GTPases Rac1 and Cdc42 are molecular switches that control many cellular processes, but are best known for their roles in the regulation of actin cytoskeleton dynamics. Here, we employed podocyte-specific Cre-lox technology and found that mice with deletion of Rac1 display normal podocyte morphology without glomerular dysfunction well into adulthood. Using the protamine sulfate model of acute podocyte injury, podocyte-specific deletion of Rac1 prevented foot process effacement. In a long-term model of chronic hypertensive glomerular damage, however, loss of Rac1 led to an exacerbation of albuminuria and glomerulosclerosis. In contrast, mice with podocyte-specific deletion of Cdc42 had severe proteinuria, podocyte foot process effacement, and glomerulosclerosis beginning as early as 10 days of age. In addition, slit diaphragm proteins nephrin and podocin were redistributed, and cofilin was dephosphorylated. Cdc42 is necessary for the maintenance of podocyte structure and function, but Rac1 is entirely dispensable in physiological steady state. However, Rac1 has either beneficial or deleterious effects depending on the context of podocyte impairment. Thus, our study highlights the divergent roles of Rac1 and Cdc42 function in podocyte maintenance and injury.

Podocyte-specific overexpression of GLUT1 surprisingly reduces mesangial matrix expansion in diabetic nephropathy in mice.

Increased expression of the facilitative glucose transporter, GLUT1, leads to glomerulopathy that resembles diabetic nephropathy, whereas prevention of enhanced GLUT1 expression retards nephropathy. While many of the GLUT1-mediated effects are likely due to mesangial cell effects, we hypothesized that increased GLUT1 expression in podocytes also contributes to the progression of diabetic nephropathy. Therefore, we generated two podocyte-specific GLUT1 transgenic mouse lines (driven by a podocin promoter) on a db/m C57BLKS background. Progeny of the two founders were used to generate diabetic db/db and control db/m littermate mice. Immunoblots of glomerular lysates showed that transgenic mice had a 3.5-fold (line 1) and 2.1-fold (line 2) increase in GLUT1 content compared with wild-type mice. Both lines showed similar increases in fasting blood glucose and body weights at 24 wk of age compared with wild-type mice. Mesangial index (percent PAS-positive material in the mesangial tuft) increased 88% (line 1) and 75% (line 2) in the wild-type diabetic mice but only 48% (line 1) and 39% (line 2) in the diabetic transgenic mice (P < 0.05, transgenic vs. wild-type mice). This reduction in mesangial expansion was accompanied by a reduction in fibronectin accumulation, and vascular endothelial growth factor (VEGF) levels increased only half as much in the transgenic diabetic mice as in wild-type diabetic mice. Levels of nephrin, neph1, CD2AP, podocin, and GLUT4 were not significantly different in transgenic compared with wild-type mice. Taken together, increased podocyte GLUT1 expression in diabetic mice does not contribute to early diabetic nephropathy; surprisingly, it protects against mesangial expansion and fibronectin accumulation possibly by blunting podocyte VEGF increases.

Rosiglitazone reduces renal and plasma markers of oxidative injury and reverses urinary metabolite abnormalities in the amelioration of diabetic nephropathy.

Recent studies suggest that thiazolidinediones ameliorate diabetic nephropathy (DN) independently of their effect on hyperglycemia. In the current study, we confirm and extend these findings by showing that rosiglitazone treatment prevented the development of DN and reversed multiple markers of oxidative injury in DBA/2J mice made diabetic by low-dose streptozotocin. These diabetic mice developed a 14.2-fold increase in albuminuria and a 53% expansion of renal glomerular extracellular matrix after 12 wk of diabetes. These changes were largely abrogated by administration of rosiglitazone beginning 2 wk after the completion of streptozotocin injections. Rosiglitazone had no effect on glycemic control. Rosiglitazone had similar effects on insulin-treated diabetic mice after 24 wk of diabetes. Podocyte loss and glomerular fibronectin accumulation, other markers of early DN, were prevented by rosiglitazone in both 12- and 24-wk diabetic models. Surprisingly, glomerular GLUT1 levels did not increase and nephrin levels did not decrease in the diabetic animals; neither changed with rosiglitazone. Plasma and kidney markers of protein oxidation and lipid peroxidation were significantly elevated in the 24-wk diabetic animals despite insulin treatment and were reduced to near-normal levels by rosiglitazone. Finally, urinary metabolites were markedly altered by diabetes. Of 1,988 metabolite features identified by electrospray ionization time of flight mass spectrometry, levels of 56 were altered more than twofold in the urine of diabetic mice. Of these, 21 were returned to normal by rosiglitazone. Thus rosiglitazone has direct effects on the renal glomerulus to reduce reactive oxygen species accumulation to prevent type 1 diabetic mice from development of DN.

Preserved expression of GLUT4 prevents enhanced agonist-induced vascular reactivity and MYPT1 phosphorylation in hypertensive mouse aorta.

We previously showed that GLUT4 expression is decreased in arterial smooth muscle of deoxycorticosterone acetate (DOCA)-salt hypertensive rats and that GLUT4-knockout mice have enhanced arterial reactivity. Therefore, we hypothesized that increased GLUT4 expression in vascular smooth muscle in vivo would prevent enhanced arterial reactivity and possibly reduce blood pressure in DOCA-salt hypertensive mice. Adult wild-type (WT) and GLUT4 transgenic (TG) mice were subjected to DOCA-salt hypertension with uninephrectomy or underwent uninephrectomy and remained normotensive. GLUT4 expression was increased more than twofold in the aortas of GLUT4 TG mice compared with WT aortas. Eight weeks after implantation of the DOCA pellets, GLUT4 expression decreased by 75% in aortas of WT hypertensive mice, but not in GLUT4 TG hypertensive aortas. Systolic blood pressure was significantly and similarly increased in WT and GLUT4 TG DOCA-salt mice compared with their respective sham-treated controls (159 vs. 111 mmHg). Responsiveness to the contractile agonist 5-HT was significantly increased in aortic rings from WT DOCA-salt mice but remained normal in GLUT4 TG DOCA mice. Phosphorylation of the myosin phosphatase targeting subunit MYPT1 was significantly enhanced in aortas of WT DOCA-salt mice, and this increase was prevented in GLUT4 TG mice. MYPT1 phosphorylation was also increased in nonhypertensive GLUT4-knockout mice. Myosin phosphatase, a major negative regulator of calcium sensitivity, is itself negatively regulated by phosphorylation of MYPT1. Therefore, our results show that preservation of GLUT4 expression prevents enhanced arterial reactivity in hypertension, possibly via effects on myosin phosphatase activity.

Reduction in podocyte density as a pathologic feature in early diabetic nephropathy in rodents: prevention by lipoic acid treatment.

BACKGROUND: A reduction in the number of podocytes and podocyte density has been documented in the kidneys of patients with diabetes mellitus. Additional studies have shown that podocyte injury and loss occurs in both diabetic animals and humans. However, most studies in animals have examined relatively long-term changes in podocyte number and density and have not examined effects early after initiation of diabetes. We hypothesized that streptozotocin diabetes in rats and mice would result in an early reduction in podocyte density and that this reduction would be prevented by antioxidants. METHODS: The number of podocytes per glomerular section and the podocyte density in glomeruli from rats and mice with streptozotocin (STZ)-diabetes mellitus was determined at several time points based on detection of the glomerular podocyte specific antigens, WT-1 and GLEPP1. The effect of insulin administration or treatment with the antioxidant, alpha-lipoic acid, on podocyte number was assessed. RESULTS: Experimental diabetes resulted in a rapid decline in apparent podocyte number and podocyte density. A significant reduction in podocytes/glomerular cross-section was found in STZ diabetes in rats at 2 weeks (14%), 6 weeks (18%) and 8 weeks (34%) following STZ injection. Similar declines in apparent podocyte number were found in STZ diabetes in C57BL/6 mice at 2 weeks, but not at 3 days after injection. Treatment with alpha-lipoic acid substantially prevented podocyte loss in diabetic rats but treatment with insulin had only a modest effect. CONCLUSION: STZ diabetes results in reduction in apparent podocyte number and in podocyte density within 2 weeks after onset of hyperglycemia. Prevention of these effects with antioxidant therapy suggests that this early reduction in podocyte density is due in part to increased levels of reactive oxygen species as well as hyperglycemia.