Inflammation and Immunity Pathways Regulate Genetic Susceptibility to Diabetic Nephropathy

Diabetic nephropathy (DN) is a leading cause of end-stage renal disease worldwide (1). Genetic factors play a significant role in the development of DN, as only a subset of patients with diabetes develops nephropathy (2). Moreover, in patients with type 1 diabetes, the incidence of DN peaks 10 and 20 years after the initial diagnosis and declines thereafter, consistent with a defined, susceptible population (3–5). Finally, familial clustering and ethnic variation of susceptibility for renal disease are seen in both type 1 (6) and type 2 diabetes (7). Unraveling the genetic basis of DN promises the potential for valuable insights into disease-promoting pathways, allowing identification of new targets for therapy and useful information for optimizing existing treatments. However, it has been difficult to precisely define genetic pathways promoting susceptibility to DN in humans. Results from several large genome-wide association studies of DN have been reported, but so far, no individual genetic loci with major effects to influence DN susceptibility in humans have been clearly identified (8–14). Possible reasons for these outcomes include imprecise phenotyping (15) and the existence of multiple genes with small effects to influence DN development (8,9).

As an alternative approach for understanding DN pathogenesis, we and others have developed mouse models of DN that could be used for mechanistic experiments and preclinical studies (16–18). However, diabetic mice typically develop only modest kidney injury. Nonetheless, several models have been developed recapitulating characteristic features of human DN by superimposing stresses such as endothelial dysfunction on DN platforms (19–21). Similar to humans, there is a strong effect of genetic susceptibility to influence development of kidney disease in diabetic mice (17,18,22), suggesting these models might be useful for understanding genetics of DN.

In humans, one of the most well-established pathways in DN pathogenesis is abnormal activation of the renin-angiotensin system (RAS) (23). This is most clearly evident from clinical trials showing the significant benefits of ACE inhibitors and angiotensin receptor blockers to reduce proteinuria and delay the progression of DN in patients with type 1 and type 2 diabetes (24–26). Accordingly, we describe a model combining type 1 diabetes and chronic RAS activation, which develops robust kidney disease characteristic of human DN including high-grade albuminuria and nodular glomerulosclerosis. In this model, genetic background has a powerful influence on the severity of nephropathy, similar to humans. Moreover, on the susceptible background, the development of DN is associated with broad upregulation of glomerular gene expression patterns associated with immune activation and inflammation, suggesting these pathways may determine vulnerability to kidney disease. Furthermore, resistance to DN is associated with coordinate downregulation of these pathways.

C57BL/6-Ins2^C96Y and wild-type (WT) mice were purchased (The Jackson Laboratory), and 129SvEv-Ins2^C96Y mice were generated (129-Akita) (18). Mice bearing the renin transgene (RenTg) were obtained from Drs. Caron and Smithies (27). Experimental animals were generated by breeding to combine mutant alleles. Animals were housed in an Association for Assessment and Accreditation of Laboratory Animal Care–accredited animal facility at the Durham VA under National Institutes of Health (NIH) guidelines. Blood glucose was measured monthly, and during the last month of the phenotyping study, systolic blood pressures and glomerular filtration rate (GFR) were measured as described (18).

At 8, 12, and 26 weeks of age, 24-h urine collections were obtained (17,18). In a separate group of 16-week-old 129–Akita-RenTg mice (n = 10), we administered the angiotensin receptor blocker (ARB) losartan (10 mg/kg/day) in drinking water for 10 days. Urine albumin and creatinine concentrations were measured (20).

In a cohort of 12-week-old mice, glomeruli were isolated with Dynabeads as described (28). RNA was extracted from these samples using the Qiagen RNeasy Micro-Kit and submitted to the Duke Microarray Core Facility for processing with MessageAmp Premier RNA Amplification (Ambion) (29). The GeneChip Mouse Genome 430A 2.0 Array (Affymetrix) was used to probe expression levels of 14,000 mouse genes. Arrays were scanned using Affymetrix equipment and protocols and data downloaded as CEL files.

Background-adjusted, quantile-normalized, and log2-transformed gene expression signals were obtained via the Robust Multichip Averaging method (30,31). A total of 5,888 probes with a normalized average expression <50 units in all groups were excluded, resulting in the retention of 16,802 probes. Statistical significance of the differentially expressed genes was ascertained via a regularized t test based on the Bayesian statistical framework (32). For each strain, gene expression changes were quantified by the difference in the log average signals between Akita-RenTg mice and corresponding WT controls. Microarray data have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus repository (accession number GSE85569).

Pathway enrichment analysis was conducted via a gene set enrichment anlysis (GSEA) tool and overrepresentation analyses (ORAs). A GSEA tool (33) was applied to the list of 16,802 probes, and a running-sum statistic was used to determine the enrichment of a priori defined pathways based on the gene ranks. Statistical significance of pathway enrichment score was ascertained by permutation testing over size-matched random gene sets, and multiple testing was controlled by the false discovery rate (FDR) (34). Pathway information was obtained from a custom database containing mouse gene-symbol equivalents of human pathways from the Kyoto Encyclopedia of Genes and Genomes (KEGG) (35).

ORA was conducted via the Ingenuity Pathway Analysis (IPA) tool (www.ingenuity.com) with a prefiltered list of nominally differentially expressed genes (P < 0.05 and absolute fold-change ≥1.5), resulting in 270 genes for the 129 comparisons and 108 genes for B6. Enrichment analysis was conducted against the repository of canonical pathways available in the Ingenuity Knowledge Base. Overrepresentation of biological pathways was ascertained via Fisher exact test and corrected for multiple testing by the Benjamini-Hochberg procedure (34).

Biofunctions were based on the curated knowledge base in IPA. Secondary functions were based on the published evidence of gene function. The enrichment P value was calculated using the right-tailed Fisher exact test. The overall likelihood of activation/inhibition of a biofunction (based on published effects of the genes on that function) is represented as a z score. Biofunctions with an absolute z score ≥2 were predicted to be activated or inhibited, based on the sign of the z score.

The IPA platform was used to predict possible upstream regulators (36). ORA (Fisher exact test) was first performed to determine whether a regulator was enriched for differential expression of its target genes. The overall activation/inhibition status of the regulator was then inferred from the preponderance of evidence for up- or downregulation of target genes, represented by a z-score statistic, similar to biofunctions.

Total RNA was extracted from kidneys of mice (n = 3 to 4/group) using the Qiagen RNeasy Mini Kit. A total of 1 μg RNA was reverse-transcribed to cDNA using iScript (Bio-Rad) and subjected to real-time PCR using SsoFast EvaGreen Supermix (Bio-Rad). CCL5 (RANTES) (37) and CXCR4 gene (38) expression levels were measured (37) and normalized to GAPDH.

Western blot was performed on total protein extracts from kidneys of mice (n = 3/group). A total of 50 µg protein/lane was separated on 13.5% SDS-PAGE gel and transferred on polyvinylidene difluoride membrane for 1 h at 100 V. The blots were immunoprobed with anti–tumor necrosis factor-α (TNF-α; Santa Cruz Biotechnology 52746 mouse) and anti-GAPDH (Santa Cruz Biotechnology 25778 rabbit) antibodies. Pierce ECL Western Blotting Substrate (catalog number 32106) was used for the detection of the horseradish peroxidase–conjugated secondary antibody; films were scanned using the RICOH scanner (600 dpi and grayscale) and analyzed with ImageJ software.

For physiological data, the values for each parameter within a group are expressed as the mean ± SEM. For comparisons between two experimental groups, a t test or Mann-Whitney test was used as appropriate to assess statistical significance. For comparisons among multiple groups, ANOVA with Tukey test was used (GraphPad).

In order to enhance the extent of kidney injury in diabetic mice, we combined the Ins2^C96Y Akita mutation with a constitutively active renin transgene (RenTg). The single-copy renin transgene generates mouse renin under control of the albumin promoter independent of typical physiological regulation (27). As genetic background has a significant impact on renal manifestations of diabetes in Akita mice (18,22), we generated Akita-RenTg mice on two inbred backgrounds, C57BL/6 and 129/SvEv, through repetitive backcrossing (N ≥ 7 generations) and compared their phenotype to WT, Akita, and RenTg mice, on each background.

Double-heterozygous Akita-RenTg mice were born in the expected frequencies and developed marked hyperglycemia (Fig. 1A) without significant differences between the two genetic strains (Table 1). Exogenous insulin was not administered. Previous studies have shown that the renin transgene causes an approximately sixfold elevation in plasma renin concentrations and moderate hypertension (27). Systolic blood pressures were significantly elevated in both strains of mice with the RenTg (P < 0.0002) and tended to be higher on the 129/SvEv genetic background (164 ± 6 mmHg) compared with C57BL/6 (153 ± 2 mmHg; P = 0.09).

We measured total albumin excretion (AE) and albumin-to-creatinine ratio as a marker of renal injury (Fig. 1B and Table 1). As reported previously (18) and shown for comparison in Table 1, WT mice on both backgrounds had barely detectable levels of albuminuria, and in the presence of diabetes associated with the Akita mutation, there was a modest, statistically significant increase in albuminuria. The presence of the RenTg alone resulted in modest albuminuria in both strains, greater in 129 compared with C57BL/6 (383 ± 42 vs. 86 ± 29 µg/24 h; P = 0.00005). Moreover, on the 129 background, the combination of the Akita mutation with the RenTg caused a dramatic acceleration of albuminuria to levels that were ∼100-fold of 129-WT controls (3,456 ± 702 µg/24 h; P < 0.001) by 6 months of age (Fig. 1B). In contrast, albuminuria was only modestly increased in C57BL/6–Akita-RenTg mice (398 ± 94 µg/24 h; P < 0.001 vs. 129–Akita-RenTg) (Table 1). On a susceptible genetic background (129), the combination of diabetes with RAS activation markedly enhances albuminuria. To determine whether marked albuminuria in 129-Akita-RenTg mice was associated with alterations in renal function, we measured GFR. Similar to other accelerated mouse models of DN (39), 7-month-old 129–Akita-RenTg mice do not have reduced GFR but exhibit significant hyperfiltration (Supplementary Fig. 5).

To evaluate the extent of renal structural changes between the groups, we examined histopathology at 6 months of age. In the mice with diabetes or RenTg alone, there were minimal changes in kidney pathology, confined to modest mesangial expansion (Fig. 1C and D and Table 1). A similar increase in mesangial matrix was seen in the C57BL/6–Akita-RenTg animals (Table 1). By contrast, we found more florid structural abnormalities in the 129–Akita-RenTg group, including frank glomerulosclerosis with interstitial infiltration by inflammatory cells, along with protein casts in tubules (Fig. 1E and F). This severe pathological change was not observed in any other experimental group. The acceleration in renal pathology was also reflected in the quantitative pathological scores in which 129–Akita-RenTg mice had higher scores for mesangial expansion (1.2 ± 0.2 vs. 0.82 ± 0.11; P = 0.06). By contrast, glomerular volumes and kidney weights were increased significantly and to a similar extent in Akita-RenTg mice on 129 and C57BL/6 backgrounds compared with WT (Table 1). Accordingly, compared with C57BL/6, the 129–Akita-RenTg mouse model recapitulates many of the features of human DN including marked albuminuria, hypertension, glomerulosclerosis, and interstitial inflammation.

Antiproteinuric and renal protective actions of ARBs are key characteristics of human DN (23). To block the RAS, we administered the ARB losartan (10 mg/kg/day) to 16-week-old 129–Akita-RenTg mice for 10 days. After 10 days of administration, the ARB caused an abrupt and significant reduction in AE from 1,049 ± 325 to 308 ± 156 µg/day (P = 0.004). Thus, similar to human DN (23), blockade of the RAS in our mouse model of DN causes dramatic reductions in proteinuria.

To identify molecular pathways involved in the development of DN, we performed microarray analysis of glomerular RNA, with particular attention to differences in gene expression patterns between resistant (C57BL/6) and susceptible (129) strains. We focused on the glomerulus, because this is the compartment where the earliest and defining pathological changes are seen, and used glomeruli isolated from younger mice at 12 weeks of age. To test for evidence of DN manifestations in young mice, we measured albuminuria in separate groups of C57BL/6– and 129–Akita-RenTg at 8 and 12 weeks of age. As shown in Supplementary Fig. 1, albuminuria was increased in the 129–Akita-RenTg compared with C57BL/6–Akita-RenTg as early as 8 weeks of age (402 ± 87 vs. 136 ± 32 µg/24 h; P < 0.025), and this trend was also observed at 12 weeks (843 ± 186 vs. 432 ± 240 µg/24 h; P < 0.04), the time of glomerular harvest.

Using a nominal P < 0.05 and absolute fold change ≥1.5-fold cutoff for differential gene expression, we first compared the overall transcriptomic response between the Akita-RenTg and WT mice in each of the two strains. Overall, a greater number of genes were differentially regulated in the 129 strain compared with C57BL/6 (270 vs.108). The level of overlap among the differentially expressed genes between the two strains was generally low: only 10 genes were shared among the 168 and 76 downregulated genes in 129 and C57BL/6, respectively, whereas there were only 12 genes in common among the 102 and 32 upregulated genes between the two strains (Supplementary Fig. 2). This modest level of overlap suggests that the biological pathways impacted by the combination of diabetes and RAS activation are quite different between the two strains. The full list of differentially regulated genes is in Supplementary Table 1.

GSEA identified a coordinate upregulation of several proinflammatory pathways with the strongest enrichment signal observed in the 129–Akita-RenTg samples (all pathways at FDR ≤0.05). These included cytokine–cytokine receptor interaction, chemokine signaling, and Toll-like receptor signaling. In contrast, on the C57BL/6 background, the Akita-RenTg combination was associated with downregulation of inflammatory pathways (also at FDR ≤0.05), some of which were identical to the upregulated pathways in the susceptible strain, including chemokine signaling and cytokine–cytokine receptor interaction, among others. The enrichment for the cytokine–cytokine receptor pathway is shown in Fig. 2, clearly demonstrating the contrasting upregulation in 129 (Fig. 2A) and coordinate downregulation in C57BL/6 (Fig. 2B). The full list of significantly enriched pathways is in Supplementary Table 2.

We compared distribution of individual genes among the top 10 most significantly enriched pathways, sorted by FDR. Some of the key genes distributed across the significantly enriched pathways in 129–Akita-RenTg included proinflammatory mediators such as TNF-α, TGFB1, TGFB3, CXCL3, CCL5, and CCL4. A similar analysis of the significantly downregulated pathways in C57BL/6–Akita-RenTg showed genes such as PI3Kγ, IL7R, and IL5 to be represented across multiple pathways. The complete gene distribution across the selected pathways is shown in Supplementary Fig. 3.

IPA was conducted with a select list of genes based on the magnitude and statistical evidence for differential expression (P < 0.05; absolute fold change ≥1.5). This analysis identified opposite effects between susceptible and resistant strains, most significantly in the inflammatory response and immune-cell trafficking biofunctions. Figure 3 presents a tree map summarizing the changes in subfunctions underlying the inflammatory response in which subfunctions such as cell adhesion, activation, and chemotaxis are clustered together, clearly highlighting the opposing patterns of gene regulation in these two strains, with consistent upregulation in 129 and downregulation in C57BL/6 (P < 0.001).

To confirm differences in expression of key proinflammatory genes, we measured mRNA levels of two chemokines, CXCR4 and CCL5, by quantitative PCR. These chemokines were chosen because CXCR4 was one of the top differentially expressed inflammatory genes, and CCL5 is a target of multiple upstream, proinflammatory regulators with exaggerated activity identified in 129–Akita-RenTg mice. As shown in Fig. 4, mRNA expression of both CXCR4 and CCL5 was significantly increased in 129– compared with C57BL/6–Akita-RenTg mice.

With the Ingenuity knowledge base, we explored upstream regulators, which might explain the observed transcriptomic responses. We identified CEBPB, IFNG, IL1A, IFNA, and TNF as putative regulators with opposing effects in the two backgrounds (P < 0.001 in all cases). Several of these genes were also functionally connected, leading to a probable mechanistic network (36) explaining the gene expression differences in the two strains. Figure 5A depicts a network for the TNF family of cytokine regulators in which TNF is postulated to activate IKBKB and CHUK, two key components of a cytokine-activated protein complex and activators of nuclear factor-κB signaling, which, in turn, regulate effectors such as NFKBIA, CEBPB, and RELA to explain the observed gene expression changes. Collectively, the set of regulators shown in Fig. 5A connects a total of 44 genes in the data set with TNF, 25 of them directly. Based on these findings, we measured TNF-α protein levels in kidneys of 129– and C57BL/6–Akita-RenTg mice. As shown in Fig. 6, we found a significant threefold increase in TNF-α protein levels in 129– compared with C57BL/6–Akita-RenTg mice (P = 0.03).

We compared the pathways identified in our mouse model with published data from a rat model of accelerated DN (40), with an inducible renin transgene combined with streptozotocin. Similar to our model, these rats developed proteinuria and glomerulosclerosis (40). Microarray data from whole kidney RNA were downloaded from the Gene Expression Omnibus (GSE85569) and subjected to signal extraction as described. We filtered differentially expressed genes using identical thresholds (P < 0.05 and absolute fold-change ≥1.5). As shown in Fig. 7A, 41 differentially regulated genes were common between the rat and 129 lists, and 34 out of 41 genes showed consistent direction of changes. In contrast, as shown in Fig. 7B, only 26 genes were shared between the rat and B6 gene lists, and the majority (24 genes) showed an opposite expression trend (i.e., genes upregulated in the rat model were downregulated in B6 nephropathy resistance and vice versa). Thus, there is congruence of gene expression patterns between the rat model of DN and the Akita-RenTg mouse, but only on the susceptible background (129).

Next, we interrogated the rat transcriptome data and identified KEGG pathways that were enriched for differentially expressed genes. These pathways were compared with the pathways identified earlier from our mouse studies and filtered at an FDR ≤30% to allow for study heterogeneity and identification of weaker, but consistent, pathway associations. A list of overlapping and unique pathways for each species is presented in Supplementary Table 3. For upregulated pathways, a core set of inflammation-related processes, such as cytokine–cytokine receptor interaction, Toll-like receptor signaling, and chemokine signaling pathways, were activated in common between the rat and 129 mice. Conversely, several proinflammatory pathways upregulated in rat were downregulated in B6 mice.

We next assessed whether the findings from the mouse model were also be recapitulated in humans with DN. For this analysis, we identified two studies from the Nephroseq database (https://www.nephroseq.org/) interrogating transcriptomes in glomeruli from patients with DN versus healthy control subjects. For each study, we identified genes nominally significant at P < 0.05 and absolute fold change ≥1.5-fold between DN and control subjects and subjected genes to ORA (Ingenuity), comparing results from the human data sets to our mouse studies (Supplementary Fig. 4). No overlap was seen between B6 mice and either of the human studies. Four pathways (P < 0.01) were identified as commonly upregulated in DN in the mouse and humans. Comparisons at the biofunction level identified 13 biofunctions jointly upregulated in the three studies (absolute z score ≥2). These biofunctions were all related to immune cell trafficking, suggesting one common mechanism behind the upregulation of proinflammatory signaling. Finally, we also identified a common set of 10 upstream regulators predicted to be activated among the three studies (z score ≥2), including several related to the regulation of proinflammatory signaling genes (TNF, IFNG, IL1A, and IL17A). Notably, there was significant overlap in components of the TNF regulatory network between the 129 mouse and humans, suggesting conservation of signaling modules in the pathophysiology of DN (Fig. 5B) and activation of TNF pathways as an important component of DN in humans (41).

We describe a mouse model with typical features of human DN, including hypertension, high-grade albuminuria, glomerulosclerosis, and responsiveness to RAS blockade. Combining chronic, low-level activation of the RAS with type 1 diabetes on a susceptible genetic background dramatically accelerates the development of kidney injury. This is consistent with previous studies in the rat (40), mouse (42), and, with the well-established role of RAS activation, in the pathogenesis of human DN (23,43). One key feature of our mouse model is the powerful impact of genetic background in determining severity of kidney injury, with striking susceptibility observed on the 129 strain background and marked resistance on the C57BL/6 background. These strain differences provide a powerful system for identifying mechanisms associated with pathogenesis of DN as well as factors that may protect against development of kidney disease in diabetes.

In order to explore mechanisms of DN susceptibility and resistance, we profiled glomerular transcriptomes from mice at 12 weeks of age, an early stage of disease, to avoid confounding effects of cellular infiltration or structural injury on identification of key causal pathways. Our most striking findings were marked differences in expression of genes involved in immune and inflammatory responses between susceptible and resistant strains, with broad triggering of immune and inflammatory gene expression profiles in the mice that would develop DN. By contrast, these pathways were coordinately downregulated in mice protected from kidney injury. Although these differences were observed between the strains in the presence of diabetes and RAS activation, they were also apparent in glomeruli from WT mice of each parental line, suggesting these are basic characteristics of the two strains.

Although DN is classically considered a bland glomerulopathy, our data suggest that inflammation and immune responsiveness may be critical features determining susceptibility. Similar patterns of immune activation were recapitulated in our comparisons with a rat model of accelerated diabetic kidney injury and human DN. Hodgin et al. (44) also identified common transcriptional networks shared between human and mouse DN, including interleukin-7 and Janus kinase/STAT pathways. Furthermore, the pattern of suppressed inflammation and immunity seen in the resistant C57BL/6 model was very different from transcriptomic profiles in rats or humans with diabetic kidney disease, suggesting that suppressed immune activation is associated with protection from nephropathy.

A role for inflammation and immune activation has been suggested by other studies implicating chemokines such as MCP-1 (45,46) and inflammatory cytokines including TNF-α (41,47,48) in the pathogenesis of DN. For example, a specific inhibitor of the MCP-1 receptor CCR2 attenuated albuminuria diabetic mice with a human CCR2 transgene (49), and specific TNF-α inhibitors reduced proteinuria in streptozotocin-treated rats (50). Accordingly, although such pathways may represent mechanisms conferring susceptibility, they may also be useful targets for interrupting pathogenesis and slowing progression of kidney injury in DN. In particular, signaling pathways connected to TNF-α were enhanced in mice, rats, and humans with DN, but are coordinately downregulated in mice resistant to development of kidney injury in diabetes. These data are consistent with recent clinical studies, suggesting that activation biomarkers for TNF pathways are associated with risk for DN susceptibility (41).

In summary, one approach for understanding root causes of DN and testing potential therapies is through animal models that faithfully reproduce the human disorder. In this study, we describe a mouse model that is simple and robust and avoids the potential for individual variability seen with chemically induced models. Furthermore, activation of the RAS is critical to the development of kidney injury, similar to the human disorder (23–26). Along with high-grade albuminuria, glomerulosclerosis, and hypertension, this model manifests another key feature of human DN, genetic regulation of susceptibility to kidney injury. The profound strain differences in susceptibility to kidney disease allow for an unbiased approach to identifying mechanisms promoting and/or protecting against the development of nephropathy. Although the precise genetic determinants of susceptibility and resistance to DN may differ between mice and humans, it is likely that the molecular pathways involved are similar. In this regard, upregulated immune and inflammatory pathways, including networks associated with TNF-α signaling, are shared among mice, rats, and humans with diabetic kidney disease and are coordinately downregulated in mice that are protected from DN. Such pathways are attractive as biomarkers for disease susceptibility (41) and targets for therapy.

Funding. This work received funding support from the NIH Animal Models of Diabetic Complications Consortium (U01-DK076136-02) and DiaComp Pilot and Feasibility Program (3 U24-DK076169-09, subcontract #25732-54), the Duke O’Brien Center for Kidney Research (NIH P30-DK-096493), a Duke/Duke-NUS Collaboration Pilot Project Award, and the Singapore National Medical Research Council (NMRC/OFLCG/001/2017).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. S.B.G. designed and conducted experiments, analyzed data, and wrote the manuscript. S.G. analyzed data and wrote the manuscript. S.A.J., K.A., R.B.S., S.E.G., M.M., and T.W.M. conducted experiments. T.M.C. designed experiments, analyzed data, and wrote the manuscript. S.B.G. and T.M.C. are the guarantors of this work and, as such, had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.