Transcriptional networks of progressive diabetic peripheral neuropathy in the db/db mouse model of type 2 diabetes: an inflammatory story

Abstract

Diabetic peripheral neuropathy is the most common complication of diabetes and a source of considerable morbidity. Numerous molecular pathways are linked to neuropathic progression, but it is unclear whether these pathways are altered throughout the course of disease. Moreover, the methods by which these molecular pathways are analyzed can produce significantly different results; as such it is often unclear whether previously published pathways are viable targets for novel therapeutic approaches. In the current study we examine changes in gene expression patterns in the sciatic nerve (SCN) and dorsal root ganglia (DRG) of db/db diabetic mice at 8, 16, and 24 weeks of age using microarray analysis. Following the collection and verification of gene expression data, we utilized both self-organizing map (SOM) analysis and differentially expressed gene (DEG) analysis to detect pathways that were altered at all time points. Though there was some variability between SOM and DEG analyses, we consistently detected altered immune pathways in both the SCN and DRG over the course of disease. To support these results, we further used multiplex analysis to assess protein changes in the SCN of diabetic mice; we found that multiple immune molecules were upregulated at both early and later stages of disease. In particular, we found that matrix metalloproteinase-12 was highly upregulated in microarray and multiplex data sets suggesting it may play a role in disease progression.

INTRODUCTION

Type 2 diabetes (T2D) is a metabolic disorder which arises primarily from obesity and is characterized by hyperglycemia, hyperlipidemia, and impaired insulin signaling. As T2D progresses, these factors lead to microvascular complications which significantly increase both the morbidity and the mortality of afflicted patients^{1, 2}. In particular, diabetic peripheral neuropathy (DPN) is one of the most common diabetic complications, with 50% of T2D patients developing DPN³. DPN is characterized by progressive loss of sensation in the limbs, pain, and allodynia. It also increases infection risk and the rate of foot ulcers that can lead to amputation ³.

Despite the prevalence of T2D and the morbidity associated with DPN, the mechanisms of neurodegeneration are poorly understood. Numerous pathways, including inflammation ⁴, reactive oxygen species (ROS) formation^{5, 6}, mitochondrial dysfunction ⁷, and endoplasmic reticulum stress ⁸ are implicated and may play a role in disease progression. To elucidate potential mechanisms, previous studies have utilized either microarray or RNA-Seq analyses to identify genes which are up- or downregulated in the peripheral nervous system during DPN^9–14. However, these studies have focused on specific time frames and do not examine changes over the course of disease; pathways that are upregulated early or late in disease may not necessarily play an important role over the entirety of the disease course.

Using microarrays as a transcriptomic platform, we identified DPN-associated pathways in the sciatic nerve (SCN) and dorsal root ganglia (DRG) at 8, 16, and 24 weeks of age in the well characterized db/db mouse model of T2D and DPN. Previously published 16 and 24 wk transcriptomic datasets^{9, 11} were preprocessed with new datasets using a unified analysis pipeline. Two forms of data analyses were then used in parallel to identify altered molecular pathways in DPN. First, using Self-Organizing Map (SOM) analysis, pathways of interest were organized into distinct clusters which followed similar kinetic patterns over the course of disease. Next, Differential Expression Analysis was used to identify up- and downregulated differentially expressed genes (DEGs) at all three time points. By comparing both sets of analyses, we found that pathways associated with inflammation and immune regulation were altered at all time points in both the SCN and the DRG. To validate these data, multiplex analysis was used to assess protein changes in the SCN during early (8 wk) and late (24 wk) stages of DPN. Several pro- and anti-inflammatory markers, including interferon (IFN)-γ and interleukin (IL)-10, were upregulated over the course of disease. Also, consistent with our previous observations^{14, 15}, we found that matrix metalloproteinase-12 (MMP-12) was highly upregulated in DPN. Together, these data suggest that the immune system plays a key role during DPN and that aspects of the immune system may represent a viable therapeutic target.

METHODS

Animals

Three cohorts of C57BLKS (BKS.Cg-m+/+Lepr^db/J; stock number 000642, Jackson Laboratory, Bar Harbor, ME) db/db mice, a well-characterized murine model of T2D, were used for the study. Control (db/+) and diabetic (db/db) mice were fed a standard diet (5LOD, 13.4% kcal fat for 8/24 wk; or AIN76A, 11.5% kcal fat for 16 wk; Research Diets, New Brunswick, NJ). Mice were cared for in a pathogen-free environment by the University of Michigan Unit for Laboratory Animal Medicine. Mice were, euthanized at 8, 16 or 24 weeks of age using a lethal injection of pentobarbital (Vortech, Dearborn, MI). Animal protocols were approved by the University of Michigan University Committee on Use and Care of Animals.

Metabolic and Neuropathy Phenotyping

Control db/+ and db/db mice were phenotyped for body weight, fasting blood glucose (FBG), and glycated hemoglobin (GHb) levels (n = 5–6 for each group per time point). FBG levels were measured with an AlphaTrak Glucometer (Abbott Laboratories, Abbott Park, IL) for 8 and 16 wk mice and a standard Glucometer (OneTouch; LifeScan Inc., Milpitas, CA) for 24 wk mice. GHb levels were determined using a Glyco-Tek Affinity column (Helena Laboratories, Beaumont, TX) at the Michigan Diabetes Research and Training Center Chemistry Core. Peripheral nerve function was assessed at 8, 16, and 24 wk according to Diabetic Complications Consortium guidelines (https://www.diacomp.org/shared/protocols.aspx). Motor (sciatic) and sensory (sural) nerve conduction velocities (NCVs), hind-paw withdrawal latency from a thermal stimulus, and intraepidermal nerve fiber density were measured for large and small fiber nerve function, and small fiber loss using our previously published protocols^16–19. Data from 16 wk and 24 wk cohorts are previously published^{9, 11}.

Microarray Data Analyses

Lumbar DRG were harvested, and nerve roots removed under a dissecting microscope. A 15 mm segment of SCN was harvested immediately proximal to the sural/tibial/peroneal trifurcation, and adherent fat and connective tissue removed under a dissecting microscope. Total RNA was collected from SCN and DRG at 8 wk (db/+, n = 5; db/db, n = 6), 16 wk (db/+, n = 6; db/db, n = 6), and 24 wk (db/+, n = 6; db/db, n = 6) using the silica gel-based isolation protocol RNeasy Mini Kit (QIAGEN, Valencia, CA) at each time point. RNA quality was measured using a 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA). Samples meeting RNA quality criteria were analyzed by microarray as previously described ¹¹. In brief, total RNA (75 ng) was amplified and biotin-labeled using the Ovation Biotin-RNA Amplification System (NuGEN Technologies Inc., San Carlos, CA) per the manufacturer’s protocol; amplification and hybridization were performed by the University of Michigan DNA Sequencing Core using the Affymetrix GeneChip® Mouse Genome 430 2.0 Array (Affymetrix, Santa Clara, CA). These studies include previously published transcriptomics datasets (16 wk SCN and DRG ⁹ ; 24 wk SCN¹¹).

Quality Control

The quality of the microarray CEL files were assessed using the affyAnalysisQC R package (http://arrayanalysis.org) with Bioconductor (www.bioconductor.org). The Robust Multi-array Average method with the BrainArray Custom Chip Definition File version 19 ²⁰ was applied to normalize each group of microarray data using GenePattern (http://genepattern.broadinstitute.org/gp/).

Self-Organizing Map Analysis

Patterns of gene expression were identified during DPN progression in SCN and DRG in an unbiased manner using SOM analysis. To compare the gene expression changes across the groups, fold-change (FC) values of all genes at 8, 16, and 24 wk for each tissue were calculated. Due to biological and technical variation resulting from different harvest time points and microarray batches, a log₂-transformed fold-change ratio between db/+ and db/db was used for SOM input data. To enhance the visualization of SOM algorithms, Melikerion (Version 1.2.1) software (http://www.finndiane.fi/software/melikerion/) was used with a 7×7 grid structure ²¹. Genes having a similar FC pattern across the groups were gathered together in cells called ‘modules’. Functional enrichment analysis using DAVID (http://david.abcc.ncifcrf.gov) based on Gene Ontology with a Benjamini-Hochberg (BH) corrected p-value (<0.05) identified the enriched functions in each module ²². Pathway enrichment analysis was then performed to identify significantly enriched canonical pathways with a BH corrected p-value (<0.05) for the clusters of interest using Ingenuity Pathway Analysis software (www.qiagen.com/ingenuity).

Differential Expression Analysis

Conventional differential analysis was performed by pairwise comparisons between groups (db/+ vs. db/db) for each group in SCN and DRG. DEGs were determined using the Significance Analysis of Microarrays tool ²³ with an estimated false discovery rate of <0.05 on the basis of 1,000 permutations. Next, DEG sets were identified at each time point to identify unique and common DEG sets between time points. To functionally characterize DEG sets common between all three time points, pathway enrichment analysis was then performed to identify significantly enriched canonical pathways with a BH corrected p-value (<0.05) using Ingenuity Pathway Analysis software (www.qiagen.com/ingenuity).

Multiplex Analysis

Cytokine levels were analyzed in SCN from an additional cohort of 8- (n = 6 per group) and 24-wk (n = 6 per group) db/+ and db/db mice as previously described ¹⁵. Briefly, SCN samples were homogenized and cytokine concentrations in the lysate were analyzed using MILLIPLEX xMAP magnetic bead technology (Millipore, Billerica, MA) with an MMP-12 (MMMP3MAG-79K) and custom cytokine/chemokine panel (MCYTOMAG-70K). Multiplex assays were run on a Bio- Plex200 multiplex array system with Bio-Plex Manager™ 6.0 software (BioRad, Hercules, CA) according to the manufacturer’s instructions. Protein levels of 17 immune markers were determined: eotaxin, IFN-γ, IL-1α, IL-1β, IL-4, IL-6, IL-10, IL-13, IP-10, KC, MCP-1, MIG, MIP-1α, MIP-1β, RANTES, TNF-α, and MMP-12.

Statistical Analysis

F-tests were used to determine whether data sets were normally distributed. For phenotypic measurements, which were normally distributed, statistically significant differences were determined using a two-tailed t-test. For Multiplex assays, which were non-normally distributed, comparisons were performed using the non-parametric Mann-Whitney U test. All tests were performed using Prism 6 software (GraphPad Software, La Jolla, CA).

RESULTS

Project Workflow

We attempted to identify key pathways in DPN by assessing gene expression changes at multiple time points in the db/db mouse model of T2D (Figure 1). First, db/+ control mice and db/db diabetic mice were phenotyped at 8, 16, and 24 wk to verify metabolic changes and the development of DPN. Next, SCN and DRG were harvested at 8, 16, and 24 wk. Gene expression was assessed using microarray technology, and parallel analyses using SOM and DEG identified pathways of interest. Results were cross-referenced and changes in the immune system were validated using multiplex protein analysis.

Figure 1. Study workflow.

Control db/+ and diabetic db/db mice underwent phenotypic assessment at 8, 16, and 24 weeks of age. Mice were euthanized at each time point and the sciatic nerve and dorsal root ganglia collected; total RNA was collected from each tissue type, gene expression assessed using microarray, processed, and checked for quality. Processed data were then analyzed in parallel using SOM or DEG analysis. Functional enrichment and canonical molecular pathways associated with altered gene expression were identified using DAVID and IPA. Finally, multiplex was used to assess protein expression levels and validate pathways detected using SOM and DEG analyses.

Metabolic and Neuropathy Phenotyping

To verify that db/db mice develop long-term metabolic changes as well as DPN, body weights, and hyperglycemia for db/+ and db/db mice at 8, 16, and 24 wk of age were measured (Supplementary Figure 1). For all time points, db/db mice were significantly heavier than db/+ mice (Supplementary Figure 1A). FBG and GHb were significantly increased in db/db mice as well, particularly at later time points (Supplementary Figure 1B, 1C).

In parallel, we measured four standard metrics of neuropathy at 8, 16, and 24 weeks of age: sciatic motor NCV (Supplementary Figure 1D), sural NCV (Supplementary Figure 1E), hind-paw withdrawal latency (Supplementary Figure 1F), and intraepidermal nerve fiber density (IENFD) (Supplementary Figure 1G). IENFD was significantly reduced in 16 and 24 wk db/db mice compared to controls, while sural and sciatic NCVs were significantly reduced at all time points. Finally, hind-paw withdrawal latency was significantly increased at 16 and 24 wk of age. Together, these data indicate that db/db mice display metabolic changes over the course of their life and develop DPN that is particularly pronounced by 16 and 24 wk of age.

Self-Organizing Map Analysis

We next attempted to identify mechanistic pathways that are altered at all time points of DPN that can therefore serve as viable therapeutic targets. We used microarray technology to assess gene expression levels in the SCN and DRG of db/+ and db/db mice. First, to verify appropriate tissue dissections, we confirmed that expression levels of the skeletal muscle-specific transcript, myogenic differentiation 1 (Myod1) was negligible in DRG and SCN tissue, while positive controls were present in DRG (advillin, Avil) and SCN (peripheral myelin protein 22, Pmp22) (Supplementary Table 1). Using SOM analysis, we examined changes in gene expression patterns across the 8, 16, and 24 wk time points. For each tissue, the 17,856 genes detected were organized into a SOM consisting of 7×7 modules (Figure 2A, B). Each module represents a unique pattern of gene expression across the three time points. To visualize changes in gene expression in db/db mice compared to db/+ mice, modules with large expression changes are represented by red, whereas genes with smaller expression changes are blue. Each module was subjected to DAVID analysis to generate heat-maps for both SCN (Supplementary Figure 2) and DRG (Supplementary Figure 3). Adjacent modules with similar patterns and large changes in gene expression were grouped together into clusters. These clusters were then analyzed using IPA to identify canonical molecular pathways that were significantly altered in SCN and DRG (Tables 1–3).

Figure 2. Self-Organizing Map analysis.

SOM analysis was used to identify coherent patterns of gene expression in sciatic nerve (A) and dorsal root ganglia (C) at 8, 16, and 24 weeks of age. Genes with similar expression patterns were organized into modules; those with small expression changes relative to control mice organize into the center of the map (blue hexagons) while those with large changes organize in the corners (red hexagons). Modules with similar kinetic patterns were organized into clusters. Clusters from each tissue type were examined at 8, 16, and 24 weeks to determine how gene expression changed over the course of disease.

Table 1.

Pathway enrichment analysis of Sciatic Nerve Cluster 2. Significantly enriched canonical pathways among the shared genes in modules 7.6 and 7.7 from the SOM analysis using IPA^*

Ingenuity Canonical Pathways	p-value	Genes
TREM1 Signaling	.0380	CXCL3, Naip1, TYROBP, TLR1, TLR7, LAT2, Tlr13, ITGAX
Granulocyte Adhesion and Diapedesis	.0380	ITGB2, CXCL3, CLDN15, IL1RN, Ccl8, CCL3L3, MMP8, FPR2, CL17, CCL22, MMP12, IL18RAP
Primary Immunodeficiency Signaling	.0380	IL7R, BLNK, PTPRC, ADA, IGHM, Ighg2b
Phagosome Formation	.0380	CLEC7A, MSR1, TLR1, TLR7, PIK3R5, FCER1G, Tlr13, Ighg2b, FCGR3A/FCGR3B
Cell Cycle: G2/M DNA Damage Checkpoint Regulation	.0380	TOP2A, CCNB2, BORA, PLK1, CDK1, CCNB1
B Cell Development	.0380	IL7R, PTPRC, SPN, IGHM, Ighg2b
Agranulocyte Adhesion and Diapedesis	.0430	ITGB2, CXCL3, MYH2, CLDN15, IL1RN, Ccl8, CCL3L3, MMP8, MYL4, CCL17, CCL22, MMP12
LPS/IL-1 Mediated Inhibition of RXR Function	.0470	FMO2, APOC4, GSTA5, APOC2, GSTO1, IL18RAP, SLC27A5, ALDH1A1, IL1RN, CD14, ABCC3, FABP9, PPARGC1A
Leukocyte Extravasation Signaling	.0490	ITGB2, RAC2, NCF1, SPN, ARHGAP9, CLDN15, TIMP1, MMP8, PIK3R5, CYBB, NCF4, MMP12

^*IPA: ingenuity pathway analysis; p-value: Benjamini-Hochberg (BH) corrected p-value.

Table 3.

Pathway enrichment analysis of Dorsal Root Ganglia Cluster 2. Significantly enriched canonical pathways among the shared genes in modules 1.6 and 1.7 from the SOM analysis using IPA^*

Ingenuity Canonical Pathways	p-value	Molecules
Granulocyte Adhesion and Diapedesis	< .0001	CLDN11, SELL, CLDN15, MMP3, IL1RL1, Cldn13, IL1R2, CXCL13, CCL3L3, CXCR2, MMP8, Ccl6, HRH2, C5AR1, SDC1, Ppbp, FPR2, FPR1, SELPLG, Ccl9, IL18RAP, ITGB2, CLDN1, CLDN3, MMP9
Leukocyte Extravasation Signaling	< .0001	RAC2, CLDN11, SPN, MMP3, CLDN15, ARHGAP4, Cldn13, NCF4, SELPLG, BTK, ITGB2, TIMP4, NCF1, ARHGAP9, CLDN1, MMP8, CYBA, CYBB, VAV1, ACTG2, MMP9, CLDN3
Agranulocyte Adhesion and Diapedesis	< .0001	AOC3, CLDN11, SELL, C5AR1, CLDN15, MMP3, Ppbp, Cldn13, Ccl9, SELPLG, ITGB2, CXCL13, CLDN1, CXCR2, CCL3L3, MMP8, ACTG2, Ccl6, MMP9, CLDN3
Atherosclerosis Signaling	.0001	ALOX15, MMP3, CD36, APOC4, CMA1, APOC2, CCR2, PDGFC, TPSAB1/TPSB2, SELPLG, ITGB2, LYZ, S100A8, MMP9, RBP4
LXR/RXR Activation	.0003	TTR, IL1RL1, CD36, APOC4, APOC2, IL18RAP, IL1R2, LYZ, LY96, CD14, S100A8, LBP, MMP9, RBP4
Coagulation System	.0019	PLG, F10, F5, VWF, F13A1, TFPI, PLAT
Phagosome formation	.0081	MRC1, CLEC7A, FCGR2A, TLR7, FCER1A, FCER1G, Tlr13, IGHG1, Ighg2b, FCGR3A/FCGR3B, INPP5D
FcÎ3 Receptor-mediated Phagocytosis in Macrophages and Monocytes	.0083	PLD4, RAC2, NCF1, FCGR2A, HCK, VAV1, ACTG2, FCGR3A/FCGR3B, INPP5D, FGR
Complement System	.0145	ITGB2, SERPING1, C5AR1, C1QC, CFH, C3AR1
Role of Macrophages, Fibroblasts and Endothelial Cells in Rheumatoid Arthritis	.0145	C5AR1, MMP3, IL1RL1, LTB, FZD9, IRAK3, IGHG1, PDGFC, IL18RAP, IL18R1, IL16, IL1R2, F2RL1, TLR7, Tlr13, Ighg2b, FZD2, FCGR3A/FCGR3B, IRAK2
Extrinsic Prothrombin Activation Pathway	.0229	F10, F5, F13A1, TFPI
LPS/IL-1 Mediated Inhibition of RXR Function	.0269	CYP3A7, FMO2, SLC27A2, IL1RL1, APOC4, GSTA5, APOC2, ALAS1, IL18RAP, IL1R2, LY96, FABP4, CD14, ALDH3B1, LBP
Production of Nitric Oxide and Reactive Oxygen Species in Macrophages	.0309	LYZ, NCF1, PTPN6, MPO, MAP3K6, CYBA, APOC4, CYBB, S100A8, APOC2, NCF4, SPI1, RBP4
Hepatic Fibrosis/Hepatic Stellate Cell Activation	.0331	CTGF, LEP, IL1RL1, PDGFC, IL18RAP, IL1R2, LY96, COL13A1, PDGFRA, CD14, LBP, MMP9, PDGFRB
Primary Immunodeficiency Signaling	.0398	PTPRC, BTK, Igha, IGHG1, Ighg2b, AICDA
Inhibition of Angiogenesis by TSP1	.0437	SDC1, CD36, GUCY1A2, MMP9, GUCY1B3
IL-10 Signaling	.0457	IL1R2, CCR1, FCGR2A, IL1RL1, CD14, LBP, IL18RAP
B Cell Receptor Signaling	.0490	PTPRC, BTK, RAC2, EBF1, PTPN6, APBB1IP, MAP3K6, FCGR2A, VAV1, IGHG1, Ighg2b, INPP5D
Communication Between Innate and Adaptive Immune Cells	.0490	CCL3L3, Igha, TLR7, FCER1G, Tlr13, IGHG1, Ighg2b, Ccl9
Natural Killer Cell Signaling	.0490	RAC2, PTPN6, KLRD1, FCGR2A, TYROBP, FCER1G, VAV1, FCGR3A/FCGR3B, INPP5D

^*IPA: ingenuity pathway analysis; p-value: Benjamini-Hochberg (BH) corrected p-value.

Within the SCN, IPA analysis of Cluster 1 detected changes in molecular pathways related to axonal loss, axonal and neuronal differentiation, neuronal projection development, and neurological system processing. These mechanisms were highly upregulated at 8 wk but downregulated by 24 wk (Figure 2B). Despite this, none of the canonical pathways identified using IPA reached the level of statistical significance at all three time points (data not shown). In contrast, pathways in Cluster 2 were upregulated at 8, 16, and 24 wk. Pathways that were significantly altered at all three time points were primarily associated with immune function including immune cell signaling, trafficking, and activity (Table 1).

Pathway changes were also detected in the DRG (Figure 2C). Cluster 1 contained pathways that were downregulated at all time points, particularly 8 and 24 wk (Figure 2D). Using IPA to detect enriched canonical pathways, we found that these downregulated pathways were related primarily to immune function, extracellular matrix function, and molecular signaling (Table 2). Pathways in Cluster 2 of the DRG paralleled those in the SCN: gene expression was upregulated at all time points (Figure 3D). Within Cluster 2 canonical pathways we observed upregulated genes primarily associated with immune functions (Table 3). In addition, several other enriched pathways, notably those related to the complement system and ROS production, play a central role in the immune response.

Table 2.

Pathway enrichment analysis of Dorsal Root Ganglia Cluster 1. Significantly enriched canonical pathways among the shared genes in modules 1.1 and 1.2 from the SOM analysis using IPA^*

Ingenuity Canonical Pathways	p-value	Genes
Hepatic Fibrosis/Hepatic Stellate Cell Activation	< .0001	MYH4, ICAM1, MYL2, LEPR, COL2A1, COL8A1, MYH7, MMP2, IGFBP5, IL6, COL15A1, MET, COL1A1, COL6A3, CYP2E1, TGFB1, TIMP1, NGFR, COL6A5, COL18A1, STAT1, MYL3, COL3A1, MYH1
Agranulocyte Adhesion and Diapedesis	.0002	MYH4, ICAM1, MYL2, CLDN19, MMP16, MYH7, MMP2, CXCL16, HRH1, MMP23B, SELP, Ccl8, Ccl2, CXCL14, ACTC1, ACTA1, MYL3, Ccl7, MYH1, CCL1
Calcium Signaling	.0028	MYH4, TNNT1, MYL2, TNNI2, TNNT3, TNNC2, TNNC1, TRDN, MYH7, ATP2A1, CASQ1, RYR1, TNNI1, ACTC1, ACTA1, MYL3, MYH1
Activation of IRF by Cytosolic Pattern Recognition Receptors	.0076	IFIH1, IRF7, ZBP1, MAVS, STAT2, IRF9, IL6, STAT1, ADAR
Granulocyte Adhesion and Diapedesis	.0191	ICAM1, CLDN19, MMP16, MMP2, ITGB3, CXCL16, HRH1, MMP23B, SELP, Ccl8, Ccl2, NGFR, CXCL14, Ccl7, CCL1
Atherosclerosis Signaling	.0195	COL1A1, PLA2G4A, PON1, ICAM1, LCAT, SELP, TGFB1, PLA2R1, COL2A1, IL6, COL18A1, COL3A1
Death Receptor Signaling	.0209	Naip1, PARP10, ZC3HAV1, APAF1, TNFSF10, PARP3, ACTC1, PARP9, ACTA1, PARP14
Dendritic Cell Maturation	.0380	COL1A1, ICAM1, DDR2, HLA-A, LEPR, NGFR, COL2A1, STAT2, IL6, COL18A1, STAT1, FCGR1A, HLA-DRB5, COL3A1
Inhibition of Matrix Metalloproteases	.0389	ADAM17, MMP23B, TIMP1, MMP16, MMP2, TFPI2
Role of Pattern Recognition Receptors in Recognition of Bacteria and Viruses	.0417	PTX3, IFIH1, IRF7, TGFB1, TLR1, CASP1, Oas1b, MAVS, IL6, OAS3, IL25
ILK Signaling	.0417	MYH4, MYL2, ACTN2, Actn3, MYH7, CCND1, ITGB3, PTGS2, PPP2R1B, ACTC1, Irs3, ACTA1, MYL3, MYH1
Epithelial Adherens Junction Signaling	.0417	MET, MYH4, RRAS2, TUBB6, MYL2, ACTN2, Actn3, MYH7, ACTC1, MYL3, ACTA1, MYH1
Intrinsic Prothrombin Activation Pathway	.0437	COL1A1, COL2A1, COL18A1, FGG, COL3A1
HER-2 Signaling in Breast Cancer	.0490	RRAS2, PARD6B, ERBB3, MMP2, ERBB2, CCND1, AREG, ITGB3

^*IPA: ingenuity pathway analysis; p-value: Benjamini-Hochberg (BH) corrected p-value.

Figure 3. Differential Expression Analysis.

Gene expression from sciatic nerve (A, C) and dorsal root ganglia (B, D) were analyzed using differential expression analysis. (A, B) Significantly up- or down-regulated genes were detected in db/db mice (compared to db/+ control mice) at 8, 16, 24 weeks of age in both tissue types. (C, D) The number of genes that were significantly up- or down-regulated were compared between time points to identify genes that were altered over the entire course of DPN (white numbers).

Differential Expression Analysis

Results of microarray analyses can be highly variable, as different analyses can often yield very different results ^{24, 25}. To complement our SOM analysis and strengthen these results, we also analyzed gene expression in SCN and DRG using Differentially Expressed Gene (DEG) analysis at 8, 16, and 24 wk and interpreted the resulting data using multiple techniques (Figure 1). We began by generating pairwise comparisons (db/+ vs. db/db) for each gene in the SCN and DRG at each time point (Figure 3A, 3B). For each comparison, if the difference between genes in db/+ vs. db/db mice are significantly different then they are considered to be differentially expressed. Thus, these genes are considered to be DEGs regardless of whether they are over- or under-expressed in db/db mice compared to controls. Using DEG analysis, we found that at 8 wk, 3,241 (SCN) and 3,438 (DRG) genes were significantly up- or down-regulated; at 16 wk, 1,756 (SCN) and 2,115 (DRG) DEGs were identified, while 3,662 (SCN) and 1,835 (DRG) DEGs were identified at 24 wk. Next, to determine which genes are differentially regulated throughout the course of disease, we compared the DEGs from all three time points in the SCN and in the DRG. Of the 8,659 DEGs detected in the SCN, we found that 225 genes were differentially expressed at all three time points (Figure 3C). Similarly, of the 7,388 DEGs in the DRG we identified 361 genes differentially expressed at 8, 16, and 24 wk (Figure 3D).

These two pools of genes (225 SCN, 361 DRG) were then analyzed using two separate methods to identify potential mechanisms of disease. First, a simple analysis was performed whereby the top 20 most upregulated and top 20 most downregulated genes were ranked for each tissue. In the SCN upregulated genes included those associated with bioenergetics but also with numerous aspects of the immune system, particularly Mmp12 (Supplementary Table 2). Downregulated genes included structural genes as well as several genes associated with immunity (Supplementary Table 3). Similarly, genes upregulated in the DRG were associated with immune function (Il1r2, Timp4) (Supplementary Table 4), while downregulated genes were associated with compromised lipid homeostasis (Them5, Pmp2) (Supplementary Table 5).

Next, IPA was used on both the SCN and DRG data sets to identify altered biological pathways within the DEGs at all three time points. Unsurprisingly, within the SCN the most highly enriched pathways were associated with atherosclerosis signaling and obesity-related leptin signaling. Also highly upregulated, however, were pathways associated with immune cell maturation, cytokine signaling, antigen presentation, the complement system, ROS, and graft-vs-host disease (Table 4). IPA analysis of DRG detected enriched pathways primarily related to biosynthesis (Table 5). However, the top enriched pathway was linked to the behavior of macrophages in rheumatoid arthritis, and several other immune-related pathways were detected including the fibrotic response, ROS activity, and pathogen-associated molecular pattern receptors. Together, these data support the SOM analysis and suggest that the immune system plays a role in DPN.

Table 4.

Pathway enrichment analysis of Sciatic Nerve DEGs. Significantly enriched canonical pathways among the shared DEGs at all time points using IPA^*

Ingenuity Canonical Pathways	p-value	Genes
Atherosclerosis Signaling	5.13E-06	Pon1, Pla2g2d, Il18, Apoc2, S100a8, Il1rn, Mmp9, Col2a1, Apoc4
Leptin Signaling in Obesity	1.05E-05	Pik3r5, Adcy2, Pdia3, Lep, Lepr, Adcy8, Npy
Dendritic Cell Maturation	1.45E-05	Il18, Il1rn, Hla-Drb5, Hla-Dmb, Pik3r5, Pdia3, Col2a1, Lep, Lepr, Trem2
LXR/RXR Activation	2.45E-04	Pon1, Il18, Apoc2, S100a8, Il1rn, Mmp9, Apoc4
Hepatic Cholestasis	2.63E-04	Il18, Il1rn, Slco3a1, Adcy2, Prkch, Abcc3, Adcy8, Esr1
Cellular Effects of Sildenafil (Viagra)	3.63E-04	Myl4, Prkg2, Adcy2, Pdia3, Myh10, Pde2a, Adcy8
IL-12 Signaling and Production in Macrophages	4.57E -04	Apoc2, Pon1, Il18, S100a8, Pik3r5, Prkch, Apoc4
Antigen Presentation Pathway	5.37E-04	Hla-Drb5, Hla-Dmb, Pdia3, Psmb9
Complement System	5.37E-04	Itgax, Cfd, C6, C3ar1
Hepatic Fibrosis/Hepatic Stellate Cell Activation	5.89E -04	Mmp9, Myl4, Igfbp3, Col2a1, Myh10, Lep, Lepr, Timp2
P2Y Purigenic Receptor Signaling Pathway	1.38E-03	Gng4, Pik3r5, Adcy2, Pdia3, Prkch, Adcy8
Graft-versus-Host Disease Signaling	1.45E-03	Il18, Il1rn, Hla-Drb5, Hla-Dmb
Sperm Motility	1.51E-03	Pla2g2d, Atp1a4, Prkg2, Pdia3, Prkch, Pde2a
GPCR-Mediated Nutrient Sensing in Enteroendocrine Cells	1.70E-03	Gng4, Adcy2, Pdia3, Prkch, Adcy8
LPS/IL-1 Mediated Inhibition of RXR Function	1.86E-03	Il18, Apoc2, Gsto1, Il1rn, Fabp9, Slc27a3, Apoc4, Abcc3
Relaxin Signaling	2.63E-03	Mmp9, Gng4, Pik3r5, Adcy2, Pde2a, Adcy8
IL-8 Signaling	2.88E-03	Mmp9, Gng4, Pik3r5, Cstb, Itgax, Angpt1, Prkch
Clathrin-mediated Endocytosis Signaling	2.95E-03	Pon1, Apoc2, S100a8, Ubd, Pik3r5, Fgf18, Apoc4
Thrombin Signaling	3.55E-03	Myl4, Gng4, Pik3r5, Adcy2, Pdia3, Prkch, Adcy8
Nitric Oxide Signaling in the Cardiovascular System	3.63E-03	Adrb3, Prkg2, Pik3r5, Prkch, Pde2a

^*IPA: ingenuity pathway analysis; p-value: Benjamini-Hochberg (BH) corrected p-value.

Table 5.

Pathway enrichment analysis of Dorsal Root Ganglia DEGs. Significantly enriched canonical pathways among the shared DEGs at all time points using IPA^*

Ingenuity Canonical Pathways	p-value	Genes
Role of Macrophages, Fibroblasts and Endothelial Cells in Rheumatoid Arthritis	1.51E-03	Sfrp5, F2rl1, Tlr7, Ccnd1, Il18rap, Lta, Prkcb, Pdia3, Il1r2, Frzb, Il6st, Pdgfc
Hepatic Fibrosis/Hepatic Cell Activation Stellate	1.58E-03	Agtr1b, Stat1, Il18rap, Il1r2, Lep, Myh8, Col13a1, Agtr1, Pdgfc
Tetrapyrrole Biosynthesis II	2.09E-03	Uros, Alas2
Superoxide Radicals Degradation	5.62E-03	Gpx7, Nqo1
Heme Biosynthesis II	7.24E-03	Uros, Alas2
Retinol Biosynthesis	1.23E-02	Lrat, Esd, Lipg
Lanosterol Biosynthesis	1.48E-02	Lss
Chondroitin Sulfate Degradation (Metazoa)	2.00E-02	Chil3/Chil4, Mgea5
The Visual Cycle	2.00E-02	Lrat, Rlbp1
Neuroprotective Role of THOP1 in Alzheimer’s Disease	2.09E-02	Hla-A, Nfya, Mapt
Dermatan Sulfate Degradation (Metazoa)	2.24E-02	Chil3/Chil4, Mgea5
S-methyl-5-thio-α-D-ribose 1-phosphate Degradation	2.88E-02	Apip
Formaldehyde Oxidation II (Glutathione-dependent)	2.88E-02	Esd
p70S6K Signaling	3.09E-02	F2rl1, Prkcb, Pdia3, Mapt, Agtr1
GADD45 Signaling	3.09E-02	Ccnd1, Gadd45b
Protein Ubiquitination Pathway	3.55E-02	Usp1, Hspa1a/Hspa1b, HlaA, Usp54, Hsph1, Hspa4l, Usp18, Usp48
Role of Pattern Recognition Receptors in Recognition of Bacteria and Viruses	3.80E-02	Tlr7, Lta, Oas2, Prkcb, Oas1b
UVA-Induced MAPK Signaling	4.07E-02	Stat1, Pdia3, Parp14, Smpd4
Inosine-5′-phosphate Biosynthesis II	4.37E-02	Atic

^*IPA: ingenuity pathway analysis; p-value: Benjamini-Hochberg (BH) corrected p-value.

We cannot exclude the possibility that genetic disruption of leptin signaling contributes to DPN progression in the db/db mouse. A comparison of Lep/Lepr gene regulation between the db/db model and non-genetic models of DPN could address this issue.

Multiplex Chemokine/Cytokine Analysis

To confirm that changes in the immune system occur in DPN, we used multiplex analysis to assess chemokine/cytokine protein levels in the SCN during early (8 wk) and late (24 wk) time points in both db/+ and db/db mice (Table 6). Consistent with our SOM and DEG analyses, we observed significant increases in immune markers, such as IL-10, eotaxin (also known as CCL11) at 24 wk, IFN-γ, and MMP-12. We next compared these increases to those observed in our microarray analysis (Figure 4). While we observed negligible increases or minor decreases in eotaxin, Ifng, and early Il10 gene expression in our micro array data sets, we observed increased gene expression of Il10 at 24 wk and increased Mmp12 expression during early and late DPN. These data further support a central role for the immune system in DPN and highlight several potential therapeutic targets.

Table 6.

Protein Expression in Sciatic Nerve During Early and Late Disease

	8 Weeks of Age			24 Weeks of Age
Immune Marker	db/+	db/db	p-value	db/+	db/db	p-value
Eotaxin (CCL11)	282.5 ± 61.87	382.2 ± 101.6	0.132	115.1 ± 21.7	168.7 ± 60.3	*0.0379
IFN-γ	0	13.4 ± 12.4	0.606	0	6.4 ± 12.7	0.4615
IL-1α	0	10.4 ± 16.5	0.4545	0	5.8 ± 15.2	0.9999
IL-1β	14.3 ± 12.7	10.8 ± 12.9	0.8052	0	7.0 16.2	0.4615
IL-4	0.8 ± 1.6	0.8 ± 1.8	0.9999	0.02 ± 0.1	2.0 ± 3.7	0.4615
IL-6	17.7 ± 28.9	39.3 ± 36	0.1277	0	14.4 ± 27.6	0.4615
IL-10	49.7 ± 8.3	97.1 ± 10.7	**0.0099	55.3 ± 12.1	80.1 ± 17.6	** 0.0099
IL-13	0	7.882 ± 19.3	0.9999	0	0	NA
IP-10 (CXCL10)	113.6 ± 19.0	124.0 ± 15.4	0.4848	65.8 ± 12.4	85.2 ± 23.2	0.1649
KC (CXCL1)	116.0 ± 23.2	98.3 ± 12.2	0.3095	42.3 ± 7.2	42.1 ± 7.6	0.9015
MCP-1	92.1 ± 33.5	103.7 ± 7.8	0.3939	0.8 ± 0.8	2.0 ± 1.9	0.6807
MIG (CXCL9)	144.0 ± 32.1	164.5 ± 22.6	0.2403	79.9 ± 26.6	87.5 ± 19.5	0.3179
MIP-1α	42.6 ± 48.5	72.4 ± 15.1	0.0649	0	13.3 ± 19.3	0.1923
MIP-1β	36.4 ± 26.9	35.5 ± 5.5	0.132	0	4.1 ± 7.3	0.4615
RANTES (CCL5)	0	0	NA	0	0	NA
TNF-α	0	0	NA	0	3.6 ± 9.5	0.9999
MMP-12	3.5 ± 6.8	334.6 ± 130.5	**0.0022	6.9 ± 8.6	38.4 ±14.6	**0.0012

Units = pg cytokine/chemokine per mg total protein lysate; data expressed as mean ± SD; comparisons made using Mann-Whitney,

^*p < 0.05,

^**p < 0.01;

NA = not applicable

Figure 4. Comparison of microarray and multiplex analyses.

Protein expression patterns of four upregulated immune markers were examined at 8 and 24 weeks of age in sciatic nerve using multiplex analysis, and compared to gene expression levels generated using microarray. Columns represent the log2 fold-change in gene expression (solid bars) or protein (checkered bars) compared to control mice at the same time point.

DISCUSSION

Previous studies have examined molecular pathways associated with DPN progression; however, they have done so by examining gene expression during specific intervals. As molecular pathways involved in DPN may be altered during specific time windows, it is therefore important to examine pathway alterations across the entire disease course. Importantly, pathways commonly regulated at all stages of DPN would represent promising therapeutic targets, and the identification of these pathways thus formed the basis of the current study. In order to identify potential pathways of interest, we compared gene expression in the SCN and DRG of control db/+ and diabetic db/db mice at 8, 16, and 24 weeks of age using microarray. Data were then analyzed using two separate methods – SOM and DEG – and functional enrichment and canonical molecular pathways were identified using DAVID and IPA. Finally, gene expression data were verified using multiplex analysis to assess immune protein levels in the SCN at early (8 wk) and late (24 wk) stages of disease.

Metabolic and neuropathic phenotyping of control and diabetic mice demonstrated significant changes in db/db mice compared to controls at every time point. SOM and DEG analysis found increases in immune-associated pathways in both SCN and DRG across all time points. Subsequent protein quantification in SCN using multiplex analysis further found that IFN-γ, IL-10, and MMP-12 are all upregulated during early and late DPN. Together, these data indicate that the immune system is involved at all stages of DPN.

Inflammation is linked to obesity, with markers of inflammation upregulated in mouse models of disease as well as in diabetic patients ^{15, 26–32}. Moreover, altering specific immune pathways can ameliorate insulin resistance and obesity in T2D ^33–37. The role of immunity in diabetic neuropathy in Type 1 Diabetes Mellitus (T1D) has been established as well ^38–40, and involvement of the immune system likely extends to DPN in T2D ^{4, 15, 31, 41, 42}. T2D patients with DPN have higher levels of circulating inflammatory cytokines than both control patients and diabetic patients lacking neuropathy ⁴³; immune cells such as monocytes – which are already more activated in patients with T2D ^44–46 – traffic at increased rates to non-adipose tissue ⁴⁷. However, the exact role of the immune system in DPN is not clear, as the immune system can protect or damage sensitive neuronal tissue depending on the local immune environment ⁴⁸. Moreover, in the case of chronic inflammation, increased levels of both pro- and anti-inflammatory signals are often present simultaneously, further complicating our understanding of potential mechanisms ⁴⁹.

The current study suggests a similar complex role for the immune system in DPN. Our SOM analysis – as well as increased eotaxin protein levels in SCN during late disease – suggests that immune cell recruitment is central to pathology. This is consistent with previous reports demonstrating that obesity results in increased levels of immune chemoattractants ^{36, 50} and that immune trafficking promotes diabetic complications including neuropathy ¹⁵ and nephropathy ⁵¹. Alternatively, the immune system may be playing – or attempting to play – a role in neuronal repair during DPN. SOM analysis and multiplex revealed that IL-10 production and associated pathways are upregulated in db/db mice during early and later stages of disease, and eotaxin – a Th2-associated chemokine – is upregulated at 24 wk as well. Both IL-10 and the Th2 response are associated with protection and tissue repair. Similarly, macrophages and neutrophils are centrally involved in neuronal repair ^52–54 and myelin remodeling ^55,56.

Involvement of macrophages in DPN is supported by previous studies as well our pathway analyses showing increased phagosome formation. Inflammation in obesity is based on the innate immune system ⁵⁷, with monocyte-derived macrophages as the primary drivers of adipose tissue inflammation ²⁶. Previous studies have found that monocytes from T2D patients have a pro-inflammatory profile ^{44–46, 58}, that high glucose levels induce monocyte production of inflammatory signals ⁵⁹, that depletion of macrophages can ameliorate neuropathic pain ⁶⁰, and that macrophages are responsible for clearing debris from dying neurons ⁶¹ and injured peripheral nerves ⁶². In addition to phagocytosis of neuronal debris, macrophages are also necessary for the growth of new axons ⁶³ and play a key role in suppressing local inflammation during the repair process via IL-10 production ⁶⁴. Combined with our analyses highlighting phagosome-associated pathways, these observations suggest that macrophages are an important component in DPN pathogenesis, though their exact role is unclear.

Similarly, neutrophils can serve multiple functions, and their involvement in DPN is supported by our pathway analyses which highlight granulocyte adhesion and diapedesis as a major factor in DPN. Though commonly associated with pro-inflammatory defense against pathogens, neutrophils have been linked to inflammation in obesity ⁶⁵. They are also key cells in initiating the repair response in neuronal tissue ^{53, 54}. Their involvement in DPN may explain the discrepancy we observed between gene expression levels and protein levels: neutrophil granules contain a broad spectrum of cytokines, chemokines, and other immune factors that can be deployed quickly without requiring de novo synthesis ⁶⁶. This aids in the rapid response to pathogens and may explain why we were able to detect increases in specific immune markers with multiplex and not microarray.

The role of innate immune cells such as macrophages and neutrophils in DPN is further reinforced by alterations in pathways associated with toll-like receptor (TLR) signaling. TLRs are used by immune cells to sense the local environment ⁶⁷ and are often upregulated in patients with neuropathy or metabolic syndrome ^{68, 69}. Receptor binding can mediate phagocytosis, and its expression can be induced on monocytes by high glucose levels ⁷⁰. Moreover, saturated fatty acids can activate innate immune cells via TLRs ^71–73 supporting the idea that innate immune cells are being differentially affected in the nervous tissue of db/db mice. However, TLRs are produced by and modulate the activity of a number of cell types, including DRG ⁷⁴ and Schwann cells ^75–77, so their role in DPN may not be limited entirely to the immune system.

Significant changes in soluble immune markers were also observed in SCN of diabetic mice during multiple stages of DPN. These changes are particularly important, as immune cell activity is strongly influenced by the local cytokine milieu ⁴⁸. During both early and late time points, we observed increased levels of IL-10 which is consistent with the SOM analysis. IL-10 is typically associated with a protective immune function, but in neuropathy it can also associated with hyperalgesia ^{78, 79}. Interestingly, we did not see any significant changes in SCN tumor necrosis factor (TNF)-α levels despite previous evidence associating elevated plasma TNF-α with obesity and T2D, and DPN. ^{68, 80–83} There are several possible reasons for this lack of expression. First, although TNF-α is involved in acute hyperalgesia and painful peripheral neuropathy ^{84, 85}, peripheral nerve TNF-α levels do not directly associate with peripheral pain phenotypes across rodent models. Jolivalt et al. reported decreased SCN levels in allodynic, hyperalgesic STZ-rats,⁸⁶ while Saleh et al. reported decreased DRG levels in hypoalgesic STZ-rats. ⁸⁷ Similarly, TNF-α inhibition for 4 weeks in a prevention paradigm improved DPN phenotypes in a rat model of HFD-STZ neuropathy. ⁸⁸ Second, many previous studies linking TNF-α to DPN were conducted in the streptozotocin-induced rodent ^{39, 86, 87, 89} which models autoimmune-driven T1D. Finally, TNF-α can drive neurite outgrowth ⁸⁷, and its absence may be a sign of reduced neuronal repair in DPN.

Finally, we also found that MMP-12 was highly elevated in db/db mice. These changes were observed at the gene expression and protein levels and are consistent with our previous observations ^{14, 15}. We also have preliminary data on MMP12 function, with in situ zymography data indicating an increase in MMP12 activity in the diabetic nerve environment (data not shown). MMP-12 can play multiple roles – both positive and negative – in neuronal tissue depending on the context ⁹⁰. It has been shown to regulate both insulin sensitivity and inducible nitric oxide synthase (iNOS) ⁹¹ – an immune-mediated source of ROS – and can exacerbate existing tissue damage ⁹². The involvement of ROS would be consistent with our RNA data as well as previous studies ^{5, 93}. However, the absence of MMP-12 also delays myelin formation ⁹⁴ and can exacerbate damage to the nervous system in mouse models of multiple sclerosis ⁹⁵. Thus, whether MMP-12 is playing a protective or destructive role in DPN is unclear. One hypothesis is that diabetes causes upregulation of plasminogen activator inhibitor-1 and MMPs, resulting in extracellular matrix remodeling and basement membrane thickening, leading to fibrosis and nerve dysfunction. Further investigation is needed to fully understand the role of MMP12 in DPN.

Together, our data point to the immune system as a major factor in DPN, with innate immune cells likely playing a central role. Multiple forms of analysis – SOM, DEG, and multiplex – all detected upregulated markers of immunity in DPN, and this involvement is consistent during all stages of disease. Interestingly, we observed that the local neuronal immune environment does not necessarily reflect previously reported global immune changes in diabetes. The exact role of the immune response in DPN therefore requires future study. However, the continued involvement of the immune system during all stages of DPN suggests that immune modification may be a viable therapeutic strategy.

Highlights.

• We examined longitudinal transcriptomics patterns in SCN and DRG of db/db mice

• Genes/pathways regulated across all stages of DPN are promising therapeutic targets

• Immune-associated pathways were increased in both SCN and DRG at all time points

• IFN-γ, IL-10, and MMP-12 proteins were upregulated in SCN during early and late DPN

• These data indicate specific immune system involvement at all stages of DPN