As DR occurs in Type I diabetes patients in spite of insulin therapy, this study sought to characterize the limitations of insulin replacement in normalizing diabetes-induced molecular changes in the retina by identifying gene expression alterations resistant to normalization by chronic insulin therapy. The use of a Type I diabetes animal model allows for the examination of retinal gene expression with and without insulin treatment and at different durations of diabetes in a manner not possible in humans.
Gene expression alterations normalized with insulin treatment
As observed in the microarray analysis, the majority of diabetes-induced changes were normalized with insulin treatment. In confirmatory qPCR experiments that also examined a one month duration of diabetes, some of these normalized genes were further subdivided into "rescued" and "prevented" categories. The expression patterns of Kcne2, Nppa, and Dcps provide examples of how restoration of insulin and euglycemia can normalize diabetes-induced gene expression alterations that have occurred before initiation of therapy to non-diabetic levels. Additionally, a number of changes not evident at the one month time point were prevented from occurring with continued diabetes by insulin therapy. The majority of these genes examined in the confirmation experiments (i.e., Car13, Carhsbp1, Chi311, Timp1, C1inh, C1s, Hspb1, Icam1, Lgals3) are proinflammatory signalling molecules and have been described elsewhere [14, 16, 18, 22, 23]. These data provide clear evidence that animal model testing of novel DR treatments should be performed in the context of insulin replacement to more closely mimic the clinical scenario and assess the combinatorial effects of any potential therapies, as insulin therapy alone is insufficient to fully restore the normal retinal transcriptomic phenotype.
Gene expression alterations not rescued by insulin therapy
In this study we identified a number of gene expression changes evident at 1 month of diabetes that were not reversed by chronic insulin therapy. The basement membrane anchoring filament protein Lad1 contributes to epithelial stability  and is regulated by both glucocorticoid receptor activation  and LPS stimulation . Similarly, Mt1a is expressed in vascular components of the blood-retina barrier  and is regulated by pro-inflammatory signalling molecules, including factors released by Müller glia . The regulation of these vascular function genes indicates a potential mechanism through which inflammation may contribute to vascular abnormalities in the diabetic retina. Several other inflammatory response genes induced with diabetes were also not rescued by insulin treatment. Litaf1, an inducer of TNFα expression , is a pro-inflammatory modulator induced in monocytes and macrophages in response to inflammatory stimuli such as LPS . Txnip inhibits the antioxidant molecule thioredoxin, increases proinflammatory gene expression and signalling cascades . Inhibition of Txnip in retinal endothelial cells prevents pathologies associated with early diabetic retinopathy (i.e., inflammation and gliosis) . The induction of Lad1, Litaf1, Mt1a, and Txnip in the retina after 1 month of diabetes and the continued induction of these transcripts at three months of diabetes despite insulin replacement for the past 1.5 months suggest persistent alterations in glial and inflammatory processes, even with normalization of blood glucose levels. Rgr, a G protein-coupled receptor expressed almost exclusively in the retina  and localized to Müller glia and the retinal pigment epithelium , encodes a photoisomerase that contributes to pigment regeneration in photoreceptors. In Rgr-/- mice, dark adaptation is delayed and attenuated , suggesting that the suppression of Rgr observed could play a role in the well described deficits in dark adaptation in early DR. The expression pattern of these genes is consistent with the metabolic memory phenotype and demonstrates that even a one month period of loss of insulin signalling and hyperglycemia can result in molecular alterations that are unresponsive to insulin treatment. Recent work suggests that these metabolic memory prototypical changes may be regulated through epigenetic mechanisms [36, 37].
Diabetes-induced gene expression changes not prevented by insulin treatment
In addition to the gene expression changes at one month of diabetes Not Rescued by insulin treatment, the gene expression alterations identified in this study Not Prevented by insulin replacement and re-establishment of euglycemia are of importance for understanding DR in the context of insulin therapy and identifying targets for adjuvant therapies. The Not Prevented phenotype genes were expressed at non-diabetic control levels at one month of diabetes, but became significantly dysregulated with a longer duration (three months) of diabetes regardless of insulin treatment for the last one and a half months. Several proinflammatory genes exhibited this expression pattern. Ccr5, which mediates leukocyte attraction via chemokine receptor signalling, plays well characterized roles in inflammation . In Type 1 and 2 diabetic patients, plasma Ccr5 is elevated and correlates to % glycosylated HbA1c levels [39, 40]. We observed a similar positive correlation of CCR5 gene expression to glycosylated hemoglobin levels. Jak3, a proinflammatory gene expressed by immune cells and activated by interleukin receptors, is regulated by inflammation and tissue damage in the retina  in models of photic damage and retinal degeneration. We have previously reported the upregulation of Jak3 at three months of untreated diabetes in the STZ rat model [14, 16] and the failure of insulin to prevent upregulation of Jak3 suggests that insulin therapy is insufficient to prevent some aspects of retinal inflammation with diabetes. Increased Qprt expression is also associated with tissue damage and neurodegeneration . The protein product of this gene catalyzes the breakdown of neurotoxic quinolinate in the tryptophan nicotinamide adenine dinucleotide pathway, preventing the intracellular accumula tion of toxic compounds and maintaining neuronal function. By regulating actin/cofilin dynamics necessary for cytoskeletal plasticity, Ssh3 influences neuronal function and has been localized to the inner retinal layer in the eye , suggesting a role in ganglion cell function.
In addition to these gene expression changes, several others were only partially prevented by insulin therapy. A2m, a cytokine transporter induced by interleukin and predominantly expressed in blood vessel walls and the retinal pigment epithelium in the retina has also been described as systemically elevated in diabetes patients . In retina of diabetic rodents, A2m expression increases in astrocytes and the inner limiting membrane, suggesting a broad inflammatory insult to both the neural and vascular retina . Increased Gfap, a hallmark of gliosis, also increases with inflammation and diabetes [46–48]. This gene is expressed by astrocytes and Müller glia throughout the retina and likely contributes to inflammatory cascades and neurovascular dysfunction in diabetic retinopathy. Npr3, contributes to vascular regulation of electrolyte concentrations and fluid balance [49, 50]. The failure of insulin to prevent dysregulated Npr3 expression may lead to an abnormal extracellular environment in the retina that has the potential to induce inflammation and neuronal dysfunction.
Together, the findings of this study demonstrate the ability of chronic insulin therapy, initiated after one and a half months of uncontrolled diabetes, to normalize the majority of retinal gene expression changes. However, chronic insulin therapy fails to completely restore the retinal transcriptome to a non-diabetic control phenotype. Additionally, at least some of these transcript changes are reflected in altered abundance of the corresponding proteins. Given the well supported hypothesis that a feed-forward cycle of inflammation and neurovascular dysfunction is a causative factor in DR development [51, 52], it is not surprising that the majority of genes dysregulated with diabetes and not normalized by insulin are implicated in inflammatory responses and neuronal and vascular function. As such, these changes may represent the first stages of this feed forward cycle. Some previously reported pro-inflammatory molecules (e.g, IL-1β and TNFα) were not detectable, suggesting a low level of mRNA expression. Recent reports have emphasized the dependence of the duration of hyperglycemia on the induction of these factors . Future studies will need to assess whether the observed induction of pro-inflammatory factors, such as Ccr5, Litaf, Txnip, and Jak3, result in further activation of prototypical inflammatory signals at a later time.
These results also agree with the clinical findings of past hyperglycemic periods continuing to confer an increased the risk of DR and other diabetic complications long after restoration of euglycemia . Additional studies are needed to examine other complications for similar molecular profiles. This ability of insulin to normalize much, but not all, of the non-diabetic control transcriptomic phenotype may attenuate or delay the development of DR pathology, but ultimately the continued dysregulation of neuronal, vascular, and inflammation related genes may lead to more classical aspects DR development such as edema, leukostasis, and neuronal apoptosis.
The regulatory mechanisms underlying the retinal gene expression changes observed in this study remain to be determined. The nearly complete normalization of the majority of diabetes-induced changes (Rescued and Prevented categories) by insulin treatment points to their regulation by insulin signalling and/or hyperglycemia. Expression of genes in the Not Rescued category was not correlated to blood glucose or glycosylated hemoglobin and first occur after just one month of diabetes and before we have observed changes in retinal vascular permeability or apoptosis . This indicates that the dysregulation of these genes is most likely not due to irreversible tissue damage. Previous reports suggest that many of the genes observed to be treatment resistant are localized to Müller, endothelial, and neuronal cell populations. While further studies are necessary to localize these changes to specific cell types, these results suggest that the observed changes are not the result of deposited leukocytes. An alternate mechanism for the un-responsiveness of these genes may be through epigenetic mechanisms such as altered chromatin status or differential DNA methylation of promoter regions [37, 54]. Genes in the Not Prevented and Partially Prevented groupings are unchanged at 1 month of uncontrolled diabetes but are dysregulated at three months of diabetes despite blood glucose control for the prior 1.5 months. These dysregulations may occur, despite insulin therapy, as a secondary effect of the genes not rescued with insulin or due to the non-physiologically regulated release of insulin in these rats.
Future studies will need to examine the effects of intensive insulin therapy or beta cell transplantation on normalization of retinal gene expression. Ultimately, protein expression studies and localization of changes to specific cellular populations will help provide the necessary distinction between adaptive and pathological changes. We have already characterized a small number of insulin therapy resistant changes in retinal protein expression with diabetes  and these transcriptomic findings suggest that the quantitative proteomic analysis of retinal protein expression not normalized by insulin treatment is warranted. Individually, the genes discovered in this study to be not normalized by insulin treatment are targets for functional analyses using genetic and pharmacological approaches. Drug discovery efforts that target diabetes-related molecular alterations that are not normalized by insulin are needed to develop an adjuvant therapy that, in conjunction with insulin replacement, restores retinal gene expression and tissue function to a healthy state.