A habit of binge-eating during low blood sugars can wreak havoc on your blood sugar levels, your energy, your weight, and your daily life.
The blood sugar roller coaster that often accompanies over-eating during hypoglycemia is exhausting.
In this article, we’ll discuss how to stop the cycle and habit of binge-eating during low blood sugars.
What is hypoglycemia?
In the human body, a blood sugar level below 70 mg/dL will interfere with your brain and your entire body’s ability to function properly. Even the most basic tasks, like walking or speaking, can become extremely difficult the lower your blood sugar drops below 70 mg/dL.
Your brain relies on a second-by-second delivery of glucose from your bloodstream in order to function. Without enough glucose (sugar) in your bloodstream, your brain and entire body will struggle to function.
The intense food cravings during a low blood sugar are really coming from your brain, pleading, “Feed me! Feed me!”
Left untreated, low blood sugars can lead to seizures and death.
Here are 5 steps to stopping the habit of binge-eating during low blood sugars.
Identify the symptoms of your low blood sugar
Acknowledge your current habit (and its consequences) around lows and food
Choose 3 specific fast-acting carbohydrates as your primary treatment for lows
Distract yourself with something else…
Reinforce this mantra: “I do have control over what I eat when I’m low.”
Let’s take a closer look.
Identify the symptoms of that specific low blood sugar
Hunger is one of the many telltale signs that your blood sugar has dropped below a safe level, but it’s important to identify other symptoms of low blood sugar to help reinforce good habits around how you treat low blood sugars in general.
By identifying the symptoms that you personally experience during a low blood sugar, you’re reminding yourself that those intense cravings for food aren’t random or normal hunger, they are directly tied to your blood sugar, and they aren’t rational.
When you do notice your blood sugar is low, take just 20 seconds to identify your symptoms — including those severe cravings — and remind yourself they are an irrational aspect of hypoglycemia.
Yes, you need some food, but you don’t need all the food.
Acknowledge your current habit (and its consequences) around low blood sugars and food
Routinely binge-eating during low blood sugars comes with some not-so-subtle consequences. And a big part of breaking that binge-eating habit comes down to fully acknowledging it.
You could write it down, make a YouTube video, or record a voice-memo about that you send to your mom — whatever works for you! The goal is to simply look at the habit and the vicious cycle it creates from a distance.
Things to consider…
How many extra calories do you consume every day or week because of binge-eating during lows?
that you’re always prepared (Gummy Lifesavers, for example, survive in hot and cold temperatures, only require 3 or 4 to treat the average low, and one bag can treat dozens of lows).
that you can control exactly how many grams of carbohydrates you consume. Mild lows may only need 6 to 8 grams of carbohydrate, while severe lows may need 15 to 20 grams.
that you’re consuming something that digests easily and quickly. Using high-fat or high-protein foods to treat lows will only make your symptoms and cravings last longer.
that you’re consuming “medicine food.” Whichever 3 foods you choose (for example: jelly beans, juice box, fruit snacks), you can think of those things as “medicine foods.”
you won’t use lows an excuse to binge-eat yummy treats like brownies or ice cream. Instead, give yourself permission to eat those treats in a more controlled environment when your blood sugar and your appetite are rational and stable.
When you choose fast-acting carbohydrates like fruit snacks, jelly beans, gummy Lifesavers, Smarties, glucose tabs, etc. to treat lows, you can prepare by storing them in your purse, desk, backpack, nightstand, gym bag, jacket pocket, and glove compartment in your car.
The more you are prepared for treating lows properly, the more easily you’ll be able to manage them properly, too!
Distract yourself with something else…
Of course, we all know the symptoms of low blood sugar can persist long after your blood sugar has risen to a safe level. And of course, extreme hunger is one of the most persistent symptoms (followed by the desperate desire for a cozy nap!)
This means that after you treat the low, you need to distract yourself with something until those intense cravings for more food quiet down and disappear.
You may find that simply noshing on something like gum or carrots can provide that satisfying feeling to your brain because you’re chewing without consuming starchy carbohydrates.
Here are a few distraction techniques:
Chug some ice-cold water
Chew on carrots or celery
Pop some gum or TicTacs
Record a voice memo to yourself about your ability to control what you eat
Clean your kitchen
Squeeze a stress ball
Lie down in bed and close your eyes
The goal is to either distract your mouth or your hands or both!
Reinforce this mantra: “I do have control 0ver what I eat when I’m low.”
Truly, the biggest contributor to your habit of binge-eating during low blood sugars is the belief that you have no control over what you’re doing. That you are helplessly putting more food in your mouth because your brain is desperately begging for more
But you do have control. You do have rational thought and logic and reason at your disposal, you just have to make the choice to use those things.
Sure, I’d like to eat 3 bowls of Honeycombs mixed with Peanut Butter Captain Crunch every time my blood sugar dips below 50 mg/dL, but I know what that will do to the rest of my day, to my blood sugar, to my energy, and how guilty and regretful I’ll feel because of those things.
If you continue to tell yourself you have no control over how much you eat during lows or that you deserve to eat as much as you want during lows, then you’ll continue to feed that belief and that vicious habit.
If you take a moment to remind yourself, “I do have control over what I eat when I’m low,” you will be taking responsibility for how you treat lows and for your actions in general.
It’s actually pretty darn empowering. And quite rewarding, because treating a low with a reasonable amount of fast-acting carbohydrates means bringing your blood sugar up to a reasonable blood sugar level.
This means you don’t have to deal with a blood sugar roller coaster and the guilt of overeating. Instead, you get to continue with your day, proud of your ability to exert control over your own behavior.
Self-discipline produces…pride. Low blood sugars could become a source of pride and integrity in how you manage your health. And that could easily start to spill-over into other areas of your health, too.
Suggested next posts:
If you found this guide to avoiding binge eating during low blood sugars useful, please sign up for our newsletter (and get a free chapter from the Fit With Diabetes eBook) using the form below. We send out a weekly newsletter with the latest posts and recipes from Diabetes Strong.
Success! Now check your email to download the eBook chapter.
To identify the factors mediating the progression of diabetic nephropathy (DN), we performed RNA sequencing of kidney biopsy samples from patients with early DN, advanced DN, and normal kidney tissue from nephrectomy samples. A set of genes that were upregulated at early but downregulated in late DN were shown to be largely renoprotective, which included genes in the retinoic acid pathway and glucagon-like peptide 1 receptor. Another group of genes that were downregulated at early but highly upregulated in advanced DN consisted mostly of genes associated with kidney disease pathogenesis, such as those related to immune response and fibrosis. Correlation with estimated glomerular filtration rate (eGFR) identified genes in the pathways of iron transport and cell differentiation to be positively associated with eGFR, while those in the immune response and fibrosis pathways were negatively associated. Correlation with various histopathological features also identified the association with the distinct gene ontological pathways. Deconvolution analysis of the RNA sequencing data set indicated a significant increase in monocytes, fibroblasts, and myofibroblasts in advanced DN kidneys. Our study thus provides potential molecular mechanisms for DN progression and association of differential gene expression with the functional and structural changes observed in patients with early and advanced DN.
Diabetic nephropathy (DN) is the most common cause of end-stage renal disease in the U.S., and its incidence is rising worldwide despite glycemic and blood pressure control regimens (1). Therefore, elucidating mechanisms that mediate the early stage of DN may help us to identify new targets for better preventive and therapeutic measures.
Genome-wide gene expression profiling can be useful in providing a global picture of the disease pathogenesis and to identify potential new biomarkers and drug targets for DN. Several previous studies examined the transcriptomes of human diabetic kidney samples. The European Renal cDNA Bank (ERCB) consortium examined the gene expression of the tubulointerstitial compartments of European Caucasian patients with DN compared with those from pretransplant donors and minimal change disease by microarray analysis, which identified the expression of genes related to the nuclear factor-κB–driven inflammatory pathway to be highly associated with the progression of DN (2). The ERCB consortium also found that vascular endothelial growth factor A (VEGFA) expression was downregulated in the tubulointerstitial compartment of DN, which is associated with interstitial vascular rarefaction (3). Berthier et al. (4) analyzed the transcriptome of glomerular and tubulointerstitial compartments in patients with early DN from a Pima Indian cohort, patients with progressive DN from ERCB, and control subjects without diabetes and identified the Janus kinase-STAT pathway to be upregulated in both glomerular and tubulointerstitial compartments in DN. Woroniecka et al. (5) analyzed the glomerular and tubular microarray data from kidney biopsy samples of patients with DN compared with samples from living allograft donors and surgical nephrectomies, which showed the upregulation of RhoA, Cdc42, integrin, and VEGF signaling in the glomerular compartment and inflammation-related pathways in the tubulointerstitial compartment in DN. Nair et al. (6) examined the microarray analysis of the tubulointerstitial compartments samples from the Pima Indian cohort, which included patients with diabetes with normoalbuminuria, microalbuminuria, and macroalbuminuria, and found that the cortical interstitial fractional volume, an index of tubulointerstitial damage, correlated significantly with the transcripts enriched for pathways associated with mitochondrial dysfunction, inflammation, migratory mechanisms, and tubular metabolic functions. In another recent study by Pan et al. (7), microarray analysis of kidney samples from patients with advanced DN and surgical nephrectomies were performed for glomerular transcriptome profiling, which identified SLIT-ROBO GTPase-activating protein 2a (SRGAP2a) as a key gene associated with proteinuria and estimated glomerular filtration rate (eGFR) in patients with DN. Thus, the transcriptomic analyses are of microarray analyses in samples from patients with advanced DN, with the exception of the Pima Indian study. In most of these previous studies, the transcriptomes were profiled in glomerular and tubulointerstitial compartments separately.
Here, we report an RNA sequencing (RNA-seq) analysis of the whole-kidney biopsy samples of patients with early and advanced DN compared with nephrectomy sample tissues from patients without diabetes. We compared gene expression profiles among normal samples, early DN, and advanced DN. We also performed a correlation analysis of genes with renal function (eGFR) and histological parameters in patients with DN. We took advantage of the recently published single-cell RNA-seq (scRNA-seq) data to perform a computational deconvolution analysis to identify the different cell types present in normal and diseased kidneys. Finally, we validated some of the key findings by immunostaining of the kidney tissues from these patients.
Research Design and Methods
Human Kidney Biopsy Sample Collection
A total of 28 patients with biopsy-proven DN hospitalized from January 2015 to December 2016 in Shanghai Jiao Tong University Affiliated Sixth People’s Hospital were enrolled in the study. Kidney tissues were collected through ultrasound-guided kidney biopsy after informed consent was obtained, according to the guidelines of the local ethics committee. Samples were quickly frozen in liquid nitrogen and stored at −80°C before use. Patients were divided into two groups—early DN (n = 6) and advanced DN (n = 22)—on the basis of urinary albumin-to-creatinine ratio (UACR) and renal function (calculated using the MDRD equation) by at least two randomized measurements. Early DN was defined as UACR between 30 and 300 mg/g, eGFR >90 mL/min/1.73 m2, whereas advanced DN was defined as UACR >300 mg/g, eGFR <90 mL/min/1.73 m2. Nine control human kidney samples were obtained from the unaffected portion of tumor nephrectomies. Demographic, blood biochemical characteristic, urine albumin excretion, and kidney function data were collected. The study was approved by the institutional review board at Shanghai Jiao Tong University Affiliated Sixth People’s Hospital.
RNA-Seq and Bioinformatics Analysis
RNA-seq was performed on 28 DN and 9 control samples. Total RNA was extracted using TRIzol (Thermo Fisher Scientific). The RNA quality was evaluated by an Agilent 2100 Bioanalyzer. The total RNA sample was treated using a Ribo-Zero Magnetic Gold Kit to deplete rRNA, and cDNA libraries were prepared and sequenced at BGI using a HiSeq 4000 system. The reads with good quality were first aligned to human reference databases including hg19 human genome, exon, and splicing junction segment and a contamination database including ribosome and mitochondria sequences using the STAR alignment algorithm (8). After filtering reads mapped to the contamination database, the reads that were uniquely aligned to the exon and splicing junction segments with a maximal two mismatches for each transcript were then counted as the expression level for the corresponding transcript. After filtering out the transcripts with low reads (<100) across all samples, the read counts were log2 transformed and quantile normalized at an equal global median value to compare the transcription level across samples. Gene expression data were first adjusted by demographic confounders such as age and sex by extracting residuals from the linear regression model. Principal component analysis was then performed to assess the overall sample distribution using adjusted expression data of all the genes. The differential analysis by limma test (9) was carried out to identify significantly dysregulated genes at P < 0.05 (for advanced DN vs. early DN, we used a false discovery rate [FDR] <0.05 because of a large amount of differentially expressed genes [DEGs]) and 1.5-fold change, which was then subjected to Gene Ontology (GO) function (10) and pathway (Kyoto Encyclopedia of Genes and Genomes, Ingenuity Pathway Analysis, BioCarta, Naba, Panther, Pathway Interaction Database, Reactome, WikiPathways) enrichment analysis by Fisher exact test. Correlation of gene expression and eGFR and histology scores were done on all DN samples. We determined eGFR-related genes at an FDR <0.05 by Pearson correlation test. Correlation with histology scores was determined by multivariate analysis using linear regression model at P < 0.05 for each score. As for deconvolution analysis, CIBERSORT (11) was used to estimate the percentage of each cell type on the basis of the expression value of markers in bulk sequencing samples. A human kidney allograft biopsy single-cell data set was obtained from the Gene Expression Omnibus (GEO) database (GSE109564) and used to identify 16 cell types by Seurat with parameters described in the original article (12). For each cell type, gene expression values were compared against other cell types. Genes with t test P < 0.01 and log2(fold change) >0.5 were selected as markers for corresponding cell types. For each marker, we calculated the mean expression value in each cell type to construct the base matrix for deconvolution analysis. The same strategy for deconvolution analysis was also performed on the basis of mouse single-cell sequencing data obtained from the GEO database (GSE107585) (13).
Histological analysis of all patients with DN was performed by investigators blinded to the experimental groups. A renal pathologist scored detailed histological features from whole-slide images of kidney biopsy samples stained with hematoxylin-eosin, periodic acid Schiff, and Masson trichrome as described previously (14).
Immunostaining of Kidney Sections
Human kidney biopsy samples from these patients were collected as described under protocols approved by the institutional review board. Biopsy samples included 5 cases of early DN, 44 of advanced DN, and 7 of normal tissue adjacent to tumor nephrectomy samples. Immunostaining was performed in all five samples of early DN, five randomly selected samples of advanced DN, and five normal samples from nephrectomies using specific primary antibodies and biotinylated secondary antibodies (Vector Laboratories Inc.). Staining was revealed with avidin-peroxidase (VECTASTAIN Elite; Vector Laboratories Inc.). Slides were mounted in Aqua Poly/Mount (Polysciences, Inc.) and photographed under an Olympus BX60 microscope with a digital camera. The following antibodies were used: retinol-binding protein 4 (RBP4) (ab133530; Abcam) and GRP1 (ab166987; Abcam). The extent of kidney staining in human biopsy samples was semiquantitatively scored on the basis of the percentage of positive staining area divided by total kidney cortex area in the kidney section for each patient. Immunostaining for immune cells was performed with anti-Mrc1 for macrophages (ab64693; Abcam), CD3 for T cells (A0452; Dako), and CD20 (M0755; Dako) for B cells.
Data and Resource Availability
The RNA-seq data set generated and analyzed in the current study is available from the GEO repository (GSE128736). All other data sets generated during the current study are available from the corresponding authors upon reasonable request. No applicable resources were generated during the current study.
RNA-Seq of Kidney Biopsy Samples From Patients With DN
Kidney biopsy samples from 28 patients with DN (22 advanced and 6 early) and 9 normal nephrectomy samples adjacent to tumors were used for RNA-seq analysis. Baseline characteristics are presented in Supplementary Table 1. Patients with early DN had an eGFR >90 mL/min/1.73 m2 and microalbuminuria (UACR <300 mg/g). Patients with advanced DN had either an eGFR <90 mL/min/1.73 m2 or UACR >300 mg/g. Patients with advanced DN had significantly lower eGFR and serum albumin but higher UACR, serum creatinine, and blood urea nitrogen levels than patients with early DN. Patients with early DN had a similar duration of diabetes and other clinical parameters to patients with advanced DN, suggesting that they were likely to be nonprogressors. In addition, their UACR and renal function in the 2–3-year follow-up from the time of collection of kidney biopsy samples remained stable (data not shown). To minimize the tissue processing (i.e., without any dissections or digestions) that may change the gene expression profiles (15), RNA-seq was performed using the whole-kidney biopsy samples, which contain primary kidney cortices. The principal component analysis of the RNA-seq data showed three distinct clusters (Supplementary Fig. 1), indicating large changes in the overall gene signatures among the three groups.
Comparison of RNA-Seq Data of Kidney Biopsy Samples Between Patients With Early DN and Control Patients
We first analyzed the DEGs in the kidney cells of patients with early DN compared with control patients without diabetes. Figure 1A shows a heatmap of the top 25 upregulated and top 25 downregulated genes in early DN, and Fig. 1B shows a volcano plot of the top 50 DEGs (limma P < 0.05 and fold change >1.5 or less than −1.5) (a complete list of DEGs is included in Supplementary Excel File 1). Of note, among the top 25 downregulated genes were transcription factors that are implicated in the pathogenesis of DN (16), such as EGR1, EGR2, EGR3, JUNB, FOS, FOSB, and ATF3. IL6 and CXCL2, inflammatory markers associated with kidney injury in DN (17,18), were also included in the top 25 downregulated genes. GO analysis of DEGs between early DN and control samples is shown in Fig. 2. The top GO terms of the upregulated genes in early DN were related to cellular contraction, containing several myosin and actin genes, and to hormonal regulation and visual perception, containing several genes in the retinoic acid pathway (e.g., RDH8, RDH12, and RBP4). Retinoic acid pathway is shown to be renoprotective in several animal models of kidney disease (19,20). Interestingly, glucagon-like peptide 1 receptor (GLP1R) expression was also highly upregulated in early DN. The pathway enrichment analysis of combined upregulated and downregulated DEGs is shown in Supplementary Fig. 2. Many of the downregulated genes were related to immune response, suggesting that the inflammatory pathway is suppressed in early and nonprogressive DN compared with control nephrectomy samples. This could be due to low inflammation status in the kidney of the patients with early DN but may be a result of mildly increased inflammation in the nephrectomy samples from the normal tissues surrounding tumor.
Differential gene expression analysis for early DN vs. nondiabetic control. A: Heatmap of the top 50 dysregulated genes (25 upregulated genes and 25 downregulated genes) in early DN samples. Data are z normalized for heatmap visualization. Each column represents an individual sample from the control or early DN group. B: Volcano plot of dysregulated genes at limma P < 0.05 and fold change >1.5 or less than −1.5. The top 50 DEGs are labeled on the plot. Log2Rat, log2 ratio; −log10(p), −log10P value.
GO pathway analysis for DEGs for early DN vs. control. The top 20 enriched GO functions for upregulated (Top_Up, pink) and downregulated (Top_Dn, blue) DEGs are shown. −log10(P), −log10P value.
Comparison of RNA-Seq Data of Kidney Biopsy Samples Between Advanced and Early DN
We next examined the gene expression between early DN and advanced DN. Figure 3 shows a heatmap and volcano plot that indicate the top 50 dysregulated genes in advanced DN versus early DN (a complete list of DEGs is included in Supplementary Excel File 1). Consistent with previous findings, many matrix- and inflammation-related genes were upregulated in advanced DN. GO analysis indicated that the top GO terms of upregulated genes are related to the immune response, whereas the downregulated DEGs are largely of transport and metabolic processes (Fig. 4). The pathway enrichment analysis for all DEGs showed that changes in the pathways were related to matrisome, fibrosis, inflammation, and metabolism (Supplementary Fig. 3). These findings are consistent with the high level of inflammation, tubular dysfunction, and metabolic dysregulation observed in the kidneys of patients with advanced DN.
Differential gene expression analysis for advanced DN vs. early DN. A: Heatmap of the top 50 dysregulated genes (25 upregulated genes and 25 downregulated genes) in advanced DN samples. Data are z normalized for heatmap visualization. Each column represents an individual sample from the early DN or the advanced DN group. B: Volcano plot of dysregulated genes at limma FDR <0.05 and fold change >1.5 or less than −1.5. The top 50 DEGs are labeled on the plot. Log2Rat, log2 ratio; −log10(p), −log10P value.
GO pathway analysis for DEGs for advanced DN vs. early DN. Top 20 enriched GO functions for upregulated (Top_Up, pink) and downregulated (Top_Dn, blue) DEGs are shown. −log10(P), −log10P value.
Comparison of RNA-Seq Data Among Control, Early, and Advanced DN Biopsy Samples
We then compared the gene expression in all three groups. Only a moderate number of genes changed in their expression progressively from control to early DN to advanced DN (five increased and seven decreased genes) (Fig. 5A and B and Supplementary Excel File 1). The analysis further identified 148 genes that increased in early DN versus control but decreased in advanced DN versus early DN (Fig. 5C and Supplementary Excel File 1). Interestingly, many of these genes were of the retinoic acid pathway, such as RDH8, RDH12, and RBP4; furthermore, GLP1R was among these genes. These data suggest that retinoic acid and GLP1R agonists might have protective effects in early DN, thus preventing patients with diabetes from the progression of DN. The top GO terms of the 148 genes are shown in Fig. 5D and a full list is found in Supplementary Excel File 1. We also identified 270 genes that were downregulated in patients with early DN versus control subjects but were significantly increased in advanced DN versus early DN (Fig. 5E and Supplementary Excel File 1). These genes were mostly related to immune response (Fig. 5F and Supplementary Excel File 1), suggesting that these genes may be involved in the promotion of DN through enhanced inflammation.
Gene expression change from control to DN states. Mean expression (black line) of genes and the SD of the mean (gray band) are shown for the control, early DN, and advanced DN groups. A and B: Mean expression of five upregulated (A) and seven downregulated (B) genes. C: Mean expression of 148 genes that were upregulated in early DN but downregulated in advanced DN. D: The top 10 GO enrichment terms for 148 DEGs in panel C. E: Mean expression of 270 genes that were downregulated in early DN but upregulated in advanced DN. F: The top 10 GO enrichment terms for 270 DEGs in panel E. −log10(P), −log10P value.
Association of Genes With eGFR in Patients With DN
Additionally, we examined the association of genes in DN with eGFR in all patients with diabetes. GO terms related to iron transport and cell differentiation were positively associated with eGFR, while the immune response was at the top of the list of GO terms that were negatively associated with eGFR (Fig. 6A). Pathway enrichment analysis of combined DEGs showed phagocytosis regulation as one of the top pathways associated with eGFR (Supplementary Fig. 4). Small G-protein regulation (RhoA and Rac1), fibrosis, and inflammatory pathways were also highly associated with eGFR. The individual list of genes associated with eGFR is included in Supplementary Excel File 2.
Top enriched GO functions for eGFR and histological parameters in all DN samples. A: The bar chart shows the top enriched functions for DEGs that correlated positively (pink) or negatively (blue) with eGFR. B: A summary heatmap of the top enriched functions shows pathways that are correlated with each histological parameter. Scale bar indicates significance (shown as −log10P value [−log10(P)]). IF, interstitial fibrosis; TA, tubular atrophy; Top_Dn, top downregulated DEGs; Top_Up, top upregulated DEGs; tub, tubular.
Association of Genes With Kidney Histological Scores in All Patients With DN
We also examined the correlation between genes and histological scores in all patients with DN. Supplementary Table 2 shows the scoring of the individual histological parameters in patients with early or advanced DN. The most enriched GO pathways that correlated with histology scores in all DN samples are summarized in Fig. 6B. Segmental glomerulosclerosis (GS), tubular atrophy, and fibrosis had several shared GO pathways, such as Ras protein signaling, cell death, and several metabolic pathways. Global GS and arteriosclerosis also shared several pathways, such as chromatin remodeling and calcium-dependent cell-cell adhesion. Regulation of microtubule cytoskeleton was a common pathway between acute tubular injury and arteriosclerosis. However, each histological parameter also had distinct GO terms. For example, segmental GS had a unique feature on Rho protein signaling, which may be related to podocyte dysfunction. Mitochondrial dysfunction and fatty acid disturbance were associated more with global GS. Multiple inflammation-related pathways were associated with tubular atrophy/interstitial fibrosis. However, the acute tubular injury was associated with actin filament, iron transport, and secretion. Arteriosclerosis had specific GO terms on blood vessel remodeling, hormonal regulation of systemic arterial blood pressure, activation of adenylate cycle activity, and regulation of osteoclast differentiation. The latter might be related to the vascular calcification. Overall, these pathway analyses highlight the potential underlying molecular mechanisms for the different pathological changes observed in DN. The full list of DEGs associated with each histological parameter is included in Supplementary Excel File 3.
Deconvolution of RNA-Seq Data for the Representation of Kidney Cell Types
Because the biopsied tissue samples used in this study contain a heterogeneous mixture of kidney cell types, we used an R-based algorithm, CIBERSORT (11), to deconvolve the current data set to estimate the relative fractions of diverse kidney cell types. For this, we used the recently published scRNA-seq data set from human kidneys as a reference from Wu et al. (12). Enumeration of the data showed distinct clusters of kidney cells in control, early DN, and advanced DN kidneys (Fig. 7A). We found a significant reduction of proximal tubular and collecting cells in advanced DN compared with early DN and control samples. Interestingly, endothelial cells appeared to be decreased in both early and advanced DN compared with control, suggesting that kidney (glomerular and peritubular) endothelial cell injury may occur in early DN (Fig. 7A). There was an increase of monocytes, fibroblasts, myofibroblasts, B cells, and plasma cells in advanced DN compared with early DN and control samples. Since the scRNA-seq data used as a reference by Wu et al. did not include macrophages as a cell cluster, we could not determine whether there was a change in the macrophage population in DN. Therefore, we used the scRNA-seq data from normal mouse kidneys from Park et al. (13) as a reference, which showed a significant increase of macrophages and a small but significant increase in both T and B cells in advanced DN compared with early DN and control (Fig. 7B). These analyses are consistent with previous histological observations of increased inflammation, fibrosis, and tubular cell loss in the kidneys of patients with advanced DN compared with those from patients with early DN and control patients.
Estimates of cell components by deconvolution analysis in control, early, and advanced DN samples. A: Box plots of deconvoluted cell population across all cell types estimated by Wu et al. (12). B: Box plots of deconvoluted cell population across immune cells estimated by Park et al. (13). CD, collecting duct; EC, endothelial cell; LOH_AL, loop of Henle, ascending limb; LOH_DL, loop of Henle, distal limb; Mono1, monocyte type 1; Mono2, monocyte type 2; NK, natural killer; PT, proximal tubule.
The above deconvolution analysis showed a significant reduction of tubular cells in advanced DN. However, we could not appreciate the changes of podocytes because of the small portion of podocytes in the kidney cortices and the limitations of the analysis. Therefore, we compared the expression of both glomerular and tubular cell–specific markers and found that there was a reduction of both podocyte- and tubular cell–specific markers in advanced DN. The full list of kidney cell–specific DEGs is presented in Supplementary Excel File 4.
Since the changes of kidney cell population may affect the data analysis, we rechecked the DEGs and the GO pathways from early DN versus control and advanced DN to early DN after adjusting for the differences in cell population, as estimated from the deconvolution analysis. The adjusted analysis (Supplementary Figs. 5 and 6) largely showed similar findings to the unadjusted analysis (Figs. 2 and 4). In addition, we compared the DEGs from this study with the previously published transcriptomic data sets from Pan et al. (7) (GSE96804), a microarray analysis of glomerular transcriptome in DN, and Woroniecka et al. (5) (GSE30122), a microarray analysis of glomerular and tubular transcriptomes. As shown in Supplementary Table 3, there was an ∼30% overlap in DEGs and >60% overlap in most of the enriched pathways compared with both sets. Given that these published data sets are of glomerular or tubular compartment–specific microarray analyses and that our data set is of RNA-seq data of biopsied whole-cortex samples, the overlap is quite significant between our and these two studies.
Validation of the Findings From Our RNA-Seq Data
We next confirmed by immunohistochemical analysis the change in expression of several of the DEGs identified. We were particularly interested in the genes that were increased in early DN but decreased in advanced DN because they may represent genes that may have a protective role against the progression of DN. Among these, we selected RBP4 and GLP1R for further validation by immunostaining in the kidney sections of patients with DN because their expression has not been well characterized in diabetic kidneys. As shown in Fig. 8A, we confirmed that RBP4 and GLP1R increased in early DN but decreased in advanced DN. RBP4 staining localized mostly in renal tubular cells in early DN, which largely colocalized with proximal tubule marker AQP1 (Supplementary Fig. 7), while the staining was very weak in the kidneys with advanced DN (Fig. 8A and Supplementary Fig. 8A). GLP1R localized in both glomerular and proximal tubular cells in early DN kidneys and colocalized with AQP1 in the tubular compartment (Supplementary Fig. 7), but the staining was also very weak in advanced DN (Fig. 8A and Supplementary Fig. 8A). For genes that are increased in advanced DN but not in early DN, we selected two immune response genes for validation by immunostaining. We found that expression of both interleukin 6 (IL-6) and IL-1B was significantly increased in advanced DN kidneys but not in early DN kidneys (Fig. 8B and Supplementary Fig. 8B).
Immunostaining of genes that are altered in DN. A: Representative images of RBP4 and GLP1R immunostaining in control and DN kidneys. B: Representative images of IL-6 and IL-1B immunostaining in control and DN kidneys. C: Representative images of CD20, CD3, and MRC1 immunostaining in control and DN kidneys. Semiquantification of immunostaining is shown in Supplementary Fig. 8.
We also performed immunostaining for the markers of immune cells to validate the findings from the deconvolution data. We found that the staining for M2-macrophage marker MRC1 increased significantly in advanced DN compared with early DN (Fig. 8C). There was also a significant increase of staining for T- and B-cell markers CD3 and CD20, respectively, in advanced DN compared with early DN kidneys (Fig. 8C and Supplementary Fig. 8C). These data are consistent with our deconvolution analysis and support an important role of these immune cells in the progression of DN.
In the current study, we performed transcriptomic studies of whole-kidney biopsy samples from patients with either early or advanced DN compared with control nephrectomy samples from patients without diabetes obtained from Shanghai Jiao Tong University Affiliated Sixth People’s Hospital. For this study, we chose to use the whole-kidney biopsy samples for analysis. The rationale for this was mainly because of the limited amount of tissue that is typically available in biopsy samples and, importantly, because the dissection and digestion processes could alter the transcriptomic profiles by induction of stress-related gene expression as described in a previous study (15). The digestion process in particular may also have differential effects between nephrectomy and native kidney biopsy samples as well as between normal and diseased kidneys, thereby creating further artificial differences in gene expression between experimental groups. However, the limitation to our approach is that since kidney cortices contain mostly a tubulointerstitial compartment, the transcriptomic data will reflect mostly the mRNA expression of cells in this compartment. Indeed, the largest population of cell types identified by deconvolution analysis was proximal tubules. While the isolation of glomeruli would yield more specific information on glomerular cells, the data will nevertheless represent those of heterogeneous glomerular cell types. To circumvent this issue, we had performed isolation of specific glomerular cells from fluorescently labeled cells in mice (21,22), which cannot be done in human tissue samples. More recently, scRNA-seq has emerged as a powerful tool for observing gene expression at a single-cell level (23,24). However, the depth and number of the genes detected by scRNA-seq are still limited compared with the bulk RNA-seq (25). As mentioned above, the scRNA-seq data can be significantly affected by the digestion process because the procedure requires more stringent digestion of kidney tissues into the single cells. In addition, digestion of the human kidney biopsy samples for scRNA-seq is challenging to perform because of the limited amount of available material (26,27). Therefore, each technical approach has distinct advantages and disadvantages but together can provide complementary information to better understand the pathogenesis of DN.
There are a few advantages of the transcriptomic analysis described in this study. First, we used the RNA-seq approach, rather than microarray analyses used in most previous studies, to study the transcriptome of diabetic kidneys. The limitation of microarray analysis is that not all genes can be detected with currently available chips. In addition, our study provides data on the changes of noncoding RNAs in the diabetic kidney. For example, we found several noncoding RNAs among the DEGs between early and advanced DN that included LOC101926964, LOC101927136, LOC100507537, LINC00417, and LINC01255. We found that miR-3189 was upregulated in early DN compared with control, and it is known that miR-3189 is a potent regulator of cell apoptosis through the p53-dependent pathway (28). Therefore, we will confirm whether mature miR-3189 is indeed upregulated in patients with early DN compared with control patients in future investigations.
Another strength of our study is that we were able to include several samples from patients with early DN in addition to samples from patients with advanced DN. Although we only had samples from six patients with clinical and biopsy-proven early DN, the transcriptomic data from these patients provide useful insight into the early disease process of DN. These six patients have the same duration of diabetes and other clinical parameters as those with advanced DN and, therefore, are likely to be nonprogressors. To further support this, we have followed these six patients for 2–3 years since their enrollment and found that they had stable renal function and UACR. Previous studies included only patients with advanced DN (4,5), and the only other kidney transcriptome of early DN was performed in the Pima Indian study (6). Because our study is of patients of Asian origin, it would be informative to compare the transcriptomic data of patients with DN from different ethnic backgrounds.
Also, by using the recently published scRNA-seq data sets from normal human and mouse kidneys, we were able to deconvolve the bulk RNA-seq data from DN samples to estimate the changes occurring in specific kidney cell types in DN. Because of the variations of methodology and analysis among the currently available scRNA-seq data sets, we used two scRNA-seq data sets as a reference: human (26) and mouse (13) whole-kidney cortices. Even though the deconvolution data generated from the two data sets had some differences, the overall results suggest that there is an increase of monocytes/macrophages, fibroblasts, and myofibroblasts in DN kidneys. We further confirmed this by immunostaining of the kidney sections from the same patients. Thus, our data further support the critical role of macrophages in the pathogenesis of human DN. We are aware of the limitations of this analysis, but we believe that this approach will be significantly improved when we have more reliable scRNA-seq data from patients with DN.
When we examined the DEGs between early and advanced DN and control, we identified a group of genes that was highly expressed in early DN but suppressed in advanced DN, such as genes in the retinoic acid pathway and GLP1R. These likely represent a group of genes that have renoprotective effects in the early stage of DN, consistent with the nonprogressive nature of these patients. In addition, we were able to validate the expression of RBP4 and GLP1R in the kidney tissues of these patients. The role of RBP4 is to deliver retinol to the target tissue (29) and has been shown to be associated with renal function in patients with diabetes (30). Retinoic acid has been shown to have protective effects in vitro and in animal models of kidney disease (19,20). Recent studies suggested that local synthesis of retinoic acid appears impaired in the diseased kidney and contributes to the progression of kidney disease (31). Interestingly, genetic variations of RBP4 are associated with early DN but not advanced DN (32). Our study suggests that local expression of RBP4 might be elevated in early DN and contribute to local retinoic acid synthesis and protects patients from the progression of DN. Future studies are required to further examine the role of RBP4 in the progression of DN. GLP1R agonists have been shown to have renal protective effects in animal models and humans with DN (33–35). Our study suggests that GLP1R expression was increased locally in the kidney to protect the kidney from injury in early DN. Immunostaining of GLP1R indicated that its expression is indeed increased in both glomerular and tubular cells in early DN. How GLP1R mediates the effects of its agonists in local kidney cells requires further studies.
Interestingly, we found that expression of the immune response or inflammatory genes were suppressed in early DN but highly upregulated in advanced DN. The suppression of immune response genes in early DN could be partially due to mildly elevated inflammatory status of control nephrectomy samples. But more likely, these patients with early DN are nonprogressors and resistant to the progression of DN, and therefore, the inflammation is suppressed. The marked increase in immune response and inflammatory genes in patients with advanced DN confirms a critical role of inflammation in the progression of DN and is consistent with the previous transcriptomic data showing that the Janus kinase-STAT and nuclear factor-κB pathways are highly enriched in diabetic kidneys (2,4) and with the findings in Woroniecka et al. (5). Consistently, our deconvolution data suggest an increase in immune cells in the diabetic kidney, such as macrophages. Together, these data support a critical role of inflammation in the progression of DN.
We also performed correlation studies of gene transcripts with renal function (eGFR). Interestingly, DEGs related to phagocytosis and inflammatory pathways were negatively correlated with eGFR, suggesting again the role of macrophages and immune response in the progression of DN. DEGs related to iron transports were positively correlated with eGFR, indicating that tubular cell injury (loss of iron transports) contributes to the progression of DN.
DN has classic pathological features, with injury in the glomeruli, tubule/interstitium, and blood vessels (14,36). Different patients may present more injury in one of these compartments than others. However, the underlying molecular mechanisms of each pathological feature in DN remain unclear. Therefore, we studied the association of renal transcripts with the histological scores of individual pathological features obtained from patients with both early and advanced DN. Interestingly, we found that the segmental GS, global GS, and arteriosclerosis share some but have their own unique GO terms. Also, tubular fibrosis score was highly associated with the DEGs related to the immune response and inflammatory pathways. Since tubular fibrosis is known to be tightly associated with eGFR, our data show that the DEGs related to inflammation pathways correlate with eGFR. Interestingly, we also found several specific GO terms associated with arteriosclerosis, such as blood vessel remodeling and hormonal regulation of blood pressure. Several genes involved in bone metabolism were also associated with arteriosclerosis, and they may be involved in vascular calcification in patients with chronic kidney disease. A similar study was performed recently in a Pima Indian cohort with DN (6). The authors reported that the cortical interstitial fractional volume, an index of tubulointerstitial damage, correlated significantly with the transcripts enriched for pathways associated with mitochondrial dysfunction, inflammation, migratory mechanisms, and tubular metabolic functions. Further studies are required to determine how these DEGs contribute to the specific pathological changes observed in DN.
In conclusion, our study provides the transcriptomic data of patients with early and advanced DN compared with normal tissues from control patients from nephrectomy samples in a Chinese population of patients with diabetes. The correlation of renal transcripts with renal function and pathological changes will help us to further understand the underlying molecular mechanisms contributing to the progression of DN. We believe that whole-kidney transcriptomic data and scRNA-seq data will be complementary, and future sophisticated computational programs could help to better dissect the mechanisms of individual kidney cell injury by combining these two data sets. Finally, we believe that the data generated here could be an important resource for the renal community to further dissect the pathogenesis of DN.
Funding. Y.F. is supported by the National Natural Science Foundation of China (81870468) and the Medical and Engineering Cross Fund of Shanghai Jiao Tong University (YG2017MS10). K.L. is supported by National Institute of Diabetes and Digestive and Kidney Diseases grant R01-DK-117913. J.C.H. is supported by a Veterans Administration Merit Award and National Institute of Diabetes and Digestive and Kidney Diseases grants 1R01-DK-078897, 1R01-DK-088541, and P01-DK-56492. N.W. is supported by the National Natural Science Foundation of China (81670657 and 81870504).
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
Author Contributions. Y.F., Z.Y., F.Z., J.W., T.Z., Z.L., L.H., and Q.Z. performed the experiments and histological scoring. Y.F., Z.Y., Z.S., W.Z., K.L., and J.C.H. analyzed the data. Y.F., K.L., J.C.H., and N.W. designed the research project. Y.F., K.L., J.C.H., and N.W. drafted and revised the manuscript. V.D.D’A. performed the histopathological scoring. Y.F., J.C.H., and N.W. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
It’s officially Brussels sprouts season! I can’t think about sprouts without thinking of my BFF Wendy. She was the only kid I knew who loved Brussels Sprouts. The rest of us wouldn’t go near them. Boy, how taste buds change! (Not Wendy’s, though.) Celebrate the season by making Roasted Brussels Sprouts with Grapes.
My husband has been on a mission to eat more cruciferous vegetables ever since he starting reading How Not to Die by Michael Greger, MD.
Cruciferous vegetables include cabbage, cauliflower, and Brussels sprouts. They contain tons of nutrients and fiber, help you feel full longer, and reduce that dreaded inflammation. Ease into eating these nutritional powerhouses with recipes like the one below. Brussels sprouts, grapes, and shallots are roasted in the oven, then combined with a touch of balsamic vinegar for a hint of sweetness.
How to Trim Brussels Sprouts
Are you intimidated by prepping Brussels sprouts? It does take a little bit of time, but it’s really very easy. All you need are a sharp knife, a cutting board, and a colander. Most markets sell sprouts individually, but some, especially farmers’ markets, sell fresh ones still on the stalk. If yours are on the stalk, simply cut them off and then proceed:
Trim a slice off the stem end. (This may not be necessary if you just cut yours off the stalk.)
Remove any unattractive outer leaves.
Cut each in half lengthwise (through the stem end).
Rinse and drain.
Hosting a holiday meal? Roasted Brussels Sprouts with Grapes would be a great vegan and gluten-free dish for your table.
Other Recipes You Might Like
Like Brussels sprouts? Check out these other recipes:
Roasted Brussels Sprouts with Grapes
Brussels sprouts with a hint of sweetness from grapes and balsamic vinegar
Author:Adapted from Prevention
Course: Side Dishes
Keyword: brussels sprouts, roasted brussels sprouts, thanksgiving side
8 teaspoons extra-virgin olive oil divided
1 pound Brussels sprouts trimmed and cut in half
1/4 teaspoon kosher salt
1/8 teaspoon freshly ground black pepper
2 cups seedless red grapes
2 shallots sliced
1/2 tablespoon balsamic vinegar
2 tablespoons sliced almonds toasted (see Notes)
Preheat oven to 425°F. Brush two baking sheets with 1 teaspoon of olive oil each.
In a large bowl, toss together the Brussels sprouts, 1 tablespoon olive oil, salt, and pepper. Spread the Brussels sprouts in a single layer on one of the prepared baking sheets, cut side down.
In the same bowl, toss together the grapes, shallots, and remaining 1 tablespoon olive oil. Transfer to the second baking sheet.
Place both sheets in the oven and roast. After 15 minutes, remove the pan with the grapes from the oven and stir. Return it to the oven and remove the pan with the Brussels sprouts and stir. Return it to the oven and continue roasting until the sprouts are golden and can be easily pierced with a fork, about 10 more minutes (25 minutes total). Remove both pans from the oven.
In a small bowl, combine the balsamic vinegar with 1 tablespoon water. Pour it over the grapes (still in the pan), then stir and scrape up any browned bits.
In a large serving bowl, combine the Brussels sprouts and the grapes with their juices. Top with the almonds and serve.
How to toast almonds: Place almonds in a nonstick skillet over medium heat. Cook, tossing frequently, until the almonds are aromatic and golden in color, about 5 to 6 minutes. They will brown quickly once the pan gets hot.
Hey guys! I’ve missed you! It was a much longer blog break than I anticipated, and well, my new laptop has taken me a while to get used to. I would download pics to my laptop, start a post and then had no idea HOW to find the pictures that I knew were on my laptop somewhere.
Not surprisingly, my brother Charlie and his family who are in town this weekend, helped me figure it out. I’ve also got a new theme to my blog, although it will be a work in progress. Many things to fix, but hopefully in the next month it will be where I want it to be. My recipes are so hard to find, and that’s number one priority.
So while I am back to posting, don’t worry if the blog looks a bit jenky for a bit!
Last week I made the most delish cheese danishes. The star of them was Safe + Fair’s chocolate candy cane granola. Holy balls this quickly shot to the top of my favorite granola. I love that I don’t get the candy cane in every bite, so it’s a surprise when I get it!
One of my high school breakfasts when Suzy-Q’s were unavailable at 7-11 (along with a Dr. Pepper!) was a Hostess cherry pie. That cherry filling, the crackly frosting on the outside – pure heaven!
These are my version of a grown up Hostess cherry pie – made with the Safe + Fair Candy Cane granola kneaded into the danish dough, filled with brie and no sugar added cherry pie filling. Then finished off with a glaze. It’s delicious. You don’t get the candy cane in every bite, so it’s a nice surprise when you get that bite. The slightly saltiness of the brie mixed with sweet/tart cherry pie filling. Well, let’s just say I am glad I only made three otherwise I’d eat a dozen.
Going forward, I think I will ditch the 1/2 teaspoon of butter that I kneaded into the dough – not sure it made that much of a difference, which would make them 4 points, not 5.
No matter what @ww plan you chose, these are each 5 points. Leave out the 1/2 teaspoon of butter brings it down to 4 points.
It’s been a busy few weeks. I was supposed to go to LA this past weekend, but the owner of Nyrvana had a family emergency, so it’s been postponed to the end of January. I was happy though since my brother and his family were in town this weekend, and I got to spend more time with him, even though I just got back from a week of visiting him. We laugh so much!
I am nearly done with my Christmas shopping. I’ve learned to kind of wait until the weekend before because otherwise I tend to spend too much money.
I’ll be back tomorrow with another recipe – my sugar free chocolate cupcakes with a sugar free vanilla frosting that’s shaped into a Christmas tree – it’s so pretty!
Happy Monday friends – hope you have an amazing day.
What’s your strategy for managing your diabetes during the holidays?
If your plan is to eat yourself into a food coma, you are not alone. Most gatherings this time of year revolve around food and drinks, and usually not the healthy stuff. And I absolutely think there should be room for that!
There should be room for indulgence, throwing your diet to the wind for a few hours and just enjoying being with family and friends.
And yes, of course we can do that even though we live with diabetes. If you are insulin-dependent like me, you just need a solid game plan and you’ll be able to get through the festivities without wreaking havoc on your diabetes management.
My plan for successfully managing my diabetes during the holidays
My plan is fairly straightforward, so it should work for you too:
Know your carb ratios (how much insulin to take to a gram of carbs) – if you don’t know them yet, get a piece of paper and take notes for a few days to get it right. Knowing your daily carb ratios will help you gain good blood sugar control year-round, not just for the holidays. You can read my post about carb ratios and sensitivities to learn more.
Know what you are eating – if you didn’t make it yourself, ask the host. You can’t always see what’s in a dish. Mashed potatoes are usually not just potatoes but also include a lot of butter, milk (or cream) and maybe even sugar.
Bring a carb counting cheat sheet – It might not be exact, but it’s better than nothing. If you go to your mom’s house during the holidays, you probably know a lot of the food she will be cooking because you have had it since childhood.
Do a little research upfront and find out how many carbs are in a serving of your favorite Christmas foods. Maybe you can even get mom’s recipe and calculate the exact carbs? You can find different cheat sheets online or simply make your own.
Work out in the morning – hit the weights before you go out. It will improve your insulin sensitivity and make it easier to manage your blood sugars during the afternoon/night. Not only will you feel better, but your muscles will love the additional carbs and proteins you’ll enjoy later in the day. Why not use all those holiday calories to build some muscles?
Consider increasing your basal rate – if you are sitting down eating for hours, your basal insulin may need to be turned up. Always consult your healthcare professional before making changes to your insulin dosage, but it might be an idea to increase your basal rate during dinner and throughout the night.
Go for a walk after dinner (or a snowball fight) – it will not only help digestion but also your blood sugar. I’m not even sure this one needs more explanation. Just be sure your sugars don’t crash!
Test, test, test and keep track of active insulin so you don’t overdose – I bring my meter to the table and rely heavily on my CGM. My sugars will have some spikes, and that’s ok, as long as they come down as planned.
Remember to bolus for alcohol – Bolusing for alcohol can be tricky because alcohol can make your sugars drop. My rule of thumb is that I bolus for everything except hard liquor (which I never drink anyway). Especially if your alcohol is mixed with juice or other carb-heavy drinks, you’ll need insulin with that.
And please drink responsibly. It’s really hard to manage anything, and especially your diabetes, if you are out of your mind drunk. If you for some reason aren’t eating but only drinking, be very careful with dosing for alcohol since you’ll have a high likelihood of going low during the night.
Alcohol is tricky, so if you’re drinking and unsure how your blood sugars will react, I suggest having a bedtime snack just to be safe.
I’ve successfully managed my diabetes during the holidays if I’ve enjoyed myself without having to worry about my diabetes too much, if I’ve been able to manage my sugars so that I don’t wake up the next morning in the 200 mg/dL (11 mmol/L) range, and if I haven’t had too many low sugars.
I doubt Santa is going to bring me a new pancreas for Christmas, so it’s a good thing I have learned to manage my diabetes on my own.
Suggested next posts
If you found this post useful, please sign up for our newsletter (and get a free chapter from the Fit With Diabetes eBook) using the form below. We send out a weekly newsletter with the latest posts and recipes from Diabetes Strong.
Success! Now check your email to download the eBook chapter.
Obesity and type 2 diabetes are emerging as global epidemics and impose huge burdens on patient families as well as society. They are leading causal factors for cardiovascular disease, neurological disease, cancer, and kidney disease and negatively impact health span and life span. However, treatment options are limited, and there is an unmet need for developing effective and safe medications. In this issue of Diabetes, Jiang et al. (1) report a new treatment strategy and a new antidiabetes agent.
Obesity arises from energy imbalance, and excessive energy is stored as triacylglycerol in fat (2). Mitochondria are an essential organelle responsible for cellular respiration and energy production. Mitochondria oxidize free fatty acids (FFAs) and glucose to produce ATP (referred to as oxidative phosphorylation). The inner mitochondrial membrane (IMM) contains an electron transport chain composed of the complexes I, II, III, IV, and V. Burning of FFAs and glucose produces high-energy electrons that are transported from complex I or II to complexes III and IV, where electron-carried energy is released to pump out protons, generating proton gradients across the IMM (Fig. 1). The IMM proton gradients drive ATP synthesis by complex V. The IMM also contains uncoupling proteins (UCPs) that mediate proton influx, thus suppressing ATP synthesis. UCPs release free energy as heat (thermogenesis) and are engaged in body temperature homeostasis. Brown adipose tissue UCP1 plays a pivotal role in the maintenance of body temperature homeostasis, particularly in rodents (3). UCP-mediated thermogenesis increases energy expenditure, thus protecting against obesity (3). In the 1930s, chemical uncoupler 2,4-dinitrophenol (DNP) was developed for obesity treatment; unfortunately, it had severe adverse effects (e.g., hyperthermia, hyperlactacidemia, death) and was stopped for clinical use (4,5). Nonetheless, the diabetes community continues to search for safe, therapeutic mitochondrial uncouplers for the treatments of obesity and type 2 diabetes. Here, Jiang et al. identified a promising one called 6j.
Protection against diabetes by dual mechanisms of mitochondrial uncoupling and pyruvate oxidation. 6j possesses dual properties of stimulating uncoupling and PDH-mediated pyruvate oxidation. Uncoupling increases mitochondrial respiration and burning of glucose and fatty acids. PDH activation reroutes glycolysis-derived pyruvate from lactate production to tricarboxylic acid cycle (TCA)-mediated catabolism, preventing hyperlactacidemia toxicity. I, II, and IV: complexes I, II, and IV.
DNP-induced uncoupling lowers ATP synthesis and ATP content, thereby increasing glycolysis to compensate for ATP deficiency. Increase in anaerobic glycolysis leads to pyruvate and lactate overproduction and hyperlactacidemia following DNP treatment (5). Of note, mitochondrial pyruvate dehydrogenase (PDH), which is tightly regulated through phosphorylation, catalyzes the first reaction of pyruvate catabolism (6). Pyruvate dehydrogenase kinases (PDKs) phosphorylate PDH at phospho-Ser293 and potently inhibit PDH (6). Several PDK inhibitors, including dichloroacetic acid (DCA), have been developed to study the function of the PDK/PDH pathway. To combat hyperlactacidemia toxicity, Jiang et al. reasoned that DCA-stimulated PDH activation might reroute pyruvate from lactate production to PDH-mediated oxidation, thereby ameliorating DNP-associated hyperlactacidemia.
Jiang et al. first established cell culture systems and validated DNP and DCA actions in vitro. As expected, DNP potently stimulates mitochondrial uncoupling, as revealed by marked increases in both oxygen consumption rate and ADP/ATP ratio. DCA inhibits PDK-induced phosphorylation of PDH at phospho-Ser293, leading to PDH activation and suppression of lactate synthesis (i.e., decrease in extracellular acidification rate) due to PDH-mediated pyruvate oxidation. Jiang et al. then tested DNP and DCA in mice. In line with previous reports, acute DNP administration induces hyperthermia, hyperlactatemia, and death at high doses. A chronic DCA treatment decreases both PDK-mediated phosphorylation of PDH (i.e., increases PDH activity) and lactate production. Notably, DCA also induces ectopic lipid accumulations in the liver and skeletal muscle. As an adaptive response to DCA-stimulated pyruvate catabolism and glucose oxidation, FFA oxidation is suppressed, contributing to muscle and liver steatosis. Additionally, pyruvate-derived acetyl-CoA may serve as a lipogenic precursor to increase de novo lipogenesis. Intracellular lipid species are known to promote insulin resistance and type 2 diabetes. For instance, diacylglycerol activates protein kinase C-θ (PKC-θ) in skeletal muscle and PKC-ε in the liver, which in turn inhibit insulin signaling and induce insulin resistance (7,8). Accordingly, the authors observed that DCA treatment activates muscle PKC-θ and liver PKC-ε and impairs insulin signal transduction in these two tissues. Next, Jiang et al. tested an innovative idea that DNP and DCA dual treatments may preserve the antidiabetes effect while eliminating the adverse consequences. The authors elegantly demonstrated that in cell cultures, DCA-stimulated pyruvate oxidation abrogates DNP-induced lactate overproduction while DNP maintains its ability to stimulate mitochondrial uncoupling. In mice, DCA pretreatment (i.e., increased pyruvate oxidation) reverses DNP-induced hyperthermia, hyperlactacidemia, and death. In mice with dietary obesity, remarkably, DNP/DCA dual treatments considerately improve hyperglycemia, insulin resistance, and glucose intolerance. Furthermore, DNP decreases DCA-induced ectopic lipid accumulations and insulin resistance in the muscle and liver. These exciting results provide proof-of-concept evidence that concomitant stimulation of both mitochondrial uncoupling and pyruvate catabolism is a viable strategy for type 2 diabetes treatment. These observations also raise an intriguing possibility that mitochondrial uncoupling and pyruvate oxidation act coordinately to regulate mitochondrial selection of fuel substrates and metabolic flexibility. Next, Jiang et al. wondered, cleverly, whether they can design an agent that possesses dual properties of stimulating both mitochondrial uncoupling and PDH activation—two birds with one stone. They successfully engineered compound 6j (Fig. 1), using high-throughput screenings and structure/activity-based chemical modifications. Remarkably, chronic 6j administrations, like DNP/DCA dual treatments, considerably improve insulin resistance, glucose intolerance, and liver steatosis in mice with either dietary or genetic (db/db) obesity. Thus, 6j and related agents hold great promise as new medications for the treatment of type 2 diabetes.
This work, like many other great studies, raises several interesting questions. ATP deficiency, due to mitochondrial uncoupling, potentially has adverse consequences on energy-demanding, ATP-sensitive cells (e.g., cardiomyocytes, neuronal subpopulations). Hence, cell type–specific 6j-related agents are expected to have additional safety properties. In line with this notion, recent studies highlight the translational significance of cell type–specific mitochondrial uncouplers (9,10). Aberrant mitochondrial uncoupling and depolarization may cause mitochondrial injury, mitochondria-originated cell death and inflammation, and/or mitophagy (11). Therefore, the potential effects of long-term 6j treatments on mitochondrial integrity, function, and mitochondrial diseases need to be further assessed. Given the paramount role of mitochondrial energy expenditure in body weight control, the impact of 6j and related agents on protection against obesity needs additional investigation. Interestingly, DCA suppresses DNP-induced hyperthermia, but the underlying mechanism is not fully understood. Finally, the direct molecular targets of 6j linking to uncoupling and PDH activation remain elusive, impeding structure/activity-based optimizations to further improve specificity and efficacy.
Funding. This work was supported by National Institute of Diabetes and Digestive and Kidney Diseases grants RO1 DK115646 and RO1 DK114220.
Duality of Interest. No other potential conflicts of interest relevant to this article were reported.
Sprinkle the chicken with the cumin and pepper. Place the water and a steamer basket in the Instant Pot. Arrange the chicken in the steamer basket. Seal the lid, close the valve, and set the Manual/Pressure Cook button to 6 minutes.
Use a natural pressure release for 5 minutes, followed by a quick pressure release. When the valve drops, carefully remove the lid. Remove the chicken and place it on a cutting board. Let stand for 5 minutes before shredding. Set aside.
Meanwhile, whisk together the yogurt, mayonnaise, sugar, curry, and salt in a medium bowl and set aside.
Place the asparagus and edamame in the steamer basket in the pot. Seal the lid, close the valve, press the Cancel button, and reset to Manual/Pressure Cook for 1 minute.
Use a quick pressure release. Transfer the asparagus mixture to a colander and run it under cold water to stop the cooking process and cool quickly; drain well.
Place equal amounts of the kale mix on each of 4 dinner plates. Top with equal amounts of the asparagus mixture.
Add the chicken and onions to the yogurt mixture and toss until well coated. Spoon equal amounts on top of each serving of the asparagus mixture, and sprinkle with cilantro.
A serving is approximately 1/2 cup of the chicken mixture, 1/2 cup of the asparagus mixture, and 1 cup of greens.
OBJECTIVE Efficacy and safety of the glucagon-like peptide 1 (GLP-1) analog oral semaglutide and the sodium–glucose cotransporter 2 inhibitor empagliflozin were compared in patients with type 2 diabetes uncontrolled on metformin.
RESEARCH DESIGN AND METHODS Patients were randomized to once-daily open-label treatment with oral semaglutide 14 mg (n = 412) or empagliflozin 25 mg (n = 410) in a 52-week trial. Key end points were change from baseline to week 26 in HbA1c (primary) and body weight (confirmatory secondary). Two estimands addressed efficacy-related questions: treatment policy (regardless of trial product discontinuation or rescue medication) and trial product (on trial product without rescue medication) in all randomized patients.
RESULTS Four hundred (97.1%) patients in the oral semaglutide group and 387 (94.4%) in the empagliflozin group completed the trial. Oral semaglutide provided superior reductions in HbA1c versus empagliflozin at week 26 (treatment policy –1.3% vs. –0.9% [–14 vs. –9 mmol/mol], estimated treatment difference [ETD] –0.4% [95% CI –0.6, –0.3] [–5 mmol/mol (–6, –3)]; P < 0.0001). The treatment difference in HbA1c significantly favored oral semaglutide at week 26 for the trial product estimand (–1.4% vs. –0.9% [–15 vs. –9 mmol/mol], ETD –0.5% [95% CI –0.7, –0.4] [–6 mmol/mol (–7, –5)]; P < 0.0001) and at week 52 for both estimands (P < 0.0001). Superior weight loss was not confirmed at week 26 (treatment policy), but oral semaglutide was significantly better than empagliflozin at week 52 (trial product −4.7 vs. −3.8 kg; P = 0.0114). Gastrointestinal adverse events were more common with oral semaglutide.
CONCLUSIONS Oral semaglutide was superior to empagliflozin in reducing HbA1c but not body weight at 26 weeks in patients with type 2 diabetes uncontrolled on metformin. At week 52, HbA1c and body weight (trial product estimand) were significantly reduced versus empagliflozin. Oral semaglutide was well tolerated within the established safety profile of GLP-1 receptor agonists.
Many patients with type 2 diabetes fail to achieve or maintain adequate blood glucose control when treated with metformin monotherapy. Injectable glucagon-like peptide 1 receptor agonists (GLP-1RAs) and oral sodium–glucose cotransporter 2 (SGLT-2) inhibitors are recommended as second-line therapy because of their ability to lower glucose without increasing hypoglycemia risk, weight loss effect, and associated cardiovascular benefits (1,2).
Semaglutide is a human GLP-1 analog currently available as a once-weekly injection associated with reduced glycated hemoglobin (HbA1c), weight loss, and fewer cardiovascular events in type 2 diabetes (3–9). Oral semaglutide is coformulated in a tablet with the absorption enhancer sodium N-(8-[2-hydroxylbenzoyl] amino) caprylate, which facilitates semaglutide absorption across the gastric mucosa (10). Oral semaglutide has demonstrated significantly greater reductions in HbA1c and body weight compared with placebo in patients with type 2 diabetes uncontrolled with diet and exercise or oral antidiabetic medication, including in patients with moderate renal impairment (11–14). Significantly greater reductions in HbA1c and body weight have also been shown with oral semaglutide, given as 7 or 14 mg/day or flexibly dosed, compared with sitagliptin in patients uncontrolled with oral antidiabetic drugs (15,16). Oral semaglutide also resulted in a noninferior reduction in HbA1c and superior weight loss versus liraglutide in patients on metformin with or without an SGLT-2 inhibitor (13). Cardiovascular safety has been confirmed, with an indication of benefit, by a nonsignificant 21% risk reduction in major adverse cardiovascular events versus placebo (17).
Empagliflozin is a widely used oral SGLT-2 inhibitor shown to improve glycemic control and body weight (18–22) and associated with a reduced risk of cardiovascular and all-cause mortality in patients at high cardiovascular risk (23). The present phase 3a trial, PIONEER 2, is the first direct comparison of oral semaglutide with an SGLT-2 inhibitor, empagliflozin, in type 2 diabetes uncontrolled with metformin monotherapy.
Research Design and Methods
This randomized, open-label, multinational 52-week trial was conducted at 108 sites in 12 countries (Argentina, Brazil, Croatia, Greece, Hungary, Italy, Poland, Russia, Serbia, Spain, Thailand, U.S.). Patients were randomized (1:1) to once-daily oral semaglutide 14 mg or empagliflozin 25 mg for 52 weeks using an interactive web response system with a further 5 weeks of follow-up (Supplementary Fig. 1). An open-label trial design was used because manufacture of placebo tablets resembling empagliflozin was not feasible within a reasonable time frame. Oral semaglutide was initiated at 3 mg once daily, escalated to 7 mg at week 4 and 14 mg after week 8. Because food impairs absorption of oral semaglutide, patients were instructed to administer oral semaglutide in the morning in a fasted state with up to 120 mL of water at least 30 min before breakfast and any other oral medication. Empagliflozin was initiated at 10 mg once daily in the morning and escalated to 25 mg at week 8.
Additional antidiabetic medication was available for patients with persistent or unacceptable hyperglycemia on trial product and for patients who prematurely discontinued trial product and remained in the trial. Additional antidiabetic medication was defined as that initiated (or intensification of existing antidiabetic background medication by a dose increase of >20%) during the planned treatment period (i.e., from randomization to the planned end-of-treatment visit) either as add-on to trial product or initiated after premature discontinuation of trial product. The subset of additional antidiabetic medication (or intensification of existing antidiabetic background medication) used as add-on to trial product is defined as rescue medication. Short-term use (≤21 days) of antidiabetic medication (e.g., in connection with intercurrent illness) was not considered as additional antidiabetic medication (including rescue medication).
Rescue criteria were fasting plasma glucose >260 mg/dL (14.4 mmol/L) from week 8 to 13, >240 mg/dL (13.3 mmol/L) from week 14 to 25, and >200 mg/dL (11.1 mmol/L) (or HbA1c >8.5% [69.4 mmol/mol]) from week 26 onward. Rescue medication was prescribed at the investigator’s discretion (excluding GLP-1RAs, dipeptidyl peptidase 4 inhibitors, and amylin analogs in the oral semaglutide arm and SGLT-2 inhibitors in the empagliflozin arm). Patients who prematurely discontinued trial product remained in the trial and could receive any other antidiabetic medications at the investigator’s discretion (excluding GLP-1RAs in the oral semaglutide arm before completion of the follow-up visit 5 weeks after the last date on trial product).
Two different questions related to the efficacy objectives were addressed through the definition of two estimands: treatment policy and trial product. Both estimands were defined based on interactions with regulatory agencies. The treatment policy estimand evaluates the treatment effect for all randomized patients, regardless of trial product discontinuation or use of rescue medication. This estimand reflects the intention-to-treat principle as defined in International Council on Harmonization (ICH) E9 (24). The estimand reflects the effect of initiating treatment with oral semaglutide compared with initiating treatment with empagliflozin, both potentially followed by either discontinuation of trial product and/or addition of or switch to another glucose-lowering drug.
The trial product estimand evaluates the treatment effect for all randomized patients under the assumption that all patients remained on trial product for the entire planned duration of the trial and did not use rescue medication. This estimand aims at reflecting the effect of oral semaglutide compared with empagliflozin without the confounding effect of rescue medication. The statistical analysis that was applied to estimate this estimand is similar to how many phase 3a diabetes trials have been evaluated, and results from such analyses are currently included in many product labels (prescribing information, U.S., and summary of product characteristics, European Union) for glucose-lowering drugs (e.g., Ozempic summary of product characteristics).
Trial product discontinuation and initiation of rescue medication are accounted for by the treatment policy strategy for the treatment policy estimand and by the hypothetical strategy for the trial product estimand as defined in draft ICH E9 (R1) (25). Further details on the use of estimands in this trial are provided in Supplementary Data, Estimands, with additional background provided by Aroda et al. (26).
The trial protocol was approved by all relevant institutional review boards/independent ethics committees, and the trial was conducted in accordance with ICH Good Clinical Practice guidelines and the Declaration of Helsinki. All patients provided written informed consent before any trial-related activity.
Eligible patients were adults with type 2 diabetes and an HbA1c of 7.0–10.5% (53–91 mmol/mol) receiving a stable dose of metformin (≥1,500 mg or maximum tolerated). Key exclusion criteria (see Supplementary Table 1 for full list) were any medication for diabetes or obesity within the previous 90 days other than metformin or short-term (≤14 days) insulin, renal impairment with an estimated glomerular filtration rate <60 mL/min/1.73 m2, proliferative retinopathy or maculopathy requiring acute treatment verified by fundus photography or dilated fundoscopy, and history of pancreatitis.
Trial End Points
The primary end point was change in HbA1c from baseline to week 26. The confirmatory secondary end point was change in body weight (kg) from baseline to week 26.
Secondary end points included changes from baseline to week 52 in HbA1c and body weight (kg) and changes from baseline to weeks 26 and 52 in fasting plasma glucose, self-measured blood glucose (SMBG) profile (7-point profile and mean postprandial increment over all meals), fasting C-peptide, fasting insulin, fasting proinsulin, fasting glucagon, HOMA of insulin resistance (HOMA-IR), HOMA of β-cell function (HOMA-B), C-reactive protein, body weight (%), BMI, waist circumference, and fasting lipid profile. Other secondary end points were the proportion of patients achieving HbA1c <7% (53 mmol/mol) or ≤6.5% (48 mmol/mol); weight loss of ≥5% or ≥10%; composite end point of HbA1c <7% (53 mmol/mol) without severe or symptomatic hypoglycemia (blood glucose <56 mg/dL [<3.1 mmol/L]) and no weight gain; composite end point of an absolute reduction in HbA1c of ≥1.0% (10.9 mmol/mol) and body weight loss of ≥3% (weeks 26 and 52); and changes from baseline to weeks 26 and 52 in the patient-reported outcomes, Short Form (SF) 36v2 Health Survey (Acute Version) (27) and Control of Eating Questionnaire (28). Further end points are listed in the Estimands section and the protocol that are included as part of the Supplementary Data.
Safety end points included the number of treatment-emergent adverse events, incidence of American Diabetes Association (ADA)–classified (29) severe or confirmed symptomatic hypoglycemic episodes (blood glucose <56 mg/dL [<3.1 mmol/L]), and changes from baseline in heart rate, blood pressure, and other clinical and laboratory assessments. An independent external event adjudication committee (EAC) performed masked validation of predefined adverse events, including deaths, selected cardiovascular events, acute pancreatitis, malignant neoplasms, acute kidney injury, and lactic acidosis.
The primary end point of change from baseline to week 26 in HbA1c was tested for both noninferiority and superiority of oral semaglutide versus empagliflozin, with a sample size calculation to ensure a power of at least 90% for testing superiority. The confirmatory secondary end point of change from baseline to week 26 in body weight was tested for superiority of oral semaglutide versus empagliflozin. The confirmation of efficacy of oral semaglutide on change in HbA1c and body weight from baseline to week 26 was based on a weighted Bonferroni closed testing strategy (30) to control the overall type I error for the hypotheses evaluated by the treatment policy estimand (Supplementary Fig. 2). Because of the potential for type I errors as a result of multiple comparisons, findings for analyses of additional secondary end points should be interpreted as exploratory.
The treatment policy estimand was estimated by a pattern-mixture model using multiple imputation to handle missing week 26 data for both confirmatory end points. Data collected at week 26, irrespective of premature discontinuation of trial product or initiation of rescue medication, were included in the statistical analysis. Imputation was done within groups defined by trial product and treatment status at week 26. Both the imputation and the analysis were based on ANCOVA models. The results were combined by use of Rubin’s rule (31). Before testing for noninferiority, a value of 0.4% (the noninferiority margin) was added to imputed values at week 26 for the oral semaglutide treatment arm only (32). The trial product estimand was estimated by a mixed model for repeated measurements that used data collected before premature trial product discontinuation or initiation of rescue medication from all randomized patients.
Further details on the statistical analyses can be found in Supplementary Fig. 2. All analyses were performed using SAS 9.4M2 statistical software.
Data will be shared with bona fide researchers submitting a research proposal approved by the independent review board. Access request proposals can be found at http://novonordisk-trials.com. Data will be made available after research completion, approval of the product, and product use in the European Union and U.S. Individual participant data will be shared in data sets in a deidentified/anonymized format using a specialized SAS data platform.
A total of 1,122 patients were screened, with 822 randomized to oral semaglutide 14 mg once daily (n = 412) or empagliflozin 25 mg once daily (n = 410). Four hundred (97.1%) patients in the oral semaglutide group and 387 (94.4%) in the empagliflozin group completed the trial (Supplementary Fig. 3). Baseline characteristics were well balanced between treatment groups (Table 1). Patients, of whom half (49.5%) were female, had a mean age of 58 years, baseline HbA1c of 8.1% (65 mmol/mol), fasting plasma glucose of 173 mg/dL (9.6 mmol/L), average duration of diabetes of 7.4 years, and mean body weight of 91.6 kg.
Baseline characteristics and demographics
Use of additional antidiabetic medication and rescue medication is shown in Supplementary Table 2. Through to week 26, 17 (4.1%) patients initiated additional antidiabetic medication in the oral semaglutide group; in 8 (1.9%) of these patients, it was rescue medication. In the empagliflozin group, 13 (3.2%) patients initiated additional antidiabetic medication through to week 26, with this being rescue medication in 5 (1.2%). Through to week 52, 52 (12.7%) patients initiated additional antidiabetic medication in the oral semaglutide group; in 31 (7.5%) of these patients, it was rescue medication. In the empagliflozin group, 56 (13.7%) patients initiated additional antidiabetic medication, with this being rescue medication in 44 (10.7%). Sulfonylureas were the most commonly used additional antidiabetic and rescue medication. Disposition of patients throughout the trial is shown in Supplementary Fig 4.
Oral semaglutide 14 mg provided a superior reduction in HbA1c compared with empagliflozin 25 mg at week 26 when evaluated by the treatment policy estimand (regardless of rescue medication use or trial product discontinuation) (–1.3% vs. –0.9% [–14 vs. –9 mmol/mol]; estimated treatment difference [ETD] –0.4% [95% CI –0.6, –0.3] [–5 mmol/mol (–6, –3)]; P < 0.0001 for noninferiority and superiority) (Fig. 1). Results from sensitivity analyses supported the results of the confirmatory analysis (Supplementary Fig. 5). When evaluated by the trial product estimand (on trial product and without the use of rescue medication), the reduction in HbA1c was significantly greater with oral semaglutide at week 26 (–1.4% vs. –0.9% [–15 vs. –9 mmol/mol], ETD –0.5% [–0.7, –0.4] [–6 mmol/mol (–7, –5)]; P < 0.0001) (Fig. 1). Significantly greater reductions in HbA1c with oral semaglutide compared with empagliflozin were also observed at week 52 (both estimands) (Fig. 1). More patients achieved the predefined HbA1c targets with oral semaglutide than with empagliflozin, and the odds of doing so were significantly greater at weeks 26 and 52 (both estimands, all P < 0.0001) (Fig. 1 and Table 2).
Glycemic control and body weight–related efficacy end points. A: Observed absolute change in HbA1c over time. B: Estimated changes from baseline in HbA1c at weeks 26 and 52. C: Observed proportions of patients achieving HbA1c <7% (53 mmol/mol) at weeks 26 and 52. D: Observed absolute change in body weight over time. E: Estimated changes from baseline in body weight at weeks 26 and 52. F: Observed proportions of patients achieving body weight reduction ≥5% at weeks 26 and 52. Treatment policy estimand: Data irrespective of discontinuation of trial product and initiation of rescue medication were included. Trial product estimand: Data collected after discontinuation of trial product or initiation of rescue medication are excluded. P values are two-sided and unadjusted. *Superiority confirmed for oral semaglutide versus empagliflozin. Observed mean change (± SEM) from baseline (A and D), estimated mean changes from baseline at week 26 and 52 (B and E), and observed proportions of patients achieving target at weeks 26 and 52 (C and F). Patient numbers represent patients contributing to the means. EOR, estimated odds ratio.
Selected secondary end points (treatment policy estimand and trial product estimand)a
Fasting plasma glucose was reduced with both treatments, with no significant difference between groups (Table 2 and Supplementary Fig. 6). Oral semaglutide resulted in significantly greater reductions in mean 7-point SMBG profiles compared with empagliflozin at both weeks 26 and 52 (Table 2 and Supplementary Fig. 6) and significantly reduced mean postprandial increments, as averaged for all meals (excluding the treatment policy estimand evaluation at week 26) (Table 2).
Superiority of body weight reduction at week 26 with oral semaglutide over empagliflozin was not confirmed (treatment policy estimand −3.8 vs. −3.7 kg; ETD −0.1 kg [95% CI −0.7, 0.5]; P = 0.7593). Results from sensitivity analyses supported the results of the confirmatory analysis (Supplementary Fig. 5). There was no difference between treatments using the trial product estimand (−4.2 vs. −3.8 kg; ETD −0.4 kg [−1.0, 0.1]; P = 0.1358) (Fig. 1). A significantly greater reduction in body weight was achieved with oral semaglutide versus empagliflozin at week 52 when evaluated by the trial product estimand (−4.7 vs. −3.8 kg; ETD −0.9 kg [−1.6, −0.2]; P = 0.0114) but not the treatment policy estimand (−3.8 vs. −3.6 kg; ETD −0.2 kg [−0.9, 0.5]; P = 0.6231). Proportions of patients achieving ≥5% or ≥10% weight loss are shown in Fig. 1 and Table 2, respectively. Reductions in waist circumference were significantly greater with oral semaglutide than with empagliflozin at week 26 (both estimands) and at week 52 (trial product estimand) (Table 2).
More patients achieved the two composite end points (HbA1c <7% [53 mmol/mol] without severe or symptomatic hypoglycemia and no weight gain and an absolute reduction in HbA1c of ≥1.0% [10.9 mmol/mol] and body weight loss of ≥3%) with oral semaglutide versus empagliflozin, and the odds of doing so were significantly greater at both weeks 26 and 52 (Table 2). Reduction in C-reactive protein was significantly greater with oral semaglutide versus empagliflozin (Table 2). Other secondary end points are presented in Table 2 and Supplementary Table 3.
For the Control of Eating Questionnaire, the domains craving control (weeks 26 and 52) and craving for savory (week 52) were significantly improved in favor of oral semaglutide versus empagliflozin (treatment policy estimand). Both domains were significantly in favor of oral semaglutide at both weeks 26 and 52 for the trial product estimand. Patient-reported outcomes are summarized in Supplementary Fig. 7.
The overall number of adverse events and proportion of patients reporting adverse events were similar with oral semaglutide and empagliflozin, and most events were mild to moderate severity (Table 3). Fewer patients experienced serious adverse events in the oral semaglutide group. There was one death in the empagliflozin group (undetermined cause). The most frequent adverse event with oral semaglutide was nausea, which was nonserious, usually mild to moderate severity and transient, and did not exceed a prevalence of 10% at any time (Table 3 and Supplementary Fig. 8). Female and male genital mycotic infections of mild to moderate severity occurred more frequently with empagliflozin than with oral semaglutide (8.5% and 6.7% vs. 2.0% and 0%, respectively) (Supplementary Table 4).
On-treatment adverse events
Adverse events resulting in trial product discontinuation were more frequent with oral semaglutide than with empagliflozin (10.7% vs. 4.4%) and were primarily related to gastrointestinal symptoms (8.0% vs. 0.7%) (Table 3). In both groups, premature discontinuations mainly occurred in the first 16 weeks of treatment.
Incidence of severe or confirmed symptomatic hypoglycemic episodes (<56 mg/dL [<3.1 mmol/L]) was low and similar in both groups (Table 3). Diabetic retinopathy–related adverse events were reported in 14 (3.4%) patients in the oral semaglutide group and in 5 (1.2%) in the empagliflozin group (in-trial period) (Supplementary Table 5). All such events were identified by routine eye examination as part of the trial protocol and were nonserious, of mild or moderate severity, and did not require treatment. EAC-confirmed malignant neoplasms were identified in seven (1.7%) patients in the oral semaglutide group and two (0.5%) in the empagliflozin group (in-trial period). There was no clustering of malignancies in any particular organ or system (Supplementary Table 6). Cardiovascular events occurred at a similar rate in both groups (EAC confirmed; oral semaglutide n = 5 [1.2%], empagliflozin n = 6 [1.5%]) (Supplementary Table 6). Other EAC-confirmed events and safety assessments are reported in Supplementary Tables 6 and 7.
Oral semaglutide is the first oral GLP-1RA to be investigated for the treatment of type 2 diabetes. In PIONEER 2, oral semaglutide was superior to empagliflozin, with meaningful reductions in HbA1c at 26 weeks in patients with type 2 diabetes uncontrolled on metformin monotherapy. Furthermore, the difference between treatments remained significant at 52 weeks. Attainment of ADA-recommended HbA1c targets at 26 and 52 weeks was also significantly greater with oral semaglutide. Reductions in fasting plasma glucose were similar in both groups, suggesting differences in glycemic control may be mostly driven by the greater reduction in postprandial glucose with oral semaglutide.
Reductions in body weight occurred with both treatments, but superiority of oral semaglutide versus empagliflozin could not be confirmed at week 26. However, weight loss in the empagliflozin group stabilized around week 26, whereas in the oral semaglutide group, weight loss continued until around week 38 and was significantly greater at 52 weeks on the basis of the trial product estimand. This significantly greater weight loss at 52 weeks with oral semaglutide on the basis of the trial product estimand reflects the treatment effect without the confounding influence of rescue medication use and treatment discontinuations. Patients discontinuing oral semaglutide could not be switched to additional antidiabetic medication with a comparable weight-reducing effect, while patients on empagliflozin could be switched to GLP-1RAs.
The safety profile of oral semaglutide was consistent with previous trials (11–16). More patients prematurely discontinued treatment because of adverse events with oral semaglutide versus empagliflozin mainly as a result of gastrointestinal symptoms associated with dose escalation. The proportion of adverse events leading to discontinuation of oral semaglutide (10.7%) was similar to previous observations with injectable GLP-1RAs (6–11%) (4,33,34).
The use of subcutaneous semaglutide has previously been associated with a higher rate of diabetic retinopathy–related complications compared with placebo, which is consistent with the phenomenon of early worsening of preexisting diabetic retinopathy secondary to an initial, rapid improvement in glycemic control (6,35). The possible effect of subcutaneous semaglutide on diabetic eye disease is being further investigated in the ongoing FOCUS trial (NCT03811561) (36). In the current trial, diabetic retinopathy–related adverse events were more frequent with oral semaglutide compared with empagliflozin, although occurrence was low in both groups (3.4% vs. 1.2%). All events were nonserious, most were mild in severity, and none required treatment or led to trial product discontinuation. All were discovered during routine end-of-treatment eye examination and were diagnosed as nonproliferative diabetic retinopathy. In a longer-term, 78-week, double-blind trial, no imbalance in the occurrence of diabetic retinopathy–related events was observed between oral semaglutide 3, 7, and 14 mg and sitagliptin (6.7%, 6.0%, 5.6%, and 7.7%, respectively) (15). Occurrence of diabetic retinopathy–related events was also similar with oral semaglutide and placebo (7.1% vs. 6.3%) in a double-blind trial that assessed cardiovascular outcomes in patients at high cardiovascular risk (17).
This trial provides a comparison of two increasingly used drug classes that are commonly added to metformin when glycemic control is not achieved. The principal limitation of the trial was the open-label design.
In conclusion, the oral GLP-1 analog oral semaglutide was superior to the SGLT-2 inhibitor empagliflozin for reduction in HbA1c, but not body weight, at 26 weeks in patients with type 2 diabetes uncontrolled with metformin. Reductions in HbA1c were significantly greater with oral semaglutide at 52 weeks. Assessed by the trial product estimand, oral semaglutide provided significant reductions in body weight at 52 weeks. Oral semaglutide was well tolerated, with a safety profile consistent with that of GLP-1RAs.
Acknowledgments. Emisphere is acknowledged for providing a license to the Eligen Technology, the sodium N-(8-[2-hydroxylbenzoyl] amino) caprylate component of oral semaglutide. The authors thank the patients, investigators, trial site staff, and Novo Nordisk employees involved in the trial. In addition, the authors thank Andy Bond of Spirit Medical Communications Group Ltd. for medical writing and editorial assistance and Brian Bekker Hansen of Novo Nordisk for reviewing the manuscript. Novo Nordisk (the sponsor) designed the trial, monitored sites, and collected and analyzed the data. Editorial support was funded by the sponsor and provided by independent medical writers under the guidance of the authors.
Duality of Interest. This trial was funded by Novo Nordisk A/S, Denmark. H.W.R. reports consulting, advisory boards, clinical research, and lecturing for AstraZeneca, Boehringer Ingelheim, Janssen, Eli Lilly, Merck, Novo Nordisk, Sanofi, and Regeneron Pharmaceuticals. J.R. reports scientific advisory boards and honoraria or consulting fees from Eli Lilly, Novo Nordisk, Sanofi, Janssen, Boehringer Ingelheim, and Intarcia and grants/research support from Merck, Pfizer, Sanofi, Novo Nordisk, Bristol-Myers Squibb, Eli Lilly, GlaxoSmithKline, AstraZeneca, Janssen, Genentech, Boehringer Ingelheim, Intarcia, and Lexicon. L.H.C. reports clinical research for Novo Nordisk, Janssen, Eli Lilly, and Sanofi and lecturing for Sanofi and Boehringer Ingelheim. C.D. reports consulting, advisory boards, clinical research, and lecturing for AstraZeneca, Boehringer Ingelheim, Janssen, Eli Lilly, Novo Nordisk, Merck, Sanofi, Takeda, and Merck Sharp & Dohme. J.G. has received speaker’s or consulting honoraria from Novo Nordisk, Eli Lilly, Servier, Merck Sharp & Dohme, Bioton (Poland), Merck (Darmstadt), Sanofi, Polpharma (Poland), Polfa Tarchomin (Poland), AstraZeneca, and Boehringer Ingelheim. S.Ø.L., A.L.S., and M.B.T. are employees of Novo Nordisk A/S. A.L.S. and M.B.T. have shares in Novo Nordisk A/S. I.L. reports consulting, advisory boards, and/or research grants from Novo Nordisk, AstraZeneca, Boehringer Ingelheim, Sanofi, Eli Lilly, Intarcia, MannKind, Valeritas, Novartis, Mylan, Merck, and Pfizer. E.M. reports scientific advisory boards, consulting, lecturing, and/or research grants from AstraZeneca, Boehringer Ingelheim, Eli Lilly, Grupo Ferrer Internacional S.A., Intarcia, Menarini, Janssen, Servier, Merck Sharp & Dohme, Novo Nordisk, and Novartis. No other potential conflicts of interest relevant to this article were reported.
Author Contributions. H.W.R., J.R., L.H.C., C.D., J.G., S.Ø.L., I.L., A.L.S., M.B.T., and E.M. were responsible for the acquisition, analysis, or interpretation of data and the drafting or critical revision of the manuscript for important intellectual content. H.W.R. and E.M. were signatory investigators on the study. S.Ø.L. and M.B.T. were involved in the concept and design of the study. A.L.S. was responsible for the statistical analysis. H.W.R. and E.M. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Prior Presentation. Parts of this study were presented in oral form at the 79th Scientific Sessions of the American Diabetes Association, San Francisco, CA, 7–11 June 2019.
Epigenetic changes may contribute substantially to risks of diseases of aging. Previous studies reported seven methylation variable positions (MVPs) robustly associated with incident type 2 diabetes mellitus (T2DM). However, their causal roles in T2DM are unclear. In an incident T2DM case-cohort study nested within the population-based European Prospective Investigation into Cancer and Nutrition (EPIC)-Norfolk cohort, we used whole blood DNA collected at baseline, up to 11 years before T2DM onset, to investigate the role of methylation in the etiology of T2DM. We identified 15 novel MVPs with robust associations with incident T2DM and robustly confirmed three MVPs identified previously (near to TXNIP, ABCG1, and SREBF1). All 18 MVPs showed directionally consistent associations with incident and prevalent T2DM in independent studies. Further conditional analyses suggested that the identified epigenetic signals appear related to T2DM via glucose and obesity-related pathways acting before the collection of baseline samples. We integrated genome-wide genetic data to identify methylation-associated quantitative trait loci robustly associated with 16 of the 18 MVPs and found one MVP, cg00574958 at CPT1A, with a possible direct causal role in T2DM. None of the implicated genes were previously highlighted by genetic association studies, suggesting that DNA methylation studies may reveal novel biological mechanisms involved in tissue responses to glycemia.
Type 2 diabetes mellitus (T2DM) is a major and increasing public health problem. Genome-wide studies have identified more than 240 genetic variants (1) that are robustly associated with T2DM. However, these only explain a minor portion of T2DM susceptibility variance (2,3). Environmental factors, including diet and physical activity, and also early life factors during fetal and early postnatal development are reported to contribute to the etiology of T2DM. Epigenetic variation can occur as a result of genetic and/or environmental factors (4). DNA methylation (DNAm) at cytosine-guanine dinucleotides (CpG sites) is the most commonly studied epigenetic mechanism to date, due to its role in expression regulation and available assays to quantify DNAm intensity at multiple sites across the epigenome that are applicable to large-scale studies. Unlike genotypic variation, DNAm intensity patterns are liable to change over time, with age or following disease or other exposure, and therefore disease-associated changes may be either causal or consequential (5).
Previous epigenome-wide association studies (EWAS) have identified seven methylation variable positions (MVPs) that are significantly associated (P < 1.0 × 10−7) with incident T2DM (6,7). However, the causal role of those markers in T2DM is unclear. Here, we aimed to elucidate DNAm determinants of T2DM by performing an EWAS for incident T2DM in the European Prospective Investigation into Cancer and Nutrition (EPIC)-Norfolk study (8). By further integrating genome-wide genetic array data, we aimed to identify methylation quantitative trait loci (methQTLs) for any T2DM-associated MVPs in order to assess the likely causal role of DNAm markers in T2DM through Mendelian randomization analyses (9).
Research Design and Methods
The discovery phase EWAS was undertaken in an incident T2DM case-cohort study nested within the EPIC-Norfolk study (8), a prospective cohort study that recruited 25,639 individuals aged between 40 and 79 years at baseline in 1993–1997. The cohort was representative of the general population of England and Wales for age, sex, anthropometric measures, blood pressure, and serum lipids but differed in that 99.7% of the cohort were of European descent. We defined a random subcohort of the whole EPIC-Norfolk study population excluding known prevalent case subjects of diabetes at baseline using the same definitions as used in the InterAct Project (10) who had available genotype data. Incident T2DM cases were ascertained from multiple sources: two follow-up health and lifestyle questionnaires providing self-reported information on doctor-diagnosed diabetes or medications, medications brought to the second clinical exam, and medical record linkage. Record linkage to external sources included the listing of any EPIC-Norfolk participant in the general practice diabetes register, local hospital diabetes register, hospital admissions data with screening for diabetes-related admissions, and Office of National Statistics mortality data with coding for diabetes. Participants who self-reported a history of diabetes that could not be confirmed against any other sources were not considered confirmed case subjects. Follow-up was censored at date of diagnosis of T2DM, 31 July 2006, or date of death—whichever came first. By definition in a case-cohort design, there are case subjects within and outside the random subcohort, but for the purposes of this analysis, we considered them in the incident case set only, with noncase subjects forming the comparison group. BMI and HbA1c levels were measured for each participant at baseline (Table 1). All participants in the EPIC-Norfolk study gave signed informed consent, and the study was approved by the local research ethics committee.
Baseline characteristics of participants in the EPIC-Norfolk, LOLIPOP, and FHS study samples
Confirmation of top signals from the discovery EWAS was sought in two further studies. The London Life Sciences Prospective Population (LOLIPOP) study is a prospective population study of Indian Asian (N = 17,606) and European (N = 7,766) individuals, recruited at age 35–75 years from the lists of 58 family doctors in west London, U.K., between 1 May 2002 and 12 September 2008. Indian Asians had all four grandparents born on the Indian subcontinent (India, Pakistan, Sri Lanka, or Bangladesh). The LOLIPOP study is approved by the National Research Ethics Service (07/H0712/150), and all participants gave written informed consent at enrolment. The LOLIPOP nested case-control study of incident T2DM has previously been described (6). Briefly, at follow-up, on 31 December 2013, individuals with T2DM were identified by primary care electronic health records and structured queries. Participants with incident T2DM were defined as those who did not have T2DM at baseline but who developed the disease during follow-up. Control subjects were identified from a random subset of 7,640 participants with a clinical assessment of fasting blood glucose concentration and HbA1c and questionnaire assessment between 11 January 2010 and 31 December 2013.
The Framingham Heart Study (FHS) is a community-based longitudinal study of participants living in and near Framingham, MA, at the start of the study in 1948 (11). The Offspring cohort comprised the children and spouses of the original FHS participants, as previously described (12). Briefly, enrollment for the Offspring cohort began in 1971 (N = 5,124), and in-person evaluations occurred approximately every 4–8 years thereafter. The current analysis was limited to participants from the Offspring cohort who survived until the eighth examination cycle (2005–2008) and consented to genetics research. DNAm data of peripheral blood samples collected at the eighth examination cycle were available in 2,741 participants. Prevalent T2DM was defined as having fasting glucose ≥7 mmol/L or as reporting taking T2DM medication at any examination cycle up to the eighth examination. All participants provided written informed consent at the time of each examination visit. The study protocol was approved by the institutional review board at Boston University Medical Center (Boston, MA).
Methylation Array Profiling
In all studies, DNAm intensity was measured using the Illumina Infinium Human Methylation 450K BeadChip array (12-sample array for FHS and 96-sample array for EPIC-Norfolk and LOLIPOP). Bisulfite conversion of DNA was performed using the EZ DNAm kit (Zymo Research, Orange, CA).
For 1,378 EPIC-Norfolk participants, DNAm was measured in DNA extracted from whole blood samples collected at baseline. Converted DNA was assayed by PCR and gel electrophoresis. Each 96-well DNA sample plate contained two duplicate samples. The average correlation between the duplicate samples was 98%.
In LOLIPOP, DNAm was measured among the first 1,074 Indian Asian participants with incident T2DM and 1,590 matched Indian Asian control subjects. Control subjects were matched to case subjects by age (5-year groups) and sex. DNAm was quantified in the baseline DNA samples collected at study enrollment. Samples were analyzed in random order, with masking to case-control status.
In FHS, peripheral blood samples were collected at the eighth examination (2005–2008). Genomic DNA was extracted from buffy coat using the Gentra Puregene DNA extraction kit (QIAGEN). Bisulphite-converted DNA samples were hybridized to the 12-sample Infinium HumanMethylation450 BeadChip array using the Infinium HD Methylation Assay protocol and Tecan robotics (Illumina, San Diego, CA). DNAm quantification was conducted in two laboratory batches.
EWAS Quality Control and Normalization
In EPIC-Norfolk, epigenome-wide DNAm data were analyzed in R (version 3.2.2). Initial quality control was performed as recommended by the array manufacturer; methylation intensity values were corrected using the Illumina background correction algorithm as implemented in minfi (13), methylation intensities with a detection P value ≥ 0.01 were set to “missing,” and methylation intensity β values were calculated for each methylation marker per sample. For duplicate samples, the sample with the lower CpG detection percentage was excluded.
Sample call rates were calculated as the proportion of missing data in each sample, by autosomal, X and Y chromosomes. For the autosomal data, 77 samples with a call rate ≤ 0.99 were excluded. All samples passed the call rate threshold on the X chromosome. For the Y chromosome, seven male samples that did not pass the call rate and two further female samples were excluded. Distributions of methylation intensities were also inspected by autosomal and sex chromosomes and separately in females and males, leading to the exclusion of two additional samples that had an unusual distribution of methylation intensities. After those quality control procedures, data on 1,290 samples remained. All further downstream analyses were restricted to autosomal methylation markers.
Marker call rates were calculated as the proportion of missing data at each CpG site. 8,775 CpGs with a call rate ≤0.95 were excluded. The R package ENmix (14) was used to identify CpG sites with multimodal distributions of methylation intensity, which typically arise from technical artifacts; 3,295 such CpG sites were excluded. A further 18,874 CpG sites with probes previously identified as mapping to more than one genomic location were also excluded (15).
To ensure reliability of the data, we repeated filtering on sample and marker call rates until all samples and all markers passed their respective call rate thresholds. After exclusion of prevalent T2DM case subjects at baseline, the final data set comprised 1,264 samples (563 incident T2D case subjects, including 22 case subjects from the subcohort, and 701 noncase subjects) with methylation intensities at 442,920 autosomal CpG sites. Quantile normalization of methylation intensity β values was applied separately to the different subgroups of markers based on color channel, probe type, and methylated/unmethylated subtypes as proposed by Lehne et al. (16)
In LOLIPOP, DNAm data were analyzed in R (version 2.15) using minfi (13) and other R scripts. Marker intensities were normalized by quantile normalization as previously described (6).
In FHS, DNAm data were normalized using the Dasen methodology implemented in the wateRmelon package (17). Sample exclusion criteria included poor single nucleotide polymorphism (SNP) matching of control positions, missing rate >1%, outliers from multidimensional scaling, and sex mismatch. Probes were excluded if missing rate was >20%. Data from laboratory batches were pooled, leaving up to 2,635 samples and 443,304 CpG probes for analysis. Additional information on DNAm, normalization, and quality control is available the previously published work by Aslibekyan et al. (18). Differences in DNAm data generation, quality control, and statistical models are summarized in Supplementary Table 1.
EWAS Statistical Analyses
In EPIC-Norfolk, to identify MVPs associated with incident T2DM, we performed a logistic regression model for each methylation marker with incident T2DM status, with adjustment for age, sex, estimated cell counts, and sample plate using the EWAC pipeline. A conservative multiple test–corrected P value threshold was applied (P < 1 × 10−7). Different methylation profiles have been observed between the different cell types in whole blood (19), and blood-based profile of DNAm was shown to predict the underlying distribution of cell types (20). To correct for cell composition variability (21), we estimated first the proportions of different cell types (CD4+ and CD8+ T-cell subtypes, natural killer cells, monocytes, granulocytes, and B cells) from DNAm data using the algorithm described by Houseman et al. (22) as implemented in the R package minfi (13). These cell count estimates were then used as covariates in the epigenome-wide regression models for incident T2DM.
We used STRING (23) to perform gene set enrichment on the significant genes associated with the 18 significant MVPs identified in the EWAS. We also performed a modified version of the MAGENTA (24) pipeline to identify the pathways associated with genes at the loci of the significant MVPs. Since MAGENTA uses SNP data to identify loci, we assigned to each CpG a “nearest SNP” based on HapMap3 data and using build 36 positions for both the CpG site and the SNPs (average distance to the nearest SNP = 4,175 base pairs [bp] [interquartile range 1,375–4,859]; 1,707 of 466,039 CpGs were not assigned a SNP). In effect, rather than using a SNP P value to rank genes to assess enrichment, we used the P value from the methylation site to run MAGENTA.
For LOLIPOP, an epigenome-wide association of DNAm was performed in Indian Asians with incident T2DM who were identified from the 8-year follow-up of the study. Differential white blood cell (lymphocyte, monocyte, and granulocyte) count was available for all participants, and epigenome-wide methylation scores were used to impute a further four lymphocyte subsets (CD4, CD8, natural killer, and B cells). Principal components analysis was performed to quantify latent structure in the data, including batch effects. Associations between incident T2DM and the 18 significant MVPs identified in EPIC-Norfolk were tested using logistic regression including intensity values from Infinium HumanMethylation450 BeadChip assay control probes, bisulfite conversion batch, measured white cells, and imputed white cell subsets and the first five principal components as covariates. Association results were corrected for the genomic control inflation factor. For testing of the predictive ability of the 18 markers for incident T2DM, univariate logistic regressions were run for each of the 18 markers to obtain individual effect sizes (β values) for incident T2DM. A weighted methylation risk score was subsequently calculated from these β values, and receiver operating curve analyses were performed to provide estimates for area under the curve (AUC).
In FHS, association between each identified MVP (associated with incident T2DM in EPIC-Norfolk) was tested for association with prevalent diabetes and glycemic traits (fasting glucose, fasting insulin, and HbA1c). The analysis of glycemic traits included only individuals without diabetes. Fasting insulin was natural log transformed. Random effects statistical models were used to analyze the data to account for sibling correlation and included adjustments for age, sex, white blood cell counts, technical covariates, batch effects, and BMI, with DNAm as the dependent variable.
We also examined each T2DM-associated MVP for additional cross-sectional association with type 1 diabetes mellitus (T1DM) in an earlier EWAS of 52 monozygous twin pairs discordant for T1DM, in cell-sorted peripheral blood mononuclear cells (monocytes, B cells, or T cells) (25). As T2DM and T1DM have largely differing aetiologies, MVPs that are consistently associated with both outcomes may indicate metabolic effects of diabetes on DNAm.
The relevance of changes in DNAm intensity in whole blood to other tissues was tested by analysis of genome-wide DNAm data, generated using the Illumina Infinium Human Methylation 450K BeadChip array, from human liver, adipose tissue, and skeletal muscle, as previously published (26). Human liver DNAm data were from participants of the Kuopio Obesity Surgery Study (KOBS); 35 with T2DM and 60 without (27). Data on adipose tissue (14 pairs), skeletal muscle (17 pairs), and blood (19 pairs) were from monozygotic twins discordant for T2DM (26,28,29). Adipose tissue and skeletal muscle from the same individual were available for most of these twin pairs (16 pairs in blood/muscle and 14 pairs in blood/fat); concordance in DNAm intensity across these tissues was tested for each highlighted MVP by Spearman correlation tests. We further tested cross-tissue correlations in DNAm at T2DM-associated MVPs between blood and other tissues of relevance to T2DM etiology, liver, and pancreas in publicly available Infinium HumanMethylation450 BeadChip array data from six cadavers sampled within 12 h postmortem (mean [SD] age 65.5 [7.2] years) (30).
Mendelian Randomization Analyses
We performed bidirectional Mendelian randomization analyses to test whether any T2DM-associated MVPs had a causal effect on T2DM or are a consequence of metabolic differences that had originated before the baseline measurement in this study. To predict the causal effect of each of T2DM-associated MVPs on T2DM, methQTLs associated with each MVP (FDR <0.05) in whole blood in 3,841 adults of European descent were identified using the BIOS QTL browser (31). To run Mendelian randomization analyses, we converted the Z score for each methQTL to β and SE using the following formulas (32):where N is the sample size and MAF is the minor allele frequency. We then tested these methQTLs in Mendelian randomization analyses (9) for T2DM. Genetic associations with T2DM were estimated in 69,677 case and 551,081 control subjects from the UK Biobank study (33), the EPIC-InterAct study (10) and the DIAbetes Genetics Replication And Meta-analysis (DIAGRAM) consortium (2). A summary statistics method (inverse variance weighted) that combines all the SNPs for each MVP as a genetic instrument was used to predict the effect of that MVP on T2DM (34). To ensure that the instruments are independent, we performed clumping. The Egger regression for Mendelian randomization was also used to assess the sensitivity of the results to violations of Mendelian randomization assumptions. Mendelian randomization analyses were run using the R package TwoSampleMR (35).
For the reverse direction causal assessment, we tested SNPs with previously reported associations with T2DM (2) or related metabolic phenotypes (BMI , fasting glucose , 2-h glucose , fasting insulin , fasting insulin adjusted for BMI , insulin resistance , insulin secretion , and waist-hip-ratio adjusted for BMI ) to test whether these traits have causal effects on methylation intensity at any T2DM-associated MVP. We used summary statistics methods (inverse variance weighted and Egger tests) that combine all the SNPs for each trait as a genetic instrument to predict the effect of that trait on each T2DM-associated MVP (34) in the cohort control samples of EPIC-Norfolk (N = 613), in which genotype data were generated using the Affymetrix Axiom UK Biobank chip. All genotypes passed standard quality control criteria as specified by the Affymetrix best practice pipeline, and SNPs with MAF <5% in this sample were excluded.
MVPs Associated With Incident T2DM
In the EPIC-Norfolk study, we identified 18 MVPs that are associated with incident T2DM at P < 1 × 10−7, including 15 novel signals (Table 2). None of these was reported to have a SNP on the target CpG (15). The two strongest associations were the previously reported signals at TXNIP (cg19693031; P = 2.7 × 10−21) and ABCG1 (cg06500161; P = 6.4 × 10−14) (6,7). We confirmed a third previously reported signal at SREBF1 (cg11024682; P = 6.0 × 10−10) and provide supportive evidence for an additional signal at PROC (cg09152259; P = 4.2 × 10−4) that had previously not been considered to be true due to lack of replication in Europeans (Supplementary Table 2).
MVPs associated with incident T2DM at P < 1.0E-07 in EPIC-Norfolk (N = 1,264)
We sought confirmation of the top 18 MVPs in data on 1,074 incident T2DM cases and 1,590 control samples from the LOLIPOP study and in cross-sectional data from FHS (403 with prevalent T2DM and 2,204 control subjects) (Table 3). All 18 MVPs showed directionally consistent associations with incident T2DM (14 at P < 0.05) and prevalent T2DM (16 at P < 0.05).
Confirmation of the top 18 T2DM-associated MVPs in LOLIPOP and FHS
Novel MVPs associated with incident T2DM include cg14476101 (P = 2.8 × 10−10), located in the gene body of PHGDH, which encodes phosphoglycerate dehydrogenase, an enzyme involved in the synthesis of l-serine and other amino acids, and cg00574958 (P = 5.2 × 10−9) in the 5′ UTR (untranslated region) of CPT1A, which encodes an enzyme that initiates mitochondrial oxidation of long-chain fatty acids (Supplementary Table 11). Four of the 18 MVPs were located within solute carrier family genes (SLC1A5, SLC43A1, SLC9A1, and SLC9A3R1), which encode plasma membrane proteins that regulate cell transport of amino acids and other metabolites.
To systematically explore the biological pathways implicated by T2DM-associated methylation signals, we first tested the 18 MVPs for gene set enrichment using STRING and identified significant enrichment for three pathways: “positive regulation of cholesterol biosynthetic process” (indicated by MVPs at ABCG1, SREBF1, and POR), “carnitine metabolic process” (indicated by CPT1A and POR), and “AMPK signaling” (indicated by PFKFB3, CPT1A, and SREBF1). We then tested the full EWAS data set in a modified MAGENTA pipeline and identified significant enrichment for T2DM-associated methylation signals in 10 pathways (Supplementary Table 4), including “insulin receptor signaling,” “IGF-1 signalling,” “erythropoietin signaling,” “JAK signaling,” and “integrin signaling.”
MVPs Associated With Glycemic Traits
In nondiabetic control FHS samples, all 18 T2DM-associated MVPs showed directionally concordant associations with fasting glucose, fasting insulin levels and BMI, and 16 of the 18 MVPs showed directionally concordant associations with HbA1c (Supplementary Table 5). In additional conditional models in the EPIC-Norfolk discovery sample, the associations of all individual 18 MVPs with incident T2DM were markedly attenuated when models were further adjusted for baseline BMI and HbA1c (median attenuation 49% [Supplementary Table 3]), indicating that these DNAm intensity changes largely reflect baseline differences between future incident T2DM cases and other cohort participants.
Furthermore, among 52 monozygous twin pairs discordant for T1DM, 7 of the 18 T2DM-associated MVPs showed cross-sectional differences in DNAm intensity in peripheral white blood cells (monocytes, B cells, or T cells) between the T1DM-affected and -unaffected twins, consistent with an effect of glycemia on DNAm intensity (at TXNIP, SLC9A3R1, SREBF1, CPT1A, C7orf50, PFKFB3, and cg08309687) (Supplementary Table 6).
Relevance of Whole Blood MVPs to Other Tissues
To explore the possible relevance of changes in DNAm intensity in whole blood to other tissues, relevant to T2DM pathogenesis, we examined these 18 MVPs in liver, adipose tissue, and skeletal muscle from individuals with and without T2DM. Nominal associations (P < 0.05) were found with only our 2 strongest whole blood MVP signals: cg06500161 at ABCG1 in adipose tissue (as previously published ) and cg19693031 at TXNIP in skeletal muscle (Table 4). Furthermore, at 12 of the 18 MVPs there was evidence for a positive correlation in DNAm intensity between whole blood and liver, pancreas, adipose tissue, or muscle (Supplementary Table 7).
Analysis of the top 18 T2DM-associated MVPs in nonblood tissues
Causal Effects on T2DM
To investigate the potential causal effects of the 18 T2DM-associated MVPs, we used the BIOS QTL browser (31) to identify methQTLs (genetic sequence variants) that are robustly associated (at P < 5 × 10−8) with DNAm intensity at any of the 18 MVPs. We found 54 methQTLs (33 cis, 21 trans), each associated with one of 16 MVPs (Supplementary Table 8). We then used these methQTLs as instrumental variables in Mendelian randomization analyses, based on aggregated publicly available GWAS data in 69,677 T2DM case and 551,081 control subjects (DIAGRAM , UK Biobank , and EPIC-InterAct ). Only one of the 16 T2DM-associated MVPs with an identified methQTL showed nominal evidence for a direct causal association with T2DM, cg00574958 at CPT1A (P = 0.01); however, for other MVPs the genetic-predicted effects overlapped with the observed effects in the LOLIPOP study (Fig. 1 and Supplementary Table 9).
Predicted causal effects of DNAm on T2DM. The scatterplot shows the genetic-predicted effects of DNAm intensity on risk for T2DM (y-axis) plotted against observed effect estimates (from the LOLIPOP confirmation phase [x-axis]) at each of 16 top-hit MVPs (see Supplementary Table 7). Effect sizes are log–odds ratios per 1-unit change in normalized methylation intensity aligned to higher observed odds of T2DM.
We performed reverse direction causal analyses to identify causal effects of BMI and glycemic traits on methylation intensity at the 18 MVPs. Among participants without T2DM in EPIC-Norfolk (N = 613), none of the genetic instruments for the tested glycemic or metabolic traits (T2DM, BMI, fasting glucose, 2-h glucose, fasting insulin, fasting insulin adjusted for BMI, insulin resistance, insulin secretion, and waist-to-hip ratio adjusted for BMI) showed a consistent association with any of the 18 T2DM-associated MVPs (Supplementary Table 10).
Prediction of T2DM
In the LOLIPOP study sample, which was independent of the discovery EWAS, the top 18 T2DM-associated MVPs in aggregate showed no predictive ability for incident T2DM (AUC = 0.53). Furthermore, the addition of these 18 MVPs did not improve on a prediction model based on other baseline phenotypes (BMI, HbA1c, age, sex: AUC = 0.761; BMI, HbA1c, age, sex, plus 18 MVPs: AUC = 0.762).
In this prospective study, we substantially increased the number of MVPs in whole blood that are robustly associated with incident T2DM. Associations for 17 of the 18 MVPs were confirmed with either incident or prevalent T2DM in two independent studies, which indicates the consistency of T2DM-associated whole blood DNAm intensity changes across different settings and ethnicities. Genetic causal modeling identified evidence to support a causal effect of DNAm on T2DM at one of these MVPs, cg00574958 at CPT1A.
The prospective designs of the EPIC-Norfolk and LOLIPOP studies aimed to identify MVPs that precede the development of T2DM. However, the identified T2DM-associated DNAm intensity changes were largely attenuated by adjustment for differences in BMI and glycemia that had developed prior to the baseline measurement in the prospective studies. Our Mendelian randomization analyses failed to find evidence for direct causal effects for the majority of T2DM-associated MVPs, as indicated by no detectable genetic-predicted effect of DNAm intensity on T2DM and a wide discordance between the observed and genetic-predicted effects. Conversely, overlap between EWAS signals for T2DM and T1DM was consistent with effects of glycemia on DNAm intensity for at least 7 of the 18 T2DM-associated MVPs.
Whether or not they show directly causal associations, these novel and consistent T2DM-associated MVPs are highly informative with regard to implicated genes and biological pathways. Notably, none of the genes implicated by this EWAS were previously identified by genetic variant association studies. This stark difference may suggest that T2DM-associated DNAm intensity changes may reveal novel biological mechanisms involved in tissue responses to glycemia rather than in the pathogenesis of insulin resistance or insulin secretion, which are implicated by those genetic studies. The topmost signal, cg19693031, which lies on TXNIP, is also the most significant observation in other T2DM EWAS studies (6,7). Phosphoglycerate dehydrogenase (PHGDH) catalyzes the first and rate-limiting step in glucose-derived serine synthesis and may indicate consequent purine and deoxythymidine nucleotide synthesis in response to hyperglycemia and potential tissue proliferative responses (43). Functional variation in carnitine palmitoyltransferase 1 (CPT1A) regulates the composition of circulating polyunsaturated n-3 fatty acids and docosahexaoenic acid (44) and is reported to activate lipolysis and mitochondrial activity in brown fat (45,46) and to maintain pancreatic islet secretion of the principal hyperglycemic hormone, glucagon (47). Solute carrier family members are sodium-dependent membrane transporters that regulate intracellular cell pH, cell volume, and other cellular events such as adhesion, migration, and proliferation and also contribute to systemic homeostasis of fluid volume, acid-base balance, and electrolytes. Specifically, SLC9A3R1 (NHERF1) binds to PTEN to activate the PI3 kinase signaling cascade involved in cell survival, growth, proliferation (48) and is a key component of insulin and IGF-1 signaling pathways that we found enriched for T2DM EWAS associations. These highlighted pathways could potentially contribute to the pathogenesis of micro- and macrovascular complications of hyperglycemia. PFBK3, a regulator of glycolysis and insulin signaling in mice, was recently highlighted by a SNP association with late-onset autoimmune diabetes, and we here provide independent evidence to support its role in human glucose regulation (49).
We recognize a number of limitations of our study. Both of the prospective study samples displayed large differences in baseline glycemia and BMI between incident T2DM case and noncase subjects. This nested prospective study design aimed to identify interactions between genetic factors and baseline lifestyle factors measured prior to the development of clinically diagnosed T2DM (10). Since it is impossible to develop T2DM except by passing through a phase of nondiabetic hyperglycemia, it is inevitable that people who go on to develop incident diabetes in a cohort study will have raised glucose levels at baseline if follow-up is of short or medium duration. Future studies that have samples stored many years prior to disease onset would be required to identify when in the development of diabetes the T2DM-MVP associations become apparent. Secondly, our assessments of other, nonblood, tissues were limited in the range of tissues and numbers of samples available. Despite concordant changes in DNAm intensity between whole blood and various tissues relevant to T2DM pathogenesis at 12 of the 18 T2DM-associated MVPs, nominal differences in DNAm were found only for our strongest two MVPs, which suggests that larger study samples are needed. We recognize that whole blood is not a tissue of interest to the pathogenesis of T2DM; however, current, and most likely future, large-scale EWAS are confined to such samples, and functional insights will depend on follow-up of whole blood signals in other tissues (50,51). The same issue of appropriate tissue of interest limits our genetic modeling approach, which identified genetic markers of DNAm intensity in peripheral blood. Furthermore, the sample size for this approach (N = 3,841 in BIOS QTL  and N = 613 in the EPIC-Norfolk cohort control group) is relatively small compared with data on QTLs for gene expression in peripheral blood (N = 8,086 in the study by Westra et al. ). Hence, we found only nominal evidence for a causal effect of DNAm at only 1 of the 18 T2DM-associated MVPs, at CPT1A, and for several MVPs the genetic-predicted effects were overlapping with the observed effects. Similarly, a recent large EWAS for BMI found a causal role of methylation at only one MVP (cg26663590 at NFATC2IP) (53). There are various possible conceptualizations of the functional interplay between SNP, MVP, and T2DM, which provide alternative explanations other than SNP-to-DNAm-to-T2D (54), but they do not limit the statistical detection of apparent causal signals. Future, larger reference data on QTLs for DNAm intensity in whole blood are being generated (Genetics of DNA Methylation Consortium [GoDMC]), which will allow more powerful tests for causality, although their relevance to DNAm in tissues of interest remains an important question. Finally, the determinants of the identified T2DM-associated MVPs remain unknown. Again, larger reference panels of GWAS and DNAm array data, as well as new methods to integrate findings across multiple methQTLs for each MVP, will inform future causal analyses. Future studies are needed to identify the potential lifestyle and developmental determinants of these T2DM-associated MVPs.
In conclusion, we identified several robust and consistent DNAm markers for incident T2DM. These appear to be related to T2DM via glucose and obesity-related pathways that had their effects before the collection of baseline samples in these cohort studies, which commenced in midlife. These associations indicate several plausible biological mechanisms involved in tissue responses and comorbidities of hyperglycemia.
Acknowledgments. The authors are grateful for all of the participants and staff of the EPIC-Norfolk, LOLIPOP, and FHS cohorts. The authors thank Dr. Stephen Burgess (University of Cambridge) for advice on methQTLs and Dr. Jan Bert van Klinken (Leiden University) for advice on the BIOS QTL data as well as Stephen Sharp and Dr. Jian’an Luan (University of Cambridge) for advice on statistical analyses and Ylva Wessman, Per-Anders Jansson, and Emma Nilsson (Lund University Diabetes Center) for help with the study on twin pairs discordant for T2DM.
Funding. EPIC-Norfolk is supported by program grants from the Medical Research Council (MRC) [G9502233, G9502233, and G9502233] and Cancer Research UK [C864/A8257]. The generation and management of the Illumina Infinium Human Methylation 450K BeadChip array data in this cohort are supported through the MRC Cambridge initiative in metabolomic science [MR/L00002/1]. The genome-wide genotyping data in EPIC-Norfolk was funded by MRC award MC_PC_13048. This work is also supported by MRC program grants MC_UU_12015/1, MC_UU_12015/2, and MC_UU_12015/5. The LOLIPOP study is supported by the National Institute for Health Research (NIHR) Comprehensive Biomedical Research Centre Imperial College Healthcare National Health Service (NHS) Trust, the British Heart Foundation (SP/04/002), the MRC (G0601966 and G0700931), the Wellcome Trust (084723/Z/08/Z, 090532, and 098381), the NIHR (RP-PG-0407-10371), the NIHR Official Development Assistance (award 16/136/68), and the European Union Seventh Framework Programme (FP7) (EpiMigrant, 279143) and Horizon 2020 Framework Programme (iHealth-T2D, 643774). We acknowledge support of the MRC-PHE Centre for Environment and Health and the NIHR Health Protection Research Unit in Health Impact of Environmental Hazards. The work was carried out in part at the NIHR/Wellcome Trust Imperial Clinical Research Facility. J.C.C. is supported by the Singapore Ministry of Health’s National Medical Research Council under its Singapore Translational Research Investigator (STaR) Award (NMRC/STaR/0028/2017). The FHS is supported by grants N01-HC-25195 and HHSN268201500001I. The laboratory work for this investigation was funded by the Division of Intramural Research, National Heart, Lung, and Blood Institute, and by the National Institutes of Health (NIH) Director’s Challenge Award (principal investigator: D.L.). The analytical component of this project was funded by the Division of Intramural Research, National Heart, Lung, and Blood Institute, and the Center for Information Technology, NIH. J.B.M. is supported by National Institute of Diabetes and Digestive and Kidney Diseases grants U01 DK078616 and K24 DK080140. Data on T1DM-discordant twin pairs arose from studies funded by the EU FP7 project BLUEPRINT (282510). The Cardiovascular Epidemiology Unit at the University of Cambridge is supported by the U.K. MRC (MR/L003120/1), British Heart Foundation (RG/13/13/30194), and NIHR (Cambridge Biomedical Research Centre at the Cambridge University Hospitals NHS Foundation Trust). Data from human tissues are from studies supported by grants from the Novo Nordisk foundation; Swedish Research Council, Region Skåne (ALF); Euoropean Foundation for the Study of Diabetes; EXODIAB; Swedish Foundation for Strategic Research (IRC15-0067); Swedish Diabetes Foundation; and Albert Påhlsson Foundation.
The views expressed are those of the authors and do not necessarily represent the views of the NHS, the NIHR, the Department of Health and Social Care, the National Heart, Lung, and Blood Institute, the NIH, or the U.S. Department of Health and Human Services. The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the manuscript.
Duality of Interest. A.Y.C. is currently employed by Merck Research Laboratories. No other potential conflicts of interest relevant to this article were reported.
Author Contributions. A.C., F.R.D., J.R.B.P., M.L., A.Y.C., B.L., D.S.P., I.D.S., and C. Liu performed data analyses. A.C., F.R.D., and K.K.O. drafted the manuscript. A.C. constructed the figure. A.C., F.R.D., J.R.B.P., N.J.W., and K.K.O. had full access to all of the data in the study. A.C. and K.K.O. had final responsibility for the decision to submit for publication. L.A.L., N.D.K., R.A.S., K.-T.K., N.G.F., C.La., M.M.M., D.L., S.B., R.D.L., J.D., J.B.M., J.S.K., M.-F.H., J.C.C., N.J.W., and K.K.O. contributed to the data collection. K.-T.K., N.G.F., J.D., J.B.M., J.S.K., C.Lin., M.-F.H., J.C.C., N.J.W., and K.K.O. contributed to the study design. All authors contributed to data interpretation and revisions of the manuscript. A.C. and K.K.O. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.