There are many different approaches to nutrition and blood sugar management for people living with diabetes.
Some people swear by low-carb or ketogenic diets, some prefer diets with more balanced macronutrients, while others are strictly vegan.
All of these diets work and we don’t advocate for one particular nutritional approach here on Diabetes Strong. Instead, we want you to know about the different options so you can make an informed choice for what works for you, your body, and your diabetes.
Today, we will take a look at the very interesting high-carb, plant-based approach from the newly released book Mastering Diabetes.
I recently had the pleasure of exploring the Mastering Diabetes approach to nutrition for people living with diabetes and it’s fascinating. It’s very different from the way I live my life, but it might just be perfect for you.
Cyrus Khambatta, Ph.D. and Robby Barbaro, MPH, the guys behind the Mastering Diabetes approach, both live with diabetes and they teach an approach that focuses on eating a plant-based diet consisting of plenty of carbohydrates and very limited fats.
Yes, you heard that right, a high-carb diet!
You might wonder how to make a high-carb diet work when living with diabetes or why that would even be something to explore. To answer some of these questions, I sat down with Robby Barbaro to discuss the new book and learn a bit more about the Mastering Diabetes approach and the philosophy behind it.
What has it meant for your personal health to follow this lifestyle?
Before adopting the Mastering Diabetes Method, I suffered from cystic acne, plantar fasciitis, chronic allergies, and glucose intolerance. Each of these conditions disappeared when I changed my lifestyle and maximized my insulin sensitivity.
I now eat over 700 grams of total carbohydrate per day and inject an average of 27 total units of insulin per day (about 13 units of long-acting insulin and 14 units of fast-acting insulin).
Cyrus, my co-founder and co-author, is an active athlete and has been playing soccer and lifting weights for more than 20 years. When he ate a low-carbohydrate diet (as his doctors had instructed him to do immediately after being diagnosed with type 1 diabetes in 2002), his energy levels decreased, his blood glucose was very challenging to control, and he wasn’t able to exercise as frequently or as intensely as he had his entire athletic career.
When he transitioned to a plant-based diet, his energy levels increased immediately, his blood glucose became significantly more controllable, and his insulin use fell by 40%.
He eats 600-700 grams of carbohydrate energy per day and injects 25-30 units of insulin per day (11 units of long-acting insulin and 14-19 units of rapid-acting insulin), and is more active than he’s ever been.
What is The Mastering Diabetes Method and how does it work?
The Mastering Diabetes Method is a system that is designed to reverse insulin resistance. First, let’s review what insulin resistance is and why it’s something worth understanding.
Insulin resistance occurs when your muscles and liver have an impaired response to the action of insulin. This is caused by the accumulation of excess fat in tissues that are not designed to store large quantities of fat, resulting in a reduced ability of insulin to promote glucose uptake in both muscle and liver cells.
Insulin resistance affects people living with all forms of diabetes (including type 1, type 1.5, prediabetes, type 2, and gestational diabetes), which dramatically increases your risk for coronary artery disease, atherosclerosis, cancer, high cholesterol, high blood pressure, obesity, peripheral neuropathy, retinopathy, Alzheimer’s disease, chronic kidney disease, and fatty liver disease.
How does this method reverse insulin resistance?
Our method reverses insulin resistance by helping you decrease the amount of excess fat that has accumulated in your muscle and liver cells. Those cells can then return to storing an appropriate amount of fat.
This is accomplished by:
Low-fat, plant-based, whole-food nutrition: We recommend eating foods that are naturally low in fat, high in fiber, water, and nutrient density to improve your insulin sensitivity dramatically. These foods include fruits, starchy vegetables, legumes, intact whole grains, non-starchy vegetables, leafy greens, herbs, spices, and mushrooms.
Intermittent Fasting: Changing the timing of your food intake is one of the most powerful ways to improve your insulin sensitivity, improve your cardiovascular health, and lose weight. Mastering Diabetes teaches you how to design and incorporate a sustainable intermittent fasting regimen to transform your metabolic health.
Daily Movement: Your body is designed for physical activity, and when you make daily movement a part of your lifestyle, you’re likely to dramatically improve your insulin sensitivity, energy levels, and mood.
How do you handle blood glucose spikes on a high-carbohydrate diet?
For those living with insulin-dependent diabetes, there are a few keys to preventing post-meal blood glucose spikes:
Do your best to make sure that your fat intake does not exceed 30 grams per day (or a maximum of 15% of total calories)
Eat whole carbohydrate-rich food that is full of water, fiber, vitamins, minerals, antioxidants, and phytochemicals
Transition to a plant-based diet slowly over the course of 1-2 months, and focus on changing only one meal at a time. This enables you to make small changes to your insulin and/or medication strategy and helps you understand how each meal affects your blood glucose control.
Pay attention to insulin timing and make sure your blood glucose is 120 mg/dL and trending down before eating a carbohydrate-rich meal (using a fingerstick blood glucose measurement).
For those living with non-insulin dependent diabetes, follow the Mastering Diabetes Method exactly the way we describe in the book by measuring your level of baseline insulin resistance.
For those with a high level of baseline insulin resistance, we recommend taking a 2-step approach by eating low-glycemic foods for the first few weeks followed by higher-glycemic foods over the course of time. This enables you to incorporate more plants into your diet while minimizing the risk of hyperglycemia.
Can you follow the Mastering Diabetes Method successfully if you’re not vegan?
Yes, absolutely. We have a traffic light food system containing green light, yellow light, and red light foods in which we encourage you to eat a diet containing as many plant foods as possible.
If you choose not to eat a 100% plant-based diet, you can still eat small amounts of animal products found in the red light category and begin reversing insulin resistance predictably.
The truth is that we are not the food police. We simply encourage you to eat a diet containing as many plants as possible to maximize your insulin sensitivity, however, your desire to minimize or eliminate animal products is entirely up to personal choice.
Are there any downsides or things one should be aware of?
The biggest downside is that you’ll have to be willing to significantly reduce your consumption of delicious high-fat foods such as avocados, nuts, seeds, and coconut meat in order to truly maximize your insulin sensitivity
Be aware of how much of those foods you are consuming in order to make sure your transition is as smooth as possible. A small amount is beneficial, but it’s very easy to overeat fatty foods, which has an immediate negative impact on blood glucose control.
Cyrus Khambatta, Ph.D., and Robby Barbaro, MPH are the cofounders of Mastering Diabetes, a platform that teaches people how to reverse insulin resistance via low-fat, plant-based, whole-food nutrition.
Cyrus has been living with type 1 diabetes since 2002. He has an undergraduate degree from Stanford University and a Ph.D. in Nutritional Biochemistry from UC Berkeley.
Robby was diagnosed with type 1 diabetes in 2000. He has an undergraduate degree from the University of Florida and an MPH from American Public University. He spent six years helping build the revolutionary Forks Over Knives empire, before turning his attention in 2016 to coaching people with diabetes full time.
Success! Now check your email to download the eBook chapter.
Wow Your Guests With Best Beef Tenderloin Recipe – Plan Ahead
When planning a dinner party, Holly’s Dijon Pepper Beef Tenderloin recipe always tops the menu. This recipe makes you appear like a chef in the kitchen and truly is effortless. This scrumptious and easy whole beef tenderloin recipe from Guy’s Guide To Eating Wellmakes such an impression. Here’s an example of Holly’s Dinner Party Recipesand start a few days prior to the party preparing my make-ahead recipes. I don’t want to leave all the work for the day of the party. Marinate meat for 48 hours and then cook the beef tenderloin in oven. Simple and a guaranteed hit!
Whole Beef Tenderloin Recipe Always On Menu For Entree
Purchase the whole tenderloin and just trim the fat off yourself or have the butcher do only a little trimming. If you go to Costco or Sam’s you get a better price on these tenderloins. They are very pricey but worth every penny! Have fun with entertaining and use Holly’s trim & terrific recipes no matter how many people you have coming.
Step By Step to Impressive Yet Easy Beef Tenderloin Recipe
First start with marinating the tenderloin. After you marinate the beef tenderloin for up to 3 days, let it come to room temperature and then cover it heavily with the Dijon and Worcestershire Sauce. Then, LOTS of pepper! Always cook the beef tenderloin on a foil lined pan for easy clean up. What’s great is you cook this beef tenderloin recipe in the oven. OMG, if beef could melt-in-your-mouth, it does when you take that first bite!
Pepper Dijon Beef Tenderloin
For dinner parties or gatherings, this most amazing meat sets the mood for a ‘real deal-maker” dinner. A no fail winning recipe and definitively on my short list
Prep Time5minutes + marinating time
5-6pound whole beef tenderlointrimmed of excess fat
salt and pepper to taste
1/2cup fat free Italian dressing
1/2cup worcestershire sauce
1/2cup dijon mustard
Coarsely cracked black pepper
Lay tenderloin in glass dish and season to taste
Cover and pat tenderloin with Italian dressing and Worcestershire sauce. Cover with plastic wrap and refrigerate 48 hours time permitting. Let meat come to room temperature before cooking (at least one hour).
Preheat oven 500°F. Pour off marinade, cover meat with Dijon mustard and heavily with cracked black pepper. Transfer to baking dish. Cook 500°F 12 minutes and reduce temperature to 275°F and cook another 25 – 30 minutes depending on doneness.
Per Serving: Calories 148, Calories from Fat 39%, Fat 6g, Saturated Fat 2g, Cholesterol 59 mg, Sodium 242 mg, Carbohydrates 1g, Dietary Fiber 0g, Protein 21g, Total Sugars 0g,Dietary Exchanges: 3 lean meat
Dijon Pepper Beef Tenderloin recipe is from Holly’s men’s cookbook, Guy’s Guide To Eating Well. Guy’s Guide contains recipes that men love to cook and eat. From the Food for the Mood Chapter, this recipe is rich in protein. Also, you’ll learn tips like “Remember to start marinating 48 hours before. Be sure to let meat come to room temperature before cooking.”
There’s chapters like Grilling & Hunting and Fix It Fast or Fix It Slow plus preventive health chapters including heart disease, joint pain, cancer and more.
Whole Beef Tenderloin In Oven Cooks Perfectly Every Time
The cooking directions might seem a little different but this is the best of all the whole beef tenderloin recipes. Sometimes, if the tenderloin is really big, you might cook it a little longer. With the Dijon pepper seasoning and marinated in Italian dressing, you won’t believe how something so simple can taste amazing!
Beef Is Good For You!
This whole tenderloin recipe has been a favorite for years. When Holly was writing Guy’s Guide To Eating Well, she decided to include her favorite beef tenderloin recipe. So, she put the recipe in the Food For The Mood Chapter as it’s the perfect recipe for a romantic dinner or any time you entertain. Red meat is a good source of zinc. So, give your libido a boost since zinc raises testosterone. Did you know that?
Find more easy and impressive dinner party recipes for easy entertaining on Holly’s healthy cooking blog.
This study aims to model genetic, immunologic, metabolomics, and proteomic biomarkers for development of islet autoimmunity (IA) and progression to type 1 diabetes in a prospective high-risk cohort. We studied 67 children: 42 who developed IA (20 of 42 progressed to diabetes) and 25 control subjects matched for sex and age. Biomarkers were assessed at four time points: earliest available sample, just prior to IA, just after IA, and just prior to diabetes onset. Predictors of IA and progression to diabetes were identified across disparate sources using an integrative machine learning algorithm and optimization-based feature selection. Our integrative approach was predictive of IA (area under the receiver operating characteristic curve [AUC] 0.91) and progression to diabetes (AUC 0.92) based on standard cross-validation (CV). Among the strongest predictors of IA were change in serum ascorbate, 3-methyl-oxobutyrate, and the PTPN22 (rs2476601) polymorphism. Serum glucose, ADP fibrinogen, and mannose were among the strongest predictors of progression to diabetes. This proof-of-principle analysis is the first study to integrate large, diverse biomarker data sets into a limited number of features, highlighting differences in pathways leading to IA from those predicting progression to diabetes. Integrated models, if validated in independent populations, could provide novel clues concerning the pathways leading to IA and type 1 diabetes.
Type 1 diabetes results from autoimmune destruction of insulin-producing pancreatic β-cells. Clinically apparent diabetes is typically preceded by a period of islet autoimmunity (IA), marked by appearance of autoantibodies against islet autoantigens (1). While there is consensus that chronic autoimmune destruction of β-cells is triggered by an interaction of environmental factor(s) with a relatively common genetic background, the specific cause remains elusive. Prospective cohort studies have reported a number of demographic, immune (2–4), genetic (5–10), metabolomic (11,12), and proteomic (13–15) predictors of IA and/or progression from IA to diabetes. Each analytic approach offers unique insights; however, single data stream analysis is unable to address the importance of technique-specific observations in the context of other analyses. Use of data fusion methods to integrate different data types can create models that are more complete and accurate than those derived from any individual source (16). Our objective was to provide proof of principle that machine learning Bayesian modeling of disparate biomarkers can yield useful integrated models for hypothesis generation. Applied to longitudinally collected biomarkers, such integrated models could provide novel clues concerning pathways leading to IA and/or diabetes. This integrative modeling approach could improve personalized prediction of progression through presymptomatic stages of type 1 diabetes.
Research Design and Methods
We performed a nested case-control study of children participating in the Diabetes Autoimmunity Study in the Young (DAISY) cohort. DAISY follows prospectively 2,547 children at increased risk for type 1 diabetes. The cohort consists of first-degree relatives of patients with type 1 diabetes (FDRs) and general population children with type 1 diabetes–susceptibility HLA DR-DQ genotypes identified by newborn screening (17,18), recruited between 1993 and 2004. Follow-up results are available through 29 September 2017. Written informed consent was obtained from subjects and parents. The Colorado Multiple Institutional Review Board approved all protocols.
Autoantibodies were tested at 9, 15, and 24 months and, if negative, annually thereafter; autoantibody-positive children were retested every 3–6 months. Radio-immunoassays for insulin (IAA), GAD (GADA), insulinoma-associated protein 2 (IA-2A), and/or zinc transporter 8 (ZnT8A) autoantibodies were conducted as previously described (19–23). Subjects were considered persistently islet autoantibody positive if they had two or more consecutive confirmed positive samples, not due to maternal islet autoantibody transfer, or had one confirmed positive sample and developed diabetes prior to next sample collection. Diabetes was diagnosed using American Diabetes Association criteria.
Selection of Subjects for Analyses
Sixty-seven children were selected in July of 2011 from the DAISY cohort for studies of metabolomic, proteomic, and immune predictors. Of those, 22 children developed diabetes (T1D group), 20 had developed persistent IA and were islet autoantibody positive at their last study visit (AbPos group), and 25 were control subjects (control group). Control subjects were frequency matched with subjects in the combined T1D and AbPos group on the HLA DR-DQ genotypes, age, sex, and FDR status. As of 29 September 2017, all control subjects have been negative for all islet autoantibodies. Of the AbPos group, four progressed to diabetes in subsequent years at median age 17.8 years. These individuals were retained in the AbPos group. Supplementary Fig. 1 describes subject selection. Supplementary Table 1 presents individual autoantibody histories of all case and control subjects at relevant time points.
Specimens for Analysis
When available, samples for each subject were analyzed for metabolomic, proteomic, and immune biomarkers at four time points: T1, earliest available sample prior to development of islet autoantibodies (typically age 9–15 months); T2, just prior to development of first autoantibody; T3, just after development of first autoantibody; and T4, just prior to diagnosis of diabetes or most recent sample for AbPos subjects at time of sample selection.
Of the subject with T1D, five were missing a T2 sample. Samples from control subjects were selected to frequency match storage time of samples from T1D and AbPos subjects combined.
Global metabolic profiling combined two separate ultrahigh-performance liquid chromatography/tandem mass spectrometry (UHPLC/MS/MS2) injections, optimized for basic and acidic species, and gas chromatography/mass spectrometry (GS/MS) (Metabolon, Durham, NC). All serum samples were stored at −80°C ≤1 h after collection, never thawed until analyses, and processed essentially as described previously (24,25). Metabolites were identified by automated comparison of ion features in experimental samples with a reference library of chemical standard entries using software developed at Metabolon (26). A total of 382 named metabolites were included in this analysis. For statistical analyses and data display, any missing values were assumed to be below limits of detection, and these values were imputed with the compound minimum (minimum value imputation).
Relative abundance of 1,001 serum proteins were measured by aptamers (Somalogic, Boulder, CO) (27). Additionally, 49 peptides (representing 24 proteins) were measured by LC-MRM/MS in the laboratory of Drs. Thomas Metz and Qibin Zhang at Pacific Northwest National Laboratory as previously described (28).
Cytokines were measured using a Human Custom Cytokine 9-Plex assay (Meso Scale Discovery, Rockville, MD) and included interferon (IFN)-α2a, interleukin (IL)-6, IL-17, IL-1β, interferon γ–induced protein (IP)-10, monocyte chemotactic protein (MCP)-1, IFN-γ, IL-1α, and IL-1ra (4).
All 106 non-HLA SNPs available for these subjects were included. Locus and reference for genotyping are shown in Supplementary Table 2. As genetic data were derived from several analyses, some genes were represented by more than one SNP and the rs2476601 SNP for PTPN22 was present in two genetic feature sets from separate analyses (Supplementary Table 2).
Individual metadata included in the model consisted of age at sampling, sex, Hispanic ethnicity, and FDR status (classified as mother with type 1 diabetes, other FDR with type 1 diabetes, or no FDR) Additionally, subjects were classified by four HLA risk categories based on typing for HLA class II alleles as previously described (29).
Statistics and Machine Learning
SAS, version 9.4 (SAS Institute, Cary, NC), was used to analyze descriptive data for groups. Integrative machine learning, based on the set of demographic characteristics, gene variants, cytokines, proteins, and metabolites, was used to evaluate whether predictive models can separate future cases from control subjects, as well as identify the primary features that distinguish the groups. Two disease stages were modeled. 1) Development of IA: transition from T1 to T2 among combined AbPos and T1D subjects versus control subjects. 2) Progression to diabetes: transition from T3 to T4 among T1D versus AbPos subjects. Transition between time points is represented in analysis by determining log fold difference of cytokine, protein, or metabolite between time points. Integrative machine learning was performed using probability-based integration of multiple data streams via the Posterior Probability Product (P3) (16,30). Feature selection was performed in the context of the integrated model using a Repeated Optimization for Feature Interpretation (ROFI) approach (31). The ROFI-P3 algorithm for this analysis is described in detail in Supplementary Fig. 2. It has several key characteristics amenable to identifying and evaluating important features of diabetes progression across disparate data sources, which include allowing each data set to be modeled with the optimal machine learning algorithm and features to be assigned importance metrics through repeated analyses.
The first step requires selection of the machine learning method to model each data set (e.g., metabolomic, proteomic, etc.). We evaluated seven machine learning classification methods with all features in each data set where a feature is the ratio of the measurements between time points for a case or control subject, including random forest (RF), logistic regression (LR), k-nearest neighbors (KNN), linear discriminant analysis (LDA), support vector machine (SVM) with a radial basis function (RBF) kernel, SVM with a linear (LIN) kernel, and naive Bayes (NB). Supplementary Fig. 3 shows average area under the receiver operating characteristic (ROC) curve (AUC) based on fivefold CV repeated 100 times. The only requirement of this step is that a machine learning classifier can output the posterior probability, defined as the probability that class i is observed given the data for subject s of data set j; P(ci|Dsj). The posterior probabilities are generated using the standard functions in the R programming language for each machine learning algorithm.
Integration via the P3 approach is a naïve product-based integration as the product of the posterior probability of each sample as related to each data sets:
(16,30). These integrated probabilities can be used to compute the accuracy of the integrated model using a standard AUC. Feature selection is performed on the integrated model to assure that features selected are those that work best in combination across disparate sources. Selection utilizes a statistical optimization algorithm, such as simulated annealing, which is not affected by the order of features in the data set and allows the algorithm to move out of local minima by updating the solution at each iteration based on the current feature state and sampling in proportion to including or excluding the variable of interest. Thus, for each feature change proposal, this is based on looking at the difference of the accuracy of the current state (AUCCurrent) and an updated solution (AUCUpdated). The updated solution is selected in proportion to:based on a uniform distribution between 0 and 1, where Δ = 0.25 for this analysis. For each run of the algorithm, we perform 100 random changes of individual features and keep or discard the change based on this exponential difference between AUC values. After each 100 proposals and potential updates, we determine whether the solution has converged based on the difference between the AUC prior to the 100 feature evaluations and the current solution. If this value is <δ, which was set to 1E-4 for this analysis, it is determined that the solution has converged (31).
Within ROFI-P3, the AUC is computed based on fivefold CV for every feature evaluation. We repeat the algorithm in conjunction with CV for 100 repetitions, each of which yields a single feature set solution. We use the 100 repetitions to obtain a feature ensemble solution, which gives the likelihood that the feature would be selected for inclusion in the model. This is represented as the percentage of times that a specific feature was selected to be in the model. This also has the additional benefit of yielding robust measures of uncertainty on our classification accuracy metrics.
To evaluate the performance of the ROFI-P3 method versus established optimization methods, we compared ROFI-P3 with standard recursive feature elimination (RFE) approaches. RFE is a method used extensively in biology in combination with various machine learning algorithms, such as LDA and SVM (16,30,32,33). RFE is readily available in most statistical programming languages and is simple to implement. It is a greedy algorithm that sequentially eliminates the feature that yields the maximum AUC.
Data and Resource Availability
The data sets generated during and/or analyzed during the current study are available from the corresponding author upon reasonable request.
Characteristics of the study subjects are presented in Table 1. Ages at visits ranged from 6 months to almost 21 years old with similar ages at time points T1, T2, T3, and T4 (Supplementary Table 3). Demographic (metadata), genetic, immune, metabolic and proteomic biomarkers were analyzed across these four distinct time points.
Characteristics of the study participants
First, we determined the optimal machine learning algorithm for each of the data types (Supplementary Fig. 3). Next, we analyzed the ability of ROFI-P3 to predict the changes leading up to two stages of type 1 diabetes: 1) seroconversion and 2) progression to diabetes. ROFI-P3 was performed, and for each repetition features were selected. Each feature is represented as the percentage of times it is selected as part of the model during 100 repetitions. For comparison, we also ran RFE for multiple repetitions, each time permuting the features, since the order of the features has a direct relationship with those selected. This allowed us also to represent RFE features as the percentage of times they were selected. To evaluate how well the methods work, a feature selection threshold is selected and an ROC curve was generated only on this reduced model using fivefold CV to build and test the model independently and to minimize overfitting. Fig. 1A and B show the results of these comparisons at a 50% frequency selection, i.e., features selected at least 50% of the time for both ROFI-P3 and RFE, as well as if no feature selection is performed. Various thresholds were evaluated, and the ROFI-P3 integrated feature selection approach was consistently more accurate than RFE. This demonstrated a clear advantage over both the RFE selection and simple combination of all features for prediction of development of IA (AUC 0.91 vs. 0.84 and 0.64, respectively, P < 0.0001) and progression (AUC 0.92 vs. 0.82 and 0.64, respectively, P < 0.0001) at a 50% feature selection threshold. We further evaluated the method in the context of the classification of specific individuals rather than a global metric of classification. If we select a defined reasonable false positive rate of 10% for the development of IA (Fig. 1A), we would correctly classify 67.6% of those who developed IA with ROFI-P3. The percentages drop to 56.5% and 31.5% for the RFE selection and simple combination approaches, respectively. The prediction is slightly better for the progression end point; ROFI-P3 correctly identifies 71.6% compared with 61.4% for RFE and 18.3% for simple combination (Fig. 1B). The top features selected by ROFI-P3 at a 50% frequency or higher for analysis of development of IA (Supplementary Fig. 4A) and progression to diabetes (Supplementary Fig. 4B) included features from five of the six data sets.
ROC curves. A: Comparing development of IA in control group vs. combined AbPos and T1D groups at transition from earliest time point (T1) to preseroconversion (T2). B: Comparing progression to T1D—transition from postseroconversion (T3) to before T1D diagnosis (T4)—in AbPos vs. T1D groups. Dotted line, prediction based on all features; gray dashed line (RFE), prediction based on features selected 50% of the time or more using recursive feature elimination; solid black line (ROFI), prediction based on features selected at least 50% of the time using ROFI-P3 algorithm; black dashed line (in panel B only), prediction based on glucose change from T3 to T4.
The top features selected by ROFI-P3 as predictors of IA are shown in Table 2. Percent selected is a measure of the number of times a particular feature was selected in the 100 iterations of the algorithm, an indication of the importance of that feature in the prediction of outcome. Supplementary Table 4 shows all 76 features selected at 50% frequency or higher. Further detail regarding these features is shown in Supplementary Tables 5–8. Box plots of the log fold change in abundance in these top metabolites, proteins, and peptides from T1 to T2 for IA and control subjects are shown in Fig. 2.
The top 16 predictors for development of IA
The top 10 protein, peptide, and metabolite predictors for development of IA. For each analyte, the box plots show log fold change from time 1 (T1) to time 2 (T2) for case and control subjects with individual values noted by circles. The value of log2(T2 − T1) is positive with increasing trajectory and negative with decreasing trajectory.
The top features selected by ROFI-P3 as predictors of progression from IA to diabetes are shown in Table 3, while all 83 features selected at 50% frequency or higher are shown in Supplementary Table 9. Characteristics of features including genotypes or direction of change in abundance from T3 to T4 are further described in Supplementary Tables 10–13. Figure 3 shows the box plots for control and AbPos groups of the log fold change of each feature from T3 to T4.
The top 16 predictors of progression from IA to diabetes
The top 12 protein, peptide, and metabolite predictors for progression to diabetes. For each analyte, the box plots show log fold change from time 3 (T3) to time 4 (T4) for case and control subjects with individual values noted by circles. The value of log2(T4 − T3) is positive with increasing trajectory and negative with decreasing trajectory.
Of note, the metabolite glucose is the top selected feature, as could be expected during the progression to T1D. To evaluate the value of adding additional features to glucose, we ran a logistic regression using glucose alone as the predictor for progression to T1D compared with that of the selected ensemble. Change in glucose alone was able to classify a majority of case subjects, but the AUC for the ROFI-P3 (0.91) was significantly higher than that for glucose alone (0.83, P < 0.00001) (Fig. 1B). As expected for the end point of development of IA, the AUC of glucose is not predictive: 0.48.
Overfitting is often an issue with machine learning, especially when sample sizes are only large enough to allow CV. To evaluate whether the top features could separate the groups with an unsupervised approach, principal component analysis (PCA) was utilized on only the top qualitative omics features in Tables 1 and 2. From the PCA plot, the first component can visually separate the two groups for both predictors of IA (Fig. 4A) and progression from IA to diabetes (Fig. 4B) without prior knowledge of the groups. This demonstrates that although there may be some overfitting, the methodology in general is identifying features that can discriminate the groups of interest.
PCA of (A) predictors of IA based on top 4 metabolites and 6 proteins and (B) progression from IA to diabetes on top 5 metabolites and 7 proteins. Open circles represent control subjects, and closed circles represent combined AbPos and T1D groups. PC1, principal component 1; PC2, principal component 2.
Identification of causative factors in the development of IA and type 1 diabetes has been elusive. Recent observations regarding the role of vitamin D in risk of IA has underlined the importance of understanding environmental exposures in the context of genetic background (34). Thus, analysis that integrates multiple data streams has the potential to identify unique ensembles of pathogenic features. This proof-of-concept analysis represents the first integration of disparate omics data sets for the prediction of IA and type 1 diabetes. The ROFI-P3 approach solves the feature selection process through hundreds of iterations, resulting in a probability measure for each individual feature. This allows the reduction of large data sets to a smaller, more informative set of features as well as a robust measure of feature-level uncertainty. The biomarker panels identified in this analysis represent an individualized prediction algorithm based on a set of disparate features (e.g., metabolites, proteins in combination with genetics, and standard risk factors) selected in at least 50% of the iterations. These models predicted development of IA and progression to diabetes with an AUC of 0.91 and 0.92, respectively. It should be noted that as the analysis incorporated change in protein, metabolite, or cytokine over time, selected features represent features whose change, not absolute value, is associated with outcome.
To predict development of IA, several metadata features were included, which serves to adjust the analysis for these factors. Among the most highly selected features were all five metadata elements: age, FDR status, Hispanic ethnicity, HLA risk group, and sex, indicating that these categories were important in conjunction with other features in the prediction of IA.
Two highly selected features were genetic markers associated with development of IA: PTPN22 (rs2476601) and CTLA4 (rs3087243 and rs231775) (5,35). Both PTPN22 (rs2476601) and CTLA4 (rs3087243 and rs231775) were selected twice in this analysis, as they were included in the feature set comprising data from multiple separate genetic analyses. The observation that the same SNPs were selected twice provides additional evidence of robustness of this analytical approach.
Many of the most frequently selected features were metabolites. The highest selected feature was ascorbate (vitamin C), an important antioxidant. Ascorbate was present at lower relative abundance in participants who developed IA at the earliest time point (T1) relative to control subjects and rose over time (Fig. 2), while control subjects started with a higher level and then showed a downward trend in ascorbate levels. Discrepant trajectories between these two groups were significantly associated with IA outcome (Supplementary Table 6). Other metabolites whose change over time predicted outcome included 3-methyl-oxobutyrate (α-ketoisovaleric acid), a degradation product from valine as well as a precursor to valine for leucine synthesis. These branched-chain amino acids are known to predict development of insulin resistance. They play an intriguing role in promoting lymphocyte growth and proliferation as well as cytotoxic T- lymphocyte activity and have been previously identified as elevated prior to seroconversion (36). 4-hydroxyhippuric acid is a microbial end product produced through polyphenol metabolism by intestinal microflora (37), and serum levels are affected by altered gut permeability in mice (38). Pyroglutamic acid is a derivative of l-glutamic acid, formed nonenzymatically from glutamate, glutamine, and γ-glutamylated peptides. Elevated blood levels of pyroglutamine may indicate problems in antioxidant glutathione metabolism (39). This metabolite increased during progression to IA in case subjects but decreased in control subjects.
Finally, among the most frequent features selected for development of IA were multiple proteins involved in immunity and inflammation: FCRL3 (Fc receptor-like protein 3), KLRK1, MMP-2, and activin A. Also selected were SSRP1, a protein involved in DNA repair, and CSK21, which plays a role in apoptosis and response to viral infections.
Of the five elements of metadata included in analysis for progression from IA to diabetes, only age and FDR status were among the highest selected features, indicating that these features were important in conjunction with the constellation of other highly selected features.
Interestingly, top-selected SNPs associated with development of IA were different from those associated with progression to diabetes. Of these, rs2157678 (HLA DQB1 8.1) is associated with the ancestral HLA DR3-B8-A1 haplotype (40).
In analysis of progression from IA to diabetes, selected metabolomic features included multiple carbohydrates: glucose, mannose, and ribose (Table 3 and Fig. 3). All three metabolites increased in abundance in children progressing to diabetes but decreased from T3 to T4 in the control group (Supplementary Table 11). Additionally, butyrylcarnitine, an acylcarnitine, was noted to increase from T3 to T4 in case subjects but decrease in control subjects. This could be explained by an overall increase in lipolysis secondary to progressive insulinopenia as one approaches clinical diabetes.
Among the top proteomic features in progression from IA to diabetes was cystatin-F, a protein that modulates natural killer and T cell cytotoxity and RAD51, which plays a role in DNA repair. Plasminogen, a protease important for lysis blood clots, also plays a role in activating the complement system. Proteins involved in cell adhesion and growth (DRR1, IL-11 RA, and spondin-1) were also among the most frequently selected features.
In summary, we demonstrated that the ROFI-P3 algorithm can identify and evaluate known and novel predictors of development of IA and progression to diabetes across disparate data sources. Importantly, in children with high-risk HLA genotypes, changes in relative abundance of certain proteins and metabolites as well as genetic markers predicted development of IA, and a distinct constellation of features predicted progression of persistent IA to diabetes. Seroconversion was associated with altered antioxidant profile, a finding that has been noted in humans (36) and NOD mice (41). Additionally, there are indications of altered gut permeability, another proposed pathogenic mechanism (42). In contrast, progression from IA to diabetes was associated with altered sugars and acyl carnitines, indicating a potential switch to alternate metabolic pathways as relative insulin deficiency becomes more prominent.
The goal of this study was to develop a robust statistical machine learning model that predicts development of IA and progression from IA to diabetes. The major advantage of this study is the prospective characterization of developing autoimmunity over a prolonged period of time, with repeat longitudinal measurements of biomarkers. The DAISY cohort has >20 years of follow-up. While a peak incidence of IA has been observed within the first 2 years of life (3,43), new seroconversion has been observed well into adolescence and beyond (44). In addition, data from multiple sources (clinical data, genetics, metabolomics, and proteomics) are available to be integrated in a machine learning framework. Limitations of this study include the relatively small numbers of subjects in each group. Larger cohort size could allow additional analysis, including examination of whether age at seroconversion or specific endotypes play a role in the features selected. A further limitation of the study is the potential bias to individuals with later seroconversion. Both IA and T1D groups (Table 1) were older than the reported median seroconversion age of 2.3 years in The Environmental Determinants of Diabetes in the Young (TEDDY) study at 7 years of follow-up (45). In contrast, studies such as DAISY (46) and BABYDIAB (43), with longer follow-up beyond the early peak in autoimmunity, observe ongoing seroconversion into later childhood. Thus, selection of participants included individuals with seroconversion at older ages. Further, attention to requisite sample availability may have biased against individuals with early seroconversion who often have exceedingly rapid progression. This may impact generalizability to such rapidly progressing individuals.
Building predictive models via machine learning is an emerging strategy for identification of predictive biomarkers in type 1 diabetes and other diseases; however, challenges remain in the integration of large and diverse data sets. Machine learning strategies that incorporate feature selection allow identification of biomarkers that perform well in combination. This not only selects the most predictive features from among many but also may lend insight into important biological mechanisms. Although P3 and ROFI have both been used previously to study omics data, this is the first combination for feature selection in an integrative fashion. The feature sets identified using the ROFI-P3 strategy perform well in prediction of both IA and type 1 diabetes outcome. Further, identification of distinct panels of predictors underlines differences between processes leading to development of IA from pathways involved in progression to diabetes. The associated measure of probability adds further information for interpreting the utility of various biomarkers and could help researchers in identifying the best candidates to focus limited resources on validation. Further studies will determine whether these selected features can be validated in independent populations to predict progression to IA or type 1 diabetes.
Funding. This work was supported by JDRF (grants 17-2013-535, 11-2010-206, and 5-ECR-2017-388-A-N) and the National Institutes of Health (grants R01 DK32493, DK32083, DK049654, K12 DK094712, and P30 DK57516).
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
Author Contributions. B.I.F., B.-J.W.-R., and M.R. contributed to the design of the study, analyzed data, contributed to the discussion, and wrote the manuscript. L.M.B. and S.M.R. analyzed data, contributed to the discussion, and reviewed and edited the manuscript. K.W., A.K.S., and J.M.N. researched data, contributed to discussion, and reviewed and edited the manuscript. M.R. is principal investigator of DAISY. All authors approved the final version of the manuscript. M.R. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Prior Presentation. Parts of this study were presented in abstract form at the 77th Scientific Sessions of the American Diabetes Association, San Diego, CA, 9–13 June 2017.
When I found a jar of sun-dried tomatoes in the pantry and some goat cheese in the refrigerator, I came up with Mediterranean Zucchini Noodles with Goat Cheese. I had read a similar recipe elsewhere that inspired me, but couldn’t find it when I wanted it (of course). I’m sure now it will appear.
The sun-dried tomatoes really make this dish by adding sweetness. The goat cheese lends creaminess and a bit of tang. The noodles make a great dinner side or a light lunch that travels well.
No spiralizer? Many grocery stores sell zucchini already cut into noodles. If yours doesn’t, check out Taste of Home’s instructions for making them with a vegetable peeler, knife, or grater.
Other Zucchini Noodle Recipes
Like the whole zoodle idea? Try these other ideas:
Mediterranean Zucchini Noodles with Goat Cheese
Zucchini noodles (“zoodles”) lightly sautéed and topped with sun-dried tomatoes, walnuts, spinach, and goat cheese
Author:Shelby Kinnaird (Diabetic Foodie)
Course: Side Dishes
Keyword: zoodles, zucchini noodles
1/4 cup walnuts chopped
4 ounces sun-dried tomatoes (packed in olive oil) thinly sliced
3 tablespoons olive oil from the jar of sun-dried tomatoes
1 pound zucchini noodles
1/2 teaspoon Herbes de Provence
2 garlic cloves chopped
freshly ground black pepper
1/4 cup white wine or water
3 ounces baby spinach
4 ounces goat cheese or feta cheese crumbled
Place the walnuts in a small nonstick skillet over medium heat. Cook, tossing constantly, until the walnuts have browned slightly and are fragrant. Remove from heat and let cool.
Heat 1-1/2 tablespoons of olive oil (from the jar of sun-dried tomatoes) in a large skillet over medium heat. Add the zucchini noodles, Herbes de Provence, garlic, salt, and pepper. Cook, stirring occasionally, until the noodles have begun to soften, about 5 minutes. Transfer the noodles to a serving bowl.
Add remaining 1-1/2 tablespoons of olive oil to the pan along with the wine, sun-dried tomatoes, and spinach. Cook, stirring occasionally, until the spinach has wilted. Top the zucchini noodles with the spinach mixture.
Garnish with walnuts and cheese.
If you can’t find pre-cut zucchini noodles, use a spiralizer or vegetable peeler and make your own.
You can also use dry sun-dried tomatoes. Cut them into strips with scissors, then submerge them in boiling water to soften. Use regular olive oil to replace the oil that would have been in the jar.
I like to think that I am a pretty good cook, and have definitely gotten better over the years. I am a visual learner so if I am new at something, I can watch a few YouTube videos and get the gist of the recipe.
Spring rolls, however, have never worked for me. The wrappers tore, or if I got it to lay flat when I went to roll it up it would stick to the board. So frustrating.
Since Hannah has reorganized my pantry, I found a brand new package of spring rolls, as if taunting me to try another stab at it.
Here are a couple things I realized: you don’t need to heat up water – just regular warm tap water will work. Also it helps if your cutting board is a bit wet – my wrappers didn’t stick!
The star of these shrimp rolls, was well – the shrimp! Mariano’s makes shrimp cocktail to take – I got over a half pound for $5 – you can’t beat that price.
Now these are probably not authentic shrimp spring rolls, because I didn’t use fresh herbs, and I used what I had on hand. A lot of spring rolls have cooked vermicelli pasta in the middle, I decided to make brown rice. But after making it, it was too hot so I put it outside to cool and forgot about it. And as I am typing this, realized that I haven’t brought that back in yet. #klassy
Another tip about the wrappers, dip in the water and pull out just as they start to wilt – they will continue to soften after you take them out of the water.
While my wrapping skills will improve, these are the best spring rolls I’ve ever made. While the filling my seem simplistic, which it is, the dipping sauce is what makes these spring rolls. Don’t make these without making the dipping sauce.
These were so good. I thought I may not be full after eating these, but they kept me full for hours – winning! Side note: don’t try to fry these – the wrapper basically disintegrated. 😛
While anyone — with or without diabetes — can experience bouts of diarrhea, it can be the direct result of diabetes for some.
Also referred to as “diabetic diarrhea,” this is one of the lesser-known complications of diabetes, but it can be incredibly difficult to manage and live with on a daily basis.
It’s estimated that about 22 percent of people who have lived with diabetes for a while experience frequent diarrhea.
In this article, we’ll look at the causes, symptoms, and treatment options for diabetes-related diarrhea.
How does diabetes cause diarrhea?
Diabetic diarrhea is likely the result of longterm high blood sugar levels damaging the nerves and overall function within your colon, small intestines, and your stomach (a condition known as neuropathy).
“Many gastrointestinal complications of diabetes seem to be related to dysfunction of the neurons supplying the enteric nervous system,” explains the 2000 edition of Clinical Diabetes.
When the nerve fibers throughout your digestive system are damaged, this can result in constipation or diarrhea (often both) at different times because some nerves stimulate movement and other nerves help to slow movement in your intestines.
The bouts of diarrhea can be ongoing or they can persist for a few months and then quiet down for a few months. The phases of diarrhea can also be countered with phases of constipation, too.
Another possible cause of diabetic diarrhea is that people with diabetes are more likely to consume large quantities of artificial sweeteners and sugar alcohols, which are known to be potent laxatives.
Diagnosing diabetic diarrhea
There is no simple way to test and determine if the cost of your symptoms are related to nerve damage or something else.
If you suspect your diarrhea is directly related to your diabetes, schedule an appointment with your healthcare team, and keep careful notes for a week or two on the frequency and severity of your bowel movements.
Depending on their assessment, your healthcare team may refer you to a gastroenterologist for further investigation.
First of all, working with your healthcare team to improve your blood sugars is a crucial part of treating diabetic diarrhea.
Without addressing your blood sugars, you’ll continue to experience damage to the nerves throughout your digestive system (and entire body), and continue to experience uncomfortable symptoms.
Secondly, your doctor may advise you to change your eating habits. You may need to eat more fiber (or less fiber), drink more water, or include more (or less) vegetables in your diet.
Finally, there are a few specific medications being used to help patients counter the effects of this nerve damage and decrease or eliminate bouts of diarrhea:
Parenteral somatostatin analog octreotide: For patients with longstanding diabetes and digestive complications, this option has proven useful. It is an injected medication that’s used primarily to treat different types of diarrhea.
Basic over-the-counter treatments: While there are a variety of over-the-counter options to treat general diarrhea, they aren’t as likely to work in a patient with diabetes struggling with severe nerve damage to their colon and overall digestive system.
Other possible causes of diabetic diarrhea
Aside from neuropathy within your colon, there are a few other things to consider when evaluating the potential cause of your digestive woes.
Metformin is one of the most prescribed drugs across the globe, and the first line of treatment for people with type 2 diabetes — but it can definitely upset your digestive system.
Actually, needing to “run to the bathroom” frequently while taking metformin is one of the reasons many patients stop taking it altogether.
The number one thing you can do to manage diarrhea resulting from metformin is to talk to your doctor about getting insurance approval for the more expensive “extended-release” version.
This is never the first version prescribed because it costs more, which means your health insurance will want you to first try the cheaper version.
If you report adverse reactions and your doctor documents this, you have a good chance of getting insurance approval for the extended-release version of metformin.
A diet high in sugar alcohols in “low-carb” products
If you’re eating a lot of low-carb protein bars, candies, ice creams and other processed products that use sugar alcohols as a sweetener, don’t be surprised if you’re running to the bathroom regularly.
Sugar alcohols are a known laxative. While most people can consume some sugar alcohol without much digestive trouble, most people will find that too much too often leave them dealing with stomach cramping, gas, and diarrhea.
Everyone has a different tolerance level for sugar alcohols. Approach these types of products with caution and take note of how your body responds to the different types of sugar alcohols and how often you’re eating it.
Celiac disease is an autoimmune condition, which means your immune system starts attacking your own body when gluten is present. This can be tested for with a simple blood test — but make sure you do not start eating a “gluten-free diet” until after you’ve had the test done, otherwise your results may not show signs of Celiac disease.
An intolerance to gluten is more like an allergy. When you eat gluten, you feel awful in any variety of ways. The only way to “test” for this is to completely remove gluten from your diet for at least 3 weeks, assess if your symptoms went away, then reintroduce it and assess if your symptoms return.
Symptoms of either condition can include: headaches, bloat, gas, tiredness, diarrhea, depression, brain fog, skin issues (eczema, acne, etc.), weight-loss, constipation, or foul-smelling stools.
Yeast infections in your GI tract
People with diabetes whose blood sugars are consistently high face a significant risk of yeast infections — including in your gastrointestinal tract. This overgrowth of yeast is the direct result of too much glucose in your bloodstream.
We need some yeast and bacteria in these areas of our body, but when blood sugars are consistently high, the excess glucose feeds the growth of the yeast.
Most likely, after being examined by your doctor, you’ll be prescribed an anti-fungal medication like nystatin, ketocanazole, or flucanazole to help kill the excessive yeast.
Fortunately, the treatment is highly effective and can resolve most yeast infections within a few days to a week. However, keep in mind that part of your treatment plan must include working to improve your blood sugar levels, otherwise, you’ll simply develop another yeast infection.
Irritable bowel syndrome
Irritable bowel syndrome (IBS) is an overarching diagnosis for a variety of symptoms related to your digestive system. It appears to be more common in the diabetes population but can develop in anyone.
The symptoms of IBS can include two specific details:
Ongoing pain in your abdomen
Frequent changes in your bowel movements (diarrhea, constipation, or both)
Discomfort after eating (but unable to specify a consistent food item as a source of pain)
Bloating and gas
Difficulty sleeping due to overall discomfort
Testing for IBS is difficult. Instead, the diagnosis is usually the result of ruling out other options, especially Celiac disease or gluten intolerance. In people with diabetes, nerve damage within the colon and small intestine will be heavily considered, too.
A diagnosis of IBS is generally the result when it’s clear there are no other signs of damage or other conditions affecting any part of your digestive system.
The treatment for IBS can include some over-the-counter medications for constipation or diarrhea, but most likely, you’ll need to take good notes on what foods cause you the most pain, consider removing gluten from your diet as it causes a great deal of digestive inflammation in many, and improve the overall amount of whole foods (versus processed foods) in your diet.
For many, following a “low-FODMAP diet” can be tremendously helpful in improving IBS-related symptoms.
Probiotics — with your doctor’s support — can also help improve the beneficial bacteria in your gut which in turn helps improve digestion.
Stress can also play a major role in IBS. If you are under a great deal of stress in your life, consider this as a potential cause of your digestive upset and talk to your healthcare team about therapy, medications, and changes in your life to reduce stress and anxiety.
Simple steps to improve your digestive health
Digestive issues as people with diabetes are common — but there are so many things you can do to prevent and protect yourself, starting with improving your blood sugar levels, eating a diet consisting of mostly whole foods, and getting regular exercise.
Like many aspects of diabetes health, improving and protecting your digestion comes down to the simplest habits and choices we make every day.
Suggested next posts:
If you found this guide to diabetic diarrhea 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.
Best Chicken Tortilla Soup Recipe Crowd Pleasing Chicken Chili Style Soup !
Best Chicken Tortilla Soup recipe makes a great tortilla soup when you have extras to feed and when you’re goal is eating healthy. You are in luck! I spoke in Clinton, LA (had so MUCH FUN) for 100 people and we served lunch from my cookbooks for healthy eating. Everyone loved my menu of healthy easy recipes and asked for the recipes! They served my Chicken Tortilla Soup recipe in my Gulf Coast cookbook and I had two bowls while signing books. I love Chicken Tortilla Soup and my recipe almost tastes like an easy White Chicken Chili. Chunky, hearty and oh soooo good! I couldn’t wait to share this my easy chicken tortilla soup ecipe with you on my healthy cooking blog. Guess what? There’s was not one drop of this delicious soup left.
Best Chicken Tortilla Soup
Leftover chicken (Rotisserie chicken), southwestern seasonings and canned broth quickly turn into a mouth-watering one-pot meal.
1(15 1/2-ounce) can Great Northern or navy white beansdrained and rinsed
2tablespoons lime juice
1 1/2teaspoons ground cumin
1tablespoon chili powder
1/2cup chopped green onions
1cup shredded reduced-fat Mexican-blend or Cheddar cheese
1 small avocadopeeled and diced
Preheat oven 350°F. Place tortilla strips on baking sheet and bake 10-15 minutes or until crisp. Set aside.
In nonstick pot coated with nonstick cooking spray, sauté onion and garlic over medium heat until tender, about 7 minutes.
Add chicken and continue cooking until chicken is done, about 5-7 minutes. Add tomatoes with juice, broth, green chilies, corn, beans, lime juice, cumin, and chili powder. Bring to boil, reduce heat and simmer about 10-15 minutes.
Serve soup topped with tortilla strips, green onions, cheese, and avocado.
Per Serving: Calories 209 Calories from Fat 17% Fat 4g, Saturated Fat 2g, Cholesterol 37mg Sodium 571mg Carbohydrates 24g Dietary Fiber 5g Total Sugars 4g, Protein 20g, Dietary Exchanges: 1 1/2 starch, 2 lean meat
Best of Chicken Chili and Tortilla Soup with Chunky Chili Flavor
A cross between an easy White Chicken Chili and Tortilla Soup, this chunky chicken tortilla soup recipe is filled with beans, corn, and southwestern flair. I think that’s the reason this recipe is the best Chicken Tortilla Soup recipe. The best of both worlds! This recipe is a thicker, chunkier soup with corn and beans. When serving the soup, I always serve my soup with condiments. I top this healthy chicken tortilla soup recipe with cheese, avocados, and tortilla strips. If you’re having trouble with deciding dinner like many of us do. I have the dinner solution of what to prepare quickly and a family one-pot meal. I believe in convenience food so pick up a rotisserie chicken and you probably have the other ingredients in your pantry.
You’ll Love My Easy White Chicken Chili Style Chicken Tortilla Soup
I know this is my Louisiana cookbook but this book really includes recipes from my Louisiana kitchen. So, you get the best healthy Cajun recipes plus lots of healthy easy recipes. There’s tons of other great soup recipes on my blog but I especially like this one. Also, you’ll find the easiest Crawfish and Corn Soup with only six ingredients and BBQ Shrimphas been a standout Cajun recipe.
Best of all, these easy Cajun recipes you can make wherever you are and they are also healthy Cajun recipes. Who said Louisiana cooking wasn’t good for you–not if you hang with me! If you like soups, check out my Chicken and Sausage Gumboor my fabulous White Chicken Chili!
I Love A White Plain Soup Bowl to Serve My Best Chicken Tortilla Soup Recipe
I really like these plain soup bowls! If you don’t have soup bowls, these bowls make a great option. The perfect size and I am partial to white bowls. Dishwasher safe and with colder weather around the corner, you’ll find you will be pulling out these bowls all the time.
They go with whatever dishes you have and this easy Chicken Tortilla Soup is a one-meal dish-even hearty enough to fill up those big eaters and so delicious!
Chicken Tortilla Soup Recipes Perfect For Tailgating in Your Home
I love serving my best Chicken Tortilla Soup recipe for company coming over to watch games or gatherings. You can even keep the soup warm in a slow cooker and serve with mugs. Have the condiments in a bowl surrounding the soup and let everyone serves themselves. Who doesn’t love Chicken Tortilla Soup and you’ll really enjoy my healthy easy tortilla soup recipe. The ingredient list might look long but the recipe is easy. Can you open cans? If you want another simple soup that’s also as easy as opening cans, try my Black Bean Soup!
Once you use these silicon pot holders, they will be your one and only kitchen pot holders for several reasons. They are easy to use and best of all, they never get dirty. My cloth pot holders end up so filthy so these colorful clean heat resistant pot holders are inexpensive and the best!
Diabetic Chicken Tortilla Soup Recipe
Yes, this delicious Chicken Tortilla Soup recipe makes an easy diabetic chicken tortilla soup. All my recipes include nutritional information and this recipe is even diabetic. It also has 5 grams fiber making it a high fiber food option and we always need to include more fiber into our diet. This is a one-meal dish and works great in a slow cooker or take a short cut with rotisserie chicken! Lots of people love a chicken tortilla slow cooker soup but honestly, I use my crock pot to serve my Tortilla soup in!
Can you eat delicious food that is also good for you? Of course! Diabetic friendly meals definitely do not have to be boring and tasteless. This Diabetic Meal Plan & Recipes Downloadable is your easy go-to guide to meal planning diabetic meals the whole family will love. This comprehensive guide includes 13 weekly recipes, from dinners, lunch, snacks and dessert.
Lipodystrophies are a group of disorders characterized by absence or loss of adipose tissue and abnormal fat distribution, commonly accompanied by metabolic dysregulation. Although considered rare disorders, their prevalence in the general population is not well understood. We aimed to evaluate the clinical and genetic prevalence of lipodystrophy disorders in a large clinical care cohort. We interrogated the electronic health record (EHR) information of >1.3 million adults from the Geisinger Health System for lipodystrophy diagnostic codes. We estimate a clinical prevalence of disease of 1 in 20,000 individuals. We performed genetic analyses in individuals with available genomic data to identify variants associated with inherited lipodystrophies and examined their EHR for comorbidities associated with lipodystrophy. We identified 16 individuals carrying the p.R482Q pathogenic variant in LMNA associated with Dunnigan familial partial lipodystrophy. Four had a clinical diagnosis of lipodystrophy, whereas the remaining had no documented clinical diagnosis despite having accompanying metabolic abnormalities. We observed a lipodystrophy-associated variant carrier frequency of 1 in 3,082 individuals in our cohort with substantial burden of metabolic dysregulation. We estimate a genetic prevalence of disease of ∼1 in 7,000 in the general population. Partial lipodystrophy is an underdiagnosed condition. and its prevalence, as defined molecularly, is higher than previously reported. Genetically guided stratification of patients with common metabolic disorders, like diabetes and dyslipidemia, is an important step toward precision medicine.
Inherited lipodystrophies are a group of genetically heterogenous disorders characterized by selective deficiency of adipose tissue in the absence of nutritional deprivation or catabolic state. This marked loss or absence of adipose tissue is commonly accompanied by hormonal and metabolic dysregulation that result in severe comorbidities due to ectopic fat accumulation in other organs such as liver and muscle (1).
Lipodystrophic disorders can be classified based on the extent and areas of fat loss that range from localized and partial lipodystrophy, generally affecting the limbs but sparing the trunk and face, to generalized lipodystrophy where lack of adipose tissue occurs in mostly the entirety of the body. Interestingly, because abnormal fat distribution is a key diagnostic feature of lipodystrophy disorders, this tends to be more recognizable in females versus males, resulting in a higher rate of reported cases for females compared with males, who are generally considered to be more muscular. Additionally, lipodystrophies can have genetic or environmental etiologies, with the latter usually a consequence of antiretroviral therapies in HIV-infected individuals (1–3). However, in the absence of significant family history or clinical signs evidencing a genetic disorder, lipodystrophy in many patients is considered to have an acquired cause or be lipodystrophy of unknown etiology (4). Nevertheless, genetic studies in families and individuals with lipodystrophy have been able to identify several genes and pathogenic variants that cause inherited lipodystrophy disorders.
Inherited lipodystrophies are classified based on their molecular etiology and adipose tissue loss distribution but can be broadly divided into congenital generalized lipodystrophies, which are mostly autosomal recessive disorders with a near complete absence of adipose tissue, and familial partial lipodystrophies (FPLDs), which are mostly inherited dominantly and present with a progressive selective loss of specific adipose depots (3,5–7). Inherited lipodystrophies are generally considered to be very rare genetic disorders with estimated prevalences ranging from 1 in 10 million for congenital generalized lipodystrophies to 1 in 1 million for FPLDs. Previous studies have estimated that the burden of inherited lipodystrophies in the general population ranges from 1.0 in 1 million (7) to 4.7 in 1 million (8). Mutations in a handful of genes have been associated with inherited lipodystrophy disorders, many of which are involved in adipocyte differentiation (BSCL2, AKT2, PPARG), maintenance (LMNA, ZMPSTE24, CIDEC, PSMB8), or function (AGPAT2, CAV1, PTRF, LIPE, PLIN1) (1,2,5,9–15). A few additional genes have been identified in patients with syndromic presentations where lipodystrophy is one of the clinical manifestations (SPRTN, PIK3R1, WRN). Mutations in LMNA and PPARG account for >50% of the reported cases of FPLD (16).
Independently from the underlying cause of the disease, the loss of or failure to develop adipose tissue in patients with lipodystrophy generally results in metabolic and endocrine abnormalities that usually encompass insulin resistance, diabetes, hypertriglyceridemia, hyperlipidemia, hypertension, hepatic steatosis, acanthosis nigricans, hormonal imbalance, and polycystic ovaries (3,17). Depending on the genetic defect, patients can also have features of cardiomyopathy, peripheral neuropathy, and skeletal abnormalities (18).
In this report, we examine the prevalence of clinically diagnosed lipodystrophy from de-identified electronic health record (EHR) data available from the Geisinger Health System (GHS) and a large claims database. Whole exome sequencing in 92,455 participants, part of the GHS-Regeneron DiscovEHR collaboration, enabled estimates of the prevalence of molecularly confirmed inherited lipodystrophies. Our findings suggest that inherited lipodystrophies are more common than previously reported.
Research Design and Methods
DiscovEHR participants are a subset of the GHS MyCode Community Health Initiative. The MyCode Community Health Initiative is a GHS-wide repository of blood, serum, and DNA samples from GHS patients that have consented to participate in research and donate samples for broad research use, including genomic analysis that can be linked to de-identified EHR information. Participants consented in accordance with the GHS Institutional Review Board–approved protocol (study number 2006-0258).
We interrogated the EHR information for 1,361,924 adult individuals (age ≥18 years) from the GHS for cases of at least one inpatient diagnosis or two outpatient diagnoses of “lipodystrophy/lipoatrophy” diagnosis codes (ICD-9 272.6/ICD-10 E88.1) excluding individuals with HIV infection diagnosis codes (ICD-9 042/ICD-10 B20). Similarly, we interrogated the Truven Health MarketScan Research Database, a database of health care claims data for 85,688,196 adult individuals enrolled in the period from 1 January 2012–30 September 2017 for the number of cases of at least one inpatient diagnosis or two outpatient diagnoses (on separate calendar days) of lipodystrophy/lipoatrophy (ICD-9 272.6/ICD-10 E88.1) with no instance of HIV infection diagnostic codes (ICD-9-CM 042/ICD-10-CM B20) during the study period divided by the total adult population enrolled during the same period. Prevalence of lipodystrophy was age and sex standardized to the U.S. population in 2017. Comorbidities including hyperlipidemia, diabetes, hypertension, and nonalcoholic fatty liver were defined as having at least one diagnosis code during the study period. Medication use including insulin, metformin, and statins was defined as having at least one dispense of each drug during the study period. A list of ICD codes used for phenotype queries can be found in Supplementary Table 1.
Sample preparation, whole exome sequencing, and sequence data production were performed at the Regeneron Genetics Center (RGC) as previously described (19). In brief, 1 μg high-quality genomic DNA was used for exome capture using the NimbleGen VCRome 2.1 or the IDT xGen target enrichment reagent. Captured libraries were sequenced on the Illumina HiSeq 2500 platform with v4 chemistry using paired-end 75 base pair (bp) reads. Exome sequencing was performed such that >85% of the bases were covered at ≥20×. Mean coverage across all samples included in the DiscovEHR cohort was ∼80×. Mean coverage for samples specifically referenced in this study was 86.4×. Raw sequence reads were mapped and aligned to the GRCh38/hg38 human genome reference assembly using BWA-MEM. Single nucleotide and indel variants and genotypes were called using the GATK HaplotypeCaller. Sequencing and data quality metric statistics were captured for each sample to evaluate capture performance, alignment performance, and variant calling. Called variants were annotated using an RGC cloud-based developed pipeline.
For downstream genetic analyses, variants were further annotated and analyzed using an in-house implemented annotation and analysis pipeline and additional customized Perl bioinformatics scripts for data processing.
For ascertainment and survey of pathogenic variants, a union of NCBI’s ClinVar pathogenic/likely pathogenic and HGMD’s (Human Gene Mutation Database) high-confidence disease-causing mutations (DM-High) reported variants with a phenotype association to “lipodystrophy” was considered. Variants were considered “pathogenic” if they were reported in both databases without conflicting interpretations.
Data and Resource Availability
The Truven Health MarketScan Research Database is available through paid license. The Exome Aggregation Consortium (ExAC) database is publicly available through exac.broadinstitute.org. Genomic variant frequency data for the first 50,000 DiscovEHR participants are available through www.discovehrshare.com. Individual phenotype and genomic data for DiscovEHR participants are not available due to privacy considerations. Genetic variants reported and discussed in this study are disclosed within the main text and figures and in Supplementary Data. Additional information for reproducing the results described in this study is available upon reasonable request and subject to a data use agreement and appropriate consent and privacy considerations.
We interrogated the EHR data for 1,361,924 adult individuals from the GHS for “lipodystrophy/lipoatrophy” diagnosis codes (ICD-9 272.6/ICD-10 E88.1) to assess the prevalence of this phenotype from a clinical perspective in an unascertained EHR-linked clinical cohort. We identified 114 adult patients with a clinical diagnosis code consistent with lipodystrophy; of these, 99 did not have a diagnosis of HIV infection and were therefore considered as likely having an inherited or genetic cause of lipodystrophy versus an acquired etiology. The calculated prevalence for lipodystrophy in the GHS clinical cohort was 7.2 in 100,000 (or 1 in 13,889). In order to replicate our observations, we similarly interrogated the Truven Health MarketScan Research Database for the number of cases of diagnoses of lipodystrophy/lipoatrophy divided by the total adult population enrolled during the same period. We identified 6,055 patients with a lipodystrophy diagnosis, of whom 4,029 did not have a diagnosis of HIV infection, resulting in an estimated prevalence for lipodystrophy of 4.7 in 100,000 (or 1 in 21,277). Consistent with previous reports on lipodystrophy, we observed a sex bias in the clinical diagnosis of lipodystrophy, with 77–80% of patients being female. Additionally, and consistent with disease presentation and known accompanying metabolic abnormalities, we observed an increased proportion of lipodystrophy patients also having diagnosis codes for comorbidities such as hyperlipidemia (ICD-9 272/ICD-10 E78), diabetes (ICD-9 250/ICD-10 E11), hypertension (ICD-9 401/ICD-10 I10), and nonalcoholic fatty liver disease (ICD-9 571.8/ICD-10 K75.8 and K76) in both databases (Table 1). The demographic and clinical characteristics of individuals with a clinical diagnosis of lipodystrophy in these two cohorts are summarized in Table 1. We further standardized the observed prevalence of the disease in the MarketScan database based on age and sex and projected this onto the U.S. population (as of July 2017) to better estimate the prevalence of lipodystrophy in the general population, resulting in an estimate of 47.3 cases per 1 million people (Supplementary Table 2) (or ∼1 in 21,142). Based on these data, we estimate the clinical prevalence of inherited lipodystrophy to be ∼1 in 20,000 individuals.
Demographics and comorbidities of lipodystrophy*and control patients from two unascertained clinical population databases
Of the 99 individuals with a likely inherited cause of lipodystrophy, 24 had been sequenced as part of the ongoing Geisinger-Regeneron DiscovEHR collaboration that links EHR and genomic data (19). Analyses of the phenotypes in these individuals showed that type 2 diabetes is significantly enriched in patients with a clinical diagnosis of lipodystrophy (odds ratio [OR] 4.28 [95% CI 1.90 – 9.64]), P = 1.27 × 10−4) (Table 2). Next, we performed genetic analyses in the 24 individuals with available genetic data in order to identify the likely causative variants of the condition in these patients. We identified four individuals who are heterozygous for a previously reported pathogenic variant in LMNA [hg38.g.chr1:156136985(G>A); c.1445G>A; p.R482Q] associated with Dunnigan FPLD (FPLD2) (11,20–22). Interestingly, an additional 12 individuals were also carriers of this genetic variant; however, they did not have any documented diagnoses of lipodystrophy in their EHR. This variant is present in our unascertained clinical cohort at a frequency of 0.000173 (16 of 92,455 sequenced individuals). The frequency of this variant appears to be higher in DiscovEHR than in other sequenced population cohorts and publicly available databases such as ExAC (minor allele frequency [MAF] = 0.000008324) and gnomAD (MAF = 0.000004063). De-identified EHR review of these 16 LMNA p.R482Q heterozygous individuals showed that they have diagnosis codes for phenotypes consistent with the metabolic abnormalities observed in lipodystrophy and that these abnormal metabolic phenotypes are significantly enriched in these individuals versus the rest of the DiscovEHR cohort, including hyperlipidemia (93.75%) (OR 11.19 [95% CI 1.47–84.76], P = 3.18 × 10−3), hypertension (93.75%) (OR 12.47 [95% CI 1.64–94.45], P = 1.65 × 10−3), diabetes (81.25%) (OR 13.26 [95% CI 3.77–46.54], P = 1.47 × 10−7), and inflammatory liver disease (25%) (OR 3.42 [95% CI 1.10–10.60], P = 0.02344) (Fig. 1 and Table 2). Their EHR also shows that they are on medications to treat these conditions, including insulin (68.75%), metformin (75%), and statins (100%). Additionally, we also observed other phenotypes less consistently associated with lipodystrophy including atherosclerosis (56.25%), hypothyroidism (37.5%), and chronic kidney disease (37.5%). Interestingly, other phenotypes of interest that were enriched in this group of patients were heart disease (68.75%) with numerous diagnoses of heart failure, myocardial infarctions and conduction disturbances, and neuropathies (50%) including carpal tunnel syndrome and one patient with a diagnosis of diabetic polyneuropathy (Fig. 1). These accompanying diagnoses would be consistent with the clinical spectrum observed for laminopathies, specifically, LMNA-associated disorders (18,23). Of note, identity by descent (IBD) metrics derived from genomic data to estimate relatedness (24) among individuals carrying the LMNA (p.R482Q) variant only identified two of these carriers to be related, a brother-sister sibling pair where, although both are carriers of the variant, only the sister (patient 4 [Fig. 1A]) has a clinical diagnosis of lipodystrophy. Examination of the brother’s EHR information showed that he has clinical diagnoses of type 2 diabetes, hyperlipidemia, and hypertension (patient 7 [Fig. 1A and B]). The remaining individuals do not appear to be related up to a fifth-degree relationship, which would be equivalent to unrelated individuals from an outbred population. We further confirmed this by exploring the IBD segments among these individuals and did not observe large shared genomic segments, except for the sibling pair previously identified and beyond the region where the LMNA variant is located. We were able to narrow the genomic region in which this variant resides to a 5.332 Mb minimum shared haplotype across all carriers (Fig. 1C). Genetic analyses of the remaining patients with a clinical diagnosis of lipodystrophy did not identify a clear pathogenic variant for the disease in known lipodystrophy disease genes.
Demographics and comorbidities of individuals with lipodystrophy and individuals carrying the LMNA (p.R482Q) variant in the DiscovEHR cohort
Phenotypic and genetic characterization of carriers of the pathogenic LMNA (p.R482Q) variant in DiscovEHR cohort. A: Major phenotypes identified through de-identified EHR review of the encounter diagnoses of DiscovEHR participants identified to be carriers of the pathogenic p.R482Q variant in LMNA. Patients are listed from lower to higher BMI as noted in the third column and colored based on sex: females, pink; males, blue. Age is listed in the second column. For the phenotypes: dark blue, presence of specific diagnosis codes for the particular disease; light blue, absence of diagnosis codes for the disease; intermediate blue, suggestive of disease due to related diagnosis codes but no listing of the specific code for the particular disease. B: Inferred pedigree for the two first-degree–related individuals among the 16 p.R482Q variant carriers, a brother-sister pair where only the female has a clinical diagnosis of lipodystrophy, despite the brother having diagnosis codes consistent with comorbidities documented in lipodystrophy patients. C: Identification of the common 5.332 Mb haplotype harboring the p.R482Q variant in LMNA by IBD segment analysis in the 16 carrier individuals.
To further explore whether additional cases of genetically driven lipodystrophy exist in our clinical population, we surveyed the DiscovEHR cohort (N = 92,455 sequenced individuals) for pathogenic and likely pathogenic variants previously reported in the HGMD or the NCBI database of Clinical variants (ClinVar) in lipodystrophy-associated genes. We identified an additional eight individuals with previously reported pathogenic variants in these databases in LMNA, PPARG, and PIK3R1, all without a clinical diagnosis of lipodystrophy (Supplementary Table 3). Similar to our observations in the LMNA (p.R482Q) variant carriers, we observed a substantial burden and similar pattern of metabolic dysregulation by EHR review in carriers of these variants consistent with a likely undiagnosed lipodystrophy disorder (Fig. 2A). We identified an individual carrying a predicted pathogenic missense variant classified as “likely pathogenic” in ClinVar [hg38.g.chr3:12392714(G>A); c.G497A; p.R166Q] affecting the highly conserved Arg166/Arg194 residue in PPARG. A previously reported pathogenic variant (p.R166W/p.R194W) affecting this same residue and shown to disrupt DNA binding activity of PPARG (25) was also found in our cohort (Fig. 2 and Supplementary Table 3). This p.R166Q/p.R194Q variant is predicted to be pathogenic by bioinformatic algorithms, including the MITER classifier tool specifically developed to assess pathogenicity of missense variants in PPARG (26). The “probability of causing FPLD3” (FPLD type 3) according to MITER for each of the three missense variants in PPARG identified in our cohort (p.R166W/p.R194W, p.R166Q/R194Q, and p.V290M/p.V318M) was 97.2%, 99.9%, and 5.6%, respectively. Based on previously reported pathogenic variants only, the prevalence of lipodystrophy in the DiscovEHR cohort is 1 in 4,020. However, we additionally identified six individuals with heterozygous predicted loss of function and expected pathogenic variants in PPARG. Loss-of-function and dominant negative variants in PPARG have been associated with FPLD3 (MIM #604367), severe insulin resistance, diabetes, and hypertension (10,14,26). These additional six carriers of expected pathogenic variants had a clinical presentation similar to those of the other pathogenic variant carriers (Fig. 2A). Four of these individuals were related and part of the same pedigree segregating a novel nonsense variant in exon 3 of PPARG [hg38.g.chr3:12379922(G>T); c.G217T; p.E73X] (Fig. 2B). Interestingly, a progression of disease and additional phenotypes consistent with FPLD3 can be observed in the older individuals of these pedigree versus the two younger sisters (Fig. 2B). Altogether, we identified 14 additional individuals who are carriers of a pathogenic, likely pathogenic, or expected pathogenic variant for lipodystrophy in our DiscovEHR cohort. These individuals have hallmarks of metabolic disease in the spectrum of lipodystrophy comorbidities but do not have a documented clinical diagnosis of the disease in their EHR. Overall, among the 92,455 sequenced participants from the DiscovEHR clinical cohort, we identified 30 carriers of pathogenic, likely pathogenic, or expected pathogenic variants in the lipodystrophy-associated genes, LMNA, PPARG, and PIK3R1, which amounts to a genetic variant carrier prevalence of 1 in 3,082. In order to evaluate the genetic prevalence of lipodystrophy in a different data set, we conducted a similar survey of lipodystrophy-associated genetic variants in the ExAC (27) database of 60,706 non–clinically ascertained individuals and identified 8 individuals carrying pathogenic or likely pathogenic variants in lipodystrophy-associated genes. This number accounts for a pathogenic variant carrier prevalence for autosomal dominant FPLD of ∼1 in 7,588 individuals in ExAc. Combining these two observations from a large clinical care cohort with available genomic data linked to longitudinal EHR information and a non–clinically ascertained large population genomic data set with broad continental ancestry representation, we estimate the molecular prevalence of lipodystrophy disorders to be 1 in 7,000 individuals.
Phenotypes in carriers of pathogenic, likely pathogenic, and expected pathogenic variants for lipodystrophy in the DiscovEHR cohort. A: Major phenotypes identified through de-identified EHR review of the encounter diagnoses of DiscovEHR participants found to be carriers of known pathogenic, likely pathogenic, and expected pathogenic variants for lipodystrophy. The gene and identified variant are listed in the first column, age is reported in the second column, and corresponding BMI is noted in the third column. Patients are colored based on sex: females, pink; males, blue. For the phenotypes: dark blue, presence of specific diagnosis codes for the particular disease; light blue, absence of diagnosis codes for the disease; intermediate blue, suggestive of disease due to related diagnosis codes but no listing of the specific code for the particular disease. B: Inferred pedigree for four identified individuals carrying a novel nonsense variant (p.E73X) in PPARG expected to be pathogenic and cause lipodystrophy. These individuals correspond to patients 8, 9, 10, and 11 from panel A listed in descending order for age. Severity of the phenotype appears to correlate with age with the older female individual being the most affected and with the youngest only having recorded diagnoses for “other and unspecified hyperlipidemia” within a short electronic health record.
Through clinical and molecular investigations in an unascertained large clinical cohort, our data show that lipodystrophy is an underestimated and underdiagnosed condition. The prevalence of disease is higher than previously estimated both from a clinical and molecular genetics perspective.
From large clinical cohorts, we have derived an estimated clinical prevalence of lipodystrophy disorders of ∼1 in 20,000 individuals, which is much higher than previously reported. This was first observed in the GHS database encompassing >1.3 million adult patients and further confirmed in the larger MarketScan database that includes de-identified data from >85 million adult individuals. This estimated prevalence is in contrast with previous reported estimates from the literature of <1.0 in a million (7) and a more recent study that similarly looked at EHR-based databases and literature to achieve an estimated prevalence of 1.3–4.7 cases per million for all types of lipodystrophy (8). Similarly to Chiquette et al. (8), we queried the databases for the corresponding clinical diagnosis codes for lipodystrophy and excluded individuals with HIV. However, instead of using additional inclusion/exclusion criteria, we decided to characterize the clinical features of this disorder by evaluating the fraction of individuals presenting with the expected associated comorbidities. We observe an enrichment of diagnosis codes for hyperlipidemia, type 2 diabetes, hypertension, and liver disease in individuals that also carry a clinical diagnosis of lipodystrophy (Table 1). This enrichment is consistent and significant both in the broader clinical lipodystrophy cohorts and in the subset of sequenced DiscovEHR participants, despite the high rate of metabolic disease in the larger DiscovEHR cohort as evidenced in Table 2 and as previously reported (19). Although the majority of the DiscovEHR cohort is clinically unascertained, a fraction of the participants is derived from the GHS cardiac catherization laboratory and the bariatric surgery clinic (19), which could contribute to an overrepresentation of metabolic abnormalities in this cohort. Furthermore, individuals who have more frequent interaction with the health care system are more likely to be approached and recruited to consent for the MyCode Community Health Initiative, which could also contribute to the enrichment of chronic diseases represented among the DiscovEHR participants.
Consistent with previous reports in lipodystrophy patients, we observed a 3:1 ratio of females being diagnosed versus males. This has been documented to be a diagnostic bias due to the increased likelihood of detecting abnormal fat distribution in females versus males due to a more muscular appearance in the absence of subcutaneous fat. It is also possible that in our database query, we are capturing cases of localized lipoatrophy or insulin lipoatrophy, which based on ICD diagnoses are not possible to differentiate and may be contributing to an increase in our estimate of clinical prevalence of lipodystrophy. However, because individuals with lipodystrophy may also use insulin as a medication, as evidenced by our data in Table 2 where 11 of the 16 (68.75%) patients with a molecular diagnosis of FPLD2 due to the p.R482Q variant in LMNA have used insulin as a medication, excluding patients on insulin may not be appropriate when assessing the clinical prevalence of this disorder.
Additionally, we performed genetic analyses to better characterize the underlying molecular etiology of lipodystrophy in patients with a clinical diagnosis of disease, as well as to explore the phenotypic associations of pathogenic and likely pathogenic variants in lipodystrophy-associated genes. In our DiscovEHR cohort, we identified a known molecular cause of disease in 16.66% (4 of 24) of our sequenced participants with a clinical diagnosis of lipodystrophy. In addition to four lipodystrophy-diagnosed patients, the p.R482Q variant in LMNA that has been previously reported and characterized as causative of FPLD2, Dunnigan type, was observed in additional patients in our cohort who did not carry a clinical diagnosis code for lipodystrophy. However, de-identified manual chart review of their EHR showed that they had clinical features consistent with disease. The p.R482Q variant was identified in probands from distinct families segregating Dunnigan-type FPLD from Canada and the U.K. (11,20). Clinical characterization of patients with this variant has shown that variant carriers have reduced plasma leptin levels by ∼60% of noncarriers and increased concentrations of fasting plasma insulin and C-peptide (28). Additionally, they often are diagnosed with dyslipidemia, hyperinsulinemia, diabetes, and hypertension (21). Our observations confirm these previously reported associations with metabolic abnormalities in patients with this variant. Furthermore, they highlight the underdiagnosis of lipodystrophy in these individuals that have all of the metabolic disease hallmarks of FPLD2 due to the p.R482Q variant in LMNA. Additional variants in LMNA including other missense variants altering the same codon of the Arg482 residue have been associated with FPLD2, suggesting an important role of this residue and the COOH-terminal globular domain of lamin A/C in adipocyte maintenance (11,22). In addition to FPLD2, mutations in LMNA have been associated with a range of genetic disorders broadly known as laminopathies, including Charcot-Marie-Tooth disease type 2B1 (MIM #605588), dilated cardiomyopathy (MIM #115200), Hutchinson-Gilford progeria (MIM #176670), and a spectrum of muscular dystrophies, among others (18). The clinical spectrum of these disorders often overlaps and involves other tissues including skeletal and cardiac muscle and the peripheral nervous system. Therefore, it is not surprising that FPLD2 patients due to mutations in LMNA often present with additional comorbidities beyond the metabolic complications common to lipodystrophy, including heart disease and peripheral neuropathy, as evidenced by our patients with LMNA mutations (23) (Figs. 1 and 2). Variability in the expression of these phenotypes in patients can be partly explained by the progressive and later-onset nature of these traits; however, the possibility of genetic and/or other environmental modifiers that may exacerbate, accelerate, or prevent the onset of additional comorbidities may be further explored in the future as more patients with these conditions are ascertained and phenotyped in detail. While lipodystrophy is not a condition currently covered by the ACMG SF v2.0 gene list (29), LMNA is one of the genes screened for clinically actionable secondary findings in genomic sequencing studies due to its association with dilated cardiomyopathy characterized by cardiac dilation and reduced systolic function (29,30). It is possible that as more genomic sequencing efforts start looking for pathogenic variants in this gene, additional cases of molecularly undiagnosed lipodystrophy patients will become evident (31), further informing the prevalence and contribution of LMNA pathogenic variation to partial lipodystrophy and metabolic disease in unascertained patients.
In addition to the p.R482Q variant in LMNA that appears to be enriched in our particular cohort compared with other publicly available population databases, we identified additional pathogenic and likely pathogenic variants in LMNA, PPARG, and PIK3R1 in patients with metabolic abnormalities in the spectrum of lipodystrophy but without a clinical diagnosis. This suggests that the prevalence of genetic lipodystrophies, as defined molecularly, might be higher than previously reported. Loss of function and pathogenic variation in PPARG and LMNA appear to be the major molecular contributors to cases of undiagnosed lipodystrophy. In our cohort, we calculate the prevalence of inherited lipodystrophy disorders to be ∼1 in 3,082. However, with exclusion of the p.R482Q LMNA variant enriched in the DiscovEHR cohort to correct for an overascertainment of individuals with lipodystrophy in this population, a carrier frequency of ∼1 in 6,604 is calculated. This lipodystrophy variant carrier frequency is more in line with the observation in a non–clinical care cohort such as ExAc, where ∼1 in 7,588 individuals carry a pathogenic or likely pathogenic variant. The resulting calculated molecular prevalence of ∼ 1 in 7,000 individuals for lipodystrophy is much higher than previously estimated. This is particularly relevant, as carriers of lipodystrophy-associated variants show metabolic abnormalities in the spectrum of lipodystrophy disease; however, they are more often diagnosed as having common metabolic diseases, such as type 2 diabetes, dyslipidemias, and unspecified metabolic syndrome (17).
Although the clinical and molecular prevalence of lipodystrophy was higher in our DiscovEHR cohort compared with broader external databases, it is likely that the clinical prevalence observed in DiscovEHR will be typical for other clinical care–ascertained cohorts, while the molecular prevalence estimate will be relevant to the implementation of precision medicine in similar clinical settings. The availability of both clinical and genetic data from a single large clinical care cohort such as DiscovEHR allows us not only to begin to estimate prevalence rates but also to explore issues of underdiagnosis, variant pathogenicity, penetrance, and expressivity of “rare” genetic disorders such as lipodystrophy. Genetic diagnosis of these conditions can inform disease course and potential disease complications in carriers of pathogenic variants. Early treatment and appropriate management can improve prognosis in patients with a molecular diagnosis of lipodystrophy before the onset or exacerbation of metabolic complications and comorbidities. Additionally, other therapies that address the molecular defect, such as the use of leptin analogs, may improve the metabolic profile of these patients. Furthermore, genetically guided stratification of patients with “common” disorders, such as diabetes and dyslipidemia, is an important step toward precision medicine and has the potential to lead to more effective therapies for patients.
Acknowledgments. The authors thank the participants of the Geisinger MyCode Community Health Initiative and the DiscovEHR Geisinger-Regeneron collaboration who contribute their genomic and clinical data for research purposes to improve and implement genomic precision medicine.
Duality of Interest. C.G.-J., W.G., J.S., C.V.H., A.Y., J.G.R., J.D.O., A.R.S., S.G., A.B., and J.A. are full-time employees of the Regeneron Genetics Center or Regeneron Pharmaceuticals, Inc., and receive stock options as part of compensation. O.G. and J.G. are former employees of Regeneron Pharmaceuticals, Inc., and received stock options as part of compensation. J.G. has recently become an employee of Exonics Therapeutics. No other potential conflicts of interest relevant to this article were reported.
Author Contributions. C.G.-J. coconceived and designed the study, performed genetic and data analyses, elaborated tables and figures, and wrote and revised the manuscript. W.G. performed phenotype data analyses. J.S. contributed to genetic analyses. C.V.H. contributed to statistical analyses and manuscript editing and revision. A.Y., R.C., J.B.L., H.L.K., and M.F.M. contributed to phenotype data analyses. J.G.R. contributed bioinformatic resources. D.J.C. contributed to study design and manuscript revision. J.D.O. contributed to sequencing of samples, data quality, and manuscript revision. A.R.S. contributed to study design and discussion and manuscript editing and revision. O.G. contributed to study design; phenotype data analyses, resources, and tools; and manuscript revision. S.G. contributed to phenotype data analyses and methods. J.G. and A.B. contributed to study design and manuscript revision. J.A. coconceived and designed the study and wrote and revised the manuscript. C.G.-J. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Prior Presentation. Parts of this study were presented in abstract form at the American Society of Human Genetics 67th Annual Meeting 2017 (Orlando, FL, 17–21 October 2017) and European Human Genetics Conference 2018 (Milan, Italy, 16–19 June 2018).
The older I get the more avocados I eat and the better I like guacamole. This Pear Guacamole with Pistachios features sweetness plus a hint of smoke from the toasted nuts.
You know avocados are good for you, right? Loaded with monounsaturated fat, vitamins, potassium, and a slew of other nutrients, they may reduce the inflammation that causes a lot of disease. Eating avocados may also lower your cholesterol and triglycerides, protect your vision, prevent cancer, and help you lose weight. But that’s not why I eat them. I eat them—a lot—because they are tasty and versatile.
How to Serve Pear Guacamole with Pistachios
You can use guacamole in lots of different ways at any meal or snack. Here are just a few ideas for serving this pear guacamole:
With vegetable sticks/strips like carrots, celery, jicama, and/or colorful bell peppers.
On avocado toast. Use whole-grain or sourdough bread, watch your portion size, and eat some protein too.
With a spoon 🙂 (It’s delicious by itself.)
Pear Guacamole with Pistachios
Creamy vegan guacamole with hints of sweetness and smoke
Author:Adapted from Food Network
Course: Appetizer, Condiments, Snack
Keyword: guacamole, pear guacamole
2 large ripe avocados halved and pitted
juice from 1 small lime
1 teaspoon chopped pickled jalapeños
2 cloves garlic minced
1/4 teaspoon kosher salt
1/2 teaspoon ground coriander
2 tablespoons chopped cilantro or parsley plus more for garnish
1 small pear cored and finely diced
freshly ground black pepper
3 tablespoons pistachios toasted and chopped
Scoop out the flesh from each avocado half and place it in a large bowl. Mash with a fork or potato masher. Stir in the lime juice.
Add jalapeños, garlic, salt, coriander, and cilantro. Mix well, then stir in the pear and half of the pistachios. Season with pepper.
Garnish with remaining pistachios and cilantro.
If you can’t find pickled jalapeños, you can use some finely minced fresh jalapeño or green bell pepper (or any other pepper you like).
Serve this guacamole with vegetable dippers like red bell pepper or jicama strips for a snack. Try it atop a Mexican salad featuring black beans. It’s also good spread on toasted whole-grain bread (just watch the carbs).
I started to do a meal plan over the weekend and was looking through sale flyers, but realized I still had plenty of food in my fridge/freezer and pantry that I didn’t need to buy anything but fresh fruit.
I spent $6.68 cents on groceries this week. I bought bananas for Hannah but they were already too ripe for her to eat – she can’t eat them if there is the tiniest bit of brown on them. Even though the stems were green, nope – she wasn’t going to eat them.
So expect a lot of banana recipes!
These pancakes are dense, not fluffy like buttermilk pancakes, but delicious. After I made them and found that I got 9 pancakes out of the recipe, I put the recipe into the recipe builder, and each one is 1 point each no matter how many you eat.
Believe me – two are plenty!
I used my stick blender to mix up the wet ingredients. Even though these aren’t fluffy pancakes, I still let the baking powder do its magic and let the batter sit 15 minutes before making the pancakes. I used 1/3 cup measuring cup for each pancake.