A long-term ketogenic diet causes hyperlipidemia, liver dysfunction, and glucose intolerance from impaired insulin secretion in mice

Abstract

Ketogenic diets (KDs)-very-low-carbohydrate and very-high-fat diets-have gained popularity as therapeutic against obesity and type 2 diabetes. However, their long-term effects on metabolic health remain understudied. Here, we show that, in male and female mice, a KD protects against weight gain and induces weight loss but over time leads to the development of hyperlipidemia, hepatic steatosis, and severe glucose intolerance. Unlike mice on conventional high-fat diet, KD-fed mice remain insulin sensitive and display low-insulin levels. Hyperglycemic clamp and ex vivo glucose-stimulated insulin secretion assays revealed systemic and cell-intrinsic impairments in insulin secretion. Transcriptomic profiling of islets from KD-fed mice indicated endoplasmic reticulum (ER)/Golgi stress and disrupted ER-Golgi protein trafficking, which were confirmed by electron microscopy showing a dilated Golgi network consistent with defective insulin granule trafficking and secretion. Together, these results suggest that long-term KD leads to multiple aberrations of metabolic parameters that caution their systematic use as a health-promoting dietary intervention.

Fig. 1.. Mice on KD are protected from weight gain as compared to mice on HFD.

(A) Diets and color scheme used throughout the paper. (B) Macronutrient composition of diets used. (C) Male fed and fasted beta-hydroxybutyrate levels taken at 4 weeks [overnight (OVN) fast] and 6 weeks [ad libitum (AL) fed]; n = 7 to 10 per group. (D) Female fed and fasted beta-hydroxybutyrate at 15 weeks on diets (n = 4 to 5 per group). (E) Male BW pooled from three independent cohorts (n = 30 to 50 per group). (F and G) Male body composition (n = 5 to 6 per group): (F) fat mass and (G) lean mass. (H) Average 24-hour food intake from 36 weeks of data collection from three independent cohorts (n = 8 to 14 cages per group). (I) Female BW pooled from two independent cohorts (n = 10 to 20 per group). (J and K) Female body composition (n = 5 to 10 per group): (J) fat mass and (K) lean mass. (L) Average 24-hour food intake from 26 weeks of data collection from two independent cohorts over (n = 2 to 4 cages per group). Statistics: One-way ANOVA was used for (H) and (L), and mixed effects ANOVA was used for all other comparisons, post Hoc testing used Tukey’s HSD test. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001; # only used for within group comparisons, #P < 0.05, ##P < 0.01, and ###P < 0.0001. Data are represented as means ± SEM.

Fig. 2.. A long-term KD causes dyslipidemia regardless of sex, and males have steatosis and liver dysfunction.

(A and B) Plasma TG and (C and D) NEFAs in [(A) and (C)] males (n = 5 to 10 per group) and [(B) and (D)] females (n = 9 to 10 per group) after 32 and 15 weeks on diets, respectively. Blood samples were collected in the fed state. (E) Representative images from males (top) and females (bottom) on diet for 28 and 33 weeks, respectively. Scale bars, 20 μm. (F and G) Liver scoring by a pathologist blinded to conditions (male, n = 5 per group; female, n = 3 to 6 per group). (F) Steatosis as estimated percent visual field containing fat. (G) Inflammation score from 0 to 3 based on average number of inflammatory foci per visual field with five visual fields scored as described in (90). (H) Liver TG in males from two independent cohorts on diets for 28 and 38 weeks (n = 10 to 11 per group) and females (n = 8 to 10 per group) on diets for 33 weeks. (I and J) Plasma alanine aminotransferase (ALT) levels in males [(I), n = 5 to 10 per group] and females [(J), n = 9 to 10 per group] after 32 and 15 weeks on diets resp. (K to M) Liver bulk RNA sequencing (males, 38 weeks on diets, n = 4 to 5 per group. (K) UpSet plot comparing all DE genes. [(L) and (M)] Heatmap of (L) the top 200 most significant DE genes and (M) selected genes associated with steatosis, fibrosis, and metabolism between LFD and KD. Color intensity represents the z score based on log₂ expression values. Statistics: [(A) to (F) and (H) to (J)] Data are represented as means ± SEM. One-way ANOVA with Tukey’s post hoc testing with P values as: *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. (G) Kruskal-Wallis test with Dunn’s multiple comparison tests for post hoc analyses. [(K) to (M)] RNA sequencing statistics were determined using DESeq2 analysis with a Wald test and Benjamini-Hochberg correction. Significant genes had an adjusted P value of <0.05.

Fig. 3.. A long-term KD causes severe glucose intolerance and suppresses insulin secretion.

(A to C) Fasted and fed BG. Fasted glucose was measured at 7 a.m. (onset of dark cycle) following a 14-hour fast, and fed glucose was measured at 11 a.m. during the dark cycle. (A) Males after 4 weeks (n = 8 to 10 per group), (B) 27 and 32 weeks on diet from two independent cohorts (n = 5 to 10 per group). (C) Females after 15 weeks on diets (n = 4 to 9 per group). (D to G) Glucose tolerance tests (GTTs) with a bolus of glucose (1.5 mg/g) given after an overnight fast (over the light phase) in: (D) males after 11 weeks on diets (n = 8 to 10 per group), (E) males after 33 weeks on diets (n = 5 to 6 per group), (F) females after 15 weeks on diets (n = 9 to 10 per group), (G) females after 31 weeks on diets (n = 8 to 10 per group). (H to M) Insulin tolerance tests with insulin (0.75 U/kg). Bar graphs [(I), (K), and (M)] show the slopes in the first 15 min (I) or 30 min [(K) and (M)]. [(H) and (I)] Males after 11 weeks on diets (n = 5 per group). [(J) and (K)] Males on diets for 31 weeks (n = 5 to 10 per group). [(L) and (M)] Females on diets for 29 weeks (n = 5 to 10 per group). (N to Q) Plasma insulin levels at baseline and 30 min after a bolus of glucose. Data in (N), (P), and (Q) are from the blood collected during the GTT presented in (D), (F), and (G), respectively. Data shown in (O) are from a separate experiment after 30 weeks on diets (n = 8 to 10 per group). Statistics: One-way or two-way ANOVA with Tukey’s HSD for post hoc testing was used. Paired t tests were used to compare insulin levels between 0 and 30 min within a group. The number of symbols represents P values as: *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. Data are represented as means ± SEM.

Fig. 4.. A HFD but not a KD increases islet size, pancreas weight, and insulin content in the pancreas.

(A to G) Analysis of chromogenic insulin/proinsulin staining in the pancreas with hematoxylin counterstain. Qupath was used to quantify islet area based on detection of the brown insulin stain. [(A) to (C)] Males, 8 to 12 slices were taken throughout the pancreas from n = 4 to 6 animals per groups for a total 5157 islets analyzed on LFD, 6718 on HFD, and 6389 on KD. [(E) to (G)] Females, six to seven slices were taken throughout the pancreas (n = 5 per group), for a total of 3749 LFMP and 3428 KD islets analyzed. [(A) and (E)] Average islet area. [(B) and (F)] Frequency distribution of islet sizes as number of islets per slice in each bin. Bins correspond to islet diameter 100 μm² = diameter of 11.3 μm or the size one β cell (and the minimum size needed to identify a nucleus and clear borders in the analysis); 170 μm² = diameter of 14.7 μm; 490.87 μm² = diameter of 25 μm; 1963 μm² = diameter of 50 μm; 4417.86 μm² = diameter of 75 μm; 7853.981 μm² = diameter of 100 μm; and 17,671 μm² = diameter of 150 μm. [(C) and (G)] Average number of islets per slice. (D) Representative images from male islets. Scale bar, 100 μm. (H) Male pancreas weight after 6 to 9 months on respective diets (n = 17 to 18 per group pooled from three independent cohorts) and (I) insulin content after 6 months on the diets (n = 5 to 6 per group). (J) Female pancreas weight (n = 4 to 8 per group) and (K) insulin content after 6 months on respective diets (n = 3 to 4 per group. Statistics: One-way or two-way ANOVA with Tukey’s HSD for post hoc testing was used. The number of symbols represents P values as: *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. Data are represented as means ± SEM.

Fig. 5.. Mice on KD lose rapid insulin secretion.

(A to F) Measures taken during hyperglycemic clamp after 25 weeks on the respective diets. n = 4 to 6 male mice per group. Time courses of BG (A), the GIR (C), and plasma insulin levels (E) obtained via tail vein sampling. Average BG (B) and average GIR (D) during the 90 to 120 min when BG was clamped. (F) Insulin levels before and 15 min into the glucose infusion. (G to I) Ex vivo GSIS (n = 3 per group). (G) Insulin level over time during the different phase of the perifusion assay, wherein insulin level is normalized to 100 insulin equivalents (IEQ; 1 IEQ represents an islet with a diameter of 150 μm). (H) Total insulin secretion during the GSIS as calculated by the area under the curve (AUC) from (G). (I) Peak insulin secretion during high glucose and KCl stimulation [insulin levels from (G) at 18 and 69 min, respectively]. Statistics: One-way ANOVA (B, D, and H) or two-way ANOVA [(A), (C), (E), (G), and (I)] with post hoc testing using Tukey’s HSD except in (H), which used Dunnett’s multiple comparison test to compare HFD and KD with the LFD group only. The number of symbols above data denote the following P values: *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. Data are represented as means ± SEM.

Fig. 6.. Islet transcriptomic analysis reveals that KD causes ER Golgi stress and impaired protein transport.

Bulk RNA sequencing from islets isolated from male mice after 36 weeks on interventions. (A, C, and F) Heatmaps showing z scores of log₂-normalized expression. (A) Heatmap of all DE genes. The Venn diagram inset shows number of genes per comparison, and the colors correspond to the colored labels on the heatmap. (B) The 397 genes that were DE in both KD versus LFD and KD versus HFD were run through Metascape, which identified ontology categories that were enriched in one group versus the other as displayed in the heatmap; the purple indicates increased expression in the corresponding group, and the color intensity represents −log₁₀(P value) of the enrichment as calculated by Metascape. (C) Using the Gene Ontology database, we identified genes involved in ER and Golgi stress genes and plotted the genes that were significantly different in KD versus LFD. Heatmap of selected genes associated with ER-Golgi stress. (D) Significantly up-regulated pathways identified using Reactome Pathway analysis of DE genes between KD and LFD. False discovery rate < 0.05 for all pathways. (E) Pathway of insulin granule generation and processing. Colors in (E) correspond to colors in (D) and (F). (F) Heatmap of selected genes involved in protein processing and vesicular transport. (G) Electron micrograph images of pancreatic β cells showing Golgi dilation in KD. (H) Mechanism linking transcriptomic findings with reduced insulin secretion. Statistics: DE genes were identified using DESeq2 analysis with a Wald test and Benjamini-Hochberg correction. Significant genes had an adjusted P value of < 0.05.

Fig. 7.. A KD for WL is less effective than a LFD and causes impaired insulin secretion.

(A and D) Body weight during weigh gain (WG), WL, and WRG intervention in males (A) and females (D). (A) 14 weeks of WG on 60% HFD (n = 30), followed by 9 weeks of WL on KD (red) or LFMP (gray) (n = 7 to 10 per group), and WRG on HFD (KD→HFD, purple and LFMP→HFD, green) in males. (B) GTT after 7 weeks of WL [week 21 dotted line on the BW graph in (A)] with glucose (1.5 mg/g). (C) In vivo GSIS: Plasma insulin during GTT in (B) and pre-WL. (D) Fifteen weeks of WG on 60% HFD (n = 28), followed by 16 weeks of WL on KD (red) or LFD (black) (n = 5 to 10 per group). At week 31 (16 weeks of WL), females were switched back to HFD (KD→HFD, purple and LFD→HFD, green) leading to WRG. (E) GTT after 14 weeks of WL [week 29 dotted line on the BW graph in (D)] with glucose (1.5 mg/g). (F) In vivo GSIS: Plasma insulin during GTT in (E) and pre-WL. (G and I) GTT with glucose (1.5 mg/g) and (H and J) plasma insulin during GTT (in vivo GSIS) in males after 4 weeks of WRG [dotted line at week 27 in (A); (G) and (H)] and in females after 10 weeks of WRG [dotted line at week 41 in (D); (I) and (J)]. (K and L) Following a year-long dietary regimen on either LFMP or KD, KD mice were switched to LFMP for 4 weeks. (K) BW before and after switching the diets. (L) GTT [glucose (1.5 mg/g)] before and 4 weeks after switching to LFMP. Statistics: One-way or two-way ANOVA with post hoc Tukey’s HSD. Number of symbols represents significance level: *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. Data are represented as means ± SEM. ns, not significant.