Blood sugar, appetite, and daily rhythm benefits of a dairy-rich diet with high-protein breakfast and early carb eating in type 2 diabetes

The study included 29 individuals (15 women and 14 men) aged 45–77 years, with a diagnosis of type 2 diabetes and HbA_1c ≥48 mmol/mol (6.5%). All were treated with stable doses (≥3 months) of oral glucose-lowering agents or were managed by diet. Additional inclusion criteria included stable weight (<5% change in the previous 3 months), normal liver, kidney and thyroid function, and eGFR>60 ml/min per 1.73 m². Concomitant medications (e.g. antihypertensive or lipid-lowering agents) were permitted if doses were stable for at least 3 months. Participants maintained a regular sleep–wake schedule (waking at 06:00–08:00 hours and sleeping at 22:00–24:00 hours). Exclusion criteria included shift work within 6 months of enrolment or recent travel across more than two time zones, type 1 diabetes, insulin therapy, active malignancy, use of psychotropic/steroid medications, substance abuse, pregnancy/lactation and known hypersensitivity to milk components. Participants were instructed to maintain their usual physical activity throughout the study. Recruitment was conducted via the Diabetes Unit at Wolfson Medical Center, regional diabetes conferences and support group meetings. All participants were Israeli, but detailed data on ethnicity, religious background and socioeconomic factors were not collected. Gender was self-reported and verified via the medical record. The study protocol was approved by the Helsinki Ethics Committee of Wolfson Medical Center (Holon, Israel) and written informed consent was obtained from all participants prior to enrolment. Recruitment began in late 2019, was paused due to the COVID-19 pandemic, and resumed after protocol amendments and re-approval. Recruitment was completed by July 2024 and the last visit occurred in October 2024. The study was registered at ClinicalTrials.gov (registration no. NCT03772067↗).

Both dietary interventions were isoenergetic with matched macronutrient content, meal timing and carbohydrate distribution, following a previously described protocol [26]. Energy intake was distributed as follows: a high-energy, carbohydrate-rich breakfast (45% of total intake); a large lunch (40% of total intake); and a low-energy, carbohydrate-free dinner (15% of total intake) (electronic supplementary material [ESM] Table 1). Carbohydrates were limited to breakfast and lunch; dinner carbohydrates were minimal and derived mainly from non-starchy vegetables (mostly attributed to dietary fibre). Calcium and fibre intake was comparable across both diets. The sole difference between diets was the protein source. Participants prepared their own meals at home according to individualised meal plans specifying exact food quantities, portion size and timing. All ingredients were purchased and prepared by the participants under twice-a-week supervision and dietary counselling by a registered dietitian. The YesMilk diet emphasised dairy-based protein sources, including at least one serving of fluid milk at breakfast, and additional dairy products throughout the day. The NoMilk diet included only non-dairy protein sources (eggs, poultry, fish, meat, legumes). A soy-based milk alternative was included in the meal plan at breakfast to match the fluid dairy component of the YesMilk diet. Prior to randomisation, a dietary anamnesis was performed by a registered dietitian and individual energy needs were estimated. Participants received dietary instructions aligned with an energy intake of approximately 7500–9200 kJ/day (1800–2200 kcal/day) (ESM Table 1).

During the first 2 weeks of each intervention phase, participants were closely monitored to ensure weight stability. Energy intake was adjusted as needed to maintain energy balance. Participants received tailored written guidelines detailing daily food quantities, individualised meal plans and a substitution list with energetic and nutritionally equivalent alternatives. They were instructed to record their daily food intake and meal timing, which were reviewed every 2 weeks at clinic visits. At least two weekly phone calls were conducted to assess real-time dietary intake. Dietary intake was assessed through food diaries, 24 h recalls, telephone interviews and in-person anamnesis. All dietary records were analysed using the Tsameret dietary software (Israeli Ministry of Health), based on national food composition data. Food-group servings were categorised according to national dietary guidelines. A second registered dietitian independently reviewed all records. Non-compliance was defined as follows: (1) >10% deviation from prescribed energy intake per meal/day; (2) failure to follow meal timing schedule or macronutrient distribution; or (3) consumption of a protein source inconsistent with the assigned diet. Participants non-compliant for >3 days/week were excluded from the study.

CGM was performed using FreeStyle Libre-2 sensors (Abbott, Alameda, CA, USA) applied at baseline, washout and every 2 weeks during each dietary phase (6 periods). A minimum of four complete 24 h recording days per phase was required for inclusion in the final data analysis, consistent with previous CGM studies [26, 27]. CGM-derived metrics included fasting glucose, 24 h mean glucose, glucose management indicator (GMI), time in range (TIR), and daily glucose fluctuations. Standard metrics, such as TIR and 24 h mean glucose, were calculated using the LibreView software (Abbott). In addition, a semi-automated algorithm developed specifically for this study was used to identify fasting glucose and postprandial glucose excursions (defined as ≥2.2 mmol/l increase from baseline) within the following predefined meal windows: 08:00–12:00 hours (breakfast); 12:00–16:00 hours (lunch); and 17:00–21:00 hours (dinner). The earliest daily increase per window was selected, according to previous CGM studies [28]. Outputs from our algorithm were cross-validated against metrics generated by Libre-derived data (ESM Fig. 1).

Peripheral blood was collected in Tempus RNA tubes (Applied Biosystems, Foster City, CA, USA), and total RNA was extracted according to the manufacturer’s instructions. One microgram of RNA was reverse-transcribed using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems). Quantitative PCR (qPCR) was performed using SYBR Green Master Mix (Thermo Fisher Scientific) on a QuantStudio 6 system. Relative expression was calculated using the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${2}^{-\Delta \Delta {\mathrm{C}}_{\mathrm{t}}}$$\end{document}2-ΔΔCt method and normalised to the reference gene actin. Candidate genes were selected a priori based on their central role in circadian regulation and metabolic control: BMAL1, REV-ERBα (also known as NR1D1), CRY1 and PER1. These genes represent key components of the core transcriptional–translational feedback loop of the circadian clock. Primer sequences spanning exon–exon boundaries are listed in ESM Table 2 as was previously reported [26]. Additional clock-related genes (CLOCK and RORα [also known as RORA]) were also examined but did not show statistically significant differences between diets. The primary outcome of the study was the change in circadian clock gene expression (BMAL1, REV-ERBα, CRY1, PER1) between the YesMilk and NoMilk dietary phases.

Appetite was assessed using validated 100 mm VAS, anchored by ‘not at all’ (0 mm) and ‘extremely’ (100 mm). Participants completed the VAS hourly from waking to bedtime, once weekly during baseline and once weekly during each dietary phase. Thus, each participant completed the full-day VAS assessment at baseline and four times for each period (YesMilk and NoMilk). This instrument has been previously validated for free-living circadian appetite monitoring [29].

The primary outcome was the change in circadian clock gene mRNA expression in response to the dietary interventions; this guided the sample size calculation. Secondary outcomes included glycaemic indices derived from CGM and subjective appetite scores. A sample size of n=19 participants completing both interventions was estimated to provide ≥80% power to detect a moderate-to-large within-participant difference of approximately 50%±30% in clock gene expression (α=0.05), based on prior data [10]. Post hoc analysis demonstrated >99% power to detect >50% differences in REV-ERBα and PER1 expression. The study was powered based on expected within-participant differences in circadian clock gene expression (primary outcome), as described above. Glycaemic and appetite outcomes were analysed as secondary endpoints without separate power estimation. To account for an anticipated dropout rate of ~20%, we aimed to recruit at least 24 participants. Twenty-nine individuals were screened and 25 were randomised, yielding an attrition rate consistent with similar studies [11, 26]. No changes were made to the primary or secondary outcomes after trial commencement.

Statistical analyses were performed using JMP Pro (18.0.2, SAS Institute, Cary, NC, USA). All analyses followed a per-protocol approach and included only participants who completed both dietary intervention phases (n=19). Given the crossover design, each participant served as his/her own control, ensuring complete paired data across conditions. Data distribution for each variable was assessed using histograms, boxplots and normal Q–Q plots. Homogeneity of variances was evaluated using Levene’s and Bartlett’s tests to guide selection of appropriate tests. Since this was a crossover design in which each participant served as their own control, all within- and between-diet comparisons were analysed using paired Student’s t tests (or paired Welch’s t tests when variances were unequal). Welch’s t test was used to compare means between groups, accounting for unequal variances; otherwise, Student’s t test (pooled variance) was applied. Tukey’s HSD test was applied for multiple time point comparisons within and between treatment arms. For time series, a one-way ANOVA was performed to analyse daily or 24 h patterns. A post hoc t test analysis was used for comparison between the YesMilk and NoMilk groups at each time point. A p value of <0.05 was considered statistically significant. Data are presented as mean ± SEM for outcome variables (unless stated otherwise) and mean ± SD for baseline characteristics. Linear mixed-model (LMM) analyses were performed, including treatment, time, period, and their interaction as fixed effects, and participant as a random intercept. Clock gene expression values were log-transformed prior to analysis. Full model outputs are presented in ESM Table 3. A significant period effect was observed for TIR and time above range (TAR) in the exploratory LMM, likely reflecting sequence imbalance and variability in CGM-derived data rather than biological carry-over. However, the absence of significant treatment × time interactions and the consistency with paired analyses substantiate the validity of the main results.

Dietary intake is summarised in ESM Table 4. Baseline food intake showed no differences in energy or macronutrient intake between the pre-YesMilk and pre-NoMilk conditions (p>0.05) (ESM Table 5). The YesMilk and NoMilk diets were isoenergetic and matched in macronutrient distribution and carbohydrate allocation per meal (ESM Tables 4, 5). The only notable difference was in calcium distribution, which was higher at breakfast in the NoMilk group and higher at dinner in the YesMilk group (p<0.05, between diets) (ESM Table 6), while total daily calcium intake remained comparable between diets (p=0.91) (ESM Table 4). Participants in the YesMilk group consumed approximately five servings of dairy protein per day, including ~1.5 servings of fluid dairy and ~3.5 servings of other dairy products (ESM Table 7). They also consumed ~5.5 servings of non-dairy protein, compared with ~10.5 servings in the NoMilk group, reflecting the substitution of dairy with other protein sources.

Meal composition analysis of the YesMilk diet intervention showed that ~73% of daily dairy protein intake occurred at breakfast, which consistently included at least one serving of liquid dairy products and was therefore rich in whey protein (ESM Table). Lunch contributed only ~4% of daily dairy protein intake, with protein intake at lunch being mostly non-dairy due to local cultural practices. The remaining daily dairy protein intake (~23%) was consumed at dinner, a carbohydrate-free meal composed mainly of casein-rich, low-lactose dairy products with minimal whey content. 8

The GMI, which estimates laboratory-measured HbA_1c from CGM data, decreased in both groups (Fig. 3c). YesMilk resulted in a greater reduction, from 6.8% to 6.25% after 2 weeks (p<0.001, within-diet, week 2 vs baseline), and to 6.06% at four weeks (p<0.001, within-diet, week 4 vs baseline), with statistically significant between-group differences at both time points (p=0.044 and p=0.0036, respectively, between diets, week 4) (Fig. 3c). Mean 24 h daily glucose levels declined in both groups (Fig. 3d). NoMilk mean glucose levels decreased from 7.9 mmol/l to 7.2 mmol/l after 2 weeks (p<0.001, within-diet, week 2 vs baseline), and to 6.9 mmol/l at 4 weeks (p=0.1, within-diet, week 4 vs week 2). YesMilk resulted in greater reductions, from 8.1 mmol/l to 6.8 mmol/l after 2 weeks (p<0.001, within-diet, week 2 vs baseline), and to 6.4 mmol/l after 4 weeks (p=0.05, within-diet, week 4 vs week 2). Between-group differences were statistically significant at both time points (p=0.0389 at week 2 and p=0.0043 at week 4). These outcomes were independently confirmed using a semi-automated algorithm developed specifically for this study (ESM Fig. 1).

Daily glucose fluctuations (24 h mean ± SEM) were similar at baseline, showing rhythmic daily fluctuations (p<0.0001) (Fig. 3e). After 2 weeks, glucose levels were lower in the YesMilk group than in the NoMilk group in the early morning (07:00–08:00 hours), early afternoon (14:00–15:00 hours) and early evening (18:00–20:00 hours) (p<0.05 between diets, at week 2) (Fig. 3f), with more pronounced reductions after 4 weeks in the mid-morning (09:00–11:00 hours), post-lunch (16:00–18:00 hours) and across the evening and night period (21:00–05:00 hours) (p<0.05 between diets, at week 4) (Fig. 3g). Both diet interventions maintained the rhythmic glucose fluctuations throughout the study (p<0.0001). The YesMilk diet improved glycaemic management not only after breakfast but also more prominently after lunch, dinner and during night-time. Glucose levels further decreased at 16:00–17:00 hours, 22:00–02:00 hours, and 03:00–05:00 hours by week 4 (p<0.05 within-diet, week 2 vs week 4 and p<0.0001 at week 4 vs baseline), with no notable changes in the NoMilk group (ESM Fig. 2a, b).

Pre- and postprandial glycaemic response analyses showed differences between the interventions, as pre-breakfast glucose levels were lower in the YesMilk group than in the NoMilk group, approaching statistical significance at 2 weeks (p=0.06, between diets) (Fig. 3h). This difference became statistically significant at 4 weeks (p=0.046, between diets) (Fig. 3i), congruent with the fasting glucose outcomes (Fig. 3b). Pre-lunch glucose levels were also lower at 2 weeks (p=0.001, between diets), with a near-significant difference before dinner (p=0.055, between diets) (Fig. 3h). Post-dinner glucose levels were ~1.1 mmol/l lower with the YesMilk diet than with the NoMilk diet at week 2 (p=0.034, between diets) (Fig. 3h). By week 4, the YesMilk group consistently showed lower glucose levels than the NoMilk group at the end of the postprandial window (3 h) after breakfast, lunch and dinner (p=0.003, p=0.033, p=0.053, respectively, between diets) (Fig. 3i). The most consistent and robust glycaemic improvement was observed during dinner (i.e. peak and late postprandial glucose values were statistically significantly reduced at both time points) (Fig. 3i).

The fullness rating increased statistically significantly only in the YesMilk group (p=0.003, within-diet, week 4 vs baseline) (Fig. 5c). Although fullness appeared to be rated higher in the YesMilk group at the end of the intervention compared with the NoMilk group (ESM Fig. 3c, d), the between-group difference did not reach statistical significance (p=0.08, between diets, at week 4) (Fig. 5c, d). The desire to eat gradually declined throughout the intervention in the YesMilk group (p=0.0002, within-diet, week 4 vs baseline), resulting in lower scores compared with the NoMilk group at both time points (p=0.0036, p<0.001, between-diet, at week 2 and week 4, respectively) (Fig. 5e, f and ESM Fig. 3e, f).

Prospective consumption (participants’ estimated capacity to eat at a given moment) showed a similar pattern, as scores declined after 2 weeks of YesMilk diet (p=0.05, within-diet, week 2 vs baseline) and further by week 4 (p<0.001, within-diet, week 4 vs baseline), resulting in overall lower daily values compared with the NoMilk group (p<0.001, between diets at week 4) (Fig. 5g, h and ESM Fig. 3g, h). The desire for sweets decreased in the YesMilk group at both time points compared with baseline (p=0.0007, within-diet at week 2; p<0.001, within-diet at week 4), while no comparable changes were observed in the NoMilk group (Fig. 5i and ESM Fig. 3i, j). Overall, hunger and sweet craving scores decreased by 15–20% (p<0.05).