Canagliflozin, an SGLT2 inhibitor, corrects glycemic dysregulation in TallyHO model of T2D but only partially prevents bone deficits

Abstract

Higher fracture risk in type 2 diabetes (T2D) is attributed to disease-specific deficits in micro-structural and material properties of bone, although the primary cause is not yet established. The TallyHO (TH) mouse is a polygenic model of early-onset T2D and obesity analogous to adolescent-onset T2D in humans. Due to incomplete penetrance of the phenotype, ~25% of male TH mice never develop hyperglycemia, providing a strain-matched, non-diabetic control. Utilizing this model of T2D, we examined the impact of glucose-lowering therapy with canagliflozin (CANA) on diabetic bone. Male TH mice with or without hyperglycemia (High BG, Low BG) were monitored from ~8 to 20 weeks of age, and compared to age-matched, male, TH mice treated with CANA from ~8 to 20 weeks of age. At 20 weeks, untreated TH mice with high BG [High BG: 687±106 mg/dL] exhibited lower body mass, decrements in cortical bone of the femur (decreased cross-sectional area and thickness; increased porosity) and in trabecular bone of the femur metaphysis and L6 vertebra (decreased bone volume fraction, thickness, and tissue mineral density), as well as decrements in cortical and vertebral bone strength (decreased yield force and ultimate force) when compared to untreated TH mice with low BG [Low BG: 290±98 mg/dL; p<0.0001]. CANA treatment was metabolically advantageous, normalizing body mass, BG and HbA1c to values comparable to the Low BG group. With drug-induced glycemic improvement, cortical area and thickness were significantly higher in the CANA than in the High BG group, but deficits in strength persisted with lower yield force and yield stress (partially independent of bone geometry) in the CANA group. Additionally, CANA only partially prevented the T2D-related loss in trabecular bone volume fraction. Taken together, these findings suggest that the ability of CANA to lower glucose and normalized glycemic control ameliorates diabetic bone disease but not fully.

1. Introduction

Humans with T2D have an increased risk for bone fracture (1, 2), a risk that is neither explained by individual bone mineral density (BMD) measurements (3) nor by an increased propensity to fall (4). Instead, both cortical and trabecular bone quality are impaired in T2D, more so in those with poorer glycemic control (5) and longer duration of disease (2). At present, little is known about the onset and natural history of skeletal changes in youth diagnosed early with T2D, or of the impact of T2D during the adolescent years which are critical to acquisition of peak bone mass (6). However, coincident with the global epidemic of obesity among the pediatric population (7), the prevalence of T2D in children and adolescents is also increasing (8), with the incidence of T2D estimated to increase at a rate of ~7% per year (9), disproportionately impacting ethnic minority groups (7). Unlike adults with T2D, youth-onset T2D also displays a more aggressive phenotype (7, 8), characterized by: 1) a more rapid decline in ß-cell function (10, 11); 2) a more rapid deterioration in glycemic control (11), requiring multi-drug therapy; and 3) an earlier onset and more rapid progression of both microvascular and macrovascular complications of all types (12, 13). With both an earlier onset and greater severity of disease, pediatric patients with T2D face the potential for lifelong exposure to antidiabetic medications. Therefore, understanding the impact of early-onset hyperglycemia and relevant pharmacotherapy on the skeletal health in youth-onset T2D is critically important.

TallyHO/JngJ (TH) mice represent a polygenic model of early-onset T2D and obesity (14) consistent, maturationally, with adolescent-onset T2D in humans. In male TH mice, rapid weight gain is apparent by 4 weeks of age. Body weight increases progressively, and by ~10-14 weeks of age, insulin resistance progressing to frank hyperglycemia [blood glucose (BG) >300 mg/dL] is demonstrable; hyperinsulinemia and hyperlipidemia are also present (15). When compared with a genetically similar, non-diabetic, control strain (SWR/J mice) (16), adult TH male mice also exhibit lower bone formation markers (17, 18), lower trabecular bone volume (17, 18), lower bone toughness (18), and reduced fracture resistance (18), suggesting that skeletal acquisition and bone integrity are impaired in this model of T2D. Such skeletal comparisons, however, are confounded by a >30% increase in body size and a >100% increase in percent body fat in the obese TH mice, compared with the non-obese SWR/J strain (17). Moreover, the observed reduction in bone toughness and fracture resistance of TH mice, compared with SWR/J mice, does not relate to duration of diabetes (18), questioning the role of hyperglycemia in these strain differences that may be due, in part, to differences in the genetic makeup of SWR/J and TH mice.

Interestingly, the penetrance of the polygenic TH phenotype is incomplete, with ~20-25% of TH male mice maintaining normal glucose levels in our hands, despite advancing age and excess weight gain. (Female TH mice also remain euglycemic throughout life.) By selectively identifying and studying a cohort of non-hyperglycemic, yet comparably obese male TH mice, the skeletal effects of juvenile-onset diabetes in TH mice can be assessed independently of the co-existent, early-onset obesity and accomplished using littermates (i.e., same genetic background). Moreover, the TH model is amenable to testing the efficacy of therapies to lower blood glucose (19). One such FDA-approved therapy, canagliflozin (CANA), is a selective sodium-dependent glucose co-transporter 2 inhibitor (SGLT2I), a class of anti-hyperglycemic medications known to effectively lower serum glucose in an insulin-independent manner by inhibiting renal glucose reabsorption and promoting enhanced urinary glucose excretion. Despite the effectiveness of SGLT2Is in controlling glycemia and in reducing the risk of cardiovascular disease (20), CANA has been associated with a higher-than-expected fracture risk among T2D patients with certain comorbidities (21), though recent reports have indicated that this association was a chance observation (22, 23). Since hypercalciuria accompanies excessive glucose excretion, SGLT2Is may also decrease the mineral content in the extracellular matrix of bone, thereby reducing bone strength in T2D. Therefore, we sought to evaluate the hypothesis that SGLT2I treatment of hyperglycemic TH mice would improve their glycemic control, but would not fully rectify the diabetes-related skeletal deficits.

2. Research Design and Methods

2.1. Animals and experimental design

Six-week old male mice, TallyHO/JngJ (JAX #5314) were obtained from Jackson Laboratories (Bar Harbor, ME). All mice were maintained in a 14-hour light (lights on at 0600 hours): 10-hour dark cycle (lights off at 2000 hours), and provided ad libitum access to food (Teklad global 18% protein rodent diet; 6.2% fat, 44.2% carbohydrate; Envigo Corp., Indianapolis, IN) and water throughout all studies, unless otherwise specified below. All animal procedures were approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Kentucky (#2015-2293). Additionally, all methods were carried out in accordance with relevant guidelines and regulations.

Following documentation of hyperglycemia, juvenile diabetic TallyHO/JngJ male mice (hereafter designated High BG; average pre-treatment blood glucose (BG) = 496±145 mg/dL; 8-10 weeks of age) along with age-matched non-converting, normal glucose TallyHO/JngJ male mice (designated Low BG; average pre-treatment BG = 245±51 mg/dL) were thereafter monitored from study start (8-10 weeks of age) until 20 weeks of age. Concurrently, a mixed population of TH male mice (also 8-10 weeks of age) were treated with canagliflozin for a duration of 10 weeks by consuming Teklad chow compounded with CANA at 100 ppm (designated CANA; average pre-treatment BG = 449 ± 179 mg/dL, in which 12 of 52 mice had pretreatment BG ≤250 mg/dL and 40 of 52 mice had pretreatment BG > 250 mg/dL); to also reach a comparable 20 weeks of age. CANA dosing (at 100 ppm) was calculated to approximate 20 mg/kg/day of drug exposure; this would be comparable to the human adult canagliflozin therapeutic dose of 100 mg/day (equaling 1.7 mg/kg for a 60 kg reference adult body weight), according to recommendations for interspecies dose extrapolation (24).

At study conclusion, any mouse initially assigned to the Low BG group, but later identified as “a converter” with documented hyperglycemia [by BG>400 mg/dL at study end; n=4] was removed from further analysis. To achieve sufficient cohort size in the Low BG group, this experimental design was replicated three times and data from all three experiments were pooled for analysis (Sample size for the three study replicates: High BG, n=7, n=16, or n=20 mice, respectively; Low BG, n=5/7/4 mice per study; CANA, n= 12/24/16 mice per study). The final cohort size for analysis (Table 1, Supplemental Table S1) included High BG, n=43 mice; Low BG, n=16 mice; CANA, n= 52 mice.

Table 1:

Effect of treatment (CANA) or high blood glucose (BG) on selected outcome measurements (mean ± SD) from the evaluations of femur mid-shafts (cortical bone) and distal femur metaphysis (trabecular bone). CANA-treatment normalizes a bone parameter if the CANA vs. Low BG comparison is not significant (NS). p<0.05 (p<0.1).

Parameter	Units	[A]	[B]	[C]	ANOVA	Adjusted p-value
		CANA	High BG	Low BG	p-value	CANA effect	CANA normalization	T2D effect
Cortical Bone		(n = 52)	(n = 43)	(n = 14-16)		A vs. B	A vs. C	B vs. C
Ma.V	---	1.17 ± 0.09	1.16 ± 0.10	1.10 ± 0.09	(0.0526)	NS	0.0462	(0.0886)
Ct.Ar	mm2	0.79 ± 0.04	0.76 ± 0.6	0.81 ± 0.04	0.0005	0.0103	NS	0.0011
Tt.Ar	mm2	1.43 ± 0.07	1.39 ± 0.07	1.41 ± 0.08	(0.0742)	(0.0592)	NS	NS
Imin	mm4	0.090 ± 0.010	0.085 ± 0.009	0.092 ± 0.010	0.0076	0.0166	NS	0.0368
Section modulus	mm3	0.157 ± 0.012	0.151 ± 0.013	0.158 ± 0.012	0.0297	0.0424	NS	NS
Ct.Th	mm	0.212 ± 0.010	0.205 ± 0.013	0.222 ± 0.009	<0.0001	0.0105	0.0073a	<0.0001a
Ct.TMD	mgHA/cm3	1282 ± 36	1268 ± 35	1289 ± 41	(0.0699)b	NS	NS	NS
Ct.Po	%	2.79 ± 0.31	2.95 ± 0.33	2.65 ± 0.34	0.0030	0.0378	NS	0.0047
Stiffness	N/mm	129 ± 15	122 ± 17	138 ± 15	0.0020	(0.070)	NS	0.0019
Yield force	N	17.6 ± 2.0	16.3 ± 2.3	19.3 ± 1.5	<0.0001	0.0091	0.0151	<0.0001
Ultimate force	N	20.5 ± 1.6	19.3 ± 2.3	21.4 ± 1.8	0.0025c	0.0332	NS	0.0048
PYD	mm	0.267 ± 0.152	0.250 ± 0.108	0.218 ± 0.121	NSb		No effects
Work-to-fracture	kJ/m2	6.18 ± 1.86	5.66 ± 1.45	5.95 ± 2.09	NSb		No effects
Yield stress	MPa	228 ± 25	220 ± 24	249 ± 20	0.0004	NS	0.0087	0.0002
Ultimate stress	MPa	262 ± 15	257 ± 24	271 ± 16	0.0486c	NS	NS	0.0430
Modulus	GPa	15.3 ± 1.7	15.4 ± 1.9	16.1 ± 1.5	NS		No effects
Toughness	MJ/m3	2.96 ± 0.86	2.84 ± 0.75	2.76 ± 0.94	NSb		No effects
PY toughness	MJ/m3	1.99 ± 0.85	1.98 ± 0.75	1.73 ± 0.92	NSb		No effects
Trabecular Bone
Tb.BV/TV	%	5.9 ± 1.6	4.6 ± 1.2	7.6 ± 1.8	<0.0001b	0.0006	0.0142	<0.0001
Conn.D	mm−3	72.8 ± 26.6	76.5 ± 23.6	80.5 ± 22.9	NS		No effects
SMI	0:plates; 3:rods	2.76 ± 0.26	2.83 ± 0.20	2.56 ± 0.24	0.0007	NS	0.0111	0.0004
Tb.N	mm−1	3.29 ± 0.36	3.32 ± 0.29	3.41 ± 0.31	NS		No effects
Tb.Th	mm	0.042 ± 0.005	0.036 ± 0.004	0.046 ± 0.004	<0.0001b	<0.0001	NS	<0.0001
Tb.Sp	mm	0.309 ± 0.036	0.303 ± 0.028	0.297 ± 0.028	NSb		No effects
Tb.TMD	mgHA/cm3	983.5 ± 40.8	959.0 ± 39.7	984.9 ± 50.3	0.0003b	0.0003	NS	(0.058)

^aP-value from Dunn’s test because only Ct.Th of TH – Low BG did not pass normality, but the residuals did pass normality.

^bP-value from Kruskal-Wallis test because residuals did not pass normality

^cP-value from Kruskal-Wallis test because the SDs were significantly different among the 3 groups

2.1.1. Sample size estimation

Based on the mean and SD values of yield force for untreated T1D mice and T1D mice treated with CANA in our previous study (25), we used a standardized mean difference (Cohen’s d) of 1.0951 in the estimate of the sample size. With an allocation ratio of 4 (penetrance estimate of 4 times more High BG mice than Low BG mice), an α error probability of 0.05, and a power (1 – β error probability) of 0.9, 12 mice in the Low BG group and 46 mice in the High BG group would be required to detect a difference in yield force that was similar to the difference in our previous CANA study involving a mouse model of T1D (25). Because of the uncertainty in which mouse would maintain High BG at the time of allocation to treatment (Rx) or no Rx during the replicate studies, our final sample size was similar but not exact. Nonetheless, by having a total sample size of 109-111 mice, we were powered (0.90) to detect an effect size f of 0.3457 (Cohen’s d = 0.6912) among 3 groups with 1 covariate at α=0.05. These power calculations were done using G*Power (v3.1.9.3, Heinrich Heine University Düsseldorf) (26, 27). The Cohen’s d for BV/TV in the previous T1D-CANA study was greater than the Cohen’s d to power the present study.

2.2. Assessment of physiology and phenotype

For all experiments, weekly (experimental replicate 1), or biweekly (experimental replicates 2, 3) animal assessments included measurement of individual body weight, food and water consumption, and random blood glucose concentration (so as to document glycemic status). Blood glucose (BG) was measured from tail vein sampling using a glucometer (AlphaTRAK 2 Veterinary Blood Glucose Monitoring System; Zoetis, Parsippany, NJ). During the 9-10^th week of treatment (mouse age ~19 weeks) mice from all 3 cohorts were transferred to individual metabolic cages for urine collection. Urine was collected over defined intervals (range 6-18 hours for experimental replicates 1 and 2; 24-hour collection for replicate 3); for each experiment, total urine volume and collection times were recorded. Using an aliquot of each collection, urine calcium and phosphate concentrations were measured using calcium and phosphate colorimetric assays (Sigma-Aldrich Corporation, St. Louis, MO; #MAK022 and #MAK308, respectively) and urine creatinine was measured using an alkaline picrate chemical assay (Exocell, Inc., Philadelphia, PA; The Creatinine Companion, #1012). Calcium and phosphate concentrations were normalized to urine creatinine concentration, and reported as a urine calcium/creatinine ratio (CA/Cr) or urine phosphate/creatinine ratio (PHOS/Cr). In the third experimental replicate, a 24-hour urine collection was used to quantify total daily urine excretion of CA, PHOS, or glucose, reported as μg/g/day for calcium or phosphate and as mg/g/day for glucose, corrected to individual mouse body mass (g). Urine glucose for this parameter was quantified using the Mouse Crystal Chem glucose enzymatic assay (Elk Grove Village, IL; #81692).

At study end, mice were euthanized by isoflurane overdose followed by decapitation, and trunk blood was collected directly into 1.5 ml Eppendorf tubes on ice. Physiological assessments (diabetes parameters and bone biomarkers) and ex vivo analyses of bone phenotype [high-resolution micro-computed tomography (μCT) and biomechanical testing] were completed on all mice, as described in subsequent sections.

2.3. Serum Assays

At study end, fasting blood glucose (4-5 hour fast with free access to water) was measured and terminal Hemoglobin A1c [HbA1c; a long-term estimate of daily averaged blood glucose; (28)] was quantified using a mouse HbA1c whole blood assay (Crystal Chem; Downers Grove, IL, #80310). Fasting serum insulin levels were also quantified at study end, using an ultrasensitive mouse insulin ELISA (Crystal Chem: #90080). As a marker of bone formation, procollagen type 1 N-terminal propeptide (P1NP) was measured in serum at sacrifice, using the Rat/Mouse P1NP Enzyme immunoassay (Immunodiagnostics Systems, Inc., Fountain Hills, AZ; #AC-33F1). As a marker of bone resorption, C-terminal telopeptides of type I collagen (CTX) were measured in serum using the RatLAPs ELISA (Immunodiagnostics Systems, Inc., Fountain Hills, AZ; #AC-06F1).

2.4. Assessment of Skeletal Microarchitecture

At study end, left femurs and lumbar vertebrae were harvested, frozen in phosphate buffered saline (PBS) and stored at −20°C until analysis. Femur length (neck to condyle groove) was measured using calipers. For bone microarchitecture analyses, the mid-shaft and distal metaphysis regions of the femurs or centrum of the L6 vertebral bodies (VB) were scanned with a micro-computed tomography (μCT) scanner (Scanco Medical μCT50, Brϋttisellen, Switzerland): isotropic voxel size of 6.0 μm (12.0 μm for VB), peak x-ray energy setting of 70 kVp/0.114 mA (55 kVp/0.200 mA for VB), integration time of 300 ms (600 ms for VB), and a sampling rate of 1162 samples (500 for VB) per 1000 projections during a full rotation of the tube holder. Each scan was calibrated to the manufacturer provided QC standard [a hydroxyapatite (HA) phantom with peak density of 1200 mgHA/cm³] to account for any beam hardening artefacts, as previously described (29). The regions of interest (ROI) included the trabecular bone of the metaphysis (3.72 mm segment above the growth plate of the distal femur) and between the VB endplates (1.40-2.25 mm segment of the VB) as well as the cortical bone of the diaphysis (1.86 mm segment centered at the mid-point of the femur).

Each ROI was analyzed after the scan reconstruction using a custom auto-contour program in Scanco Image Processing (IPL) Software v6.6 to segment out the bone against the background. The entire image stack (310 axial slices) of the femur mid-shaft was contoured to capture the cortical bone properties, while for the image stack of the distal femur (350 slices after excluding 50 slices proximal to primary spongiosa) and L6VB (115-185 slices after excluding endplates) only the trabecular bone was included. Using our previously published methods for contouring these regions (18, 29) and then applying a Gaussian noise filter (sigma of 0.8 with support of 2 for cortical bone and sigma=0.2, support=1 for trabecular bone) and a global threshold (429.4 mgHA/cm³ and 900.5 mgHA/cm³ for trabecular and cortical bone, respectively), each ROI was evaluated using standard Scanco evaluations scripts IPL program applied to determine the architectural and structural properties of trabecular and cortical bone. Also, the femur mid-shafts dual contour (encompassing only the intra-cortical region) was used to estimate the porosity by subtracting its cortical bone volume fraction from unity.

2.5. Assessment of Bone Biomechanical Properties

After μCT analysis, each hydrated left femur mid-shaft was loaded to failure at 3 mm/min in three-point bending with a span of 8 mm, following our previously described procedures (30, 31). The stiffness (slope of the linear portion of the curve), yield force (0.2% offset method), and ultimate force (maximum or peak load endured by the bone before fracture) were determined from the resulting force vs. displacement curve. These structural-dependent properties were converted to estimates of material properties (modulus, yield stress, and ultimate stress, respectively) using the flexural equations from Euler-Bernoulli beam theory and the geometrical properties of each femur mid-shaft from the μCT evaluations, namely the moment of inertia for the bending orientation and the distance between the neutral axis and the periosteal surface in the direction of loading (29). Because the flexural equations are less valid after the bone tissue yields, we estimated the toughness and post-yield toughness as 3 × W_f / Ct.Ar / Span where Wf is the work-to-fracture (area under the entire force vs. displacement curve) or the post-yield work-to-fracture (area under the force vs. displacement curve between the yield point and failure point) and Ct.Ar is the bone cross-sectional area of the femur mid-shaft (32). Each hydrated L6 VB was loaded at 3.0 mm/min in compression to determine the ultimate compressive force as the measurement of whole-bone strength (18).

2.6. Statistical Analysis

Outcomes of interest were measured as pre-treatment (pre-treatment body mass, blood glucose), near end-of-experiment (urine CA/Cr, CA/day; urine PHOS/Cr, PHOS/day; urine glucose/day) or at the time of death for each mouse (post-treatment blood glucose and all other parameters). For these quantitative variables, group-level means and standard deviations (SD) were calculated as well as group-level medians and first and third quartiles when warranted. The approximate normality of data was checked prior to data analyses using the Anderson-Darling test. For normally distributed variables with homoscedasticity across groups, analysis of variance (ANOVA) was employed. If the group means were significantly different, the 3 pairwise comparisons of groups were subjected to Tukey’s multiple comparison test. For variables not normally distributed or heteroscedastic, Kruskal-Wallis non-parametric testing was employed, followed by Dunn’s multiple comparison test. Statistically significant group differences were defined as p≤0.05 unless otherwise stated in table or figure legends. A Spearman’s rank correlation coefficient was also calculated between variables of interest, given that most correlations were either heteroscedastic or not normally distributed. In a sub-group analysis comparing daily mineral excretion to daily glucose excretion for untreated and treated mice, the residuals from the ordinary least squares regression either did not pass the Anderson-Darling normality test or the runs test for homoscedasticity. Therefore, the data was bootstrapped with 1000 replicates to determine the p-values for the slope of the regression line(s) and the interaction between glucose excretion and group (untreated or CANA-treated). To determine whether group-level differences were independent of final body mass and whether the linear relationship between yield force (or BV/TV) depended on treatment, we used analysis of covariance, bootstrapping with 1000 replicates when the residuals did not pass the Anderson-Darling test or were heteroscedastic. Analyses of the group differences, Spearman correlations, and linear regressions were completed using GraphPad Prism 8.1.1 statistical software (San Diego, CA). To bootstrap the data for estimating the p-values of the regression coefficients, we used Stata 11.0 data analysis (College Station, TX).

3. Results

3.1. Physiology:

3.1.1. A comparison of mice with hyperglycemia vs. normoglycemia confirms differences in body mass and metabolic phenotype, by 20 weeks of age.

At the start of the experimental period, pre-treatment measurements of body mass for the 3 groups were similar (Fig. 1A; CANA: 33.3 ± 2.6 gm; High BG: 33.5 ± 2.5 gm; Low BG: 31.9 ± 3.3 gm; ANOVA p=0.1304). Additionally, at study start, mice in the High BG and CANA groups had similarly high pre-treatment blood glucose levels (Fig. 1C), compared with the designated Low BG group. Over the next 10 weeks, untreated hyperglycemic TH mice (High BG) demonstrated a failure to gain weight (Fig. 1A compared to Fig. 1B and Fig. S1 in supplemental materials); and, as shown in Figure 1, at study end, in addition to the ~15% lower body mass, High BG mice maintained a significantly higher blood glucose (Fig. 1D and Table S1 in supplemental materials), along with an elevated HbA1c (Fig. 1F) and a lower serum insulin concentration (Fig. 1E), compared with the non-diabetic cohort (Low BG). By comparison, following 10 weeks of treatment with canagliflozin (CANA), body mass (Fig. 1B), post-treatment BG (Fig. 1D) and HbA1c values (Fig. 1F) normalized, and serum insulin levels improved (Fig. 1E), compared with the High BG group (Table S1). Final femur lengths, however, were not significantly different among the 3 groups (CANA: 14.94 ± 0.33 mm; High BG: 15.01 ± 0.20 mm; Low BG: 14.87 ± 0.17 mm; Kruskal-Wallis test p=0.0624) indicating comparable linear bone growth across the groups by 20 weeks of age.

Figure 1: Differences in selected physiological measurements at pre-treatment or post-treatment.

Pre-treatment body mass was similar among the 3 groups (A); 10-12-weeks later, untreated mice with high blood glucose (BG) weighed less than CANA-treated mice or mice with low blood glucose (B). Average pre-treatment BG was significantly elevated in both the CANA-treated and High BG groups, compared with the designated Low BG group (C); 10-12-weeks later, BG was similar between CANA-treated mice and Low BG mice (D). [Fig. 1D shows 1 outlier for the High BG cohort, with a post-treatment BG of 250 mg/dL. For this mouse, BG at the time of group assignment was 493 mg/dL with subsequent measurements of 505, 363, 399, and 368 mg/dL throughout the study. Because the average BG for study duration was 394 mg/dL, this mouse was retained in the High BG group despite an inconsistent final BG reading.] Accompanying the reduction in BG, circulating insulin (E) and HbA1c (F), a marker of glycemic control, was higher and lower, respectively, for CANA-treated than untreated mice with high BG. Group mean ± SD is shown. **** p<0.0001.

Individual values for HbA1c and post-treatment BG were correlated (Table S3; Spearman r = 0.689, p<0.0001), demonstrating consistency across these two measures of glycemic control. Additionally, individual values for final body mass were inversely correlated with both HbA1c (r = −0.676, p<0.0001) and post-treatment BG (r = −0.734, p<0.0001), supporting the expectation that chronic, untreated hyperglycemia, where present, contributed to the poor weight gain across time.

3.1.2. Hypercalciuria, hyperphosphaturia, and reduced bone formation are evident in diabetic mice; these abnormalities persist in CANA-treated mice despite a drug-related glycemic correction.

At study end, High BG mice had significantly higher urine calcium (CA) and urine phosphate (PHOS) excretion, compared with the Low BG group (Table S1; Fig. 2A and 2C, respectively). However, unlike the effects of CANA on glucose control, urinary mineral loss did not normalize with CANA. Instead, daily calcium excretion (urine CA/day) was highly correlated with daily glucose excretion (urine glucose/day), regardless of whether glucosuria was the result of untreated diabetes or of SGLT2I exposure (Fig. 2B). Daily phosphate excretion was also highly correlated with daily glucose excretion, although CANA treatment affected the relationship (Fig. 2D) such that there was less urine PHOS/day for a given urine glucose/day in CANA-treated mice, relative to untreated mice with hyperglycemia.

Figure 2: Differences in urine and serum bone biomarkers.

Urinary excretion of calcium (CA) normalized to creatinine (Cr), measured at study end, was elevated in CANA-treated mice and untreated High BG mice, compared to Low BG mice (A). Examined in a subset of mice, the daily excretion of CA was directly proportional to daily excretion of glucose (B) with one linear curve being required to explain the relationship for both untreated and treated mice (i.e., the interaction term was not significant, p=0.1960, and therefore removed). Urinary excretion of phosphorous (PHOS/Cr) was also higher in the CANA-treated mice and untreated High BG mice, compared with Low BG mice (C). In the same subset of mice, the daily excretion of PHOS was directly proportional to daily excretion of glucose (D), but the linear relationship was significantly different between untreated and treated mice (i.e., the interaction term was significant, p<0.0005). The serum marker of bone formation (P1NP) was reduced in the High BG group (E). There were no differences in the serum marker of bone resorption (CTX) between groups (F). Group median with the intra quartile range is shown. **** p<0.0001; ** p<0.01; *p≤0.05; and dashed line when adjusted p-value was greater than 0.05 but less than 0.1. In the linear regression graphs (B, D), closed symbols indicate untreated mice and the open triangle indicate CANA-treated mice.

Compared to the Low BG group, the mean for P1NP (i.e., the N-terminal propeptide) a marker of bone formation, was lower for both the CANA group and the untreated High BG group. (Table S1; Fig. 2E). However, CANA improved bone formation as P1NP was significantly higher in the CANA group compared with the High BG group. Consistent with these between group differences for bone formation, P1NP was also inversely correlated with HbA1c (Table S3, r= −0.452, p<0.0001) and with post-treatment BG (r= −0.473, p<0.0001), supporting the finding that bone formation is reduced in diabetic mice.

Mean values for CTX (i.e., the C-terminal telopeptide), a marker of bone resorption, were not significantly different between the groups (Table S1, Fig. 2F).

3.2. Skeletal Phenotype

3.2.1. Micro-computed tomography (μCT) analysis of the femur diaphysis (cortical bone) demonstrates deficits in cortical bone with diabetes, which are only partially corrected with CANA treatment.

When comparing structural properties of the diaphysis between the High BG vs. Low BG groups (Table 1), hyperglycemic mice exhibited a significant reduction in cortical area (Ct.Ar) and cortical thickness (Ct.Th), and in moment of inertia about the minimum axes (I_min), along with an increase in our estimate of cortical porosity (Ct.Po), as evidence of a diabetic bone phenotype in the TH model.

Mice treated with CANA showed a normalization of Ct.Ar and I_min after 10 weeks of treatment (Table 1). However, the lower Ct.Th and higher medullary volume (Ma.V) seen in High BG mice persisted in CANA-treated mice. Since mice with High BG did not gain weight while CANA treatment improved weight gain (Fig. 1B), the effect of T2D and the effect of CANA on these structural properties of bone could be due to differences in weight gain or lack thereof. When final body mass (BM) was included as a covariate in the analysis (Table S2), only Ct.Ar and Ct.Th depended on group (CANA, High BG, or Low BG). In effect, Ct.Ar and Ct.Th remained lower for CANA-treated mice than for untreated High BG mice, when accounting for the positive contribution of BM to these bone properties (Table S2). In contrast, the variance in I_min and Ct.Po was explained by BM, without any significant contribution from group (Table S2).

3.2.2. Micro-computed tomography (μCT) analysis of the femur metaphysis (trabecular bone) demonstrates deficits in trabecular bone with diabetes, which persist despite CANA treatment.

Hyperglycemic TH mice also exhibited significant deficits in trabecular bone (Table 1 and Fig. 3A, 3B), as a component of the diabetic bone phenotype. Specifically, reductions in trabecular bone volume fraction (Tb.BV/TV) and trabecular thickness (Tb.Th), along with an increase in structural model index (SMI) indicating a more rod-like characteristic, were apparent when comparing High BG and Low BG mice (Table 1 and Fig. 3C).

Figure 3: Differences in trabecular bone architecture and mineral density.

Representative micro-computed tomography evaluations of distal femur metaphysis are shown for the 3 groups (A) along with 3D renderings of the compartment with blue indicating low trabecular thickness and red indicating high trabecular thickness (B). There was less trabecular bone volume fraction (BV/TV) in the High BG group compared to the Low BG group and CANA group; CANA-treatment partially rescued BV/TV (C). Tissue mineral density of the trabecular bone (Tb.TMD) was also less in the High BG group compared to the other groups with no difference between CANA-treated and Low BG mice (D). Group median with the intra quartile range is shown. **** p<0.0001; ***p<0.001; *p<0.05; and dashed line when adjusted p-value was greater than 0.05 but less than 0.1.

While a trend toward improvement in trabecular parameters was seen in the CANA group (Table 1; significant differences noted between CANA and High BG), at study end Tb.BV/TV and SMI did not normalize with CANA treatment, but remained significantly different from the Low BG group, despite the period of improved metabolic control with CANA treatment. When adjusting for final body mass, group still significantly explained some of the variance in Tb.BV/TV, SMI, and Tb.Th (Table S2), but CANA treatment only affected Tb.Th (higher compared to High BG). In the multivariable regressions, mice with Low BG had significantly higher Tb.BV/TV and Tb.Th as well as lower SMI (Table S2) compared to mice with High BG (i.e., persistent T2D).

3.2.3. Micro-computed tomography (μCT) analysis of the L6 vertebra (trabecular bone) also demonstrates deficits in trabecular bone with diabetes, which persist despite CANA treatment.

Similar to the trabecular bone of the femur, deficits in trabecular bone of the L6 vertebra were noted with diabetes, including reductions in Tb.BV/TV, Tb.Th, and Tb.TMD along with an increase in SMI, when comparing High BG and Low BG mice (Table 2). And again, a trend toward improvement in these trabecular parameters was noted in the CANA group, but CANA treatment did not normalize the L6 VB trabecular bone (Table 2).

Table 2:

Effect of treatment (CANA) or high blood glucose (BG) on selected outcome measurements (mean ± SD) from the analysis of the L6 vertebral body (trabecular bone with a thin cortical shell). CANA-treatment normalizes a bone parameter if the CANA vs. Low BG comparison is not significant (NS). p<0.05 (p<0.1).

Parameter	Units	[A]	[B]	[C]	ANOVA	Adjusted p-value
		CANA	High BG	Low BG	p-value	CANA effect	CANA normalization	T2D effect
Centrum		(n = 52)	(n = 43)	(n = 14-16)		A vs. B	A vs. C	B vs. C
Tb.BV/TV	%	25.4 ± 3.0	22.9 ± 3.5	28.8 ± 2.3	<0.0001	0.0049a	0.0031a	<0.0001
Conn.D	mm−3	213.3 ± 36.8	232.6 ± 21.8	210.1 ± 24.0	0.0038b	0.0136	NS	0.0201
SMI	0:plates; 3:rods	0.80 ± 0.28	1.06 ± 0.31	0.49 ± 0.18	<0.0001	<0.0001	0.0002	<0.0001
Tb.N	mm−1	4.74 ± 0.33	4.81 ± 0.32	4.84 ± 0.24	NSb		No effects
Tb.Th	mm	0.056 ± 0.004	0.052 ± 0.005	0.061 ± 0.003	<0.0001c	0.0006	0.0024	<0.0001
Tb.Sp	mm	0.209 ± 0.017	0.205 ± 0.015	0.205 ± 0.013	NSb		No effects
Tb.TMD	mgHA/cm3	876.0 ± 17.0	862.7 ± 22.3	891.5 ± 10.2	<0.0001c	0.0167	0.0055	<0.0001
Ax.Led	mm	1.80 ± 0.17	1.76 ± 0.15	1.85 ± 0.26	NSa		No effects
Whole Bone
Yield force	N	21.8 ± 6.6	15.5 ± 4.5	27.9 ± 7.3	<0.0001c	<0.0001	(0.0594)	<0.0001
Ultimate force	N	26.0 ± 7.1	19.5 ± 5.4	32.1 ± 6.4	<0.0001	<0.0001	0.0011	<0.0001

^aP-value from Dunn’s test because only BV/TV of CANA did not pass normality, but the residuals did pass normality.

^bP-value from Kruskal-Wallis test because residuals did not pass normality.

^cP-value from Kruskal-Wallis test because the SDs were significantly different among the 3 groups.

^dAxial length (Ax.Le) is the distance between endplates as determined by μCT evaluations.

3.2.4. Biomechanical testing demonstrates deficits in bone strength with diabetes.

Biomechanical testing of the femur demonstrated a reduction in both bone stiffness and bone strength in High BG mice compared with Low BG mice (Table 1), as indicated by significant reductions in stiffness (Table 1), yield force (Fig. 4A), ultimate force (Table 1), yield stress (Fig. 4B), and ultimate stress (Table 1) with T2D. In contrast, overall resistance to fracture (work-to-fracture, toughness, and post-yield toughness) was not significantly different between High BG and Low BG mice (Table 1). Biomechanical testing of the L6 VB also demonstrated reductions in bone strength (yield force, ultimate force) with T2D (Table 2).

Figure 4: Differences in bending strength.

The force at yielding (proportional limit when permanent damage occurs) was highest in the Low BG group and lowest in the High BG group (A). The stress at yielding (partially independent of bone structure) was also highest for the Low BG group, compared with the other two groups. (B). ****p<0.0001; ***p<0.001; ** p<0.01; *p≤0.05.

In CANA treated mice, despite metabolic improvement with drug therapy, neither of the material strength estimates (yield stress or ultimate stress) of the femur improved; however, some increase in yield force (Fig. 4A) and ultimate force (Table 1) was apparent when comparing the CANA-treated mice to untreated High BG mice. When accounting for differences in final body mass between High BG and CANA and between High BG and Low BG, group was still a significant factor affecting yield force and yield stress, but the coefficients for the CANA effect or the T2D effect were not strictly significant. The effect of hyperglycemia on the strength measurements was not independent of its negative effect on weight gain (Table S2).

Similar to the femur, some improvement in strength of the L6VB was noted when comparing the CANA-treated mice to untreated High BG mice, although drug therapy did not normalize these parameters (Table 2).

3.2.5. The relationships between yield force and several outcomes differ between non-treated TH mice and CANA-treated TH mice.

To identify potential reasons for the treatment effect on yield force, we performed linear regressions pooling High BG and Low BG mice as an untreated group (no CANA) and included CANA treatment (Rx) as a potential explanatory factor. As expected, yield force was positively associated with body mass, but Rx was a significant factor such that for a given BM, mice treated with CANA tended to have weaker femur mid-shafts in bending than untreated mice (Table 3 and Fig. S2). Yield force was negatively associated with blood glucose, but again Rx was a significant factor such that this measure of whole bone bending strength tended to be less in the treated than in non-treated mice for a given BG. With respect to HbA1c as a potential predictor, the interaction term was significant (Table 3) indicating that the negative association between glycemic control and yield force was only significant for the untreated mice (slope = −0.8 N/%) but much less so for treated mice (slope = −0.002 N/%). For all other potential predictors – negative associations for CTX, uPHOS/Cr, and Ct.Po and positive associations for P1NP, section modulus or I_min/c_min, Ct.Ar, and Ct.Th – Rx was not a significant factor suggesting that the CANA-related increase in P1NP, section modulus, Ct.Ar, and Ct.Th improved bone strength. Likewise, the inability of CANA to improve uPHOS/Cr and CTX had a negative effect on yield force. Lastly, CANA treatment reduced the strength of the already weak negative association between uCA/Cr and yield force (slope = −0.001 N/μg/mg for untreated mice and slope = −0.0002 N/μg/mg).

Table 3:

Coefficients and p-values from linear regression models in which yield force was the dependent variable.

Model	Ref if not Rx	p-value	Add to Ref if Rx a	p-value	Slope	p-value	Slope if Rx b	p-value	Adj-R2
Rx + Body mass	+2.57	0.1388	−1.27	0.0020	+0.45	<0.0001	NI		0.414
Rx + Glucose	+21.58	<0.0001	−1.61	0.0015	−0.007	<0.0001	NI		0.293
Rx + HbA1c	+23.3	<0.0001	−5.3	0.0030	−0.8	<0.0001	−0.002	0.0130	0.257
Rx + Insulin c	+16.94	<0.0001	zero	0.6500	+0.19	0.0069	NI		0.063
Rx + P1NP d	+16.09	<0.0005	zero	0.246	+0.24	<0.0005	NI		0.309
Rx + CTX c	+18.12	<0.0001	zero	0.5131	−0.10	0.0128	NI		0.051
Rx + uCA/CR	+18.64	<0.0001	zero	0.5005	−0.001	<0.0001	−0.0002	0.0047	0.142
Rx + uPhos/CR	+17.95	<0.0001	zero	0.1287	−0.023	0.0005	NI		0.123
Rx + Imin/cmin	+2.2	0.3633	zero	0.9103	+99.7	<0.0001	NI		0.272
Rx + Ct.Ar	−8.3	0.0005	zero	0.7698	+33.3	<0.0001	NI		0.544
Rx + Ct.Th	−13	<0.0001	zero	0.6894	+146	<0.0001	NI		0.611
Rx + Ct.Po	+29.0	<0.0001	zero	0.5960	−4.1	<0.0001	NI		0.338

^aIf zero, the contribution of the coefficient for treatment (Rx) to the prediction of yield force is negligible (not significant). Ref is the intercept term.

^bThe interaction term was not included (NI) if it was not significant. When included, the coefficient of the slope is different between untreated mice and the CANA-treated mice.

^cThe regression model explaining the variance in yield force is not robust (p<0.025 instead of p≤0.0006 like the other models)

^dP-values were generated from bootstrapped regression with 1000 replicates because residuals were heteroscedastic.

3.2.6. The relationships between Tb.BV/TV and several outcomes differ between non-treated TH mice and CANA-treated TH mice.

Treatment with CANA also affected relationships between trabecular bone volume fraction in the distal femur metaphysis and several outcome measurements (Table 4). It reduced the slope of the linear relationship between BV/TV and final body mass and caused a negative offset of the relationship between BV/TV and BG as well as between BV/TV and HbA1c. Essentially, a negative association exists between glycemic control and trabecular bone, and so the CANA-related improvement in HbA1c and BG helps preserve BV/TV but not fully, as evident by the significant coefficient for Rx. There were positive associations with P1NP and insulin; and CANA had little effect on their linear relationship with BV/TV. The linear relationships between BV/TV and a resorption marker or mineral excretion were not significant or very weak (Table 4).

Table 4:

Coefficients and p-values from linear regression models in which trabecular bone volume fraction (BV/TV) was the dependent variable.

Model	Ref if not Rx	p-value	Add to Ref if Rx a	p-value	Slope	p-value	Slope if Rx b	p-value	Adj-R2
Rx + Body mass d	−7.48	0.002	+6.63	0.050	+0.39	<0.0005	+0.18	0.036	0.335
Rx + Glucose d	+9.18	<0.0005	−1.41	0.001	−0.007	<0.0005	NI		0.402
Rx + HbA1c d	+9.51	<0.0005	−3.00	0.028	−0.58	<0.0001	−0.11	0.062	0.221
Rx + Insulin dc	+4.97	<0.0005	zero	0.381	+0.17	<0.0005	NI		0.093
Rx + P1NP d	+4.32	<0.0005	+0.47	0.094	+0.20	<0.0005	NI		0.365
Rx + CTX dc	+6.02	<0.0005	zero	0.292	−0.08	<0.0005	NI		0.065
Rx + uCA/CR dc	+5.77	<0.0005	+0.63	0.059	Zero	0.148	NI		0.072
Rx + uPhos/CR dc	+5.77	<0.0005	+0.58	0.086	zero	0.119	NI		0.084

^aIf zero, the contribution of the coefficient to the prediction of yield force is negligible (not significant).

^bThe interaction between the categorical variable (Rx) and the continuous variable was not included (NI) if it was not significant. When included, the coefficient of the slope is different between untreated mice and the Rx-treated mice.

^cThe regression model explaining the variance in yield force is weak (p<0.025 instead of p≤0.0006 like the other models)

^dP-values were generated from bootstrapped regression with 100 replicates because residuals were heteroscedastic.

3.2.7. Correlations between metabolic parameters and bone parameters reveal different relationships when examined across all untreated mice (Table 5) vs. all CANA-treated mice (Table 6).

Table 5.

Spearman r between metabolic properties vs. bone properties pooling data from untreated mice (High BG + Low BG groups).

N = 51 to 59. Bold font: r ≥ ∣0.6. n.s. = not significant.

X property	Glucose		HbA1c		Insulin		P1NP		CTX		uCA/Cr		uPHOS/Cr
Y property	r	p-Value	r	p-Value	r	p-Value	r	p-Value	r	p-Value	r	p-Value	r	p-Value
Ma.V	n.s.	0.3434	n.s.	0.1865	n.s.	0.6010	n.s.	0.2538	0.276	0.0342	n.s.	0.1708	0.413	0.0019
Ct.Ar	−0.453	0.0003	−0.570	< 0.0001	0.357	0.0069	0.544	< 0.0001	n.s.	0.4325	−0.431	0.0007	−0.420	0.0016
Tt.Ar	n.s.	0.1233	−0.312	0.0258	n.s.	0.1354	0.305	0.0189	n.s.	0.9692	n.s.	0.1684	n.s.	0.5498
Imin	−0.329	0.0108	−0.433	0.0015	0.385	0.0034	0.457	0.0003	n.s.	0.7817	−0.299	0.0227	−0.306	0.0244
Imin/cmin	−0.278	0.0329	−0.420	0.0022	0.343	0.0096	0.462	0.0002	n.s.	0.8298	−0.297	0.0238	n.s.	0.0596
Ct.Th	−0.536	< 0.0001	−0.609	< 0.0001	0.356	0.0071	0.593	< 0.0001	−0.293	0.0243	−0.511	< 0.0001	−0.620	< 0.0001
Ct.TMD	n.s.	0.4809	n.s.	0.9163	n.s.	0.1724	n.s.	0.2816	−0.334	0.0098	n.s.	0.9129	n.s.	0.1231
Ct.Po	0.345	0.0074	0.506	0.0002	n.s.	0.4924	−0.597	< 0.0001	0.576	< 0.0001	0.340	0.0091	0.509	0.0001
Stiffness	−0.462	0.0003	−0.469	0.0006	0.278	0.0399	0.638	< 0.0001	−0.338	0.0095	−0.328	0.0127	−0.330	0.0158
Yield force	−0.639	< 0.0001	−0.705	< 0.0001	0.469	0.0003	0.618	< 0.0001	n.s.	0.1560	−0.641	< 0.0001	−0.623	< 0.0001
Ultimate force	−0.480	0.0001	−0.634	< 0.0001	0.345	0.0098	0.620	< 0.0001	−0.303	0.0208	−0.476	0.0002	−0.459	0.0005
PYD	n.s.	0.0689	n.s.	0.0987	n.s.	0.6638	n.s.	0.5152	n.s.	0.9802	n.s.	0.1413	n.s.	0.0727
Work-to-fracture	n.s.	0.9628	n.s.	0.4714	n.s.	0.5531	n.s.	0.1427	n.s.	0.2292	n.s.	0.7098	n.s.	0.9954
Length	n.s.	0.2426	n.s.	0.2369	−0.293	0.0298	n.s.	0.7809	n.s.	0.3194	n.s.	0.0810	0.436	0.0011
Yield stress	−0.555	< 0.0001	−0.575	< 0.0001	0.350	0.0088	0.410	0.0014	n.s.	0.0871	−0.558	< 0.0001	−0.573	< 0.0001
Ultimate stress	−0.349	0.0073	−0.513	0.0001	n.s.	0.2806	0.459	0.0003	−0.376	0.0036	−0.368	0.0049	−0.344	0.0116
Modulus	n.s.	0.1000	n.s.	0.2307	n.s.	0.7506	0.352	0.0067	−0.464	0.0002	n.s.	0.4836	n.s.	0.2665
Toughness	n.s.	0.3294	n.s.	0.7275	n.s.	0.8667	n.s.	0.7005	n.s.	0.2471	n.s.	0.5190	n.s.	0.3715
PY toughness	n.s.	0.0834	n.s.	0.0545	n.s.	0.8749	n.s.	0.5631	n.s.	0.9063	n.s.	0.1207	n.s.	0.0812
BV/TV	−0.702	< 0.0001	−0.653	< 0.0001	0.486	0.0001	0.515	< 0.0001	n.s.	0.5056	−0.567	< 0.0001	−0.585	< 0.0001
Conn.D	n.s.	0.4290	n.s.	0.1502	n.s.	0.8887	n.s.	0.2338	n.s.	0.5152	n.s.	0.5703	n.s.	0.6154
SMI	0.456	0.0003	0.498	0.0002	n.s.	0.3765	−0.316	0.0149	n.s.	0.5441	0.446	0.0004	0.340	0.0118
Tb.N	n.s.	0.2440	−0.296	0.0350	n.s.	0.6555	n.s.	0.1477	n.s.	0.8174	n.s.	0.1207	n.s.	0.1580
Tb.Th	−0.706	< 0.0001	−0.555	< 0.0001	0.568	< 0.0001	0.532	< 0.0001	n.s.	0.3755	−0.505	0.0001	−0.562	< 0.0001
Tb.Sp	n.s.	0.2997	n.s.	0.0557	n.s.	0.7443	n.s.	0.1422	n.s.	0.9434	n.s.	0.1703	n.s.	0.2110
Tb.TMD	−0.356	0.0056	n.s.	0.0884	n.s.	0.6681	0.369	0.0040	−0.447	0.0004	n.s.	0.1774	−0.373	0.0054

Table 6.

Spearman r between metabolic properties vs. bone properties pooling data from CANA-treated mice. N = 47 to 51.

X property	Glucose		HbA1c		Insulin		P1NP		CTX		uCA/Cr		uPHOS/Cr
Y property	r	p-Value	r	p-Value	r	p-Value	r	p-Value	r	p-Value	r	p-Value	r	p-Value
Ma.V	0.424	0.0019	n.s.	0.2970	n.s.	0.2568	n.s.	0.4481	0.299	0.0348	n.s.	0.1616	0.304	0.0318
Ct.Ar	n.s.	0.6254	n.s.	0.8661	n.s.	0.5555	0.421	0.0021	n.s.	0.5427	n.s.	0.4518	n.s.	0.3168
Tt.Ar	n.s.	0.1676	n.s.	0.4537	n.s.	0.5672	n.s.	0.1593	n.s.	0.3900	n.s.	0.6783	n.s.	0.5101
Imin	n.s.	0.7943	n.s.	0.0854	n.s.	0.9167	0.400	0.0036	n.s.	0.9644	n.s.	0.3935	n.s.	0.5570
Imin/cmin	n.s.	0.8584	n.s.	0.1289	n.s.	0.8616	0.421	0.0021	n.s.	0.9487	n.s.	0.6311	n.s.	0.7599
Ct.Th	−0.478	0.0004	n.s.	0.8736	n.s.	0.1071	0.595	< 0.0001	−0.367	0.0088	−0.359	0.0103	−0.495	0.0003
Ct.TMD	−0.359	0.0097	n.s.	0.0869	n.s.	0.7934	0.311	0.0265	−0.339	0.0160	n.s.	0.8781	n.s.	0.1887
Ct.Po	0.580	< 0.0001	n.s.	0.9596	n.s.	0.2719	−0.483	0.0003	0.411	0.0030	0.319	0.0237	0.440	0.0014
Stiffness	n.s.	0.2150	n.s.	0.4363	n.s.	0.2243	0.332	0.0185	n.s.	0.7634	n.s.	0.0628	−0.293	0.0407
Yield force	−0.344	0.0143	n.s.	0.6657	n.s.	0.8018	0.521	0.0001	n.s.	0.0761	n.s.	0.6391	n.s.	0.0780
Ultimate force	n.s.	0.3848	n.s.	0.3729	n.s.	0.5486	0.453	0.0009	n.s.	0.1445	n.s.	0.1637	n.s.	0.0692
PYD	n.s.	0.6910	n.s.	0.4411	n.s.	0.5584	n.s.	0.8503	n.s.	0.1997	n.s.	0.9039	n.s.	0.4678
Work-to-fracture	n.s.	0.9277	n.s.	0.5639	n.s.	0.5559	n.s.	0.3993	n.s.	0.0731	n.s.	0.7599	n.s.	0.5978
Length	n.s.	0.2645	n.s.	0.2414	n.s.	0.8665	n.s.	0.9554	n.s.	0.8036	n.s.	0.8634	n.s.	0.8039
Yield stress	−0.395	0.0046	n.s.	0.1510	n.s.	0.5340	0.298	0.0359	n.s.	0.0967	n.s.	0.4658	−0.291	0.0422
Ultimate stress	n.s.	0.4066	n.s.	0.6522	n.s.	0.4023	n.s.	0.4477	n.s.	0.0881	n.s.	0.3606	n.s.	0.0976
Modulus	n.s.	0.1355	n.s.	0.3817	n.s.	0.1155	n.s.	0.8037	n.s.	0.9791	n.s.	0.0718	−0.328	0.0214
Toughness	n.s.	0.9045	n.s.	0.5031	n.s.	0.5674	n.s.	0.4689	n.s.	0.0678	n.s.	0.5675	n.s.	0.4610
PY toughness	n.s.	0.3609	n.s.	0.6412	n.s.	0.4800	n.s.	0.7906	n.s.	0.4964	n.s.	0.3704	n.s.	0.1666
BV/TV	−0.421	0.0021	n.s.	0.4652	n.s.	0.2573	0.516	0.0001	n.s.	0.0508	−0.431	0.0018	−0.434	0.0016
Conn.D	n.s.	0.7887	n.s.	0.9605	n.s.	0.2717	n.s.	0.8907	n.s.	0.9091	n.s.	0.2826	n.s.	0.6273
SMI	n.s.	0.0816	n.s.	0.4301	n.s.	0.0761	n.s.	0.2041	n.s.	0.2300	0.306	0.0308	n.s.	0.0699
Tb.N	n.s.	0.7510	n.s.	0.1363	n.s.	0.4336	n.s.	0.1056	n.s.	0.0987	n.s.	0.1557	n.s.	0.3236
Tb.Th	−0.463	0.0006	n.s.	0.6739	n.s.	0.9073	0.488	0.0003	n.s.	0.4949	n.s.	0.1284	−0.351	0.0126
Tb.Sp	n.s.	0.7052	n.s.	0.1704	n.s.	0.3960	n.s.	0.0930	n.s.	0.0713	n.s.	0.1672	n.s.	0.3333
Tb.TMD	−0.508	0.0001	n.s.	0.9652	n.s.	0.6142	0.490	0.0003	−0.308	0.0293	n.s.	0.6746	n.s.	0.2383

Among untreated mice (High BG + Low BG groups combined), HbA1c and post-treatment BG, as markers of glycemic control, along with uCA/Cr and uPHOS/Cr, as markers of urine mineral loss, correlated with deterioration of structural properties of cortical and trabecular bone (Table 5). With respect to cortical bone, HbA1c and BG were negatively correlated with Ct.Ar, Ct.Th, and I_min, yet were positively correlated with Ct.Po. Similarly, uCA/Cr and uPHOS/Cr were negatively correlated with Ct.Ar, Ct.Th and I_min and positively correlated with Ma.V (uPHOS/Cr only) and Ct.Po (Table 5). With respect to trabecular bone, all 4 parameters (HbA1c, BG, uCA/Cr, uPHOS/Cr) correlated negatively with Tb.BV/TV, Tb.Th and Tb.TMD (BG and uPHOS/Cr only) and correlated positively with SMI (Table 5). Similarly, these parameters negatively correlated with indicators of bone strength (yield force, ultimate force, yield strength, ultimate strength) and bone stiffness.

In contrast, P1NP, a marker of bone formation, correlated with improvement in structural properties of cortical bone (positively correlated with Ct.Ar, Ct.Th, I_min; negatively correlated with Ct.Po) (Table 5) or trabecular bone (positively correlated with Tb.BV/TV, Tb.Th, Tb.TMD; negatively correlated with SMI) (Table 5). P1NP also correlated positively with bone strength and stiffness.

When examined within the CANA-treated mice alone (Table 6) these same relationships between metabolic properties and bone properties were often not significantly correlated. Only P1NP exhibited associations with bone parameters which were comparable in strength and directionality to those associations seen with untreated mice. (Spearman correlations between measurements of metabolism or urine mineral excretion and bone properties when all 3 groups are combined are included in Tables S4 and S5.)

4. Discussion

In contrast to previous reports by our groups and others comparing skeletal features between the TallyHO/JngJ mice with diabetes and SWR/J mice without the disease (17, 18), unique features of the TH polygenic, incompletely penetrant model of type 2 diabetes have allowed us to compare the impact of diabetes, independent of genetic strain differences, on skeletal phenotype. Herein, we have focused on skeletal differences between subgroups of TH mice, either with or without hyperglycemia and with or without exposure to glucose-lowering therapy with an SGLT2-inhibitor, specifically canagliflozin. In so doing, we have demonstrated that by 20 weeks of age, male TH mice with diabetes had comparatively lower body mass, and exhibited decrements in cortical and trabecular bone structure, along with an increase in cortical porosity, compared to non-diabetic TH mice. The urinary excretion of critical bone minerals, specifically calcium and phosphorus, was exaggerated in diabetic mice, and related to the degree of glucose excretion. Additionally, bone formation was reduced. Moreover, these changes in metabolism were associated with a reduction in overall bone strength and bone stiffness as determined by three-point bending tests of femur mid-shafts. Interestingly, however, bone toughness, or overall resistance to fracture, was not altered, perhaps reflecting a lack of T2D-related changes to the organic matrix that would impart brittleness (e.g., accumulation of advanced glycation end-products) and/or a decrease in mineral from the organic matrix at local sites (e.g., peri-lacunar) that would confer more strain capacity to the bone.

Treatment with an SGLT2-inhibitor such as canagliflozin normalizes blood glucose levels by enhancing urinary glucose excretion, rather than by modifying insulin secretion or sensitivity. In our experiments, SGLT2I treatment, as expected, was shown to normalize serum blood glucose levels and body weight in TH mice, most of whom (>75%) would have exhibited persistent diabetes out to 20 weeks of age without treatment. However, consistent with a drug-induced glucosuria, both calcium and phosphate wasting persisted in CANA-treated mice at study end. Additionally, the increase in cortical porosity estimate, the deficits in trabecular bone structure, and the reduction in bone strength and bone stiffness seen in the diabetic bone phenotype were not fully rectified with canagliflozin, despite good glycemic effect of the drug.

A reduction in bone stiffness and strength, as was seen in either TH mice with diabetes or in CANA-treated euglycemic mice, would be anticipated in disorders where a decrease in tissue mineralization exists (33). Evidence in clinical literature supports an association between glucosuria, with or without hyperglycemia, and urinary mineral excretion. For example, adolescent females with type 1 diabetes were shown to have negative calcium retention, attributed to higher urinary calcium excretion, which was positively correlated with HbA1c values, suggesting an increase in calcium excretion in those individuals with poorer glycemic control (34). Similarly, historical data examining children with more poorly controlled diabetes mellitus (mean HbA1c = 10.0 ± 0.3%) demonstrated simultaneous hypercalciuria and hyperphosphaturia, related to glucosuria, which was closely associated with the severity of linear growth failure (i.e. skeletal growth failure) in these children (35). Children with isolated renal glucosuria, without diabetes, also exhibit hypercalciuria (36). Finally, in patients with renal Fanconi syndrome who exhibit glucosuria and phosphaturia, hypophosphatemic osteomalacia or rickets is seen (37, 38).

In the present study, the excess urinary mineral loss of both calcium and phosphate, documented at study end, occurred in mice exhibiting glucosuria (i.e., those with diabetes, or those receiving SGLT2I treatment) such that the magnitude of mineral wasting was highly correlated with the magnitude of glucose excretion (Fig. 2B, 2D); therefore, some demineralization of bone could be expected in these mice over time, unless offset by increased mineral acquisition in the diet, or unless our end-point measures were not consistent with mineral excretion at earlier times (i.e. start of drug therapy). We also cannot ignore the observed relative reduction in daily PHOS excretion per daily glucose excretion in CANA-treated mice (Fig 2D) compared with High BG mice, given that SGLT2I mechanism of action would predict increased sodium-dependent phosphate reabsorption in the renal proximal tubule in CANA-treated mice, despite glucosuria, at least at the initiation of therapy (39). Interestingly, the extent of mineral loss from the organic matrix was not sufficient to significantly affect μCT-derived Ct.TMD (Table 1).

An association between excess calcium excretion, specifically, and subsequent bone disease has been validated clinically. For example, patients with idiopathic hypercalciuria demonstrate reduced bone density (40, 41). Population studies of women with post-menopausal osteoporosis also confirm an increased prevalence of PTH-independent hypercalciuria (42, 43). Moreover, bone biopsy specimens from hypercalciuric patients display evidence of a mineralization defect, consistent with osteomalacia (44, 45). Other studies have also shown that the rate of urine calcium excretion in humans appears to correlate with the degree of bone loss (46, 47). Together, these studies support the expectation that hypercalciuria will negatively impact bone mineralization; hence, a similar skeletal outcome from glucosuria-associated hypercalciuria in TH mice could be anticipated.

Our results in the TallyHO model of T2D are consistent with previous reports by us and others demonstrating hypercalciuria in rodent models of diabetes or SGLT2I-related glucosuria. For example, in DBA/2J mice treated with streptozotocin (STZ), significant glucosuria and hypercalciuria were evident in diabetic mice, as well is in canagliflozin-treated diabetic and non-diabetic DBA/2J mice (25, 48). In pregnant rats, maternal diabetes is also associated with marked hypercalciuria, when compared with non-diabetic pregnant rats (49). Furthermore, genetic ablation of SGLT2 function in the Jimbee mouse model was found to result in glucosuria and hypercalciuria, along with growth stunting and a reduction in tissue mineral density in male mice (50). A study by Wang, et al, in the Zucker Diabetic Fatty (ZDF) rat model of T2D also demonstrated evidence of hypercalciuria and hyperphosphaturia in diabetic rodents although glucosuria was not specifically documented in their publication (51); however, in their study, treatment with dapagliflozin (another SGLT2I), resulted in alleviation of hypercalciuria. Reasons for this discrepancy are unclear, but could reflect drug-specific differences on renal handling of calcium (canagliflozin vs. dapagliflozin), differences in the T2D model (rat vs. mouse), or the small sample size used in the Wang, et al study (n=6-7 rats/group) (51), compared with our study.

In other rodent models of diabetes, the reduction in the fracture resistance of long bone has been attributed to the hyperglycemia-mediated accumulation of advanced glycation end-products, namely pentosidine, a fluorescent collagen crosslink (52, 53). While we did observe a decrease in bone strength in hyperglycemic TH mice, we did not observe a decrease in work-to-fracture or post-yield toughness in these mice; additionally, we did not see full improvement in bone strength with canagliflozin despite normalization of glucose control. Finally, while bone strength parameters were inversely associated with HbA1c values in non-treated mice (Table 5), these parameters were unrelated to HbA1c in CANA-treatment mice (Table 6). Moreover, these parameters did not improve in conjunction with the excess daily glucose excretion seen in CANA-treated mice. Together, these observations suggest that changes in stiffness, strength (yield force and yield stress), and overall fracture resistance across the groups cannot be explained by hyperglycemia alone, or hyperglycemia mitigation, in this model.

An increase in cortical porosity is a common finding among rodent models of T2D, including the ZDF rat (54), the Zucker Diabetic Sprague-Dawley (ZDSD) rat (53), and the Otsuka Long-Evans Tokushima Fatty rat (55). An association between porosity and severity of dysglycemia has been demonstrated in these same studies. Similarly, an increase in distal radial cortical porosity has been demonstrated in elderly females with T2D (56-59), particularly in those with history of fragility fractures (60). An association between higher porosity and higher HbA1c has also been demonstrated in some clinical studies (61). However, to our knowledge, an association between exaggerated urinary mineral loss and cortical porosity has not been reported, though lacunar and canicular porosity has been observed to increase in lactating mice (62). In the present study, Ct.Po estimate was significantly higher (11.3% on average) in High BG mice compared to Low BG mice, and persisted in CANA-treated mice. In our prior TH study (18), Ct.Po appeared to be similar between the 4-5 TH mice per age group (16-wks. and 34-wks.) that had high blood glucose levels – but not as high as the TH mice in the present study – and the 7-10 TH mice per age group that had low (16-wks.) or moderate BG levels (34-wks.). This supports the notion that the T2D-related increase in cortical porosity is the result of severe hyperglycemia (i.e., frank, uncontrolled diabetes).

Interpretation of our results are limited by several factors. First, this study was not designed to provide a comprehensive assessment of calcium or phosphate metabolism with respect to bone strength. Specifically, we did not measure serum calcium or phosphorus concentrations. Additionally, we did not rigorously quantify dietary mineral intake (calcium or phosphorus), gastrointestinal absorption, or total body mineral retention. Estimated food intake available from 2 of 3 experimental replicates indicated that food consumption was increased in both the High BG groups, and the CANA groups, relative to the Low BG group [CANA: 0.18 ± 0.04 g (chow)/g (mouse)/day (providing 18 mg/kg/day of canaglifozin); High BG: 0.23 ± 0.06 g (chow)/g (mouse)/day; Low BG: 0.12 ± 0.03 g (chow)/g (mouse)/day; ANOVA p<0.0001], resulting in a relative increase in daily dietary calcium and phosphorus intake in mice that consumed more chow. However, total body mineral balance was not specifically quantified. Therefore, any link between mineral excretion and skeletal mineralization remains indirect, observational and hypothetical at present. Additionally, at this time we have studied only one of the SGLT2I medications (canagliflozin), given at only one dose. Therefore, our findings are not generalizable to other members of the SGLT2I drug class, and our results differ, as noted above, from those reported for dapagliflozin (51). Although canagliflozin therapy did not affect the material strength (ultimate stress) of cortical bone in TH mice at the apparent level, we do not know whether this SGLT2 inhibitor affected mechanical or compositional properties at the tissue-level. In another pre-clinical model of diabetes involving male mice (fed a high fat diet and administered a low dose of STZ), dapagliflozin and liraglutide rescued or partially rescued the mineral-to-matrix ratio, crystal size, and glycation (Fourier transformed infrared spectroscopy or FTIR) as well as hardness and modulus (nanoindentation) of newly formed tissue (i.e., tissue between double labels) (63). Treatment effects on composition were not observed, however, when a quadrant of cortical bone (femur mid-shaft) comprising more existent than newly formed tissue was analyzed by FTIR imaging. Finally, our results are pertinent only to male TallyHO/JngJ mice. Female TH mice develop obesity, moderate hyperinsulinemia, and hyperleptinemia, but do not progress to frank diabetes (14).

In summary, this study has now delineated the diabetic bone phenotype, independent of obesity, in the TallyHO/JngJ model of early onset type 2 diabetes mellitus involving male mice. We have also demonstrated that mitigation of hyperglycemia in this model by an SGLT2-inhibitor only partially rectifies the bone deficits that occur when blood glucose levels are very high. Instead, glucosuria-associated mineral wasting persisted with long-term SGLT2I exposure, likely complicating the ability of this anti-diabetic therapy to restore bone despite glucose normalization.

HIGHLIGHTS.

• TallyHO (TH) mouse is a polygenic, incompletely penetrant model of early onset T2D

• Deficits in cortical bone structure, strength exist in diabetic TH femurs at 20 wks

• Deficits in trabecular bone exist in diabetic TH femurs, L6 vertebrae

• Canagliflozin treatment of diabetic TH mice for 12 weeks normalizes glucose control

• However, bone deficits and urine mineral wasting persist with SGLT2I treatment