This study is an ancillary work based on the ADIMOS cohort (Clinical Gov NCT03219125), including postmenopausal women enrolled by the Department of Rheumatology at Lille University Hospital (France) between October 2018 and June 2021 (10). Two groups were compared. Exposed participants were postmenopausal women with a fragility fracture that occurred within the 12 previous months. In contrast, Non-exposed participants corresponded to postmenopausal women with osteoarthritis without any fragility fracture history. The study protocol was approved by the local Institutional Review Board (2017-A00472-51), and the study procedures complied with the ethical standards of the relevant institutional and national Human Experimentation Ethics Committees. All patients provided their written informed consent.

All participants underwent a complete musculoskeletal physical examination by an experienced rheumatologist with expertise in managing patients with osteoporosis (JP, 13 years of experience). Inclusion and exclusion criteria were reviewed. Other comorbidities were looked for (e.g., type 2 diabetes mellitus, chronic pulmonary disease, cardiovascular events), and the Charlson Comorbidity Index (CCI) was calculated. Lifestyle characteristics were assessed using the leisure time activity score. Medication data and relevant disorders were collected from all patients.

Bone mineral density (BMD) was measured at the lumbar spine (L1–L4) and the non-dominant hip by dual-energy X-ray absorptiometry (Hologic Discovery A; Hologic, Marlborough, Marlborough, USA). The machine was calibrated daily, and quality-assurance tests were carried out daily and weekly. WHO criteria were used to define osteoporosis and osteopenia based on BMD (T-score ≤ −2.5 and T-score between −1.0 and −2.5, respectively). Body composition derived from DXA acquisitions was obtained in all patients. It comprised BMI calculation (kg/m²) from the subject’s height and weight, as well as an estimate of the visceral adipose tissue (VAT, cm²), total body fat (TBF, %), and total and appendicular lean masses (kg).

After fasting for at least 8 hours, blood samples were obtained. Routine assays assessed creatinine and high-sensitivity C-reactive protein (hs-CRP). The estimated glomerular filtration rate (GFR) was calculated using the CKD formula (mL/min). Intact parathyroid hormone (PTH) was measured using a chemoluminescent immunoassay on an Architect automatic analyzer (Abbott Laboratories; Chicago, Illinois, USA). 25-hydroxyvitamin D was measured by a competitive chemoluminescent immunoassay on an IDS-iSYS instrument (IDS; Pouilly-en-Auxois, France). Procollagen 1 intact N-terminal (P1NP) and serum cross laps (CTX) were measured by chemiluminescence assay using the IDS-iSYS Multi-Discipline Automated Analyzer (Immunodiagnostic Systems, Inc., Fountain Hills, AZ).

Qualitative variables are reported as frequency (percentage). Continuous variables are reported as mean ± standard deviation in case of a normal distribution or as median [25th to 75th percentiles] otherwise. The normality of distributions was assessed using histograms and using the Shapiro-Wilk test. At baseline, the patient’s general characteristics and biochemistry results were compared between the two groups (exposed/non-exposed) using Chi-Square or Fisher exact test for categorical variables and Student t-test in case of a normal distribution or Mann-Whitney U test otherwise for continuous variables. Postmenopausal women’s paravertebral and psoas muscle features were compared between the two groups using analysis of covariance without and with adjustment for age, weight, and height. The relationship between myosteatosis and bone marrow adiposity, myosteatosis and VAT, myosteatosis and TBF, bone marrow adiposity and VAT, and bone marrow adiposity and TBF were investigated with and without adjustment for predefined confounding factors in fractured/non-fractured group separately, using analyses of variance. For each model, the normality of residuals was checked graphically. We performed two different adjusted models. The first was only adjusted for age. The second was adjusted for age, weight, height, and adding the bone mineral density according to the studied bone marrow adiposity. In each group, the intra- and inter-observer agreement for postmenopausal women’s paravertebral and psoas muscle features was calculated using the concordance correlation coefficient and its 95% confidence interval (14). Values were interpreted using Landis & Koch classification (0, disagree; 0-0.20, very low agreement; 0.21-0.40, low agreement; 0.41-0.60, moderate agreement; 0.61-0.80, important agreement; 0.81-1, almost perfect agreement). All statistical tests were done at the two-tailed α-level of 0.05 using the SAS software version 9.4 (SAS Institute, Cary, NC).

Table 1 shows the baseline characteristics. Among the initial 199 participants, 97 postmenopausal women were excluded from the original ADIMOS cohort (paravertebral acquisitions missing or insufficient quality due to respiratory artifacts). In the remaining 102 postmenopausal women with no recent use of bone-active medication included in this ancillary study, non-fractured women (n=46) were significantly younger (p=0.001), taller (p=0.010), and heavier (p=0.032) than fractured participants (n=56). Comorbidities and osteoporosis risk factors were comparable in both groups. In parallel, the CCI was lower in the non-fractured group (p<0.001). Regarding biochemistry results, non-fractured postmenopausal women had significantly lower PINP (p<0.001) than fractured patients. In the fractured group (n=56), we included 28 women with at least one vertebral fracture (median [min-max]: 1 (1–4)), 28 women with non-vertebral fractures (11 forearm/wrist fractures, 9 hip fractures, 5 pelvis fractures, and 3 proximal humerus fractures), and 31 (55.4%) had a history of osteoporotic fractures.

Body composition parameters were comparable in both groups, except for total lean mass, which was higher in the non-fractured group compared to fractured women (36.5 kg (IQR: 33.4 to 39.1) versus 33.8 kg (IQR: 31.2 to 36.1), p=0.038).

Non-fractured postmenopausal women had significantly higher BMD at the lumbar spine, femoral neck, and total hip than in the fractured group (p<0.05 for all). Regarding bone marrow PDFF, no significant differences were found between those with and without fracture ( Table 1 ). PDFF at the lumbar spine was higher in the fractured group than in non-fractured subjects but did not reach statistical significance (58.3 ± 10.2% versus 54.9 ± 9.8%, p=0.091).

The mean PDFF of the psoas and paravertebral muscles was higher in the fracture group than in non-fractured patients, even after adjustment for age, weight, and height (PDFF_Psoas: 17.1 ± 6.1% versus 13.5 ± 4.9%, p=0.004, and PDFF_{Paravertebral}: 34.4 ± 13.6% versus 24.9 ± 8.8%, p=0.002). No significant differences between fractured and non-fractured participants could be detected in CSA and CMI values of the psoas and paravertebral muscles after adjustment for age, weight, and height ( Table 2 ).

In the fracture group, there was no significant relationship between PDFF in the psoas or paravertebral muscles and marrow adiposity, either at the lumbar spine or the femoral neck ( Table 3 ). Surprisingly, in the non-fractured group, we found a significant relationship between higher PDFF_{Paravertebral} and lower PDFF at the lumbar spine (β = -6.80 ± 2.85, p=0.022) after adjustment for age, weight, height, and lumbar spine BMD. Expressed otherwise, each standard deviation increase in PDFF_{Paravertebral} was associated with lower PDFF lumbar spine (-6.8%).

In both groups, there was a significant relationship between higher PDFF_Psoas and higher VAT after adjustment for age, weight, and height (β = 20.27 ± 9.62, p=0.040 in the fracture group, and β = 37.49 ± 8.65, p<0.001 in the non-fractured group) ( Table 4 ). Each SD increase in PDFF_Psoas was associated with higher VAT (20.27 cm² and 37.49 cm², respectively).

In the non-fractured group, there was a significant relationship between higher PDFF_{Paravertebral} and higher TBF after adjustment for age, weight, and height (β = 6.57 ± 1.80, p<0.001). Each SD increase in PDFF_{Paravertebral} was associated with higher TBF (+6.57 kg).

After adjusting for age, weight, height, and BMD in both groups, there was no significant relationship between PDFF lumbar spine and femoral neck with other fat depots ( Table 5 ).

Intraclass correlation coefficients were excellent (over 0.90) for PDFF and CMI measurements in the psoas and paravertebral muscles. Intraobserver agreement was good for CSA measurements in the psoas muscles (0.89, with a 95% confidence interval of [0.78-0.95]) and excellent for the interobserver agreement (0.95, with a 95% confidence interval of [0.91-0.98]). Similarly, interobserver agreement was good for CSA measurements in the paravertebral muscles (0.89, with a 95% confidence interval of [0.79-0.95]) and excellent for intraobserver agreement (0.96, with a 95% confidence interval of [0.91-0.98]).

Fatty infiltration of multifidus and erector spinae muscles is well known to precede their atrophy after vertebral fracture, with no CSA change six months from onset (15). Accordingly, we observed higher fat deposits in paravertebral muscles among patients with a recent fragility fracture compared to controls without significant changes in CSA. This natural course highly suggests bone-muscle-fat interactions, supported by previous research on postmenopausal women with osteoporosis. The AMBERS cohort is indicative of this hypothesis. Using MRI and pQCT, this cross-sectional study reported that a higher amount of fat within muscles was associated with higher bone marrow adiposity at the mid-tibia, adjusting for age, weight, height, average daily energy expenditure, hypertension, and diabetes (7). However, no adjustment for BMD was performed, whereas a robust negative correlation has been observed with bone marrow adiposity (16–18). Our analysis considered BMD a potential confounder and did not reveal a relationship between bone marrow adiposity and myosteatosis in patients with fragility fractures. Interestingly, a higher amount of fat within the paravertebral muscles (erector spinae and multifidus) was associated with a lower PDFF at the lumbar spine among non-fractured women, adjusted for age, weight, height, and lumbar BMD.

The underlying mechanism explaining the differential effect between groups is unclear. The decreased physical activity with aging and muscle disuse makes the erector spinae and multifidus more vulnerable to fat infiltration than the psoas and lower limb muscles (19). A similar observation was reported in healthy men, with an increase in intermuscular adipose tissue of erector spinae but not psoas muscles with aging (20). In patients with a recent fragility fracture, one hypothesis is that the trends in a higher bone fatty infiltration compared to controls might mask the potential but weak bone-muscle interaction. Nevertheless, this novel finding supports an existing but fragile interaction between bone marrow adiposity at the lumbar spine and paravertebral myosteatosis in the absence of fragility fracture, with numerous potential actors at the cellular and hormonal levels (1).

Contrary to the underestimated importance of bone marrow adiposity, the role of ectopic fat infiltration into muscle –i.e., myosteatosis– has gained increasing attention in the scientific literature. Miljkovic et al. reported an independent association between abdominal myosteatosis (all muscles of the abdominal wall), hyperinsulinemia, and insulin resistance, even after adjusting for lifestyle factors, TBF, and visceral or subcutaneous adipose tissues (21). Indeed, skeletal muscle plays a crucial role in insulin resistance and may contribute to developing type 2 diabetes mellitus (22).

Furthermore, a strong link exists between myosteatosis and body fat. We observed that fatty infiltration in psoas muscles increased in patients with more extensive VAT, even after adjusting for age, weight, and height. Among postmenopausal women without fragility fractures, an association was found between PDFF in the erector spinae and multifidus muscles and TBF, based on the same adjustments. Although conducted in a cohort comprising only men over 65, these observations comply with the work from Miljkovic et al., reporting a relationship between psoas or paraspinal intermuscular adipose tissue with TBF and VAT (21). This proximity between muscle and visceral adipose tissues is supported by fundamental studies that observed similar molecular profiles (23). However, a more nuanced picture was depicted by Correa-de-Araujo et al., the association between myosteatosis and metabolic disorders being unclear when adjusting for visceral and ectopic fat depots (24). Nonetheless, our data suggest that skeletal and visceral fat depots share close functions from a metabolic perspective.

Bone marrow adiposity is a recently recognized tissue, underestimated for a long time. Its peculiar relationship with skeletal health is best described by the paradoxical accumulation despite the depletion of other fat tissues in women with anorexia nervosa (25). Bone marrow adiposity may contribute to vertebral bone weakness and is inversely correlated with lower trabecular BMD in older women (26, 27). Potential clinical implications have been reported, but data are still scarce or inconsistent regarding its relationship with other fat depots (3).

Our analysis did not highlight any significant relationship between lumbar or femoral neck bone marrow adiposity and other fat depots after age, weight, height, and BMD adjustment. In a study based on a cohort of premenopausal women, Bredella et al. reported a positive correlation between vertebral bone marrow adiposity and VAT, even after adjusting for BMD (28). In patients with anorexia nervosa, an inverse correlation between the VAT and bone marrow adiposity at the vertebral level or proximal femur was observed without reaching significance (25). These inconsistencies in the investigated relationship between bone marrow adiposity and other fat depots suggest that bone marrow adiposity is uniquely regulated, with differential tissue connections depending on associated disorders such as obesity, anorexia nervosa, or osteoporosis.

Our study comprises several strengths. (1) This ancillary work of the ADIMOS cohort yielded an original sample comprising prospectively included only postmenopausal women with or without a recent fragility fracture. This population is, therefore, homogeneous, limiting biases secondary to estrogen contribution. (2) Chemical shift encoding-based water-fat imaging was performed, considering the T1 bias and T2* decay, providing an accurate PDFF at the lumbar spine and paravertebral muscles (29). Despite the superior accuracy of MR spectroscopy, the use of chemical shift-based water-fat imaging allowed straightforward measurements achievable on most MR clinical systems while maintaining sufficient accurateness. Its main drawback was the lack of information provided about fat composition. Furthermore, although pQCT would bring more precise data on bone microarchitecture, MRI is considered as the gold standard and was preferred in our study to allow the exploration of the axial skeleton (30). Finally, DXA and MRI examinations were acquired on the same machines with the same parameters, reinforcing the comparability of the subjects’ data. (3) Literature is scarce on fat depots comparison. Our original approach compared myosteatosis with bone marrow adiposity in addition to conventional white adipose tissues (i.e., TBF and VAT). (4) Due to the strong correlation between bone marrow adiposity and BMD, our statistical model included this latter parameter to assess the relationship between the different fat depots independently.

However, a few limitations are also inherent to the study design and participants’ characteristics. (1) Although prospectively included, this study’s cross-sectional design and ancillary aspect restrained the temporal control between the occurrence of a vertebral fracture and imaging acquisitions. Nevertheless, the inclusion criteria outlined that only participants with a recent fracture (within 12 months) could be included. (2) Age and body parameters (weight and height) differed between groups. However, we adjusted for these characteristics in each statistical model we performed. Emergent anthropometric indices may be considered in future studies, such as the subcutaneous fat index (31). (3) Only a manual segmentation was achieved to quantify bone marrow adiposity and myosteatosis based on selected slices from MRI acquisitions. Semi-automatic or fully automated segmentation of vertebrae or paravertebral muscles would provide volumic features. However, we applied previously published methodological recommendations for assessing intramuscular fatty infiltration of paravertebral muscles (12). We also included the contractile mass index parameter (CMI) in our analyses to depict better the lean mass of the psoas, erector spinae, and multifidus muscles (9). Ultimately, intra- and interobserver agreements were good to excellent for MRI segmentation in our sample.