Effects of long‐term voluntary wheel running and selective breeding for wheel running on femoral nutrient canals

Abstract

The nutrient artery provides ~50%–70% of the total blood volume to long bones in mammals. Studying the functional characteristics of this artery in vivo can be difficult and expensive, so most researchers have measured the nutrient foramen, an opening on the outer surface of the bone that served as the entry point for the nutrient artery during development and bone ossification. Others have measured the nutrient canal (i.e., the passage which the nutrient artery once occupied), given that the external dimensions of the foramen do not necessarily remain uniform from the periosteal surface to the medullary cavity. The nutrient canal, as an indicator of blood flow to long bones, has been proposed to provide a link to studying organismal activity (e.g., locomotor behavior) from skeletal morphology. However, although external loading from movement and activity causes skeletal remodeling, it is unclear whether it affects the size or configuration of nutrient canals. To investigate whether nutrient canals can exhibit phenotypic plasticity in response to physical activity, we studied a mouse model in which four replicate high runner (HR) lines have been selectively bred for high voluntary wheel‐running behavior. The selection criterion is the average number of wheel revolutions on days 5 and 6 of a 6‐day period of wheel access as young adults (~6–8 weeks old). An additional four lines are bred without selection to serve as controls (C). For this study, 100 female mice (half HR, half C) from generation 57 were split into an active group housed with wheels and a sedentary group housed without wheels for 12 weeks starting at ~24 days of age. Femurs were collected, soft tissues were removed, and femora were micro‐computed tomography scanned at a resolution of 12 μm. We then imported these scans into AMIRA and created 3D models of femoral nutrient canals. We tested for evolved differences in various nutrient canal traits between HR and C mice, plastic changes resulting from chronic exercise, and the selection history‐by‐exercise interaction. We found few differences between the nutrient canals of HR versus C mice, or between the active and sedentary groups. We did find an interaction between selection history and voluntary exercise for the total number of nutrient canals per femur, in which wheel access increased the number of canals in C mice but decreased it in HR mice. Our results do not match those from an earlier study, conducted at generation 11, which was prior to the HR lines reaching selection limits for wheel running. The previous study found that mice from the HR lines had significantly larger total canal cross‐sectional areas compared to those from C lines. However, this discrepancy is consistent with studies of other skeletal traits, which have found differences between HR and C mice to be somewhat inconsistent across generations, including the loss of some apparent adaptations with continued selective breeding after reaching a selection limit for wheel‐running behavior.

We investigated femoral nutrient canal phenotypic plasticity in response to physical activity in mice from four replicate high runner (HR) lines bred for voluntary wheel running, along with four non‐selected control (C) lines. Shown are 3D models of femoral nutrient canals, re‐oriented to measure transverse slices for cross‐sectional area. Many nutrient canal shapes were encountered. Voluntary wheel running for 12 weeks increased the number of canals in C mice but decreased it in HR mice.

1. INTRODUCTION

Bones are dynamic, constantly remodeling in response to changing mechanical needs during growth, load bearing, and locomotion (Frost, 1987). Mechanical loading plays an important role in building and maintaining both skeletal mass and strength (Frost, 1997; Huiskes et al., 2000; Mori et al., 2003; Newhall et al., 1991; Rubin & Lanyon, 1984). When bones experience loading from mechanical forces, the resulting strain induces microdamage to the bone tissue (Seref‐Ferlengez et al., 2015). Osteocytes (mechanosensory cells that sense fluid flow associated with strain) then translate mechanical strain to biochemical signals and initiate bone remodeling (Bonewald, 2007; Yu et al., 2018). Over the course of bone remodeling, osteoblasts and osteoclasts add or remove bone, respectively (Katsimbri, 2017). Thus, exercise that increases the mechanical loading (and strain) on bones can induce changes to traits related to biomechanical properties, such as bone mineral density, mass, and tensile strength (Gómez‐Cabello et al., 2012; Jones et al., 1977; Karlsson & Rosengren, 2020; Yuan et al., 2016).

Coincident with increased bone remodeling, mechanical loading from exercise also results in increased levels of regional bone and marrow blood flow (Jones et al., 1977; Stabley et al., 2014). Although nutrient, epiphyseal–metaphyseal, and periosteal arteries all supply blood to long bones (Rhinelander, 1972), the nutrient artery is often the main source of blood (Gümüsburun et al., 1994; Trueta, 1963). The Hagen–Poiseuille equation describes how factors can influence the flow rate through a pipe (e.g., an artery):

$$Q=\frac{\mathit{\Delta p\pi }{R}^4}{8\mathit{\mu L}},$$

where flow (Q) is increased when the difference in pressure (Δp) is positive, the radius of the vessel (R) is increased, the viscosity of the medium (μ) is decreased, or the length of the vessel (L) is decreased. In particular, changes to the radius of the vessel can have a large effect on flow due to their exponential relationship. Therefore, increased blood perfusion (e.g., due to exercise) can be reasonably attributed to at least some increases in arterial vessel size. Conversely, a study which unloaded the hindlimbs of adult rats by suspending them for 2 weeks found a significant decrease in nutrient artery maximal diameter compared to controls (Prisby et al., 2015). No studies have yet directly investigated the effect of exercise on nutrient artery diameter. However, measuring the nutrient artery in vivo is difficult, impractical, or in the case of deceased organisms, impossible. Instead, researchers have measured the nutrient foramen (an opening on the outer surface of the bone that is formed when the nutrient artery enters cartilage during endochondral ossification) or the nutrient canal (i.e., the passage which the nutrient artery once occupied) as a proxy for nutrient artery size and, by extension, blood flow to long bones.

Nutrient canals are located in the diaphysis of long bones and are the entry point for arteries, veins, and peripheral nerves to the medullary cavity (Houssaye & Prévoteau, 2020). Given that bone is a hard tissue, one might expect that nutrient canal size limits the maximum size of any vessels that pass through it and so would be positively related to maximal blood flood (e.g., as required during intense periods of bone growth or remodeling). A study of chickens found that nutrient canal area was significantly positively correlated (r = 0.51, p = 0.02) with the area of the nutrient artery lumen in the femur, although the nutrient artery lumen occupied only ~20% of the nutrient canal cross‐section (Hu et al., 2022). This indicates that femoral blood flow rate can, to some extent, be estimated from nutrient canal size. Additionally, nutrient canal size (adjusted for body size) was correlated with the whole‐body maximum rate of oxygen consumption during exercise in comparisons among species of mammals but was less correlated with resting metabolic rate (Seymour et al., 2012). As a result, some have proposed that nutrient canal size can be used as a proxy for the metabolic intensity of extinct animals, such as dinosaurs (Seymour et al., 2012). In kangaroos, during in‐pouch development, high growth rate appears to be the main factor determining femoral bone blood flow, while in the post‐pouch life stage, micro‐fracture repair is most influential on blood flow requirements (Hu et al., 2018). However, despite the intuitive importance of blood flow in the early phases of bone growth and ossification, or whenever bone is remodeling (Gross et al., 1981; Sim & Kelly, 1970), relatively little research has been conducted on nutrient canals overall, and, in particular, on whether nutrient canals exhibit phenotypic plasticity in response to behaviors expected to increase blood flow to long bones (e.g., exercise).

The high runner (HR) mouse model can be used to simultaneously study potential genetic and training effects on bone. Briefly, four replicate lines of mice have been bred for voluntary wheel‐running behavior based on the total number of revolutions run on days 5 and 6 of a 6‐day exposure to wheels. Four non‐selected control (C) lines are bred without regard to wheel running. HR mice run approximately three times more than C mice on a daily basis, mainly from increased average speed rather than duration (Copes et al., 2018; Garland et al., 2011; Hiramatsu et al., 2017; Kelly et al., 2017; Swallow, Carter, & Garland, 1998). HR lines reached an apparent selection limit for wheel running after ~17–27 generations, depending on the replicate line and sex (Careau et al., 2013), and have since continued to run ~2.5–3‐fold more than C lines (currently at 102 generations of selection) (Copes et al., 2018; Garland et al., 2011; Hiramatsu et al., 2017; Kelly et al., 2017).

High runner mice have higher running endurance (Meek et al., 2009) and maximum aerobic capacity (VO₂max) during forced treadmill exercise (e.g., Cadney et al., 2021; Hiramatsu et al., 2017; Schwartz et al., 2023; Singleton & Garland, 2019; Swallow, Garland, et al., 1998), among a number of other anatomical, physiological, neurobiological, behavioral, and genetic differences from C mice (e.g., see Cadney et al., 2021; Hillis & Garland, 2023; Rhodes et al., 2005; Singleton & Garland, 2019; Wallace & Garland, 2016). The HR and C lines have also been shown to differ in skeletal morphology (Castro et al., 2022; Kelly et al., 2006; Middleton et al., 2010; Schutz et al., 2014; Wallace et al., 2012; Young et al., 2009). For example, adjusting for variation in body mass, and depending on the generation studied (See Castro et al., 2021), HR mice have been reported to have increased diameter and mass of hindlimb bones (Kelly et al., 2006), wider distal femora (Middleton et al., 2008), significantly larger periosteal areas, endocortical areas, and polar moments of area in the femur (Wallace et al., 2012), significantly larger femoral condyles (Garland & Freeman, 2005), and to lack significant hindlimb directional asymmetry, which is present in control mice (Garland & Freeman, 2005). However, hindlimb length and metatarsal‐to‐femur ratio, which are classic indicators of cursoriality, were not increased in HR mice (Castro et al., 2022; Garland & Freeman, 2005). A previous study of the nutrient canals of both HR and C mice was conducted at generation 11, prior to the selection limit, finding that HR mice had significantly higher nutrient canal total cross‐sectional area (CSA) in both sexes (Schwartz et al., 2018).

The purpose of the present study was to investigate (1) whether previously studied differences in nutrient canal size between HR and C lines (Schwartz et al., 2018) were still present at a later generation (i.e., after selection limits had been attained), and (2) whether nutrient canal cross‐sectional area exhibits phenotypic plasticity in response to chronic exercise. We studied the same 100 females from generation 57 that have been previously examined in Copes et al. (2015, 2018), Lewton et al. (2019), and Castro et al. (2022). Given that the HR mice used for the present study ran much more than C mice from weaning (3 weeks) to 15 weeks of age (Copes et al., 2018), we expected that any effects of chronic exercise, if present, would be greater in HR than in C mice.

2. METHODS

2.1. Selection experiment background and experimental design

As noted above, the specimens used here are the same as in Copes et al. (2018) and were sampled from generation 57 of the HR selection experiment (Swallow, Carter, & Garland, 1998), as outlined in Section 1. (The selection experiment has since passed 100 generations.) Briefly, we sampled a total of 100 females, equally divided between the four replicate HR and four non‐selected control (C) lines, except that HR line 6 included four extra mice because it is polymorphic for the mini‐muscle phenotype (see below). Mice were weaned and weighed at 21 days of age.

The 12 weeks of experimental procedures began when the mice were 24–27 days old, and then housed individually, half in cages with an attached wheel (see below) and half without. Mice reach sexual maturity at ~6 weeks of age (Jilka, 2013) and experiments involving bone changes in mice typically last 8–12 weeks, because bone growth slows substantially after puberty (Bourguignon, 1988; Jilka, 2013). Weekly procedures included weighing of each mouse and food hopper, from which apparent food consumption was determined (Swallow et al., 2001), as reported elsewhere (Copes et al., 2015, 2018). All experimental procedures were approved by the Institutional Animal Care and Use Committees at the University of California, Riverside and Arizona State University.

2.2. Wheel running

At ~24 days of age, half of the mice were given wheel access, as used in the selection protocol (1.12 m circumference) (Swallow, Carter, & Garland, 1998). Each of the four groups (Control No Wheel, Control Wheel, HR No Wheel, HR Wheel) began with 25 mice. The term “active” will be used to refer to the groups with access to wheels whereas those without access to wheels are referred to as “sedentary.” Each day, a computer recorded wheel revolutions in 1‐min intervals over a period of 23.5 h. We calculated the total number of revolutions, the number of 1‐min intervals with at least one revolution (minutes of wheel activity), the mean speed of running (revolutions/intervals), and the single interval with the greatest number of revolutions (maximum speed) using SPSS (IBM). We used average values for wheel running across 12 weeks, which have been reported previously (Copes et al., 2018), as covariates to predict bone traits.

2.3. Spontaneous physical activity in the home cage

All 100 cages were fitted with a passive infrared sensor placed in a corner and housed in a wire mesh protective enclosure (Copes et al., 2015). Total home‐cage activity (HCA) was taken as the sum of all activity over 23.5 h, HCA duration was calculated as the number of 1‐min intervals during which any activity was registered, and mean intensity of activity was calculated as total HCA divided by minutes of activity. Similarly to wheel running, we used the average values of HCA over 12 weeks, reported in Copes et al. (2018), as covariates in statistical analyses.

2.4. Dissections and specimen preparation

Over the course of the experiment, three mice died of various causes. The remaining 97 were euthanized by CO₂ overdose. The triceps surae muscles were weighed and their mass was used to determine the number of mice with the mini‐muscle phenotype (Kelly et al., 2013) (see below); 18 mice were found with the trait in this sample. Any tissue not taken at dissection was discarded or removed via soaking of the carcass in a 1% solution of enzymatic detergent (marketed as Tergazyme by Alconox). Starting with N = 100 mice, three mice died prior to collecting femurs, and three of the CT scans (see below) had too much static, making them unusable for measuring nutrient canals, for a final N of 94 for measuring nutrient canals.

2.5. Micro‐computed tomography scanning

The right femur of each specimen was micro‐computed tomography (micro‐CT) scanned at the University of Calgary (Viva‐CT40; Scanco Medical AG, Basserdorf, Switzerland) at 12‐μm resolution (55 kV, 145 mA, 500 projections). Copes et al. (2018) erroneously listed the resolution as 21 μm. The femur was chosen because, along with the humerus, they (1) are the largest long bones with the greatest attached muscle mass, (2) are the most frequently examined in studies of the effects of exercise on bone morphology (Ferguson et al. 2003; Jepsen et al., 2009; Judex et al., 2004; Tommasini et al., 2008; Yang et al., 2007), and (3) were previously used to study nutrient canal morphology in HR mice (Schwartz et al., 2018).

2.6. AMIRA 3D modeling

For each specimen, the raw data were reconstructed as 16‐bit TIFF image sequential stacks using ImageJ software (Schneider et al., 2012). Image stacks were imported into Thermo Scientific AMIRA 5.6 Software (Thermo Fisher Scientific, Waltham, MA, USA) for visualization and segmentation.

We followed the protocols previously established in Schwartz et al. (2018) for 3D modeling of femurs from CT scans and nutrient canals. The Supporting Information contains a PDF file with detailed instructions. Using the Isosurface module in AMIRA, surface renderings of the femur were created, and the external morphology of the nutrient foramina [defined here as the superficial openings through which the nutrient artery(s) is presumed to pass] was inspected. Examination for nutrient foramina was restricted to a portion of the bone inferior to the femoral neck and superior to the proximal edge of the patellar groove (Figure 1). This restriction was used in order to exclude metaphyseal and epiphyseal blood vessels, which typically penetrate bone outside of this region, although at least in mice the bottom third of the femur can also contain metaphyseal vessels (Bab et al., 2007; Brookes, 1958; Prisby, 2020). Our criteria for identification of a nutrient canal required a continuous absence of cortical bone from the periosteal (external) border of the cortex, through the cortex, and past the endosteal surface toward the medullary cavity. After identification, the empty space of the nutrient canal was manually selected slide‐by‐slide in AMIRA (Figure 2). Using the Label Field module, a 3D surface model of the nutrient canal was created. Once all the nutrient canals in the bone had been selected and modeled, each canal was isolated from the femur and virtually re‐oriented using the Align Principal Axes function. The nutrient canals were re‐oriented so that a transverse cross‐section could be obtained perpendicular to the long axis of the canal (Figure 3). This was necessary to avoid elliptical cross‐sections, which would overestimate the area compared to the correct circular cross‐section. Ten cross‐sections of the nutrient canal were measured for area, and the minimum cross‐sectional area of the total nutrient canal was recorded. The minimum cross‐sectional area was chosen because the flow through a cylindrical pipe is limited by the smallest cross‐sectional area as described by the Hagen–Poiseuille equation: $Q=\left(\mathit{P\pi }{r}^4\right)/\left(8\mathit{L\eta }\right)$, where $Q$ is flow rate, $P$ is the difference in blood pressure, L is vessel segment length, $\eta$ is blood viscosity, and $r$ is the radius of the vessel.

FIGURE 1.

Zone within which femoral nutrient canals were measured. 3D model of mouse femur, medial view, with distal end on the left, and proximal end on the right. Nutrient canals are circled in red. Following Schwartz et al. (2018), measurements were restricted to the region above the patellar groove and below the base of the femoral neck (as indicated by the yellow dashed lines) to prevent inclusion of metaphyseal and periosteal vessels which frequently penetrate bone outside this defined area.

FIGURE 2.

Example of a femoral nutrient canal seen across serial computed tomography scan slices. Sequential transverse slices of a mouse femur showing the nutrient canal (outlined in green).

FIGURE 3.

Re‐oriented nutrient canals of various shapes with transverse slices. 3D models of femoral nutrient canals, re‐oriented to properly measure the transverse slices for cross‐sectional area. Many different nutrient canal shapes were encountered, including, straight, curved, and branched. Mouse identification numbers are (left to right) 60204 (Line HR7), 60275 (Line C4), and 60447 (Line C2).

Because of the varied size and shape of nutrient canals, including both non‐linear or curved shapes and bifurcations, certain canals required multiple rounds of re‐orientation as described above. For nutrient canals with a curved shape, the long axis was re‐oriented several times along the length of the canal at each major inflection point (assessed visually) so that multiple perpendicular cross‐sections could be obtained. Only the smallest cross‐sectional area was used for further analysis.

In nutrient canals with bifurcation, all the branches of the canal were measured for minimum cross‐sectional area as described above. If the sum of the minimum cross‐sectional area of the branches was greater than that of the source trunk, then the branches were not considered blood‐flow limiting structures, and the minimum cross‐sectional area of the trunk was recorded (see slide 37 in Data S1). If the cross‐sectional area of the source trunk was greater than the sum of its branches, then the branches were considered as blood‐flow limiting. In this case, both branches were counted as distinct nutrient canals, each with its own minimum cross‐sectional area.

Although not measured in Schwartz et al. (2018), the location of each nutrient canal was also recorded by noting the slide on which the nutrient canal started and ended. Those numbers were averaged to obtain the midpoint of the nutrient canal. The start and end slide of the whole femur was also recorded. The nutrient canal's position along the bone as a proportion of its length was calculated.

As compared with Schwartz et al. (2018), another change to the methodology included the adjustment of the Zoom and Data Window as well as the Display and Masking parameters. Previously, these parameters were both adjusted on a case‐by‐case basis. This parameter is of particular importance because it determines what AMIRA considers to be bone versus empty space, which will directly affect the size of the minimum cross‐sectional area of all nutrient canals measured for that bone. Compared to the previous study, the CT scans from the current study had much more static noise, which made measurements more sensitive to variation in masking values. To lower possible sources of error, we elected to use standardized Zoom and Data Window as well as Display and Masking values across the entire data set. These values were obtained by having two researchers each produce two replicates of the aforementioned values for a total of 4 values for every parameter of every bone. The values were then compared, discrepant values were re‐examined, and all values were then averaged across the entire data set.

Additionally, having the correct number of nutrient canals is important for measuring the total cross‐sectional area correctly, as well as for analyzing the number of nutrient canals themselves. We confirmed each nutrient canal between two researchers for each of the 94 bones, checking that the nutrient canals were within the previously established borders, as well as fully penetrating the periosteal border, past the endosteal surface to the medullary cavity. This step was included because many nutrient canals were found much closer to the borders previously established in Schwartz et al. (2018).

2.7. Statistical analyses

Following numerous previous studies of these lines of mice (e.g., Castro et al., 2022; Copes et al., 2015, 2018; Lewton et al., 2019), data were analyzed as mixed models in SAS Procedure Mixed, with REML estimation and Type III Tests of Fixed Effects. Main effects were linetype (selected HR lines vs. non‐selected C lines), activity (active vs. sedentary), and the mini‐muscle phenotype (see below). Replicate line was nested within linetype as a random effect. Degrees of freedom for linetype, activity, and the linetype‐by‐activity interaction were 1 and 6. Analyses were done with and without body mass as a covariate. Additional analyses were done with wheel running and/or HCA (averaged across the entire 12‐week exposure to wheels) as covariates (Copes et al., 2018). Statistical significance was evaluated at p < 0.05. Outliers were removed if the absolute value of their standardized residual exceeded ~3 and/or the value was >1 standard deviation from the next value. For analysis of canal branching, scored as 0 for none or 1 if one or more branching canals occurred in a given femur, we used similar mixed models, but with SAS Procedure GLIMMIX.

The mini‐muscle phenotype (Garland et al., 2002) is caused by a Mendelian recessive mutation (Kelly et al., 2013) that halves hindlimb muscle mass, primarily due to a great reduction in the number of Type IIb muscle fibers (Talmadge et al., 2014), with many pleiotropic effects, such as generally larger internal organs (Garland et al., 2002; Kelly et al., 2017; Swallow & Garland, 2005). Various skeletal traits are altered in mini‐muscle individuals, including lengthening and narrowing of the femur (e.g., Kelly et al., 2006), lower femoral cortical areas and bending moments of inertia (Copes et al., 2018), as well as smaller femoral third trochanters (Castro et al., 2022). Mini‐muscle mice also have smaller ilium cross‐sectional properties, including cortical area, total periosteal area, polar section modulus, polar moment of area, and cortical area robusticity index (Lewton et al., 2019). The underlying allele was initially present at a frequency of ~7% in the base population. The mini‐muscle phenotype was observed in two of the four HR lines, eventually becoming fixed in one HR line (HR line 3) and remaining polymorphic in another (HR line 6). The mini‐muscle phenotype was observed in one C line for at least 22 generations, then lost. Of the 94 mice analyzed here for femoral canal properties, 11 in HR line 3 and 5 in HR line 6 were mini‐muscle.

3. RESULTS

3.1. Body mass, body length, and femur length

No statistically significant main effects or interactions were found for body mass, body mass with body length as a covariate, body length, or femur length with body mass or body length as a covariate (Table 1). Results were similar when the physical activity covariates were included (Table S1). The sample size for these traits differs by a few mice because three of the CT scans had scanning errors and were unusable for measuring nutrient canals. Additionally, a few of the mice that were used for measuring nutrient canals did not have measurements for femur and/or body length. In any case, the findings for body mass and femur length are consistent with those reported in Copes et al. (2018), while analysis of body length was not previously reported.

TABLE 1.

Results from mixed models for body mass and femur length (SAS PROCEDURE MIXED).

Variables	n	Transform or modification	Skew	Linetype effects			Activity			Linetype × activity			Mini‐muscle effects			Body length			Slope	Body mass			Slope
				df	F	p	df	F	p	df	F	p	df	F	p	df	F	p		df	F	p
Body mass	94	N/A	0.26	1,6	4.64	0.0748	1,6	3.53	0.1092	1,6	0.62	0.4612	1,77	0.30	0.5836
Body mass	91	N/A	0.15	1,6	4.16	0.0874	1,6	1.07	0.3406	1,6	0.02	0.8842	1,73	0.58	0.4474	1,73	37.60	<0.0001	+
Body length	91	N/A	0.21	1,6	3.78	0.0999	1,6	1.40	0.2810	1,6	2.14	0.1940	1,74	0.69	0.4095
Femur length	89	N/A	−0.18	1,6	0.05	0.8245	1,6	0.08	0.7844	1,6	0.11	0.7520	1,71	0.12	0.7318	1,71	21.46	<0.0001	+
Femur length	92	N/A	−0.57	1,6	0.07	0.7987	1,6	0.06	0.8140	1,6	0.00	0.9460	1,74	0.21	0.6482					1,74	56.85	<0.0001	+

Note: Significance levels (p values; bold indicates p < 0.05, unadjusted for multiple comparisons) from two‐way nested analysis of covariance models implemented in SAS PROC MIXED.

3.2. Basic characteristics of nutrient canals

As noted previously (Schwartz et al., 2018), nutrient canals in mouse femora are diverse in shape, as well as size and number. In the present study, canals varied in shape from straight tubes through the bone to complex curved, looped, or branched canals (Figure 3). Canal numbers ranged from 0 to 5 in the proximal region, 0 to 5 in the distal region, and 1 to 7 in total.

Some canals were bifurcated, but the number was relatively small (7 of 167 canals in the 47 HR mice [4.2%], 29 of 188 canals in the 47 C mice [15.4%]). When analyzed as a 0–1 variable indicating whether a given mouse had any bifurcated canals (SAS PROC GLIMMIX) and with line nested within linetype and no covariates, HR mice (5 of 47 had at least one bifurcated canal) tended to have fewer bifurcated canals (p = 0.0631) than did C mice (18 of 47 had at least one bifurcated canal), with no effect of activity, no linetype‐by‐activity interaction, and no effect of mini‐muscle (Table 3). No variance was associated with line‐within‐linetype, and an analysis without line nested and no covariates indicated a linetype effect (p = 0.0252), again with no other significant main effects. Models with mass as a covariate indicated no effect of body mass (Table 3).

TABLE 3.

Results from mixed models for nutrient canal bifurcations (SAS PROCEDURE GLIMMIX).

Variables	n	Transform or modification	Linetype effects			Activity			Linetype × activity			Mini‐muscle effects			Mass
			df	F	p	df	F	p	df	F	p	df	F	p	df	F	p
Bifurcated number (line nested)	94	N/A	1,6	3.59	0.1070	1,6	1.46	0.2717	1,6	1.88	0.2192	1,76	0.70	0.4049	1,76	0.14	0.7123
Bifurcated number (line nested)	94	N/A	1,6	5.18	0.0631	1,6	1.54	0.2606	1,6	1.81	0.2266	1,77	0.64	0.4252
Bifurcated number (line not nested)	94	N/A	1,88	3.59	0.0614	1,88	1.46	0.2294	1,88	1.88	0.1736	1,88	0.70	0.4046	1,88	0.14	0.7122
Bifurcated number (line not nested)	94	N/A	1,89	5.18	0.0252	1,89	1.54	0.2175	1,89	1.81	0.1814	1,89	0.64	0.4249

Note: Significance levels (p values; bold indicates p < 0.05, unadjusted for multiple comparisons) from mixed models in SAS PROC GLIMMIX analyzing the presence/absence of any bifurcated nutrient canals, with and without body mass as a covariate. If line is not considered as a nested random effect within linetype, then statistical significance is attained for the linetype effect.

Some canals were curved: 63 of 167 canals in HR lines (37.7%) and 88 of 188 canals in C lines (46.8%). Canals were classified as either straight or curved, with straight canals having no noticeable curve or bend, and everything else being classified as curved (curved, looped, etc.). In analyses from SAS PROC MIXED, we found no statistical effects on the percentage of curved canals.

Location of each nutrient canal was recorded, but for ease of analysis, nutrient canals were designated as proximal or distal based on their respective location, and the average proximal and distal locations were calculated. Average distal canal location had no main effects but was associated with body mass (p = 0.0379), with distal canals being located closer to the midpoint of the bone as mass increased. This effect was equally strong in analyses with the activity covariates (p = 0.0283) (Table S1), and in those analyses average distal canal location was more medial as wheel running increased (p = 0.0350).

3.3. Nutrient canal numbers

The total number of nutrient canals per femur was affected by a significant linetype‐by‐activity interaction (Table 1: p = 0.0175). Specifically, for C mice, wheel access increased canal number, whereas for HR mice, wheel access decreased the total number of canals (Figure 4). This interaction also affected proximal and distal numbers of canals in the same manner, but statistical significance was not attained (p = 0.1378 and 0.0949, respectively). Body mass was a negative predictor of proximal number (p = 0.0188) but a positive predictor of distal number (p = 0.0056), resulting in no significant relation with total canal number (p = 0.6773). Results were similar when body mass was not included as a covariate (Table S1).

FIGURE 4.

Nutrient canal characteristics in relation to body mass. Relation between (a) total canal number, (b) proximal canal number, and (c) distal canal number and body mass for mice from four experimental groups (see text). Solid lines are simple least‐squares linear regressions. Body mass was a negative predictor of proximal number (p = 0.0188) but a positive predictor of distal number (p = 0.0056), resulting in no significant relation with total canal number (p = 0.6773). With body mass as a covariate, total canal number was affected by an interaction between linetype and wheel access (p = 0.0175): voluntary exercise increased numbers in C mice, but decreased numbers in HR mice (Table 2) and (d), which shows least squares means and standard errors from SAS procedure mixed). The interaction also affected proximal and distal numbers of canals in the same manner, but statistical significance was not attained (p = 0.1378 and 0.0949, respectively).

Percent distal number (distal canals/total canals) had an effect of linetype (p = 0.0493), with C mice having lower percentage of distal canals and HR mice having higher percentage of distal canals. However, this effect was only present in the analysis with mass, wheel running, and home cage activity as covariates. Percent distal number decreased with home cage activity (p = 0.0330). Without mass as a covariate, this effect lost its statistical significance (p = 0.0822).

3.4. Nutrient canal cross‐sectional areas

Total nutrient canal area, as well as proximal and distal canal area, were unaffected by linetype, activity, linetype‐by‐activity interaction, or mini‐muscle. However, body mass was a significant positive predictor (p = 0.0032) of distal canal area, but not total or proximal area (p = 0.2026 and 0.2186, respectively) (Table 2). Additionally, percent distal cross‐sectional area (distal area/total area) also increased with mass (p = 0.0342). These effects were equally strong when activity covariates were included.

TABLE 2.

Results from mixed models for nutrient canal characteristics (SAS PROCEDURE MIXED).

Variables	n	Transform or modification	Skew	Linetype effects			Activity			Linetype × activity			Mini‐muscle effects			Mass			Slope
				df	F	p	df	F	p	df	F	p	df	F	p	df	F	p
Total area (mm2)	94	N/A	0.33	1,6	0.01	0.9236	1,6	3.40	0.1148	1,6	1.46	0.2730	1,76	0.12	0.7248	1,76	1.65	0.2026
Proximal area (mm2)	93	Outlier	0.20	1,6	0.00	0.9933	1,6	2.37	0.1743	1,6	0.42	0.5405	1,75	1.26	0.2652	1,75	1.54	0.2186
Distal area (mm2)	94	N/A	0.24	1,6	0.02	0.8919	1,6	0.64	0.4539	1,6	2.27	0.1827	1,76	1.06	0.3069	1,76	9.27	0.0032	+
Percent distal CSA	94	N/A	0.04	1,6	0.21	0.6657	1,6	0.01	0.9351	1,6	0.65	0.4510	1,76	0.41	0.5252	1,76	4.65	0.0342	+
Total number	94	Log10	−0.06	1,6	0.72	0.4272	1,6	1.55	0.2593	1,6	10.56	0.0175	1,76	1.02	0.3146	1,76	0.17	0.6773
Proximal number	94	N/A	0.33	1,6	2.71	0.1511	1,6	0.55	0.4877	1,6	2.91	0.1387	1,76	0.97	0.3268	1,76	5.76	0.0188	−
Distal number	94	Log10	0.16	1,6	0.41	0.5471	1,6	0.58	0.4769	1,6	3.92	0.0949	1,76	0.94	0.3351	1,76	8.14	0.0056	+
Percent distal number	94	N/A	0.35	1,6	2.77	0.1469	1,6	0.65	0.4495	1,6	0.64	0.4553	1,76	2.50	0.1180	1,76	10.64	0.0017	+
Average proximal location	93	Log10	0.22	1,6	0.01	0.9217	1,6	1.45	0.2737	1,6	0.40	0.5497	1,75	0.84	0.3611	1,75	0.96	0.3294
Average distal location	88	Power 2	−0.32	1,6	0.60	0.4669	1,6	0.18	0.6880	1,6	0.02	0.8982	1,70	0.72	0.3981	1,70	4.48	0.0379	−
Average CSA per canal	94	Log10	−0.35	1,6	0.57	0.4787	1,6	0.21	0.6619	1,6	1.76	0.2324	1,76	0.16	0.6930	1,76	0.14	0.7051
Proximal avg. CSA per canal	94	Rank	−0.04	1,6	1.18	0.3196	1,6	0.87	0.3859	1,6	0.89	0.3828	1,76	0.07	0.7990	1,76	0.06	0.8079
Distal avg. CSA per canal	94	Rank	0.05	1,6	0.72	0.4300	1,6	0.06	0.8186	1,6	0.09	0.7792	1,76	0.28	0.5972	1,76	0.01	0.9407
Percent curved Canal number	94	Power 0.7	−0.23	1,6	1.25	0.3067	1,6	3.07	0.1303	1,6	0.29	0.6091	1,76	0.46	0.4997	1,76	0.00	0.9623

Note: Significance levels (p values; bold indicates p < 0.05, unadjusted for multiple comparisons) from two‐way nested analysis of covariance models implemented in SAS PROC MIXED.

4. DISCUSSION

We studied the number, location, size, and shape of femoral nutrient canals from four replicate HR lines of house mice that had been selectively bred for voluntary wheel‐running behavior for 57 generations, and compared with four non‐selected control (C) lines. Half of the mice were housed with wheels (active group) and half without (sedentary group) for 12 weeks starting shortly after weaning. With this experimental design, we were able to study evolved differences related to selection for high voluntary wheel‐running behavior (HR vs. C lines), phenotypic plasticity in response to chronic exercise across a key stage of ontogeny (active vs. sedentary), and potential interactions between the two factors.

A previous study examined femoral nutrient canal morphology at generation 11 and found that HR lines had a significantly greater total canal area than C lines. However, the HR lines from that study had not yet reached a selection limit (plateau) for voluntary wheel‐running behavior (which would not occur until ~10 generations later; Careau et al., 2013). Therefore, we expected that the HR versus C difference might have increased by our sampling at generation 57. In addition, presuming that the size (but not number) of canals can change between weaning and the attainment of full skeletal growth, we expected that voluntary exercise (especially in the HR lines) would lead to changes in nutrient canal size. Contrary to our expectations, we found little evidence of differences between the HR and C lines at generation 57, nor of an exercise‐training effect, although we did find a significant linetype‐by‐activity interaction for the total number of canals in the femur (see Section 4.2).

4.1. Summary of nutrient canal characteristics

Historically, nutrient canals have been described as a single foramen, with no branching, oriented at a right angle to the long axis of long bones, which develop an oblique orientation over time due to asymmetric bone growth (Brookes, 1958; Brookes & Harrison, 1957; Greene, 1935; Henderson, 1978; Rogers & Gladstone, 1950; Singh et al., 1991). In general, previous studies of larger‐bodied animals (e.g., humans, pigs, horses) have described most long bones as possessing only one nutrient foramen, though some may have two or none at all (Campos et al., 1987; Carroll, 1963; Payton, 1934). However, more recent studies using micro‐CT scans and 3D modeling software have shown great amounts of variation in both the number and structure of nutrient canals, including the presence of branching in some nutrient canals (Houssaye & Prévoteau, 2020; Hu et al., 2018; Schwartz et al., 2018). In the femur, nutrient canals are most often oriented proximo‐distally, from the inside to the outside of the bone (Houssaye & Prévoteau, 2020). Studies with multiple species have shown a great amount of both inter‐ and intra‐specific variation in the number of nutrient canals per long bone (Houssaye & Prévoteau, 2020). Our previous study of 137 mice found that femurs averaged between four and five nutrient canals and that nutrient canals were located near the proximal or distal ends, with no nutrient canals being found in the middle of the diaphysis (Schwartz et al., 2018). The current study found nutrient canal locations to be the same as previously noted.

As expected from our previous study (Schwartz et al., 2018), we encountered a large diversity of nutrient canal shapes, ranging from straight tubes to curved, looped, or bifurcating canals (see also Houssaye & Prévoteau, 2020). Only three of 94 mice had exclusively straight canals (i.e., nutrient canals that neither curved nor branched); 88 had at least one “curved” canal (which we defined as any canal needing more than one round of re‐orientation during the measurement process, as described in Section 2.6), and 38 mice had more than one “curved” canals. With regard to bifurcation, 71 mice had no branched canals, 21 had one, and only two had two. Although the functional significance of canal bifurcation is unknown, we found that HR mice tended to have fewer bifurcated canals compared to C mice (Table 3, SAS PROC GLIMMIX, p = 0.0631). Something interesting about the aforementioned bifurcation of nutrient canals is that these bifurcations occur within the cortical bone. Although noted in Schwartz et al. (2018), we add here that this phenomenon is odd in that bifurcation before or after entering the cortical bone would be, at least in principle, a simpler and more efficient process. Whether this has any functional implications is currently unknown.

4.2. Nutrient canal numbers

Total nutrient canal number was affected by a significant linetype‐by‐activity interaction (Table 1: p = 0.0175 without correction for multiple comparisons), with wheel access increasing canal number for C mice (+15%) but decreasing it for HR mice (−6%). This result is perhaps surprising, especially given that no such effects, nor indeed any effects of exercise, were found for femur length, cortical cross‐sectional area, or polar moment of inertia (Copes et al., 2018). Similarly, no effects of exercise nor any interactions were found for the sizes of three femoral muscle attachment sites (Castro et al., 2022).

Nutrient canals are first formed during development, when the nutrient artery penetrates the cartilaginous femur prior to endochondral ossification (Ahn, 2013). In mice, ossification of the femur starts at around 14.5 days post‐coitum (Barle & Piano, 2008). No mechanism for new nutrient canals to form after ossification is presently known. Formation of new blood vessels within bone is possible, referred to as intraosseous angiogenesis (Laroche, 2002; Rodrigues et al., 2022; Rumney et al., 2019). However, this process refers to the growth of new capillaries from preexisting capillaries or from post‐capillary venules, and whether or not this process can occur within nutrient canals is unknown (Laroche, 2002). Despite this, we can speculate as to how chronic exercise on wheels might affect the number of nutrient canals. Perhaps mice at weaning have more canals than needed, with some closing as they grow and age to sexual maturity and beyond. If all mice have an excess number of nutrient canals at weaning (consistent with the idea of “momentarily excessive construction” in Gans, 1979), then perhaps those from the relatively low‐activity C lines need to keep more canals to accommodate chronic voluntary wheel running, whereas HR mice, which run much more, need to divert more blood flow to the trabecular bone at the ends of the femur, which occurs via canal closing.

Closing of a canal would decrease blood flow to the center of the femur and thus divert flow to the trabecular bone at the ends of the femur via metaphyseal and epiphyseal blood vessels. Ten weeks of treadmill running in young growing rats resulted in increased trabecular bone mass, from creation of new trabeculae, as well as increased trabecular thickness (Joo et al., 2003). A closing mechanism might involve arteries and/or veins that run through the canals withering, followed by the empty canal being filled in by ossification.

The mechanism of arteries/veins closing could be similar to a process known as vascular rarefaction, which occurs in arterioles and capillaries (Rosei & Rizzoni, 2007). Vascular rarefaction occurs in hypertensive animals (Goligorsky, 2010; Liang et al., 2019). However, the HR mice have not been found to be hypertensive (Kolb et al., 2013), so although the mechanism of nutrient canals closing may not be exactly as follows, we believe it is important to acknowledge the possibility. Rarefaction can occur in two ways, functional and structural rarefaction, where functional rarefaction is a reversible reduction in perfusion and structural rarefaction is an anatomical loss of vessels (Chen et al., 1981). Structural rarefaction is likely preceded by functional rarefaction (Prewitt et al., 1989). However, we know of no evidence that closing of canals occurs in mice after weaning, regardless of the mechanism. This could be an area for future study.

A recent comparative study analyzed nutrient canals in the femur and humerus from 23 different quadrupedal mammal species, including 10 mustelids (Houssaye & Prévoteau, 2020). The study group was phylogenetically diverse, as well as diverse in size, morphology, and method of locomotion, as it included terrestrial, semi‐aquatic, and aquatic organisms (Houssaye & Prévoteau, 2020). Some of the species were represented by more than one femur sample, giving a total sample size of 48 femurs. The number of nutrient canals found in a single femur ranged from 1 to 4, with an average number of 2 canals per femur (mean = 1.96). The species from this study were all larger than the mice in the present study, which had between 1 and 7 nutrient canals (mean = 3.78). Taking averages from each species from their data, ours, a value for rats of one canal (Brookes, 1958; Henderson, 1978; Prisby, 2020; Prisby et al., 2015), and an average of two for humans (Gupta & Ambekar, 2016), the correlation between the average number of femoral nutrient canals and average body length (determined from Wikipedia) was −0.11, which is not statistically significant (N = 25 species but 26 data points [mice represented twice], p = 0.59). Although not discussed in Houssaye and Prévoteau (2020), allometry does not seem to be a factor in explaining interspecific variation in nutrient canal number.

Another possibility to consider would be the presence of metaphyseal canals in the diaphysis, where nutrient canals were being measured. Metaphyseal canals exiting the bone in the diaphysis have been observed by Houssaye and Prévoteau (2020) in multiple species of mammals. In further support of this possibility, the metaphyseal zone in murine femurs extends partly into the diaphysis (Bab et al., 2007). Metaphyseal canals in the diaphysis would be a possible imperfection to the data, which could introduce error in the number and total area of nutrient canals. However, potential metaphyseal canals were inspected by two researchers on the basis that only nutrient canals would be (a) complete from the outer surface of the cortical bone to the medullary cavity and (b) oriented toward the midpoint of the diaphysis (i.e., pointed away from the metaphyseal region of bone). Therefore, we believe that the chance of including metaphyseal canals in this data set is minimal.

4.3. Nutrient canal cross‐sectional areas

Our results did not replicate those of a previous study conducted at generation 11, prior to when the HR lines reached selection limits (plateaus), which found HR mice to have significantly higher total cross‐sectional area of femoral nutrient canals compared to C mice (Schwartz et al., 2018). This discrepancy is perhaps not surprising, given that another study found skeletal differences between HR and C mice to fluctuate across generations (Castro et al., 2021).

4.4. Concluding remarks and future directions

In the present study of 16 traits related to femoral nutrient canals, we found no statistically significant (p < 0.05) effects of linetype, activity, or the mini‐muscle phenotype, and only a single linetype‐by‐activity interaction (p = 0.0175). This number of significant effects (1/64 = 1.6%) is lower than observed for other bone traits studied in these same individual mice, especially for the mini‐muscle effect (Table 4). Tallying across three previous studies that reported 22 bone traits (Castro et al., 2022; Copes et al., 2018; Lewton et al., 2019), the number of traits with p < 0.05 was 0/22 for linetype, 3/22 (14%) for activity, 0/22 for the linetype‐by‐activity interaction, and 8/22 (36%) for the mini‐muscle phenotype (grand total 11/88 = 12.5%). With respect to body mass as a covariate, we also found a smaller number of significant effects (4% here vs. 86%) (Table 4). Taken together, these results indicate that, at least for these mice and this type of exercise exposure, nutrient canals are both less phenotypically plastic and less likely to respond evolutionarily to selection for increased locomotor activity than many other osteological traits. However, phenotypically plastic characteristics of nutrient canals may be more closely related to factors in embryonic development when bone is first changing from cartilage into ossified bone, a life stage that was not examined in the current study. In addition, characteristics might have changed at the histological level (e.g., bone remodeling), but again we did not examine that here.

TABLE 4.

Summary of results for bone traits analyzed for this set of mice.

Reference	Linetype (HR vs. C lines)	Activity	Linetype‐by‐activity interaction	Mini‐muscle	Body mass
Copes et al. (2018)	0/12 (0%)	0/12 (0%)	0/12 (0%)	2/12 (17%)	10/12 (83%)
Lewton et al. (2019)	0/6 (0%)	2/6 (33%)	0/6 (0%)	5/6 (83%)	6/6 (100%)
Castro et al. (2022)	0/4 (0%)	1/4 (25%)	0/4 (0%)	1/4 (25%)	3/4 (75%)
This study of canals	0/14 (0%)	0/14 (0%)	1/14 (7%)	0/14 (0%)	6/14 (43%)
Total	0/36 (0%)	3/36 (8%)	1/36 (3%)	8/36 (22%)	26/36 (72%)

Note: Summary of statistical results for four studies that have measured bone traits in the same set of female mice from generation 57 of the high runner selection experiment. Overall, the present study found fewer statistically significant effects than the other three studies. Cell entries are the number of p values <0.05 divided by the total number of traits. Body mass was used as a covariate in all analyses compared here.

Abbreviations: C, controls; HR, high runner.

A possible explanation for the lack of skeletal change induced by 12 weeks of wheel‐running in these mice is that their bones were affected by activity, just not in traits that were measured. For example, forced treadmill exercise for 5 weeks in young mice increased tibial bone strength and post‐yield behavior without significant changes in bone mass or architecture (Gardinier et al., 2018). More specifically, no exercise effects were found for cortical area or polar moment of inertia, which is consistent with the findings of Copes et al. (2018), who analyzed the femur and humerus. However, Lewton et al. (2019) did find that activity increased the cortical area of the ilium. Bone strength and post‐yield behavior have not been studied in this set of mice, and given the way the bones were cleaned and stored, it will be impossible to do so in the future, so any effect of activity on those traits in this specific set of mice is unknown.

Although the nutrient artery is the primary source of blood to long bones, metaphyseal and epiphyseal arteries also supply blood to long bones (Brookes, 1958; Gümüsburun et al., 1994; Prisby, 2020; Trueta, 1963). For example, when the nutrient canal of day‐old rabbits was occluded, adult femurs were only 3% shorter compared to controls (Brookes, 1957). Thus, the metaphyseal and epiphyseal arteries were able to accommodate and supply most of the blood that would have otherwise been provided by the nutrient artery (Brookes, 1957). Additionally, mice have hundreds of capillaries that fully cross the cortical bone from the endosteum to the periosteum (Grüneboom et al., 2019). Perhaps these transcortical vessels were able to supplement the blood flow to long bones in addition to the nutrient artery. The current study used 12‐μm resolution CT scans, which are not sufficient to detect capillaries, which are usually 8–10 μm in diameter. Some studies suggest that the periosteum, the outer layer of long bones, may be permeable in certain conditions (Evans et al., 2013; Li et al., 1987; Qin et al., 2003). For example, the periosteum was found to increase its permeability with loading when compared to unloaded bone, as well as exhibiting differences in directional permeability, dependent on flow rates (Evans et al., 2013; Knothe Tate et al., 1998). If the periosteum is permeable, then nutrients and waste products could enter and exit through the bone itself instead of having to pass through the nutrient canal. This would reduce or neutralize the need for nutrient canal plasticity to accommodate changing demands in bone growth and remodeling. However, this diffusion mechanism of transport is likely more useful for transporting small molecules, such as amino acids, rather than larger molecules, such as proteins (Knothe Tate et al., 1998). Additionally, the permeability of the periosteum in a mouse model has not yet been studied. Future studies should examine the adaptability of the bone's blood supply, and how this adaptability varies depending on genetic or environmental factors.

How nutrient canals can change in size and number throughout development needs further investigation. For future studies, vascular contrast perfusion in conjunction with CT scans (or perhaps contrast‐enhanced magnetic resonance imaging) could be used to more precisely study nutrient canals in vivo and the various nerves, arteries, and veins that pass through them. Vascular contrast perfusion has been used in chickens to study the size of the nutrient artery in relation to its nutrient canal (Hu et al., 2022), but chickens differ in the number of nutrient canals compared to mice, with a maximum of three per femur and most femurs having only one nutrient canal. 18F‐labeled sodium fluoride ([¹⁸F]NaF) imaging with positron emission tomography could be utilized to investigate possible shifting of perfusion and remodeling activity between the diaphysis and metaphyseal region of the femur. In addition, microsphere injection could be used to measure blood supply to the bone. Another area that needs further investigation is the functional significance of nutrient canal number.

Although nutrient canal number clearly varies both among and within species (Houssaye & Prévoteau, 2020; Schwartz et al., 2018; present study), symmetry between bones has not yet been examined. Given that the asymmetry of hindlimb bone length was found to be reduced in the HR lines of mice at generation 11, this would be an interesting area for future research.

AUTHOR CONTRIBUTIONS

All authors designed the research. L.E.C. and T.G. produced the mice. L.E.C. obtained the CT scans. B.B.T. and N.E.S. designed the methodology used to acquire data and acquired the data. B.B.T. and T.G. performed the analyses and drafted the manuscript. All authors revised the manuscript.

FUNDING INFORMATION

Funding was provided by U.S. NSF grants DDIG 0925793 to Lynn E. Copes, as well as IOS‐1121273 and IOS‐2038528 to Theodore Garland, Jr.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.

Supporting information

Table S1.

Data S1.

Tan, B.B. , Schwartz, N.E. , Copes, L.E. & Garland, T. Jr. (2024) Effects of long‐term voluntary wheel running and selective breeding for wheel running on femoral nutrient canals. Journal of Anatomy, 244, 1015–1029. Available from: 10.1111/joa.14021

DATA AVAILABILITY STATEMENT

Data are available on request from the authors.