Bone's mechanostat: A 2003 update


  • Harold M. Frost

    Corresponding author
    1. Department of Orthopaedic Surgery, Southern Colorado Clinic, Pueblo, Colorado
    • Department of Orthopaedic Surgery, Southern Colorado Clinic, 3676 Parker Blvd., Pueblo, CO 81008-9000
    Search for more papers by this author


The still-evolving mechanostat hypothesis for bones inserts tissue-level realities into the former knowledge gap between bone's organ-level and cell-level realities. It concerns load-bearing bones in postnatal free-living bony vertebrates, physiologic bone loading, and how bones adapt their strength to the mechanical loads on them. Voluntary mechanical usage determines most of the postnatal strength of healthy bones in ways that minimize nontraumatic fractures and create a bone-strength safety factor. The mechanostat hypothesis predicts 32 things that occur, including the gross anatomical bone abnormalities in osteogenesis imperfecta; it distinguishes postnatal situations from baseline conditions at birth; it distinguishes bones that carry typical voluntary loads from bones that have other chief functions; and it distinguishes traumatic from nontraumatic fractures. It provides functional definitions of mechanical bone competence, bone quality, osteopenias, and osteoporoses. It includes permissive hormonal and other effects on bones, a marrow mediator mechanism, some limitations of clinical densitometry, a cause of bone “mass” plateaus during treatment, an “adaptational lag” in some children, and some vibration effects on bones. The mechanostat hypothesis may have analogs in nonosseous skeletal organs as well. Anat Rec Part A 275A:1081–1101, 2003. © 2003 Wiley-Liss, Inc.

So much has happened since the last description of the seminal mechanostat hypothesis (Frost, 1996) that an update of its evolving status seemed desirable. Parts I and II of this article describe the chief mechanical function of the biologic machinery that determines the postnatal strength of most of our bones, how that machinery works, and what would comprise bone's mechanostat. Part III provides predictions of the mechanostat hypothesis, along with comments about its bearing on some clinical and research matters, and on searches for the cellular and molecular-biologic support for its tissue-level functions.

Table 1 defines frequently-used abbreviations in this article.

Table 1. Frequently used abbreviations
BMU: Basic multicellular unit of bone remodeling.
Fx: a bone's fracture strength, its ultimate strength.
LBB: a load-bearing bone on which the major loads are TPVMLs.
MBC: mechanical bone competence.
MDx: microscopic fatigue damage in bones.
MESm: bone's modeling threshold range (“MES” here and below stands for Minimally Effective Strains or other Stimuli (Frost, 1990a)). Genetic information may encode that threshold in some cell or cells.
MESp: bone's operational microdamage threshold range. Genetic information may provide that threshold.
MESr: the threshold range for disuse-mode BMU-based bone remodeling.
TPVML: typical peak voluntary mechanical load on a load-bearing bone during typical physical activities (it implies muscle forces and excludes loads from ultrasound and trauma, and from rare but exceptionally strenuous voluntary activities such as jumping from great heights).


1) Four observations helped to suggest the mechanostat hypothesis. (A) In all contemporary bony vertebrates of any size, age, and sex, healthy load-bearing bones (LBBs) and their trabeculae have more strength than is needed to keep typical peak voluntary mechanical loads (TPVMLs) on them from causing nontraumatic fractures, which are often called “spontaneous” fractures because they are not caused by injuries. In this article “TPVMLs” refers to the largest repeated and intentional loads on bones exerted by intentional activities during a typical week or month. Thus, TPVMLs would exclude loads caused by injuries, rare actions (such as jumping from great heights), or high-frequency vibration. Consequently, TPVMLs would come mainly from intentional skeletal muscle contractions. (B) In all bony vertebrates, the lifelong number of traumatic fractures far exceeds the lifelong number of nontraumatic fractures. (C) As experienced physicians know, situations A and B persist during most debilitating diseases that reduce TPVMLs. (D) The fossil record suggests that situations A and B existed at least by the beginning of the Cretaceous period, ∼145 million years ago (Enlow and Brown, 1958; Romer, 1966).

2) Here Question 1 becomes pertinent: What could constitute the chief mechanical function of bone's biologic machinery?

Observations such as those in Section 1 above suggested this answer: Bone's biologic machinery would make healthy postnatal human load-bearing bones (LBBs) and their trabeculae strong enough to keep typical peak voluntary mechanical loads (TPVMLs) from breaking them suddenly or in fatigue, whether those loads are chronically subnormal, normal or supranormal, or chronically small or large. That statement could define “mechanical bone competence” (MBC), and could apply equally well to the bones of mice, humans, elephants, and dinosaurs. Some of its implications follow.

(i) MBC would depend on the three-way relationship between a bone's strength, the size and kinds of the TPVMLs on it, and nontraumatic fractures caused by those TPVMLs.

(ii) Stress and pathologic fractures excepted, diseased bone-biologic machinery would cause a failure to achieve MBC. Nontraumatic (spontaneous) fractures would herald such failures (Frost, 2002a) because all traumatic fractures are caused by injuries. Nontraumatic fractures affect both extremity bones and the spine in osteogenesis imperfecta (Sillence, 1981; Rauch et al., 2000) and idiopathic juvenile osteoporosis (Dent, 1977; Dimar et al., 1995). Yet in some pre- and postmenopausal women and in some aging men, nontraumatic “fractures” only affect thoracic and lumbar vertebral bodies (see Section 4 in Part III) (Eastell et al., 1991; Riggs and Melton, 1995; Marcus et al., 1996). Quotes embrace those vertebral “fractures” because “ … the majority of (such) spine fractures are asymptomatic … ” (Siris et al., 2001; p. 2819), and experienced orthopedists would realize that is very unusual for an acute traumatic fracture of any spinal bone. Accordingly, questions arise as to whether these “fractures” are true acute fractures (Eastell et al., 1991; Frost, 1998c, 2002a); most of them could represent slow fatigue changes of vertebral body morphology.

(iii) MBC could provide a functional definition of bone “quality” that supplements other definitions (Luck, 1950; Bradenburger, 1993; Heaney, 1993; Riggs and Melton, 1995; Boivin and Meunier, 2002; Burr, 2002; Weinans, 2002).

(iv) MBC could also define an LBB's health in functional terms instead of in terms of its bone “mass,” bone mineral “density,” T or Z score, strength, stiffness, fragility, “crystallinity” (Boskey, 2002), etc.

(v) In physiologic importance, the above mechanical functions of LBBs would rank above nonmechanical functions, such as homeostasis (Parfitt, 1993, 1996) and hematopoiesis.

3) Here Question 2 becomes pertinent: How do bones achieve MBC? Nine features (i–ix below) would let most bones achieve it (the following discussion ignores trauma, longitudinal bone growth, cranial sutures (Mao, 2002), and synovial joints).

(i) Their chief functions show we have two kinds of bones. (A) Healthy LBBs and their trabeculae have enough strength to carry TPVMLs without letting those loads break them suddenly or in fatigue. Human LBBs include femurs, tibias, vertebrae, scapulae, humeri, radii, phalanges, zygomas, mandibles, metacarpals, etc., so they are not limited to weight-bearing bones. (B) However, a few bones have other chief functions. These bones include the cranial vault, nasal bones, turbinates, cribriform plates of the ethmoids, and inner ear ossicles. Different rules appear to determine the strength of the “A” and “B” bones (Frost, 1987a; Rawlinson et al., 1995; Currey, 2003). It seems noteworthy that in all “osteoporoses,” no matter how one defines them, the so-called osteoporosis fractures only affect the “A” bones (meaning the LBBs).

This mechanostat update concerns the LBBs.

(ii) The maternal environment may exert some effects, but chiefly gene expression patterns in utero created our body's baseline conditions by the time of birth (Hanes and Mohidden, 1965; Cruess, 1982; Moore, 1988). Those conditions include our basic bony anatomy and anatomical relationships, our basic neurologic and muscular anatomy and physiology, and the biologic machinery that can increase the strength of LBBs after birth, in part by adding adaptations to postnatal TPVMLs to the baseline conditions (see Section 4, Subpart 2 in Part II below).

That biologic machinery includes the next seven tissue-level features. (iii) A multicellular mechanism called “modeling by formation and resorption drifts” (Jee, 1989), when averaged over a whole bone, can increase an LBB's strength (Frost, 2001a). This sluggish process can take many weeks. (iv) Over weeks, months, or years, another multicellular mechanism called “disuse-mode BMU-based remodeling” (Jee, 2001) can decrease a hollow LBB's strength by removing some endocortical and trabecular bone (Jaworski et al., 1980; Frost, 1998b). Bone modeling and remodeling have other functions as well (Frost, 2001a). (v) After birth, dynamic strain-dependent bone signals (Fukada and Yasuda, 1957; Rubanyi et al., 1990; Martin et al., 1998; Burger and Klein-Nuland, 1999; Lanyon and Skerry, 2001) can reveal how each LBB's strength relates to the size of the TPVMLs on it. Persistently small signals would indicate too much bone strength for those TPVMLs and a large “bone-strength/bone-load” ratio, while very large signals would indicate too little bone strength for those TPVMLs and a small ratio.

(vi) Aided by signaling systems that detect and process the above signals (an activity often called “mechanotransduction” (Doty, 1981; El Haj, 1990; Weinbaum et al., 1994; Marotti, 1996; Martin et al., 1998; Skerry, 2002), threshold ranges of those strain-dependent signals (the MESm for modeling and the MESr for disuse-mode remodeling) apparently help to switch the two whole-bone-strength functions on and off; see Sections 8B and 9 (ii) in Part III (Torrance et al., 1994; Martin et al., 1998). “Whole bone” distinguishes bones as organs from bone as a tissue or structural material. Figure 1 shows how these features would usually affect an LBB's postnatal strength. The strain span between the MESr and the MESm in Figure 1 could define two things: a region of naturally acceptable whole-bone strength relative to the TPVMLs on a bone, and the span of a normal “bone-strength/bone-load” ratio. In Part III, Section 2 concerns that ratio and Section 3 concerns the evaluation of whole-bone strength and that ratio. The MESm would designate the strain region where mechanically-controlled modeling begins and above which its activity could increase, while the MESr would designate the strain region in and below which maximal disuse-mode remodeling activity occurs, and above which it could begin to decrease.

Figure 1.

Combined modeling and remodeling effects on LBB strength. The horizontal line at the bottom suggests typical peak bone strains from zero on the left, to the fracture strain on the right (Fx), plus the locations of the remodeling, modeling, and microdamage thresholds (MESr, MESm, and MESp, respectively). The horizontal axis represents no net gains or losses of an LBB's strength. The lower dotted line curve suggests how disuse-mode remodeling would remove bone next to marrow when strains stay below the MESr range, but otherwise would tend to maintain existing bone and its strength. The upper dashed line curve suggests how modeling drifts would begin to increase bone strength where strains enter or exceed the MESm range. The dashed outlines suggest the combined modeling and remodeling effects on a bone's strength. Such a curve was originally suggested by Carter (1984). Beyond the MESp range, woven bone formation usually replaces lamellar bone formation (Frost, 2001a). At the top, DW = disuse window; AW = adapted window, as in normally adapted young adults; MOW = mild overload window, as in healthy growing mammals; and POW = pathologic overload window (Frost, 1992a). The strain span between the MESr and MESm represents the span between those features in bone's general biomechanical relation. This scheme omits possible nonmechanical effects on muscle and the nervous system (reproduced by permission (Frost, 1997)).

Lanyon and Smith (1970) initiated the in vivo bone strain studies that helped to reveal many of these things.

(vii) Repeated bone strains can cause microscopic fatigue damage in bones (microdamage (MDx)) (Schaffler et al., 1995; Burr and Milgrom, 2000). Bone MDx has a threshold strain range (the MESp) (Burr et al., 1983, 1997; Martin, 1995, 2000). Normally, living LBBs can detect and repair the small amounts of MDx caused by strains that stay below the MDx threshold. Osteocytes could provide that detection (Frost, 1963; Verborgt et al., 2000), and remodeling BMUs provide at least most of that repair (Mori and Burr, 1993; Burr and Milgrom, 2000). Yet strains in and above the MESp range can cause enough MDx to escape repair and accumulate (Martin, 2000). Normally the MESm's magnitude would lie below the MESp's magnitude, which would lie below an LBB's ultimate strength, Fx (thus MESm < MESp < Fx). Accumulated MDx causes or helps to cause pathologic fractures; nontraumatic fractures in true osteoporoses (see Section 4 in Part III) (Frost, 2001a); stress fractures in athletes, special forces trainees, and horses (Burr and Milgrom, 2000); pseudofractures in osteomalacia (Snapper, 1957; Aurbach et al., 1992); subchondral bone collapse in idiopathic aseptic necroses of the femoral head (Frost, 1986); osteochondritis dissecans (Nambu et al., 1988); and nontraumatic fractures of large load-bearing bone allografts used in some tumor surgeries (Aho et al., 1994) and in some revisions of total joint replacements (Tsahakis et al., 1994).

Learning how to minimize MDx in such situations could have therapeutic uses. As an aside, bone MDx depends very sensitively on load magnitudes. Doubling the loads that originally cause 2,000 microstrain in tension or compression can increase bone MDx by more than 400 times (Pattin et al., 1996). Loads that fracture a healthy bone cause strains centered near 25,000 microstrain in young adults.

Mechanotransduction and MDx have become separate fields of study in bone physiology and biomechanics. Mechanotransduction should participate in other features of bone physiology as well. Table 2 lists the nature and number of these features.

Table 2. Bone features that should depend on some kind of mechanotransduction (which nonmechanical agents could also affect)
Replacing primary spongiosa with secondary spongiosa beneath growth plates and in some tumors.
Detecting MDx.
Repairing MDx.
Creating new remodeling BMUs.
Controlling the switch between disuse-mode and conservation-mode remodeling.
Creating modeling formation drifts.
Creating modeling resorption drifts.
Controlling the switch that determines if woven bone, plexiform bone or lamellar bone formation occurs.
Determining the values of bone's three thresholds (MESm, MESr, MESp).
Determining different bone responses on different bone envelopes.
Turning the formation of fracture callus on.
Replacing fracture callus with lamellar bone.
Turning the regional acceleratory phenomenon on and off.

(viii) Except in cases of trauma, lever-arm and gravitational effects cause muscles to put the largest loads on the LBBs, even on weight-bearing bones (Inman, 1947; English and Kilvington, 1979; Afoke et al., 1980; Currey, 1984; Kannus et al., 1996; Martin et al., 1998; Jee, 1999). Thus bone's biologic machinery would adapt a postnatal LBB's strength more to muscle strength (and perhaps to muscle power) than to body weight or other sources of bone loads (Burr, 1997; Anderson et al., 2000). One can express momentary muscle strength in Newtons, and muscle power in Newton-meters/sec, Joules/sec, or watts. For such reasons the dynamic TPVMLs on a soccer player's femur during practice or a game can often (albeit briefly) exceed 5× body weight (Crowninshield et al., 1978; English and Kilvington, 1979; Lu et al., 1997). Muscle forces cause the TPVMLs mentioned above, and in aging humans they tend to decrease in size and power, as well as in their 20–60-Hz frequency components (Aniansson et al., 1984; Burr, 1997; Rittweger et al., 2000; Rubin et al., 2003). Neither genes nor bones could foresee and adapt to one-time loads from injuries or from rare but exceptionally vigorous activities. Natural selection during evolution may have minimized fractures from such loads (Currey, 2003), and Section 2 in Part III describes a possible mechanism for this.

(ix) A “general biomechanical relation” (GBR) can indicate how the magnitudes of some of the above features would normally ladder in relation to each other (Frost, 2000a). MESr indicates the threshold strain range below which the mechanically-controlled disuse-mode remodeling function of decreasing a hollow LBB's strength usually acts maximally, and above which it begins to decrease (conservation-mode remodeling may replace the disuse mode when typical peak bone strains rise to or above −400 microstrain). “E” indicates the typical peak dynamic strains caused by TPVMLs on a healthy LBB; MESm indicates the strain threshold range in and above which the mechanically-controlled modeling function of increasing a bone's strength usually begins (its activity would increase as repeated strains increased in size toward the MESp); MESp indicates bone's MDx threshold range in and above which unrepaired MDx can begin to accumulate; and Fx indicates an LBB's fracture strength.

Hence the GBR: MESr < E < MESm < MESp < Fx. Because its entries constitute ranges with uncertain widths, in a first approximation the centers of those ranges could define their “set points” as the corresponding strains, stresses, or unit loads in Table 3. One could also write the GBR as: Fx > MESp > MESm > E > MESr.

Table 3. Set point values for bone's thresholds and ultimate strength (in microstrain, stress and unit-load terms)*
  • *

    mpa = megapascal = 106 Newtons/meter2. kg/mm2 = “unit loads”. Unit loads on a bone correspond to the stresses they cause in the bone. The above values apply mainly to cortical lamellar bone in healthy young adult mammals, based on currently available information. The values might vary somewhat in different individuals in the same species, and perhaps in different species too. The values show that above the yield point bone strains and stresses do not stay linearly proportional to each other. While some suggested the typical peak allowed bone stress could correspond to ∼60–75 mpa, such values did not depend on “TPVMLs” as this text defines them in Part I. Nor did they account for the “adaptational lag” in developing whole-bone strength during childhood (Frost, 1997), nor for the distinction between physiologic and nonphysiologic loading mentioned in Part I, Section #4, C–E of this article.

MESr: 50–100 microstrain; ∼1–2 mpa, or ∼0.1 kg/mm2 (one can argue for a value of ∼400 microstrain (Frost, 2000a)).
MESm: 1000–1500 microstrain; ∼20 mpa, or ∼2 kg/mm2.
MESp: ∼3000 microstrain; ∼60 mpa, or ∼6 kg/mm2. This also approximately equals bone's yield point (Biewener, 1993; Martin et al., 1998), and the point where woven bone formation begins to replace lamellar bone formation.
Fx: ∼25,000 microstrain in young adults (a bit more during growth and a bit less in aging adults); ∼120 mpa or ∼12 kg/mm2 in cortical bone of healthy young adult mammals.

The GBR could define some constraints an LBB should satisfy to achieve MBC. Presumably, these constraints would have the same rank and magnitude in small and large LBBs in a given species, and perhaps in different species as well.

4) Related Observations

(A) Healthy LBB design may rank minimizing fatigue failures above providing great whole-bone strength (Alexander, 1984; Carter, 1984). This may be at least one function of the (E < MESm < MESp) arrangement in the general biomechanical relation. Failure to minimize fatigue fractures might allow frequent fatigue fractures before reproductive age prevent the survival of an affected species during evolution. (B) Researchers still study how strain magnitudes, rates, frequencies, total numbers, kinds (including shear), modes of vibration, continuous and intermittent loads, and other stimuli might affect an LBB's biologic machinery and strength (Fukada and Yasuda, 1957; Becker et al., 1964; Currey, 1984; Wood, 1987; Beaupre et al., 1990; O'Cormor et al., 1982; Huiskes, 1995; Flieger et al., 1998; Mullender et al., 1998; Lanyon and Skerry, 2001; Rubinacci et al., 2000; Rubin et al., 2002a,b, 2003; Hamrick, 2003; Rosen and Wuster, 2003; Torvinen et al., 2003). Such things represent some of the “devils in the details” of the mechanostat hypothesis. At present their roles in bone physiology are unclear.

(C) In healthy free-living bony vertebrates, physiologic muscle loads on LBBs have the properties of magnitude, rate of increase in magnitude, frequency of loading events, accumulated number of loading events, and power. This mechanostat update concerns physiologic-loading effects on LBBs.

(D) Some nonphysiologic loading modes include large one-time loads from trauma, vibration of upper extremity bones during jack-hammer operation (Kolar et al., 1965), exceptionally vigorous but rare actions (such as jumping from a great height), head-butting by some goats, vibration of other bones, and ultrasound. Some of these nonphysiologic modes may have therapeutic uses. For example, ultrasound treatment probably can improve bone healing (Glazer et al., 1998; Rubin et al., 2002a), and it may increase trabecular bone “mass” and strength (Rubin et al., 2002b), and/or decrease trabecular bone loss (Flieger et al., 1998; Rubin et al., 2001). Ultrasound causes tiny bone loads and strains at very high frequencies and strain rates, and it can cause huge cumulative numbers of loading events (see part E below). Because the LBBs of free-living animals should seldom if ever experience that loading mode, one could view it as “nonphysiologic” even if it had therapeutic uses.

(E) As some authors note (Torvinen et al., 2003; p. 877), others suggested that “ … extremely low magnitude strains … can efficiently increase bone mass … if applied at high frequency.” However, while 10–20 large loads and strains per week can increase bone strength after a few weeks (Umemura et al., 1997; Lanyon and Skerry, 2001), and even one loading session might do that (Li et al., 2002), it takes many hundreds of thousands of repeated very small strains from low-magnitude vibrations to have that kind of effect, and that can take months. Some examples follow: (i) A 50,000-Hz ultrasound stimulus for 5 min, 5 days a week, for 4 months would produce 1.2 × 109 loading events on a bone (each vibration pulse would constitute one loading event). (ii) In one study, a 30-Hz vibration for 20 min a day, 5 days a week, for 12 months produced over 8,000,000 loading events on sheep bones, which led to some increase in epiphyseal trabecular “mass” but not in cortical bone “mass” (metaphyseal trabecular bone “mass” was not evaluated in this study) (Rubin et al., 2002b). (iii) In another study, a ∼30-Hz vibration for 4 min a day, four times a week, for 8 months produced over 1,000,000 loading events on human bones, which did not detectably increase cortical or metaphyseal trabecular bone “mass” (epiphyseal trabecular bone “mass” was not evaluated in this study) (Torvinen et al., 2003). In Part III, Section 1, Group 1 (9) an explanation is suggested for the apparent discrepancy in those results.

While some researchers have suggested that the 20–60-Hz frequency components in normal skeletal muscle contractions could increase whole-bone strength as “efficiently” as a few large TPVMLs per week, this notion conflicts with the above observations and with the fact that some very active humans who lack very strong muscles, such as marathon runners, lack the whole-bone strength of weight lifters. Thus the “devils” in this matter's details need clarification, and using “efficient” in the above context might startle the publishers of Webster's dictionary.

Nota bene: Present evidence indicates that bone's biologic machinery can respond to both large and small repeated loads, but it clearly ranks the importance of a few large loads per week far above equal numbers of very small loading events per week. As noted above, the roles of strain rate and kind, including shear, remain to be determined.

The failure to distinguish physiologic from nonphysiologic loading modes, and physiology from other things that might have therapeutic uses, helped to create controversies about this matter (but may have yielded new insights too). Discussions in some articles, including some of mine (mea culpa; we live and learn and, one hopes, learn to view the errors of others as we hope they will view ours), illustrate this problem (Frost, 1987b, 1990ac; Höcker, 1995; Rubin and McLeod, 1995; Burr et al., 1996, 2002; Ohashi et al., 2002; Turner, 2002a).


Graphic relationships can suggest the feedback between the above parts of bone's biologic machinery described above. Hence, see Relation 1 in Figure 2 and Relation 2 below. Because a bone's strength correlates strongly with its stiffness, in most circumstances either term can provide a useful surrogate for the other (Yamada, 1970; Currey, 1984, 2003; Ferretti et al., 2002).

Figure 2.

Relation 1. In this relation, boldface capitals denote the mechanically-dedicated message traffic, and BONE indicates an LBB. CNS, central nervous system; PNS, peripheral nervous system; PNE, peripheral nerve endings; MC, muscle contraction forces; MU, mechanical usage with its TPVMLs and other loads on LBBs. “L” signifies local nonmechanical agents (genes, cytokines, ligands, receptors, paracrine and autocrine effects, apoptosis, etc.). “S” signifies systemic, blood-borne nonmechanical agents (hormones, minerals, vitamins, drugs, nutrients, etc.). The “highways” include any mediators and modulators of the strain-dependent signals, plus the threshold ranges that help to control the modeling and remodeling bone-strength functions. The mechanical feedback loop (m) concerns bone modeling by drifts, while the mechanical feedback loop (r) concerns bone remodeling by BMUs. Italics signify nonmechanical things that could modulate the mechanically-dedicated message traffic without participating in it. Many such things act as permissive agents that bone adaptations to mechanical influences need in order to occur (see Section #5 in Part III). Arrows indicate the “message traffic” routes and the accompanying feedback (reproduced by permission (Frost, 1996)).

1) The relationships in Figure 2 suggest the feedback that connects the various parts of bone's biologic machinery. The simpler scheme in Relation 2 may help to elucidate the rather “busy” (complicated) Figure 1. Boldface type identifies the mechanical feedback loop in both relations.

Thumbnail image of

In Relation 2 some nonmechanical agents could affect the central nervous system (CNS) and the muscles that cause TPVMLs on bones, as well as the dynamic strain-dependent signals that would help to control bone's biologic mechanisms (Otter et al., 1985; Forwood and Turner, 1995; Lanyon and Skerry, 2001). Some agents that affect both muscle and bone include growth hormone (Chen et al., 1995; Yeh et al., 1994; Forwood et al., 2001), androgens (Wilson and Foster, 1992), calcium (Wortmann, 2000), and vitamin D (Snapper, 1957; Avery and First, 1993; Vaughn et al., 1975; Lee et al., 1996; Glerup et al., 2000).

2) Now Question 3 seems pertinent: What comprises bone's mechanostat? A combination of all the features in Parts I and II above may form the mechanostat. If so, this combination could substitute for bone's “biologic machinery” throughout this article. Michael Parfitt called it the “ … most important unsolved problem in bone physiology … ” (Parfitt, 2000; p. 10).

3) Related Observations

(A) Similar to the functions of nephrons, the mechanostat's functions require explanation at the cellular and molecular-biologic levels, and its functions should depend on many kinds of cells and (probably) on numerous genes (Ferster and Spruston, 1995; Frost, 2003a). (B) The mechanostat's responses to dynamic strain-dependent “error signals” from an LBB would tend to reduce any mechanical error and the signals it caused. Such behavior characterizes error-driven negative feedback systems (Wiener, 1964; Regling, 1993), which bone's mechanostat should exemplify. (C) While that mechanostat's presumed chief function would minimize nontraumatic fractures, the “MESm < Fx” situation in the general biomechanical relation would create a strength-safety factor for LBBs that could minimize traumatic fractures as well, as described in Section 2 of Part III. Because the lifelong number of traumatic fractures of LBBs in all bony vertebrates far exceeds the lifelong number of nontraumatic fractures, LBBs minimize nontraumatic fractures more efficiently than traumatic ones. (D) LBBs are made from very fatigue-prone lamellar bone (Schaffler et al., 1995; Pattin et al., 1996; Burr and Milgrom, 2000). However, fatigue fractures per 10,000 LBBs per year are uncommon in healthy bony vertebrates, including humans, so in life their healthy LBBs have very long fatigue lives.

Hence Question #4: How could fatigue-prone bones have very long fatigue lives?

In answer, living bone's ability to detect and repair limited amounts of MDx, plus the MESm's ability to keep an LBB's typical peak strains below bone's MDx threshold (MESm < MESp) and thereby limit MDx to amounts its repair mechanism can handle, may cause very long fatigue lives. Parenthetically, if osteocytes detect and signal the presence of MDx, the lack of live osteocytes in necrotic bone (which includes bone allografts) should prevent MDx detection and repair, and thus load-bearing necrotic bone in living subjects should be (and is) more prone to fatigue fractures than living bone.

(E) If an LBB's chief mechanical function is to have enough strength to keep TPVMLs from breaking the LBB, it is intriguing that signals that depend on the relationship between the LBB's stiffness and its TPVMLs would control most of its strength. (F) The mechanostat could provide optimal whole-bone strength with a minimum amount of structural tissue, and it should limit peak strains caused by TPVMLs on LBBs. In vivo strain gage studies show that limitation does occur, as noted in (I) below. (G) As noted years ago (Frost, 1986, 1987c), the strain-dependent signals that help to control the bone-strength functions of modeling and disuse-mode remodeling have a microstructural basis that mechanically amplifies bone strains by more than 30× (Cowin and Weinbaum, 1998). (H) Most trabeculae in LBBs transfer the loads from TPVMLs back and forth between cortical bone on the one hand, and joints or tooth sockets on the other hand. When these trabeculae do not carry TPVMLs, disuse-mode remodeling usually removes them slowly, which may explain why diaphyses contain little or no spongiosa (Frost and Jee, 1994a,b; Frost, 2002b) (an exception is the trabeculae in our cranial vault, which is one of the “B” bones mentioned in Section 3 (i) in Part I of this article). (I) “Connecting the dots” in many in vivo strain-gage studies showed that in healthy young-adult, free-living mammals, TPVMLs rarely caused bone strains above 1,000–1,500 microstrain in LBBs. This should indicate the region of bone's MESm and put it below both bone's MESp and its yield point, and help to validate “MESm < MESp” in the general biomechanical relation above. However, strenuous activities may cause larger strains in rapidly growing mammals (Nunamaker et al., 1990). For comparison, healthy bone's ultimate strength expressed as a strain (when bone is loaded in compression parallel to its grain) is centered near 25,000 microstrain in young adults (Table 3), but is somewhat higher during growth and somewhat lower in aged adults (Yamada, 1970).

4) Recapitulation

Subpart 1.

An elegant engineering stratagem would make the TPVMLs on LBBs and their trabeculae determine most of their postnatal strength. Cybernetic considerations (Wiener, 1964) suggest that this would require at least five things: (i) Mechanisms that could change whole-bone strength after birth (as modeling and disuse-mode remodeling can do), (ii) ways to monitor the relationship between an LBB's strength and the size and kinds of the TPVMLs on it (as strain-dependent signals can do), (iii) special criteria for acceptable and unacceptable whole-bone strength (bone's three thresholds could provide them), (iv) cells and mechanisms that could detect and respond to those signals and thresholds (mechanotransduction could do that), and (v) feedback between those features (which the mechanostat could provide as in Relations 1 and 2).

Subpart 2.

Accordingly, bone's mechanostat could indeed make TPVMLs determine most of the postnatal strength (x) of our LBBs. Why not determine all of that strength? The baseline conditions might provide a minor part (y) of it (Frost, 2001a), which might increase after birth in response to nonmechanical agents, including genes, while postnatal adaptations to TPVMLs (z) were being added to it. Or, “x = y + z”.

This raises Question #5: why do bones never disappear in totally and permanently paralyzed limbs?

In answer, the mechanostat hypothesis predicts that the baseline conditions, or the “y” part, remain while the mechanostat tries to negate the added postnatal, or “z,” adaptations in whole-bone strength. A natural experiment for studying such phenomena could involve the permanent and total lower-limb paralyses caused by myelomeningoceles.

Subpart 3.

Most authorities on biomechanics now agree that bones adapt their strength to their TPVMLs (Kannus et al., 1996; Martin et al., 1998; Lanyon and Skerry, 2001; Currey, 2003); thus the remaining issues concern how they do it and what function(s) that process serves. Subpart 1 above summarized how LBBs could adapt their strength to their TPVMLs, in terms of “functional adaptations” (Lanyon and Skerry, 2001) or “structural adaptations to mechanical usage” (SATMU) (Frost, 1990ac). One function of these adaptations would create MBC, and an LBB that did not develop nontraumatic fractures from its TPVMLs could have achieved that MBC regardless of its absorptiometric Z score, as suggested below in Section 4 of Part III.

Early analysts of these matters did not know of the features described in Parts I and II of this article (Wolff, 1892; Roux, 1983; Koch, 1917; Gelser and Trueta, 1958), and some of these features were not known or recognized when bone's mechanostat was first described (Frost, 1987b). While a few early researchers realized that muscles put the largest loads even on weight-bearing bones (trauma excepted) (Inman, 1947; Frankel and Burstein, 1970; Pauwels, 1973), most skeletal scientists and clinicians did not know or accept that before ∼1980, and some of them may still doubt it and/or its role in affecting postnatal bone strength.


1) Predictions by the Mechanostat Hypothesis of Things That Should Occur

A physiological hypothesis becomes more useful when it can predict as well as explain things. PREdictions differ from POSTdictions. Predictions from first principles tell what should occur, whether or not one already knows that it does occur. Postdictions usually try to explain known phenomena with generally-accepted knowledge and ideas, and in the past most postdictions in bone matters ignored bone's tissue-level biologic machinery. Table 4 lists 32 predictions by the updated mechanostat hypothesis of phenomena that should and do occur. Many of them have clinical significance, and they support (but do not validate) the mechanostat hypothesis. Group 1 below explains some of those predictions using the same numbers as in Table 4, while Groups 2 and 3 mention others that remain to be tested or verified. At present I do not know of any erroneous predictions made by this updated hypothesis.

Table 4. Thirty-two correct predictions of the mechanostat hypothesis
1) LBBs should have a strength-safety factor (see Section #2 in Part III).
2) During growth whole-bone strength could lag behind growing mechanical needs in some children (see Section #1 in Part III).
3) Diaphyseal forearm fractures from falls should increase during our adolescent growth spurt (see Section #1 in Part III).
4) Diaphyseal forearm fractures from falls should decrease in young adults.
5) Peak bone strains from TPVMLs during growth should exceed those in adults (Section #1 in Part III).
6) Large TPVMLs should have more effects on bone strength than equal numbers of small loads (Section #4, E in Part I).
7) The bone strength and “mass” effects of some bone-active agents could plateau (Section #9, (iii) in Part III).
8) A functionally-based classification of osteopenias and osteoporoses should exist (Section #4 in Part III).
9) Epiphyseal and metaphyseal spongiosas could respond differently to mechanical and some nonmechanical influences (Frost and Jee, 1994a and b).
10) Histomorphometric norms for the Jamshidi and Bordier-Meunier iliac biopsies should differ.
11) Small strains should potentiate bone healing (Jensen, 1998; Meyer et al., 1999).
12) Rare strenuous voluntary activities could cause larger strains than the MESm.
13) Peak bone strains caused by TPVMLs should remain below the MESp.
14) Bone “mass” factors alone could not evaluate whole-bone strength (Section #3 in Part III).
15) Speed-of-sound values could not evaluate whole-bone strength (Section #4 in Part III).
16) Bones should not disappear after permanent-total paralyses (see Subpart #2, Section #4 in Part II).
17) Most gross-anatomical bone features of osteogenesis imperfecta.
18) Living bones should have very long fatigue lives (Section #3, D in Part II).
19) Necrotic bone should be more prone to fatigue fractures than living bone.
20) X-ray absorptiometry could not evaluate MBC or whole-bone strength (Section #3 in Part III).
21) Ultrasonography could not evaluate MBC or whole-bone strength (Section #4,F,G below).
22) Cross sectional amounts of primary spongiosa should exceed the amounts of secondary spongiosa below growth plates.
23) Osteoporosis fractures from falls should affect mainly metaphyseal bone regions (Section #4 in Part III).
24) Diaphyseal marrow cavities should contain little or no spongiosa (Section #3, H in Part II).
25) The mechanostat should need permissive agents in order to function properly (Section #6 in Part III)..
26) Something in marrow can modulate mechanical and nonmechanical effects on modeling and remodeling of bone touching or close to marrow (Section #9, (i) in Part III).
27) X-ray absorptiometry cannot distinguish physiologic osteopenias from true osteoporoses (Section #4, F in Part III).
28) Current ultrasound methods cannot distinguish physiologic osteopenias from true osteoporoses (Section #4, F in Part III).
29) Clinical densitometry cannot diagnose physiologic osteopenias or true osteoporoses (Section #4,F in Part III).
30) In healthy, free-living, young-adult mammals peak bone strains from TPVMLs would seldom exceed the MESm range and would rarely approach the MESp range (Section #3,I in Part II).
31) Ignoring the mechanostat when analyzing the relationships between cell-level data and their organ-level effects usually commits jumping frog errors (Section #8 in Part III).
32) Some drugs could decrease traumatic but not spontaneous fractures, or conversely (Section #4,E in Part III).

Group 1.

(2) The sluggish and “error-driven” nature (Wiener, 1964) of mechanically-controlled bone modeling should produce an “adaptational lag” in some children such that increases in the strength of their LBBs lag behind the needs of the increasing size of the TPVMLs on those LBBs (Frost and Jee, 1994a,b). (3) This lag should increase during our adolescent growth spurt when body weight, muscle strength, and bone length increase faster than before (Frost, 1997). If so, corresponding increases in human forearm diaphyseal fractures from falls should accompany the above changes. (4) Yet in young adults, when body weight and muscle strength usually stop increasing, the lagging strength-safety factor should “catch up” and peak in value. If so, diaphyseal forearm fractures from falls should decrease in young human adults. Regardless of whether this explanation is correct, age-related human fracture patterns correlate very well with these predictions (Blount, 1955; Wiley and McIntyre, 1980; Rockwood and Green, 1997a,b; van Staa et al., 2001), and some studies support them (Sumner and Andriacchi, 1996; Rauch et al., 2001). While metabolic explanations for increased fractures during our adolescent growth spurt were proposed in “premechanostat” times (Uhthoff, 1986), the above biomechanical explanation seems plausible. (5) Furthermore, peak bone strains from TPVMLs during growth should and do exceed those in adults (Nunamaker et al., 1990). (9) Epiphyseal and metaphyseal spongiosas should respond differently to mechanical loading (Frost and Jee, 1994a,b), which seems to be the case (Flieger et al., 1998; Rubin et al., 2001, 2002b; Torvinen et al., 2003). This may help to explain apparent discrepancies in vibration effects on bones reported by various groups (see Section 4E in Part I).

(10) Histomorphometric norms for vertical iliac bone biopsies by the Jamshidi method should differ from norms for transiliac biopsies by the Bordier-Meunier method. The former biopsies contain both apophyseal and metaphyseal iliac spongiosas, while the latter biopsies only contain metaphyseal iliac spongiosa (Minaire, 1973; Recker, 1983). Previous studies have explained why that difference occurs (Frost and Jee, 1994a,b). (11) Small strains should and do potentiate fracture healing (Aspenberg et al., 1996; Frost, 1998d; Jensen, 1998; Meyer et al., 1999). (12) Rare activities, such as jumping from a high place or running very strenuously up a hill, could and do cause larger bone strains than the MESm (Hillam, 1996; Burr et al., 1999). (14) Absorptiometric evaluations of bone “mass” factors alone cannot reliably evaluate whole-bone strength (see Section 3 below), (15) nor can current ultrasound methods (Gonnelli and Cepollaro, 2002). (17) The mechanostat's set point and modeling-rate-limitation features can predict the gross anatomical bone abnormalities found in osteogenesis imperfecta (Frost, 1987d, 2003c), something no other present hypothesis can do. (22) As explained elsewhere (Frost and Jee, 1994a,b), below active growth plates the cross-sectional area of primary spongiosa normally exceeds that of the adjacent secondary spongiosa (Kimmel and Jee, 1980). (23) Osteoporosis fractures from falls mainly affect the metaphyses of long bones (see Section 4C below) (Marcus et al., 1996; Rockwood and Green, 1997a). (25) The mechanostat needs many permissive agents in order to function properly (see Section 6). (29) Clinical densitometry cannot diagnose physiologic osteopenias and true osteoporoses as Section 4 defines them. Only clinical histories and/or diagnostic X-rays can determine whether nontraumatic fractures occurred, and such fractures would identify true osteoporoses. (32) Some drugs that decrease traumatic fractures would not decrease nontraumatic fractures, and vice versa (see Section 4I).

Group 2 (two untested predictions).

(i) A modestly elevated modeling threshold (MESm) could decrease an LBB's strength-safety factor (see Section 2 below), which should weaken affected bones (and often reduce bone “mass”) and make them more prone to traumatic and stress fractures (Martin et al., 1998). (ii) A modestly decreased modeling threshold (MESm) could increase a bone's strength-safety factor, which should strengthen affected bones (and often increase bone “mass”) and make them more resistant to traumatic and stress fractures.

Whether that explanation is correct or not, a few individuals do seem unusually prone to stress and traumatic fractures (Schönau et al., 1996; Burr and Milgrom, 2000), while others seem unusually resistant to them. Johnson et al., (2002) described some people who might belong in the (ii) category above. While the mechanostat hypothesis predicts these two situations, Table 4 omits them because we do not know if MESm changes cause them, although “biologic variation” suggests that such changes could occur.

Group 3 (more predictions that still need testing).

(i) In both Group 2 situations above, affected bones need not show an abnormal histology, composition, metabolism, tissue dynamics, or material properties. The causative abnormality should lie chiefly in the MESm threshold, and as Section 9 (ii) below indicates, the genes and cell(s) responsible for that threshold remain unknown. (ii) The mechanostat hypothesis predicts that learning how to change the MESm's value may have therapeutic uses. As examples, lowering that threshold would enable growing children to accumulate more bone, so they would enter adult life with larger and stronger bone “banks” (Schönau et al., 1996); it should minimize nontraumatic fractures in true osteoporoses (Burr and Milgrom, 2000; Martin et al., 1998), as well as decrease traumatic fractures; it should decrease stress fractures in athletes, special forces trainees, horses, and greyhounds (Burr and Milgrom, 2000); and it might help to prolong the service lives of some load-bearing joint replacement and dental endoprostheses (Frost, 1992b).

Hence Question #6: How could those things work?

In answer, decreasing the MESm should enable smaller strains than before to make modeling strengthen bones relative to the same TPVMLs (Martin et al., 1998). That would increase the “bone-strength/bone-load” ratios of affected bones.

(iii) The mechanostat hypothesis suggests that the amounts of trabecular or cortical bone in cross sections of LBBs from skeletal remains, including fossilized bones, could indicate the TPVMLs on those bones in life (Frost, 1999), which is a matter of interest to forensic pathologists and paleontologists. (iv) At present, the effects of constant and repeated dynamic loads and strains on LBBs remain unclear. This article concerns repeated dynamic loading and strain effects. (v) The mechanostat hypothesis suggests that in physiologic importance whole-bone strength should rank above the features that contribute to it. Such features include bone “mass,” bone mineral content, absorptiometric bone mineral “density” (Nielsen, 2000), outside bone diameter, cortical thickness, trabecular connectivity and thickness, the shape and architecture of bone, etc. If the hypothesis is correct, whole-bone strength could become an important datum in much future work, especially in matters that concern “osteoporosis,” stress fractures, the design of load-bearing endoprostheses (Frost, 1992b), and bone healing. Section 3 below concerns ways to measure whole-bone strength and to evaluate it noninvasively. In this article “osteoporosis” and “osteopenia” in quotes refer to current conventional definitions (Kanis, 1994; Riggs and Melton, 1995), which differ from the italicized definitions in Section 4 below.

2) Bone's Strength/Safety Factor (SSF)

Healthy LBBs have an SSF (Alexander, 1984; Cowin, 1990; Currey, 2003; Frost, 2003b) that defines how much more strength they have than the minimum needed to keep TPVMLs from breaking them suddenly or in fatigue. That SSF could define an LBB's “bone-strength/bone-load” ratio as well.

(A) This raises Question #7: What creates the SSF?

In answer, the MESm would determine a healthy LBB's largest normally-allowed strain or stress caused by its TPVMLs in free-living bony vertebrates (Frost, 2001a). Then as long as MESm < Fx, an SSF must exist (although the existence of a SSF would not prove that an MESm exists too), and it could equal a bone's ultimate strength divided by its modeling threshold, or SSF = FX ÷ MESm. When the latter two terms are expressed as stresses, healthy young-adult mammalian LBBs should have ∼6× more strength than the minimum needed to keep TPVMLs from breaking them (from Table 3, 120 mpa ÷ 20 mpa = 6). This “bone-strength/bone load” ratio assumes physiologic-mode TPVMLs, and it should help to minimize traumatic fractures from low-energy trauma also. Hence #1 in Table 4.

(B) A value of ∼2 suggested for bone's SSF (Biewener, 1993) used bone's yield point of ∼60 mpa (Table 3) instead of its MESm to calculate the SSF. Bone's yield point is a range on bone's stress-strain graph below which the relationship between a bone's strain and its stress remains linearly proportional or “Hookean” without a permanent set or plastic flow, and above which strain increases faster than stress (Yamada, 1970; Reilly and Burstein, 1974; Martin et al., 1998; Ferretti et al., 2000). Whether by coincidence or not, in Table 3 that yield point range for lamellar bone equals bone's MESp range, and also the lower bound of the pathologic overload window (POW in Fig. 1) where bone strains begin to cause woven rather than lamellar bone formation (Rutishauser and Majno, 1950; Frost, 2002b; Li et al., 2002). As noted in Part II, Section 3I, in healthy LBBs typical peak bone strains from TPVMLs nearly always stay below the MESp and yield point ranges listed in Table 3.

Aging effects on bone's SSF remain uncertain at present, except for the “adaptational lag” mentioned above in Section 1, Group 1 of Part III.

(C) The mechanostat hypothesis predicts that anything that would decrease an LBB's SSF towards or below unity would potentiate nontraumatic fractures. The causes of this phenomenon could include inadequate modeling, excessive disuse-mode remodeling, impaired detection and/or repair of MDx, degraded material properties of bone that potentiated MDX production, and combinations thereof (see Section 4D below).

3) Whole-Bone Strength and Its Evaluation

If whole-bone strength becomes an important factor in many future studies, one should know how to evaluate it noninvasively and how not to evaluate it. This requires one to know its determinants. Parts I and II of this article describe its chief tissue-level biologic determinants.

Whole-bone strength has four chief physical determinants (Yamada, 1970; Currey, 1984; Ferretti, 1999; Ferretti et al., 2002). Changes in bone's collagen, microstructure, and mineral crystals with aging and in disease have much smaller effects on whole-bone strength than these four determinants, which include (A) the properties of bone as a tissue or structural material, including its yield point, stiffness, ultimate strength, and fatigue life (the material properties factor). These material properties vary much less than the other three physical determinants in different bony vertebrates, in different bones, during aging, in different sexes, and in varied diseases, excepting osteomalacia. The other physical determinants include (B) the amount of MDx in a bone (the MDx factor); (C) the amount and kind of bone tissue in a bone (woven, plexiform and lamellar bone, compact and trabecular bone, the bone mass factor); and (D) the cross and longitudinal shape and size of a bone, and how its bone tissue distributes in space (the architectural factor, which Ferretti (1999) calls the geometric factor). For example, doubling the diameter of a hollow long bone while keeping the same amount of bone in its cross section so that the “C” factor did not change would leave its strength in uniaxial loading nearly unchanged, but would increase its bending strength ∼8X (a beam with a rectangular cross section increases its bending strength as the fourth power of the height of that section, but a cylinder's bending strength increases as the cube of its diameter). Engineering figures of merit called “cross-sectional moments of inertia” can help to predict and calculate the bending and torsional strengths of intact hollow bones (Ferretti, 1999).

Currently there are no noninvasive methods that can evaluate the MDx and material-properties factors reliably in patients or experimental animals. Dual-energy X-ray absorptiometry (DEXA) can evaluate the “mass” or (C) factor noninvasively as bone mineral content values. The currently popular bone mineral “density” (BMD) values, which provide another “mass” or (C) factor, provide very unreliable indicators of whole-bone strength and bone “mass,” as do speed-of-sound values. The notion that BMD values can provide useful surrogates for whole-bone strength (Turner, 2002b) is false (Ferretti et al., 1995; Augat et al., 1996; Banu et al., 1999; Nielsen, 2000; Beck et al., 2001). For example, healthy rat and horse femurs would have similar BMD values (and speed-of-sound values), yet their strengths differ by more than 1000 times. Hence #20 and #21 in Table 4.

However, with suitable software, quantitative computed tomography (CT) or peripheral quantitative CT (pQCT) can provide bone strength indicators (BSIs) that evaluate whole-bone strength quite well. Their calculation accounts for both the “mass” and the architectural factors in whole-bone strength (Schiessl and Willnecker, 1999; Ferretti, 1999). Partly for such reasons, pQCT-derived BSIs (by that or any other name (others called them strength-strain indices or SSIs) (Schiessl and Willnecker, 1999; Ferretti, 1999)) are being used increasingly in noninvasive evaluations of whole-bone strength. Different bone “mass” and architectural factors, rather than different material or MDx properties, cause the huge strength differences between mouse and elephant femurs. Hence #20 and #21 in Table 4.

This raises Question #8: What characterizes a reliable BSI?

In answer, when multiplied by the same constant (k) a reliable BSI (or SSI) would correctly predict the hugely different fracture strengths of rat ribs and giraffe tibias; or, k(BSI) = Fx (Frost, 2003b). As noted elsewhere (Frost, 2001a), bone metaphyses and diaphyses might need different BSIs.

Other investigators have described ways to evaluate whole-bone strength ex vivo (Evans, 1957; Yamada, 1970; Reilly and Burstein, 1974; Hasegawa et al., 2001; Banu et al., 1999). Some noninvasive methods for use in patients and experimental animals have limitations (Cann et al., 1995; Ferretti et al., 1995, 1998; Schönau et al., 1996, 2000; Schiessl et al., 1998; Ferretti, 1999, 2000; Jiang et al., 1999; Wilhelm et al., 1999; Neu et al., 2001; Nielsen, 2000; Schneider et al., 2001). As an example, many biomechanical and risk-of-fracture arguments and conclusions depended on BMD values as indicators of whole-bone strength and of bone “mass” (Marcus et al., 1996; Kanis et al., 1997; Meunier et al., 1999). But, to repeat, BMD data provide very unreliable indicators of both features (Kröger et al., 1992; Augat et al., 1996; Banu et al., 1999; Nielsen, 2000; Beck et al., 2001). This would weaken arguments and conclusions that depend on BMD data.

While many researchers trust risk-of-fracture analyses that depended on BMD data (Kanis, 1994; Faulkner, 2000), many of them may have assumed that strong associations indicate cause–effect relationships. Yet the extremely strong association between skin cancer and breathing (r > 0.9999, P < 0.00001) does not mean that breathing causes skin cancer. When risk-of-fracture analyses depend on BSIs instead of BMD data, and when they also account for the risk of falls (see Section #4B next), their reliability ought to improve (another prediction of the mechanostat hypothesis that awaits testing).

All of the above material indicates that neither x-ray absorptiometry nor ultrasonography, alone, can evaluate MBC as Part I of this article defines it. Hence #20 and #21 in Table 4 (see Section #4E–G below).

4) The Mechanostat and “Osteoporosis”

(A) After menopause, most women lose ∼15% of their former bone “mass” (Marcus et al., 1996). This is often referred to as postmenopausal “osteoporosis” (Barzel, 1970). The same kinds of phenomena happen more slowly and less markedly in most aging men, resulting in what many call male “osteoporosis” (Orwoll, 2002). In both situations the loss comes only from bone that is touching or is close to marrow (from trabecular and endocortical bone) (Trotter et al., 1960; Takahashi and Frost, 1966; Garn, 1970; Trotter and Hixon, 1974; Keshawarz and Recker, 1984; Smith and Gilligan, 1989; Frost, 1998b), as noted in Section 9 (i) below.

(B) Nevertheless, the majority (>70%) of such people never develop nontraumatic fractures (Riggs and Melton, 1995; Marcus et al., 1996), so by the earlier definition of MBC they would have mechanically-competent LBBs, even with negative Z scores (Jonsson et al., 1993; Kanis, 1994; Aspray et al., 1996). Falls cause most traumatic fractures (of hips, wrists, humeral surgical necks, ankles, etc, (Jaglal et al., 1993; Kannus et al., 1999) in this population, so decreasing falls should also decrease fractures, even when sizeable bone “mass” deficits coexist. Various impairments of balance, muscle strength (and power?), neuromuscular coordination, and vision lead to an increased number of falls by aging humans (Overstall et al., 1977; Tinetti et al., 1988, 1994; Lappe et al., 1989; Greenspan et al., 1994; Guralnik et al., 1995; Marcus et al., 1996; Runge, 2002). This should make falls the largest “risk factor” for traumatic osteoporosis fractures, because without falls fractures caused by them do not occur whether or not some MDx increases the bone fragility associated with all osteopenias. Yet a recent review of risk factors for osteoporosis fractures in over 200,000 American women over 50 years of age did not even mention falls or their causes (Siris et al., 2001), and no current absorptiometric method can evaluate such falls. After I observed in China in 1999 that a drug that decreased falls by 30% should similarly decrease traumatic osteoporosis fractures, investigators began to note how drugs intended to treat or prevent “osteoporosis” affect the frequency of falls. Hence another “devil” in bone physiology's details.

(C) Falls and other trauma can put magnitudes, kinds, and combinations of one-time loads (Pinilla et al., 1995) on parts of LBBs to which they could not have adapted. Most so-called osteoporosis fractures of extremity bones affect their metaphyses rather than their diaphyses (Ferretti et al., 1995; Riggs and Melton, 1995; Rockwood and Green, 1997a; Frost, 2001a).

Hence Question #9: Why do extremity-bone osteoporosis fractures preferentially affect long-bone metaphyses?

The human radius can illustrate a biomechanical answer to that question. The distal radial metaphysis would have adapted mainly to uniaxial compression TPVMLs from muscles, because the low friction of the adjacent wrist joint would protect the metaphysis from large shearing, bending, and torsional loads from TPVMLs. Yet the radius' diaphysis would normally adapt to repeated combined uniaxial, bending, and torsional TPVMLs from arm and forearm muscles (Frost, 2001a). When a fall puts a one-time large, combined uniaxial, bending, shearing, and torsional load on the whole radius, the part least adapted to the load's bending, torsional, and shearing components (the metaphysis) usually fractures first. Indeed, most so-called Colle's fractures represent metaphyseal fractures caused by bending and/or shearing loads on the metaphysis (Rockwood and Green, 1997a,b). While premechanostat explanations have been suggested for this phenomenon (Uhthoff, 1986), the above explanation seems plausible. Hence #23 in Table 4.

(D) Such findings suggested that osteopenias without nontraumatic fractures should be called physiologic osteopenias regardless of their T or Z scores (Frost, 1998c). Presumably, combinations of age-related muscle weakness and loss of estrogen in women and androgens in men cause most such osteopenias. They would be “physiologic” because healthy mechanostats should cause osteopenias in which LBBs still satisfied the earlier definition of MBC. In this article, osteopenia without quotation marks refers to a condition of less bone and less whole-bone strength than is found in most people of the same age, sex, race, species, and occupation, regardless of the severity of the loss of bone strength and bone “mass.”

One could name osteopenias in which nontraumatic fractures do occur true osteoporoses, again regardless of their T or Z scores (Frost, 1998c). Presumably, still-enigmatic mechanostat diseases cause them, because nontraumatic fractures should herald diseased mechanostats and, usually, excessive bone MDx. Furthermore, spontaneous fractures should show that the strength-safety factor in affected bones had decreased toward or below unity (see Section #2C in Part III above). Hence #8 in Table 4.

(E) Because traumatic and nontraumatic osteoporosis fractures have different causes, the mechanostat hypothesis predicts that some “antiosteoporosis” drugs could affect these two kinds of fractures differently. For example, they might decrease hip and wrist fractures but not spine “fractures,” or vice versa. Such differences have been observed (Seeman, 2001). Hence #32 in Table 4.

(F) Physiologic osteopenias and true osteoporoses, as defined above, could have equally small or equally large whole-bone strength and bone “mass” deficits, and equally negative T and Z scores. Ergo, neither x-ray absorptiometry nor ultrasonography can diagnose them or distinguish them from each other. Hence #27–#29 in Table 4.

(G) Some authorities may question certain ideas in this article because these ideas challenge accepted “wisdom.” I respect such people and their views (Kanis et al., 1997; Meunier et al., 1999; Raisz and Seeman, 2001; Siris et al., 2001; Turner, 2002; Rosen, 2003), and time and more work should resolve our differences (see Section 9 (iv) D below).

(H) Different definitions of disease may contribute to this problem. (i) In one view, any abnormality stems from disease; therefore all osteopenias, all negative T and Z scores, and all wrist and hip fractures would represent a bone disease. (ii) However, in the functionally-based mechanostat view, LBBs that provide MBC would be healthy regardless of any negative Z scores (Frost, 2003a), and injuries rather than bone disease usually cause wrist and hip fractures. Neurophysiologic, muscular, and visual disorders, rather than bone disease, lead to most of the falls that cause those fractures. An accompanying osteopenia only makes a fall more likely to cause a fracture, but osteopenias per se do not cause fractures.

The literature shows growing recognition of the merit of many of these statements (which are less controversial now than they were 15–25 years ago), and efforts to evaluate the causes of falls and to minimize them and/or their bone effects in aging humans have increased considerably since 1980.

(I) Three matters should be commented on at this point. First, ovariectomized or orchidectomized mice and rats develop osteopenias but not nontraumatic fractures, so they should provide animal models of physiologic osteopenias rather than true osteoporoses, according to their italicized definitions above (Jee, 1995; Yao et al., 2000; Frost, 2001c). Second, much of what people think is known about the nontraumatic “fractures” of thoracic and lumbar vertebral bodies in true osteoporoses (as defined above) came from data obtained from human transiliac bone biopsies (Eriksen, 1986; Mellish et al., 1989; Vesterby et al., 1989; Balena et al., 1992, 1994). Yet the ilium rarely develops nontraumatic fractures in that disease so it could not reveal their causes. Third, the pathogeneses of traumatic and nontraumatic fractures unquestionably differ; as the poet Robert Burns might have put it, they are different “beasties.” Excessive amounts of MDx and declines in the strength-safety factor of affected bones may help to cause the latter fractures. Lumping both kinds of fractures together as “osteoporosis fractures,” as was often done in the past, would in a sense be equating oranges with roses.

5) On Genetic Roles in Whole-Bone Strength

The mechanostat hypothesis for bones is an important part of the Utah paradigm of skeletal physiology (Takahashi, 1995; Frost, 2000b,c). I suggest that this hypothesis relates to bone physiology and disorders in the same way that nephrons relate to renal physiology and disorders. Bone's mechanostat is still a hypothesis, but perhaps in the same sense that E = mc2 technically still represents a hypothesis. More than 85 years ago, “connecting the dots” between varied physics data provided by others, and thinking “outside the box” of then-conventional wisdom, allowed a Swiss postal clerk to find E = mc2 hiding in those data (Calaprice, 2000).

That said, (A) in some premechanostat views that may still linger, postdiction reasoning that ignores the mechanostat suggests that genes predetermine optimal postnatal bone features such as bone “mass,” bone mineral “density” or content, trabecular thickness and/or connectivity, outside bone diameter, trabecular number, cortical thickness, bone's histologic structure, a bone's strength and fragility, etc. (Garn, 1970; Kleerekoper et al., 1985; Arden et al., 1996; Seeman et al., 1996; Jaffurs and Evans, 1998; Deng et al., 2000; Nguyen et al., 2000; Ridley, 2000; Campbell et al., 2001).

In this article “mass” in quotation marks has an absorptiometric meaning and is not a true physical mass (Marcus et al., 1996); similarly “density” in quotes has an absorptiometric meaning and is not a true physical density (Ferretti, 1999; Jiang et al., 1999; Nielsen, 2000).

(B) However, the mechanostat hypothesis suggests that genetics predetermine a bone's baseline conditions (y) in utero, including its mechanostat. To that “y” part the mechanostat would then add any adaptations (z) needed to fit a postnatal LBB's strength (x) to the size and kinds of the TPVMLs on it after birth. Thus again, x = y + z (Frost, 2003a). Some authorities question this idea as well, and I respect them and their views (see Section 4G above and Section 9 (iv) D below). In a personal communication in 1999, Prof. J.L. Ferretti noted, plausibly in my view, that the phenomena described in Parts I and II above made sense only if “(B)” represents the better view. If genes predetermine whole-bone strength, why would it change in response to changes in postnatal TPVMLs in paralyses and mechanical disuse? Yet such changes do occur (Kiratli, 1996; Sievanen et al., 1996).

(C) So far, few investigations into the genetic influences on “osteoporosis”, bone “mass,” bone strength, and/or BMD values, have considered how genetic influences on muscle could affect such features. Yet in identical twins, genetic influences on muscles and the mechanostat could cause very similar muscle strength, bone strength, and bone “mass” indicators (Arden et al., 1996; Seeman et al., 1996). Hence this is something new for twin-study advocates to look at, and another “devil” in bone physiology's details.

6) On Permissive Agents

In some premechanostat views that may still linger, postdiction reasoning that ignores the mechanostat suggests that nonmechanical influences on osteoblasts, osteoclasts, and/or their precursor cells by factors such as genes, hormones, calcium, vitamins C and D, some drugs, and some cytokines dominate the control of postnatal bone health (Frost, 1961, 1962; McLean and Urist, 1961; Reifenstein, 1962; Nordin et al., 1966; Barzel, 1970; Bronner, 1994; Deng et al., 2000; Whedon, 1984; Huffer, 1988; Gallagher, 1990; Nordin and Heaney, 1990; Amling et al., 2000; Pietschmann et al., 2002; Rosen, 2003). By, implication such factors would also dominate control of an LBB's postnatal MBC.

Yet most such nonmechanical agents could act as “permissive” ones that the mechanostat needs in order to make LBBs mechanically competent relative to the TPVMLs on them. By analogy, cars need hundreds of things (fuel, motor, wheels, gears, steering, batteries, bolts, etc.) in order to be driven. For the driving function, all such things would represent permissive ones that do not drive the car or choose its destinations. Reaching a destination by car is similar to achieving MBC by a bone.

This raises Question #10: How can one recognize permissive bone-active agents?

In answer, deficiencies of these agents can cause big health problems but in healthy people an excess of such agents has little or no effect, or may have different kinds of effects, including toxicity (Lloyd et al., 1993; Frost and Schönau, 2000). Some examples follow.

(i) Vitamin C deficiency causes scurvy, but its excess has little effect on a healthy body. (ii,iii) Vitamin D and thyroxine deficiencies cause short stature (Vaughn et al., 1975; Wilson and Foster, 1992; Avery and First, 1993), but their excesses does not cause giantism, and can cause toxicity. (iv) According to long-accepted wisdom, growth hormone (GH) compels whole-bone strength to increase regardless of the loads on bones (Bouillon, 1991; Auerbach et al., 1992; Wilson and Foster, 1992; Inzucchi and Robbins, 1994; Andreasson et al., 1995; Ohlsson et al., 1998; Prakasam et al., 1999). However, a clever Australian study found that without increased bone loads, GH does not significantly increase whole-bone strength (Forwood et al., 2001). An earlier study reached the same conclusion (Halloran et al., 1995); therefore, GH could indeed play a permissive role in this matter, as suggested several years ago (Frost, 1998a).

This permissive role may help to explain some disappointing results of treating “osteoporosis” with GH (Wilton, 1999; Rosen and Wuster, 2003). When people with “osteoporosis” do not exercise against resistance to increase their muscle strength (and their typical peak bone strains), GH's permissive role might manifest itself as a failure to increase LBB strength enough to decrease traumatic osteoporosis fractures. Effects on LBB strength by agents such as testosterone, calcium, and vitamin D, and perhaps by many cytokines, chemokines (Li et al., 2002), cell receptors, ligands, genes, etc., might also depend on such permissive roles. Indeed, the increased muscle strength usually associated with GH or testosterone treatment may depend on permitting rather than compelling muscle strength to increase in response to exercise against resistance.

Forwood et al. (2001) found that without TPVMLs on a bone, GH did not increase its strength. Similar experiments could reveal further permissive effects on bones and increase our understanding of the mechanisms of drug actions on bone and its mechanostat. The idea of permissive roles in skeletal physiology is old, but after 1890 few skeletal physiologists even mentioned them (although mechanostat physiology may depend strongly on them). Another model for such studies might compare an agent's effects on the strength of the femur (an LBB) and the turbinates (which are not LBBs).

This is another “devil” in bone physiology's details. As a car needs hundreds of permissive things in order to be driven, the mechanostat needs many permissive agents to create MBC for the LBBs. Hence #25 in Table 4.

7) On Proof of the Modeling-Remodeling Distinction, and Collaboration Between In Vitro and Live-Animal Research

(A) Most readers would agree that all the above features need more study. However, bone's tissue-level mechanisms, which include the mechanostat, do not function normally in present cell-, tissue-, and organ-culture systems (Frost, 1986; Parfitt, 1995; Jee, 2001). Therefore, their study requires more live-animal research.

(B) Since the early 1960s, W.S.S. Jee's laboratory has led the way in showing how to do live-animal bone research (Jee, 1995, 1999; Jee and Yao, 2001). They tested the 1964 idea that bone modeling and remodeling are independent mechanisms that can respond independently to various stimuli (Frost, 1964a,b). They found that with partial disuse, cortical modeling decreased or turned off, while BMU-based remodeling of spongiosa increased (Jee and Li, 1990; Li et al., 1990; Jee et al., 1991; Li and Jee, 1991; Yao et al., 2000). An endocrinologic challenge resulted in the same findings (Chen et al., 1995; Yeh et al., 1995). Connecting the dots shows that others found the same phenomena in studies too numerous to cite here, although their authors seldom remarked upon the fact. Studies by some of Professor Jee's former pupils support his original conclusions (Wronski and Morey, 1982, 1983a,b; Wronski et al., 1986, 1989, 1993a,b; Kimmel et al., 1990; Jerome, 1994; Wronski and Yen, 1994; Yeh et al., 1994; Ke et al., 1997).

Such studies proved that in the same bone at the same time, modeling and remodeling can respond in opposite ways to the same stimulus, although both mechanisms appear to utilize the same kinds of osteoblasts and osteoclasts (Jee, 2001). This verifies the 1964 idea, even though some people may still not accept it, or may not know about it yet.

(C) Such findings suggest that combining cell-biologic and molecular-biologic expertise with live-animal research and insights into the mechanostat might considerably enhance the productivity of some bone research. A recent article exemplified the use of both such a combination (Kousteni et al., 2002) and “designer drug” development (Economides et al., 1995; Frost, 1998c).

8) Macrocosms, Microcosms, and “Jumping Frog” Errors

For over 60 years, researchers who predicted an agent's in vivo effects on bones from its in vitro effects on osteoblasts, osteoclasts, and/or on their precursor cells erred, as explained below.

In physics and astronomy, “microcosms cannot predict macrocosms,” although the former can usually help to explain the latter after other things have revealed the latter's features, which would represent a postdiction (Schermer, 2002). Thus, predicting galaxies and cars solely from a knowledge about atoms has a vanishingly small chance of success, yet in postdictions atoms can help to explain already-known features of galaxies and cars. Investigators who tried to predict in vivo bone effects from in vitro data often attempted, usually erroneously, to predict a skeletal macrocosm from a microcosm while ignoring bone's tissue-level mechanostat. The following examples include one of my own errors (mea culpa—again).

(i) Recognition in the early 1960s that calcitonin hinders osteoclastic, but not osteoblastic, activities in vitro suggested that calcitonin could increase bone “mass” and cure “osteoporosis” when given in vivo. Yet it did not. This attempt to predict skeletal macrocosms from a skeletal microcosm caused a “jumping frog” error (Frost, 2000c). (ii,iii) In 1935–1955, some people thought that estrogen or supplemental dietary calcium would also increase bone “mass” and cure “osteoporosis” (since their deficiencies cause bone loss). Yet they did not. Hindsight shows that permissive functions of these agents were mistaken for compelling determinants of whole-bone strength and bone “mass” (just because a lack of something causes bone loss does not necessarily mean that its excess will increase bone “mass”). (iv) The authors of a study of mechanical loading on mammalian long-bone growth plates decided that even small loads depressed their growth (Ohashi et al., 2002). If that was the case, bones in paralyzed growing limbs would become longer than corresponding bones in normal limbs. However, connecting the dots reveals that bones in paralyzed human limbs grow shorter than in normal limbs, while bones in deloaded limbs in growing experimental animals never grew longer than corresponding bones in control limbs. (v) One study concluded that cells in deloaded bones become “resistant” to GH's presumed ability to compel whole-bone strength to increase, regardless of a bone's TPVMLs (Halloran et al., 1995). Yet that study really indicated GH's permissive role in that situation, as in Section 6 above. (vi) In the 1970s, I thought that manipulating the activation-resorption-formation sequence in the remodeling BMU could restore bone to, and thereby cure, “osteoporotic” skeletons (Frost, 1979). However, that “coherence treatment” did not work as hoped. Hence, a jumping frog error resulted from my attempt to predict a skeletal macrocosm from a microcosm in a way that, in retrospect, ignored bone's tissue-level mechanostat. Interestingly, in Anderson's (Anderson et al., 1984) test of that idea in humans, it did increase trabecular bone “mass,” but the increase disappeared after the treatment stopped. Perhaps (in retrospect) this was because the mechanostat's thresholds were not affected by the treatment (see Section 9 (ii) below). Hence #31 in Table 4.

Such errors seldom stemmed from inaccurate data. They usually stemmed from combinations of faulty interpretations of data, from not “connecting the dots,” from trying to predict macrocosms from microcosms, from not thinking “outside the box” of long-accepted wisdom and ideas, from confusing transients with steady states, from a lack of essential information and/or ideas, and from ignoring bone's mechanostat.

In that regard, today most physiologists would find it far beyond naive to try to explain renal physiology and its disorders while ignoring nephrons. May I suggest that trying to explain bone physiology and its disorders while ignoring bone's mechanostat—a tissue-level “nephron-equivalent” mechanism in bones (Frost, 2000c)—could be equally naive? Predictions and postdictions that ignore bone's mechanostat will lead to further jumping frog errors, just as ignoring nephrons would do when analyzing renal physiology and disorders. Hence #31 in Table 4.

Given that, in some premechanostat views, osteoblasts and osteoclasts “drive” the bone “car” and skeletal microcosms can predict skeletal macrocosms (Albright and Reifenstein, 1948; Weinmann and Sicher, 1955; Snapper, 1957; Aegerter and Kirkpatrick, 1958; Frost, 1961, 1962; Rasmussen and Bordier, 1974; Jowsey and Offord, 1978; Canalis, 1993; Pietschmann et al., 2002; Rosen, 2003).This idea led to studies and reviews that ignored bone's mechanostat and assumed that independently-working osteoclasts cause the bone losses that lead to “osteoporosis,” and thus depressing the activity of these cells should cure the disease (Rohdal et al., 1960; McLean and Urist, 1961). Yet it seems clear now that disuse-mode remodeling causes those bone losses, which occur chiefly in bone close to or next to marrow, and not from the periosteal envelope (see Section 9 (i) below). However, some investigators may still not know or accept that (see Section 4G in Part III above and Section 9D below).

9) Some General Observations

(i) On a probable mediator mechanism in marrow.

LBBs have four “envelopes” (the periosteal, haversian, endocortical, and trabecular surfaces) (Sedlin, 1964; Jee, 2001). In all adult-acquired osteopenias, the bone loss occurs in bone that is touching or is close to marrow, i.e., in trabecular and endocortical bone (Frost, 1998b). In steady states (Frost, 1973; Heaney, 1984) it does not come from the periosteal envelope (which may actually expand while the osteopenia occurs (Sedlin, 1964; Garn, 1970)), and only minor amounts come from the haversian envelope. This observation applies to all adult-acquired osteopenias from any cause (Marcus et al., 1996). However, pathologic observations show that throughout life, all four bone envelopes retain functional osteoclasts and/or the precursor-cell mechanisms that can create them when and where they are needed (Jaffe, 1972; Aegerter and Kirkpatrick, 1975; Anderson and Kissane, 1977; Bogomil and Schwamm, 1984; Jubb et al., 1985; Jee, 2001). Furthermore, during puberty human females deposit extra bone next to marrow, which is lost during menopause (Schiessl et al., 1998). Bones in men show the same phenomena, but the process is shower and less marked (Marcus et al., 1996; Frost, 1998b). Estrogen and androgen effects on bone next to marrow may explain those features. An aromatase that can convert testosterone to estrogen complicates our understanding of androgen effects on bones in men (Smith et al., 1994).

Accordingly, something associated with marrow may modulate some mechanical and other influences on LBBs (Frost, 1966, 1998b; Raisz and Seeman, 2001). The periosteal envelope may even have an analogous mechanism (Horiuchi et al., 1999; Shimizu et al., 2001; Frost, 2002b). Such a mechanism could “target” some drug and disease effects to particular bone envelopes. As examples, in birds estrogen evokes woven bone formation on the endocortical and trabecular envelopes, but not on the periosteal envelope (Bogomil and Schwamm, 1984); woven bone formation occurs mainly where bone lies next to marrow in human myelofibrosis (Anderson and Kissane, 1977); mainly periosteal new-bone formation occurs in human pulmonary hypertrophic osteoarthropathy (Jaffe, 1972); a woman's usual postmenopausal bone loss occurs in bone next to marrow, not from periosteal bone; and “saber shin” deformities of the tibia in lues and Paget's disease stem mainly from periosteal new-bone accumulations (Luck, 1950; Putschar, 1960; Jaffe, 1972; Kuhlencordt and Bartelheimer, 1981). Hence #27 in Table 4.

Of course, differences in bone-cell differentiation on bone's different envelopes may help to explain some effects attributed to this marrow mediator mechanism.

Partly for the above reasons, some values for cortical bone in Table 3 may differ for endocortical and trabecular bone.

(ii) Concerning criteria for “naturally acceptable” whole-bone strength, and the cell(s), mechanism(s), and genes that should provide those criteria.

Bone's MESm and MESr threshold ranges could provide criteria for distinguishing between enough and too little or too much whole-bone strength relative to the size of the TPVMLs on an LBB. (Hence Section 4, Subpart 1 (iii) in Part II above). If nontraumatic fractures did not occur in an LBB with a negative Z score (Kanis, 1994), to evaluate its MBC one would need to compare its strength to the TPVMLs on it. Among others, Banu et al. (1999), Ferretti et al. (2002), Schönau et al. (1996), Schiessl et al. (1998), and Schiessl and Willnecker (1999) have shown how to establish norms for such comparisons, and how to make such comparisons. When such a comparison reveals a subnormal bone-strength/bone-load ratio, this may suggest an impending loss of MBC and a decreased strength-safety factor in an affected bone.

The ability to permanently prevent or normalize bone-strength deficits in osteopenias, osteoporoses, or paralyses, and reliably potentiate bone healing and prolong the service lives of some load-bearing skeletal and dental endoprostheses, may depend strongly on learning how to control bone's MESm and MESr thresholds (Frost, 1992b, 1998d, 2001a; Gasser, 1998; Jensen, 1998).

That leads to Question # 11: What makes these thresholds so important?

In answer, these thresholds are like the thermostats that switch a room's heating and cooling mechanisms on and off, where “heating” is analogous to increasing a bone's strength by modeling, and “cooling” is like decreasing a hollow bone's strength by disuse-mode remodeling. The relationship between the British thermal units (BTUs) of heat in a room and the room's temperature would be similar to the relationship between the size of the TPVMLs on an LBB and its strength. Small rooms need fewer BTUs than do large rooms to reach the same temperature; equally, bones that carry small TPVMLs need less strength than bones that carry large TPVMLs to keep their peak strains within the same MESr-MESm span (Fig. 1). In this analogy, a bigger furnace, or a more potent fuel for it, would not determine the room's temperature. Such things would only affect how quickly the thermostat could return the temperature to its original setting.

However, the things that comprise these bone thresholds or “thermostat equivalents” remain unknown, including any genes that might encode them. The mere fact that osteocytes can respond to bone strains (Aarden et al., 1994; Weinbaum et al., 1994; Parfitt, 2002; Skerry, 2002) does not prove they contain these thresholds as well, although they might. Consequently, finding and studying what contains these thresholds should pose an important problem for bone research. Michael Parfitt (2000) may have had that in mind when he made the remark quoted in Part II, Section 2.

(iii) Concerning therapeutic plateaus.

The effects on whole-bone strength (and bone “mass”) of agents that only change the activities of existing osteoblasts or osteoclasts usually plateau instead of continuing during prolonged treatment. In one postdiction explanation for such plateaus, these cells become resistant to a drug (Halloran et al., 1995) or they “escaped” its effects (Tashjian et al., 1978). Yet a healthy mechanostat can cause such plateaus (Martin et al., 1998), which represent “signatures” of negative feedback systems at work (Wiener, 1964). As an explanation, say that an agent only increased osteoblastic activity so that bone strength began to increase. This would make the same TPVMLs as before cause smaller bone strains, which could make the mechanostat increase disuse-mode remodeling to decrease bone strength, and tend to depress subsequent modeling. Eventually the combined effects could make the whole-bone strength plateau.

Bone “mass” plateaus follow treatment with many bisphosphonates (Cointry et al., 1995; Fleisch, 1995), some prostaglandins (Ke et al., 1991), calcitonin (Avioli and Gennari, 1993; Mosekilde et al., 1994), and calcium and vitamin D (Bronner, 1994). Apparently, parathyroid hormone (Takahashi et al., 1991; Capozza et al., 1995; Gasser, 1998) can bypass the mechanostat to evoke new modeling formation drifts on the periosteal, endocortical, and trabecular envelopes (Frost, 1998c; Zhang et al., 1999), but when treatment is stopped, disuse-mode remodeling begins to decrease the added bone strength.

The room-temperature/bone-strength analogy in (ii) above might help to explain mechanostat-induced plateaus. If something (e.g., an open door on a very cold or hot day) changed a room's temperature without changing the two thermostat settings, the thermostats would make the heating and cooling systems try to return the room's temperature to its previous state. Equally, if something changed a bone's strength without changing the MESm and MESr, those thresholds would make modeling and remodeling try to return the bone's strength to its previous state. This appears to happen after bone “mass” is increased with bone-anabolic agents, such as parathyroid hormone and prostaglandin E-2, and it may have happened after the coherence-treatment trial mentioned above (Anderson et al., 1984). Hence #7 in Table 4.

If the effect of a bone-active agent plateaus while treatment with that agent continues, it could mean that bone's mechanostat caused the plateau.

(iv) Other matters.

(A) Elsewhere and in brevity's interest “proposition #1” signified MBC (Frost, 1998c). (B) The literature discusses further things. They include the “bone anabolic” effects of parathyroid hormone (Takahashi et al., 1991; Jerome, 1994; Reeve, 1996) and some prostaglandins (Norrdin et al., 1990; Jee, 1995); and estrogen effects on bone (Lindsay and Tohme, 1990; Turner et al., 1994; Jee, 1995; Bilezikian et al., 1996; Schiessl et al., 1998). Other studies discussed androgen, growth hormone, bisphosphonate, and other effects on bone and its mechanostat (Riggs et al., 1990; Andreassen et al., 1995; Frost, 1998ac; Gasperino, 1995; Frost and Schönau, 2000); bone's transient and steady states, its regional acceleratory phenomenon, and its chondral modeling barrier (Frost, 1986); bone's innervation and vibration effects on bones (Rubin et al., 2003; Torvinen et al., 2003); and a so-far unstudied mechanism that could cause bone losses close to marrow (Jaworski et al., 1972).

(C) Currey (2003) recently suggested that evolution made bones strong enough to carry muscle forces without breaking to minimize traumatic fractures, and to let a minimum mass of bone do those things, which endowed healthy bones with a strength-safety factor. He also noted that beneficial developments in physiology achieved by natural selection should reside in the genome. If so, ultimately genes should determine the features of his views and of the updated mechanostat hypothesis. Currey (2003) suggested what natural selection achieved during evolution, whereas the mechanostat hypothesis suggests how that was achieved in ways that have predictive power and that could provide nearly the same strength-safety factors for large and small LBBs in both mice and humans. That traumatic fractures far outnumber nontraumatic ones in bony vertebrates suggests that bone design minimizes the latter fractures more effectively than the former ones. Again, time and more work must resolve any seeming conflicts between these views. Although “here be dragons” may characterize some future debates about these views (James, 2001), eventually a consensus about their merits should emerge.

(D) Until such a consensus is reached, it seems appropriate to note five things. Reasonable people can devise more than one explanation for the same facts; it is more difficult to validate a hypothesis than to validate the facts that led to it; controversies in a field usually mean that progress is being made; poor interdisciplinary communication in skeletal science and medicine usually delays progress; and no hypothesis can invalidate another one.

10) Mechanostats in Extraosseous Load-Bearing Skeletal Organs?

Replacing the word “bone” in Relation 2 with fascia, joint, ligament, tendon, or growth plates could allow that relation to apply to them as well. They may have their own mechanostats with analogs of the presumed chief mechanical function of bone's mechanostat, and they may have the same general biomechanical relation, but with currently unknown values for its entries for nonosseous tissues and organs (Frost, 2000b,d, 2001a,b, 2003b; Takahashi, 1995).


This update represents the end of a personal journey that began in 1945. Legions of researchers helped by providing information and finding flaws in early speculations. This update identifies numerous subjects for future study and discussion, and it suggests some of the mechanostat's implications for clinical, pharmaceutical, and research matters, as well as for nonosseous skeletal organs. This work provides a foundation to build on. It is a privilege to share its features with today's skeletal science and clinical communities.