1. Evolutionary Biology
  2. Human Biology and Medicine
Download icon

Computed tomography shows high fracture prevalence among physically active forager-horticulturalists with high fertility

  1. Jonathan Stieglitz  Is a corresponding author
  2. Benjamin C Trumble
  3. HORUS Study Team
  4. Caleb E Finch
  5. Dong Li
  6. Matthew J Budoff
  7. Hillard Kaplan
  8. Michael D Gurven
  1. Université Toulouse 1 Capitole, France
  2. Institute for Advanced Study in Toulouse, France
  3. Arizona State University, United States
  4. University of Southern California, United States
  5. Emory University, United States
  6. Harbor-University of California at Los Angeles Medical Center, United States
  7. Chapman University, United States
  8. University of California, Santa Barbara, United States
Research Article
  • Cited 2
  • Views 435
  • Annotations
Cite this article as: eLife 2019;8:e48607 doi: 10.7554/eLife.48607

Abstract

Modern humans have more fragile skeletons than other hominins, which may result from physical inactivity. Here, we test whether reproductive effort also compromises bone strength, by measuring using computed tomography thoracic vertebral bone mineral density (BMD) and fracture prevalence among physically active Tsimane forager-horticulturalists. Earlier onset of reproduction and shorter interbirth intervals are associated with reduced BMD for women. Tsimane BMD is lower versus Americans, but only for women, contrary to simple predictions relying on inactivity to explain skeletal fragility. Minimal BMD differences exist between Tsimane and American men, suggesting that systemic factors other than fertility (e.g. diet) do not easily explain Tsimane women’s lower BMD. Tsimane fracture prevalence is also higher versus Americans. Lower BMD increases Tsimane fracture risk, but only for women, suggesting a role of weak bone in women’s fracture etiology. Our results highlight the role of sex-specific mechanisms underlying skeletal fragility that operate long before menopause.

https://doi.org/10.7554/eLife.48607.001

Introduction

Compared to other hominoids and extinct hominins, modern humans have postcranial skeletons that are more gracile (i.e. lower bone mass and strength for body size) (Cotter et al., 2011; Ruff et al., 2015; Ryan and Shaw, 2015). Declining skeletal strength has been documented in the cortical structure of long bone diaphyses (e.g. size or shape in a cross-section) and in trabecular bone micro-structure (e.g. thickness, bone volume fraction) and is particularly evident in the later Pleistocene or Holocene (Chirchir, 2019; Chirchir et al., 2015; Ruff et al., 2015; Ryan and Shaw, 2015). Physical inactivity is the most common explanation for skeletal fragility, based largely on the well-established principle that impact forces from load bearing and muscle contraction trigger bone deposition (Carter and Orr, 1992). According to this principle, which has been documented in numerous studies of competitive athletes and exercise interventions (e.g. Polidoulis et al., 2012; Warden et al., 2014), and inferred from skeletal remains of prehistoric populations (e.g. Macintosh et al., 2017; Ruff et al., 2015), bone responds to physical activity demands by adding tissue and altering cross-sectional distribution in the direction of highest bending strains (i.e. change in length per unit length) (but see Demes et al., 2001; Lieberman et al., 2004; Lovejoy et al., 2003 and references therein). This mechanical response of bone to loading is variable throughout the body, depending on types of weight-bearing activity and muscle function.

Evolutionary life history theory provides a broad explanatory framework that incorporates ultimate and proximate levels of analysis for understanding variability in bone strength (i.e. ability to withstand an applied load). In all organisms, limited resources are allocated to competing metabolic demands so as to optimize biological fitness. Due to higher fitness gains of reproduction earlier versus later in life (Williams, 1957), natural selection often prioritizes investments in earlier reproduction over somatic maintenance. Organisms may thus increase fertility at the expense of maintenance (Kirkwood and Austad, 2000), and we should expect greater energetic investments in reproduction to trade-off against investments in maintenance (Stearns, 1992). Consistent with life history theory, reproductive effort is expected to moderate effects of physical activity on adult bone strength, which is an indicator of energetic investment in somatic maintenance.

Tests of this life history trade-off in humans are inconclusive (Le Bourg, 2007), in part because many studies focus on mortality rather than investments in maintenance per se (but see Ryan et al., 2018; Ziomkiewicz et al., 2016), precluding direct analysis of whether greater reproductive effort inhibits maintenance. Bone tissue is ideal for examining metabolic trade-offs between reproduction and maintenance. Constant remodeling is necessary to maintain bone strength, and the skeleton is a general mineral reservoir for the competing metabolic demands of maternal maintenance and fetal bone accretion or lactation (Stieglitz et al., 2015). The average full-term human fetus has ~30 g calcium, 20 g phosphorus and 0.8 g magnesium, and at least 80% of these macro-minerals are accreted in the third trimester (see Kovacs, 2016 for a comprehensive review of bone metabolism during pregnancy, lactation and post-weaning). For an average-sized fetus this corresponds at week 24 of gestation to a mean calcium [phosphorus] transfer rate of ~60 mg/day [~40 mg/day] and between weeks 35–40 of 300–350 mg/day [200 mg/day]. In the third trimester, hourly fetal transfers of calcium and phosphorus are between 5 and 10% of that present in maternal plasma, which is enough to provoke maternal hypocalcemia and hypophosphatemia. Generally, patterns of bone turnover are similar comparing pre- to early pregnancy states, but turnover increases during the third trimester to create a net resorptive state. During the first six months [second six months] postpartum, ~200 mg/day [~120 mg/day] of calcium is secreted into human breast milk. Analyses of bone turnover markers, bone mineral density (BMD), and bone structure by high-resolution peripheral quantitative computed tomography (HR-pQCT) suggest that lactating women are in negative calcium balance, especially when milk production is elevated. Longitudinal studies show consistent declines in lactating women’s BMD or bone mineral content, with mean declines of 3–10% after 3–6 months of lactation. The greatest BMD losses (5–10%) occur in the lumbar spine, with modest losses (<5%) occurring at sites with less trabecular bone, and smaller losses (<2%) at sites containing mostly cortical bone. The few studies utilizing HR-pQCT in the limbs (radius, femur) show smaller (<2%) reductions in trabecular thickness and cortical thickness and volume, with greater reductions among women who lactate longer. Thus, because mineral allocations to maternal maintenance and reproduction draw from the same skeletal reservoir, direct trade-offs between these competing demands should manifest in bone. These trade-offs are expected to manifest in the longer term, regardless of whether maternal bone tissue fully or only partially recovers from mineral losses following a specific bout of gestation and lactation. Life history theory makes no assumptions or predictions about the extent of bone mineral recovery (i.e. whether full or partial) after weaning a specific child. Accordingly, a general hypothesis from life history theory is that greater reproductive effort constrains the ability of bone tissue to respond to mechanical loading and high physical activity levels (PALs). This hypothesis is not an alternative to and may complement other hypotheses of bone structural variation emphasizing developmental factors affecting the trade-off between investment in growth and reproduction (Macintosh et al., 2018). But unlike other hypotheses of bone structural variation derived from life history theory or proximate explanations (focusing, for example on nutrition, inflammation, hormones), the hypothesis emphasizing effects of reproductive effort uniquely predicts sex differences within and between populations, the magnitude of which should be influenced by relative investment in reproduction.

Timing of reproduction, in addition to lifetime reproductive effort, is also expected to affect bone strength (Macintosh et al., 2018). Peak bone mass is typically not achieved until the late 20s, so earlier pregnancy and lactation can disrupt maternal bone growth and/or mineralization (Madimenos et al., 2012; Stieglitz et al., 2015), potentially reducing peak bone mass and thus later-life bone strength (but see Chantry et al., 2004). In addition to early onset of reproduction, short interbirth intervals (IBIs) can potentially generate unbalanced cycles of maternal bone resorption and formation, limiting maternal skeletal recovery before subsequent pregnancy (Stieglitz et al., 2015). Whether the maternal skeleton is fully or partially restored post-weaning to its prior mineral content and strength, despite potentially lasting micro-architectural changes shown in recent imaging studies, is currently debated (Clarke and Khosla, 2010; Kovacs, 2016; Kovacs, 2017; Wysolmerski, 2010) and physiological mechanisms underlying post-weaning maternal skeletal recovery are not well understood. Dual-energy X-ray absorptiometry data suggest that for most women, lactation-associated BMD losses and micro-architectural deficits are reversed by 12 months post-weaning, although patterns can vary by skeletal site and most studies are conducted among well-nourished women with low fertility (Kovacs, 2016). Restorative capacity partly depends on lactation duration; HR-pQCT studies of the radius and femur show recovery of trabecular and cortical micro-architecture in women lactating for shorter periods, but incomplete recovery in women lactating for longer periods. A recent HR-pQCT study found incomplete recovery of trabecular and cortical micro-architecture in the tibia and radius after a median of 2.6 years post-weaning for women exclusively breastfeeding for five months (Bjørnerem et al., 2017). Incomplete restoration of lumbar spine BMD to pre-pregnancy values by 12 months post-partum is observed among rural Gambian women practicing on-demand breastfeeding for about two years (Jarjou et al., 2010).

A relevant literature on ‘maternal depletion syndrome’ (Jelliffe and Maddocks, 1964) examines trade-offs between reproductive effort and maternal health more broadly in high fertility contexts. Currently, evidence for maternal depletion is mixed (Gurven et al., 2016; Tracer, 2002), though most prior studies focus only on the period covering one or two births rather than the cumulative long-term effects on health of repeated pregnancies. Moreover, most of these prior studies focus on maternal anthropometric status (e.g. weight, adiposity) rather than bone tissue per se. Bone tissue fluctuates less than anthropometric markers with short-term changes in energy balance, rendering studies of bone less susceptible to sampling biases.

Here, we examine in a natural fertility population, Tsimane forager-horticulturalists of Bolivia, whether greater reproductive effort compromises bone strength, particularly for women given their greater energetic costs of reproduction. Tsimane are an ideal population to test whether women’s greater reproductive effort compromises bone strength. Tsimane fertility is high (total fertility rate = 9 births per woman), birth spacing is short (Stieglitz et al., 2015), breastfeeding is on-demand, effective birth control is rare, and PALs are high (Gurven et al., 2013), as is typical of other small-scale rural subsistence populations. In a population-representative sample of adults aged 40+ years, who mostly have completed their reproduction, we utilize thoracic computed tomography (CT) to measure two primary indicators of bone strength in thoracic vertebrae: BMD, which accounts for ~70% of the variance in bone strength (NIH Consensus Development Panel on Osteoporosis Prevention, Diagnosis, and Therapy, 2001), and fracture prevalence and severity. We focus on thoracic vertebrae since spontaneous thoracic vertebral fractures are among the most common osteoporosis-related fractures in humans (Sambrook and Cooper, 2006). Such fractures have not been observed in wild or captive apes, even in individuals with severe osteopenia (Gunji et al., 2003), suggesting that modern humans are more susceptible than other primates to osteoporosis-related fractures (Cotter et al., 2011).

Most activities of daily living, including sitting, walking, running and lifting, generate loads on human vertebrae (Myers and Wilson, 1997; Rohlmannt et al., 2001; Stewart and Hall, 2006), and thus thoracic vertebrae track mechanics of both lower and upper limbs. Even regular breathing appears to generate intradiscal pressure and some loading in the thoracic spine (Polga et al., 2004). Vertebral bodies, which are composed mostly of trabecular bone surrounded by a thin cortical shell, function largely as shock absorbers and can deform to a greater degree than tubular bones (Seeman, 2008); deformation facilitates spinal flexion, extension and rotation. The major loading mode on human vertebral bodies is axially compressive, and most axial force is carried by the trabecular bone (Myers and Wilson, 1997). For many activities (e.g. neutral standing, standing with weight, mild trunk flexion and extension, lifting objects above the head), the greatest compressive loads along the spine are generated in the thoracolumbar region (Bruno et al., 2017). The greatest compressive vertebral loads occur during activities in which body mass or externally applied weights are shifted anteriorly, such as during trunk flexion or carrying weight in front of the body (Bruno et al., 2017; Polga et al., 2004; Rohlmannt et al., 2001). To accommodate forces, the architecture of the vertebral body trabecular bone consists of thick vertical plates and columns supported by thinner horizontal trabeculae. This trabecular structure changes with age, such that vertical plates are successively perforated during remodeling and converted into columns, whereas horizontal trabeculae perforate and disappear (Mosekilde et al., 1987). These age-related trabecular structural changes can result in vertebral strength declines that are greater than predicted from bone mass estimation alone. Vertebral strength is compromised more by loss of trabecular connectivity than by trabecular thinning, and women are more susceptible than men to age-related horizontal trabecular perforation and disappearance (Mosekilde, 1989).

In this paper, we first test, among Tsimane women, whether greater reproductive effort – indicated by earlier age at first birth, higher parity and shorter IBIs – is associated with reduced thoracic vertebral BMD. We then test whether Tsimane BMD is lower, particularly for women, than a matched American sample with directly comparable CT-derived indicators of bone strength. This latter prediction follows from the hypothesis that greater reproductive effort compromises bone strength, which is consistent with a trade-off between investment in reproduction and maintenance as posited by evolutionary life history theory. In contrast, a simple prediction from a physical inactivity hypothesis for compromised bone strength posits the opposite, that is, lower BMD for Americans than Tsimane, for both sexes, given lower PALs, on average, among Americans. Regarding the second measured bone strength indicator, thoracic vertebral fracture, we test whether Tsimane fracture prevalence is higher than matched Americans, particularly for women. To determine whether Tsimane thoracic vertebral fracture results from compromised bone strength as opposed to trauma, we test whether fracture risk is inversely associated with thoracic vertebral BMD. Lastly, we test whether Tsimane women’s fracture risk increases with reproductive effort, even after adjusting for BMD, which is expected if greater reproductive effort compromises bone micro-architecture in complex ways beyond just reducing mineral density (e.g. by reducing trabecular thickness or connectivity density).

Results

Tsimane women’s thoracic vertebral BMD declines with early age at first birth and short IBI

Earlier age at first birth is associated with reduced BMD (Std. βAge at 1st birth [years, logged]=0.099, p=0.036, controlling for age and fat-free mass, adj. R2 = 0.51, n = 213; Appendix 1—table 1). Back-transforming logged age at first birth values into observed values and holding controls at sample means, there is a BMD difference of 0.57 SDs for women with maximum versus minimum age at first birth (37 versus 12 years, respectively).

Parity (continuously or categorically operationalized) is not associated with BMD controlling for age at first birth, age and fat-free mass (Appendix 1—table 2), nor does parity interact with any indicator of reproductive effort to predict BMD.

Shorter mean IBI (<29.7 months) is associated with lower BMD (Std. βShorter mean IBI=-0.201, p=0.032, controlling for age at first birth, age and fat-free mass, adj. R2 = 0.52; Appendix 1—tables 34). Mean IBI also interacts with age at first birth: BMD is 0.28 SDs higher for women with longer mean IBI and later age at first birth versus women with shorter mean IBI and earlier age at first birth (interaction p=0.027, controlling for age and fat-free mass; see Figure 1 and Appendix 1—figure 1).

Standardized residual thoracic vertebral BMD (mean ±SEM; controlling for age and fat-free mass) by reproductive effort (n = 212 Tsimane women).

Earlier vs. later age at first birth is defined as <18 vs. ≥18 years (median split), respectively; lower vs. higher parity is defined as ≤9 vs. >9 live births (median split), respectively; and shorter vs. longer mean IBI is defined as <29.7 months vs. ≥29.7 (median split), respectively.

https://doi.org/10.7554/eLife.48607.002

Additionally controlling for indicators of modernization, that is, residential proximity to the closest market town of San Borja, Spanish fluency and schooling, which could reflect differential activity levels, diet and/or other factors affecting bone strength (e.g. infectious burden), strengthens the association between BMD and both mean IBI and age at first birth (comparing estimates in Appendix 1—table 5 to those in Appendix 1—table 4). BMD is not significantly associated with any modernization indicator. Neither young age at menarche nor menopause is associated with BMD (Appendix 1—table 6), nor does either variable interact with any indicator of reproductive effort to predict BMD.

Thoracic vertebral BMD is lower for Tsimane than Americans, but only for women

For women, age-standardized mean BMD is 8.9% lower for Tsimane than Americans (Figure 2; Appendix 1—table 7). BMD is significantly lower (all p’s < 0.01) for Tsimane than Americans at all ages, and this population-level difference increases with age in the cross-section. Mean BMD for Tsimane women is 5.6% lower than age-matched Americans aged 40–49 years, 9.8% lower for ages 50–59, 13.7% lower for ages 60–69, and 18.8% lower for women aged 70+. For men, age-standardized mean BMD is 0.24% lower for Tsimane than Americans. Tsimane BMD is not significantly different at ages 40–49 (1.0% lower for Tsimane, p=0.547), 60–69 (5.2% lower for Tsimane, p=0.069) and age 70+ (2.5% lower for Tsimane, p=0.543); Tsimane men aged 50–59 have 4% higher mean BMD (p=0.027) than age-matched Americans.

Tsimane versus American mean thoracic vertebral BMD (95% CIs) by age and sex.

The ‘total’ category is age-standardized (see Appendix 1—table 7 for values and details).

https://doi.org/10.7554/eLife.48607.003

Tsimane thoracic vertebral fracture prevalence is higher than Americans

For women, age-standardized prevalence of any thoracic vertebral fracture (i.e. grade ≥1; including mild, moderate or severe) for Tsimane and Americans is 18% and 9%, respectively (adjusted RRTsimane [vs. US]=1.81, 95% CI: 1.16–2.83, p=0.009, controlling for age, n = 491; see Figure 3 and Appendix 1—table 8). Using a more conservative fracture definition (i.e. grade ≥2; including only moderate or severe), age-standardized prevalence for Tsimane and Americans is 6% and 2%, respectively (adjusted RRTsimane [vs. US]=2.69, 95% CI: 1.07–6.75, p=0.035; Appendix 1—table 9). For men, age-standardized prevalence of any fracture for Tsimane and Americans is 36% and 11%, respectively (adjusted RRTsimane [vs. US]=3.30, 95% CI: 2.26–4.82, p<0.001, n = 524; Appendix 1—table 8), and using a more conservative fracture definition, 10% and 2%, respectively (adjusted RRTsimane [vs. US]=3.79, 95% CI: 1.78–8.09, p=0.001; Appendix 1—table 9).

Thoracic vertebral (T6–T12) fracture prevalence (95% CIs) for Tsimane (in gray) and American (in green) women (A) and men (B) by fracture grade.

Prevalence based on a less (grade ≥1) and more (grade ≥2) conservative fracture definition is shown. Prevalence is estimated from log-binomial generalized linear models adjusting for age.

https://doi.org/10.7554/eLife.48607.004

For both sexes, Tsimane are significantly more likely than Americans to present moderate (i.e. grade 2) but not severe (grade 3) fracture (Appendix 1—tables 1011). Tsimane men but not women are significantly more likely than Americans to present borderline deformity (grade 0.5) and mild fracture (grade 1).

Lower thoracic vertebral BMD increases risk of thoracic vertebral fracture among Tsimane, particularly for women. Short IBI additionally increases Tsimane women’s fracture risk

Tsimane women with thoracic vertebral fracture (grade ≥1) have lower BMD (0.46 SDs; p=0.005), higher parity (0.25 SDs; p=0.071) and shorter mean IBI (0.22 SDs; p=0.011) than women without fracture (Appendix 1—table 12). There are no significant differences between women without versus with fracture in terms of age, anthropometrics, ages at menarche, menopause or first birth, or modernization indicators. Women’s BMD is inversely associated with fracture risk (adjusted RRBMD per SD increase = 0.542, 95% CI: 0.352–0.837, p=0.006, controlling for age, height and fat mass, n = 219); this association remains (adjusted RRBMD = 0.540, 95% CI: 0.345–0.845, p=0.007; Figure 4) after adding to the model mean IBI, which is also inversely associated with fracture risk (adjusted RRMean IBI per SD increase = 0.379, 95% CI: 0.165–0.866, p=0.021; Appendix 1—table 13: Model 2). The inverse association between women’s BMD and fracture risk strengthens with higher fracture grades (Appendix 1—table 14), whereas the inverse association between mean IBI and fracture risk is strongest for mild (i.e. grade 1) fracture. In multivariate models, we found no effect on women’s fracture risk of either parity, ages at menarche, menopause or first birth, or modernization indicators (controlling for BMD, mean IBI, age, height and fat mass).

Association between thoracic vertebral BMD and thoracic vertebral (T6–T12) fracture risk (95% CI) for Tsimane.

Log-binomial generalized linear models are used to estimate relative risk after adjustment for age, height and fat mass. Mean IBI is also included as a covariate for women. Parameter estimates are shown in Appendix 1—tables 1314 (women), and Appendix 1—tables 1617 (men). Severe fractures (grade 3) are omitted due to their relative scarcity.

https://doi.org/10.7554/eLife.48607.005

Tsimane men with fracture are shorter (0.23 SDs; p=0.038), have higher adiposity (0.37 SDs; p=0.025) and have higher BMI (0.34 SDs; p=0.012) than men without fracture (Appendix 1—table 15). As with women, men without versus with fracture do not significantly differ in terms of age, weight, fat-free mass or modernization indicators. But unlike women, there is no significant difference in BMD between men without versus with fracture, and men’s BMD is not associated with risk of any fracture (adjusted RRBMD per SD increase = 1.046, 95% CI: 0.862–1.269, p=0.650, controlling for age, height and fat mass, n = 227; Appendix 1—table 16). Men’s BMD is, however, inversely associated with risk of borderline (i.e. grade 0.5) deformity (Appendix 1—table 17; Figure 4). Modernization indicators are not associated with men’s fracture risk, nor do they meaningfully moderate associations between BMD and fracture risk.

Discussion

This paper makes five empirical contributions. We find that: 1) Tsimane thoracic vertebral BMD is lower among women with early age at first birth and short IBIs; 2) Tsimane BMD is lower than a directly comparable American sample, but only for women; 3) thoracic vertebral fracture prevalence is higher for Tsimane than Americans; 4) among Tsimane, lower BMD is associated with higher fracture risk, particularly for women; and 5) short IBIs are associated with Tsimane women’s fracture risk even after adjusting for BMD.

These results are consistent with a life history trade-off between reproductive effort and somatic maintenance (Kirkwood and Austad, 2000; Stearns, 1992; Stieglitz et al., 2015). Maternal physiology uniquely responds to the greater mineral demands of pregnancy and lactation by mobilizing skeletal mineral stores (Kovacs, 2016; Nelson et al., 2002; Prentice, 2003). Maternal regulatory mechanisms can compensate for acute mineral losses, for example, by retaining excess mineral in circulation to facilitate storage. But in energy-limited settings (due to high energy expenditure relative to consumption), earlier age at first birth and shorter birth spacing can potentially reduce peak bone mass and generate unbalanced cycles of bone resorption and formation, compromising bone strength long before menopause. In addition to direct negative effects of reproductive effort on bone strength, indirect negative effects are possible via reductions in women’s PALs and mechanical loading on bone, given nursing women’s reduced food acquisition efforts relative to non-nursing women (cf. Hurtado et al., 1992). Tsimane women’s participation in certain subsistence activities ostensibly entailing high-impact and high-magnitude loadings (e.g. rice-pounding using thick wooden bats) may be curtailed during pregnancy and early lactation; meta analyses of controlled exercise trials indicate that high-impact and high-magnitude loadings are especially effective at preserving women’s BMD at the lumbar spine, femoral neck and total hip (Martyn-St James and Carroll, 2009; Martyn-St James and Carroll, 2010). In the present study, we also found that parity per se is not inversely associated with BMD (Appendix 1—tables 23; but see Stieglitz et al., 2015). Prior studies of parity-specific effects on maternal BMD do not reveal consistent associations (Kovacs, 2016), although most studies are conducted in developed countries with lower fertility and greater energetic surpluses. These conditions can obscure expected energetic trade-offs between reproduction and maintenance.

The finding that thoracic vertebral BMD is lower for Tsimane than in a comparable sample of American women, whereas for men few population-level differences are apparent, despite higher mean PALs of Tsimane women and men, contributes to a growing literature emphasizing sexual dimorphism in skeletal responses to environmental stimuli (e.g. Macintosh et al., 2017), mediated in part by sex and growth hormones, growth factors and their receptors (Devlin, 2011; Gosman et al., 2011). Dietary or other systemic population-level differences (e.g. in infectious burden) do not easily explain Tsimane women’s lower BMD because such systemic factors should affect both sexes. Recent dietary analyses of Tsimane and Americans (NHANES) based on 24 hr dietary recalls indicate minor sex differences in proportional macronutrient intake for both populations (Kraft et al., 2018). In absolute terms, Tsimane mean daily per-capita energy, protein and carbohydrate intake actually exceeds that of Americans for both sexes, as does Tsimane intake of the bone-forming minerals magnesium, phosphorus and zinc. Nevertheless, prior Tsimane research indicates that energetic limitation and greater immune activation from high pathogen exposure partly explain why mean PALs above those of industrialized societies do not inevitably yield elevated peak bone mass or protect against age-related bone loss for either sex (Stieglitz et al., 2015; Stieglitz et al., 2016). In a prior cross-sectional study of women utilizing ultrasound, we found reduced calcaneal strength for Tsimane versus Americans that is already apparent in the 20s, with population-level differences increasing with age (Stieglitz et al., 2015). Between- and within-population analyses in the present study suggest that sex-specific costs of reproduction contribute to skeletal sexual dimorphism, and that sexually dimorphic responses to reproduction manifest early in adulthood, depending on age at first birth and birth spacing (Madimenos et al., 2012; Stieglitz et al., 2015).

That Tsimane fracture prevalence is higher than a comparative American sample (Figure 3 and Appendix 1—tables 811) may be surprising in light of the presumably protective higher lifetime moderate physical activity of Tsimane (Gurven et al., 2013), and their minimal exposure to other osteoporosis risk factors found in industrialized societies (e.g. glucocorticoid therapy, excessive smoking or alcohol consumption). By restricting the Tsimane sample to older adults (aged 40+ years), we minimize potential for cohort effects related to changing lifestyles associated with modernization. This is supported by the fact that study findings hold even after controlling for modernization indicators, which are not associated with thoracic vertebral BMD or fracture risk (Appendix 1—tables 56, 12, 15). Higher fracture prevalence among Tsimane versus Americans is noteworthy in light of relatively scant bio-archaeological evidence of osteoporotic fracture prior to industrialization (Agarwal, 2008; Curate et al., 2010) and evidence of increasing age-specific osteoporotic fracture incidence rates over time in Western populations (Cooper et al., 2011). But these results are not surprising in light of high Tsimane fertility and life history trade-offs between energetic investments in reproduction and bone growth and/or maintenance, and other factors (e.g. calcium deficiency, chronic immune activation due to high pathogen exposure), which may interact with high reproductive effort to further constrain the ability of bone tissue to respond to mechanical loading and high PALs (cf. Armelagos et al., 1972).

The fact that Tsimane women’s vertebral BMD is inversely associated with vertebral fracture risk (cf. Mays, 1996) (Figure 4) suggests a major role of compromised bone strength in precipitating fracture, rather than traumatic injury of otherwise healthy bone. For Tsimane men, it is likely that trauma plays a major role in precipitating fracture due to: 1) a weak association between BMD and fracture risk (Figure 4); 2) a high fracture prevalence relative to a comparative American sample (Figure 3) despite minimal BMD differences (Figure 2); and our anecdotal observations of high levels of mechanical stress on Tsimane men’s vertebrae from frequent heavy load carrying (e.g. of hunted game, timber for constructing houses). This of course does not preclude a contributing role of compromised bone strength (e.g. from micro-architectural deficiencies) in precipitating Tsimane men’s fracture. Likewise, for women, the present results do not preclude a contributing role of trauma in precipitating fracture. Future research is needed to determine the extent to which women’s subsistence involvement and frequent carrying of young children and other loads (e.g. woven bags filled with harvested cultigens) influence bone structural integrity. Some of the greatest compressive vertebral loads occur when weights are carried in front of the body (Bruno et al., 2017; Rohlmannt et al., 2001), which is how Tsimane women routinely carry infants and toddlers. The fact that Tsimane women’s shorter IBIs predict increased fracture risk even after adjusting for BMD, which also remains a significant predictor (Appendix 1—table 1314), suggests that shorter IBIs compromise multiple aspects of vertebral micro-architecture (e.g. trabecular thickness or connectivity density), although further research is needed to examine this possibility.

Inferring behaviors underlying morphological variation in past human populations

While our results do not directly address debates over the timing of and prior selection pressures underlying transition to skeletal gracility in past human populations, our results do provide insight into lifestyle factors affecting bone strength which may have been relevant during this transition. It has been hypothesized that subsistence transition from hunting and gathering to more sedentary agriculture, and increasing reliance on labor-saving technology, caused reductions in mechanical loading on bone and PALs, leading to modern human skeletal gracilization (Chirchir, 2019; Chirchir et al., 2015; Chirchir et al., 2017; Ruff et al., 2015; Ryan and Shaw, 2015). Increasing reliance on agriculture may have been associated with reductions in terrestrial mobility, and thus reduced mechanical loading of lower limbs, although the pace and magnitude of these changes likely varied temporally and spatially. Changes in upper body activities may have been much more variable during subsistence transitions, so upper limb loading may have actually increased with agriculture in some regions (Bridges, 1989). Nevertheless, comparisons of skeletal remains of hunter-gatherers and either full- or part-time agriculturalists indicate among agriculturalists reduced femoral strength, as indicated by trabecular bone structure or external size dimensions (Larsen, 1981; Ryan and Shaw, 2015), and accelerated age-related decline in radial bone mineral content (Perzigian, 1973).

Yet a physical inactivity explanation for modern human skeletal gracility – rooted in a subsistence transition from foraging to farming – is puzzling for several reasons. Evidence that agriculturalists are more sedentary than hunter-gatherers is not particularly strong: PALs and time allocation to work vary substantially within a subsistence regime, and both measures are actually higher among agriculturalists (Gurven et al., 2013; Leonard, 2008). Moreover, children in agricultural societies generally begin work earlier than hunter-gatherer children (Kramer, 2005). This is significant because higher PALs in childhood and early adulthood, particularly for higher-impact activities producing high peak stresses, increase peak bone mass, size and later-life bone strength (e.g. see Warden et al., 2014 and references therein). Furthermore, in a prospective Tsimane study we found that time spent in horticulture positively predicted ultrasound-derived indicators of radial strength (Stieglitz et al., 2017), whereas tibial strength was not predicted by extent of involvement in any subsistence activity including hunting. While various studies show reduced lower limb strength among agriculturalists relative to pre-agriculturalists, as indicated by cross-sectional diaphyseal structure of cortical bone or trabecular bone volume fraction or thickness (May and Ruff, 2016; Saers et al., 2016), other studies using similar diaphyseal structural properties or external size dimensions show no differences in limb strength or dissimilar patterns by subsistence regime (Bridges, 1989; Ruff, 1999). Taken together, these observations create uncertainty over the timing of and selection pressures underlying modern human skeletal gracility. This uncertainty is exacerbated given evidence of sex-specific skeletal responses to mechanical loading and physical activity (Bridges, 1989; Macintosh et al., 2017; Ruff, 1999). Additional sex-specific factors may have contributed to skeletal gracility during transition to farming, either independently and/or in interaction with PALs.

Our results instead suggest that fertility increases associated with subsistence transition from foraging to farming (Bentley et al., 1993) contributed to modern human skeletal gracility, particularly for women. Numerous proximate determinants have been proposed to explain fertility increases that accompanied greater energetic surpluses from agriculture, including earlier age of menarche and first birth, and shorter IBIs (e.g. due to earlier weaning and supplementary infant feeding) (e.g. Campbell and Wood, 1988). Regardless of proximate fertility determinants, given the trade-off between energetic investment in reproduction and somatic maintenance, our results suggest that women’s skeletons were especially susceptible to gracilization during subsistence transition from foraging to farming in light of the associated increases in fertility, and reduced mobility resulting from increased fertility (cf. Hurtado et al., 1992). Sex-specific mechanisms beyond menopause underlying skeletal gracility during this subsistence transition should thus be considered, in addition to explanations emphasizing increased sedentism, reliance on labor-saving technology and associated reductions in mechanical loading on bone. Nevertheless, since skeletal gracilization is observed in both sexes during this subsistence transition (Ruff et al., 2015), any explanation solely resting on changes in reproductive effort cannot explain these morphological changes in men.

Study limitations

The cross-sectional study design using retrospective demographic data limits our ability to establish that greater reproductive effort causes BMD reductions and fracture. The results presented here may thus be consistent with alternative interpretations derived from life history theory or proximate explanations, although such alternatives must address the observed population-level sex differences. Another limitation is that our measures of reproductive effort are indirect measures of reproductive costs, and we lack data on lactation duration or intensity. However, all Tsimane women breastfeed their infants and it is common for women to breastfeed exclusively for about four months (no study participant bottle fed an infant). We also lack estimates of vertebral size and geometry, which affect vertebral strength, and we lack individual-level data on activity level, nutrient intake and pathogen burden for population-level comparisons of BMD and fracture prevalence and severity. We also do not consider whether genetic diversity determines heterogeneity in BMD or fracture prevalence. Bone strength indicators such as BMD and fracture are heritable and the frequency of alleles affecting BMD differs between ethnically distinct populations (see Wallace et al., 2016 and references therein). But while genetic factors can account for a sizable portion of variance in bone strength within populations, there is little evidence that heterogeneity in bone strength between populations is due to stochastic genetic diversity.

Conclusion

This study examines direct indicators of bone strength from a clinically and mechanically relevant anatomic region using in vivo imaging in a physically active population with high fertility. Results suggest a trade-off between reproductive effort and bone strength, and that greater reproductive effort constrains the ability of bone tissue to respond to mechanical loading and high physical activity. Results also raise the possibility that increased fertility associated with subsistence transition from foraging to farming promoted modern human skeletal gracility, particularly among women. Because of the complex nature of lifestyle transformations during subsistence transitions, including apparent increases in infectious disease, nutritional deficiencies and dental decay (Cohen and Armelagos, 1984), an expanded conceptual framework incorporating diverse lifestyle factors that may constrain the ability of bone to respond to mechanical loading (e.g. high fertility, nutrient deficiency, infection-induced inflammation (Madimenos et al., 2012; Stieglitz et al., 2015; Stieglitz et al., 2016) can improve our understanding of morphological transformations associated with transition to farming. Of course, our ability to make inferences about the past using data collected in contemporary populations is limited. Tsimane are neither ‘pure’ hunter-gatherers nor agriculturalists, and they may differ in important ways from ancestral human populations in terms of residential mobility, fertility, diet and disease exposures. Yet no single population represents the range of experiences across different environments that shaped the evolution of our species over the millennia in which ecologies fluctuated. In vivo study of bone strength in well-characterized, population-representative, non-industrialized societies provides an opportunity to examine lifestyle factors that are often invisible to bio-archaeological inquiry but nonetheless relevant to understanding selection pressures over human history.

Materials and methods

Study population

Request a detailed protocol

Tsimane forager-horticulturalists of lowland Bolivia are semi-sedentary and live in >90 villages. Their diet consists of cultigens grown in small swiddens (62% of total calories; mostly plantains [60% of cultigen-derived calories], rice, sweet manioc and corn), freshwater fish (16%), meat from hunting and domesticated animals (14%), market foods (8%; mostly pasta [55% of market-derived calories]) and wild fruit and vegetables (<1%) (Kraft et al., 2018). Relative to Western dietary recommendations calcium intake is low (~240 mg/day), but intake of other bone-forming minerals is ample (magnesium: ~525 mg/day; zinc: ~14 mg/day) or high (phosphorus: ~1,550 mg/day). Women’s PAL is in the ‘moderate to active’ range (PAL = 1.7–1.9) and remains constant throughout adulthood. Men’s PAL is ‘vigorously active’ (PAL = 2.0–2.2), and declines by 10–20% from the peak (in the late 20s) to older adulthood (age 60+ years) (Gurven et al., 2013).

Study design and participants

Request a detailed protocol

The Tsimane sample includes all individuals who met the inclusion criteria of self-identifying as Tsimane and who were aged 40+ years (n = 507; 48% female; age range: 41–94 years; see Appendix for additional details and Appendix 1—table 18 for descriptives of all study variables). 185 of the 245 participating women (76%) were post-menopausal. No participant reported ever using hormonal contraception or dietary supplements with consistency. No Tsimane was excluded based on any health condition that can affect BMD or fracture risk.

Comparative American BMD data were collected among asymptomatic subjects from greater Los Angeles as part of a different study (described in Budoff et al., 2010). Briefly, 9585 subjects (43% female; mean age = 56) underwent coronary artery calcification (CAC) scanning for evaluation of subclinical atherosclerosis, after exclusion of participants with vertebral deformities or fractures. Subjects had no known bone disease (see Appendix 1—table 7 for additional details). Two American data sources are used to compare Tsimane and American fracture prevalence: a subset from the MESA study (Budoff et al., 2011) and a subset reported in Budoff et al. (2013) (see Appendix 1—table 8 for additional details); data from these American subsets were matched to Tsimane by age, sex and weight (±5 kg), and then merged to create a single American comparison sample.

Institutional IRB approval was granted by UNM (HRRC # 07–157) and UCSB (# 3-16-0766) for the Tsimane research, as was informed consent at three levels: (1) Tsimane government that oversees research projects, (2) village leadership and (3) study participants.

Thoracic computed tomography (CT)

Request a detailed protocol

Tsimane CT scans were conducted at the Hospital Presidente German Busch in Trinidad, Bolivia using a 16-detector row scanner (GE Brightspeed, Milwaukee, WI, USA). A licensed radiology technician acquired a single, ECG-gated non-contrast thoracic scan as part of a broader project on atherosclerosis, including CAC assessment (see Kaplan et al., 2017 and Appendix for additional details). Typical multi-detector CT protocols used for evaluation of CAC include imaging the mid-thoracic spine in the reconstructed field of view, facilitating simultaneous evaluation of thoracic vertebrae during a single examination without additional radiation exposure. Tsimane CT settings were: 250 ms exposure, 2.5 mm slice thickness, 0.5 s rotation speed, 120 kVp, and 40 mA with prospective triggering. Refer to the Appendix for details on CT parameters for the comparative American samples; American CT data were all collected at the same institute (Los Angeles Biomedical Research Institute).

Thoracic vertebral bone mineral density (BMD)

Request a detailed protocol

Vertebral BMD was measured manually in each of three consecutive thoracic vertebrae (T7-T10 range) by a radiologist with 20+ years of experience (see Appendix for additional details). BMD measurement started at the level of the section that contained the left main coronary artery (LMCA) caudally (beginning at either T7 or T8, depending on the origin of the LMCA). The center of the region of interest was located at the center of each vertebrae, with a 2–3 mm distance from the cortical shell; this distance ensured that BMD measurements within the vertebral body excluded the cortical bone of the vertebral shell. For each vertebrae, the radiologist manually positioned a circular region of interest while demarcating cortical from trabecular bone based on visual inspection. Any area with large vessels, bone island fractures and calcified herniated disks were excluded as much as possible from the region of interest with use of the manual free tracing protocol. Mean BMD for the three consecutive thoracic vertebrae was then calculated. This BMD measure is strongly positively correlated (Pearson r’s > 0.9) with lumbar vertebral BMD (Budoff et al., 2012). CT-derived BMD estimates can be obtained with and without calibration phantoms. Phantomless BMD estimates correlate strongly (Pearson r = 0.99) with standard phantom-based CT BMD estimates (Budoff et al., 2013). Hounsfield units were converted to BMD (mg/cm3) using a calibration phantom of known density or a scanner-specific mean calibration factor for the T7-T10 vertebrae from scans performed without the phantom. All BMD measurements used in this study were performed at the Los Angeles Biomedical Research Institute.

Thoracic vertebral fracture

Request a detailed protocol

For each subject the radiologist classified seven vertebrae (T6-T12) according to Genant’s semi-quantitative technique (GST) (Genant et al., 1993). While there is no consensus regarding the radiologic definition of vertebral fracture, the GST provides highly reproducible diagnosis of fractures, is the current clinical technique of choice for diagnosing fracture, and is the most widely used technique for identifying fracture (Shepherd et al., 2015). Based on visual inspection, each vertebra is rated according to severity of loss of vertebral height and other qualitative features, including alterations in shape and configuration of the vertebra relative to adjacent vertebrae and expected normal appearances. Each vertebra is classified into one of five categories: normal (grade 0); mild fracture (grade 1; approximately 20–25% reduction in anterior, middle, and/or posterior vertebral height, and a 10–20% reduction in projected vertebral area); moderate fracture (grade 2; 25–40% reduction in any height and a 20–40% reduction in area); and severe fracture (grade 3; >40% reduction in any height and area). A grade 0.5 indicates borderline deformed vertebra (<20% reduction in any height) that is not considered to be a definitive fracture (see Appendix for additional details). Each subject is assigned one grade representing a summary measure of all seven vertebrae. Subjects with >1 vertebral deformity are classified according to their most severe deformity. Subjects are considered to present vertebral fracture if any vertebral body is graded at least mildly deformed (i.e. grade ≥1); subjects are considered to present no fracture if graded 0 or 0.5. Given recent analyses (Lentle et al., 2018) showing lower observer agreement for mild fractures (the most common) relative to moderate and severe fractures, we repeat analyses using a more conservative fracture definition (i.e. grade ≥2). All fracture measurements used in this study were performed at the Los Angeles Biomedical Research Institute.

Socio-demographics and anthropometrics

Request a detailed protocol

Individuals for whom reliable ages could not be ascertained are not included in analyses (see Appendix for additional details). Reproductive histories were elicited in the Tsimane language. IBI refers to the number of months between live births for women with ≥2 live births. Self-reported ages at menarche and menopause were recorded during medical exams conducted by physicians of the Tsimane Health and Life History Project (THLHP). During annual THLHP census updates we also coded for each participant their village of residence (from which we derived via GPS residential proximity to the closest market town of San Borja), self-reported Spanish fluency (0 = none; 1 = moderate; 2 = fluent) and schooling (# years) as indicators of modernization.

Height and weight were measured during THLHP medical exams using a Seca stadiometer (Road Rod) and Tanita scale (BC-1500). The scale uses a method of bioelectrical impedance analysis to estimate percent body fat. Using weight and percent body fat we calculated fat mass (weight*percent body fat) and fat-free mass (weight – fat mass).

Data analysis

Request a detailed protocol

The two outcome variables indicating bone strength are thoracic vertebral BMD and thoracic vertebral fracture. Sexes are analyzed separately given the sex-specific nature of hypotheses and to minimize confounding by unobserved factors. General linear models are used to test for associations between Tsimane women’s BMD and reproductive effort (see Appendix for additional details). We compared BMD of Tsimane and age- and sex-matched Americans using age-standardized means. We use a parametric test (one sample t test) to evaluate whether population-level differences in mean BMD within each decade are significant (p<0.05), specifying as the test value the US means published in Budoff et al. (2010). We compare Tsimane and American fracture prevalence using age-standardized values, and using log-binomial generalized linear models (GLMs). GLMs are used to test for effects of BMD and women’s reproductive effort on the probability of fracture. Both continuous and categorical (e.g. median split) measures of reproductive effort are used to analyze their associations with BMD and fracture risk. Unless otherwise noted, anthropometric and socio-demographic covariates are included in a stepwise fashion in regressions (see Appendix for descriptive analyses of Tsimane BMD by age and sex, and by anthropometrics [Appendix 1—tables 1920]). Participants with any missing values are removed from analyses.

Appendix

Additional results

Tsimane women’s thoracic vertebral BMD declines with early age at first birth and short IBI

Appendix 1—table 1
General linear model: effect of age at first birth on thoracic vertebral BMD for Tsimane women.
https://doi.org/10.7554/eLife.48607.008
ParameterStd. βp
Age at 1st birth (years, loggeda)0.0990.036
Age (years)−0.665<0.001
Fat-free mass (kg)0.1060.031
Adjusted R20.508
N213
  1. aAge at first birth is not normally distributed and is thus log-transformed (Std. βAge at 1st birth [years]=0.110, p=0.019, controlling for age and fat-free mass).

Appendix 1—table 2
General linear models: effects of parity and age at first birth on thoracic vertebral BMD for Tsimane women.

Continuous and categorical parity measures (models 1–2, respectively) were added to the baseline model shown in Appendix 1—table 1.

https://doi.org/10.7554/eLife.48607.009
Model 1: continuous parity measureModel 2: categorical parity measure
ParameterStd. βpStd. βp
# birthsa−0.0690.216----------
>9 births (vs. ≤ 9 birthsb)----------−0.1230.219
Age at 1st birth (years, logged)0.0750.1450.0810.105
Age (years)−0.656<0.001−0.655<0.001
Fat-free mass (kg)0.1080.0280.1140.021
Adjusted R20.5090.509
N213213
  1. aLogging # births does not substantively affect results.

    bSubstituting other categorical measures of parity (e.g. quartiles) does not substantively affect results.

Appendix 1—table 3
General linear models: effects of mean IBI, parity and age at first birth on thoracic vertebral BMD for Tsimane women.

Continuous and categorical mean IBI measures (models 1–2 and models 3–4, respectively) were added to the models presented in Appendix 1—table 2. Variance inflation factors (all <1.6) do not indicate a high degree of multicollinearity.

https://doi.org/10.7554/eLife.48607.010
Model 1Model 2Model 3Model 4
ParameterStd. βpStd. βpStd. βpStd. βp
Mean IBI (monthsa)0.0740.2460.0740.185--------------------
Short mean IBI (<29.7 months; vs. ≥29.7b)--------------------−0.1850.084−0.1840.075
# births−0.0250.717----------−0.0190.774----------
>9 births (vs. ≤ 9 births)----------−0.0680.528----------−0.0410.712
Age at 1st birth (years, logged)0.0900.0920.0890.0760.1020.0600.1020.046
Age (years)−0.664<0.001−0.661<0.001−0.661<0.001−0.659<0.001
Fat-free mass (kg)0.1070.0290.1110.0250.1060.0310.1080.029
Adjusted R20.5100.5110.5140.514
N212212212212
  1. aLogging mean IBI does not substantively affect results.

    bMean IBI is not calculated for parous women with only one birth (n = 1), hence the reduction in sample size relative to Appendix 1—tables 12.

Appendix 1—table 4
General linear model: reduced form including only significant (p<0.05) predictors of Tsimane women’s thoracic vertebral BMD.
https://doi.org/10.7554/eLife.48607.011
ParameterStd. βp
Short mean IBI (<29.7 months; vs. ≥ 29.7)−0.2010.032
Age at 1st birth (years, logged)0.1090.022
Age (years)−0.663<0.001
Fat-free mass (kg)0.1050.031
Adjusted R20.517
N212
Appendix 1—figure 1
Estimated marginal mean thoracic vertebral BMD (SD units; 95% CI) by mean IBI and age at first birth (n = 212 Tsimane women).

Values are obtained from a general linear model including the following parameters: mean IBI <29.7 months dummy, age at first birth <18 years dummy, mean IBI <29.7 months dummy*age at first birth <18 years dummy, age, and fat-free mass. Mean IBI <29.7 months dummy*age at first birth <18 years dummy interaction p=0.027. Controls (age and fat-free mass) are held at sample means.

https://doi.org/10.7554/eLife.48607.012
Appendix 1—table 5

General linear models: reduced model shown in Appendix 1—table 4 also controlling for indicators of modernization, that is, residential proximity to the closest market town of San Borja (models 1–3), Spanish fluency (model 2) and schooling (model 3).

https://doi.org/10.7554/eLife.48607.013
Model 1Model 2Model 3
ParameterStd. βpStd. βpStd. βp
Short mean IBI (<29.7 months;vs. ≥ 29.7)−0.2080.026−0.2220.020−0.2130.026
Age at 1st birth (years, logged)0.1240.0100.1410.0040.1390.005
Age (years)−0.665<0.001−0.673<0.001−0.675<0.001
Fat-free mass (kg)0.1050.0310.1120.0230.1130.022
Close residential proximity to market town (<34.7 km; vs. ≥34.7a)0.1650.0800.1470.1260.1570.102
Spanish fluency (moderate or high vs. noneb)----------−0.0060.954----------
Schooling (any vs. noneb)--------------------0.0220.853
Adjusted R20.5210.5210.522
N212207207
  1. aMedian split.

    bData are missing for five women.

Appendix 1—table 6
General linear models: reduced model shown in Appendix 1—table 4 also controlling for residential proximity to the closest market town of San Borja (models 1–2), young age at menarche (model 1) and young age at menopause (model 2).
https://doi.org/10.7554/eLife.48607.014
Model 1Model 2
ParameterStd. βpStd. βp
Short mean IBI (<29.7 months;vs. ≥ 29.7)−0.1770.068−0.2500.020
Age at 1st birth (years, logged)0.1320.0080.1630.002
Age (years)−0.659<0.001−0.676<0.001
Fat-free mass (kg)0.1330.0090.1360.017
Close residential proximity to market town (<34.7 km;vs. ≥ 34.7)0.1120.2530.0780.464
Young age at menarche
(<13.3 years vs. ≥ 13.3 a,b)
0.0080.933----------
Young age at menopause
(<50 years vs. ≥ 50 a,c)
----------−0.1760.120
Adjusted R20.5430.559
N188152
  1. aMedian split.

    bData are missing for 24 women.

  2. cSome women either haven’t yet experienced menopause or cannot recall when they experienced menopause, hence the reduction in sample size for model 2.

Thoracic vertebral BMD is lower for Tsimane than Americans, but only for women

Appendix 1—table 7
Mean thoracic vertebral BMD (mg/cm3) for Tsimane and Americansa by age category and sex.
https://doi.org/10.7554/eLife.48607.015
Age category (years)WomenMenTotal
Tsimane (N)USc(N)% difference (from US baseline)dTsimane (N)USc
(N)
% difference (from US baseline)dTsimane (N)USc
(N)
% difference (from US baseline)d
40–49192.89 (80)204.33 (492)−5.60
(p=0.004)
192.33 (77)194.32 (895)−1.02
(p=0.547)
192.62 (157)197.87 (1387)−2.65
(p=0.040)
50–59160.43 (89)177.79 (1164)−9.76
(p<0.001)
180.06 (98)173.13 (1708)4.00
(p=0.027)
170.72 (187)175.02 (2872)−2.46
(p=0.076)
60–69132.79 (49)153.90 (1204)−13.72
(p=0.001)
150.26 (56)158.46 (1422)−5.18
(p=0.069)
142.11 (105)156.37 (2626)−9.12
(p<0.001)
≥70b103.29 (27)127.14 (1048)−18.76
(p<0.001)
135.82 (31)139.27 (1097)−2.48
(p=0.543)
120.68 (58)133.34 (2145)−9.50
(p=0.005)
Totale163.85 (245)179.91 (3908)−8.93
(-----)
176.21 (262)176.64 (5122)−0.24
(-----)
170.37 (507)177.61 (9030)−4.08
(-----)
  1. aUS data represent asymptomatic subjects from greater Los Angeles (described in Budoff et al., 2010). Briefly, 9585 subjects (43% female; mean age = 56) underwent coronary artery calcification (CAC) scanning for evaluation of subclinical atherosclerosis, after exclusion of participants with vertebral deformities or fractures. Subjects had no known bone disease.

    bMaximum age for Tsimane women and men = 91 and 94 years, respectively (US maximum age = 90 for both sexes).

  2. cUS means are weighted by sample sizes (shown in Table 1 of Budoff et al., 2010). This table reports age sub-groups in two-year intervals (e.g. 41–42, 43–44, etc.); for two-year intervals spanning multiple decades that overlap with the age categories shown in this table (i.e. 39–40, 49–50, 59–60 and 69–70), we assume that each year of the two-year interval contributes 50% of the sample.

    dP-value from a one-sample t test, including as the test value the weighted mean from Budoff et al. (2010).

  3. eAge-standardized means are shown to account for differences in age distributions across populations. We use the Tsimane adult age distribution (calculated from the 2015 THLHP census) as the standard. To calculate age-standardized means, means for each age category and population (i.e. unadjusted means for Tsimane and weighted means for US) are multiplied by the proportional representation of that age category in the 2015 THLHP census, and then summed across all age categories.

Thoracic vertebral fracture prevalence is higher for Tsimane than Americans

Appendix 1—table 8
Age-specific thoracic vertebral (T6-T12) fracture prevalence (% with fracture grade ≥1) for Tsimane and Americansa by sexb.
https://doi.org/10.7554/eLife.48607.016
Age category (years)Women (n = 491)Men (n = 524)Total (n = 1,015)
Tsimane (95% CI)US
(95% CI)
Tsimane (95% CI)US
(95% CI)
Tsimane
(95% CI)
US
(95% CI)
40–4914 (6–21)4 (<1–8)34 (23–45)12 (4–19)24 (17–30)8 (3–12)
50–5925 (16–34)15 (7–23)37 (27–46)9 (4–15)31 (24–38)12 (7–16)
60–6920 (9–32)13 (4–22)36 (23–49)11 (2–20)29 (20–37)12 (6–18)
≥70c15 (<1–29)9 (<1–21)43 (25–62)16 (2–29)30 (18–42)13 (4–22)
Total (crude)19 (14–24)11 (7–14)36 (31–42)11 (7–15)28 (24–32)11 (8–14)
Total (age-standardizedd)18936112710
  1. aUS data are from two sources: 1) a subset of MESA study participants (described in Budoff et al., 2011), and 2) a subset of study participants in the greater Los Angeles area (described in Budoff et al., 2013). Briefly, the MESA cohort is a longitudinal, population-based study of 6814 adults (54% female; mean age = 62) free of clinical cardiovascular disease, representing six areas in the US: Baltimore, MD, Chicago, IL, Forsyth County, NC, Los Angeles, CA, New York, NY, and St. Paul, MN. Regarding participants from the greater Los Angeles area, data were collected among 4126 asymptomatic subjects (51% female; mean age = 64) who underwent CAC scanning for evaluation of subclinical atherosclerosis. MESA and Los Angeles-only subsets were first matched for age and sex with the Tsimane sample, and then merged to create a single US comparison sample.

    bFor women (n = 245 Tsimane and 246 US), sample sizes for ages 40–49 are 80 Tsimane and 77 US, for ages 50–59 89 Tsimane and 86 US, for ages 60–69 49 Tsimane and 60 US, and for ages 70+ 27 Tsimane and 23 US. For men (n = 261 Tsimane and 263 US), sample sizes for ages 40–49 are 77 Tsimane and 69 US, for ages 50–59 98 Tsimane and 108 US, for ages 60–69 56 Tsimane and 54 US, and for ages 70+ 30 Tsimane and 32 US.

  2. cMaximum age for Tsimane women and men = 91 and 94 years, respectively. Maximum age for US women and men = 92 and 91 years, respectively.

    dAge-standardized prevalences are shown to account for differences in age distributions across populations. We use the Tsimane adult age distribution (calculated from the 2015 THLHP census) as the standard. Prevalence for each age category and population is multiplied by the proportional representation of that age category in the 2015 THLHP census, and then summed across all age categories.

Appendix 1—table 9
Age-specific thoracic vertebral (T6-T12) fracture prevalence (% with fracture grade ≥2) for Tsimane and Americans by sex.
https://doi.org/10.7554/eLife.48607.017
Age category (years)Women (n = 491)Men (n = 524)Total (n = 1,015)
Tsimane (95% CI)US
(95% CI)
Tsimane (95% CI)US
(95% CI)
Tsimane
(95% CI)
US
(95% CI)
40–495 (<1–10)0 (--)5 (<1–10)1 (<1–4)5 (2–9)1 (<1–2)
50–598 (2–14)3 (<1–7)14 (7–21)2 (<1–4)11 (7–16)3 (<1–5)
60–694 (<1–10)3 (<1–8)9 (1–17)7 (<1–15)7 (2–12)5 (1–9)
≥7011 (<1–24)4 (<1–13)23 (7–39)3 (<1–9)18 (7–28)4 (<1–9)
Total (crude)7 (3–10)2 (<1–4)11 (8–15)3 (1–5)9 (7–12)3 (1–4)
Total (age-standardized)6210283
Appendix 1—table 10
Log-binomial generalized linear models: women’s adjusted relative risk of fracture (95% CI) by population and fracture grade controlling for age (n=491).
https://doi.org/10.7554/eLife.48607.018
ParameterModel one outcome: grade 0.5 fracture (borderline)Model two outcome: grade one fracture (mild)Model three outcome: grade two fracture (moderate)Model four outcome: grade three fracture (severe)
Tsimane
(vs. American)
1.14
(0.80–1.64)
1.56¶
(0.91–2.65)
3.01*
(1.11–8.17)
1.33
(0.09–18.92)
Age (per SD increase)1.04
(0.87–1.23)
1.05
(0.82–1.34)
1.02
(0.68–1.55)
6.26**
(1.46–26.79)
  1. **p≤0.01 *p≤0.05 ¶p≤0.1

Appendix 1—table 11
Log-binomial generalized linear models: men’s adjusted relative risk of thoracic vertebral (T6-T12) fracture (95% CI) by population and fracture grade controlling for age (n = 524).
https://doi.org/10.7554/eLife.48607.019
ParameterModel one outcome: grade 0.5 fracture (borderline)Model two outcome: grade one fracture (mild)Model three outcome: grade two fracture (moderate)Model four outcome: grade three
fracture (severe)
Tsimane
(vs. American)
2.27***
(1.58–3.27)
3.12***
(1.97–4.95)
4.70***
(1.98–11.17)
1.48
(0.21–10.63)
Age (per SD increase)0.88
(0.74–1.04)
0.99
(0.81–1.19)
1.16
(0.86–1.56)
4.79***
(1.94–11.81)
  1. ***p≤0.001 **p≤0.01 *p≤0.05 ¶p≤0.1

Lower thoracic vertebral BMD increases risk of thoracic vertebral fracture among Tsimane, particularly for women. Short IBI additionally increases Tsimane women’s fracture risk

Appendix 1—table 12
Sample characteristics for Tsimane women (n = 245) with and without any thoracic vertebral (T6-T12) fracture (grade ≥1).
https://doi.org/10.7554/eLife.48607.020
A) Fracture
(n=47a)
B) No fracture
(n=198a)
A vs. Bb
VariableMeanSEMeanSE
Thoracic vertebral BMD (mg/cm3)142.665.93163.133.20p=0.006
Age (years)56.671.4556.090.73p=0.490
Height (cm)149.790.63150.470.38p=0.631
Weight (kg)56.071.7154.400.68p=0.404
BMI (kg/m2)24.940.7223.990.27p=0.355
Body fat (%)26.831.3425.490.55p=0.344
Fat mass (kg)15.721.2314.370.48p=0.403
Fat-free mass (kg)40.140.9340.030.37p=0.930
Age at menarche (years)13.210.0913.320.04p=0.352
Age at menopause (years)49.760.6449.160.30p=0.408
Age at first birth (years)18.490.6718.910.28p=0.427
Parity (# births)9.720.548.890.22p=0.072
Mean inter-birth interval (months)30.692.6134.341.17p=0.010
Residential proximity to market (km)49.015.2850.632.71p=0.955
Spanish fluency (1 = fluent/moderate; 0 = none)0.240.070.370.04p=0.114
Schooling (# years)0.360.150.590.10p=0.318
  1. aRepresents maximum possible sample size (i.e. no missing data).

    bP-value from Mann-Whitney U or χ² test.

Appendix 1—table 13
Log-binomial generalized linear models: effects of thoracic vertebral BMD (models 1–2) and mean IBI (model 2) on the probability of thoracic vertebral (T6-T12) fracture (grade ≥1) for Tsimane women.

Relative risk (95% CI) is shown per SD increase.

https://doi.org/10.7554/eLife.48607.021
Model 1: controlling for age and anthropometricsModel 2: + mean IBI
ParameterExp(β)95% CIpExp(β)95% CIp
Thoracic vertebral BMD (mg/cm3)0.5420.352–0.8370.0060.5400.345–0.8450.007
Mean IBI (months)---------------0.3790.165–0.8660.021
Age (years)0.6760.449–1.0170.0610.6400.423–0.9680.034
Height (cm)0.8680.660–1.1410.3100.8810.662–1.1720.385
Fat mass (kg)1.2780.978–1.6700.0731.2911.006–1.6580.045
N219212
Appendix 1—table 14
Log-binomial generalized linear models: effects of thoracic vertebral BMD and mean IBI on the probability of thoracic vertebral (T6-T12) fracture for Tsimane women (n = 212) by fracture grade.

Grade 3 (severe) fracture risk is not modeled because no grade three cases are present for the sample of women with complete data. Relative risk (95% CI) is shown per SD increase.

https://doi.org/10.7554/eLife.48607.022
ParameterModel one outcome: grade 0.5 fracture (borderline)Model two outcome: grade one fracture (mild)Model three outcome: grade two fracture (moderate)
Thoracic vertebral BMD (mg/cm3)0.92
(0.62–1.36)
0.73
(0.43–1.24)
0.27**
(0.10–0.71)
Mean IBI (months)1.02
(0.81–1.29)
0.33*
(0.12–0.93)
0.49
(0.09–2.72)
Age (years)1.06
(0.76–1.49)
0.85
(0.51–1.41)
0.34**
(0.16–0.72)
Height (cm)1.01
(0.76–1.34)
1.00
(0.71–1.40)
0.63
(0.32–1.25)
Fat mass (kg)0.77¶
(0.57–1.05)
1.26¶
(0.97–1.63)
1.25
(0.64–2.45)
  1. **p≤0.01 *p≤0.05 ¶p≤0.1

Appendix 1—table 15
Sample characteristics for Tsimane men (n = 261) with and without any thoracic vertebral (T6-T12) fracture (grade ≥1).
https://doi.org/10.7554/eLife.48607.023
A) Fracture
(n=95a)
B) No fracture
(n=166a)
A vs. Bb
VariableMeanSEMeanSE
Thoracic vertebral BMD (mg/cm3)169.613.84173.642.80p=0.454
Age (years)57.041.0755.740.71p=0.392
Height (cm)160.600.53161.860.43p=0.034
Weight (kg)63.330.9461.960.57p=0.303
BMI (kg/m2)24.490.2923.630.19p=0.014
Body fat (%)19.050.7616.800.46p=0.026
Fat mass (kg)12.460.6310.560.35p=0.034
Fat-free mass (kg)51.220.6851.390.47p=0.936
Residential proximity to market (km)51.283.8056.603.07p=0.315
Spanish fluency (1 = fluent/moderate;0 = none)0.770.050.820.03p=0.404
Schooling (# years)2.020.381.700.24p=0.639
  1. aRepresents maximum possible sample size (i.e. no missing data).

    bP-value from Mann-Whitney U or χ² test.

Appendix 1—table 16
Log-binomial generalized linear model: effect of thoracic vertebral BMD on the probability of thoracic vertebral (T6-T12) fracture (grade ≥1) for Tsimane men (n=227a).

Relative risk (95% CI) is shown per SD increase.

https://doi.org/10.7554/eLife.48607.024
ParameterExp(β)95% CIp
Thoracic vertebral BMD (mg/cm3)1.0460.862–1.2690.650
Age (years)1.0100.822–1.2410.924
Height (cm)0.8220.701–0.9630.015
Fat mass (kg)1.3661.176–1.587<0.001
  1. aFracture data are missing for one man with BMD and anthropometric data. We also omitted a man whose height (138.2 cm) was 4.3 SDs below the mean.

Appendix 1—table 17
Log-binomial generalized linear models: effect of thoracic vertebral BMD on the probability of thoracic vertebral (T6-T12) fracture for Tsimane men by fracture grade.

Relative risk (95% CI) is shown per SD increase.

https://doi.org/10.7554/eLife.48607.025
ParameterModel one outcome: grade 0.5 fracture (borderline)Model two outcome: grade one fracture (mild)Model three outcome: grade two fracture (moderate)Model four outcome: grade three fracture (severe)
Thoracic vertebral BMD (mg/cm3)0.76*a
(0.60–0.96)
0.96
(0.73–1.26)
1.24a
(0.83–1.84)
0.49
(0.02–11.09)
Age (years)0.81¶a
(0.64–1.03)
0.82
(0.61–1.10)
1.40¶a
(0.93–2.12)
2.87
(0.24–34.91)
Height (cm)1.18¶b
(0.98–1.42)
0.75**
(0.60–0.93)
0.94c
(0.65–1.37)
0.70
(0.11–4.60)
Fat mass (kg)0.80*a
(0.65–0.995)
1.41***
(1.15–1.73)
1.26a
(0.90–1.78)
0.81
(0.09–7.46)
N227a227227a227
  1. ***p≤0.001 **p≤0.01 *p≤0.05 ¶p≤0.1

    aModel parameters include BMD, age and fat mass (not height). Joint inclusion of both height and fat mass yields invalid estimates.

  2. bModel parameters include BMD (adjusted RR = 0.75, 95% CI: 0.61–0.92, p=0.006), age (adjusted RR = 0.84, 95% CI: 0.66–1.07, p=0.159) and height (not fat mass, n = 253).

    cModel parameters include BMD (adjusted RR = 1.21, 95% CI: 0.81–1.82, p=0.358), age (adjusted RR = 1.40, 95% CI: 0.92–2.14, p=0.116) and height (not fat mass, n = 253).

Additional materials and methods

Study design and participants

Since 2002 the Tsimane (population ~16,000) have participated in the ongoing Tsimane Health and Life History Project (THLHP; see Gurven et al., 2017 for a project overview). All Tsimane residing in study villages are eligible to participate, and most choose to do so at least once. Project physicians have conducted annual medical exams on Tsimane of all ages since 2002. A mobile team of physicians, biochemists, and Tsimane research assistants collects data in Tsimane villages on medical and reproductive histories, functional ability, and other aspects of lifestyle (e.g. food production and sharing), in addition to collecting biological specimens (e.g. serum, urine, feces) among a subset.

Between July 2014 and September 2015, men and women aged 40+ years from 59 Tsimane villages were invited to participate in the CT scanning project. At the time, there were 1214 eligible people living in these villages (the only eligibility criteria were being 40+ years old, self-identifying as Tsimane and willing to participate). 731 individuals were present in their villages at the time and subsequently received a CT scan. Transporting participants from their village to the nearby market town of San Borja was logistically complicated (requiring trekking through the forest, dug-out canoes, rafts propelled by poles pushed off the river bottom, trucks, and cars) and can require up to two days of travel each way. From San Borja to the Beni department capital of Trinidad (where the hospital containing the CT scanner is located) is an additional 6 hr car ride. Due to these logistical complications, participants not in their village at the time we arrived were not sampled. The Tsimane are semi-mobile and often build secondary houses deep in the forest near their horticultural plots, not returning to their village for extended periods of time. Hunting and fishing trips can last days or weeks, and some men engage in wage labor in San Borja or elsewhere (e.g. rural cattle ranches). In an average village, approximately one-third of individuals are away hunting, fishing, working in their horticultural plots, or in San Borja at any given time. Additionally, a major flood in February 2014 resulted in mass migration from some villages, and the creation of several new villages that were not sampled as part of this study, further reducing the number individuals that could be sampled in this study. To address potential sources of sample bias, analyses comparing Tsimane who received CTs and those who did not but who participated in the THLHP’s medical exams in Tsimane villages were conducted. There were no significant differences in sex, blood pressure or body fat (see Kaplan et al., 2017 for additional participant details) and thus the CT sample analyzed here is thought to be representative of all Tsimane aged 40+ years.

Of the 731 Tsimane who received a CT scan, CT data from 507 (69%) were selected with no particular criteria to estimate thoracic vertebral bone mineral density (BMD) and fracture (scans from 224 Tsimane were not assessed due to radiologist time constraints). Among these 507 individuals, some data were missing because of either broken equipment, missing supplies, participant recall problems, absent or sick team personnel who were unable to collect data, or because all relevant thoracic vertebrae did not appear in the CT image field of view (see Appendix 1—table 18 for sample sizes and descriptives for all study variables).

Appendix 1—table 18
Descriptives for all study variablesa.
https://doi.org/10.7554/eLife.48607.026
VariableNMeanSDMinMax
Thoracic vertebral BMD (mg/cm3)507165.8541.2768.91314.99
Any thoracic vertebral (T6-T12) fracture (% grade ≥ 1)5060.280.450.001.00
Age (years)50756.259.9541.0094.00
Sex (% male)5070.520.500.001.00
Height (cm)493156.057.60136.00176.30
Weight (kg)49358.709.7334.6096.90
BMI (kg/m2)49324.053.3316.3338.11
Body fat (%)44821.578.005.0046.70
Fat mass (kg)44812.886.041.9542.12
Fat-free mass (kg)44845.807.8527.8173.08
Age at menarche (years)21413.300.5611.7516.29
Age at menopause (yearsb)17349.283.5637.0055.50
Age at first birth (years)23518.833.9112.0037.00
Parity (# live births)2459.053.270.0017.00
Mean inter-birth interval (months)23133.6816.2714.02142.37
Residential proximity to market (km)50752.6238.224.71153.61
Spanish fluency (1 = fluent/moderate;0 = none)4510.580.500.001.00
Schooling (# years)4501.192.470.0015.00
  1. aData were missing for various reasons (see supplementary text for details).

    b185 of the 245 participating women (76%) were post-menopausal. 12 women were post-menopausal but could not recall their age at menopause. The remaining 60 women (245-185) were either pre-menopausal (n = 36) or missing data regarding menopausal status (n = 24). Women were categorized as post-menopausal if they reported during THLHP medical exams not having experienced a menstrual cycle in the past year, and were neither pregnant nor lactating at the time of the study.

Thoracic computed tomography (CT)

Tsimane CT scans were supervised and reviewed by at least one of the HORUS study team cardiologists and radiologists. Breath-hold instructions were given in the Tsimane language to minimize respiratory motion artifact and misregistration (for additional methodological details see Kaplan et al., 2017).

CT scanning procedures used to collect BMD data from greater Los Angeles involved use of electron-beam CT scanners (C-300; GE-Imatron, South San Francisco, California) and a 64–detector row CT scanner (LightSpeed VCT; GE Medical Systems) at the Los Angeles Biomedical Research Institute. Parameters for electron-beam CT scanning were 130 kVp, 630 mA, and 2.5 or 3 mm slice thickness. The multidetector CT parameters were 120 kVp, 200–600 mA, and 2.5 mm slice thickness. 156 subjects underwent scanning with electron-beam CT and 64–detector row CT on the same day for comparison and normalization of BMDs between scanners. CT scanning to collect MESA sub-sample data, used here to assess fracture prevalence and severity, was performed at the Los Angeles Biomedical Research Institute; procedures involved the use of a C-150 GE-Imatron electron-beam scanner and were otherwise identical to those described in Budoff et al. (2010). Supplementary CT data for assessment of fracture prevalence and severity were also collected at the Los Angeles Biomedical Research Institute using similar parameters.

Thoracic vertebral bone mineral density (BMD)

CT-measured vertebral BMD is increasingly used for osteoporosis screening because of its ability to provide three-dimensional information compared to traditional dual x-ray absorptiometry two-dimensional images (Budoff et al., 2012). The left main coronary artery (LMCA) was set as the reference site to allow reproducible detection of a spinal level for use with cardiac CT scanning. The LMCA is covered in 100% of images and the field of view can be completely reconstructed; therefore, it is an optimal reference point to locate the starting measurement level. The most common location of the LMCA origin is at the lower edge of T7.

Intra-observer variability in BMD measurements using the same measurement technique in a different sample (Budoff et al., 2010) was tested on 120 scans by one observer from the Los Angeles Biomedical Research Institute, with 1 week intervals between the two readings. To measure inter-observer variability on 67 randomly selected scans, the results obtained by two radiologists from the Los Angeles Biomedical Research Institute who were blinded to all clinical information and prior measurements were compared using this other sample. Intra- and inter-observer variations in BMD measurements were 2.5% (Bland-Altman plot ratio: 1.00, 95% CI: 0.99–1.00) and 2.6% (Bland-Altman plot ratio: 1.00, 95% CI: 0.99–1.00), respectively.

Thoracic vertebral fracture

Genant’s semi-quantitative technique does not distinguish between wedge (i.e. reduced anterior height), biconcave (i.e. reduced central height) or crush (i.e. reduced posterior height) fractures; most fractures contain a combination of these features and are influenced by the local biomechanics of the spinal level involved. Vertebral fractures are differentiated from other, non-fracture deformities (e.g. osteoarthritis), although these other deformities were not systematically coded.

Socio-demographics and anthropometrics

Birth years were assigned based on a combination of methods described in detail elsewhere (Gurven et al., 2007), including using known ages from written records, relative age lists, dated events, photo comparisons of people with known ages, and cross-validation of information from independent interviews of kin. Each method provides an independent estimate of age, and when estimates yielded a date of birth within a three-year range, the average was generally used.

Outcomes of each pregnancy reported during reproductive histories were recorded as either ending in a live birth or terminating pre-term. Whether miscarriages (including stillbirths) are included or omitted from parity counts does not affect results, and results reported here reflect only live births.

Data analysis

With respect to general linear models of BMD, variance inflation factors were checked to assess degree of covariance and ensure minimal collinearity. Crude mean BMD values are used to compare populations for each decade, and age-standardized means are used to compare populations across all decades. We use the Tsimane adult age distribution, estimated from the THLHP 2015 census, as the standard.

Appendix 1—table 19
General linear models: anthropometric predictors of thoracic vertebral BMD (mg/cm3) for women.

Height is included in models 1 and 3, weight in models 2, 3 and 5, body mass index (BMI) in model 4, body fat (% or fat mass) in models 5–6, and fat-free mass in model 6. Age is controlled in all models. Standardized betas are shown (intercepts omitted).

https://doi.org/10.7554/eLife.48607.027
ParameterModel 1Model 2Model 3Model 4Model 5Model 6
Height (cma)0.040-----0.013c---------------
Weight (kga)-----0.080¶0.075c-----0.130¶c-----
BMI (kg/m2a)---------------0.070----------
Body fat (%b)--------------------−0.061c-----
Fat mass (kgb)-------------------------0.021c
Fat-free mass (kgb)-------------------------0.087c
Age (years)−0.679***−0.670***−0.667***−0.681***−0.661***−0.659***
Adjusted R20.4840.4890.4870.4880.4980.498
N238238238238219219
  1. ***p≤0.001 **p≤0.01 *p≤0.05 ¶p≤0.1

    aData are missing for seven women.

  2. bData are missing for 26 women.

    cVariance inflation factors do not indicate a high degree of multicollinearity.

Appendix 1—table 20
General linear models: anthropometric predictors of thoracic vertebral BMD (mg/cm3) for men.

Height is included in models 1 and 3, weight in models 2, 3 and 5, body mass index (BMI) in model 4, body fat (% or fat mass) in models 5–6, and fat-free mass in model 6. Age is controlled in all models. Standardized betas are shown (intercepts omitted).

https://doi.org/10.7554/eLife.48607.028
ParameterModel 1Model 2Model 3Model 4Model 5Model 6
Height (cma)0.028-----−0.062c---------------
Weight (kga)-----0.137**0.170**c-----0.177**c-----
BMI (kg/m2a)---------------0.135**----------
Body fat (%b)--------------------−0.072c-----
Fat mass (kgb)-------------------------0.022c
Fat-free mass (kgb)-------------------------0.163**c
Age (years)−0.578***−0.556***−0.561***−0.567***−0.508***−0.505***
Adjusted R20.3070.3250.3250.3240.3000.302
N255255255255229229
  1. ***p≤0.001 **p≤0.01 *p≤0.05 ¶p≤0.1

    aData are missing for seven men.

  2. bData are missing for 33 men.

    cVariance inflation factors do not indicate a high degree of multicollinearity.

https://doi.org/10.7554/eLife.48607.007

References

  1. 1
    Light and Broken Bones: Examining and Interpreting Bone Loss and Osteoporosis in Past Populations
    1. S Agarwal
    (2008)
    In: M Katzenberg, S Saunders, editors. Biological Anthropology of the Human Skeleton. Hoboken: Wiley. pp. 387–410.
    https://doi.org/10.1002/9780470245842.ch12
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
    Fertility in Traditional Societies
    1. KL Campbell
    2. JW Wood
    (1988)
    In: P Diggory, M Potts, S Teper, editors. Natural Human Fertility: Social and Biological Determinants. London: Palgrave Macmillan. pp. 39–69.
    https://doi.org/10.1007/978-1-349-09961-0_4
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
    Paleopathology at the Origins of Agriculture
    1. MN Cohen
    2. GJ Armelagos
    (1984)
    Gainesville: University Press of Florida.
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
    Estrogen, exercise, and the skeleton
    1. MJ Devlin
    (2011)
    Evolutionary Anthropology: Issues, News, and Reviews 20:54–61.
    https://doi.org/10.1002/evan.20299
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
    Lifestyle, diet, and disease: Comparative perspectives on the determinants of chronic health risks
    1. W Leonard
    (2008)
    In: S Stearns, J Koella, editors. Evolution in Health and Disease (Second Edtion). Oxford: Oxford University Press. pp. 265–276.
    https://doi.org/10.1093/acprof:oso/9780199207466.003.0020
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
    Bioarchaeological Reconstruction and Interpretation
    1. CB Ruff
    (1999)
    Skeletal Structure and Behavioral Patterns of Prehistoric Great Basin Populations Prehistoric Lifeways in the Great Basin Wetlands, Bioarchaeological Reconstruction and Interpretation, Salt Lake City, University of Utah Press.
  64. 64
  65. 65
  66. 66
  67. 67
  68. 68
  69. 69
  70. 70
  71. 71
    The Evolution of Life Histories
    1. SC Stearns
    (1992)
    Oxford: Oxford University Press.
  72. 72
  73. 73
  74. 74
  75. 75
  76. 76
  77. 77
  78. 78
  79. 79
  80. 80
  81. 81

Decision letter

  1. Diethard Tautz
    Senior and Reviewing Editor; Max-Planck Institute for Evolutionary Biology, Germany
  2. Jay Stock
    Reviewer; University of Cambridge, United Kingdom

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Computed tomography shows high fracture prevalence among physically active forager-horticulturalists with high fertility" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by Diethard Tautz as the Senior and Reviewing Editor. The following individual involved in review of your submission has agreed to reveal his identity: Jay Stock (Reviewer #1).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary and essential revisions:

Based on the reviewer comments, the discussion of the comments and the editors' assessment, we recommend to revise the manuscript into a version that does not try to test a specific hypothesis, but is more open to different interpretations. The main issue is that the current version favors a life history explanation (which is interesting and compelling), but has not demonstrated that this can be differentiated from other existing explanations. There is already evidence for life history parameters influencing bone mechanics independent of physical activity levels: Evol Med Public Health;2018(1):167-179.

The full comments of the reviewers are attached to this letter. Please respond to them in detail when submitting the revised version.

Reviewer #1:

This paper aims to test the hypothesis that modern human skeletal fragility results from greater reproductive effort. Bone mineral density was measured in thoracic vertebrae of Tsimane forager-horticulturalists and compared to American men and women from Los Angeles. No systematic differences were found between Tsimane and American men, but Tsimane women, who have higher rates of fertility, younger ages at first birth, and shorter interbirth intervals, were found to have lower thoracic bone mineral density than American women. Most of the analyses in the paper, however, focus on internal variation within the Tsimane sample, where the results demonstrate that among Tsimane women, earlier age at first birth and shorter interbirth intervals are associated with reduced BMD. Age at menarche and menopause are not associated with BMD, but lower BMD and shorter interbirth intervals increases fracture risk. The results of the analyses of the Tsimane men and women are very interesting and provide an important contribution to the literature, but the most original component of the paper is the interpretation of variation within the Tsimane women and between Tsimane and American women, within the framework of life history theory. In my opinion, Life History Theory is a powerful explanatory model that has application across a wide range of biological phenomenon. The key to evaluating this paper is whether LHT provides a testable alternate hypothesis relative other mechanisms. Does it explain the observed variation or are other mechanisms a parsimonious explanation?

Overall this paper presents an important dataset, and I find the analyses appropriate, however the interpretations need further delimitation with regards to other mechanisms.

With regards to the analyses of variation within the sample of Tsimane women, the authors interpret the results as being related to life history factors: higher fertility, earlier first reproduction, and shorter interbirth intervals. This is an innovative interpretation of intrapopulation variation in bone mass, but to make the argument that life history explains variation requires an assumption that bone has the capacity to recover from mineral losses associated with pregnancy and lactation. The authors cite a few papers in support of the 'full recovery' of BMD following lactation (Prentice, for example), but I am highly skeptical of this literature. In general, trabecular bone lost during adulthood, for any reason, is never fully regained. The studies used to justify this interpretation are dated and use very coarse methods for the characterization of bone quality. I am not convinced that this literature is sufficient to rule out the Tsimane bone loss as being the cumulative effect of multiple periods of gestation and lactation 'raiding' the skeleton for resources. If there was anything less than 100% recovery with each pregnancy, you would find losses in bone as a simple result of maternal provisioning and the direct 'cause' would not necessarily be interbirth intervals, but simply the cumulative bone loss with subsequent pregnancies due to the skeleton being used as a mechanism of resource storage. This is not necessarily incompatible with life history theory, but it is difficult to see how one would differentiate whether resource depletion during pregnancy would be differentiated from other life history parameters in the current dataset. The more pregnancies a woman has, the shorter the interbirth interval and the less bone might recover. In essence, this itself is a 'life history' trait, of maternal sacrifice of bone quality and reinvestment of these resources in offspring, but it doesn't fit the model and interpretation as the authors present it.

The comparison between the Tsimane and American populations is used to draw more general conclusions about variation in humans, and forms the basis of the most significant inferences made in the paper. Here I find the analyses somewhat problematic. The results of these comparisons demonstrate that Tsimane BMD is lower than the Americans, but only for women; and that Tsimane thoracic vertebral fracture prevalence is higher than Americans. These results are interpreted within the context of life-history theory, and it is argued that the lower BMD and higher facture prevalence among the Tsimane women is associated with greater reproductive effort.

The most significant issue is with the characterization of the differences in physical activity levels (PALs) between populations. A few authors have previously discussed broad changes in PAL relative to bone robusticity, but most authors do not use this sort of characterisation. The authors argue that the literature is largely based around the assumption that the transition to agriculture is characterized by a reduction in physical activity levels, and as they consider in the Discussion, this isn't really appropriate. They are correct to point this out in the Discussion, but in setting up the paper around this assumption they are both mischaracterizing most of the literature and setting up a straw man argument. What the literature does suggest is a reduction in terrestrial mobility (lower limb loading) associated with the origins of agriculture in most regions but this is also regionally variable. Change is upper body mechanical loading, however, are much more variable across major dietary transitions, so labour/activity intensity may significantly increase with subsistence transitions in some areas. Virtually no research in the prehistoric biomechanics literature tries to pool upper and lower body (systemic) variation in bone mechanical properties and interpret it in the context of overall PALs. Mischaracterizing the literature this way provides an easy target to knock down. One thing that virtually all biomechanics literature agrees upon is that the mechanical response to loading is variable throughout the body relative to local factors of weight bearing activity and muscle function. Within the context of the current study, it is not clear whether thoracic vertebrae would be expected to track either the mechanics of the lower limb (mobility) or the upper limb (manual labour), or some other more general factors. Many might expect it would track different influences on bone quality such as diet. As the study is pitched, the authors characterize that thoracic BMD should track PALS, and they characterize the literature on terrestrial mobility as being an indicator of PALs. This neither accurately characterizes current knowledge, nor reflects the subtlety of what we do know about bone variation in the past.

A third issue is that of the characterization of the Los Angeles comparative group. The variance in biomechanical properties is usually much higher in contemporary populations on account of the internal genetic, dietary and behavioural diversity. This makes characterization of activity extremely difficult. The authors characterize the 'physical inactivity' hypothesis as predicting that modern women from Los Angeles would have lower BMD, but I'm not sure this is necessarily true. Throughout the body of literature on bone biomechanics within recent human populations, declines in bone strength are not always consistent between hunter-gatherers and agricultural populations, and is highly variable throughout the Holocene. Most notably, contemporary human populations have highly variable skeletal strength properties that are dependent upon individual variation in genetics, diet and activity. The range of variation among contemporary populations is typically so high that one rarely finds statistical differences between modern humans and prehistoric populations that are less variable, as the range of variation in the more homogenous populations is found within contemporary society. Imagine, if you will, trying to characterise the 'behaviour' of a typical person in Los Angeles. Could you do it? For any person chosen at random, there is a chance that they may be virtually sedentary, with extremely low PALs, another may be an ultra-marathon runner with extensive athletic training and competition. As a result, it is generally accepted among bone biomechanics researchers that making general assumptions about activity in contemporary populations is challenging at best. This issue could be addressed with any individual activity data for modern subjects, either by recall or activity trackers. Does any such data exist?

The authors note that Tsimane women have moderate to active PAL levels, while men are highly active. Depending on the model of mechanically induced bone remodeling that one accepts, moderate levels of activity may not be sufficient to engender functional adaptive response in the mechanical properties of bone. When compared to the American sample, the specific PAL and reproductive history of the women is unknown. While it is probably safe to assume that the women did have lower fertility and longer inter-birth intervals, it may not be as safe to assume differences in PALs. The Tsimane women have moderate PALs, the American women may have had highly variable PALs with the same population, and activity levels that may have also been 'moderate'. Very little information is provided here about the American data, either in terms of population characteristics and context, or methods, so the reader cannot evaluate whether the characterization of the American group is appropriate

In terms of methods, the information on how the American data was collected is insufficient as presented. The only details given are – "a matched American sample with directly comparable CT derived indicators of bone strength measured by the same laboratory". Specific details of the equipment and settings are provided for the Tsimane data, were the Los Angeles data collected on the same equipment? With the same settings? Were any objects/participants scanned in each location to validate pooling data from different equipment? This needs to be expanded upon, if only briefly in the text or in supplementary information, so that the reader can evaluate for themselves comparability of approaches.

Overall, while my review has been critical of several aspects of the study and interpretations, I would summarize the work more positively. The Tsimane research has provided an extremely important dataset with very interesting social and biological context. The analyses are appropriate and reveal some very interesting trends in bone mass. My concerns with the paper are primarily about interpretation and assumptions relating to activity levels in the American dataset. Both of these concerns mean it is difficult to evaluate the 'life history' hypothesis relative to other proximate mechanisms, or even more broad variation in genetics, habitual activity (and what influence it may have on thoracic vertebrae), early life nutrition, growth plasticity, prevalence of bottle feeding, or extended lactation that may be associated with bmd losses. It is not clear how life history variation necessarily provides a stronger explanation for the data than would any one of these other factors, so alternate explanations should be considered and set up, where possible, as alternate explanations of the data with different predictions of the underlying variation. I believe this dataset and analysis is important and publishable, but it requires more nuanced hypothesis testing to differentiate life history mechanisms from competing hypotheses.

Reviewer #2:

This study investigates a Tsimane population sample and compares it to that in an American cohort with specific questions addressed relating to BMD and fracture risk in thoracic vertebrae of living subjects. Overall, I think this is a robust and timely study that provides alternative explanations for the observed low BMD (or gracilization) in some populations (present and past) beyond the effect of activity levels. It is a well-designed study; the sample sizes are satisfactory, and while there are numerous variables to control for, the authors undertook sufficient analyses. The manuscript is well written and overall was a pleasure to read. The implications of the study are significant in our attempts to understand the observed gracilization in modern humans. One of the highlights of the study is the finding that the Tsimane compared to an American population (sample) show lower BMD especially in women. I also appreciate that the authors outlined the study's limitations that are inherent in this type of study of living subjects. Even so, I have a few comments for the authors that I hope will help improve their manuscript.

1) There are a number of predictions made at the beginning of the manuscript (Introduction) and consequently there are many results to report and discussion to follow. I recommend that the Discussion section follow the order in which the predictions were made so that it is easier for the reader to follow and tie in the predictions and findings.

2) Additionally, in the Introduction there is extensive discussion about gracilization especially in the fossil/archaeological record, however in the Discussion section the subject is a little glossed over. I would suggest that the implications and/or the complexity of interpreting gracilization from a broader perspective be discussed. I realize that the authors do this e.g., in the Discussion, but a little more could be added as a way to explain/justify the value of this study which goes beyond investigating physical activity levels unlike some of the previous work in the fossil and archaeological record.

3) While the authors describe how BMD was measured using the computed tomography, I would suggest that a brief description of whether BMD estimates includes both cortical and trabecular bone and if not, how were the two demarcated? The information regarding how the scanner estimates BMD is found in the supporting information section, if there is room, I would include this in the main Materials and methods section. Also, BMD in three thoracic vertebrae was measured, I suppose the BMD reading is the mean of the three and if so, is this a reading provided automated from the scanner or is this done by the researchers manually after the scans are completed?

4) I found Figure 4 to be a bit confusing to follow and that is because there are multiple variables included in the analysis. Would it be possible to have the y-axis reading be "relative risk of fracture" without the per SD increase? This way the data are more easily interpreted?

Reviewer #3:

In this article, Stieglitz et al. present data from the Tsimané of Bolivia demonstrating that in thoracic vertebrae of adults >40 years old, early age at first reproduction and low interbirth interval are associated with lower BMD, and that in Tsimané women but not men, BMD is lower and fracture incidence is higher compared to a control population from the US. The authors argue that this pattern is consistent with tradeoffs in energy allocation between reproduction and somatic maintenance. Overall this is a well written paper and the data will be of significant interest to the field. I have several major comments for the authors to consider.

1) In the Introduction (first paragraph) and Discussion (fourth paragraph), more specificity is needed in the discussions of bone strength. "Bone strength" can refer to 1) midshaft robusticity or cross-sectional geometry (the shape and area of cortical bone in a given plane) vs. 2) bone mineral density (grams of mineral in a given area or volume of bone, which can be measured in cortical or trabecular bone). They are both generally correlated with strength but can't be directly compared to each other. This is important because the background largely focuses on studies of midshaft cortical bone, but the vertebra, which was measured in this study, is primarily trabecular bone surrounded by a thin cortical shell. Note also that the Chirchir references use the word 'density' to refer to trabecular bone volume fraction (BV/TV (%), or the amount of bone in a region of interest) while most publications use density to refer to the amount of bone mineral per area or volume of bone.

2) Introduction, second paragraph, Discussion first and third paragraphs: Osteogenic responses to mechanical loading occur predominantly in the ~2 years before and after puberty, and are markedly attenuated after menarche (although mechanical loading is essential for maintaining existing bone). See review, MacKelvie et al., 2002, http://dx.doi.org/10.1136/bjsm.36.4.250 Also, while cortical bone is highly responsive to mechanical loading around puberty, much less is known about trabecular bone response to mechanical loading. After menarche, physical activity does not provoke a substantial osteogenic response, and consistent with the life history framework presented in this article, energy expended on physical activity might arguably inhibit skeletal recovery (as in the Female Athlete Triad).

3) Previous studies in industrialized populations have looked for relationships between BMD and age at first birth, parity, etc. The authors may wish to cite/discuss Parazzini et al. 1996 J Epidemiol Community Health; We et al. 2018 doi: 10.3346/jkms.2018.33.e311; Chantry et al., 2004; Bexerra et al. 2002 DOI: 10.1093/jn/132.8.2183; etc.

4) The idea that these data reflect tradeoffs in energy allocation to maintenance vs. reproduction is intriguing and makes intuitive sense. It raises a crucial question, however: is energy intake limited in the Tsimané? If not, then why must there be such tradeoffs? Given all the data available to the authors, it would be helpful to identify specific biological mechanisms that are mediating the effects of age at first birth and IBI on thoracic vertebral BMD and fracture incidence in Tsimané vs. American women (or at least frame hypotheses about what they are). For example, what are estradiol levels in Tsimane women, and how do they compare to those of the American women described in this study?

https://doi.org/10.7554/eLife.48607.033

Author response

Summary and essential revisions:

Based on the reviewer comments, the discussion of the comments and the editors' assessment, we recommend to revise the manuscript into a version that does not try to test a specific hypothesis, but is more open to different interpretations. The main issue is that the current version favors a life history explanation (which is interesting and compelling), but has not demonstrated that this can be differentiated from other existing explanations. There is already evidence for life history parameters influencing bone mechanics independent of physical activity levels: Evol Med Public Health;2018(1):167-179.

In the revised Introduction and Discussion, we state that the proposed life history interpretation focusing on reproductive effort is not an alternative to and may complement other hypotheses of bone structural variation, including those derived from life history theory (e.g. proposed in Macintosh et al., 2018, which we now cite). We also point out that the proposed life history interpretation

focusing on reproductive effort is unique relative to other explanations (ultimate and proximate) because it focuses on sex differences, both within and between populations, given women’s greater energetic costs of reproduction. Other explanations (e.g. focusing on nutrition, genetics, inflammation, hormones, habitual physical activity level) cannot easily explain the observed population-level sex differences (i.e. Tsimane BMD is lower versus Americans, but only for women, whereas minimal BMD differences exist between Tsimane and American men despite large differences in their physical activity levels). According to these other explanations, population-level differences should be directionally consistent and of similar magnitude for both sexes.

In addition to revising the Introduction and Discussion, we also acknowledge in both sections other proximate explanations which may complement the proposed life history interpretation (i.e. “maternal depletion syndrome”, and factors such as calcium deficiency and/or chronic immune activation which may interact with each other and with reproductive effort).

The full comments of the reviewers are attached to this letter. Please respond to them in detail when submitting the revised version.

Reviewer #1:

[…] The key to evaluating this paper is whether LHT provides a testable alternate hypothesis relative other mechanisms. Does it explain the observed variation or are other mechanisms a parsimonious explanation? Overall this paper presents an important dataset, and I find the analyses appropriate, however the interpretations need further delimitation with regards to other mechanisms.

Thank you, and we agree that further delimitation is needed (see above comments to the editor and more below).

With regards to the analyses of variation within the sample of Tsimane women, the authors interpret the results as being related to life history factors: higher fertility, earlier first reproduction, and shorter interbirth intervals. This is an innovative interpretation of intrapopulation variation in bone mass, but to make the argument that life history explains variation requires an assumption that bone has the capacity to recover from mineral losses associated with pregnancy and lactation. The authors cite a few papers in support of the 'full recovery' of BMD following lactation (Prentice, for example), but I am highly skeptical of this literature. In general, trabecular bone lost during adulthood, for any reason, is never fully regained. The studies used to justify this interpretation are dated and use very coarse methods for the characterization of bone quality. I am not convinced that this literature is sufficient to rule out the Tsimane bone loss as being the cumulative effect of multiple periods of gestation and lactation 'raiding' the skeleton for resources. If there was anything less than 100% recovery with each pregnancy, you would find losses in bone as a simple result of maternal provisioning and the direct 'cause' would not necessarily be interbirth intervals, but simply the cumulative bone loss with subsequent pregnancies due to the skeleton being used as a mechanism of resource storage. This is not necessarily incompatible with life history theory, but it is difficult to see how one would differentiate whether resource depletion during pregnancy would be differentiated from other life history parameters in the current dataset. The more pregnancies a woman has, the shorter the interbirth interval and the less bone might recover. In essence, this itself is a 'life history' trait, of maternal sacrifice of bone quality and reinvestment of these resources in offspring, but it doesn't fit the model and interpretation as the authors present it.

In the comment above reviewer 1 makes three important and related points: 1) that a life history theory (LHT) interpretation requires an assumption, which may not hold, that bone can fully recover from mineral losses associated with pregnancy and lactation; 2) that in the Introduction we cite dated studies using crude methods when referring to maternal skeletal changes associated with pregnancy and lactation to justify a LHT interpretation; and 3) that another mechanism, namely the cumulative effect of multiple periods of gestation and lactation “raiding the skeleton for resources”, can explain Tsimane bone loss without requiring the assumption of full bone mineral recovery following pregnancy and lactation, and that this other mechanism may be consistent with predictions from LHT. In what follows we deal with each of these three points in turn.

Regarding the first point, we agree with reviewer 1 that it is important to clarify whether bone can fully recover from mineral losses associated with pregnancy and lactation. However, we respectfully disagree with the argument that a LHT interpretation requires that bone can fully recover in this manner. LHT posits that energetic resources invested in reproduction should trade-off against resources invested in somatic maintenance under a wide range of conditions. This broad logic – which was not developed to explain variation in bone strength per se – motivates the hypothesis that greater reproductive effort compromises bone strength (where adult bone strength represents an indicator of energetic investment in somatic maintenance). LHT makes no specific predictions about whether bone fully or partially recovers from mineral losses associated with pregnancy and lactation. This is now clarified in the paper (Introduction), and we delete any mention of “full recovery” in this context (there was one such instance in the original manuscript). Reviewer 1 rightly points out that this greater clarity is needed to inform one’s interpretation of our study’s findings, even though this question of whether bone fully or partially recovers from mineral losses associated with pregnancy and lactation is not the empirical focus of our study (most study participants are post-menopausal).

We also appreciate reviewer 1’s skepticism of the relevant bone metabolism literature that we originally cited, and we share this skepticism, in part for reasons reviewer 1 rightly mentions (e.g. loss during adulthood of trabecular structural connectivity is generally irreversible, as lamellar new bone can be added only on existing surfaces; the trabecular network can therefore not be completely restored, even if bone mass increases). This is why we originally cited the Jarjou et al., 2010 study, which found evidence among Gambian women of incomplete restoration of lumbar spine BMD to pre-pregnancy values by 12 months post-partum. In fact, there is a massive literature on maternal skeletal changes during pregnancy, lactation and post-weaning recovery—spanning studies of diverse species (e.g. humans, rats, mice, monkeys), diverse physiological processes (e.g. changes in mineral ions, calciotropic and phosphotropic hormones, sex steroids and other hormones, upregulation of intestinal absorption of calcium and phosphorus, changes in renal mineral handling) and spanning diverse measures of bone quantity and quality (e.g. from DXA, micro-CT, HR-pQCT, microradiography, ultrasound, histology, histomorphometry). A recent comprehensive review of this literature by Kovacs (https://www.ncbi.nlm.nih.gov/pubmed/26887676) includes >1,000 references. The five bullet points below effectively summarize from this review the current understanding of post-weaning skeletal recovery in humans. We apologize if this bullet point summary is verbose, but it directly addresses reviewer 1’s skepticism, and thus merits inclusion in our response. Kovacs writes:

- “The available DXA data suggest that lactational loss of bone density is completely reversed by 12 mo after weaning in most women…whereas recovery has been incomplete at some skeletal sites when assessed at 6 mo or less after weaning…A small longitudinal study of 22 women found a marked loss of 8% in the distal radius that was still reduced 5% by 12 mo after weaning, which suggests that the radius is slower to recover BMD. Note that these studies report the mean responses of a cohort; the response of individual women will vary, such that some women may not return to baseline after lactation, while others may end up with a BMD even higher than prior to pregnancy. Recovery is fast enough that women who breastfeed for 6 mo or more and have a second pregnancy within 18 mo do not have reduced BMD of the spine or hip at the end of the second pregnancy”.

- “HR-pQCT of the radius and ultradistal femur reveal recovery of trabecular microarchitecture and cortical parameters in women who lactate for shorter intervals, but incomplete recovery in women who lactate for longer. However, follow-up in these studies was for less than 6 mo after lactation ended for women who lactated for a longer duration, which is not sufficient time for recovery or to determine if permanent changes in structure resulted. Similar to studies in animals, some clinical studies have found that the cross-sectional diameter of the femur increased after lactation or post-weaning recovery, and that cortical bone area is restored or increased. An increase in bone volumes will improve bone strength and may compensate for any permanent loss of trabecular microarchitecture”.

- “Hip structural analysis of data from DXA scans provides direct measurements of femoral geometry and derived measures of femoral strength. A longitudinal study obtained these measurements at 2 wk, 6 mo, and 12 mo postpartum in lactating women, while similar timing of measurements were carried out in nonpregnant, nonlactating women. In lactating women there was an ~3% decrease in cross-sectional area, a 1.7% decline in cortical thickness, a 2.1% decrease in section modulus (bending strength), and a 2.3% increase in buckling ratio (instability) by 6 mo of lactation. These changes reversed by ~6 mo after lactation ceased, during which no changes were seen in the controls”.

- “If a permanent reduction in BMD or skeletal strength were to occur from lactation, then lactation should be a strong risk factor for low BMD or fracture in women of all ages. However, over five dozen epidemiologic studies of pre-, peri-, and postmenopausal women have found a neutral effect or a protective effect of lactation on peak bone mass, BMD, and fracture risk…There are, of course, a few contrary studies that suggest lactation predicts lower BMD”.

- “Extensive animal data and human data indicate that bone turnover is also increased during the post-weaning phase, but uncoupled in the reverse direction from lactation to favor bone formation. The maternal skeleton undergoes substantial improvements in bone mass, mineralization, and strength. Rodents appear to fully regain mineralization and strength within 2–8 wk, whereas women likely achieve this closer to 1 yr after weaning. In both the animal and human studies, there may be incomplete recovery of the trabecular microarchitecture of the femora and tibiae, the effects of which may be offset by an increase in the cross-sectional diameters of these bones. Although no formal measurements of bone strength after weaning have been done in women, the fact that more than five dozen epidemiological studies found that lactation confers a neutral or protective effect against low BMD and fragility suggests that recovery is effectively complete, without long-term adverse effects on bone strength”.

In the revised Introduction we expand our treatment of this topic and summarize several of these empirical findings (we now refrain from stating that post-weaning restoration of bone strength is necessarily fully complete, or even net positive), and we summarize findings from more recent studies that are not in the Kovacs 2016 review, including Bjornerem et al., 2017 (see Introduction).

Regarding the second point, that we cite dated studies using crude methods when referring to maternal skeletal changes associated with pregnancy and lactation, we have revised the relevant sections of the Introduction (third paragraph) to include more recent citations and studies using higher resolution methods (e.g. HR-pQCT of the appendicular skeleton).

Regarding the third point, that the cumulative effect of multiple periods of gestation and lactation “raiding the skeleton for resources” can explain Tsimane bone loss (without requiring the assumption of full bone mineral recovery following pregnancy and lactation), this logic is consistent with a LHT interpretation and the expected trade-off between energetic investment in reproduction and somatic maintenance. There is no reason to consider this an alternative mechanism that either falls outside the scope of a LHT interpretation or requires a restating of a LHT interpretation. For this reason we discussed in the original Introduction the “maternal depletion syndrome” literature, which focuses precisely on the cumulative long-term effects on maternal health (including bone health) of repeated pregnancies. This literature and its associated logic complements a LHT interpretation, including the shared prediction of an inverse association between lifetime reproductive effort and bone strength (however operationalized), rather than providing an alternative interpretation with different predictions. Prior maternal depletion researchers (e.g. see Figure 1 in Winkvist et al. 1992:692; https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1694126/) have considered different patterns of energy balance (i.e. from positive to negative) over time for a given reproductive cycle, but these researchers have not made specific predictions or assumptions about whether bone tissue in particular fully recovers from mineral losses following specific bouts of pregnancy and lactation. While “maternal depletion syndrome” is characterized by net negative energy balance over time for a given reproductive cycle (line D in Figure 1 in Winkvist et al.), predictions regarding whether bone fully or partially recovers do not directly follow. Interestingly, as we and others (e.g. Kovacs 2016) have noted, many studies show null associations between parity and BMD or fracture risk, and other studies show positive (or negative) associations between parity and BMD. Therefore, in the current literature, evidence of a cumulative negative effect of multiple periods of gestation and lactation on maternal bone strength – ignoring reproductive timing (e.g. age at first birth, IBIs) – is not particularly strong. Evidence of parity-specific effects on bone strength indicators are even mixed within the Tsimane population; in a prior study of Tsimane women (pre- and post-menopausal) we found an inverse association between parity and calcaneal qUS measures (Stieglitz et al. 2015), whereas our present study finds no parity effect on CT-derived thoracic vertebral BMD or fracture risk.

Lastly, reviewer 1 is correct in pointing out the difficulties of differentiating a “pure parity effect” vs. a “pure IBI effect” on bone strength, as higher parity women also have shorter IBIs on average. A strength of our Tsimane dataset is that we can leverage observed variation to explore bone strength indicators for high parity women with shorter vs. longer IBIs; for this reason we included Supplementary figure 3 from the original manuscript, which remains in the main text of the revised manuscript as Figure 1. These results suggest that the association between parity and BMD is complex and merits simultaneous consideration of reproductive timing and relevant physiological mechanisms (the latter fall outside the scope of our paper).

The comparison between the Tsimane and American populations is used to draw more general conclusions about variation in humans, and forms the basis of the most significant inferences made in the paper. Here I find the analyses somewhat problematic. The results of these comparisons demonstrate that Tsimane BMD is lower than the Americans, but only for women; and that Tsimane thoracic vertebral fracture prevalence is higher than Americans. These results are interpreted within the context of life-history theory, and it is argued that the lower BMD and higher facture prevalence among the Tsimane women is associated with greater reproductive effort.

This characterization is accurate. We try to avoid excessive speculation when making general conclusions about human variation based on our study’s results.

The most significant issue is with the characterization of the differences in physical activity levels (PALs) between populations. A few authors have previously discussed broad changes in PAL relative to bone robusticity, but most authors do not use this sort of characterisation. The authors argue that the literature is largely based around the assumption that the transition to agriculture is characterized by a reduction in physical activity levels, and as they consider in the Discussion, this isn't really appropriate. They are correct to point this out in the Discussion, but in setting up the paper around this assumption they are both mischaracterizing most of the literature and setting up a straw man argument. What the literature does suggest is a reduction in terrestrial mobility (lower limb loading) associated with the origins of agriculture in most regions but this is also regionally variable. Change is upper body mechanical loading, however, are much more variable across major dietary transitions, so labour/activity intensity may significantly increase with subsistence transitions in some areas. Virtually no research in the prehistoric biomechanics literature tries to pool upper and lower body (systemic) variation in bone mechanical properties and interpret it in the context of overall PALs. Mischaracterizing the literature this way provides an easy target to knock down. One thing that virtually all biomechanics literature agrees upon is that the mechanical response to loading is variable throughout the body relative to local factors of weight bearing activity and muscle function. Within the context of the current study, it is not clear whether thoracic vertebrae would be expected to track either the mechanics of the lower limb (mobility) or the upper limb (manual labour), or some other more general factors. Many might expect it would track different influences on bone quality such as diet. As the study is pitched, the authors characterize that thoracic BMD should track PALS, and they characterize the literature on terrestrial mobility as being an indicator of PALs. This neither accurately characterizes current knowledge, nor reflects the subtlety of what we do know about bone variation in the past.

We thank reviewer 1 for making two important points. The first point is that a more nuanced perspective is needed when characterizing the prehistoric biomechanics literature, particularly in the Introduction, that considers differential physical activity-induced mechanical loading regimes for lower and upper limbs associated with subsistence transitions. Furthermore, while both lower and upper limb loadings are expected to vary geographically during subsistence transitions, upper limb loading variability is much greater than that for the lower limb; in some areas, upper body activity intensity and loading may actually increase with subsistence transitions, and so we agree that it is misleading and inaccurate to interpret both upper and lower body (i.e. systemic) variation in bone mechanical properties in the context of overall PALs. This more nuanced perspective suggested by reviewer 1 is important because it identifies a flawed assumption – that transition to agriculture caused reductions in PALs – underlying a broader straw man argument.

We regret that in setting up the paper we mischaracterized the prehistoric biomechanics literature on changes in PALs relative to bone strength during subsistence transitions.

Our intention was not to set up a straw man argument, and in fact, a previous version of this manuscript (that wasn’t submitted to eLife) included a longer, more detailed description of the relevant prehistoric literature that addressed some of reviewer 1’s comments. Ultimately, because our study does not directly test any hypotheses of human morphological variation in prehistoric populations (noted in the Discussion), we elected to limit mention of the relevant prehistoric literature on subsistence transitions in setting up the empirical focus of our paper. In so doing we gave a mistaken impression, which we now correct. The revised Introduction no longer references changes in PALs relative to bone strength during prehistoric subsistence transitions, nor does it invoke the misguided assumption (or otherwise imply) that prehistoric agricultural transition entailed global PAL reductions and systemic (i.e. lower and upper limb) reductions in bone strength. We also now state in the Introduction that mechanical response of bone to loading is variable throughout the body relative to local factors of weight-bearing activity and muscle function (Introduction, first paragraph). In addition, we now state in the Discussion that during subsistence transitions, changes in upper body activities were likely much more variable than lower body activities, so upper limb mechanical loading may have actually increased with agriculture in some regions (Discussion, last paragraph).

The second point is that it is unclear whether thoracic vertebrae track lower and/or upper limb mechanics, or other factors. Providing this clarity is important for interpreting our study’s results in light of hypotheses of bone structural variation, and for making more general inferences about human bone structural variation. We now clarify in a new paragraph in the Introduction (seventh paragraph) that thoracic vertebrae track both lower and upper limb loadings. We provide information on thoracolumbar loading in the context of activities of daily living, vertebral architecture to accommodate these loads, and changes in architecture with age and sex. Inclusion of this new paragraph provides some justification for our reasoning, which was implied in the original manuscript but never explicitly stated, that thoracic BMD should track both lower and upper limb mechanics.

A third issue is that of the characterization of the Los Angeles comparative group. The variance in biomechanical properties is usually much higher in contemporary populations on account of the internal genetic, dietary and behavioural diversity. This makes characterization of activity extremely difficult. The authors characterize the 'physical inactivity' hypothesis as predicting that modern women from Los Angeles would have lower BMD, but I'm not sure this is necessarily true. Throughout the body of literature on bone biomechanics within recent human populations, declines in bone strength are not always consistent between hunter-gatherers and agricultural populations, and is highly variable throughout the Holocene. Most notably, contemporary human populations have highly variable skeletal strength properties that are dependent upon individual variation in genetics, diet and activity. The range of variation among contemporary populations is typically so high that one rarely finds statistical differences between modern humans and prehistoric populations that are less variable, as the range of variation in the more homogenous populations is found within contemporary society. Imagine, if you will, trying to characterise the 'behaviour' of a typical person in Los Angeles. Could you do it? For any person chosen at random, there is a chance that they may be virtually sedentary, with extremely low PALs, another may be an ultra-marathon runner with extensive athletic training and competition. As a result, it is generally accepted among bone biomechanics researchers that making general assumptions about activity in contemporary populations is challenging at best. This issue could be addressed with any individual activity data for modern subjects, either by recall or activity trackers. Does any such data exist?

Unfortunately we lack individual-level activity data for both the Los Angeles comparative sample and Tsimane study participants (but accelerometry and recall activity data are currently being collected among Tsimane study participants). We now acknowledge this lack of activity data as a study limitation (subsection “Inferring behaviors underlying morphological variation in past human populations”). Reviewer 1 makes several valid points with respect to high documented variability in activity levels and bone strength properties in contemporary populations; due to this high variability, comparisons between contemporary and more homogenous prehistoric populations (e.g. in skeletal strength parameters) often do not reveal statistical differences. We do find statistical differences in BMD across populations, but reviewer 1 rightly points out that it is difficult to characterize activity patterns at the population level. Indeed, Tsimane physical activity levels are not outside the range of some industrialized populations (https://www.ncbi.nlm.nih.gov/pubmed/23383262). Reviewer 1 thus wonders whether our characterization of the physical inactivity hypothesis of skeletal fragility as predicting that women from Los Angeles have lower BMD than Tsimane is accurate.

There is ample evidence that Tsimane are, on average, more active than Los Angeles residents. A telephone survey of 8,354 randomly selected adults in Los Angeles County (http://publichealth.lacounty.gov/ha/reports/habriefs/v3i2_phys/physact.pdf) found that 61% of adults did not get enough weekly physical activity to meet recommended guidelines of the Centers for Disease Control and Prevention, the American College of Sports Medicine (ACSM) and the National Institutes of Health. ACSM guidelines recommend that all adults perform ≥30 minutes of moderate intensity physical activity (defined as the effort expended by an average adult in walking 1.5-2 miles in half an hour) on most and preferably all days – either in a single session or accumulated throughout the day in multiple bouts. Each activity should last at least 8-10 minutes. 41% of LA respondents were classified as “sedentary” and 20% engaged in only irregular activity. Respondents classified as “irregular exercisers” reported vigorous physical activity on 1-3 days and walking for <30 minutes on a typical day or strength training. While the irregular exercise group had some form of physical activity, it was not deemed sufficient to meet current recommendations.

In contrast, based on a large sample of systematic direct behavioral observations (see Gurven et al., 2013), Tsimane activity profiles far exceed ACSM recommendations. Men and women spend about five and two hrs/day, respectively, in food production (i.e. hunting, fishing, foraging and farming). Women and men spend an additional 4–6 hrs and 1.5 hrs/day, respectively, in domestic tasks (e.g. collecting firewood and water, food processing, childcare). By age 15, both sexes are already close to adult work levels in terms of time allocation. About 4–6 hrs/day for women and 6–7 hrs/day for men are spent in lifestyle-moderate activity and relatively little time is spent sedentary. We can also directly compare activity levels – indicated by steps per day from waist-worn accelerometers – between the US NHANES sample (n=1,512 men and 1,545 women) and a different Tsimane sample (i.e. not receiving CT scans, and much larger [n=231 men and 232 women] than the accelerometry sample from the Gurven et al., 2013 paper). There are large population-level differences in the expected direction that are maintained throughout adulthood (see Author response image 1). Unfortunately no comparable data exist yet for children, but we are currently collecting accelerometry data among a large sample of Tsimane children.

For these reasons, even though we cannot make direct activity comparisons between the two samples in our current study, we think it is entirely consistent with available empirical data to reason that Tsimane are, on average (variability notwithstanding), more physically active than Los Angeles residents. We now specify throughout the revised manuscript that our characterization of activity levels across the two populations refers to mean differences. We also state that the prediction that Los Angeles residents have lower BMD than Tsimane is a simple (i.e. not nuanced) prediction from a physical inactivity hypothesis of skeletal fragility.

The authors note that Tsimane women have moderate to active PAL levels, while men are highly active. Depending on the model of mechanically induced bone remodeling that one accepts, moderate levels of activity may not be sufficient to engender functional adaptive response in the mechanical properties of bone. When compared to the American sample, the specific PAL and reproductive history of the women is unknown. While it is probably safe to assume that the women did have lower fertility and longer inter-birth intervals, it may not be as safe to assume differences in PALs. The Tsimane women have moderate PALs, the American women may have had highly variable PALs with the same population, and activity levels that may have also been 'moderate'. Very little information is provided here about the American data, either in terms of population characteristics and context, or methods, so the reader cannot evaluate whether the characterization of the American group is appropriate.

For Tsimane men and women, PALs (defined as the ratio of total daily energy expenditure to daily basal metabolism) are about 0.1–0.3 units higher than those of industrialized populations (Gurven et al., 2013). A difference of only 0.1 PAL may seem small, but for a typical Tsimane man (62 kg) or woman (56 kg) amounts to an additional 144 and 120 calories expended per day in activity, respectively. As we acknowledge above and in the revised manuscript, a limitation of our study is that we lack activity data (frequency and intensity) at the individual level for both the comparative American and Tsimane samples. However, we can examine previously published activity data from the Multi-Ethnic Study of Atherosclerosis (MESA) sample – which represents one of the American comparative samples in our study (for vertebral fracture prevalence) – to get a sense of American activity patterns. Nearly all MESA participants completed the Typical Week Physical Activity Survey (n=5,829), which was designed to identify time spent in various activities during a “typical week in the past month” (https://www.ncbi.nlm.nih.gov/pubmed/27403323). Only one-third of MESA participants reported any vigorous activity in a typical week, and mean ± SD hours per day in “leisure sedentary behaviour” (i.e. television watching and reading) was 3.4 ± 2.2, including 2.1 ± 1.5 hours/day watching television (see Table 1 in this Joseph et al., 2016 paper). These data are not directly comparable to Tsimane data but nevertheless reveal important population-level differences in activity patterns (Tsimane study participants do not watch [or own] televisions and most do not read recreationally). Instead, Tsimane adults typically spend their leisure time either visiting family or friends, playing soccer or even going fishing or hunting, each of which entails greater activity than watching television or reading. For this reason and those mentioned above, we think our characterization of mean activity differences between populations is appropriate, while acknowledging that we do not characterize sample variances due to lack of individual-level activity data.

In terms of methods, the information on how the American data was collected is insufficient as presented. The only details given are "a matched American sample with directly comparable CT derived indicators of bone strength measured by the same laboratory". Specific details of the equipment and settings are provided for the Tsimane data, were the Los Angeles data collected on the same equipment? With the same settings? Were any objects/participants scanned in each location to validate pooling data from different equipment? This needs to be expanded upon, if only briefly in the text or in supplementary information, so that the reader can evaluate for themselves comparability of approaches.

Thanks for pointing out this oversight. We now provide details in the supplement on how American data were collected (subsection “Thoracic computed tomography (CT)”).

Overall, while my review has been critical of several aspects of the study and interpretations, I would summarize the work more positively. The Tsimane research has provided an extremely important dataset with very interesting social and biological context. The analyses are appropriate and reveal some very interesting trends in bone mass. My concerns with the paper are primarily about interpretation and assumptions relating to activity levels in the American dataset. Both of these concerns mean it is difficult to evaluate the 'life history' hypothesis relative to other proximate mechanisms, or even more broad variation in genetics, habitual activity (and what influence it may have on thoracic vertebrae), early life nutrition, growth plasticity, prevalence of bottle feeding, or extended lactation that may be associated with bmd losses. It is not clear how life history variation necessarily provides a stronger explanation for the data than would any one of these other factors, so alternate explanations should be considered and set up, where possible, as alternate explanations of the data with different predictions of the underlying variation. I believe this dataset and analysis is important and publishable, but it requires more nuanced hypothesis testing to differentiate life history mechanisms from competing hypotheses.

Thank you very much for your thoughtful, wide-ranging and constructive comments. Please refer to the comment entitled “SUMMARY and Essential revisions”) for a more in-depth response to your concerns expressed in the above paragraph.

Reviewer #2:

[…] 1) There are a number of predictions made at the beginning of the manuscript (Introduction) and consequently there are many results to report and discussion to follow. I recommend that the Discussion section follow the order in which the predictions were made so that it is easier for the reader to follow and tie in the predictions and findings.

We appreciate reviewer 2’s helpful recommendation to enhance organization of the Discussion section and paper overall. As recommended, we have restructured the Discussion to more closely follow the order in which the predictions are outlined in the Introduction. To further enhance clarity and organization in the Discussion we added sub-section headings.

2) Additionally, in the Introduction there is extensive discussion about gracilization especially in the fossil/archaeological record, however in the Discussion section the subject is a little glossed over. I would suggest that the implications and/or the complexity of interpreting gracilization from a broader perspective be discussed. I realize that the authors do this e.g., in the Discussion, but a little more could be added as a way to explain/justify the value of this study which goes beyond investigating physical activity levels unlike some of the previous work in the fossil and archaeological record.

Again, we appreciate reviewer 2’s suggestion to improve the Discussion section. Reviewer 2 states that a notable imbalance exists between the Introduction and Discussion in the mention of gracilization in the fossil/archaeological record, with the imbalance favoring the Introduction over the Discussion. In fact, whereas the original Introduction contained ~5 lines of text regarding gracilization in the fossil/archaeological record (half of the first paragraph), the Discussion contains >40 lines of text, which now – after incorporating reviewer 2’s helpful comment #1 above – includes an entire sub-section with three paragraphs. We thus do not feel as though we paid insufficient attention to this issue in the Discussion relative to the Introduction. In the Introduction, when motivating our study’s objectives, we do not pay more attention to gracilization in the fossil/archaeological record because our study’s focus is on contemporary populations, and we cannot directly test hypotheses about behavioral factors underlying gracilization in past human populations. Lastly, we refrained from further discussing gracilization in the Discussion or Introduction because we wanted to respect eLife’s suggested maximum word count.

3) While the authors describe how BMD was measured using the computed tomography, I would suggest that a brief description of whether BMD estimates includes both cortical and trabecular bone and if not, how were the two demarcated? The information regarding how the scanner estimates BMD is found in the supporting information section, if there is room, I would include this in the main Materials and methods section. Also, BMD in three thoracic vertebrae was measured, I suppose the BMD reading is the mean of the three and if so, is this a reading provided automated from the scanner or is this done by the researchers manually after the scans are completed?

We thank reviewer 2 for their attention to detail. We now clarify the subsection “Study design and participants”, whether BMD measurements include cortical bone, and we clarify how cortical bone was demarcated from trabecular bone. We also moved additional details on BMD estimation from the supplement to the main Materials and methods section (subsection “Thoracic vertebral bone mineral density (BMD)”), as suggested similarly by reviewer 3. Lastly, we now clarify that BMD was measured manually by the radiologist (subsection “Thoracic vertebral bone mineral density (BMD)”).

4) I found Figure 4 to be a bit confusing to follow and that is because there are multiple variables included in the analysis. Would it be possible to have the y-axis reading be "relative risk of fracture" without the per SD increase? This way the data are more easily interpreted?

This figure shows the bivariate association between BMD and fracture risk; associations are shown for women and men, and are adjusted for potential confounders which are described in the figure caption. Each point estimate shown and its associated 95% CIs are derived from a different regression model whose output is included in the supplement; to enhance clarity, we have modified the figure caption to inform the reader precisely where in the supplement those regression results are located.

Reviewer #3:

[…] 1) In the Introduction (first paragraph) and Discussion (fourth paragraph), more specificity is needed in the discussions of bone strength. "Bone strength" can refer to 1) midshaft robusticity or cross-sectional geometry (the shape and area of cortical bone in a given plane) vs. 2) bone mineral density (grams of mineral in a given area or volume of bone, which can be measured in cortical or trabecular bone). They are both generally correlated with strength but can't be directly compared to each other. This is important because the background largely focuses on studies of midshaft cortical bone, but the vertebra, which was measured in this study, is primarily trabecular bone surrounded by a thin cortical shell. Note also that the Chirchir references use the word 'density' to refer to trabecular bone volume fraction (BV/TV (%), or the amount of bone in a region of interest) while most publications use density to refer to the amount of bone mineral per area or volume of bone.

We appreciate reviewer 3’s concern, and we now specify indicators of bone strength in these manuscript sections. We did not intend to imply that disparate measures of bone strength (e.g. trabecular bone volume fraction and diaphyseal size or shape) can be directly compared to each other, and we fully agree with reviewer 3 that precise terminology is needed to minimize confusion given the numerous ways that researchers measure bone strength. Reviewer 3 is correct in stating that “the background largely focuses on studies of midshaft cortical bone”, whereas our study focuses on vertebrae, which is “primarily trabecular bone surrounded by a thin cortical shell.” The reason for this discrepancy is that, generally speaking, studies of human midshaft cortical bone are more prevalent in the bone functional adaptation literature within anthropology than studies of trabecular bone. While there has been a recent surge in studies of human trabecular bone, including vertebrae (many of which we cite), studies of midshaft cortical bone still dominate this literature (particularly with respect to bone responsivity to mechanical loading, as noted by reviewer 3 in their next comment below). Our study’s background thus reflects our reading of the current state of the relevant literature, rather than reflecting an attempt to motivate our study’s objectives and situate our findings based on invalid direct comparisons with prior studies utilizing disparate methods.

2) Introduction, second paragraph, Discussion first and third paragraphs: Osteogenic responses to mechanical loading occur predominantly in the ~2 years before and after puberty, and are markedly attenuated after menarche (although mechanical loading is essential for maintaining existing bone). See review, MacKelvie et al. 2002, http://dx.doi.org/10.1136/bjsm.36.4.250 Also, while cortical bone is highly responsive to mechanical loading around puberty, much less is known about trabecular bone response to mechanical loading. After menarche, physical activity does not provoke a substantial osteogenic response, and consistent with the life history framework presented in this article, energy expended on physical activity might arguably inhibit skeletal recovery (as in the Female Athlete Triad).

We thank reviewer 3 for providing the MacKelvie et al. review paper on bone responsivity to weight-bearing exercise among children, which we have now read. We fully agree with reviewer 3 that osteogenic responses to mechanical loading are much stronger in childhood vs. adulthood (particularly in the years just before and after puberty), and that among older adults – the focus of our study – the principal somatic effect of mechanical loading is to preserve existing bone rather than add new bone. The marked post-pubertal attenuation of human osteogenic responses to mechanical loading noted by reviewer 3 is evident from observational studies (cross-sectional and longitudinal) and controlled exercise trials. Meta analyses of controlled exercise trials indicate that adult participation in high-impact activities with high-magnitude loading is especially effective at preserving BMD at the lumbar spine and femoral neck (e.g. https://www.ncbi.nlm.nih.gov/pubmed/18981037 and https://www.ncbi.nlm.nih.gov/pubmed/20013013); both of these skeletal sites include substantial amounts of trabecular bone, suggesting that bone responsivity to mechanical loading is not limited to cortical bone. In response to reviewer 3’s comment, we have modified the text in several ways. We eliminated from the entire manuscript the phrase, “osteogenic response to mechanical loading and high PAL”, as it can imply, inaccurately in the present context, a somatic outcome regarding addition of new bone rather than preservation of existing bone. The ensuing revised text (Introduction, third paragraph, Discussion, third paragraph, and subsection “Study limitations”) is consistent with reviewer 3’s comment that under some circumstances “energy expended on physical activity might arguably inhibit skeletal recovery”. In addition, we now describe and cite the Martyn-St James and Carroll meta-analyses of controlled exercise trials mentioned above (Discussion, second paragraph).

3) Previous studies in industrialized populations have looked for relationships between BMD and age at first birth, parity, etc. The authors may wish to cite/discuss Parazzini et al. 1996 J Epidemiol Community Health; We et al. 2018 doi: 10.3346/jkms.2018.33.e311; Chantry et al. 2004; Bexerra et al. 2002 DOI: 10.1093/jn/132.8.2183; etc.

Thank you for providing these references. We have now read each of these papers. In the revised manuscript we now cite the Chantry et al., 2004 paper (Introduction, fourth paragraph) when mentioning potential effects of early-life pregnancy and lactation on later-life bone strength. Our mention of the Chantry paper is part of our more general summary of findings reported in Kovacs’ extraordinary 2016 review (which includes >1,000 references) on changes in maternal bone mineral during pregnancy, lactation and post-weaning.

4) The idea that these data reflect tradeoffs in energy allocation to maintenance vs. reproduction is intriguing and makes intuitive sense. It raises a crucial question, however: is energy intake limited in the Tsimané? If not, then why must there be such tradeoffs? Given all the data available to the authors, it would be helpful to identify specific biological mechanisms that are mediating the effects of age at first birth and IBI on thoracic vertebral BMD and fracture incidence in Tsimané vs. American women (or at least frame hypotheses about what they are). For example, what are estradiol levels in Tsimane women, and how do they compare to those of the American women described in this study?

Our characterization of Tsimane as “energy-limited” refers to their high energy expenditure relative to consumption (Discussion, second paragraph), and the general lack of food storage. High energy expenditure results largely from a combination of their physically active subsistence lifestyle, high infectious burden requiring energetic investments in immune responses (https://www.ncbi.nlm.nih.gov/pubmed/27375044) and high fertility rates. Recently published data from a small sample using the doubly-labeled water technique suggest a median total energy expenditure for Tsimane women and men of 2,186 and 3,065 kcal/day, respectively (n=20/sex) (https://www.ncbi.nlm.nih.gov/pubmed/27375044). Per-capita energy intake for Tsimane women and men is 2,422 and 2,736 kcal/day (https://www.ncbi.nlm.nih.gov/pubmed/30383188). These estimates were collected in different Tsimane sub-samples and can vary by season and other factors (e.g. pregnancy or lactation status), but nonetheless suggest that Tsimane (particularly men) are often in negative energy balance or only minimally achieve energetic surplus. Tsimane food insecurity is also relatively common (https://www.ncbi.nlm.nih.gov/books/NBK242449/), which reflects in part residential detachment from the market economy. For these reasons we characterize Tsimane as energy-limited compared to industrialized populations.

With respect to identification of specific physiological mechanisms mediating effects of reproductive effort on thoracic vertebral BMD and fracture, there are a plethora of candidates reviewed by Kovacs 2016. These include mineral ions and calciotropic and phosphotropic hormones, parathyroid hormone and parathyroid hormone-related protein, calcitriol and calcidiol, calcitonin, fibroblast growth factor-23, sex steroids and other hormones, upregulation of intestinal absorption of calcium and phosphorus, altered renal mineral handling, and other biochemical markers of bone turnover. At present we either have none of these relevant data for Tsimane, or we cannot directly compare Tsimane vs. American levels (e.g. because assays were performed in different labs using different protocols). In his comprehensive 2016 review, Kovacs notes that relevant physiological mechanisms such as estradiol (since it was mentioned by reviewer 3) operate in an interactive fashion (e.g. with parathyroid hormone-related protein), thus making it difficult to isolate the independent effect of each mechanism. Kovacs also writes, “It is difficult to analyze the independent effects of low estradiol on bone turnover during lactation. More intense lactation causes more breast milk output and bone loss, but more intense lactation is also associated with lower estradiol and prolonged amenorrhea. The impact of isolated low estradiol is that it causes 1–2% annual losses of BMD in recently menopausal women, but the older ages of these women may not reflect what happens to reproductive age women with low estradiol.” These physiological processes are obviously complex and in our view outside the scope of the present study. We thus prefer to avoid speculation regarding how physiological mediators may operate between populations.

https://doi.org/10.7554/eLife.48607.034

Article and author information

Author details

  1. Jonathan Stieglitz

    1. Université Toulouse 1 Capitole, Toulouse, France
    2. Institute for Advanced Study in Toulouse, Toulouse, France
    Contribution
    Conceptualization, Resources, Data curation, Software, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing—original draft, Project administration, Writing—review and editing
    For correspondence
    jonathan.stieglitz@iast.fr
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-5985-9643
  2. Benjamin C Trumble

    1. Center for Evolution and Medicine, Arizona State University, Tempe, United States
    2. School of Human Evolution and Social Change, Arizona State University, Tempe, United States
    Contribution
    Funding acquisition, Methodology, Supervision, Project administration, Writing—review and editing
    Competing interests
    No competing interests declared
  3. HORUS Study Team

    Contribution
    Supervision, Validation, Investigation, Methodology
    Competing interests
    No competing interests declared
  4. Caleb E Finch

    Leonard Davis School of Gerontology, University of Southern California, Los Angeles, United States
    Contribution
    Project administration, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-7617-3958
  5. Dong Li

    School of Medicine, Emory University, Atlanta, United States
    Contribution
    Data curation, Investigation, Methodology
    Competing interests
    No competing interests declared
  6. Matthew J Budoff

    Los Angeles Biomedical Research Institute, Harbor-University of California at Los Angeles Medical Center, Torrance, United States
    Contribution
    Resources, Supervision, Methodology, Project administration
    Competing interests
    No competing interests declared
  7. Hillard Kaplan

    Economic Science Institute, Chapman University, Orange, United States
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Investigation, Methodology, Project administration
    Competing interests
    No competing interests declared
  8. Michael D Gurven

    Department of Anthropology, University of California, Santa Barbara, Santa Barbara, United States
    Contribution
    Funding acquisition, Investigation, Supervision, Project administration, Writing—review and editing
    Competing interests
    No competing interests declared

Funding

National Institutes of Health (R01AG024119)

  • Jonathan Stieglitz
  • Benjamin C Trumble
  • Caleb E Finch
  • Hillard Kaplan
  • Michael D Gurven

Arizona State University

  • Benjamin C Trumble

University of California, Santa Barbara

  • Michael D Gurven

Agence Nationale de la Recherche (ANR-17-EURE-0010)

  • Jonathan Stieglitz

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank the Tsimane for participating and THLHP personnel for collecting and coding data. Jay Stock and two anonymous reviewers provided detailed comments that improved the quality of this manuscript. Funding was provided by the National Institutes of Health/National Institute on Aging (R01AG024119), the Center for Evolutionary Medicine at Arizona State University, and the University of California-Santa Barbara Academic Senate. JS acknowledges IAST funding from the French National Research Agency (ANR) under the Investments for the Future (Investissements d’Avenir) program, grant ANR-17-EURE-0010.

Ethics

Human subjects: Institutional IRB approval was granted by UNM (HRRC # 07-157) and UCSB (# 3-16-0766), as was informed consent at three levels: (1) Tsimane government that oversees research projects, (2) village leadership and (3) study participants.

Senior and Reviewing Editor

  1. Diethard Tautz, Max-Planck Institute for Evolutionary Biology, Germany

Reviewer

  1. Jay Stock, University of Cambridge, United Kingdom

Publication history

  1. Received: May 20, 2019
  2. Accepted: August 14, 2019
  3. Accepted Manuscript published: August 16, 2019 (version 1)
  4. Version of Record published: September 4, 2019 (version 2)
  5. Version of Record updated: September 9, 2019 (version 3)

Copyright

© 2019, Stieglitz et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 435
    Page views
  • 50
    Downloads
  • 2
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Developmental Biology
    2. Evolutionary Biology
    Mai P Tran et al.
    Research Article Updated
    1. Cell Biology
    2. Evolutionary Biology
    Mukund Thattai
    Insight