Summary of sample size findings

A typical study with sex-based analyses involved four groups, with average minimum group size of 4 and average maximum group size of 8 per sex. Among studies that pooled sexes for analyses, the average number of groups was 2, with average minimum group size of 6 and average maximum group size of 8. Studies that performed sex-based analyses tended to involve more groups since males and females constituted separate groups on top of experimental treatments. Most often the groups consisted of a single experimental treatment compared against controls, although 15 studies of maternal over-nutrition investigated dietary reversal, with over-nutrition and control diets alternated at different timepoints.^{Reference Sullivan, Rivera and True1,Reference Thompson, Valleau and Barling17,Reference Grant, Nicol, Thorn, Grove, Friedman and Marks18,Reference Cox, Williams, Grove, Lane and Aagaard-Tillery24,Reference Grant, Gillingham and Batra28,Reference Roberts, Pound and Thorn41–Reference Suter, Chen and Burdine43,Reference McCurdy, Bishop and Williams45,Reference Fan, Lindsley and Comstock52–Reference Suter, Takahashi, Grove and Aagaard54,Reference Comstock, Pound and Bishop58,Reference Suter, Bocock and Showalter59,Reference Pound, Comstock and Grove61} The maximum sized group in a study was most frequently the control group. Few studies that pooled sexes reported the numbers of males and females – only 12 of 42.

All studies reported significant differences between groups. It would be useful to review sample sizes in studies that produced significance between groups versus those that did not. However, this is impossible given that very few papers report nonsignificant findings. This publication bias is well known.^{Reference Dwan, Gamble, Williamson and Kirkham62} Readers may wonder why we do not report outcome means, effect sizes, or levels of significance in the tables to pinpoint sample sizes needed for adequate statistical power in each of these specific outcomes. There are two reasons: (1) Each study reported numerous different outcomes, and the minimum and maximum sample sizes often applied to multiple outcomes per study. (2) More importantly, outcomes reported in all the different studies were mostly unique, with almost no replication of work by other investigators. NIH expressly requests studies addressing reproducibility. Conducting a power analysis typically requires prediction of sample means. The studies reported herein could be used for preliminary data, but since there is no replication of findings, we cannot generate predicted means across multiple studies to enhance power analyses. In other words, because of the relative scarcity of data, there are few replicative studies that can be compared, so assembling these data at this point would not advance the results of this paper. Given the lack of replication, the best we can do currently is to report typical sample sizes across multiple outcomes to give investigators a clue as to how many animals have been used before to produce successful studies. As the body of research grows and work is replicated, more sophisticated approaches to sample size determination will become possible. For the same reasons, meta-analysis is not possible at this stage.

It is reassuring that the studies we investigated produced a wide variety of significant findings with sample sizes of 4–8 animals per group. Our own experience and discussions with other investigators in the field have indicated that peer-reviewers often criticize these numbers as too small and disapprove the lack of power analysis. We document here that sample sizes of 4–8 are not small but instead average (and often successful) for this type of research, as well as the legitimate reasons for not conducting power analyses.

Challenges of NHP developmental programming studies

NHP studies present challenges not present in studies of smaller common laboratory animal models like rodents, including body size, lifespan, and gestation length. All investigations reported here studied particularly large-bodied primates, Japanese macaques, and olive baboons. Both Japanese macaques^{Reference Paterson63} and olive baboons^{Reference Coelho and Watts64} complete skeletal growth at 7 years in females and 8 years in males; at these ages, young adult captive olive baboons at SNPRC weigh on average 29 kg in males and 17 kg in females,^{Reference Coelho and Watts64} while young adult captive Japanese macaques at ONPRC weigh on average 16 kg in males and 12 kg in females.^{Reference Paterson63} Lifespan of captive olive baboons at SNPRC is 21 years on average, with a maximum lifespan of 33 years.^{Reference Bronikowski, Alberts, Altmann, Packer, Carey and Tatar65} In the semi-free ranging colony of Japanese macaques of Arashiyama West, mean lifespan is 25 years and maximum lifespan 33 years.^{Reference Pavelka and Fedigan66} Gestation length in Japanese macaques is approximately 173 days and in olive baboons 180 days.^{Reference Harvey and Clutton-Brock67} Both species typically give birth to single offspring.^{Reference Harvey and Clutton-Brock67} The life history features of baboons and macaques make them costly to raise in captivity; they require a large amount of space due to their large body size, must be housed for many years to study development and aging, and reproduce slowly. Typical grant periods of 2–5 years are therefore insufficient to conduct even short-term studies, much less longitudinal analyses, so researchers must be prepared to seek many years of new funding and renewals. Despite these practical hurdles, baboons and macaques remain the nearest NHP to humans since great apes are not used in biomedical research. They are also excellent translational models of developmental programming precisely because of their body size, lifespan, gestation length, and singleton births, which make them much more similar to humans than are rodents. Moreover, their hormonal profiles, complex social behavior, and brain size exhibit a high degree of similarity to humans, indeed because they are the animals most closely related to humans other than apes. It would be useful if studies described in greater detail how the animals were housed, environmental enrichment procedures, selection of animals, preparation of animals for study, veterinary care employed, and metrics of health assessed to evaluate animal well-being. These specifics would be valuable for planning future studies.

As referred to above, given the sociality of NHPs, they ideally are kept in social groups requiring large enclosures. This leads to another obstacle: controlling food intake. This is most challenging in studies of maternal under-nutrition, since animals must be fed individually. Feeding in social groups does not allow modification of quantity of diet because even if less food is provided to the group, some individuals will monopolize food access, leading to dominant individuals eating more and subordinate individuals eating less, if at all. Special provisions are needed to control feed access. We have addressed this issue at SNPRC by building a facility that allows for social housing but separation during feeding time.^{Reference Schlabritz-Loutsevitch, Howell and Rice68} Large social cages are connected to chutes leading to individual feeding cages. Baboons are easily trained to run through the chutes to and from the feeding cages, where intake can be closely monitored. This system is much better for animal welfare than individual indoor housing, since animals receive the great benefits of social outdoor lives. New technology is transforming ability to control intake. Research Diets has developed the BioDAQ NHP system, which consists of a feeding station that records feeding bouts with RFID microchips implanted into the monkeys. The system identifies individuals and then weighs and dispenses food accordingly, recording date, time, and amount taken by each monkey. BioDAQ NHP is in use at Yerkes National Primate Research Center with rhesus macaques.

The strength of all studies depends on a homogenous experimental subject base. This is particularly true with regard to programming, where features of the parental phenotype – such as maternal BMI, age, previous obstetric history, and even now it is believed paternal phenotype – can differentially affect outcomes in offspring and constitute a major reason for uniformity of subjects. Variance in the mothers produces greater variance in the offspring. Thus, ensuring uniformity of the parental generation prior to any experimental manipulations is a crucial hurdle.

All 61 studies discussed herein appear to have been performed in two laboratories, one using macaques as a model at ONPRC and one using baboons as a model at SNPRC. Based on ages studied and the need for tissue collection at each age, each laboratory bred at least five groups of offspring. The ONPRC-based group studied fetal macaques at gestational days 30, 90, 130, and 180, as well as postnatal macaques. In the baboon experiments based at SNPRC, animals were studied at gestational days 90, 120, 165, and 175, as well as postnatally. This point highlights both a challenge and strength of NHP studies. Given all the requirements of NHPs discussed above, such as long length of studies and complex housing, NHPs are quite costly to study, presenting a challenge that few laboratories have been able to embrace. However, this is also a strength since breeding animals under the same or highly similar conditions enhances uniformity of the studies, allowing researchers to build a comprehensive phenotype integrating several physiological systems in the similar cohorts of animals. This strength is further enhanced when the same animals are used for studies on different systems, such as brain,^{Reference Franke, Clarke and Dahnke9} carbohydrate metabolism,^{Reference Choi, Li, McDonald, Comuzzie, Mattern and Nathanielsz57,Reference Kuo, Li and Mattern15} and cardiovascular system,^{Reference Kuo, Li, Li, Huber, Nathanielsz and Clarke11,Reference Kuo, Li, Huber, Schwab, Nathanielsz and Clarke12} thereby allowing rigorous integration within a cohort as opposed to integrating data on different systems from different cohorts whose individuals may differ in other confounds.

Deficiencies in the literature and directions for future work

The challenges of longitudinal studies have led to deficiencies in the current literature of NHP developmental programming by maternal nutrition. Studies are overwhelmingly categorical, rather than regression-based, since regression analyses require larger sample sizes and longer time frames to investigate different stages of development. Categorical studies provide less statistical power and lower ability to understand changes over the course of development and aging. There is also a lack of a wide range of ages studied, with most studies focusing on gestational days 90, 130, and 165. Only 22 of the 62 studies reported here investigated postnatal animals; of these, just 7 studies involved offspring older than 7 years, the age at which female macaques and baboons complete skeletal growth.^{Reference Paterson63,Reference Coelho and Watts64} There is growing interest in interactions of developmental programming and aging,^{Reference Franke, Clarke and Dahnke9,Reference Kuo, Li, Li, Huber, Nathanielsz and Clarke11,Reference Salmon, Dorigatti, Huber, Li and Nathanielsz16,Reference Huber, Kuo and Li69,Reference Kuo, Huber, Li, Li, Nathanielsz and Clarke70} so future work must involve study of adult aging and aged NHPs from different developmental backgrounds. No studies have investigated F2, so inter-generational effects are entirely unstudied in NHPs. The importance of inter-generational studies is shown by the multiple reports of intergenerational and transgenerational epigenetic and non-epigenetic transfer in guinea pigs,^{Reference Bertram, Khan, Ohri, Phillips, Matthews and Hanson71} rodents,^{Reference Aiken and Ozanne72–Reference Burdge, Slater-Jefferies, Torrens, Phillips, Hanson and Lillycrop75} and sheep.^{Reference Pankey, Walton and Odhiambo76,Reference Shasa, Odhiambo, Long, Tuersunjiang, Nathanielsz and Ford77}

Another major weakness in the literature to date is investigation of sex as a biological variable. Twice as many studies reported here pooled sexes for analysis rather than tested for differences between sexes. Although studies often state that sexes were pooled because no sex-based differences were found, it is unlikely that sample sizes were large enough per sex for determination of the potential existence of sex-based differences, given the mean minimum sample size of 6 and mean maximum sample size of 8, which leaves only 3–4 individuals per sex, assuming the number per sex was equal. It is impossible to know from most of these studies whether the number per sex was equal because less than a third of studies reported the numbers of males and females per pooled group. There is substantial evidence of differential vulnerability to effects of programming between males and females,^{Reference Zambrano and Nathanielsz78,Reference Aiken and Ozanne79} with males seen as more prone to negative effects of prenatal stress and females as more adaptable.^{Reference Sandman, Glynn and Davis80} However, studies have shown that both sexes are sensitive but in different ways, especially in susceptibility to cardiovascular disease^{Reference Dasinger and Alexander81–Reference Ojeda, Intapad and Alexander83} and dysregulated metabolism.^{Reference Dearden, Bouret and Ozanne84}

An additional challenge of sex as a biological variable in NHP programming studies is the possibility that maternal nutrition biases the birth sex ratio.^{Reference Huber, Bartlett, Smith, Li, Jenkins and Nathanielsz85} If one sex is born more often than the other in NHP programming studies, it becomes difficult to balance the sexes well enough for incorporation of sex as a biological variable, leading to a need for even larger sample sizes. In both baboons and macaques, but especially so in baboons, males are much larger than females. In species in which males are substantially larger than females, male growth and development may be faster and energetic costs higher.^{Reference Clutton-Brock, Albon and Guinness86} In circumstances of resource scarcity (e.g., maternal under-nutrition), the higher costs of raising the larger sex may result in differential perinatal mortality between sexes, leading to fewer males being born to mothers in poorer physical condition.^{Reference Clutton-Brock, Albon and Guinness86} An alternative hypothesis is that it is more costly to raise the sex with stricter inheritance of the dominance hierarchy, which is females in both baboons and macaques.^{Reference van Schaik and Hrdy87,Reference Silk and Brown88} Investigation of birth sex ratio in developmental programming by maternal nutrition is also fraught with challenges of sample size. According to one analysis, a sample size of over 2000 animals would be needed to have adequate power to detect birth sex ratio effects of developmental programming by maternal nutrition.^{Reference Huber, Bartlett, Smith, Li, Jenkins and Nathanielsz85} This study presents a good example of how preliminary data are used for power analysis. The researcher first assembled data on sex produced by 319 births to mothers on three different diets (under-nourished, over-nourished, and control). Then, using the determined level of difference in numbers of males and females born to each group, a power analysis could be conducted to determine power of the current study (15%) as well as sample size needed for a higher powered (80%) future study. It was only after the initial study was conducted that the data necessary for the power analysis were available.

Another goal of future work should be incorporation of more diverse topics. Metabolism and metabolomics have been by far the best studied, undoubtably due to the high level of interest in predisposition to obesity and metabolic disease. Epigenomics, neurological outcomes, and cardiovascular outcomes have received moderate attention. Endocrinology of growth, microbiome, and reproduction have been little studied, and therefore there are no data for power analysis. A promising method that has been little used to date is telemetry. Implanting small telemeters (<60 g) into NHP allows for continuous remote collection of a wide range of data (e.g., blood pressure, activity levels, glucose levels, temperature, respiration, electromyography, electrocardiography) while allowing animals to live in social groups with minimal sedations. Once a greater diversity of studies is achieved, as well as replication of studies, it will be possible to perform meta-analyses. Meta-analysis is highly desirable to determine variability in outcomes and true effect sizes.

A major objective of translational developmental programming studies is development of interventions to ameliorate undesirable outcomes. However, only two studies of NHP programming by maternal nutrition have reported effects of an intervention, in both cases, maternal resveratrol supplementation to counteract the effects of an obesogenic Western-style diet.^{Reference Roberts, Pound and Thorn41,Reference Pound, Comstock and Grove61} The results of resveratrol supplementation were mixed, with one study finding improvements to maternal and placental phenotype and to fetal liver development, but inexplicable and potentially harmful changes to fetal pancreatic development.^{Reference Roberts, Pound and Thorn41} The other study reported that the Western-style diet led to decreased pancreatic islet capillary density; diet reversal normalized islet vascularization, while resveratrol supplementation resulted in hypervascularization beyond the level of controls, which could have detrimental consequences.^{Reference Pound, Comstock and Grove61} In neither of these studies was amount of resveratrol consumed reported. As discussed above, controlling and measuring food intake is a major challenge in NHP research due to the need to feed individually in a species where dominance rank can prevent animals from receiving their portion. One of the studies^{Reference Roberts, Pound and Thorn41} calculated blood resveratrol concentrations during an intravenous infusion of resveratrol in a subset of animals. In rodent studies, resveratrol,^{Reference Vega, Reyes-Castro and Rodríguez-González89,Reference Zou, Chen and Yang90} maternal and offspring exercise,^{Reference Vega, Reyes-Castro, Bautista, Larrea, Nathanielsz and Zambrano91,Reference Santos, Rodríguez-González, Ibáñez, Vega, Nathanielsz and Zambrano92} and coenzyme Q10^{Reference Tarry-Adkins, Fernandez-Twinn and Madsen93,Reference Tarry-Adkins, Fernandez-Twinn and Hargreaves94} have shown promising results in mitigating the effects of less-than-ideal maternal diet during pregnancy. Future NHP work should aim to further explore similar interventions and treatments, especially given the potential for interventions like resveratrol to lead to both advantageous and damaging outcomes.

Rodent models of developmental programming have been highly valuable due to the ease of complex manipulations to genetics and diet, fast reproduction, and shorter lifespans when compared to NHPs. However, rodent models are limited in capacity for translation to humans. Some examples of the differences between humans and rodents that are important to programming studies include that rodent lipid metabolism is such that atherosclerosis is usually achieved with knocking out of genes and dietary manipulation, rodents give birth to litters of offspring with the added complication of within-womb competition between litter-mates, rodents are so small that tissue collection is inherently limited, and rodents lack the extended childhood period seen in humans and NHPs.^{Reference Rabadán-Diehl and Nathanielsz95} For these reasons and many more, NHP models are necessary to bridge the gap between rodent and human developmental programming.