Cross-species comparisons of host genetic associations with the microbiome

Introduction

Microbes coat the surfaces of the human body in highly adapted microbiomes. The vast majority of human-associated microbial cells are bacterial, with archaeal and eukaryotic cells also present in lower numbers. Together they reach their highest biomass in the distal gut. These microbial communities vary in composition across the body depending on niches defined physically, chemically, and immunologically. Although a handful of animal species have been reared germfree under highly artificial axenic conditions, germfree animals are physiologically abnormal in several fundamental ways (1). Human biology and health assume the presence of a “healthy” microbiome, one whose interactions with its host are generally beneficial.

The gut microbiome is said to encode a second genome, and its functions expand the host’s physiological potential. In the gut, microbes extend digestive capabilities, prime the immune system, produce vitamins, degrade xenobiotics and resist colonization by pathogens. In principle, selection for host genotypes that promote a beneficial microbiome is possible. This might be expected for microbial functions (e.g., production of metabolites) or taxa that encode functions beneficial to fitness. When functions are restricted phylogenetically to specific taxa, such as methanogenesis to Euryarchaeota, associations between host alleles and specific taxonomic abundances could emerge. Conversely, functional gene counts or expression levels may associate with host genotypes when those functions are encoded across disparate taxa. There are examples of specific host genes whose variants are associated with different gut microbiomes, particularly for immune genes implicated in disease or secretor status (2, 3).

Genome-wide scans hold the promise of finding novel associations between host genes and the microbiome. A major concern with this approach in humans is that environmental factors, such as diet, can strongly and rapidly alter gut community composition and function (4). Nonetheless, recent studies highlight a select suite of taxa whose abundances are partially genetically influenced, as well as the human genes involved. Cross-study comparisons show some of the same taxa to be influenced by host genetics. Comparisons across species, including plants, reveal a common theme: Host genes associated with microbiome variation are involved in immune regulation and barrier defense. In human populations where the diet is unrestricted, genes related to diet preference and metabolism also emerge.

Heritability: Estimates of the strength of the host genetic effect on microbiota

Heritability is the proportion of variance in a host trait, such as height or BMI, measured across a population, that is explained by genetic rather than environmental effects. Note that heritability, a term widely used in genetics, is unrelated to the concept of inheritance (vertical transmission from parent to offspring). Height, for example, is highly heritable, meaning that variation in height across a population has strong genetic underpinnings. Components of the microbiome, such as taxa, can be quantified across subjects and treated as quantitative traits just like height or BMI in estimates of their heritability. Statistical models use the known genetic relatedness of twin pairs, or SNP genotype data that allows direct assessment of the genetic relatedness between individuals, to calculate the heritability (ranging from 0 to 1) for each microbial trait. For twins, the vertical transmission of microbes from parents is assumed to be equivalent and is thus controlled for.

For several decades twins served to address the question of whether host genetic variation is associated with the microbiome. Identical (monozygotic; MZ) twin pairs share 100% of the genes across their genome, whereas fraternal (dizygotic; DZ) twins share on average 50%. This, combined with the assumption that twins raised together experience similar environments, has formed the basis of twin heritability studies. For quantitative traits derived from the microbiome, such as bacterial relative abundances, greater similarity for MZ twins compared to DZ twins can be ascribed to shared genes, and by definition such bacteria are heritable. With a large enough population sample, the heritabilities of taxon abundances or other quantitative aspects of the microbiome, such as species richness, can be quantified.

Early studies in twins included small numbers of subjects and employed either culture or DNA fingerprinting-based techniques, and results suggested a genetic effect on the gross composition of the gut microbiome (5, 6). Later, studies using sequence-based techniques and larger cohorts reported a similar trend, albeit weaker. Turnbaugh et al. (2009) and Yatsunenko et al. (2012) sampled equivalent numbers of young-adult twin pairs (~50) from Missouri, and characterized their gut microbiomes by sequencing partial 16S rRNA genes amplified from fecal DNA (7, 8). Comparisons of microbiomes using the UniFrac metric in both studies showed the trend that MZ twin microbiomes were more similar than those of DZ twins, however this result was not statistically significant. Together these studies highlight a global impact of environmental factors on the composition of the microbiome. They also hinted at a host genetic effect on gut microbiome variation.

Accurate assessment of genetic effects requires larger sample sizes. The work of Goodrich and colleagues included an initial sample size of 416 twins pairs (9). The patterns from earlier twin studies were successfully replicated: MZ twin microbiomes were overall more similar than those of DZ twins, but with the increased sample size the difference reached statistical significance (9). More importantly, the larger sample size allowed heritability estimates for many individual taxa to be calculated.

The taxon with the highest heritability was Christensenellaceae, a family within the Firmicutes that forms a co-occurrence consortium with other heritable taxa including the dominant human gut methanogen Methanobrevibacter smithii. MZ twins were previously shown to have greater concordance for the carriage of M. smithii compared to DZ twins (10). Studies across mammalian species (11) and within bovine lines (12) have also shown that host genetics influence levels of gut methanogens. Previous work in humans has also associated methanogens and species richness with leanness (13), and Christensenellaceae with low serum triglyceride levels (14). Transplant experiments of feces from an obese human donor lacking this consortium to germfree mice were conducted with and without addition of Christensenella minuta. Addition of C. minuta resulted in reduced adiposity in the recipient mice (9), suggesting that host genes may influence phenotype via control of microbiome components.

Cross-study comparisons

Quantitative measures of the microbiome constitute a novel complex trait in human genome-wide association studies (GWAS). Microbiome data are costly and cumbersome to generate for large numbers of subjects. So far, compared to the accepted norm in the GWAS field, sample sizes for microbiome GWA studies have been small, and as a result findings may be spurious. Until sample sizes increase and meta-analyses are conducted, cross-validation is important to cement confidence in the results. Comparisons can be made for human and mouse studies that have (i) estimated taxon heritabilities directly, or (ii) identified taxa linked to host genes in genetic association analyses (human GWAS or quantitative trait locus (QTL) analysis in mice and plants). Cross-kingdom comparisons with similar studies in other genetic models are also possible. The advantage of model systems is that environmental conditions are controlled, however the extent of the genetic variation examined is less than the variation segregating in outbred populations (15) (Figure 1).

Several of the same taxa have been estimated as heritable or linked to host genes in at least two human GWA or mouse QTL studies. The human-based studies include the UK Twins (9), the Human Microbiome Project (HMP) subjects (16), and the Hutterites (17); the mouse QTL studies include advanced intercrossed lines (18, 19), the Hybrid Mouse Diversity Panel (20), collaborative cross/diversity outbred mapping panels (21), and recombinant inbred strains (22). The majority of these heritable taxa belong to the phylum Firmicutes, while Bacteroidetes are generally not heritable. Technical issues preclude direct comparisons for some taxa, for instance, not all taxonomies presently include the Christensenellaceae. Also, heritability estimates tend to be higher on average for mice, most likely because environmental variability is controlled. Nevertheless, some of the same taxa are identified repeatedly across studies.

Specific taxa related to host genotype across multiple studies

Improving GWAS in humans requires large sample sizes

Despite a sample size of ~3000 subjects, the TwinsUK study still fails to deliver study-wide significance in genome-wide and microbiome-wide association tests. While tests focused on a few candidate host genes or candidate microbes have produced encouraging results, genome-wide tests provide the best opportunity for new discoveries. While meta-analyses of human GWA studies have included pooled sample sizes exceeding 350,000 (38), such aggregation across multiple disparate microbiome studies will present its own set of challenges. But it appears inescapable that improved power for genome-wide tests requires increasing the sample size, and doing so with a unified sampling, sequencing and analysis pipeline will require sample processing infrastructures that include roboticized sample handling. Studies of skin and oral microbiota, which are more easily obtained compared to gut microbiota, may be the first to reach that mark.

The microbiome as a complex trait

What is the best microbiome “trait” for GWAS? The microbiome is a complex, high-dimensional trait that can be described in myriad ways. Specific functional interactions likely underlie host gene-microbe associations, and 16S rRNA gene data, which are used in GWAS now, are often a poor proxy for function. This problem is similar to modeling BMI – it is easy to obtain measurements in large numbers, but it is not the best measure for adiposity. Metagenomes provide functional gene counts, but are costly to generate. When predicted from 16S rRNA data, they are unreliable for genes that are variable among the genomes of strains, which may be the best phenotypes to model, along with their gene expression patterns. As sequencing and computational technologies continue to evolve, the goal of obtaining gene-specific and even strain-specific counts may soon be achievable.

To date gut microbiome GWAS have relied on stool samples. Fecal microbiota are a mix of mucosally-associated microbes, mostly from the colon, and lumenal microbes (39). At least half of the cells may be dead, and many are enriched in stool compared to farther up the GI tract. Small intestinal microbiota are rare in stool, and important stomach microbes such as H. pylori may be undetectable. Thus, small-scale interactions between the microbiome and mucosal surfaces that underlie genetic associations may not be detectable from stool data. Studies using alternative sampling techniques, such as ingested capsules that capture small intestinal biopsies, might reveal associations lost in stool.