01.02-study_design_considerations.Rmd

# Study design considerations {#study-design-considerations}

:::: {.graybox}
The contents of this section have been extracted and modified from the article [Disentangling host–microbiota complexity through hologenomics](https://www.nature.com/articles/s41576-021-00421-0) published in ***Nature Reviews Genetics*** in 2022 by the authors of the **Holo-omics Workbook**.
:::

Holo-omic approaches can be used to understand how the combined features of hosts and microorganisms shape biological processes relevant for hosts (such as adaptation), for microorganisms (such as meta-community dynamics) or both [@Alberdi2022-ay].

Depending on the aims and features of the study system, holo-omics can be implemented using different study designs, model systems and techniques. This landscape of possibilities is shaped around five essential questions that need to be considered when designing and interpreting hologenomic studies, which relate to five core topics:

1. **[Hologenomic complexity](#hologenomic-complexity)**
2. **[Control of variables](#control-of-variables)**
3. **[Molecular resolution](#molecular-resolution)**
4. **[Spatiotemporal factors](#spatiotemporal-factors)**
5. **[Explanatory and response variables](#explanatory-and-response-variables)**

![](images/holo-omics_five_questions.png "Author: Antton Alberdi")

## Hologenomic complexity {#hologenomic-complexity}

:::: {.graybox}
The contents of this section have been extracted and modified from the article [Disentangling host–microbiota complexity through hologenomics](https://www.nature.com/articles/s41576-021-00421-0) published in ***Nature Reviews Genetics*** in 2022 by the authors of the **Holo-omics Workbook**.
:::

Hologenomic complexity can be broadly defined as the amount of information relevant to the study that the biological system under analysis contains and it can be decomposed into three major elements: host genomic, microbial metagenomic and environmental complexity [@Alberdi2022-ay]. Within each of these elements, two sources of complexity can be defined: the intrinsic complexity of the system under study, including host genome size and number of bacterial genomes, and the complexity introduced by the degree of difference between the organisms under comparison such as gene expression differences versus distinct genomes.

![](images/holo-omics_complexity.png "Author: Antton Alberdi")

:::: {.imagenote}
**Decomposition of hologenomic complexity.** **(a-c)** The design and interpretation of hologenomic studies depend on the host genomic (part a), microbial metagenomic (part b) and environmental (part c) complexity of the system under study. Within each axis of complexity, two types of gradients can be defined based on whether the features are intrinsic to the system or introduced by the researcher through the selection of groups under comparison. **(d)** Six examples of study systems with different levels of genomic, metagenomic and environmental complexity. **(e)** Three-dimensional representation of the complexity of the examples. The area of the plain represents the combined host genomic and microbial metagenomic complexity of the system, while the height represents the environmental complexity. The combined three-dimensional volume represents the overall hologenomic complexity of the system. HMP: Human Microbiome Project.
:::

## Control of variables {#control-of-variables}

:::: {.graybox}
The contents of this section have been extracted and modified from the article [Disentangling host–microbiota complexity through hologenomics](https://www.nature.com/articles/s41576-021-00421-0) published in ***Nature Reviews Genetics*** in 2022 by the authors of the **Holo-omics Workbook**.
:::

Controlling the complexity of hologenomic variables is essential for addressing specific research questions. Broadly speaking, the more detailed and mechanistic the question under study, the greater the required control. For instance, research on specific biomolecular processes using laboratory models will require a higher level of control than studying biogeographical patterns of host–microbiota interactions in wild organisms. The control of hologenomic variables can be achieved through a number of strategies.

### Controlling host genomes {#controlling-host-genomes}

The control over host genomic complexity largely depends on the model organisms studied and the technical approaches employed. In laboratory organisms that can reproduce asexually, such as water fleas (Daphnia, Crustacea) and Lamiaceae plants, absolute control over host genotypes can be achieved by using clonal organisms [@Mushegian2019-md]. When clones cannot be used, inbred laboratory animals can provide a high level of genomic homogeneity. The use of groups of genetically homogeneous hosts allows the effects of contrasting environmental conditions or specific microbial comunities to be compared. Clonal and inbred models also enable the effects of a specific host genetic factor to be studied in a controlled genomic background through the application of targeted techniques for modulating gene expression (such as RNA-mediated interference) or for genomic engineering (such as CRISPR–Cas9). Working with humans and wild organisms does not enable such a degree of control over the genotypes studied unless in vitro models, such as organ-on-a-chip co-cultures of animal tissues and microbial communities, are generated62. When this level of control is not possible, coarse control over host genotypes can be achieved through contrasting animals from different populations or from closely related species63, while greater control can be achieved through comparing individuals across different degrees of kinship, such as monozygotic versus dizygotic twins38, and family members to other individuals64.

### Controlling microbial metagenomes {#controlling-microbial-metagenomes}

Control over microbial metagenomic complexity is usually achieved through modulating microbial communities. Some strategies, such as modification of dietary regimes or the administration of microbiota-targeted additives or prebiotics, aim to modify microbial ecosystems by changing nutrient availability. However, unless compounds that match unique enzymatic capabilities of specific microorganisms are used, it is difficult to accurately modulate the microbiota owing to the complexity of ecological relationships among microorganisms. Alternative approaches to modify microbial communities include inoculation of target bacteria (such as probiotics) and faecal microbiota transplantation. The efficacy and accuracy of these methods is also variable; there is no guarantee that inoculated bacteria will establish or modulate the microbiota, while transplantation does not enable accurate control over the microbial community introduced or the secondary elements that are transplanted along with bacteria. These issues complicate interpretation of results; for example, bacteriophages transferred alongside bacteria may severely impact the gut microbiota composition. A higher level of control could potentially be achieved through transplanting synthetic microbial communities. While this approach has been successfully implemented in diverse in vitro setups the complexity of microbial communities still hinders its efficient use as a routine scientific procedure in live animals.

### Controlling the environment {#controlling-environment}
In most laboratory studies, environmental complexity is reduced so that no, or very few, environmental parameters (usually only experimental treatments) vary among groups and subjects. Climate chambers and aquaria enable experiments by providing absolute control of abiotic conditions, such as light/dark cycles, humidity and temperature variations. Outdoor common garden experiments do not provide full control over environmental factors, but they ensure the effect on the systems being compared is identical. Some natural systems can also provide special conditions that enable environmental features to be controlled, such as cuckoo nestlings that are bred by other birds or salmon populations that breed in the same rivers in alternating years. Research on wild organisms usually incorporates more complex and dynamic environmental conditions. When controlling them is not possible, collection of relevant environmental metadata to be incorporated as covariates in the statistical analyses is useful. A century of ecological research has revealed the advantages of each of these approaches. On the one extreme, laboratory microcosms allow the most reductive control. On the other extreme, studies in the macrocosm of the real world provide perspective on emergent properties of natural ecosystems that cannot be anticipated solely based on microcosms.

## Molecular resolution {#molecular-resolution}

:::: {.graybox}
The contents of this section have been extracted and modified from the article [Disentangling host–microbiota complexity through hologenomics](https://www.nature.com/articles/s41576-021-00421-0) published in ***Nature Reviews Genetics*** in 2022 by the authors of the **Holo-omics Workbook**.
:::

The complexity of a study system is not only determined by its inherent properties and study design, but also the techniques and procedures employed to analyse it. Researchers can decide how much a system is simplified by altering the resolution of the hologenomic features under study; in essence, zooming in or zooming out.

### Resolution of host genotypes {#host-genotype-resolution}

In host-microbiota studies, host genotypes can be defined at different levels, including species, breeds, populations, strains, sex or individuals. Genotypes can be defined as categorical variables, without analysing the differences between them, or can be studied in more detail through considering their actual genetic content and establishing correlations among them. When using an evolutionary perspective, phylogenetic relationships between genotypes are established based on phylogenomic markers, which usually vary above population and species level, but not among individuals. This implies that genomic variability among the individuals included within each genotype is overlooked. Studying the effect of interindividual genomic variability on host-microbiota systems, such as identifying candidate host genomic variants associated with microbial features, requires a higher level of resolution. This is achieved through defining genotypes at the individual level, and using techniques based on whole genome resequencing that enable the complexity of host genomes to be screened at a much finer level, so that differences between the individuals contrasted are not only defined based on their kinship, but also the functional properties of their genomic variants. Currently, this approach requires high quality reference genomes from which high density SNP profiles of individuals can be generated, for example through SNPchip or resequencing studies. The genomic resolution could be further refined by incorporating structural variants, methylation patterns, or even, we hypothesise, chromosome 3D folding structure as revealed through techniques such as Hi-C. In doing so, researchers can identify associations between SNPs or gene variants and specific microbiota traits, such as the relative abundance of certain taxa or the enrichment of a given function, and thus identify mechanisms by which a host exerts control over composition and function of its associated microbiota

### Resolution of microbial metagenotypes {- #microbial-metagenotype-resolution}

The structure and resolution at which microbial metagenotypes are defined also affects the complexity of the metagenome under analysis. Metagenotypes can be defined as arrays of microbial taxa, microbial genes or a combination of both. The most common approach to define them is to rely on short marker sequences targeted for metabarcoding purposes, such as the 16S rRNA or the internal transcribed spacer (ITS). However, these procedures often do not enable reliable taxonomic assignment at genus or species level, do not capture strain level community dynamics, and are prone to generate biased functional inferences, as bacteria with identical marker genes (particularly those associated with wild taxa) might carry very different catalogues of genes. Thus, while useful for estimating microbial diversity and obtaining preliminary insights into functionality, targeted sequencing approaches do not provide conclusive evidence about the metabolic capabilities of the microbiota, particularly when working with non-human systems.

By contrast, if appropriate strategies and adequate sequencing depths are employed, shotgun metagenomics enables bacterial genome sequences to be recovered, from which genes can be predicted and annotated to create a gene catalogue that can define a metagenotype. However, these genes are not randomly distributed, but enclosed within genomes of specific bacteria or other microorganisms, with a particular combination of genes that shape their expression and the specific biological features (such as oxygen affinity, reproduction time, metabolic capacity) that determine their ecology. Hence, a more refined characterisation of microbial metagenotypes can be achieved through binning algorithms that enable bacterial genome reconstruction from metagenomic mixtures, yielding metagenome-assembled genomes (MAGs). Nevertheless, unless short-read sequencing is combined with long-read approaches, it is challenging to capture multi-copy genes such as the 16S rRNA marker gene 103, which is often employed in metabarcoding studies and therefore represents a useful link to a large number of existing studies. Machine learning-based solutions to link 16S rRNA marker gene sequences with MAGs are, however, being developed 104. Finally, regardless of the approach used to define the microbial metagenotype, the complexity of microbial communities will often require dimensionality reduction to increase statistical power 105,106. This can be achieved by defining co-abundance clusters, ecological guilds or more complex strategies that also consider temporal features of microbiota variation, such as compositional tensor factorisation.

### Resolution of envirotypes {- #envirotype-resolution}

Characterisation of environmental factors that affect the host-microbiota system under study enable the definition of envirotypes, a term drawn from crop sciences that is useful for accounting for the environmental factors in the hologenomic context. Any different physical place, or place sampled at different time points, will be exposed to a different environment, as conditions will seldom be identical between two spatial and temporal points. Hence, the resolution at which the composite of environmental factors is considered will define whether these two environments will be considered different envirotypes or not. For example, if only considering water temperature, killer whales sampled in the Arctic and the Antarctic seas experience the same envirotype. However, if the biotic composition is also considered in the definition of the environment, the Arctic and the Antarctic will need to be split into two distinct envirotypes, as some killer whales will have access to penguins while others will not. The same principle applies to laboratory setups or mesocosm experiments: a temperature shift of 2-3 ºC might not be considered relevant under some experimental setups, while it can define different envirotypes under other study designs. Finally, failure to recognise environmental factors that affect host-microbiota interactions, and thus define relevant envirotypes, can lead to increased noise and decreased capacity to achieve statistical significance.

## Spatiotemporal factors {#spatiotemporal-factors}

:::: {.graybox}
The contents of this section have been extracted and modified from the article [Disentangling host–microbiota complexity through hologenomics](https://www.nature.com/articles/s41576-021-00421-0) published in ***Nature Reviews Genetics*** in 2022 by the authors of the **Holo-omics Workbook**.
:::

### Spatial factors {- #spatial-factors}

Spatial resolution. Microbial communities associated with animal and plant hosts vary not only across coarse body parts, but also at the micro-scale, such as between the lumen and the intestinal crypts. Thus, the resolution at which a body site is defined will also determine how a hologenomic system is characterised. For example, the animal gastrointestinal tract can be considered a single sampling unit, 4-5 units or hundreds of micro-units, depending on the sampling and data processing strategies employed. Naturally, each level of resolution will allow different questions to be addressed and will require the use of different technologies and analytical approaches.

### Temporal factors {- #temporal-factors}

Temporal features to be considered include when, how often, and for how long host-microbiota systems are to be analysed. When a host is first exposed to microbes with regard to temporal benchmarks (number of days or years) must be considered, as should the order in which it is exposed to them. Priority effects relate to how the order of species arrivals in an ecosystem shape the potential for subsequently arriving taxa to establish themselves. Although originally discussed at the macroorganismal level in the context of plant communities, the phenomenon is also relevant for building host-associated microorganism communities, for example as documented in the human gut. In addition, microbial communities are known to vary daily, seasonally and relative to life-stage patterns. Hence, the extent and frequency of sampling determine which of these dynamics will be observed or, conversely, missed. Finally, it is important to consider that the consequences of changes at one time period or life stage may appear only later in time, thus detection of such effects obviously requires that the subsequent period is also studied. For example, interventional animal experiments show that when the immune system develops early in life, there is a window of opportunity where the gut microbiota composition shapes the risk of developing diseases in the future.

## Explanatory and response variables {#explanatory-and-response-variables}

:::: {.graybox}
The contents of this section have been extracted and modified from the article [Disentangling host–microbiota complexity through hologenomics](https://www.nature.com/articles/s41576-021-00421-0) published in ***Nature Reviews Genetics*** in 2022 by the authors of the **Holo-omics Workbook**.
:::

Host genomic and microbial metagenomic data generated under hologenomic setups can take on different roles when generating statistical models. While the environment is most often considered as an explanatory variable (though one can also study how the hologenome affects the environment), the host genome and the microbial metagenome are sometimes viewed as explanatory and sometimes as response variables, depending on the aim of the research. In many cases, directionality is set by the researcher rather than the biological system itself, as host-microbiota systems contain many bi-directional interactions and circular processes, which complicate the establishment of causal relationships. Here, we define three basic models in which the three main variables (genome, metagenome and environment) are assigned different roles to address different types of fundamental questions.

![](images/hologenomics_response_variables.png "Author: Antton Alberdi")

:::: {.imagenote}
Examples of biological processes addressed by the different models of host-microbiota interactions. **a)** How does the hologenome shape animal phenotypes? Only the combination of specific host genomic (G) and microbial metagenomic (MG) features, probably developed due to a selective force exerted by the presence of predators (E) enables rough-skinned newts to have skin toxicity, an ecologically relevant phenotypic trait (P). **b)** How do the microbial metagenome and environment shape host genomic features? SCFA-producing bacteria along with a fibre-rich diet enhance chromatin accessibility and thus activate immune gene expression. **c)** How do the host genome and the environment shape microbial genomic features? Only the combination of a lactase nonpersister genotype combined with the milk-drinking envirotype generates a microbial metagenotype characterised by enrichment of Bifidobacterium.
:::

### Phenotype as a product of genotype, metagenotype and envirotype {- #p-g-m-e}

This is the main model used when hologenomics is conducted to ascertain how genome-metagenome-environment interactions affect the biological properties of a host, such as disease susceptibility, performance or fitness. It is an especially common and relevant model for health, agricultural, and ecological and evolutionary research 19,125–127. One clear example of a phenotype shaped by host genomic, microbial metagenomic and environmental factors was recently reported for rough-skinned newts. The study showed that bacteria on the skin of the newts produce a deadly neurotoxin from which the newt is protected by mutations in five host genes that encode the NaV channels normally targeted by the toxin. Thus, this ‘toxic newt’ phenotype is the result of both host and microbial genes, which likely evolved under the pressure exerted by an environmental factor, namely the presence of predators.

### Genotype expression influenced by metagenotype and envirotype {- #g-m-e}

When studying how core host genomic features, which contribute to shaping phenotypes, are affected by the microbiota, host genomic features become the response variable. Unlike the microbial metagenome, the genome sequence of the host organism is not variable, but microorganisms can induce chromatin remodelling and DNA methylation, and thus modulate the bioactivity of molecular receptors and host gene expression. A well-studied pathway that links the microbiota with host gene expression involves modulation of the activity of host histone deacetylases (HDAC) by short chain fatty acids (SCFA) produced by intestinal microorganisms. HDACs remove histone lysine acetyl groups, which leads to chromatin condensation and transcriptional silencing of genes. Increased SCFA concentrations inhibit histone deacetylases, thereby enhancing chromatin accessibility and activating gene expression. A metagenotype with a higher capacity to produce SCFAs combined with an envirotype characterised as a fibre-rich diet (required to produce SCFAs), therefore contributes to boost immune response through activating host immune gene expression.

### Metagenotype as a product of genotype and envirotype {- #m-g-e}

This model assumes the inverse causal directionality between the host genome and microbial metagenome to that described above. Candidate host genes related to microbiota features can be identified through GWAS in which the metagenotype (or derived metrics such as diversity or abundance of specific microbial taxa, genes or metabolic functions) are treated as a phenotypic trait. For instance, the increased abundance of lactose degrader Bifidobacteria in humans has been shown to be associated with lactase nonpersister genotype and consumption of milk (envirotype). Once candidate genes are known, targeted analyses in which natural or human-controlled genomic variability (such as the number of copies of the amylase-encoding gene in humans) can be contrasted under controlled environmental conditions to ascertain the effect on metagenotypes (such as the abundance of Ruminococcaceae bacteria in the gut microbiota).