We generally concur with the descriptions by Hooks et al. on microbiota-gut-brain (MGB) research: There are numerous literature reviews, scientific conclusions are often subjected to over-interpretation, and there is a lack of detail about the microbiome itself. The goal of the present commentary is to recommend advancements in experimental design that could enhance microbiome research.

The four top-cited papers considered by Hooks et al. used microbiome intervention to demonstrate a neurological effect or association. Two papers described a germ-free model (Heijtz et al. Reference Heijtz, Wang, Anuar, Qian, Björkholm, Samuelsson, Hibberd, Forssberg and Pettersson2011; Sudo et al. Reference Sudo, Chida, Aiba, Sonoda, Oyama, Yu, Kubo and Koga2004), whereas the other studies used a probiotic intervention (Bravo et al. Reference Bravo, Forsythe, Chew, Escaravage, Savignac, Dinan, Bienenstock and Cryan2011; Hsiao et al. Reference Hsiao, McBride, Hsien, Sharon, Hyde, McCue, Codelli, Chow, Reisman, Petrosino, Patterson and Mazmanian2013). These studies effectively demonstrated that the microbiome can cause changes in neurological function and behavior of the host. Although two of these studies provide information about microbiome composition, via culturing or 16S rRNA sequencing (Hsiao et al. Reference Hsiao, McBride, Hsien, Sharon, Hyde, McCue, Codelli, Chow, Reisman, Petrosino, Patterson and Mazmanian2013; Sudo et al. Reference Sudo, Chida, Aiba, Sonoda, Oyama, Yu, Kubo and Koga2004), this level of experimental design only provides general information about the composition and diversity of the microbiome, while failing to capture details about functional activities of this dynamic microcosm. The primary focus of these studies is to connect the microbiome to a specific physiological effect in the host. This is a critical first step in identifying whether a host symptom is affected by the microbiome but is insufficient to ascertain important details of the specific cause.

Deriving a more detailed understanding of how the metabolic activities of microbial communities affect host physiology requires an expanded experimental design. Many studies have demonstrated that microbial community composition varies across individuals but clusters as a function of age, geography, diet, co-habitation, and health (David et al. Reference David, Maurice, Carmody, Gootenberg, Button, Wolfe, Ling, Devlin, Varma, Fischbach, Biddinger, Dutton and Turnbaugh2014; Halfvarson et al. Reference Halfvarson, Brislawn, Lamendella, Vázquez-Baeza, Walters, Bramer, D'Amato, Bonfiglio, McDonald, Gonzalez, McClure, Dunklegarger, Knight and Jansson2017; The Human Microbiome Project Consortium 2012; Song et al. Reference Song, Lauber, Costello, Lozupone, Humphrey, Berg-Lyons, Caporaso, Knights, Clemente, Nakielny, Gordon, Fierer and Knight2013; Yatsunenko et al. Reference Yatsunenko, Rey, Manary, Trehan, Dominguez-Bello, Contreras, Magris, Hidalgo, Baldassano, Anokhin, Heath, Warner, Reeder, Kuczynski, Caporaso, Lozupone, Lauber, Clemente, Knights, Knight and Gordon2012). This implies that all these factors have an impact on the host's microbial composition, and that replicating the composition of the microbiome without exact replication of experimental conditions (e.g., individuals, laboratories, and diet) is very difficult. This is a formidable problem even with large sample sizes. This limitation suggests that there is less of a reproducibility crisis (Sze & Schloss Reference Sze and Schloss2016) and more of a need to design more elaborate and appropriate experiments (Knight et al. Reference Knight, Vrbanac, Taylor, Aksenov, Callewaert, Debelius, Gonzalez, Kosciolek, McCall, McDonald, Melnik, Morton, Navas, Quinn, Sanders, Swafford, Thompson, Tripathi, Xu, Zaneveld, Zhu, Caporaso and Dorrestein2018).

There is now published evidence that despite significant variation in microbial composition, broad biological function appears to remain similar (The Human Microbiome Project Consortium 2012). Hence, microbiome research will be greatly propelled by experimental designs that combine intervention studies with systems-biology techniques that seek to characterize overall microbiome function. Currently, there are four “-omics” techniques that capture “detailed” molecular-level information about a microbial community: whole community genome (metagenome) sequencing (The Human Microbiome Project Consortium 2012), metatranscriptome sequencing (Jorth et al. Reference Jorth, Turner, Gumus, Nizam, Buduneli and Whiteley2014), metaproteome identification (Erickson et al. Reference Erickson, Cantarel, Lamendella, Darzi, Mongodin, Pan, Shah, Halfvarson, Tysk, Henrissat, Raes, Verberkmoes, Fraser, Hettich and Jansson2012), and metabolome characterization (Karu et al. Reference Karu, Deng, Slae, Guo, Sajed, Huynh, Wine and Wishart2018). Each of these techniques has advantages and limitations. However, combination of these techniques should lead to high-resolution portrayal of microbial function and, potentially, microbial metabolic reconstruction at the community level.

Whole metagenomic sequencing provides a backbone for investigating microbiome function via creation of a composite gene catalog. Initially, reference sequence databases were used to characterize the taxonomy and function of a microbial community (The Human Microbiome Project Consortium 2012; Stewart et al. Reference Stewart, Ajami, O'Brien, Hutchinson, Smith, Wong, Ross, Lloyd, Doddapaneni, Metcalf, Muzny, Gibbs, Vatanen, Huttenhower, Xavier, Rewers, Hagopian, Toppari, Ziegler, She, Akolkar, Lernmark, Hyoty, Vehik, Krischer and Petrosino2018) but were limited by the range of known reference genomes and functional ortholog databases. Advancements in the understanding of genomic features allow combining tetranucleotide frequency with sequence coverage to bin the results from metagenomic assemblies into metagenome assembled genomes (MAGs) de novo (Kang et al. Reference Kang, Froula, Egan and Wange2015). This approach provides a set of low- to high-quality draft genomes, which can provide subspecies and strain-resolved information about the gene repertoire of a microbial community (Brown et al. Reference Brown, Xiong, Olm, Thomas, Baker, Firek, Morowitz, Hettich and Banfield2018). Although metagenomic sequencing allows categorizing a set of genes into their respective taxonomy and predicting their function, metagenomics is limited to only predicting potential function because it contributes no information about actual translation of genes into proteins.

Although the laboratory technique behind metatransciptomics is more similar to metagenomic sequencing, metatranscriptomics and metaproteomics explore a similar question: Which predicted genes are being transcribed and subsequently translated into proteins? Transcriptome sequencing identifies genes that are being transcribed into RNA and generally produces more total identifications, whereas liquid chromatography tandem mass spectrometry (LC-MS/MS)–based metaproteomics ultimately identifies proteins, thus providing confirmation that a metagenomics predicted gene is translated into protein (Erickson et al. Reference Erickson, Cantarel, Lamendella, Darzi, Mongodin, Pan, Shah, Halfvarson, Tysk, Henrissat, Raes, Verberkmoes, Fraser, Hettich and Jansson2012; Jorth et al. Reference Jorth, Turner, Gumus, Nizam, Buduneli and Whiteley2014). Although the number of identified proteins tends to be fewer than transcripts, it is unclear how many proteins microbes actually use in a community. Strides have been made by integrating MAGs with metaproteomes in lower-diversity communities, but much work remains to answer this question (Xiong et al. Reference Xiong, Brown, Morowitz, Banfield and Hettich2017). Theoretically, metaproteomics should provide a better connection to phenotype because functional products of gene expression are actually measured.

Gas (GC) and liquid (LC) chromatography paired with mass spectrometry provide an approach for detecting metabolites within microbial communities (Karu et al. Reference Karu, Deng, Slae, Guo, Sajed, Huynh, Wine and Wishart2018). Although these molecules are difficult to connect to their organism of origin without complementary gene information, metabolomics provides a framework to evaluate metabolic hypotheses derived from genetic or protein information. Enzyme-to-compound databases, such as Kyoto Encyclopedia of Genes and Genomes (KEGG) (Kanehisa et al. Reference Kanehisa, Sato, Kawashima, Furumichi and Tanabe2016) and MetaCyc (Caspi et al. Reference Caspi, Billington, Fulcher, Keseler, Kothari, Krummenacker, Latendresse, Midford, Ong, Ong, Paley, Subhraveti and Karp2018), can annotate recognized genes with an enzyme identifier, thereby connecting specific taxa, genes, and proteins to specific molecular functions at the compound level. Ultimately, these features can create a framework by which metabolomic data can be integrated with microbiome data.

Using constrained experimental conditions combined with specific hypothesis-driven intervention, researchers can join the power of cause-and-effect experimental design with integrated “-omics” to fully characterize the microbiome under different interventions. Time-course studies would be ideal because the subject can function as its own internal control. Eventually, as the community of investigators transitions into the next phase of microbiome research, general intervention-based studies and broad characterization of microbial composition at the population level will no longer be sufficient to provide novel insights into the MGB axis. Holistic metabolic modeling through systems-biology measurements of individual microbiomes combined with intervention-based experimental design provides a path forward.