Mouse Microbiota Models: Comparing Germ-Free Mice and Antibiotics Treatment as Tools for Modifying Gut Bacteria

As the intestinal microbiota has become better appreciated as necessary for maintenance of physiologic homeostasis and also as a modulator of disease processes, there has been a corresponding increase in manipulation of the microbiota in mouse models. While germ-free mouse models are generally considered to be the gold standard for studies of the microbiota, many investigators turn to antibiotics treatment models as a rapid, inexpensive, and accessible alternative. Here we describe and compare these two approaches, detailing advantages and disadvantages to both. Further, we detail what is known about the effects of antibiotics treatment on cell populations, cytokines, and organs, and clarify how this compares to germ-free models. Finally, we briefly describe recent findings regarding microbiota regulation of infectious diseases and other immunologic challenges by the microbiota, and highlight important future directions and considerations for the use of antibiotics treatment in manipulation of the microbiota.


INTRODUCTION
Over the past several decades, there has been a dramatic increase in both scientific and popular interest in the effects of the intestinal microbiota on human health. The microbiota, consisting of the bacteria, viruses, fungi, and archaea that inhabit different niches in the human body, has been implicated in regulation of inflammatory, infectious and metabolic diseases, and appears to play a critical role in potentially causing, propagating, or preventing human illnesses (Lai et al., 2014;Norman et al., 2014;Palm et al., 2015). With the surge of enthusiasm to understand this new and massively complex factor in human health has come the need to effectively model it. Specifically, the development of small animal models of the microbiota permits testing of subsets of the microbiota as causative vs. correlative factors in disease states, as well as offering a system to uncover putative therapeutics.
Two main methods have emerged to explore the effects of the microbiota on physiology and disease in mice: germ-free models and antibiotics treatment regimens. Both approaches have strengths and weaknesses. Here we will discuss commonly used regimens and methods to deplete the microbiota, the effects of these approaches on host physiology including cellular composition, signaling pathways, and organ function, and briefly describe what has been found using these two different methods to model the effects of the microbiota on human disease.

GERM-FREE AND ANTIBIOTICS TREATMENT MODELS
Germ-free mice are bred in isolators which fully block exposure to microorganisms, with the intent of keeping them free of detectable bacteria, viruses, and eukaryotic microbes. Initially conceptualized by Louis Pasteur in 1885, colonies of germfree rodents were first established in the 1940s (Yi and Li, 2012;Al-Asmakh and Zadjali, 2015). Germ-free mice allow for study of the complete absence of microbes or for the generation of gnotobiotic animals exclusively colonized by known microbes. However, generating and maintaining these mice requires specialized facilities, and the cost, labor, and skills required to maintain them can make these models inaccessible to many researchers. Germ-free mice must be monitored regularly for contamination using a combination of culturing, microscopy, serology, gross morphology, and sequencing-based detection techniques (Fontaine et al., 2015;Nicklas et al., 2015). For example, Charles River, one common germ-free vendor, routinely uses a serologic assay to test for viral contamination including murine norovirus, mouse rotavirus, and mouse cytomegalovirus; PCR (both 16S and pathogenspecific), microscopy, and culturing to test for bacteria; and gross examination of animals to test for parasites (Charles River Germ-Free Mouse Report). Additionally, any unique mouse strain to be studied under germ-free conditions must be rederived in these facilities, and this limits the number of different genotypes that are feasible to study. Further, maintenance of mice in isolators may make it impractical or challenging to conduct some studies (for example, behavioral testing or pathogen infections).
An alternate method that has emerged to avoid some of these complications has been the use of antibiotics treatment (Figure 1). Treatment with broad-spectrum antibiotics is commonly used to deplete the gut microbiota of mice, and can be readily applied to any genotype or condition of mouse. Unlike germ-free conditions, under which complete sterility is maintained throughout life, antibiotics can deplete bacterial populations in mice which were normally colonized since birth. Germ-free animals are broadly impaired in many aspects of development and early immune education, whereas antibiotics treatment in adult mice specifically allows for study of the role of bacteria in maintaining cell functionality and signaling pathways after development. Alternatively, some studies deliver antibiotics in drinking water to pregnant dams to limit maternal transfer of microbes and then maintain the cage on the regimen during weaning to study the effects of bacterial depletion early in development (Lamousé-Smith et al., 2011;Deshmukh et al., 2014;Gonzalez-Perez et al., 2016;Li et al., 2017).
Due to differences in mechanism of action, antibiotics can selectively deplete different members of the microbiota. For example, metronidazole and clindamycin both target anaerobes, vancomycin is only effective against gram-positive bacteria, and polymyxin B specifically targets gram-negative bacteria (Atarashi et al., 2011;Schubert et al., 2015). Individual antibiotics can be used to shift the composition of the gut microbiota in order to identify classes of bacteria relevant to different phenotypes (Schubert et al., 2015;Zackular et al., 2016). In contrast, a cocktail of different classes of antibiotics can be used to broadly deplete the gut microbiota. Researchers have used various regimens which differ in antibiotic combination, dose, and length of treatment ( Table 1). All of these combinations broadly target Gram-positive, Gram-negative, and anaerobic bacteria. Often, antibiotics are diluted in drinking water and mice are allowed to drink ad libitum throughout the course of treatment; therefore, actual delivered doses can vary. Some protocols additionally include antifungals in the cocktail to avoid fungal overgrowth during treatment (Reikvam et al., 2011;Grasa et al., 2015;Zákostelská et al., 2016). Many also add sweeteners such as sugar, Splenda, or Kool-aid to mask any bitterness and ensure mice drink the antibiotics-containing water Baldridge et al., 2015;Emal et al., 2017). However, there are reports of mice avoiding water and becoming dehydrated when provided antibiotics in this manner (Hill et al., 2010;Reikvam et al., 2011;Zákostelská et al., 2016). Daily oral gavage can prevent dehydration and allow delivery of a precise dose of antibiotics, so this method is sometimes used alone or in combination with delivery in drinking water, though it is more labor-intensive (Kuss et al., 2011;Reikvam et al., 2011).
Validation of bacterial depletion can be performed with culture-based methods by assessing the colony-forming units (CFUs) from fecal samples plated in aerobic and/or anaerobic conditions on non-selective media. However, this method only accounts for cultivatable microbes. Quantitative PCR of the gene encoding 16S rRNA allows for culture-independent assessment of gastrointestinal bacterial load. Broad-spectrum antibiotics treatment can decrease bacterial load by multiple orders of magnitude in 2 weeks of treatment or less (Baldridge et al., 2015;Gonzalez-Perez et al., 2016;Brown et al., 2017). Both germfree and antibiotics-treated mice allow for the introduction of microbes in which the contributions of defined bacterial species or consortia can be studied (Tan et al., 2016;Staley et al., 2017). Although some phenotypes seen with antibiotics treatment are attributed to removal of a single bacterial species, many differences that occur are due to broad decreases in bacterial load. Consistent with this, treating bacterially-depleted mice with conserved pattern recognition receptor ligands such as flagellin (Ichinohe et al., 2011;Oh et al., 2014) or CpG dinucleotides (Ichinohe et al., 2011;Hill et al., 2012) can restore some defects, even in the absence of microbiota restoration.
Although most studies attribute phenotypic differences after antibiotics treatment to the depletion of gut microbes, some studies have assessed how regimens affect commensal populations at other sites. Oral antibiotic regimens can decrease culturable bacteria in the respiratory tract (Ichinohe et al., 2011;Abt et al., 2012;Brown et al., 2017) and the vagina of mice (Oh et al., 2016), but do not affect skin bacterial communities (Naik et al., 2012). Importantly, though rarely quantified, antibiotics treatment also likely drastically affects bacteriophage populations, though there is debate in the literature about whether phage play important roles in transfer of antibiotic resistance genes (Modi et al., 2013;Enault et al., 2017;Górska et al., 2018). Antibiotics treatment can allow for the outgrowth of commensal fungal species, potentially confounding results as FIGURE 1 | Comparison of the advantages and disadvantages of germ-free and antibiotics-treated mouse model systems.
these organisms can alter immune function, hence the inclusion of antifungals in some antibiotics treatment regimens (Noverr et al., 2004;Kim et al., 2014). An increasing appreciation for important roles for the virome and mycobiome may lead to enhanced interrogation of these effects in the future, as well as the potential impact of antibiotics on eukaryotic viruses and archaea (Norman et al., 2014). An important final potential disadvantage of antibiotics treatment can be the evolution or development of antibiotic-resistant bacteria and their subsequent selection and outgrowth in mouse intestines Morgun et al., 2015). Depending upon the starting bacterial composition, antibiotic cocktail, duration of treatment, and phenotypic readout, antibiotic-resistance may confound findings in experiments, especially if resistant bacteria are present in only a subset of tested mice. Longitudinal analysis of bacterial populations in all experimental groups can aid in detection of resistance and analysis of whether resistant bacteria may affect experimental results.
Mice on antibiotics are not completely cleared of bacteria, but significant reductions in bacterial load are associated with shifts in cell populations, signaling pathways, and organ morphology, with results often paralleling what is seen in germ-free mice.

EFFECTS OF THE MICROBIOTA ON CELL POPULATIONS AND CYTOKINES
Although many aspects of murine physiology are affected by microbial populations, the effects of antibiotics treatment on immune cell populations are some of the most wellstudied (Figure 2). The immune system constantly responds to both pathogenic and commensal microbial populations, and shifts after antibiotics treatment reflect the dependence of cell populations and function on bacterial signals. While results of cell composition analysis are not uniform across studies, we will describe the most prominent and consistent observations (see Table 2 for exact findings by different groups).

MYELOID CELLS
Innate immune cells lack antigen-specific receptors, responding instead to broadly conserved microbial patterns. As innate cells continuously interface with the microbial populations constituting the microbiota, sensing of these microbes via pattern recognition receptors is essential for maintenance of normal host physiology (Chu and Mazmanian, 2013;Fawkner-Corbett et al., 2017).
bone marrow, and peripheral tissues (Zhang et al., 2015;Hergott et al., 2016;Ekmekciu et al., 2017) and others reporting no significant differences at baseline or after infection Oh et al., 2016;Brown et al., 2017;Robak et al., 2018). Even when cell numbers are similar, macrophages are often less mature after antibiotics treatment, impairing their responses to    Ivanov et al., 2008;Li et al., 2017 Similar in skin Similar or ↑ in SI ↓ IL-17+ in SI, skin, liver Bandeira et al., 1990;Ivanov et al., 2008;Ismail et al., 2011;Naik et al., 2012;Li et al., 2017 Less activated and more apoptotic in liver, ↓ production of antimicrobials in SI Ismail et al., 2011;Li et al., 2017 ↓ production of antimicrobials in SI, less activated in liver, diminished response to mucosal injury in colon Ismail et al., 2009Ismail et al., , 2011  pathogens Brown et al., 2017;Emal et al., 2017). Bulk granulocytes decrease in the bone marrow of antibioticstreated mice, though their numbers in peripheral blood are similar (Balmer et al., 2014;Josefsdottir et al., 2017). Neutrophils decrease in bone marrow and in peripheral sites, with an increased rate of apoptosis and decrease in aged neutrophils in the bloodstream after microbiota depletion (Deshmukh et al., 2014;Zhang et al., 2015;Hergott et al., 2016). However, neutrophil populations are not diminished at the site of infection after pathogen exposure in antibiotics-treated mice Oh et al., 2016;Brown et al., 2017;Robak et al., 2018). In contrast to neutrophils, the proliferation of basophil precursors in the bone marrow is increased after antibiotics treatment, associated with increased basophils in the periphery and an enhanced response to allergen exposure (Hill et al., 2012). Similarly, eosinophils in various tissues are enhanced at baseline (Suárez-Zamorano et al., 2015;Oh et al., 2016), and in the lung after inhaled allergen exposure (Hill et al., 2012). These alterations in granulocyte populations are consistent with a shift from type 1 to type 2 immune responses after depletion of the commensal microbiota.
Various dendritic cell subsets are also reduced after antibiotics treatment at both mucosal and systemic sites (Ichinohe et al., 2011;Ekmekciu et al., 2017;Thackray et al., 2018), although these differences may not be apparent after infection Oh et al., 2016). Differences in dendritic cell numbers have not been reported in germ-free mice, though impairment in priming has been seen (Walton et al., 2006;Ganal et al., 2012).
Reductions in innate immune cell number and function, characteristic of both germ-free mice and antibiotics-treated mice, may be explained by diminished cytokine and chemokine levels, which are necessary for normal cell recruitment, differentiation, and functionality (Mortha et al., 2014;Brown et al., 2017). Reductions in myeloid populations are likely not explained by decreases in progenitor populations: although antibiotic exposure beginning in utero can reduce postnatal granulocytosis (Deshmukh et al., 2014), treatment in adult mice does not reduce myeloid progenitor populations in the bone marrow (Josefsdottir et al., 2017;Thackray et al., 2018).

LYMPHOID CELLS
In contrast to what is seen with myeloid progenitors, common lymphoid progenitors are reduced in the bone marrow after microbiota depletion (Josefsdottir et al., 2017;Thackray et al., 2018), consistent with what is seen in some, but not all, germfree models (Balmer et al., 2014;Iwamura et al., 2017). Total lymphocytes are similarly reduced in the peripheral blood after antibiotics treatment (Josefsdottir et al., 2017).
Similar to CD4+ T cells, cytotoxic CD8+ T cells generally decrease in the intestine after antibiotics, though results at other sites are more varied (Ochoa-Repáraz et al., 2009;Kernbauer et al., 2014;Ekmekciu et al., 2017;Josefsdottir et al., 2017;Thackray et al., 2018). Proinflammatory cytokine production from cytotoxic T cells is not diminished at baseline after antibiotics treatment, but has been reported to decrease in response to infection with some pathogens (Ichinohe et al., 2011;Abt et al., 2012;Kernbauer et al., 2014;Oh et al., 2016).
Reports of B cell population shifts are varied, with some groups seeing declines in the blood, bone marrow, and tissues after antibiotics treatment, and others noting similar numbers regardless of microbiota depletion (Ochoa-Repáraz et al., 2009;Yoshiya et al., 2011;Zhang et al., 2015;Ekmekciu et al., 2017;Josefsdottir et al., 2017;Li et al., 2017;Thackray et al., 2018). Likewise, shifts in antibody responses are inconsistent-in general, total IgG and IgM levels remain similar in different sites analyzed, but secretory and serum IgA levels tend to decrease and serum IgE levels increase after microbiota depletion (Hill et al., 2010;Oh et al., 2014;Uchiyama et al., 2014;Adami et al., 2018;Lynn et al., 2018;Robak et al., 2018). Antigen-specific response to infection or vaccination vary by pathogen, mouse age, and time point after exposure analyzed, but neonates in particular generally produce a less robust response to vaccination after exposure to antibiotics (Ichinohe et al., 2011;Lamousé-Smith et al., 2011;Abt et al., 2012;Diehl et al., 2013;Oh et al., 2014;Uchiyama et al., 2014;Li et al., 2017;Lynn et al., 2018). (Constantinides, 2018). Double-positive CD4+CD8αα+ T cells serve regulatory functions in the small intestinal epithelium and are diminished in antibiotics-treated mice, associated with a reduction in the bacterium Lactobacillus reuteri which induces this cell type (Cervantes-Barragan et al., 2017). γδ T cells are present in epithelial tissues, mediating tissue repair and monitoring microbial populations. Although studies in germ-free mice suggest that the microbiota is dispensable for these cells to home to the intestine or skin (Bandeira et al., 1990;Ismail et al., 2011;Naik et al., 2012), both germ-free and antibiotics-treated models indicate that microbial colonization is necessary for normal activation and production of antimicrobial compounds by these

Broad-spectrum antibiotics
References Germ-Free References

CYTOKINES
Although shifts in cytokine levels after antibiotics treatment are variable (Table 3), most studies that report differences describe a shift away from proinflammatory cytokines. Many associate microbiota depletion with decreases in the production of IL-1 family cytokines, Th1 cytokines such as IFNγ and TNFα, and IL-17 family cytokines. The production of these cytokines is often similar or reduced specifically in the gastrointestinal tract in naïve animals, but the diminished response becomes apparent after challenge with a pathogen, often at the site of infection. For example, although IL-1 family cytokine levels are similar, pro-IL-1 and pro-IL-18 are reduced in the vaginal mucosa and lung in antibiotics-treated mice at baseline, associated with a reduced production of IL-1 family cytokines in response to infection at each site (Ichinohe et al., 2011;Abt et al., 2012;Oh et al., 2016;Robak et al., 2018). Similarly, type 1 cytokines such as IFNγ and TNFα are generally present at similar levels in tissues after antibiotics treatment in uninfected mice but reduced at the site of infection in microbiota-depleted mice Oh et al., 2016;Robak et al., 2018). Inflammatory Th17 cytokines such as IL-17 and IL-22 are generally reduced in the intestines even at baseline in antibiotics-treated mice (Hill et al., 2010;Deshmukh et al., 2014;Ekmekciu et al., 2017), whereas differences in tissues such as the lung become apparent after infection at that site (Deshmukh et al., 2014;Suárez-Zamorano et al., 2015). In parallel with this decrease in inflammatory cytokines, some reports suggest that there is an increase in the expression of Th2 family cytokines such as IL-4, IL-5, and IL-13, especially after allergen exposure, consistent with a shift from Th1 to Th2-type immunity after microbiota depletion (Hill et al., 2010;Suárez-Zamorano et al., 2015;Oh et al., 2016). Broadly, there is reasonable concordance between germ-free and antibiotics treatment mouse models in alterations of cellular compartments and cytokines. However, different groups have reported disparate findings with different antibiotics treatment regimens, making it challenging to definitively categorize microbiota-mediated modulatory effects. We propose that distinct starting microbiota composition and distinct regimens likely underlie this variability, and highlight this as an area of much-needed standardization.

MICROBIOTA EFFECTS AT THE ORGAN LEVEL
In addition to shifts in cell populations and signaling pathways, antibiotics treatment has been seen to affect organ morphology more broadly, both in the gastrointestinal tract as well as in extra-intestinal organs ( Table 4). As the bulk of commensals reside in the gastrointestinal tract where they assist with digestion and interact closely with epithelial cells, it is not surprising that many changes are seen in intestinal physiology after microbial depletion. The length of the whole intestine or the colon is not affected, but the cecum becomes dramatically larger, transit time increases, and fecal pellet frequency and consistency can be altered (Grasa et al., 2015;Suárez-Zamorano et al., 2015;Ge et al., 2017). Moreover, villi become narrower (Kernbauer et al., 2014), cellular proliferation decreases (Reikvam et al., 2011;Ekmekciu et al., 2017), and features such as tuft cells (Wilen et al., 2018) or goblet-cell antigen passages (Knoop et al., 2015) are affected in specific regions of the gastrointestinal tract. Immune function in the intestines is also affected, as the production of antimicrobial peptides is reduced (Brandl et al., 2008;Vaishnava et al., 2008;Kinnebrew et al., 2010;Reikvam et al., 2011), Paneth cells granules are diminished (Kernbauer et al., 2014), Peyer's patches become less abundant and decrease in cellularity (Reikvam et al., 2011;Grasa et al., 2015;Hashiguchi et al., 2015), expression of Tolllike receptors is altered (Grasa et al., 2015), and tolerance to the commensal intestinal microbiota is impaired .
Non-gastrointestinal organs also depend on ongoing bacterial signals to maintain normal morphology. The spleen (Ochoa-Repáraz et al., 2009;Reikvam et al., 2011;Yoshiya et al., 2011;Zhang et al., 2015;Josefsdottir et al., 2017;Thackray et al., 2018), thymus (Josefsdottir et al., 2017), and lymph nodes (Ichinohe et al., 2011;Durand et al., 2018) may decrease in size and/or cellularity after antibiotics treatment. Liver regeneration is impaired in antibiotics-treated mice (Wu et al., 2015) and bile acid synthesis is altered (Sayin et al., 2013;Zhang et al., 2014). Additionally, fat pads diminish and bone mass increases, consistent with a role for the microbiota in maintaining normal body composition (Suárez-Zamorano et al., 2015;Yan et al., 2016). There have been a number of intriguing studies recently exploring the role of the microbiota in regulating brain function and behavior via the gut-brain axis; this complex topic has been recently reviewed elsewhere (Cryan and Dinan, 2012;Liu and Zhu, 2018).

MICROBIOTA REGULATION OF IMMUNE CHALLENGES
As might be expected, given the systemic and tissue-specific differences in immune function, antibiotics-treated mice are more susceptible to a variety of pathogens. For example, microbiota-depleted mice are more susceptible to bacterial pathogens such as vancomycin-resistant enterococcus, Salmonella, and Clostridium difficile in the gastrointestinal tract (Kinnebrew et al., 2010;Fernández-Santoscoy et al., 2015;Theriot et al., 2016), a variety of pneumonia-causing bacteria in the respiratory tract (Brown et al., 2017;Robak et al., 2018), and systemic Escherichia coli (Deshmukh et al., 2014). Additionally, after antibiotics treatment, mice are impaired in their response to vaginal HSV-2 (Oh et al., 2016), flaviviruses (Thackray et al., 2018), influenza (Ichinohe et al., 2011;Abt et al., 2012), and cutaneous Leishmania (Naik et al., 2012). However, microbiotadepleted mice actually become less susceptible to enteric viral pathogens including murine norovirus and poliovirus (Kuss et al., 2011;Uchiyama et al., 2014;Baldridge et al., 2015), possibly due to direct interactions between viral pathogens and enteric bacteria or due to loss of specific cell types required for viral infection. Antibiotics-treated mice are additionally impaired in their development of tolerance to food antigens (Bashir et al., 2004;Kim et al., 2018) and are more prone to allergic diseases (Hill et al., 2012;Adami et al., 2018).

CONSIDERATIONS FOR THE FUTURE
While antibiotics treatment offers an inexpensive and accessible alternative to germ-free models, results obtained using these regimens come with the caveats of potential off-target drug effects and incomplete or inconsistent ablation of microbes. Additionally, because so many groups use distinct treatment regimens and mouse microbial populations may be institutionspecific, antibiotics studies are much more challenging to compare than germ-free mouse studies. We suggest that some standardization of antibiotics treatment regimens would be helpful; for example, if a standard cocktail were employed to demonstrate an initial finding, this could be compared to other studies, and subsequent follow-up experiments could be done with modified cocktails as necessary. Additionally, we suggest that at least a limited assessment of the replicability of findings in antibiotics-treated mice and germfree mice would be of high value for most studies, to rule in or out potential off-target drug effects or developmental differences between germ-free and standard pathogen-free mice that may be important for a phenotype. Finally, it will be critical for investigators to ensure that microbial loads are consistently monitored in both antibiotics treatment and germ-free models to identify any effects of contaminants or antibiotic-resistant microbes.
Ensuring that we are able to interpret the contribution of an individual study to the field of microbiota research will require careful planning and execution of these experiments on the part of investigators. As we continue to uncover additional health and disease states in which the microbiota plays a role, the use of these models will become increasingly common.

AUTHOR CONTRIBUTIONS
EK and MB wrote and edited the manuscript. KK edited the manuscript.