Altered Microbiota Composition Mediates Depressive BehaviorDuring Chronic Stress

Depressive disorders often run in families. In
addition to the genetic component, this may point to the microbiome
as a causative agent. In my thesis work, I employed a combination
of behavioral, molecular and computational techniques to test the
role of the microbiota in mediating stress-induced despair
behavior, as an animal model for depression. In chronically
stressed mice displaying despair behavior, we found that the
microbiota composition and the metabolic signature dramatically
change. Specifically, we observed reduced Lactobacillus and
increased circulating kynurenine levels as the most prominent
changes in stressed mice. Restoring intestinal Lactobacillus levels
was sufficient to improve the metabolic alterations and behavioral
abnormalities. Mechanistically, we identified that
Lactobacillus-derived reactive oxygen species may suppress host
kynurenine metabolism, by inhibiting the expression of the
metabolizing enzyme, IDO1, in the intestine. Moreover, maintaining
elevated kynurenine levels during Lactobacillus supplementation
diminished the treatment benefits. Collectively, our data provide a
mechanistic scenario for how a microbiota player (Lactobacillus)
may contribute to regulating metabolism and resilience during
stress.

I would first like to recognize my first adviser, Jony, for his infectious passion for progressive science. Jony drew me into his group by presenting me with a question I had never thought of before and throughout the years I have seem him pose such questions many times. To dream of things that never were and then try to pursue them takes real talent and dedication. I hope some of it rubbed off on me. I am grateful to Jony for gathering such amazing individuals as I have been working with in the many iterations of the lab and allowing me to follow my own path among them. Jony provided great opportunities during my PhD, including meeting amazing scientists and support for attending many great conferences. Perhaps most importantly, I am grateful for the sharp questions and timely criticism, which not only made me a more thorough scientist, but also taught me to continuously consider the relevance of the questions I ask and the experiments I perform.
Also first, I ought to recognize my second (acquired) adviser, Alban. One day, I went to Alban for help troubleshooting a sensitive PCR reaction and from then on, the trips to his office increased logarithmically. Alban's constant optimism and excitement for doing science is inspiring, particularly during the hardest times. His ability to break complex questions down to simple ones has helped me to both plan better experiments and analyze more critically. Alban has a wealth of technical tricks that he pulls from up his sleeve just when needed and an amazing talent to find in 1-2 minutes information I spent hours looking for. Above all, I am grateful to Alban for his dedication, availability, and constant willingness to think and plan experiments and ideas with me.

Several friends in the lab have been instrumental in my training as a scientist. Big thanks go to: Noel Derecki for teaching me many many techniques as well as always asking me what the hypotheses behind my experiments were; Sachin
Gadani for all the joint efforts early on in the lab and many "Sunday" experiments; Jim Cronk for teaching me the importance of counting how many time I triturate or mix any cell suspension; Antoine Louveau for a great example of a tasteful approach to science. To every member of the Kipnis and Gaultier labs: thank you for your honest feedback, questions and sharing your inspiration. It's been an honor to learn from and along with each of you. Thank you to my college research adviser, Lisa Gabel, for her continuous support and guidance. To everyone in the lab and in the greater BIG Center: thank you for the laughs, collegiality and willingness to help.
A PhD is not just a time of scientific growth. By the nature of the scientific method, as well as our imperfect approaches, it is a time riddled with both highs and lows. This rollercoaster ride thus ends up also being a feat of mental strength and a period of personal growth (hopefully!). So I must also acknowledge everyone who supported me on this front. Friends from high-school (Sinzi and Elena) and college (Nafis, Joaquin, Pooja, Brook, Hannah, Farah, Farai, Elliott, Ahsan, Basu) who stood to remind me of my strengths, beliefs, and abilities -thank you for the talks, the visits, the messages. Friends in Charlottesville -thank you for climbing mountains (hi Irene), sharing Thanksgiving meals, and being my community. Sachin, thanks for making bugs a little less scary, for letting me use your car for the longest time, and for picking up the phone when I was walking late at night. A, thank you for teaching me so many things about gender and pharmacology! Antoine, thank you for all the coffee trips, that one time you cooked me mussels, and naming your T cell "Boo". And thank you, Anthony! I love you the most. Thank you for your love, support, and our wonderful family.
I purposefully saved my family for last, because they are the most important. My mother likes an approximate quote she heard: "many generations before you have prayed so you could achieve what you achieved." I know it to be true. I am the first in my family to get a college degree, but I have been fortunate to have many of my small family teach me about perseverance and reaching higher. My grandparents have been examples of hard work, humanity, and legacy. I am grateful for giving me strong roots. Above all, however, I wouldn't have been anywhere near where I am today without my mother. She has this extraordinary ability of balancing between pushing me further and giving me confidence with very careful timing, to yield positive results. She taught me to be kind, curious, and perseverant. Though  3. How can peripheral kynurenine modulate brain activity? 47 Depression is one of the most prevalent mental disorders, with a lifetime prevalence of 16%, in the United States, and is often accompanied by other mental disorders, such as anxiety (almost 60%) and substance abuse (almost 25%) 4 . Whether presence of such psychiatric comorbidities warrants classification of more specific types of depression has not yet been addressed by either clinicians or scientists.

Modeling depression in animals
With depression being a mental health disorder, diagnosed solely based on patientreported symptoms, it is unsurprisingly difficult to study in humans. Not only are we limited in the experimental tools available, but also the subjective nature of the disease makes it difficult to obtain reproducible results. While using laboratory animals can overcome both of these problems, their main downside is their validity in completely modeling depressive behavior.
Three main experimental paradigms are accepted as tests for depressive behavior: the forced swim test, the tail suspension test, and the sucrose preference test. The first two assays entail putting an animal in an uncomfortable situation and measuring how long it struggles to escape and/or when it ceases to struggle, with the expectation that an animal exhibiting depressive behavior will struggle less. These tests aim to measure human equivalents of despair and, since the introduction of the forced swim test in 1977 5 , both or either test have been used to quantify depressive behavior, as well as to screen candidate antidepressant drugs. The forced swim test is, by far, the most used test for depressive behavior, due to its increased reliability compared to the tail suspension test 6 . The sucrose preference test entails presenting an animal with a choice between regular and sweetened water and measuring its preference for the sweetened choice. The test aims to measure anhedonia, one of the core symptoms of depression, with the expectation that an animal exhibiting depressive behavior will have a decreased preference for sucrose. However, the reliability of the test is very low, with most animal models not showing decrease in their sucrose preference (perhaps due to the salience of the sweetened water) 6,7 . This often leads researchers to resort to modified versions of the test, such as the sucrose consumption test, which just measures the intake of sweet water after a period of fasting. Moreover, considering that sucrose intake is a food-based behavior, its validity for measuring anhedonia can also be questioned. Many depression patients will exhibit a lack of engagement in hobbies and pleasurable activities, but may actually overeat or binge on pleasurable foods [1][2][3] . In addition to such tests that are accepted measures of depressive behavior, researchers often supplement their behavior testing with anxiety assays (such as the elevated plus maze, novelty suppressed feeding, marble burying, etc) based on the high comorbidity of anxiety and depressive disorders 1,4,6 .
In terms of the animal models exhibiting depressive behavior, these are generally based on environmental manipulations, the most popular of which are stress-based paradigms 6,8 .
The most widely used assays include the chronic unpredictable stress, chronic restraint stress, repeated social defeat, and learned helplessness after unescapable foot shock. All of these models and their variants are based on the observation that stressful life events are a risk factor for developing depression 1,9 . This basis on repeated stress gives them limited validity, nevertheless superior to older models such as olfactory bulbectomy or corticosterone injections.
Another valuable model is that of selective breeding of animals displaying high depressive behavior. While this paradigm aims to model the importance of a genetic predisposition for depression, it is unfortunately less widely used due to the expenses and the time requirement associated with breeding and screening outbred animals 6,10 .

Neurobiology of depression
Most scientists and clinicians would now agree that depression is ultimately a product of altered brain function, caused by defective circuitry and/ or chemical imbalances. This view is actually relatively young, gaining popularity following initial studies with early antidepressant drugs in the 1950s. At the time, the prevalent hypothesis was that depression "emerged from intrapsychic conditions and conflicts," the therapies of choice being psychotherapy and electric shock therapy 11 . In this context, perhaps unsurprisingly, the first antidepressant drugs, iproniazid and imipramine, were discovered accidentally 11 . Iproniazid was initially used to treat tuberculosis, while imipramine was being tested for schizophrenia, when clinicians observed their mood enhancing effects [11][12][13] .

The monoamine deficiency theory
Not only did the subsequent approval of antidepressant drugs revolutionize treatment, but also further studies on their mechanism of action provided the first neurobiological theory of depression: the monoamine deficiency theory, which remains one of the fundamental theories of depression 9,11,14 . This theory postulates that depression is caused by a general deficit in monoamine neurotransmitters: serotonin, dopamine, norepinephrine. Interestingly, this theory was developed at a time when serotonin was mostly known for its peripheral effects, such as in intestinal function and blood pressure control, while its roles in the brain were still speculative 15,16 . Yet, the success of these early antidepressants led to rapid expansion of the tricyclic family (TCAs), followed by the selective serotonin reuptake inhibitors (SSRIs, starting in the 1970s), and other atypical neurotransmitters, all aimed at boosting amounts of brain amine neurotransmitters (serotonin, dopamine, norepinephrine) 11,14 . There is no doubt that antidepressant compounds can affect brain function and influence behavior in both positive (mood elevation, anxiolysis) and negative (many peripheral side effects) ways. However, whether they correct an underlying dysregulation is debatable and now beginning to be challenged 14,17,18 . In fact, evidence that depressed patients have a deficiency in brain levels of any of the monoamines is very scarce. Studies trying to reduce amine neurotransmitter levels by acute depletion of their amino acid precursors report a fast induction of negative affect only in patients with a history of depression 19 . Moreover, chronic tryptophan depletion in laboratory animals has yielded inconsistent results regarding their depressive phenotype 20,21,22 .
In fact, many knockout animal models of enzymes involved in aminergic transmission show a depressive-resistant phenotype, rather than a pro-depressive phenotype (reviewed by Cryan and Mombereau 6 ). Finally, animal studies investigating a circuitry of depression (discussed in more detail below), are now showing that both increases and decreases in the levels of released amine neurotransmitters can be observed in different brain regions in animal models of depression, indicating that it is the circuit activity that regulates neurotransmitter balance 23-25 .
With these accumulating data, and considering the rapid raise in amine levels after antidepressant treatment, it is now believed that the delayed efficacy of the drugs is due to a progressive change in neuronal circuits due to sustained elevated levels of amines, rather than to correcting underlying deficits 14,26,27 . This new view would also account for the high relapse rates after treatment cessation.

The excitatory synapse theory
Unfortunately, while this drug-directed approach to seeking mechanisms of depression has proven unfruitful, scientists are still advocating for its pursuit, now based on an even broader acting drug -ketamine 28 . Ketamine is an NMDA receptor antagonist that was developed in the 1960s and soon approved for its use as an anesthetic 29,30 . In the past two decades, the drug also started receiving attention for its rapid antidepressant effects after a single administration, albeit with a transient effect [30][31][32] . Similarly to the amine-boosting antidepressant drugs, following this promising clinical success, researchers are now studying the molecular effects of ketamine administration with a view to understand mechanisms of depression 33-35 . Not only are these studies focusing on the acute effects of a single ketamine administration, but also to expect that the effects of a drug targeting a receptor expressed almost ubiquitously in the brain will unveil underlying causes of depression is unrealistic.
Moreover, the main mechanistic readouts of ketamine effects focus on levels of BDNF. Brainderived neurotrophic factor (BDNF) is a favorite among researchers as a target-mechanism for altered neuronal processes. Early studies in the function of the molecule showed that BDNF signaling leads to increase neuronal activity via increased synaptic release and spine size 36 indicate that glucocorticoid responses after stress are probably beneficial, and their balance in the brain is necessary for adaptive behaviors. Moreover, similar to the weak evidence that cortisol levels are higher in depressed patients, the evidence is also weak that corticosterone levels are elevated in chronically stressed mice 41 . In fact, studies show that the corticosterone release is the highest after the first episode of stress, and lower with every subsequent stress episode 45,63 . Given the fact that depressive behavior develops after several stress episodes, it may even be possible that the reduced corticosterone release (rather than its increase) is what contributes to depressive behavior.

The circuitry of depression
Perhaps the most recent theory of depression is that abnormal activity in a specific set of neuronal circuits yields the behavioral outputs characteristic of depression 14,28,64 . The appeal of this hypothesis is that it makes sense to explain the underlying cause of depression as a standalone disorder, versus a global molecular dysregulation, which would have an array of functional manifestations, one of which would be depression. The circuitry of depression is based largely on animal studies, as clinical research is limited by the available techniques.
Imaging techniques have yielded areas with different activity in depressed patients, including the prefrontal cortex and all areas of the limbic system 9 . However, such studies cannot inform about how the identified areas are wired to cause depression. Fortunately, not only are studies in laboratory animals are making progress in resolving a circuitry for depression, but also their findings largely align with the human observations 9,14,65 .
Using animal models involving different types of stress protocols, as well as animals bred for congenic depressive phenotypes, studies identified deficits in reward circuitry to be associated with depressive-like behavior. Specifically, increased activity in the lateral habenula (LHb) increases activity in the ventral tegmental area (VTA), potentiating projections to the nucleus accumbens medium spiny neurons and weakening projections to the medial prefrontal cortex (mPFC) 65 . Evidence is now accumulating that the habenula is acting as a "disappointment" center of the brain, with neurons in this region being excited in the absence of reward and inhibited in rewarding situations 66-69 . Animal models displaying depressive behavior also show increased activation of lateral habenular neurons, in turn feeding to excite the VTA [70][71][72] . Moreover, experiments involving deep brain stimulation of the habenula in both rats and humans were able to reduce neuronal activity and improve depressive symptoms 69,70 .
From the lateral habenula, excitatory projections potentiate the activity of dopaminergic neurons in the VTA 70 . In an elegant study, Krishnan 74 . On the other hand, under stress conditions, which already increase neuronal activity in this region, inhibiting the VTA is beneficial, while stimulating it induces depressive-like behavior in previously resilient mice 24,25 . Interestingly, though the VTA projects to areas across the brain, its projections to the nucleus accumbens seem to drive much of the depressive and resilient phenotypes, with selective stimulation of these projections during stress making the difference between the two behavioral outcomes 23,24 .
While the past decade has shown great progress in unveiling parts of the circuitry of depression, we are still mostly understanding how mono-synaptic circuits are altered in depression animal models and whether manipulating them has any causative effects, either pro-depressive or anti-depressive. Being able to integrate these mono-synaptic circuits into a fuller "depression diagram" will be an important advance in identifying whether there is a true starting point that could trigger an altered depression circuit, or whether depression is due to generalized deficits throughout the brain, with specific modulation of certain circuits due to environmental factors (i.e. stress). For instance, hypothesizing that sustained activity in the habenula is what triggers a downstream "circuit of depression", is this sustained activity due solely to environmental inputs -such as repeated negative reinforcement? Or could there be a generalized susceptibility in the brain (such as genetic, metabolic etc), which manifests itself through depressive behavior when further challenged (i.e. with stress). Perhaps this second explanation would be more likely to account for the difference between mice resilient and susceptible to depressive-like behavior. Considering presumed identical genetics, rearing and exposure to stress, what would cause the selective upregulation of synaptic maintenance molecules in a subset of animals, leading to their resilient phenotype?

Susceptibility to depression
Unfortunately, the etiology of depression remains very elusive, with many possible theories and contributors. Given that the disorder is diagnosed purely based on reported symptoms, it also very likely that different underlying causes may lead to similar symptoms.
Under these circumstances, one cannot talk about causes of depression, but rather discuss risk factors associated with the disorder.
The strongest association with depressive disorders is, without doubt, genetic, with a calculated heritability of depression of 37-66% 9,75 . Despite this strong association, GWAS studies have had limited success in identifying susceptibility genes, leading to the hypothesis of complex genetic interactions accounting for an increased sensitivity to environmental factors 9 . Gender is another risk factor, with women having a 70% higher lifetime prevalence of depression 4 . Though mechanisms for this increased susceptibility have been only sparsely investigated, this higher prevalence is attributed to both genetic and socioeconomic factors 76 .
In recent years, inflammation has been a factor more and more associated with depression. While the DSM-5 excludes patients with inflammatory disorders from being diagnosed with depressive disorders, the manual also acknowledges that many patients suffering of chronic inflammation experience depression symptoms 1 . For instance, up to half of the patients suffering from multiple sclerosis experience depression 77,78 . The exclusion of such patients from being diagnosed with depressive disorder is due to the fact that sustained inflammation induces an array of CNS symptoms in addition to depression, i.e. cognitive impairment, increased pain sensitivity, decreased locomotion [79][80][81] . Therefore, while inflammation may induce depression symptoms, the underlying neurobiology may be different from that of non-inflammatory unipolar depression. Elevated levels of cytokines, both acutely and chronically, have been extensively documented to affect synaptic transmission. I have previously reviewed such neuro-immune interactions extensively 82,83 , attached in Annex A.
Environmental factors are also thought to be critical for susceptibility to depression 1,9 . Early life trauma or abuse, as well as a stressful lifestyle are considered the highest environmental risk factors for developing depressive disorders 1,4,9 . In addition, factors such as exercise and metabolism have recently been receiving a lot of attention for the roles they might play in modulating depressive symptoms [84][85][86][87] . Exercise has been shown to have many beneficial effects for depression outcomes in patients, with pre-clinical studies indicating that exercise can improve many aspects of physiology, from metabolism to adult neurogenesis 84,88,89 . On the other hand, in a nutrition and obesity-conscious society, nutritional imbalances are increasingly suspected of mediating or modulating depressive behaviors [85][86][87]90 . Mechanisms of how metabolism can affect brain function are just starting to be investigated 87,91,92 . At the same time, evidence has been mounting for a new player, the gut microbiota, as a key regulator of both metabolism and the immune system, and thus with the ability to affect the function of many organs, including the brain 93-96 .

An unusual candidate -the gut microbiota
The main function of the gastrointestinal tract is digestion. Essentially a very long tube, the gut receives food, breaks it down mechanically, then enzymatically and through fermentation, absorbing nutrients and water along the way. These functions are accomplished in distinct segments of the tract, each with their own cellular specializations and accessory organs: the pancreas, liver and gall bladder for the upper GI tract, and the gut microbiota for the lower tract. The gut also possesses its own nervous system -the enteric nervous system, that develops independently from the central and peripheral nervous systems, and which is responsible for intestinal motility. Moreover, it receives extensive sympathetic and parasympathetic innervation, which controls intestinal secretions and motility [97][98][99][100] . While digestion in the upper tract is mostly mechanical and enzymatic, in the lower tract, bacterial fermentation is necessary for finishing the process, particularly for breaking down complex carbohydrates through fermentation, as well as for synthesizing micronutrients such as B vitamins and vitamin K 101,102 . Though we have known about the bacterial symbioses in the intestine for a long time, it has only been recently, with the advances in sequencing technology, that we have begun to gain an appreciation for the complexity and importance of the gut microbiota beyond its function in digestion 103,104 .
During in utero development the fetus is sterile and bacterial colonization begins at birth when the fetus acquires a large proportion of the mother's flora 101,105 . Diet is one of the major regulators of the intestinal microbiota, and so the newborn flora changes with the introduction of new foods, as well as post weaning 101,105 . At the same time, the genetic makeup of the host also plays a role in shaping the microbiota composition [106][107][108] . While the composition of the microbiota is highly dynamic throughout life and dependent on external and internal factors, it is largely dominated by two bacteria phyla -the anaerobes Bacteriodetes and Firmicutes.
The gram negative Bacteriodetes are primarily starch and polysaccharide degraders, whereas the gram positive Firmicutes are primarily proteolytic and amino acid fermenters 101 95,102 . The presence of the abundant microbiota (up to 10 12 organisms per cm 3 ) is a byproduct of the slow transit through the large intestine, combined with the permissive pH (not as acidic as the stomach).

Human vs mouse microbiome
As with many areas of biomedical science, animal studies have been instrumental in furthering our understanding of the microbiome development and function. Fortunately, the same dominant phyla are also present in mice, the laboratory model of choice in the microbiome field. Moreover, an effort comparing the microbiota composition across mouse strains to that of humans found that, functionally, the core human and mouse microbiomes are more than 95% similar. Yet, the two species only share 4% of microbial genomic sequences, indicating large differences in the bacterial strains they harbor 108 . So far, researchers have been embracing the results found in mice, particularly considering that most microbiota studies do not have the power to taxonomically identify microbial strains, or even species. However, lowering the costs of high throughput metagenomic sequencing experiments will no doubt allow us to uncover specific functions of different strains.
Another important aspect of microbiome research is the widespread use of germ-free (GF) animals. These are animals reared in a sterile environment, as opposed to specific pathogen free (SPF), with no bacterial or fungal colonization. Studying germ-free animals has helped us appreciate the importance of the microbiome and the dangers of widespread antibiotic treatment. This model has also allowed scientists to selective colonize them with humanderived microbiota and further understand its function in ways that human studies do not allow [109][110][111] . For instance, in one of the first studies of this kind, Smith and colleagues showed that it is poor diet-associated microbiota that drives growth stunting 110 . However, the boon of germfree mice may also lead to many misleading conclusions. Considering the many contributions of the microbiome to host physiology, missing it completely can be thought of equivalent to missing a kidney or the spleen. Though the organism can function without it, all the other systems will be affected, often by secondary effects, potentially leading to over-interpretation regarding the direct functions of the organ. Moreover, in studies involving "humanizedmicrobiome" mice, it is likely that the genetic differences and restricted diet of the mice alters the human inoculum they receive 109,112 . Unfortunately, most studies do not check for such differences, drawing conclusions that may not actually pertain to the initial hypothesis 109 . It is, therefore, important to exercise caution when using such reductionist animal models 112 .

Contributions of the gut microbiota to host physiology
As previously discussed, the main contribution of the gut microbiota is to host metabolism 95,102 . Commensal bacteria not only help break down carbohydrates, but also, through their metabolism, release modulatory molecules. Perhaps the most widely studied compounds are short chain fatty acids, which are absorbed by the host and further impact the metabolic output of the liver 95 . Another direct role of the microbiota is in preventing pathogenic bacterial infections 113 . While a part of this function is accomplished by the simple competition for space and resources, another important mechanism is by the production of antimicrobial compounds that bacteria themselves synthesize and secrete 113 . Finally, colonization of the intestinal tract is now accepted as a requirement for the maturation and function of the immune system 96,[114][115][116] . In addition to its importance for pathogen defense, this interaction further impacts the entire organism, as it can modulate clinical outcomes such as allergy, autoimmunity or immunodeficiency.

Microbiota-immune interactions
The relationship between the intestinal microbiota and the immune system is a very intimate one. The immune system is necessary for maintaining the microbiota, and the microbiota allows the immune system to mature and function normally 96  On the other hand, the presence and composition of the microbiota is necessary for normal immune system function. While the fact that germ-free mice have impaired immune function has been appreciated for years 122 , it has only been recently that significant studies started accumulating 96 . Works from the group of Dennis Kasper show that not only is intestinal colonization important, but also the composition of the microbiota is important for proper maturation of the immune system 123,124 . The best characterized phenotype in germ free mice is their decreased numbers of intestinal (and possibly systemic) T cells and the impairment of the T cells in maturing and acquiring memory phenotypes 123,124 . Perhaps as a compensatory mechanism, these mice have a much larger population of colonic natural killer T cells (NKTs), which predisposes them to pathology 125 . Interestingly, even though mouse, rat, and human microbiota successfully colonize the intestines upon fecal transfer, only the mouse flora can rescue the immune system deficits 124 . Two species of bacteria have been identified separately to be able to mediate this rescue: segmented filamentous bacteria, SFB, 124 and B. fragilis 123 .
Both bacteria have been shown to not only increase T cell numbers, but also restore inflammatory Th17 T cells and FoxP3+ regulatory T cells (Tregs), respectively 124,126,127 . In addition, members of the Clostridia genus were shown by another group to be potent inducers of regulatory T cells 128,129 . These are, of course, only a few examples of specific members of the microbiota identified in the past decade to be important for regulating immune function.
of microglia, the brain resident macrophages 116 . While the current literature barely scratches the surface, the bidirectional relationship between the gut microbiota and the immune system suddenly seems evident and opens doors to a lot of exciting investigations.

Ripples of microbiome effects on the brain
Through its secreted metabolites and its immunomodulatory factors, the microbiome can affect the function of virtually every organ in the body 94,95 . In terms of interactions between the gut microbiota and the brain, "gut feelings" has been a topic of many review and The fact that the gut microbiota is necessary for normal behavior comes from studies investigating the reductionist germ free or antibiotic-treated animals, which have pointed several behavioral and physiological differences compared to their SPF counterparts.
Specifically, germ free animals display increased activity and reduced anxiety behaviors 111,[140][141][142] , while also showing an exacerbated ACTH and corticosterone release in response to acute stress 143,144 . Germ free mice also show a deficit in social preference compared to SPF raised mice 145 . Interestingly, all of these deficits can be recovered by colonization with complete flora before or shortly after weaning time (4 weeks), but not once the mice reach adulthood (8 weeks and after) 143,145 . Partial recovery in some behaviors can even be achieved by colonization with a single or a subset of bacteria 143 . In terms of brain physiology, several deficits have been observed in germ free animals, including increased blood brain barrier permeability, altered microglia activity and, more molecularly, decreased levels of BDNF in various brain regions 116,140,143,144,146 . While such studies indicate that the microbiome is necessary to establish normal brain function, they lack the mechanistic insights into what products of the microbiome regulate specific aspects of brain function. Our hypothesis was that chronic stress induces changes in intestinal microbiota composition, which, in turn, alter metabolism or the immune system and feed back on the brain to sustain the development of maladaptive depressive behaviors. We tackled this hypothesis with the following specific aims: Specific aim #1: To identify the changes in microbiota composition following chronic stress and what mediates them.
Specific aim #2: To determine whether the depressive phenotype may be associated with the microbiota composition.
Specific aim #3: To distinguish contributions of microbiota-affected metabolites and peripheral immune system in mediating the communication between the gut microbiota and the brain.

Chronic antibiotic treatment induces depressive behavior
In order to investigate whether general microbiome dysregulation plays a direct role in mediating depressive behavior we first used antibiotic-mediated disruption of the microbiota.
While broad-spectrum antibiotic treatment does not fully deplete the microbiota, it does create a severe dysbiosis, making it the second best option to using germ free animals (unavailable to us). Treatment with a broad spectrum antibiotic cocktail for four weeks caused a significant reduction in bacterial load and diversity, as shown by 16s rRNA sequence analysis ( Fig. 1a-b).
Interestingly, antibiotic-treated mice developed spontaneous depressive behavior, as measured by the forced swim test (Fig. 1c). The assay measures the amount of time an animal struggles to escape an uncomfortable situation, a behavior typically affected in most models of depression and corrected by anti-depressant treatment. To further verify that this was a microbial-dependent effects and rule out a direct effect of antibiotics on the host, mice were treated with the antibiotic cocktail for two weeks and then allowed to recover for 8 weeks.
Despite the cessation of antibiotic administration, bacterial load remained reduced (Fig. 1d) and the animals continued to display depressive behavior (Fig. 1e). These findings support our hypothesis that intestinal dysbiosis may contribute to depressive behavior.

Microbiota composition is altered by unpredictable chronic mild stress
To determine whether chronic stress can directly affect the microbiota, we chose the appropriate due to the length and variety of the stress protocol (Fig. 2a). Consistent with previous reports, this protocol effectively induced despair behavior, as measured by the forced swim ( Fig. 2b) 14, 155,156 . We verified that the test results were measures of true despair behavior, as the animals showed normal activity and locomotion in the open field test (Fig. 2 c-d).
In order to assess the changes in microbiota composition that occur during chronic stress, we performed 16S rRNA sequencing on genomic DNA isolated from the fecal samples of naïve and stressed mice. The quantity of bacteria in fecal pellets was not affected by stress, as demonstrated by 16S qPCR (Fig. 2e). In terms of microbiota composition, principal coordinate analysis shows distinct clustering between samples from naïve and stressed mice, indicating a significant difference between the groups (Fig. 2f). A more in-depth taxonomic analysis of bacterial types revealed several changes in the microbiota composition ( Fig. 2g(i) shows one experimental cohort, Fig. 2g(ii) shows a different experimental cohort; bacterial classes are shown for ease of visualization). In our sequencing runs we observed between 14 and 29 significantly different genera between the naïve and stressed conditions. The variability in the starting microbiota (of naïve mice) and its changes (after stress) is not unexpected, as different shipments of mice, even from the same vendor, can have different microbiota compositions 157,158 .
Overall, the most conserved microbiota change across all independent experiments was a decrease in bacillus class members (both Lactobacillus and Turicibacter) in stressed mice (Fig. 2g).
Due to the abundance of literature linking Lactobacilli and behavior and the lack of studies and tools regarding Turicibacter species, we further focused on Lactobacillus as a confident potential player in the despair phenotype.
We verified the net loss of Lactobacillus by qPCR (Fig. 3a) and selective fecal sample cultures (Fig. 3b). These results demonstrate that chronic stress disturbs the microbiota homeostasis, in particular by decreasing the Lactobacillus levels. Correlation analysis returned a positive correlation between the relative Lactobacillus load and the escape behavior displayed by a mouse (Fig. 3c). Our observation was not limited to C57BL/6J, as BALB/cJ (a different inbred strain from Jackson Labs) and C57BL/6N (the same genetic strain from Taconic Farms) mice also show significant correlation between Lactobacillus levels and their escape behavior (Fig. 3d).
Interestingly, C57BL/6N mice had very low starting levels of Lactobacillus, which corresponded to low escape behavior even in the absence of stress. Our data are in agreement with recent studies showing associations between lower Lactobacillus levels and stress 149,151 .
Increased intestinal motility, rather than the immune system, may lead to the microbiota dysregulation To gain insight into potential causes for the changed microbiota composition, we further characterized intestinal immunity and physiology. Surprisingly, intestinal immunity was largely unaltered (Fig. 4a and data not shown), with the exception of a reduction in the Th17 population ( Fig. 4b-c). To test whether the Th17 deficit could lead to dysbiosis and depression, we sought a mouse model with increased levels of Th17 cells, hypothesizing that it would be protected against depressive behavior. Mice supplied from Taconic Farms are known to have increased type 3 immunity, due to their colonization with segmented filamentous bacteria (SFB) 126 . We compared C57BL/6J mice obtained from Jackson (Jax) to C57BL/6N mice obtained from Taconic (Tac) in their response to UCMS. Tac mice displayed depressive behavior at baseline, which was further exacerbated by UCMS (Fig. 4d). When we compared the Lactobacillus load in the two groups, we observed that Tac mice had significantly reduced levels of the bacteria (Fig. 4e) despite their high Th17 levels. This indicated that the relative abundance of Lactobacilli, rather than the Th17, correlates with the depressive phenotype. To further substantiate this interaction, we attempted to boost Th17 levels in Jackson mice by colonizing them with SFB (Jax+SFB, Fig. 4f). Despite elevated Th17 levels, stressed Jax+SFB mice remained more resilient than their Tac counterparts, while also maintaining higher Lactobacillus levels ( Fig. 4g-h). This led us to conclude that it is the Lactobacillus load and not the Th17 levels that correlates with depressive behavior.
To test our hypothesis that the immune system may contribute to the microbiota dysregulation, we subjected SCID mice (lacking adaptive immunity) to UCMS. SCID mice developed UCMS induced depressive behavior to a greater extent than immuno-competent animals (Fig. 5a). To our surprise, though their starting microbiota profile was different from their wildtype counterparts, SCID mice also displayed a significant decrease in Lactobacillus levels after stress (Fig. 5b).  These data indicate that, while the adaptive immune system may have an important role in mediating resiliency, the chronic stress induced decrease in Lactobacillus is independent of the adaptive immune system. Moreover, supplementing stressed mice with cultured Lactobacillus replenishes their Th17 levels ( Fig. 5c-d), indicating that the immune changes observed after stress are likely a product of the altered microbiota.
Given our results, the immune system is the unlikely mediator of the Lactobacillus deficit observed after chronic stress. We therefore turned to our alternative hypothesis, of altered intestinal physiology mediating the microbiota changes. Similarly to previous reports using stress models 159,160 , intestinal motility was significantly increased in the stressed animals (Fig.   6a). Furthermore, we observed an increase in the total size and cellular content of the stressed small intestines (Fig. 6b-c). These changes in intestinal physiology in response to stress may underlie the differences in microbiota composition. To more directly test this hypothesis, we treated naïve animals with mineral oil, a non-absorptive lubricant laxative. Increasing intestinal motility in this fashion led to a significant reduction in Lactobacillus levels, as well as in the escape behavior of the treated mice. These results are preliminary and a more thorough analysis of the microbiota composition, as well as of the behavior would paint a clearer picture. It is likely that Lactobacillus is not the only bacterial genus affected by laxative treatment, though its nutritional requirements do render it especially dependent to a stable environment ( [161][162][163][164] , further expanded in the discussion).
Nevertheless, these results indicate that fluctuations in intestinal motility can play a role in maintaining the composition of the microbiota.

Treatment with a Lactobacillus species ameliorates despair behavior by restoring kynurenine metabolism
To assess whether Lactobacillus levels may play a role in mediating despair behaviors, we attempted to replenish the levels of the bacteria and then measure escape behavior. To this end, we subjected mice to the UCMS protocol for three weeks and then supplemented their diet with live cultures of L. reuteri, while continuing the stress protocol for additional 3 weeks (Fig. 7a). Lactobacillus reuteri (ATCC 23272) is a species that colonizes several vertebrate hosts, including rodents and humans and has been previously used as a probiotic in mice 165 . This regimen indeed elevated Lactobacillus levels in stressed mice (Fig. 7b). Moreover, in our experimental paradigm, L. reuteri supplementation ameliorated the despair behavior induced by UCMS (Fig. 7c), indicating that Lactobacillus levels may be mediating, at least in part, the behavior.
To get an insight into the mechanism of Lactobacillus-supported resiliency, we performed untargeted metabolomics analysis of serum samples to identify if and how metabolites composition was altered after chronic stress. Principal component analysis showed that the metabolic profile of stressed mice is significantly different from that of naïve mice (Fig. 7di).
Treatment with L. reuteri converted the metabolic profile of stressed mice to an intermediate profile, suggesting that some of the stress associated metabolic alterations may be a consequence of decreased Lactobacillus levels (Fig. 7dii). While several molecules were significantly different in our analysis (234 out of 4900 spectra), most returned spectra were not confidently matched to known molecules due to limitations in the metabolite library.
Of the identified compounds, we mined those increased in stressed animals and normalized by L. reuteri treatment. Among them, metabolites in the tryptophan-kynurenine pathway presented this pattern (Fig. 7e). Evidence of dysregulation of this pathway in depressed patients, as well as its recently described role in the initiation of despair behaviors 84 , have made the tryptophankynurenine pathway particularly compelling as part of the potential mechanism mediating the microbiome effects on behavior. Intriguingly, a recent study 166 showed that Lactobacillus can directly modulate kynurenine metabolism, by inhibiting the pathway initiating enzyme IDO1 via production of reactive oxygen species (i.e. H2O2, Fig. 8a). IDO1 is the main enzyme metabolizing L-tryptophan to kynurenine outside of the liver (where TDO is the primary enzyme) 167 .
We verified that cultured L. reuteri produced a significant amount of H2O2 in vitro, when compared to E.coli (Fig. 8b). We next measured peroxide levels in the fecal contents of the stressed mice and discovered that H2O2 levels were decreased in the stressed mice; in agreement with our hypothesis, therapeutic administration of L. reuteri significantly raised the level of H2O2 in vivo (Fig. 8c). Moreover, we observed a significant correlation with the drop in Lactobacillus levels and the levels of H2O2 (Fig. 8d). We further verified the kynurenine pathway dysregulation by probing for ido1 mRNA in the intestine, as well as the brain. Our results show increased ido1 expression in the intestines after stress, while it remained almost undetectable in the brain (Fig. 8e). To further investigate the causative role of kynurenine metabolism in mediating despair behavior we treated naïve mice with L-kynurenine daily (Fig. 8f) and observed a significant reduction in escape behavior at the end of the 4-week protocol (Fig. 8g). Moreover, in order to show that the benefit of Lactobacillus supplementation is by reducing kynurenine levels, we treated stressed mice simultaneously with L.reuteri and L-kynurenine, expecting kynurenine to bypass the benefits of L.reuteri supplementation.
Indeed, while L. reuteri alone increased the escape behavior of stressed mice, kynurenine administration abrogated the beneficial effect of L.reuteri (Fig. 8h), even with elevated levels of Lactobacillus and H2O2 (Fig. 8i-j).
We further attempted to solidify our results by blocking the IDO1-kynurenine pathway.
In a preliminary experiment, mice were subjected to the UCMS protocol for 4 weeks, after which they received an inhibitor of IDO1 (1-methyl-tryptophan, 1-Me-Trp), while continued being stressed (Fig. 9a). When tested in the forced swim test, all animals showed reduced escape behavior at the 4-week timepoint. However, after 1-Me-Trp administration, the treatment group showed a trend towards improved escape behavior (p=0.17, Fig. 9b).
Though preliminary, the lack of a larger effect is likely due to the very low dose of 1-Me-Trp used. Further literature search revealed that doses of up to 400 mg/kg can be administered without toxic effects 168,169 . This result complements the rest of data in supporting a role for elevated kynurenine metabolism in mediating depressive behavior.
While so far the data point to a strong association between intestinal Lactobacilli and resilience, it does not show a causal relationship. While selective depletion of Lactobacilli is not feasible with current resources, colonization of germ-free mice with a single or a select set of bacteria has been accepted in the microbiome field. In the absence of a germ-free facility, we attempted to re-colonize antibiotic treated mice with L. reuteri, using a Bacteroides (a genus not changed in stressed animals) species as a control. We first verified selective re-colonization with L. reuteri and B. vulgatus in antibiotic treated mice ( Fig. 10a-b). When subjected to the forced swim test, the re-colonized mice showed only mild improvement, irrespective of the bacterial genus they received (Fig. 10c). This result is likely due to the devastating dysbiosis that broadspectrum antibiotic treatment causes, which cannot be restored by a single bacterial species.  Dramatic cecal enlargement is a phenomenon observed in antibiotic-treated mice, as well as germ-free mice, and is a measure of their intestinal health 119, 170 . The re-colonized mice remained enlarged, showing only a mild amelioration in their cecal size (Fig. 10d). This data indicate the sustained presence of dysbiosis despite the supply of beneficial bacteria.

r e u t e r i
Considering the downsides of this model,

Part III: Discussion and Future Directions
Putting the data in context Taken together, our results demonstrate that microbiome homeostasis was robustly altered in animals undergoing UCMS, with a consistent decrease in Lactobacilli. This phenomenon is likely due to the increased intestinal motility associated with stress. Moreover, our data suggest that the production of H2O2 by Lactobacillus may be protective against the development of despair behavior by direct inhibition of IDO1 expression and decrease in the circulating level of kynurenine, a metabolite associated with depressive behavior (Fig. 11) 84,[172][173][174] .
Our results are in agreement with recent literature demonstrating that microbiome composition is modified with acute and chronic stress [149][150][151]175 . Microbiome dysbiosis is also detected in humans affected by major depressive disorders and the transplantation of the flora from these patients in germ free mice can induce despair behavior 111,136,137  While Lactobacilli are able to control other microbial communities through secretion of antimicrobial factors, genetic limitations make them more sensitive to environmental conditions. In particular, many Lactobacillus genus members are unable to synthesize amino acids and purines and thus rely on nutrient rich environments and other bacteria for supply of essential building blocks [161][162][163][164] . We hypothesized that, in the context of increased intestinal motility such as the one observed in stressed animals, fluctuating availability of nutrients and symbiotic bacteria will impact the renewal of the Lactobacillus niche 183 . Indeed, our results show that directly increasing intestinal transit is sufficient to significantly reduce Lactobacillus levels.
We sought a mechanistic explanation for how the microbiota, and Lactobacilli, in particular, could impact behavioral outputs. By investigating peripheral metabolism, we found that the level of circulating kynurenine is increased after chronic stress, in a manner dependent on Lactobacillus levels. Kynurenine can readily cross the blood-brain barrier to drive depressive behavior, supposedly by disrupting neurotransmitter balance and driving neuroinflammation, within the CNS 84,167,174,[184][185][186] . A recent study by Agudelo et al. 84 identified this pathway as also being disrupted in stressed mice using the same model of UCMS. The study further shows that peripheral increases in kynurenine translate in dysregulation of downstream pathway metabolites in the brain, possibly altering neurotransmission. Taken together, these new findings point to disruptions in tryptophan-kynurenine metabolism as an important factor in mediating despair behavior. IDO1 is the main enzyme responsible for conversion of tryptophan to kynurenine and its expression and activity are inhibited by reactive oxygen species (ROS) 166 . Members of the Lactobacillus family have the capacity to produce high levels of ROS, as a way of fending off other bacteria and maintaining their niche 177,178 . Though difficult to show, it is conceivable that ROS molecules can diffuse from the lumen to signal to the intestinal epithelial cells. In our study, we have shown that decreased levels of ROS in stressed animals correlate with an increase in intestinal ido1 transcripts, thus potentially explaining our observed increase in circulating kynurenine.
Altogether, our results indicate that the microbiome can play a role in the development and symptomatology of depression.

Open-ended questions
While the data stemming from this project provide new insights into how specific aspects of the gut microbiota can affect behavior, they opened more questions than they answered.

Is loss of Lactobacillus sufficient to induce depressive behavior?
Perhaps the most direct question is that of the causality between the level of intestinal Lactobacillus and depressive behavior. The fact that supplementing reduced levels of Lactobacillus with exogenous bacteria improves behavioral outcome is a good indication that the bacteria has a beneficial role for the host. However, it may not necessarily be representative of the homeostatic state. As previously discussed, since selective depletion of Lactobacillus is not currently possible, the ideal experiment would be selective reconstitution of germ-free mice with defined complete flora containing or lacking Lactobacillus components. Another option would be reconstituting germ free mice with fecal contents from naïve and stressed mice.
However, previous work shows that, once the new microbiota is established in the new host, its composition changes from the original inoculum 109,111 . Therefore, it would be difficult to make conclusions about the original components of the transferred microbiota.

Is Lactobacillus-derived ROS the mechanism behind altered kynurenine metabolism?
Another open-ended question is the connection between Lactobacillus and kynurenine metabolism. Both our and previous data show that Lactobacillus can produce high amounts of ROS 166,176,178 . Previous evidence also shows that that the molecule can inhibit IDO1 activity 166 . However, whether it is indeed the ROS produced by Lactobacillus in the lumen, diffusing into the epithelial cell layer to inhibit IDO1, or whether other mechanisms are involved, remains unknown. One approach to understand this process would be to genetically manipulate the Lactobacilli to silence or boost their ROS production. Such an experiment would be technically challenging, given that Lactobacilli evolved several genes involved in ROS secretion 161,162,176,178 . Nevertheless, it would definitely be informative about the importance of Lactobacillus-derived ROS for regulating kynurenine metabolism. Another approach to understanding whether luminal ROS can impact intestinal IDO1 and kynurenine metabolism would involve local delivery or buffering of ROS directly in the lumen, and measuring IDO1 activity in the intestinal tissue. The gastroenterology field has developed several assays to study localized responses in the intestines, including creating temporary intestinal loops or catheter implantation 187,188 . This strategy would allow control over luminal ROS levels, as well as their timing, for effective differentiation of their effects. Moreover, the timed control would further allow investigation into the cell types that might respond with modulated IDO1 activity.

How can peripheral kynurenine modulate brain activity?
More directly relevant to depression etiology is a discussion regarding the role of kynurenine metabolites in the brain. In the absence of inflammation, IDO1 expression in the brain is very low and kynurenine is absorbed from peripheral circulation 167,184 . Evidence shows that kynurenine is trafficked through the large neutral amino acid transporter (LAT1), the same transporter used for its precursor, tryptophan 185,189 . LAT1 may be expressed by all the cell types in the brain, but highest expression has been reported on endothelial cells, as well as astrocytes and microglia 190,191 . The two glial cell types reportedly process kynurenine differently. Astrocytes express KAT2 and metabolize kynurenine to kynurenic acid, whereas microglia express KMO and have quinolinic acid as its product 167,184,191,192 . Both molecules have neuromodulatory effects, as they can both alter glutamatergic transmission: kynurenic acid is considered an NMDA receptor antagonist, whereas quinolinic acid is considered an agonist 167,184,193 . These experiments stem from exogenous application of the molecules, showing that kynurenic acid has inhibitory effects, whereas quinolinic acid has excitatory effects 167,184,193,194 . However, nothing is known about the regulation and interaction of the molecules at the synapse. Studies show that, with increased peripheral kynurenine metabolism, both kynurenic and quinolinic acid levels are simultaneously increased in areas of the brain outstanding questions involve discerning the role of kynurenine metabolism in different glial cell types and their contribution to synaptic activity. Given the fact that the glial cells express a variety of glutamate receptors 197 , kynurenine metabolites may also impact the function of the glial cells themselves. Moreover, the regulation and timing of release, as well as clearance from the extracellular environment of these molecules remains to be investigated. Are kynurenic and quinolinic acid released simultaneously at the same synapse, or is their timing carefully controlled? Furthermore, if they both indeed bind the same glutamate receptors (as it is assumed), which molecule wins the competition? Finally, more thorough analysis indicates that kynurenine metabolism is not upregulated indiscriminately throughout the brain, but may be limited to discrete areas 84 . Understanding how this process is regulated (is it dependent on neuronal activity?) will also further our knowledge of how kynurenine metabolism may lead to depressive behavior.

What is the role of the immune system in maintaining resilience?
While evaluating the contribution of the immune system to the microbiota changes during chronic stress, we observed that mice lacking their adaptive immune system (SCID mice) develop an exacerbated depressive phenotype. Giving the vast literature showing that activation of the immune system (both adaptive and innate) is detrimental [77][78][79][80][81] , this is an intriguing finding. Nevertheless, full immune competence has also been shown to be necessary for other behaviors, such as learning and memory [198][199][200][201][202] . Despite these previous studies, we are still lacking the understanding of how the adaptive immune system supports brain activity. One of the studies showed that immunodeficient mice have decreased adult neurogenesis 200,201 , a process often associated with depressive behavior due to its recovery by antidepressant treatment 203,204 . Though disrupted adult neurogenesis is often cited as accompanying depressive behavior in animal models 14, 89,203,204 , it is unclear whether or how this process may contribute to the depressive behavior. Most adult generated neurons are inhibitory interneurons 205,206 . Given the increased activation of circuits in the hippocampus, as well as the habenula and the VTA, one possibility is that adult neurogenesis is necessary to maintain a stable tone within brain circuitry. Of course, it is also possible that the increased activation of hippocampal circuits inhibits the integration of inhibitory neurons.
Another study investigating cognition in immunodeficient mice showed that the phenotype of the immune cells in the meningeal spaces is associated with the learning outcome of the animals 198,207 . Specifically, mice displaying a meningeal pro-inflammatory skew (such as the SCID mice, IL4 deficient mice) were poor learners compared to the mice that could skew their meningeal repertoire to a pro-homeostatic phenotype (immunocompetent mice).
Whether the inflammatory phenotype in the meninges is sufficient to drive cytokine levels high enough to change synaptic activity as seen during systemic inflammation is unknown.
Nevertheless, preliminary data shows that SCID mice have higher circulating levels of lipopolysaccharide (LPS), as well as decreased levels of meningeal myeloid cells (Fig. 13a-b).
LPS is considered a potent activator of the immune system, usually leading to increased proinflammatory cytokines and blood brain barrier (BBB) permeability 81,208 . The effects of tonically elevated LPS levels, however, are unknown. It is likely that sustained exposure to LPS mediates a wide array of effects, such as a more reactive microglia phenotype, as seen in brain areas with a more permeable BBB 116,209 . Interestingly, this hypothesis involves the co-existence of immunodeficiency and inflammation, two phenomena that are not usually associated together.
Undoubtedly, these are just few of the hypotheses to be considered regarding the role of the immune system in maintaining brain function. Following up on this topic will bring important advances about the balance between beneficial and detrimental immune responses for depressive behavior, as well as other aspects of host physiology.

Can gut microbiota dysregulation mediate human depression?
Whether the activity of a system so far removed from the brain as the gut microbiota can alter its activity to the point of causing a disorder as complex as depression is an argument to consider. In the introduction, I described how the causes of depressive disorder are still unclear.
Associated risk factors range from intrinsic (genetics) to environmental ones (stress). Moreover, many diseases, such as autoimmune multiple sclerosis, are hallmarked by depressive behavior 77,78 . In the case of multiple sclerosis, for instance, we are still uncertain of what brain changes lead to depressive behavior, specifically. However, it is believed that widespread neuroinflammation affects neuronal activity leading to a plethora of altered behaviors, including cognition, resilience and exploration 79,81 . In this context, it is possible that microbiota dysregulation, through widespread metabolism dysregulation, can also alter neuronal activity and lead to behavioral deficits.
Can this happen in humans? The wide array of depression symptoms and co-morbidities indicates that multiple underlying causes might lead to depressive behavior. Therefore, it is conceivable that a subset of patients might suffer from dysbiosis, (perhaps with low levels of Lactobacillus) which underlies or contributes to their depression symptoms. The scenario presented by our experimental model is definitely reductionist and much simpler that any scenarios involving human biology. Nevertheless, in favor of our hypothesis, human studies show high incidence of depression in patients of chronic intestinal disease [210][211][212] .
In this study, we identified Lactobacillus as a genus of bacteria important for mediating resilience, potentially by maintaining kynurenine metabolism. To my knowledge, this is the first study associating a specific type of bacteria with depressive behavior. The finding is undoubtedly dependent on our experimental procedure. However, this is likely just one of many bacteria, which, when dysregulated, may mediate depressive behavior 111 . These remain to be discovered.

New avenues: the role of kynurenine metabolism in multiple sclerosis associated depression
Much of our current understanding about the role of kynurenine metabolism in depression comes from models of inflammation 173,213,214 . Of particular interest is multiple sclerosis (MS), an autoimmune disease hallmarked by inflammatory bouts and CNS demyelination, leading to sensory and motor dysfunction. As already mentioned, many MS patients additionally suffer from psychiatric symptoms, such as depression and anxiety [215][216][217] . Therefore, a potential new avenue stemming from my project would be the investigation of the role of kynurenine metabolism in depression during multiple sclerosis.
Extensive research has been dedicated to discovering pathways and molecules acting on the immune system to limit demyelination during MS attacks 218,219 . Interestingly, one of such promising targets is the tryptophan-kynurenine pathway 216,220 . During inflammation, IDO1, the enzyme that metabolizes tryptophan to kynurenine, is upregulated by various immune cells 173,[221][222][223] . Kynurenine and kynurenine associated metabolites have a robust anti-inflammatory activity and are protective in the mouse model of MS: experimental autoimmune encephalitis (EAE) 220,224 . Furthermore, mice lacking IDO1 present with a more severe clinical form of the disease 222 . It is also suspected that IFNb, one of the most widely used MS treatments, could be acting by inducing IDO1 expression in immune cells 216,224 .
While IDO1 expression and the presence of kynurenine metabolites is beneficial on the immune system in the context of MS 220,222,224 , the same player has now been shown to have negative impacts on mental health 84,173,186,193,195,196 . Teasing out the cellular biology of kynurenine metabolism may be particularly impactful for managing depressive behavior in MS/EAE. While kynurenine metabolism is increased during MS/EAE, whether this increase is resulting only from IDO activity in immune cells, or whether cells in the brain also increase IDO expression in response to inflammation, is currently unknown. Answering this question could help targeting kynurenine metabolism to harness its beneficial effects in the immune system, while limiting its harmful effects for the CNS.
Recently, loss of oligodendrocyte progenitor cells (OPCs) was shown to initiate depressive behavior in animal models 225 . This subset of glial cells, precursors to myelinating oligodendrocytes, fluctuates in number during demyelinating inflammation, particularly at lesion sites 226,227 . Though kynurenine metabolism in oligodendrocyte lineage cells has not been investigated, it is plausible that kynurenine and/or its metabolites are toxic to these cells and impair their neuro-supportive function.
Therefore, one possible hypothesis is that, in the context of EAE inflammation, IDO/ kynurenine metabolism dampens inflammation at the expense detrimental effects on brain function leading to depressive-like behavior. Specifically, we hypothesize that kynurenine metabolites mediate this behavior in part by impairing the function of oligodendrocyte lineage cells. To address this hypothesis, we will first resolve the contribution of the immune system versus the brain (and other organs) to IDO1/ kynurenine metabolism increase during EAE inflammation and how each contributes to depressive behavior.
To this end, we will generate bone marrow chimeras, in which the host and the immune system will be either wildtype or IDO1 knockout. Upon EAE induction, we will evaluate the animals for disease progression (clinical score, severity of inflammatory response, depressive phenotype). We will also interrogate IDO1 and kynurenine levels in different compartments and cells types associated with autoimmune inflammation and/or kynurenine metabolism. (i.e.

peripheral immune, central immune, neural and glia cells, intestinal and liver cells). The results
should indicate whether the source of IDO1/ kynurenine is solely from immune cells, or whether other cells also upregulate it in response to inflammation. For instance, a chimera with IDO1 knockout immune system, and a wildtype host should present higher clinical scores, but no depressive behavior. In this scenario, reducing kynurenine uptake in the CNS in a wildtype animal (such as by pharmaceutically blocking the LAT1 transporter) could prevent or alleviate depressive behavior, while maintaining its benefits on the immune system.
For the second part of the hypothesis, we will seek to understand the role and effects of IDO1/kynurenine metabolism in oligodendrocyte lineage cells and how it relates to depressive behavior during EAE. To this end, we will use the NG2-eGFP reporter to sort cells at different stages of the oligodendrocyte lineage and screen them for enzymes in the IDO1/kynurenine pathway. We will also use in vitro cultures to determine the effects of kynurenine and its downstream metabolites (kynurenic acid, quinolinic acid, anthranilic acid), on OPC and oligodendrocyte fuction, measuring outcomes such as survival, proliferation, differentiation, myelin production, etc. Such experiments could yield particularly important insights into oligodendroglia function during EAE, as both OPCs and oligodendrocytes express various glutamate receptors, which can bind both kynurenic and quinolinic acid. Recently, activation of NMDA receptors on oligodendrocytes were shown to be important for providing metabolic support to active axons, and deleting these receptors on oligodendrocytes can lead to impaired axonal transmission. It is plausible, therefore, that accumulation of kynurenic acid, an NMDA receptor antagonist, also impairs oligodendrocyte function in providing metabolic support.
Finally, following our preliminary experiments regarding kynurenine metabolism in oligodendrocyte lineage cells, we will attempt to block the detrimental effects of kynurenine during EAE in vivo by specifically knocking out the identified enzymes using viral delivery of floxed constructs in animals expressing NG2-driven cre protein.
Altogether, these experiments will not only differentiate the source of kynurenine during neuroinflammatory episodes, but will also explore a completely new area of study, the role of kynurenine metabolism in oligodendrocyte lineage cell function. stoichiometrically. Due to their poor solubility, and to avoid using organic solvents with potential effects on the microbiota, the drugs were thoroughly blended with pulverized dry food (at 1mg/g each, equivalent to 168 ) and subsequently mixed as soft food with water. The treatment was administered for three weeks, with the mice receiving fresh food every 1-2 days.

Materials and Methods
Intestinal transit time measurement. Mice were briefly anesthetized with isofluorane and a 3 mm diameter glass bead was inserted 2 cm inside the rectum with a vaseline lubricated rod. The mice were then placed in an empty cage without bedding and the time to bead expulsion was measured.
Intestinal tissue analysis. The small intestine was dissected by excising under the stomach and before the cecum. Mesenteric fat and Peyer's patches were carefully removed using fine forceps. The intestine was opened longitudinally and the contents were removed in two PBS washes. Excess liquid was gently absorbed using kimwipes and the tissue was weighed.
Intestinal cells were then isolated as previously described 232  Metabolite analysis. Blood was collected by cardiac puncture and centrifuged at 10,000 x g for 3 minutes in gel tubes for serum preparation. Frozen (-80°C) serum samples were shipped to the University of Michigan Metabolomics Core for untargeted metabolomics analysis.
Metabolites isolated by positive and negative ion selection were analyzed by mass spectrometry. Mass spectrometry peak intensities were further analyzed using the MetaboAnalyst online software. Peak values were filtered using the interquantile range, normalized to the group sum, then log transformed and auto-scaled for principal component analysis and further statistical tests 233 . Identifiable significant metabolites were analyzed through pathway analysis, followed by further manual pathway enrichment.

ROS quantification.
Fresh fecal samples were collected in sterile 2 mL tubes, weighed, and resuspended in 1mL sterile PBS. After brief sedimentation of insoluble particles, 500 µL of bacterial slurry were incubated at 37°C for 30 minutes. Following bacterial precipitation by centrifugation, 50 µL of supernatant was reacted using the Amplex Red hydrogen peroxide/ peroxidase assay kit (Thermo Fisher, cat. A22188), according to manufacturer's protocol.
qPCR. Frozen tissues (brain and intestine) were homogenized by bead beating in RNA TRI Reagent (Life technologies) and RNA was extracted according to manufacturer's protocol.
cDNA was synthethized with the High Capacity cDNA Reverse Transcription kit (Life Technologies). cDNA was amplified using the Sensifast Sybr NO-ROX kit (Bioline), according to manufacturer's instructions. Gapdh was measured as a normalizer for each sample. Results were analyzed by the relative quantity (ΔΔCt) method 234 . For total 16S rRNA quantification, we followed the protocol described by Liu et al., 235 . L. reuteri DNA was used for the standard curve and specific primers for the 16S rRNA V3-V4 region. For relative quantification of Lactobacillus, the ΔΔCt method was used to compare Lactobacillus-specific amplification to that of the 16S rRNA gene. Both reactions were performed using the Sensifast Sybr NoROX kit from Bioline (BIO-98005). Primer sequences are available in Table 1.

Statistical analysis.
All statistical analyses were performed in Prism. Analyses involving two groups were performed using a two-tailed t-test. If the variances between groups were significantly different, a Welch correction was applied. For experiments involving stress and another variable (e.g. L.reuteri treatment or vendor), data were analyzed with a two-way ANOVA, followed by Bonferroni post-hoc tests. For the metabolomics experiments involving only 3 groups, a one-way ANOVA was utilized. Outliers were excluded if they fell more than two standard deviations from the mean. For all analyses, the threshold for significance was at p<0.05. Repeats for each experiment, if performed, are specified in the figure legend corresponding to the respective panel.