Modulation of sensory behavior and food choice by an enteric bacteria-produced neurotransmitter

Animals coexist in commensal, pathogenic or mutualistic relationships with complex communities of diverse organisms including microbes1. Some bacteria produce bioactive neurotransmitters which have been proposed to modulate host nervous system activity and behaviors2. However, the mechanistic basis of this microbiota-brain modulation and its physiological relevance is largely unknown. Here we show that in C. elegans, the neuromodulator tyramine (TA) produced by gut-colonizing commensal Providencia bacteria can bypass the requirement for host TA biosynthesis to manipulate a host sensory decision. Bacterially-produced TA is likely converted to octopamine (OA) by the host tyramine beta-hydroxylase enzyme. OA, in turn, targets the OCTR-1 receptor on the ASH/ASI sensory neurons to modulate an aversive olfactory response. We identify genes required for TA biosynthesis in Providencia, and show that these genes are necessary for modulation of host behavior. We further find that C. elegans colonized by Providencia preferentially select these bacteria in food choice assays, and that this selection bias requires bacterially-produced TA. Our results demonstrate that a neurotransmitter produced by gut microbiota mimics the functions of the cognate host molecule to override host control of a sensory decision, thereby promoting fitness of both host and microbe.

lacp :: mCherry JUb39-grown OP50-grown (17)   Numbers in parentheses indicate the number of animals; 3 independent assays. P-value is derived from an ordinal regression. 1 We next performed chemotaxis assays with animals fed on mCherry-labeled JUb39, and quantified intestinal bacterial cells in animals that had navigated either toward octanol or toward the control. We found that animals navigating toward octanol consistently contained more gut bacteria (Fig. 1e). We conclude that JUb39 colonizes the worm gut and the extent of colonization is correlated with decision-making in response to octanol.
We investigated the mechanistic basis for Providencia-mediated octanol modulation. Octanol avoidance is subject to extensive modulation directly and indirectly via multiple biogenic amines including tyramine (TA) and octopamine (OA) (Fig. 2a) as well as neuropeptides 9,[18][19][20][21][22] . TA is produced from Tyrosine (L-Tyr) via the activity of a tyrosine decarboxylase (TDC; encoded by tdc1 in C. elegans); TA is subsequently converted to OA via a tyramine beta hydroxylase (encoded by tbh1) 23 (Fig. 2a). Consequently, all tbh1 mutant phenotypes resulting from lack of OA are expected to be shared by tdc1 mutants 23 . Unexpectedly, we found that while tdc1 mutants grown on JUb39 continued to exhibit octanol modulation, the modulation exhibited by tbh1 mutants was significantly reduced (Fig. 2b). Mutations in the cat2 tyrosine hydroxylase 24 and tph1 tryptophan hydroxylase 25 enzymes required for the production of biogenic amines dopamine and serotonin in C.
elegans, respectively, did not affect octanol modulation (Fig. 2b). These results raise the possibility that C. elegans-produced OA, but not TA, is partly necessary for JUb39-mediated octanol modulation.
To account for these observations, we hypothesized that JUb39 may produce TA that functionally compensates for the host tdc1 mutation. tdc1 mutants grown on OP50 have been reported to display more rapid aversive responses to dilute (30%) octanol 18 . Exogenous TA suppresses this increased aversion of tdc1 animals but does not alter wild-type responses under the same conditions 18 . To test if JUb39 is able to suppress behavioral defects of tdc1 mutants, we performed short-range acute avoidance assays [the "smell-on-a-stick" (SOS) assay] 9,26 . In this assay, the strength of avoidance is inversely correlated with reversal latency when the animal encounters the repellent as it is moving forward. As expected, tdc1 mutants grown on OP50 responded more rapidly to 30% octanol than wild-type animals (Fig. 2c). This enhanced aversion was suppressed upon growth on JUb39 (Fig. 2c). These results are consistent with the notion that bacterially-produced TA functionally complements for the loss of host-derived TA in driving a sensory behavioral decision.
To directly test for the production of TA by JUb39, we measured succinyl-TA levels using high-resolution HPLC-MS in wild-type and tdc1 worms grown on either OP50 or JUb39 27 . Levels of succinyl-TA were comparable in wild-type animals grown on either OP50 or JUb39, whereas tdc1 mutants grown on OP50 had no detectable succinyl-TA, consistent with previous reports 23 (Fig. 2d, Fig. S2). Notably, succinyl-TA levels were  Positive numbers indicate reduced avoidance of octanol. Errors are SEM. Gray thin and thick vertical bars at right indicate Bayesian 95% and 66% credible intervals, respectively. P-values between the indicated conditions are from a GLMM with Dunnett-type multivariate-t adjustment. P-value in red indicates Wald z-statistic for the magnitude of the JUb39 effect in tbh-1 compared to wild-type. c, e) Reversal response latency of animals of the indicated genotypes grown on OP50 or JUb39 on NGM (c) or NGM + 0.5% L-Tyr (e) in response to 30% octanol (c) or 100% octanol (e) in SOS assays. Each dot is the response time of a single animal. The Y-axis is log 10 -scaled for these log-normal distributed data, and normalized to the indicated control group for each experimental day. Numbers in parentheses indicate the number of worms tested in assays over at least 3 independent days. Boxplot indicates median and quartiles, whiskers indicate the data range, excluding outliers. Gray thin and thick vertical bars at right indicate Bayesian 95% and 66% credible intervals for the difference of means, respectively. P-values between the indicated conditions are from a linear-mixed effects regression on log-transformed data (LMM). P-value in red (e) indicates Wald F-statistic for the effect of the indicated genotypes on the magnitude of the JUb39 effect. d) Quantification of succinyl-TA in wild-type and tdc-1 mutant animals grown on either OP50 or JUb39. Data are averaged from three independent replicates each. ND, not detected. P-values are from two-way ANOVA. 1 partly restored in tdc1 mutants grown on JUb39 (Fig. 2d). OA was not detected under these conditions in either wild type or tdc1 mutants. We conclude that JUb39 in association with C. elegans produces TA which can accumulate in the host.
Although TA biosynthesis in bacteria has been demonstrated in some gram-positive genera, production appears to be uncommon in gram-negative bacteria which include Providencia 28,29 . TA production in grampositive strains is induced upon supplementation with L-Tyr 30 . We found that growth on L-Tyr-supplemented media enhanced octanol modulation by JUb39 in SOS assays (Fig. S3). Under these conditions, mutations in tbh1 fully suppressed octanol modulation in SOS assays, whereas consistent with our observations in longrange chemotaxis assays, tdc1 mutants continued to exhibit robust octanol modulation (Fig. 2e). Octanol avoidance behaviors of tdc1; tbh1 double mutants were similar to those of tbh1 mutants alone (Fig. 2e), indicating that the lack of host-derived OA, and not accumulation of TA due to loss of TBH-1 23,27 accounts for the reduced octanol modulation in tbh1 mutants.
Biogenic amines are typically generated from aromatic amino acids and L-glutamate by pyridoxyl phosphate (PLP)-dependent group II aromatic amino acid decarboxylase enzymes (AADCs) in both eukaryotes and bacteria 31 ( Fig. S4a- Table S1). In gram-positive Enterococcus and Lactobacillus (Lb), TA production is mediated by the TDC-encoding (tyrDC) AADC and tyrosine permease/transporter (tyrP) genes present in an operon; this operon is inducible by L-Tyr ( Fig. 3a- Table S1) 32,33 . Although genes related to Ente rococcus tyrDC and tyrP were largely absent in Gammaproteobacteria (Fig. 3b), we confirmed the presence of homologous operons containing tyrDC and tyrP in JUb39 and PYb007 in de novo genome assemblies via whole genome sequencing (Fig. 3a, Fig. S4b, Table S1). tyrDC homologs were also identified in the genomes of additional members of the Morganellaceae family, although the operon structure was conserved in only a subset of these genomes (Fig. 3a, Fig. 3c, S4b, Table S1).
Providencia TyrDC is highly homologous to the Lb enzyme, which has been well characterized with respect to substrate specificity 34 . Protein modeling using the crystal structure of Lb-TyrDC 34 as a guide (see Methods) indicated that JUb39 TyrDC shares most known catalytic sites with Lb-TyrDC (Fig. 3d). Interestingly, JUb39 TyrDC contains a substitution at A600 (S586 in Lb-TyrDC; Fig.3d), a variant demonstrated to enhance specific catalytic activity of Lb-TyrDC for tyrosine 34 . We infer that JUb39 TyrDC likely generates TA from tyrosine.  Morganella strains (Morganellaceae family) have been reported to produce TA under certain conditions 28 , despite having no discernible tyrDC orthologs (Fig. 3c, Fig. S4a-b, Table S1). Instead in Morganella, we identified an AADC-encoding gene (hereafter adcA) with~29% and 27% sequence identity to Enterococcus TyrDC and human GAD67, respectively, in an operon upstream of a gene encoding a TYT-1 family tyrosine permease (Fig. 3a, Fig. 3c). An adcA homolog is also present in Providencia genomes including in JUb39 but is not adjacent to a tyrosine transporter (Fig. 3a, Fig. S4a-b, Table S1). We conclude that Providencia encodes at least two AADCs with the potential to generate TA, and the phylogenetic incongruence suggests that both tyrDC and adcA genes may have either been lost or acquired in the Morganellaceae family via horizontal gene transfer.
Together, these results indicate that TA produced by multiple AADC enzymes in Providencia is both necessary and sufficient to modulate octanol avoidance by wild-type C. elegans.
We next identified the molecular targets of Providencia-mediated octanol modulation in the host. As bacterially-produced TA is likely converted to OA via the host TBH-1 enzyme 23 to mediate octanol modulation, we focused primarily on host OA receptors. The bilateral ASH nociceptive neurons located in the head amphid organs of C. elegans have been implicated in sensing octanol 9,19,26 . These neurons express multiple TA and OA receptors, a subset of which is required for octanol modulation by these monoamines 18,35 . Among ASH-expressed OA receptors, mutations in octr1, but not ser3, abolished JUb39-mediated octanol modulation, without altering the extent of gut colonization (Fig. 4a, Fig. S5). We also observed an effect on octanol modulation in tyra2 TA receptor mutants, primarily due to decreased octanol avoidance upon growth on OP50 (4.3 ± 0.25s for tyra2 vs. 2.9 ± 0.13s for WT). TYRA-2 has recently been shown to mediate responses to an OA-linked pheromone 36 , although the reason for the observed effect in OP50-grown animals is currently unclear. Expression of octr1 cDNA in the ASH/ASI sensory neurons restored octanol modulation (Fig. 4a).
Next we investigated the biological relevance of the JUb39-directed decrease in octanol aversion by C. el egans. While many gram-negative enteric bacteria produce long-chain alcohols including octanol 37 , whether Providencia produces this chemical is unknown. However, JUb39 and other Providencia strains can produce the branched alcohol isoamyl alcohol (IAA), which is aversive to C. elegans when concentrated 38,39 . Similar to octanol, avoidance of high IAA concentrations is also mediated by the ASH sensory neurons 39 . We hypothesized that reduced avoidance of JUb39-produced aversive alcohols or other odorants may preferentially bias JUb39-grown C. elegans to select these bacteria in food choice assays. Indeed, animals grown on JUb39 more strongly preferred JUb39 compared to OP50-grown worms, which showed a slight preference for JUb39 in a short-range food choice assay (Fig. 4c). The bias towards JUb39 was eliminated in animals grown on JUb39 ∆tyrDC::cmR ∆adcA, suggesting that bacterial TA production is necessary for this food preference (Fig 4b).
Together, these results imply that TA produced by JUb39 reduces ASH/ASI-mediated avoidance of JUb39produced aversive cues such as concentrated alcohols, to allow preferential selection of these bacteria.
Our observations support a model in which the neurotransmitter TA produced by intestinal Providencia bacteria modulates aversive responses of C. elegans to the enteric bacteria-produced volatile metabolite octanol, likely via subverting host-dependent TA production. Bacterially-produced TA is converted to OA by C. elegans TBH-1; OA subsequently acts on the ASH/ASI neurons via the OCTR-1 OA receptor to decrease aversion of octanol. Bacterially-derived TA also increases the preference of C. elegans for Providencia in food choice assays (Fig. 4d). We speculate that the preference for Providencia upon colonization of C. elegans by these bacteria promotes increased consumption leading to stable association 4,40 and bacterial dispersal. As Providencia is a rich food source for C. elegans 6

Fig. 4. Modulation of octanol avoidance by Providencia requires the OCTR-1 OA receptor in the ASH/ASI sensory neurons. a-b)
Reversal response latency of animals of the indicated genotypes grown on the shown bacteria in control conditions of NGM + 0.5% L-Tyr (a) or supplemented with TA + 0.5% L-Tyr (b) to 100% octanol using SOS assays. Each dot is the response time of a single worm. Y-axis is log 10 -scaled for these log-normal distributed data, and normalized to the indicated control group for each experimental day. Numbers in parentheses indicate the number of worms tested in assays over at least 3 independent days. Boxplot indicates median and quartiles, whiskers indicate the data range, excluding outliers. Gray thin and thick vertical bars at right indicate Bayesian 95% and 66% credible intervals for the difference of means, respectively. P-values between indicated conditions are from a LMM with Tukey-type multivariate-t adjustment. c) (Left) Cartoon depicting assay setup of the short-range bacterial choice assay. (Right) Preference index of animals grown on the indicated bacteria for the test bacteria JUb39. Each dot represents one assay of at least 10 animals; assays were performed over at least 4 independent days. Y-axis is on log-odds (logit) scale. Errors are SEM. Gray thin and thick vertical bars at right indicate Bayesian 95% and 66% credible intervals, respectively. P-values represent difference of means relative to JUb39-grown animals from a GLMM with Dunnett-type multivariate-t adjustment. d) Cartoon of working model. JUb39 colonizes the C. elegans intestine and produces TA via the TyrDC and AdcA enzymes. TA is converted to OA by C. elegans TBH-1 and acts via the ASH neuron-expressed OCTR-1 OA receptor to modulate octanol avoidance and food choice.
Bacteria: For all experiments, bacterial strains were streaked from glycerol stocks prior to use and grown to saturation in LB media at 37ºC. For conditioned media, bacteria were grown to saturation in NGM media overnight at 37ºC, then cleared by centrifugation at 14,000g for 3 minutes. Prior to use, conditioned media or NGM was supplemented with 5x concentrated OP50 from a saturated LB culture to prevent starvation. To expose animals to bacterial odors, worms were grown on seeded NGM plates whose lids were replaced with NGM plates containing the test bacteria; these were sealed with parafilm. For L-Tyr and TA supplementation experiments, 0.5% L-Tyr (Sigma T3754) or 4mM or 10mM TA (Sigma T2879) were added to the NGM media and agar prior to pouring plates. Plasmids were transformed into JUb39 and OP50 via electroporation. Deletions in JUb39 were induced using homologous recombination with the temperature-sensitive pSC101 replicon at 42ºC, and sacB-sucrose counter-selection at 30ºC, in the absence of NaCl as described 41 , with the exception that bacteria were incubated for 1 hour at room temperature in the presence of 10mM arabinose for lambda Red induction prior to selection at 42ºC. Deletions were confirmed by sucrose resistance and kanamycin sensitivity, followed by PCR and sequencing of deleted intervals.

Molecular biology
The octr1 cDNA was a gift from Dr. Richard Komuniecki. The cDNA was amplified by PCR and cloned homology arms were 701 and 422bp, respectively, flanking a 1398bp deletion of the adcA CDS. For expression of mCherry in OP50 and JUb39, a pUCP20T-mCherry plasmid 44 was modified to replace bla(ampR) with aph(kanR).

Microscopy
All fluorescence microscopy was performed using animals anesthetized with 100 mM levamisole (Sigma Aldrich). Animals were imaged on 2% agarose pads using an upright Zeiss Axio Imager with a 63X oil immersion objective.
Quantification of intestinal bacterial cell numbers: All rod-shaped punctae in the intestines of young adult worms of approximately 1-2µm were included in the quantification. Each animal was recorded in one of three categories containing 0, <10, or >10 cells per animal. Exact numbers in animals bearing over 10 cells were not recorded, but rarely exceeded approximately 100 cells.
Fluorescence intensity measurements: All images were collected in z-stacks of 0.5 µm through the heads of young adult worms. Quantification was performed using ImageJ (NIH). Fluorescence was quantified by identifying the focal plane in which the cell soma was visible, followed by manually drawing an ROI around the soma. Mean pixel intensity was recorded for each neuron pair per animal and the average of fluorescence in each animal is shown.
Worms were cultured for 1 generation with the relevant bacteria prior to the assay. Assays were performed using 10cm square NGM plates. The number of worms in two horizontal rows adjacent to the odor and ethanol spots were quantified.
SOS assays: Smell-on-a-stick (SOS) assays in response to 1-octanol or 2-nonanone were performed as described 9,26 . NGM plates were pre-dried for 1 hour prior to assays. Age-matched young adult animals were picked from food to a clean transfer plate and allowed to briefly crawl away from food for approximately 1 min. Animals were then transferred to another clean NGM plate for 15 minutes prior to assaying responses to 100% octanol (Sigma O4500) and 100% 2-nonanone (Sigma 108731), or 20 minutes for 30% octanol assays.
Shortrange bacterial choice assay: Animals were raised and prepared identically to those used in longrange chemotaxis assays, with the exception that the final wash with water was omitted. NGM plates con-taining 2 15µL spots of overnight-grown bacterial food concentrated to OD600~10 placed 2cm apart were allowed to dry, then incubated with a closed lid for 5 hrs at room temperature. Approximately 30 animals were placed between the two spots, and excess liquid was removed. Animals were allowed to navigate for 15 minutes following which 2µL of sodium azide was applied to each spot to anesthetize worms. Very little lawn-leaving behavior was observed during this short time period. Adult animals on the control spot and test spot were counted.
Osmotic avoidance assay: Animals off the bacterial food on the cultivation plate were picked using a 10% methyl cellulose polymer solution and placed in the center of an NGM plate with a ring of 8M glycerol containing bromophenol blue (Sigma B0126). The number of worms inside and outside of the ring were counted after 10 mins. TyrDCs -denoted "Enterococcus-type TDC", (2) Eukaryotic AADCs denoted "Eukaryotic-type AADC", and (3) Morganella AdcA and Providencia AdcA.

Bacteria genome sequencing
Based on this initial tree, a second tblastn search was used to determine the presence or absence of homologous genes among complete Gammaproteobacteria genomes. Enterococcus faecalis TyrDC and C. elegans TDC-1 were used as tblastn search query sequences. Hierarchical search was performed as described above, limited to an e-value cutoff of 10-5. A maximum of 2 highly similar sequences were retained per genus for phylogenetic analysis as listed in Table S1.
A final phylogenetic tree was constructed using the amino acid sequences derived from these tblastn queries. These were assembled into a consensus alignment using the Phylomizer workflow as described above.
ProtTest 53 (https://github.com/ddarriba/prottest3) was used to identify the optimal model for likelihood estimation, using Aikake Information Criterion (AIC) values for selection. The model selected and subject to PhyML analysis was an LG model with discrete gamma distribution, an estimated proportion of invariant sites (+I), empirical frequencies of amino acids (+F), estimated gamma shape parameter (+G) for rate variation among sites with the default 4 substitution rate categories, and the subtree pruning and regrafting (SPR) algorithm. 100 bootstrap pseudoreplicates were analyzed. Representatives from the resulting phylogeny were used to categorize and compile the cladogram in Fig. 3b-c. Adjacent genomic sequences, up to 3 CDS 5' or 3', were examined for genes encoding amino-acid permeases or transporters in an apparent operon as defined by close proximity and same orientation with respect to each tblastn hit (Table S1).

Molecular modeling
The putative amino acid sequence for JUb39 TyrDC was used to model active site residues using the Lb-  34 for illustrative purposes only.

Statistical analyses
All statistical analyses were performed in R (https://www.R-project.org/) and RStudio (http: //www.rstudio.com). For modulation index and relative latency figures, data were were normalized to the relevant control group mean value for each experimental day on the log scale via subtraction. Outliers in boxplots were defined as greater than 1.5*interquartile range, but were included for analysis. All statistical analyses were performed on raw, non-normalized data. To avoid inflated P-values and account for non-independence of observations, we employed mixed-effects regression analysis in lieu of simple ANOVA and t-tests. For behavioral assays, frequentist statistical comparisons were performed using a binomial generalized linear mixed-effects model (GLMM) with a logit link function for chemotaxis and food choice assays, while a linear mixed-effects model (LMM) on log 10 -transformed data was used to analyze SOS assays using the 'lme4' package. In all cases, a random intercept term for assay plate was used to account for non-independence of animals on each assay plate and random intercept for date was used to account for day-to-day variability. In the presence of interactions, for example the effects of bacterial strains across different odorants in Fig. 1a, a random slope term per date was also used when appropriate. Estimated P-values for pairwise comparison of fixed effects were determined using Kenward-Roger approximated degrees of freedom as implemented in the 'emmeans' and 'pbkrtest' packages. In nearly all cases, inclusion of random effects model terms resulted in conservative P-value estimates compared to a simple ANOVA. In the event of singular model fit, any random slope term, followed by random date effect terms were removed to allow convergence. For Wald statistics of model terms, packages 'lmerTest' or 'car' were used.
Additionally, for each dataset, a maximal Bayesian model was fit using the 'rstanarm' and 'rstan' packages. package and 'polr' function. Categories of cell numbers were considered ordered factors of 'none', 'some' or 'many' cells.

Sample preparation for HPLCMS
Approximately 10,000 mixed-staged worms in 1.5mL microfuge tubes were lyophilized for 18-24 hrs using a VirTis BenchTop 4K Freeze Dryer. After the addition of two stainless steel grinding balls and 1mL of 80% methanol, samples were sonicated for 5 min (2 sec on/off pulse cycle at 90 A) using a Qsonica Q700 Ultrasonic Processor with a water bath cup horn adaptor (Model 431C2). Following sonication, microfuge tubes were centrifuged at 10,000 RCF for 5 min in an Eppendorf 5417R centrifuge. 800µL of the resulting supernatant was transferred to a clean 4mL glass vial, and 800µL of fresh methanol added to the sample. The sample was sonicated and centrifuged as described, and the resulting supernatant was transferred to the same receiver vial and concentrated to dryness in an SC250EXP Speedvac Concentrator coupled to an RVT5105 Refrigerated Vapor Trap (Thermo Scientific). The resulting powder was suspended in 120µL of 100% methanol, followed by vigorous vortex and brief sonication. This solution was transferred to a clean microfuge tube and subjected to centrifugation at 20,000 RCF for 10 min in an Eppendorf 5417R centrifuge to remove precipitate. The resulting supernatant was transferred to an HPLC vial and analyzed by HPLC-MS.

HPLCMS analyses
Reversed-phase chromatography was performed using a Vanquish LC system controlled by Chromeoleon Software (ThermoFisher Scientific) and coupled to an Orbitrap Q-Exactive High Field mass spectrometer controlled by Xcalibur software (ThermoFisher Scientific). Methanolic extracts prepared as described above were separated on an Agilent Zorbax Eclipse XDB-C18 column (150 mm x 2.1 mm, particle size 1.

Data availability
All statistical analysis code and raw data necessary to reproduce these analyses are available [https:// github.com/SenguptaLab/ProvidenciaChemo.git]. Draft genome assemblies will be deposited in Genbank.