286 | Nature | Vol 578 | 13 February 2020
Article
(which encodes neuron-specific collagen) and Pantr2 (which is a Pou3f3-
adjacent non-coding transcript). To validate the nRNA-seq results, we
used fluorescence RNA in situ hybridization (RNAscope) to analyse,
at single-cell resolution, the relative expression of these genes in the
myenteric plexus of the small intestine and colon. As expected, the
hybridization signal was considerably stronger in enteric neurons of
the colon relative to the small intestine (Extended Data Fig. 2a). These
experiments demonstrate that the ostensibly homogeneous intrinsic
neural networks along the mammalian intestine express segment-
specific transcriptional programs.
To assess the contribution of microbiota to shaping the transcrip-
tional landscape of neural circuits along the gut, we also compared the
nuclear transcriptomes of enteric neurons from the small intestine and
colon of germ-free mice. Using the same statistical criteria, we identi-
fied 122 CUEGs that are upregulated in colonic neurons of germ-free
mice (hereafter, germ-free CUEGs) (Fig. 1c, Supplementary Table 2).
The differential expression along the gut of many SPF CUEGs, including
those we analysed by in situ hybridization, was maintained in germ-free
mice (Extended Data Fig. 2b). These findings are consistent with parallel
studies that demonstrated a similar expression of pan-neuronal and
neuron-subtype markers in the myenteric plexus of SPF and germ-free
mice (Extended Data Fig. 3a–f ); this suggests that the transcriptional
regionalization of the ENS along the mammalian gut is largely independ-
ent of microbial colonization, and that the effects of microbiota on
ENS physiology are mediated by a small number of critical molecular
pathways. To identify these pathways, we next compared directly the
nuclear transcriptomes of colonic neurons from SPF and germ-free
mice. We identified 25 genes (which we term microbiota-dependent
CUEGs), the transcripts of which were more abundant in colonic neu-
rons from SPF mice relative to germ-free mice (Fig. 1d, Supplementary
Table 3). Several of the microbiota-dependent CUEGs (such as Fam20c)
were absent from the list of SPF CUEGs, which suggests that they are
likely to be expressed at comparable levels throughout the ENS but are
under regulation by the microbiota specifically in colonic neurons.
However, three fully annotated genes—Ahr, Dand5 and Prr5—were also
present in the list of SPF CUEGs (Supplementary Table 1), indicating
that these genes are upregulated specifically in colonic neurons in
response to microbiota colonization. RNAscope analysis confirmed
the higher expression of Ahr, Prr5 and Fam20c in colonic neurons from
SPF mice relative to germ-free mice (Fig. 1e). Together, these studies
reveal a previously unappreciated complexity of gene expression in
the mammalian ENS and demonstrate that the transcriptional land-
scape of neural circuits along the gastrointestinal tract is shaped by the
integrated effect of host-specific genetic programs and environmental
factors such as the microbiota.
Next, we investigated the dataset of microbiota-dependent CUEGs as
a potential source of regulatory and effector genes that link the micro-
bial environment of the distal intestine with the functional output of
colonic neural circuits. Initially, we focused on Ahr because it encodes
a transcription factor with activity that is regulated by a broad range
of microbial, dietary and endogenous metabolites (AHR ligands), and
which functions as a biosensor that is critical for intestinal epithelial-
and immune-cell homeostasis^15 –^17. Upon ligand binding, cytosolic AHR
translocates to the nucleus and induces expression (among others) of
genes that encode cytochrome P450 (CYP1) enzymes, which metabo-
lize AHR ligands and thereby terminate AHR signalling^16. To provide
evidence for a microbiota–AHR–neural-output axis in the gut, we immu-
nostained the outer muscular layer (which includes the myenteric
plexus) from the intestine of SPF and microbiota-manipulated mice
for AHR. In the muscularis externa of the colon of SPF mice, AHR was
expressed predominantly in myenteric neurons, which indicates that
these neurons represent the main target of AHR ligands in this gut layer
(Fig. 2a, Extended Data Fig. 4a–c). Almost all colonic myenteric neurons
identified by the expression of the pan-neuronal markers HuC and HuD
(designated hereafter as HuC/D) and subtype-specific (ChAT, nNOS,
calretinin, calbindin and NF-M) markers were positive for AHR (Fig. 2a,
Extended Data Fig. 4d–g). Neurons exhibited either a cytoplasmic or a
nuclear signal (Fig. 2a), suggesting differential activation of AHR across
the population of enteric neurons. In contrast to the colon, myenteric
neurons in the duodenum and the jejunum did not exhibit an AHR signal
(Fig. 2b, Extended Data Fig. 4h), although a relatively weak signal was
detected in the neurons of the distal ileum (Extended Data Fig. 4i). The
lower expression of Ahr in myenteric neurons of the small intestine in
comparison to the colon was also confirmed by quantification of the
RNAscope in situ hybridization signal for AHR transcripts (Extended
Data Fig. 4a, j–l). Together, these findings suggest that the expression
of Ahr in enteric neurons is commensurate to the microbial load along
the gut. In support of the microbiota-dependent expression of Ahr,
myenteric neurons in the colon of germ-free mice and antibiotic-treated
mice had reduced levels of AHR transcripts and a considerably weaker
immunostaining signal, which was reinstated after colonization with
the microbiota of SPF mice (Figs. 1 e, 2c–e, Extended Data Fig. 4m–r).
On the basis of these experiments, we suggest that microbiota-induced
expression of Ahr in colonic neurons would enable lumen- or tissue-
derived ligands to activate AHR signalling in colonic neural circuits, and
thus contribute to the molecular profile and functional specialization
of these circuits.
To identify potential target and effector genes of AHR signalling in
the ENS, we next compared the nuclear transcriptomes of myenteric
neurons from the colon of control mice and mice treated with the
AHR ligand 3-methylcholanthrene (3MC)^18. Among the 30 genes that
showed the highest fold change with this treatment (which we term
AHR-induced CUEGs) (Extended Data Fig. 5a) were Ahrr and Cyp1a1,
which are known transcriptional targets of Ahr in several types of
non-neuronal cells^15 and which have important roles in the feedback
regulation of AHR signalling—either by repressing Ahr-dependent gene
expression (Ahrr) or by metabolizing AHR ligands (Cyp1a1)^15 ,^16. AHR-
dependent induction of Cyp1a1 in enteric neurons was confirmed by
quantification of the Cyp1a1-specific RNAscope signal in muscularis
externa from the colon of control and 3MC-treated mice (Fig. 2f–h),
and enhanced yellow fluorescent protein (eYFP) immunostaining
of the myenteric plexus from 3MC-treated Cyp1a1:cre;Rosa26eYFP
reporter mice^16 , in which activation of AHR signalling results in per-
manent expression of eYFP (Extended Data Fig. 5b–d). Querying the
list of AHR-induced CUEGs for potential regulators of neuronal func-
tion downstream of the microbiota–AHR axis, we identified Kcnj12
(Extended Data Fig. 5a), a gene that encodes the inwardly rectifying K+
channel, subfamily J member 12 (Kir2.2) that regulates the excitability
of cardiac muscle and neuronal cells^19. Previous studies have detected
an inwardly rectifying current in mammalian enteric neurons^20 , and
the addition of a specific blocker (ML-133) of the current driven by
the Kir2.x subfamily (Kir2.1, Kir2.2, Kir2.3 and Kir2.6)^21 to live prepa-
rations of myenteric plexus altered the electrically evoked firing of
enteric neurons (Extended Data Fig. 5e–g). Kcnj12 was found among
the top 30 microbiota-dependent CUEGs after marginal relaxation of
the P-value criteria (P < 0.06) (Supplementary Table 3), which raises the
possibility that this gene is regulated by microbiota and AHR signalling.
In support of this idea, RNAscope experiments showed covariance of
Kcnj12 and Cyp1a1 transcript levels in individual neurons in vivo after
administration of 3MC (Fig. 2i–k) and—similar to Ahr—expression of
Kcnj12 was reduced in myenteric neurons from germ-free mice and
antibiotic-treated mice (Fig. 2l–n, Extended Data Fig. 5h–j). We also
observed a correlation between Ahr and Kcnj12 expression in the enteric
neurons of the colon of SPF mice under normal conditions (Extended
Data Fig. 5k–m). To confirm that Ahr signalling regulates expression of
Kcnj12 in colonic neurons, we next used AAV-mediated gene transfer
(Fig. 1a) to generate mice with enteric-neuron-specific deletion of Ahr
(termed AhrEN-KO mice). An AAV9 vector expressing Cre recombinase
under the control of the neuronal CaMKII promoter (AAV9-CaMKII-
Cre), was administered to conditional Ahr mutant (Ahr fl/fl)^22 and control