We paired homologous mouse and human neuron clusters by correlating mCH levels at homologous genes and found expanded neuronal diversity
in human FC relative to mouse FC (Fig. 2H and
fig. S12A) (19). Multiple human neuron clusters
showed homology to mouse L5a excitatory neurons (mL5-1), L6a pyramidal neurons (mL6-2), or
VIP, PV, and SST inhibitory neurons (Fig. 2H).
We found a unique gene-specific mCH pattern
and superenhancer-like mCG signatures in a potential human-specific inhibitory population (hPv-2;
figs. S12B and S16J) (9).
Although we detected substantial mCH in all
human and mouse neurons, cell types varied over
a wide range in terms of their genome-wide mCH
level (1.3 to 3.4% in mouse, 2.8 to 6.6% in human)
(fig. S13, A to F). The sequence context of mCH
was similar across all neuron types and consistent
with previous reports (fig. S13, I and J) (5, 7). Interestingly, global and gene-specific mCH differences were found in PV and SST inhibitory neurons
located in different cortical layers (fig. S14) (9).
Genes with low mCH in superficial-layer PV+
neurons are enriched in functional annotations including neurogenesis, axon guidance functions, and
synaptic component (fig. S14, F to H) (9), suggesting
layer-specific epigenetic regulation of synaptic
functions in inhibitory neurons.
A key advantage of single-cell methylome analy-
sis is the ability to obtain regulatory information
from the vast majority of the genome [>97% (19)]
not directly assessed by RNA sequencing. By pool-
ing reads from all neurons in each cluster, we
could find statistically significant differentially
methylated regions with low mCG in specific
neuronal populations (CG-DMRs), which are reli-
able markers for regulatory elements (5). We found
575,524 mouse (498,432 human) CG-DMRs with
average size of 263.6 bp (282.8 bp), covering 5.8%
(5.0%) of the genome (Fig. 3A, fig. S15A, and tables
S5 and S6). Most CG-DMRs (73.2% in mouse, 68.6%
in human) are located >10 kb from the nearest
annotated transcription start site (fig. S15, B to
E). mPv and mVip CG-DMRs showed the strongest
overlap with ATAC-seq peaks and putative enhancers identified from purified PV+ and VIP+
populations, respectively (fig. S15, G and H) (9, 20).
Hierarchical clustering of mCG levels at CG-DMRs
grouped neuron types by cortical layer and inhibitory neuron subtypes (fig. S15, I and J). Thus, neuron
type classification is supported by the epigenomic
state of regulatory sequences.
We inferred transcription factors (TFs) that play
roles in neuron type specification by identifying
enriched TF-binding DNA sequence motifs in CG-
DMRs (Fig. 3, B and C, and fig. S15K). We identified
known transcriptional regulators and observed that
several TF-binding motifs were enriched in human
but depleted in mouse CG-DMRs in homologous
clusters (Fig. 3C). The binding motif of NUCLEAR
FACTOR 1 (NF1) was enriched in CG-DMRs for two
human inhibitory neuron subtypes (hVip-2, hNdnf)
but was depleted in homologous mouse clusters
(mVip, mNdnf-2), suggesting a specific involve-
ment of NF1 in human inhibitory neuron speci-
fication. Thus, although the TF regulatory circuits
governing tissue types are conserved between
mouse and human (21), fine-grained distinctions
between neuronal cell types may be shaped by
species-specific TF activity.
Superenhancers are clusters of regulatory elements, marked by large domains of mediator binding and/or the enhancer histone mark H3K27ac,
that control genes with cell type–specific roles (22).
Extended regions of depleted mCG (large CG-DMRs) are also reliable markers of superenhancers
(fig. S16, A to C) (9, 23). Therefore, we used our
neuron type–specific methylomes to predict superenhancers for each mouse and human neuron
type (fig. S16, D to I, and tables S7 and S8). For
example, superenhancer activity was indicated by
a large CG-DMRs at Bcl11b (Ctip2) in a subset
of deep-layer neurons (fig. S16, F and G) and
broad H3K27ac enrichment in mouse excitatory
neurons (fig. S16F). Superenhancers overlap with
key regulatory genes in the associated cell type,
such as Prox1 in VIP+ and NDNF+ neurons (fig.
S16, H and I).
Global mCH and mCG levels were correlated
between homologous clusters across mouse and
human (Pearson r = 0.698 for mCH, r = 0.803 for
mCG; P < 0.005), suggesting evolutionary conservation of cell type–specific regulation of mC (Fig.
4A and fig. S13, G and H). Examining 12,157
orthologous gene pairs, we found stronger correlation of gene body mCH between homologous clusters in mouse and human (median Spearman r =
0.236; Fig. 4, B and C) than between different
cell types within the same species (r = –0.050,
SCIENCE sciencemag.org 11 AUGUST 2017 • VOL 357 ISSUE 6351 603
Fig. 3. Conserved and divergent neuron type–specific gene regulatory
elements. (A) Heat map showing differentially methylated regions (CG-DMRs)
hypomethylated in one or two neuron clusters; categories of DMRs
containing >1000 regions are shown. (B) TF binding motif enrichment
in CG-DMRs of homologous mouse and human clusters (false discovery
rate < 10−10). (C) Mouse- or human-specific enrichment and depletion of TF
binding motifs. Asterisks indicate TF binding motifs that are significantly
enriched in one species but depleted in the other.