mouse; r = –0.068, human). For homologous
clusters, we found shared and species-specific
CG-DMRs based on sequence conservation (
lift-over; fig. S17, A and B). Cross-species correlation
of mCG at CG-DMRs was significantly greater
for inhibitory than for excitatory neurons (P <
0.001, Wilcoxon rank sum test; Fig. 4D and fig.
S17C) (9). Greater sequence conservation at inhibitory neuron CG-DMRs could partly explain
the greater regulatory conservation (P < 0.001,
Wilcoxon rank sum test; Fig. 4E). Sequence conservation was observed only within 1 kb of the
center of inhibitory neuron CG-DMRs and did not
extend to the flanking regions (fig. S17G). These
results support conservation of neuron type–
specific DNA methylation, with greater conservation of inhibitory than of excitatory neuron
Single-cell methylomes contain rich information
enabling high-throughput neuron type classifi-
cation, marker gene prediction, and identifica-
tion of regulatory elements. Applying a uniform
experimental and computational pipeline to mouse
and human allowed unbiased comparison of neu-
ronal epigenomic diversity in the two species. The
expanded neuronal diversity in human, revealed
by DNA methylation patterns, is consistent with
more complex human neurogenesis, such as the
presence of outer radial glia and the potential
dorsal origin of certain interneuron subtypes
(15, 24, 25). Further anatomical, physiological, and
functional experiments are needed to charac-
terize the DNA methylation–based neuronal
populations defined by our study. Single-neuron
epigenomic profiling allowed the identification
of regulatory elements with neuron type–specific
activity outside of protein-coding regions of the ge-
nome. We expect that the single-nucleus methylome
approach can be applied to studies of disease, drug
exposure, or cognitive experience, thereby enabl-
ing examination of the role of cell type–specific
epigenomic alterations in neurological or neuro-
REFERENCES AND NOTES
1. B. J. Molyneaux, P. Arlotta, J. R. L. Menezes, J. D. Macklis,
Nat. Rev. Neurosci. 8, 427–437 (2007).
2. A. Zeisel et al., Science 347, 1138–1142 (2015).
3. B. Tasic et al., Nat. Neurosci. 19, 335–346 (2016).
4. B. B. Lake et al., Science 352, 1586–1590 (2016).
5. A. Mo et al., Neuron 86, 1369–1384 (2015).
6. A. Kozlenkov et al., Nucleic Acids Res. 44, 2593–2612
7. R. Lister et al., Science 341, 1237905 (2013).
8. A. Mo et al., eLife 5, e11613 (2016).
9. See supplementary materials.
10. S. A. Smallwood et al., Nat. Methods 11, 817–820
11. M. Farlik et al., Cell Rep. 10, 1386–1397 (2015).
12. C. Angermueller et al., Nat. Methods 13, 229–232 (2016).
13. Y. Huang et al., PLOS ONE 5, e8888 (2010).
14. L. van der Maaten, G. Hinton, J. Mach. Learn. Res. 9,
15. C. P. Wonders, S. A. Anderson, Nat. Rev. Neurosci. 7, 687–696
16. M. Ester, H. Kriegel, J. Sander, X. Xu, KDD’96 Proceedings of
the Second International Conference on Knowledge Discovery
and Data Mining (Association for the Advancement of
Artificial Intelligence, 1996), pp. 226–231; www.aaai.org/
17. E. S. Lein et al., Nature 445, 168–176 (2007).
18. S. A. Sorensen et al., Cereb. Cortex 25, 433–449
19. F. Yue et al., Nature 515, 355–364 (2014).
20. Y. He et al., Proc. Natl. Acad. Sci. U.S.A. 114, E1633–E1640
21. A. B. Stergachis et al., Nature 515, 365–370 (2014).
22. W. A. Whyte et al., Cell 153, 307–319 (2013).
23. M. D. Schultz et al., Nature 523, 212–216 (2015).
24. D. V. Hansen, J. H. Lui, P. R. L. Parker, A. R. Kriegstein, Nature
464, 554–561 (2010).
25. A. A. Pollen et al., Cell 163, 55–67 (2015).
We thank C. O’Connor and C. Fitzpatrick at Salk Institute Flow
Cytometry Core for sorting of nuclei; U. Manor and T. Zhang
at Salk Institute Waitt Advanced Biophotonics Core for assisting
with imaging; R. Loughnan for assistance of data analysis;
and J. Simon for assisting with illustration. Supported by
NIH BRAIN initiative grants 5U01MH105985 and 1R21MH112161
(J.R.E. and M.M.B.) and 1R21HG009274 (J.R.E.) and by NIH
grant 2T32MH020002 (C.L.K.). T.J.S. and J.R.E. are investigators
of the Howard Hughes Medical Institute. Data can be downloaded
from NCBI GEO (GSE97179) and http://brainome.org. L.K.
is an inventor on a patent application (US 14/384,113)
submitted by Swift Biosciences Inc. that covers Adaptase.
The source code for bioinformatic analyses is available
at https://github.com/mukamel-lab/snmcseq and
Materials and Methods
Figs. S1 to S17
Tables S1 to S9
31 March 2017; accepted 13 July 2017
604 11 AUGUST 2017 • VOL 357 ISSUE 6351 sciencemag.org SCIENCE
Fig. 4. Gene body mCH and CG-DMRs conserved
between mouse and human. (A) Global mCH and
mCG levels are strongly conserved within homologous
cell types between mouse and human. (B) Cross-species correlation of gene body mCH at orthologous
genes shows cell type–specific conservation. Black
boxes denote homologous neuron clusters. (C) The
median correlation of gene body mCH for homologous
clusters is higher than the within-species correlation
for distinct clusters. (D) Cross-species correlation
of mCG at neuron type–specific CG-DMRs. (E) Sequence
conservation at neuron type–specific DMRs.