574 | Nature | Vol 582 | 25 June 2020
Article
microglia in human embryos, we focused on macrophages sampled
from head tissues at CS11–CS23. Clustering analysis revealed that these
macrophages could be divided into five groups (HeadMac0–Head
Mac4), with features highly related to their developmental stage. The
signature of Head_Mac0 overlapped completely with that of YS_Mac1;
together with the observation that the yolk sac was the sole source
of macrophages at this time-point, this suggests that YS_Mac1 gives
rise to the macrophages found in the head at CS11 (Fig. 4a). A small
population of monocytes derived from YSMPs was also seen at CS17
in the head. However, the disparity between their number and those
of the other Head_Mac populations, as well as their late appearance,
together suggested that they are likely to make only a minor contribu-
tion (Extended Data Fig. 7a, b). The sequential emergence of the five
head macrophage populations indicated that these populations were
part of a continuum of developing cells, and their expression of classical
microglia-associated genes including CX3CR1, SALL1 and SPP1 further
supported the idea that these cells were moving towards a microglial
fate^22 ,^26 (Fig. 4b).
Next, we studied the dynamics of microglial specification by gener-
ating a profile of gene expression changes across the five Head_Mac
clusters based on their sequential real-time emergence between CS11
and CS23 (Fig. 4c, Supplementary Table 7). The changes largely followed
five patterns. Most genes in patterns 1 and 2 were downregulated and
some of them, such as CD163 and CD1C, were associated with immune
activity. We also witnessed increased expression of tissue development
and neurodevelopmental genes such as IGF1^26 and TMSB4X^27. A survey
of the transcription factor landscape revealed a similar trend (Fig. 4d),
with head macrophages losing expression of inflammatory transcrip-
tion factors such as IRF7 and STAT 1, and gaining microglial identity with
expression of BHLHE41, JUN, FOSB, NR4A1 and SALL1^22 ; these changes
mirror the pattern observed during mouse microglial development.
Finally, we integrated our embryonic data with publicly available
data for adult TRMs from the head, liver and lungs, along with our own
data for skin sampled from children aged eight and ten. The results
confirmed that specification had already occurred in the microglial
population in embryos, with the embryonic and adult microglia cluster-
ing together (Extended Data Fig. 8, Supplementary Table 8). Likewise,
specification towards liver TRM fate had begun, although to a lesser
extent than for head macrophages (Extended Data Fig. 7d–g, Supple-
mentary Table 9). This is in contrast with skin TRMs, where there was
only minor expression of CD207 at CS23 (Extended Data Fig. 7h) and
slight overlap between the embryonic and paediatric samples in the
integrated dataset (Extended Data Fig. 8a, b). Similarly, no commitment
had yet occurred in the embryonic lung (Extended Data Figs. 7c, 8a, b,
Supplementary Table 8).Discussion
Much of our current knowledge of embryonic haematopoiesis is based
on findings in animal models such as mice or zebrafish. This study
paves the way for a wide range of explorations and analyses that were
previously difficult to approach owing to the ambiguity surrounding
human yolk sac-derived macrophages and their progenitors. Although
it is widely acknowledged that mammalian haematopoiesis is highly
conserved^3 , the characterization of these cells in humans has so far
been restricted to either microscopic observations^3 ,^28 or explant experi-
ments^29. By leveraging the maturation of single-cell sequencing tech-
nology and bioinformatics, our analyses shed light on this issue in an
unbiased and unsupervised manner, while maintaining tissue site and
temporal information. Although it is still difficult to determine with
certainty the ontogeny of the various human TRM subsets without the
use of fate-mapping tools typically used in mouse models, we have iden-
tified two distinct HSC-independent waves of macrophages in humans
that correspond to those seen in mice. This is especially important in
the clinical context, as macrophages are essential regulators of tissue
development and homeostasis^30 , and understanding their functions
and developmental pathways is key to the diagnosis and treatment of
pathologies caused by their dysregulation. The contribution of these
HSC-independent waves must be considered when characterizing the
subsets of macrophages found in disease, as numerous studies have
established that HSC-independent macrophages maintain a distinct
transcriptomic and epigenetic identity from their HSC-derived coun-
terparts^31. Specifically, the human YSMPs that we have transcriptomi-
cally and functionally characterized here might correspond to mousea bCDH5 MYB AZU1 MPOUMAP2UMAP1CCR2 HLA−DRA CD163 MRC1log normalized expression^012345YS_Mac2
Liver_MacHead_Mac1Head_Mac2Skin_MacYS_Mac1Blood_MacLung_MacHead_Mac3Head_Mac4cd eUMAP1Myeloid groups reanalysedUMAP2Cluster
YSMP
GMP
Myeloblast
Monocyte
Blood_MacLiver_Mac
Lung_Mac
Skin_Mac
YS_Mac1
YS_Mac2
Head_Mac1
Head_Mac2
Head_Mac3
Head_Mac41 2 3 4 5 6 7 8 910
11
12
13
141
23
4567
89111012
13
14Cluster TMEM132ENDST3SPICID1CCL13LGALS9CCYP26A1BMXRNASE1PYGM
MMP1MSLN
HBE1PLOD2
BNIP3GSTA1FGAFGG
MYH6MZT2BSNRPFSEZ6LZNF300P1SLED1MAP6SOX2HAMPPDPNGLDNFLJ^41200
Liver_Mac
Blood_Mac
Lung_Mac
Skin_Mac
YS_Mac1
YS_Mac2
Head_Mac1
Head_Mac2
Head_Mac3
Head_Mac4–1.5–1.0–0.500.51.01.5Scaled expressionCS15
CS17
OthersPrimitive macrophageMacrophage(^) 1st
wave
(^2) nd wave
YSMP
Monocyte
UMAP1
UMAP2
CS11
CS12
CS13
Stage
Fig. 3 | Two distinct waves of yolk sac-derived macrophages contribute to
TRM populations in human embryos. a, UMAP visualization of
myeloid-related cells (n = 8 biologically independent embryo samples and 782
cells) with 14 re-clustered populations mapped on. b, Heat map showing scaled
expression of the top three DEGs between macrophage populations (n = 450
cells). DEGs were detected using FindAllMarkers function in Seurat (one-sided
Wilcoxon rank-sum test, with P value adjusted for multiple testing using
Bonferroni correction), and the genes with fold change >1.5 and adjusted
P < 0.05 were selected (Supplementary Table 6). c, Hierarchical clustering
based on Euclidean distance of macrophage populations (n = 450 cells).
d, Expression of the indicated genes projected on to UMAP of myeloid-related
clusters. e, UMAP visualization of myeloid-related clusters with cells from CS11
to CS17 (n = 6 biologically independent embryo samples and 268 cells) mapped
on, showing sequential appearance of macrophage populations.