(P < 0.001; fig. S6D). Finally, pure pDCs induced
undetectable or low levels of T cell proliferation
in response to LPS or LPS + R848 stimulation,
respectively (P < 0.05; Fig. 5D). We conclude that
“pure” IFN-a–producing pDCs (depleted of AS
DCs) do not up-regulate CD86 (fig. S6, C and E),
are diminished in their ability to induce T cell
proliferation, and that contamination of AS DCs
within the traditionally defined pDC gate is likely
responsible for T cell stimulation activities measured in prior reports (18–20).
AS DCs stimulate T cell proliferation and
are present in tonsils
Because AS DCs expressed the costimulator CD86
and components of antigen presentation, we hypothesized that they could stimulate T cell proliferation (fig. S6, A, C, and E). Strikingly, both
AS DC subtypes were potent stimulators of allogeneic CD4+ and CD8+ T cell proliferation, unlike pDCs (P < 0.01), and were marginally superior
to CD1C+ and CLEC9A+ DCs (Fig. 5E).
Similar to other DCs, AS DCs expressed CLA
and CD62L but not CCR7 protein (fig. S6F), suggesting potential homing to peripheral tissue
such as skin and lymph node from the circulation. Because CD123+ pDCs were observed in the
T cell area of the human tonsil (21), we evaluated
whether CD123+ AS DCs were also present by
staining human tonsils with antibodies to CD123
and AXL. We found AS DCs adjacent to CD3+
T cells, admixed with CD123+AXL– pDCs (Fig. 5F).
Flow cytometry confirmed this finding, showing
that the CD123+CD11C–/lo AS DCs represented 0.7%
and CD123–CD11C+ AS DCs represented 1.7% of the
CD45+LIN–HLA-DR+ fraction (Fig. 5F). Thus, AS
DCs are able to stimulate T cells and are present
in the T cell zones of tonsils.
Identification of circulating
CD100hiCD34int cDC progenitors
Finally, we interrogated CD11C–CD123– cells within the HLA-DR+CD14– gate used for isolating DCs
that were not considered in the initial analysis
because they were not previously thought to include DCs (red dashed gate in Fig. 1B and updated
gate in Fig. 6A used for these experiments).
Analysis of CD11C–CD123– scRNA-seq data revealed six clusters in this gate (fig. S7, A and B).
Cells in cluster 6 expressed genes associated with
hematopoiesis, DC progenitors, and genes essential for DC development (e.g., SATB1, RUNX2, KIT,
HLX, ID2) (22–25) and were marked by high expression of the cell surface protein SEMA4D
(CD100). We therefore hypothesized that cluster
6 could represent a progenitor population.
To assess the progenitor potential of this compart-
ment, we cultured FACS-purified CD11C–CD123–
cells with MS5 stromal cells and cytokines that
induce DC differentiation (6), based on a pub-
lished human DC progenitor differentiation assay
(26). After several days in culture, the cells were
evaluated by flow cytometry, using a panel of
antibodies that identify pDCs and CD1C+ and
CD45+ immune cells for a more comprehensive
assessment. For comparison, under the same con-
ditions, we monitored the differentiation potential
of isolated pDCs, CD1C+ and CLEC9A+ DCs, and AS
DC subtypes (see fig. S7, C and D).
After 7 days of culture, cells isolated from the
CD11C–CD123– gate gave rise to CLEC9A+ and CD1C+
DCs but not pDCs, according to flow cytometry
and scRNA-seq analyses (Fig. 6B). We narrowed
down the search for the progenitor cells to the
CD45RA+CD39–CD100+ pool of cells based on
the unique cluster-6 marker CD100/SEMA4D
(fig. S7B), along with candidate markers that we
tested [based on DC progenitors in the bone marrow (CD45RA) and tissue DC (CD39) markers]
(Fig. 6C, fig. S5J, fig. S6, B and F, and fig. S7, B to
H). After iteratively testing each sorted population for differentiation potential, we discovered
that only the CD100hiCD34int cells generated
CLEC9A+ and CD1C+ DCs (Fig. 6C and fig. S7F).
scRNA-seq of CD100hiCD34int cells mapped these
cells to the original cluster 6, including the expression of the same DC differentiation and progenitor function genes (fig. S7B).
We validated the existence of CD100hiCD34int
progenitors in 10 individuals, with a frequency of
~0.02% of the LIN–HLA-DR+ fraction of PBMCs
(Fig. 6D). These cells were morphologically primitive, possessing high nuclear-to-cytoplasmic ratios and circular or indented nuclei (Fig. 6D),
in contrast to AS DCs, pDCs, and CD1C+ and
CLEC9A+ DCs (Fig. 5B). Although CD100hiCD34int
cells expressed HLA-DR and low levels of the
costimulatory molecule CD86 (fig. S6E) and lymph
node homing gene CCR7 (fig. S7, B and H), they
had low T cell stimulatory potential (Fig. 5C),
which suggests that these cells are not functional
cDCs. Furthermore, CD100hiCD34int cells retained
significant proliferative capacity (P < 0.05; Fig.
6E), in accordance with their primitive morphology, phenotype, and expression profile. Although
CD100hiCD34int cells were CD117/KIT+CD45RA+
and CSF1R/CD115–, CD1C–, CD141–, CD123–—a profile
similar to that of a previously reported circulating human DC progenitor (24, 27, 28)—they differ
from the published progenitor in having a more
primitive morphology and lacking CSF2R/CD116
and FLT3/CD135 expression (fig. S7, G and H).
Differentiation potential of AS DCs
When we seeded cultures with pDCs and CD1C+
and CLEC9A+ DCs, we found that they generally
retained the same phenotype throughout the
differentiation assay (Fig. 6F and fig. S7, I and J).
Upon observing a gene expression spectrum of
AS DC states that includes pDC-like and CD1C+-like
DC signatures (fig. S5, C to F), we also seeded AS
DCs to assess their potential to transition toward
other DC subsets [ensuring no contamination
with CD1C+ and CLEC9A+ DCs (fig. S7, I and J)].
After 7 days in culture, we observed cells with
high levels of CD1C (frequency 40 to 50%, n = 6
donors) and rare cells with surface CLEC9A and
CADM1 (0.5 to 0.8%) expression (Fig. 6F), irres-
pective of the FLT3L concentration used (Fig. 6F)
or whether the culture was seeded with either of
the two AS DC subpopulations representing both
ends of the spectrum (fig. S7K). Notably, both
AS DCs at day 0 and the cells differentiated from
AS DCs did not express BATF3 (a transcription
factor required for terminal differentiation of
CLEC9A+ DCs), CADM1, or XCR1, which are key
CLEC9A+ DC discriminative markers (table S2)
(23, 29–33) (fig. S5, D and E).
We found that AS DCs did not divide during
the transition into CD1C+ DCs, in contrast to
CD100hiCD34int cells that divided and differentiated into CD1C+ as well as CLEC9A+ DCs.
Furthermore, CD100hiCD34int differentiation into
CD1C+ DCs is not likely to transition through AS
DCs, because CD100hiCD34int did not express AXL
or SIGLEC6 genes at day 0 or during differentiation. AS DCs are thus functional cDCs that exist in
a continuum of states in vivo (fig. S5, C to F), with
the potential to transition toward CD1C+ DCs.
Mapping malignant cells from patients
to the healthy DC atlas
We leveraged our human DC atlas to compare
pathogenic cells driving blastic plasmacytoid
dendritic cell neoplasm (BPDCN), a rare and aggressive hematological malignancy previously
known as NK cell leukemia/lymphoma (34, 35),
to healthy DC populations. Because the ontogeny
of these cells remains unclear (34–38), we performed scRNA-seq on CD45+HLA-DR+CD123+
blasts from four BPDCN patients (n = 174 cells)
(6). The first principal component highlighted gene
sets clustering all four patients together with
healthy blood pDCs (Fig. 6G). Analysis of BPDCN
samples together with healthy DCs showed the
highest overlap with pDC and AS DC gene expression signatures (fig. S8A). Because pure pDC
and AS DC subsets coexpress many genes yet
have distinct biological functions (Figs. 4 and 5),
we further analyzed the genes overlapping among
BPDCN, pure pDCs, and cDCs (fig. S8B). Despite
sharing some pDC genes (e.g., NRP1, IL3RA, DERL3,
LAMP5, PTCRA, and PTPRCAP), several key genes
essential for pDC function were missing or were
expressed only slightly in patient cells (e.g., GZMB,
IRF7, CLEC4C/CD303, IRF4, and SLC15A4; fig.
S8B). Only a small number of cDC genes were expressed in patient cells, including SIGLEC6, LTK,
FCER1A, CD59, CADM1, and TMEM14A. Note that
all four patient samples shared a set of discriminative genes (fig. S8B and table S9) that included
several genes expressed in B cells (e.g., FCRLA,
IGLL1, TCL1A, and IGLL5; fig. S8C) or with hematopoietic progenitors (e.g., SOX4 and CLEC11A).
Collectively, our analysis suggests that although
BPDCN malignant cells express some key B cell
markers, they are most closely related to pDCs.
DCs and monocytes are defined according to a
combination of molecular markers, functional
properties, and ontogeny (39). However, it remains
unclear whether the expression of existing markers
tracks with the more complex internal states of cells.
To address this question, we determined the states
of blood DCs/monocytes through comprehensive
profiling of gene expression at single-cell resolution,
empirically inferred cell subtypes, identified optimal surface markers for purifying the hypothesized cell subtypes, and showed that prospectively
Villani et al., Science 356, eaah4573 (2017) 21 April 2017 7 of 12