Article
(R&D Systems, AF1700, polyclonal, KWT0419021, 1:200), mouse anti
EZRIN (Sigma-Aldrich, E8897, 3C12, 117M4875V, 1:500), rabbit anti-
CER1 (Sigma-Aldrich, HPA019917, polyclonal, R10176, 1:50), phalloi-
din (F-actin) Alexa Fluor488 (Thermo Fisher Scientific, A12379, direct
labelled, 1749905, 1:300), rabbit anti HESX 1 (Abcam, AB246949, poly-
clonal, GR3267093-1, 1:100), goat anti-Brachyury (T) (R&D Systems,
AF2085, polyclonal, KQP0617031, 1:200), goat anti FLK1 (R&D Systems,
AF357, polyclonal, Cvisceral endoderm0617081, 1:100) and WGA Alexa
Fluor 647 (Invitrogen, W32466, direct labelled, 1988457, 1:500). The
secondary antibodies were incubated at room temperature for 2 h,
and the slices were washed three times with 0.05% Tween-20. Pictures
were taken by Leica SP8 laser confocal microscope.
Whole-embryo staining and 3D reconstruction
Fixed embryos were permeabilized by 0.5% Triton X-100 in PBS over-
night in a 4 °C refrigerator. Embryos were blocked with 3% BSA in
PBS for 4 h at room temperature and then transferred to a new well.
Embryos were incubated with primary antibodies for 16–18 h at 4 °C.
Embryos were washed three times in PBS including 0.05% Tween-20,
and incubated the secondary antibodies for 4 h in room temperature.
Embryos were washed three times in PBS including 0.05% Tween-20
and transferred to a well of 8-well IbiTreat μ-plates (IB-80826, Ibidi
GmbH) with an aqueous solution of 60% glycerol aqueous solution to
take photographs.
To make 3D videos, a multiphoton microscope Leica TCS SP8 DIvis-
ceral endoderm was used. Embryos were mounted in 80% glycerol aque-
ous solution. Imaging was performed with a Leica TCS SP8 DIvisceral
endoderm multiphoton microscope with a HC Fluotar VISIR 25×/1.00
NA CLARITY-optimized immersion objective with motorized correc-
tion ring (Leica Microsystems). All z-stack images were acquired with
at a 1,024 × 1,024 pixel resolution and with a z-step of 1 μm. Three-
dimensional data were deconvoluted with Lightning module of LAS X
software (Leica Microsystems). Because of limited penetration ability
of multiphoton microscope, we took 200-μm-thick pictures in Z-axis.
Isolation of single cells
Embryos were washed in PBS three times, washed in 0.25% trypsin
(T4799; Sigma-Aldrich) twice, incubated with 0.25% trypsin for 15 min
at 37 °C and terminated by DFBS. Embryos were dissociated into single
cells by repeated pipetting and dispersed in 1% DFBS in PBS. A single
cell was pipetted into a PCR tube. All above operations were performed
using a Nikon SMZ645 microscopy.
Cell number counts and embryo diameter measurements
To count cell numbers, two protocols were used. First, the whole
embryo was dissociated into single cells by digestion with 0.25% trypsin,
and the total cell number per embryo was counted. The second was
used to count numbers of specific cell types. After staining, cell num-
bers of OCT4+ EPIs, GATA6+ PrE and CK7+ TrBs were analysed by ImageJ
software (v.1.51 j8). To measure the diameter of developing human
embryos, embryos were photographed every day and their diameters
were measured by the ImageJ software.
RNA sequencing of single cells
Isolated single cells were washed in DPBS (Gibco 14190-144) and picked
up using Pasteur pipettes under a dissecting microscope. The syn-
thesis and amplification of full-length cDNAs were performed follow-
ing Smart-seq2 protocol^42. In brief, single cells were washed in DPBS
(Gibco 14190-144) and picked into lysis buffer using Pasteur pipettes
under a dissecting microscope. Reverse transcription reactions and
pre-amplification were performed using SuperScript II (Invitrogen
18064-014) and KAPA HiFi HotStart ReadyMix (KAPA Biosystems
KK2601), respectively. The quality of the cDNAs was evaluated by Bio-
analyzer 2100. Library construction and sequencing were performed
by Annoroad Gene Technology (http://www.annoroad.com/) or BGI
(https://www.bgi.com/). Sequencing was performed on an Illumina
X-ten platform or a BGISEQ-500 sequencing platform (BGI). Pair-end
reads were obtained, and the number of the reads was more than 7 mil-
lion for every individual cell.Quality control, alignment of the scRNA-seq profiles and
stringent filtering
The sequencing qualities of 557 scRNA-seq profiles were examined with
the FASTQC (https://www.bioinformatics.babraham.ac.uk/projects/
fastqc/) and MULTIQC (v.1.6)^43. The annotation of RefSeq genes were
downloaded from UCSC Genome Browser^44. RefSeq exons were used
to build databases of exon and splice sites with the extract_exon.py
and extract_splice_sites.py in the hisat2 package^45. HISAT2 (v.2.1.0)^45
was used to align the scRNA-seq profiles to the human genome. The
alignment results of HISAT2 in the SAM format were converted to BAM
format and sorted with SAMTools (v.1.1)^46. Stringtie (v.1.3.4)^47 was used
to calculate the abundances of genes (in FPKM) annotated in GENCODE
(v.29)^48 ,^49 using the options of ‘-G gencode.v29.gtf -B -e –v’. Because the
cells with limited number of expressed genes were potentially caused
by RNA degradation, two scRNA-seq profiles with numbers of genes
with abundance levels more than 1 FPKM were smaller than 2,000 were
eliminated in further analysis. The basic information of the 555 remain-
ing scRNA-seq profiles were available in Supplementary Table 8.
Qualimap 2 (v.2.2.2-dev)^50 was used to calculate the number of
reads mapped to the genes in GENCODE genes with options of ‘–java-
mem-size=40G comp-counts -bam -pe’. Then, we prepared a plot of the
number of genes versus sequencing depth (the number of sequenc-
ing reads in the scRNA-seq library) with the command of ‘qualimap
counts -d -k 5’.The t-SNE and trajectory analysis of the scRNA-seq profiles
Genes with dispersion values of at least 0.5 in a particular cell type were
selected with the Seurat package (v.2.3.4) in R^51. The top variable genes
were used to classify cells with the FindClusters function in the Seurat
package. Cell types were defined by expression of selected markers.
Genes filtered with Seurat were used to perform t-SNE analysis for the
cell types under consideration using the RunTSNE function in the Seurat
package of R. Monocle (v.2.4.0)^52 ,^53 was used to perform a trajectory
analysis for the cell type under consideration. The heat map function
of the R platform was used to generate the heat map of selected marker
genes. Because AME did not express pluripotency genes (or expressed
them only at low levels), we sorted cells from the ICM and EPI clusters by
NANOG expression in analysing dynamics of pluripotency and primi-
tive streak genes to exclude the AME and intermediate state cells. One
hundred and thirty-six cells with FPKM values of NANOG more than 1
were specifically used for violin plots of pluripotency genes and gene
regulatory networks analysis.Identifying genes related to different cell types
A feature vector of one cell type was defined as a binary vector with
values of 1 for the cell types under consideration and 0 for other cell
types. Genes with Pearson’s correlation coefficients of least 0.4 with
the feature vector for a particular type of cell were regarded as genes
related to the cell type under consideration. One hundred and thirty-six
cells with FPKM values of NANOG more than 1 were specifically used for
violin plots of pluripotency genes and gene regulatory networks analy-
sis (136 cells reclassified were used in Extended Data Figs. 5g, l 9e, f–i).Comparisons with publicly available scRNA-seq from pre-
implantation human embryos
The previously described analytical strategies and datasets^15 were used
to combine and analyse our scRNA-seq data from 6–9-d.p.f. embryos
and the data (later blastocyst or 6–7-d.p.f. blastocysts) from three pre-
vious studies with PCA^16 –^18. The The 12 previously described lineage
marker genes^15 were used in the PCA analysis. The combined dataset