Supplementary tables & figures: High resolution
pseudotemporal ordering of single-cell RNA-seq
profiles
Kieran Campbell, Chris P Ponting, Caleb Webber
[email protected]
Sept 15
Supplementary tables
Monocle classification
Embeddr classification
Number of cells
Non-mesenchymal
Mesenchymal
69
Mesenchymal
Mesenchymal
43
Non-mesenchymal
Non-Mesenchymal
159
Mesenchymal
Non-Mesenchymal
0
Supplementary Table 1 Classification of different cell types using Monocle and Embeddr. Embeddr
classifies significantly more cells as mesenchymal compared to non-mesenchymal.
Gene cluster
1
Category
Log10(q-val)
Description
GO:0000278
-43.79775
mitotic cell cycle
GO:0007049
-39.25601
cell cycle
GO:0000280
-38.43667
nuclear division
GO:1903047
-36.59466
mitotic cell cycle
process
2
GO:0048285
-35.83482
organelle fission
GO:0022402
-34.52133
cell cycle process
GO:0006613
-13.67789
cotranslational
1
protein targeting to
membrane
GO:0045047
-13.67789
protein targeting to
ER
GO:0030198
-13.67789
extracellular matrix
organization
GO:0043062
-13.67789
extracellular
structure
organization
GO:0006614
-13.67789
SRP-dependent
cotranslational
protein targeting to
membrane
GO:0072599
-13.56176
establishment of
protein localization
to endoplasmic
reticulum
3
GO:0061061
-24.65233
muscle structure
development
GO:0007517
-22.65526
muscle organ
development
GO:0003012
-19.99099
muscle system
process
GO:0006936
-19.03772
muscle contraction
GO:0030049
-18.35997
muscle filament
sliding
GO:0033275
-18.35997
actin-myosin
filament sliding
4
GO:0051084
-4.037972
‘de novo’
posttranslational
protein folding
GO:0006458
-3.962691
‘de novo’ protein
folding
GO:0006457
-3.556899
protein folding
Supplementary Table 2 Enriched GO biological process terms in different gene clusters. Only the top 6
in each cluster with a Benjamini-Hochberg corrected p-value < 0.01 are reported.
2
Marker genes identified by Embeddr and
Quake et al.
Marker genes identified by Quake et al. only.
Marker genes identified by Embeddr only
S100g Fabp5 Lamp3 Cd36 Scd1 Sftpb
Slc34a2 Sftpa1 Egfl6 Soat1 Bex2 Sftpc
Lcn2 Hc Trf Lyz2 Lyz1 Abca3 Muc1
4930420K17Rik Acsl4 Cited2 Cxcl15 Dlk1
Dram1 Etv5 Lpcat1 Mid1ip1 Napsa Nek6
Pgm2 Pla2g1b Ppp1r14c Rrp1 Sgms1
Supplementary Table 3 Marker genes for AT2 cells identified by differential silencing of genes on the
BP-AT1 transition.
Marker genes identified by Embeddr and
Quake et al.
Marker genes identified by Quake et al. only.
Marker genes identified by Embeddr only
Pdpn Rtkn2 Emp2 Cav1 Clic5 Lmo7 S100a6
Col4a3 Akap5 Cryab Sdpr
Aqp5 Ager S100a14
1190002H23Rik 2200002D01Rik
A930038C07Rik Ablim1 Actb Agrn Ahnak
Akap2 Anxa3 Arhgap29 Bola2 Cav2 Cd9
Cish Clic3 Col4a4 Crip2 Cttnbp2nl Dpysl2
Elovl5 Errfi1 Fam108c Fam174b Fbln5
Frmd4a Gm6548 Gprc5a Hes1 Hopx Hs2st1
Hspg2 Igf2bp2 Igfbp6 Il18r1 Ip6k2 Itgb6
Krt7 Lgals3 Limch1 Lims1 Magi3 Mal2
Mir682 Msln Myh14 Myl12a Myo1c Phyh
Pmp22 Ppp3ca Prdx6 Ptrf Pxdn Rps19
Rps19-ps3 Rps4y2 S100a11 Scnn1a Sec14l3
Sema3e Serpinb6b Sfn Slc1a5 Slco3a1 Sparc
Tcf7l2 Tgfb2 Timp3 Tinagl1 Tmem37
Trp53bp2 Tspan8 Usp46 Vegfa Yes1 Zfp703
Supplementary Table 4 Marker genes for AT1 cells.
3
Supplementary figures
Supplementary Figure 1 Pseudotime ordering of single-cell RNA-seq data. Beginning with a set of
unordered profiles, the algorithm will assign to each cell a pseudotime after which differential
expression across pseudotime can be found.
4
Supplementary Figure 2 Principal curve fitting in pseudotime. A principal curve (black solid line) is fit
through the centre of the 2-dimensional laplacian eigenmaps embedding of the cells (red). The arclength from the beginning of the curve to the orthogonal project of each cell (blue dotted line) is used
as to assign each cell a pseudotime.
5
Supplementary Figure 3 Laplacian eigenmaps emedding using entire gene set and clusters assigned
using reduced (high variance) genes.
6
Supplementary Figure 4 Expression of mesenchymal marker genes. A large number of cells designated
as non-mesenchymal by monocle show high expression of mesenchymal marker genes.
Supplementary Figure 5 Principal curve and pseudotime fitting to clusters 1 & 2 (differentiation
trajectory) of the Monocle dataset.
7
Supplementary Figure 6 Scatterplot of pseudotimes assigned by Monocle and Embeddr using the
Monocle dataset. Spearman correlation coefficient 0.8.
8
Supplementary Figure 7 Four clusters identified using complete linkage clustering on the predicted
values of the tobit regression models for pseudotime dynamics.
9
Supplementary Figure 8 Laplacian eigenmaps embedding and principal curve fitting (dashed, black)
for 5, 6, 8, 10, 15, 20 nearest neighbours in the embedding. The clusters are coloured by their original
assignment (k = 6), showing the method is highly robust to the choice of k.
10
Supplementary Figure 9 Embeddings and principal curves re-fit removing a randomly selected 50% of
cells. In each case cluster assignments remain well separated and the pseudotime trajectory fits
correctly.
11
Supplementary Figure 10 Spearman correlations of 30 pseudotime fittings using only 50% of cells
compared to the pseudotime fit using the full set. Median spearman correlation 0.98.
12
Supplementary Figure 11 Spearman correlations of 30 pseudotime fittings using only 50% of genes used
for original ordering. Median spearman correlation 0.91.
13
Supplementary Figure 12 Cell density as a function of pseudotime. BP cells are spread out along the
beginning of the pseudotime trajectory before moving into the AT1 and AT2 states with high density.
This suggests BP cells display more variability than AT cells, with AT1 cells representing the highest
transcriptional state. Note that in this representation, cells begin as BP and move `forwards' in
pseudotime to differentiate into AT1 cells or `backwards' in pseudotime to differentiate into AT2 cells.
14
Supplementary Figure 13 Pseudotime gene models for differentially expressed genes in the BP-AT1
transition. This transition is marked exclusively by the selective silencing of genes.
15
Supplementary Figure 14 Clusters of co-expression in the BP-AT2 transition. Three non-degenerate
clusters were identified corresponding to up-regulation (1), down-regulation (2) and transient upregulation (3).
Supplementary Figure 15 Differential gene expression as BP cells differentiate into AT cells. BP cells
(blue) differentiate into AT1 cells (purple) or AT2 cells (pink) mainly through selective silencing of
genes. The figure shows previously identified ‘perfect’ marker genes identified (purple), including
those missed by Embeddr (orange), as well as genes discovered exclusively by Embeddr (green). While
16
cells are placed horizontally in pseudotime order they were assigned but that the width of each is
constant for visualisation purposes (cellular density is not taken into account).
17