PreprintPDF Available
Preprints and early-stage research may not have been peer reviewed yet.

Abstract and Figures

In colorectal cancer, oncogenic mutations transform a hierarchically organized and homeostatic epithelium into invasive cancer tissue lacking visible organization. We sought to identify differences in cellular composition between normal colon and colorectal cancer, and to define signals controlling cancer cell development. We used single cell RNA and protein profiling to analyze tumors and matched normal tissues of twelve colorectal cancer patients. RNA metabolic labelling followed by single cell RNA sequencing in patient-derived normal colon and colorectal cancer organoids was employed to define colorectal cancer cell developmental trajectories. We find that colorectal cancer tissues exhibited consistent changes in cellular composition in the epithelial, immune and stromal compartments across patients compared to normal colon. Tumor epithelial cells displayed patient-specific gene expression often correlating with somatic copy number alterations, but mainly organized into patient-overarching clusters. These clusters were defined by cell type-specific transcriptional programs related to stem, transient-amplifying and immature goblet cells, and showed differential expression for signatures of oncogenic traits such as replication stress. Patient-derived colorectal cancer organoids exhibited developmental trajectories forming along a gradient of mitogen-activated protein kinase activity that were Wnt-independent. Likewise, colorectal cancer cell types of patient samples were organized by mitogen-activated protein kinase activity. Our single-cell analyses provide a subtyping system for colorectal cancer cells based on transcriptional readout of morphogenetic signals and oncogenic traits. We provide evidence that mitogen-activated protein kinase signaling, a key pathway for targeted therapy, is a main driver of colorectal cancer cell plasticity.
Content may be subject to copyright.
1
A census of cell types and paracrine interactions in colorectal cancer
Florian Uhlitz1,2,3,*, Philip Bischoff1,*, Anja Sieber1,2, Benedikt Obermayer5, Eric Blanc5, Mareen
Lüthen1,3, Birgit Sawitzki6, Carsten Kamphues3,4, Dieter Beule5, Christine Sers1,3,7, David Horst1,3,7, Nils
Blüthgen1,2,3,7,# and Markus Morkel1,3, 7,#
1 Institute of Pathology, Charité Universitätsmedizin Berlin, Charitéplatz 1, 10117 Berlin, Germany
2 IRI Life Sciences, Humboldt University of Berlin, Philippstrasse 13, 10115 Berlin, Germany
3 German Cancer Consortium (DKTK), German Cancer Research Center (DKFZ), 69120 Heidelberg,
Germany
4 Department of Surgery, Charité Universitätsmedizin Berlin, Hindenburgdamm 30, 12203 Berlin
5 Core Unit Bioinformatics (CUBI), Charité Universitätsmedizin Berlin, MDC Berlin, and Berlin Institute
of Health, Charitéplatz 1, 10117 Berlin
6 Institute of Medical Immunology, Charité Universitätsmedizin Berlin, Augustenburger Platz 1,
13353 Berlin
7 Berlin Institute of Health, Anna-Louisa-Karsch-Straße 2, 10178 Berlin, Germany
* joint first authors
#corresponding authors: markus.morkel@charite.de, Tel: ++49-30-450 536 107;
nils.bluethgen@charite.de, Tel: ++49-30-2093 92 390
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
2
Abstract
In colorectal cancer, oncogenic mutations transform a hierarchically organized and homeostatic
epithelium into invasive cancer tissue. To define differences in cellular composition between the
normal colon and colorectal cancer, and to map potential cellular interactions between tumor cells
and their microenvironment, we profiled transcriptomes of >50,000 single cells from tumors and
matched normal tissues of eight colorectal cancer patients. We find that tumor formation is
accompanied by changes in epithelial, immune and stromal cell compartments in all patients. In the
epithelium, we identify a continuum of five tumor-specific stem cell and progenitor-like populations,
and persistent multilineage differentiation. We find multiple stromal and immune cell types to be
consistently expanded in tumor compared to the normal colon, including cancer-associated
fibroblasts, pericytes, monocytes, macrophages and a subset of T cells. We identify epithelial tumor
cells and cancer-associated fibroblasts as relevant for assigning colorectal cancer consensus molecular
subtypes. Our survey of growth factors in the tumor microenvironment identifies cell types responsible
for increased paracrine EGFR, MET and TGF-β signaling in tumor tissue compared to the normal colon.
We show that matched colorectal cancer organoids retain cell type heterogeneity, allowing to define
a distinct differentiation trajectory encompassing stem and progenitor-like tumor cells. In summary,
our single-cell analyses provide insights into cell types and signals shaping colorectal cancer cell
plasticity.
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
3
Introduction
All cells in the human body exist in contact with other cells in finely tuned microenvironments.
Paracrine communication between cells ensures tissue homeostasis. Cancer cells are compromised in
their ability to maintain homeostasis, as the oncogenic mutations activate signaling pathways cell-
intrinsically and render cancer cells less reliant on paracrine signals1. Furthermore, cancer cells induce
remodeling of neighboring tissues, for instance by secreting growth factors not found in the normal
environment2. Thirdly, cancer cells are often immunogenic or associated with inflammation, and
therefore attract immune cells3. These processes intersect and result in the emergence of a
qualitatively and quantitatively unbalanced cellular ecosystem in cancer tissue. Interactions between
immune, stromal and cancer cells are critical for tumor progression and therapy response4,5.
Colorectal cancer (CRC) most often initiates via mutations activating Wnt/β-catenin signaling that
maintains stem cells in the normal colon epithelium, while subsequent mutations deregulate further
signaling pathways such as the RAS-RAF-MEK-ERK signaling cascade6,7. Less frequently, CRC can arise
by initiating mutations in BRAF, or from chronic inflammation increasing the mutation rate in the
tissue. Regardless of the order of oncogenic mutations, genetic CRC drivers have direct and indirect
effects on the cellular composition of CRC and its microenvironment. There is substantial evidence for
the existence of tumor cell subpopulations in CRC. For instance, cancer stem cells with high clonogenic
potential can be sorted from CRC based on surface markers like PROM1 (also known as CD133) or
LGR58–10. Furthermore, CRC cells at the invasive front express matrix metalloproteinases such as MMP7
and display epithelial-to-mesenchymal transition, in contrast to cells residing in more central locations
in the tumor 1113. However, it has not been investigated systematically how cell types and cell plasticity
differ between the normal colon and CRC.
Here, we use droplet-based single-cell RNA sequencing to profile cell types and their differentiation
states in normal colon and tumor tissues of eight CRC patients, and in matching CRC organoids. We
identify consistent changes occurring in epithelial differentiation programs between normal and tumor
epithelium in the colon, resulting in the emergence of tumor-specific epithelial cell clusters expressing
genes relevant for cancer traits such as stemness, invasion, and epithelial-to-mesenchymal transition.
We also catalog cell types expanded in the tumor microenvironment, for instance, stromal cancer-
associated fibroblasts and several types of immune cells. We provide evidence for cancer-cell-specific
re-wiring of morphogenetic signaling informed by oncogenic mutations and differences in paracrine
signaling networks.
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
4
Results
Single-cell RNA sequencing of CRC
To capture the cellular diversity in CRC and track changes from normalcy to disease, we performed
single-cell transcriptome analysis of eight previously untreated CRC patients (Fig. 1A). We utilized
tissue samples that included the invasive tumor front and matched normal tissues (Supplementary Fig.
1). Tumors under investigation encompass stages pTis (Tumor in situ) to pT4, that is, from cancer
confined within the lamina propria to invasive through the visceral peritoneum, with or without
metastasis, and with various locations along the cephalocaudal axis of the colon. Panel sequencing of
genomic tumor DNA uncovered mutations in APC, KRAS and/or TP53 in tumors P008, P009, P013, P016,
and P017; these mutations are characteristic for the canonical CRC progression pathway initiated by
loss of APC. P007 harbored BRAFV600E and TP53 mutations; this mutational pattern is in line with a
tumor initiated by BRAF activation. P008 carried a TP53 mutation and was colitis-associated. Tumor
P014 contained putative driver mutations in APC, BRAF, HRAS, and PIK3CA, albeit at a lower frequency,
suggesting the possibility of distinct subclones contributing to this tumor.
We enzymatically dissociated the fresh normal and cancer tissues to single cells, produced single-cell
transcriptome libraries from each tissue using a commercial droplet-based system, and sequenced the
libraries to obtain transcriptomes covering 500 to 5,000 genes per cell. Singe-cell profiles were
partitioned into epithelial, immune or stromal transcriptome subsets for each library, using known
marker genes14, and then merged. We observed varying fractions of cell types per library, but stromal
cells were generally less abundant (in total across all libraries: >25,000 epithelial cells; >25,000 immune
cells and 2,691 stromal cells). Fluctuations in epithelial, immune or stromal cell abundance between
libraries could reflect differences in tissue cell content or technical variances related to ischemic time
during operation or the dissociation process.
We observed that single-cell transcriptomes derived from all patients intermingled within the
epithelial, immune, and stromal compartments (Fig. 1B). When distinguishing normal versus tumor
samples, distributions of single-cell profiles largely overlapped, although several regions within each
plot were preferentially inhabited by transcriptomes derived from either normal or tumor tissues (Fig.
1C). This indicates that our data are generally free from patient- or sample-specific batch effects
confounding the cell type distributions, but that general differences occur between normal and tumor
transcriptome distributions.
Cell type census in CRC versus normal colon
We next clustered the single-cell profiles of the epithelial, immune and stromal compartments, and
used cell-type-specific signatures and marker genes to annotate cell types14 (Fig. 2A; Supplementary
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
5
Fig. 2 for genes over- or underrepresented between tumor and normal samples). In the normal
epithelium, we identified a zone populated by profiles of undifferentiated cells with high activity of
stem cell markers such as OLFM4 and transiently-amplifying proliferative markers. This region
bordered on transcriptome clusters annotated as enterocyte progenitors, and, ultimately, mature
absorptive enterocytes with high expression of markers such as KRT20 and FABP1. BEST4- and OTOP2-
expressing enterocytes formed a discrete cluster (Fig. 2A, identified by the lightest shade of green) 15.
Further separate epithelial clusters were populated by profiles annotated as immature and mature
secretory goblet cells defined by expression of MUC2, TFF1, and TFF3, and TRMP5-expressing tuft cells.
In the tumor samples, the zone of undifferentiated epithelial cells was expanded by five largely tumor-
specific clusters (TC1-5, Fig. 2A, B; Supplementary Fig. 3 for data per patient). In contrast, clusters of
differentiated absorptive and secretory cell transcriptomes were reduced in size. Profiles representing
tuft cells and BEST4/OTOP2-positive enterocytes were vastly underrepresented in the tumor cell
libraries.
We used immunofluorescence to verify the spatial distributions of epithelial cell types, using marker
genes identified in the single-cell data (Fig. 2C, D). We detected the stem cell marker OLFM4 exclusively
at the base of normal crypts. However, in tumor sections, OLFM4, as well as the proliferation marker
MKI67, stained cells scattered throughout the epithelium, as validated by co-staining with the
epithelial marker EPCAM. The goblet cell and enterocyte differentiation markers TFF3 and FABP1
stained preferentially cells in the lower and upper crypt of the normal colon, respectively. In contrast,
TFF3- and FABP1-positive tumor cell populations were not clearly organized in domains. Clusters of
TFF3- and FABP1-positive cells that were largely negative for MKI67 suggest the presence of
differentiated cells in CRC, in agreement with the single-cell sequencing data.
The five tumor-specific epithelial cell clusters TC1 to TC5 were represented in different proportions in
all eight CRCs under investigation (Fig. 2E). TC1 and TC2 were assigned as highly stem cell-like using
prior classifiers14, while TC3-4 showed the strongest similarity to transiently-amplifying cell types and
TC5 shared similarity with both, transient-amplifying and stem cells. Transcriptome-based cell cycle
analysis revealed that TC1 is highly proliferative (Fig. 2F). Furthermore, TC clusters shared a couple of
defining genes that have previously been linked to oncogenic processes in CRC (Supplementary Fig. 4;
Supplementary Table 1). The CRC stem cell marker CD4416 was expressed prominently in TC1 and TC4.
MMP7, encoding a matrix metalloproteinase responsible for CRC invasion11, was expressed highest in
TC4, but also in TC1 and TC3. VIM17 and S100A418, key markers of the epithelial-to-mesenchymal
transition, were among the genes defining TC2. PLA2G2A, encoding a phospholipase that controls
inflammation and homeostasis in the intestinal stem cell niche19, and the intestinal stem cell marker
OLFM4 was overrepresented in TC5.
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
6
We annotated immune transcriptome clusters by lineage-specific marker genes (Fig. 2A,
Supplementary Fig. 5). Two main clusters of myeloid cells were assigned as monocytes and
macrophages by expression of CD14 and CD68. ITGAX (encoding CD11c) was also expressed in this
domain of the UMAP, indicating that these clusters, in addition, also encompass dendritic cells. A large
cluster of T cell profiles, as identified by CD3, could be subgrouped into CD8-positive cytotoxic T cells
and CD4-positive T helper cells. Subclusters of conventional and regulatory T helper cells were assigned
by higher relative levels of IL7R (encoding CD127), FOXP3 and IL2RA (encoding CD25). B cell and plasma
cell clusters were defined by CD19, MS4A1 (encoding CD20) and SDC1 (encoding CD138), respectively.
Among T, B, and plasma cells we could distinguish several subclusters that were represented in tissue
samples across the patients. Six of the 26 immune cell clusters in our analysis were expanded in all
eight CRC. These comprise the macrophage/monocyte clusters, regulatory T cells, two clusters of
plasma cells (termed Plasma 5 and 8, see Supplementary Table 2) and one cluster of CD8-positive T
cells expressing high levels of IL17A (CD8+ cluster 4, Supplementary Table 2). This interleukin has been
implicated in CRC progression20, and a similar type of T cell was recently found expanded in single-cell
analyses of colitis patients14.
Among stromal transcriptomes, we annotated an interconnected supercluster of fibroblasts.
Strikingly, one fibroblast cluster was confined to the tumor samples and was therefore designated as
cancer-associated fibroblasts (CAFs). CAF transcriptomes were defined by high expression of matrix
metalloproteinase-encoding genes such as MMP1, MMP11, MMP3 and MMP2, suggesting roles of
these cells in the degradation of the tumor extracellular matrix. The fibroblast supercluster, in addition,
contained profiles of putative crypt base fibroblasts of the stem cell niche expressing the Wnt ligand
WNT2B and the Wnt amplifier RSPO321,22, upper crypt fibroblasts expressing genes such as BMP2 and
BMP4 encoding differentiation-associated growth factors23 (Fig. 2A, Supplementary Fig. 6,
Supplementary Table 3) and a further small cluster of fibroblasts positive for various chemokine ligands
and receptors including CCL2, CCL8, CCL11, CCL13, CXCL1 and CXCL14. Further distinct clusters of
stromal cells were composed of myofibroblasts, possibly intermingled with smooth muscle cells,
expressing ACTA2 and DES and pericytes marked by MCAM (encoding MUC18/CD146) and STEAP4. We
also detected small numbers of endothelial cells and glial cells, respectively. Pericytes and endothelial
cells were more frequent in the tumor samples, but, in contrast to CAFs, these cells were also present
in normal tissue samples at low frequencies.
Paracrine signaling in CRC ecosystems
As we discovered multiple clusters of epithelial tumor cells (TC1-5) and expanded clusters of stromal
and immune cells in their microenvironment, we investigated possible paracrine interactions. For this,
we mapped cognate ligand-receptor pairs in our single-cell data, taking into account expression levels
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
7
of ligands, the prevalence of the ligand-expressing cell (that is, cluster size for that cell type), and
fractions of receptor-expressing cells. We focused on ligand-encoding genes active in immune or
stromal cells and genes encoding matching receptors in proliferative epithelial cells, that is, in stem/TA
cells and the five tumor-specific clusters TC1-5 (Fig. 3A, Supplementary Fig. 7A, and Supplementary
Table 4). Possible ligand-receptor connections appeared relatively sparse in the normal tissue.
However, we found many more potential paracrine signaling connections in the tumor. This was mainly
due to three features of the tumor ecosystem: Firstly, CRC contained novel epithelial cell types TC1-5
expressing multiple receptors, including high levels of the receptor tyrosine kinase MET24 and others
(Supplementary Fig. 7B). Secondly, several immune and stromal cell populations expressing cognate
ligands for the receptors in the epithelium were expanded (Supplementary Figs. 3 and 7B;
Supplementary Table 5). In particular, CAFs dominated signaling in the microenvironment of tumors
P008 and P016, and, to a lesser extent, P007 and P017, and express genes encoding ligands of cancer-
relevant signaling pathways. Among these are GREM1, WNT2B, WNT5A, HGF, multiple EGFR ligands
including AREG and many others that have been linked to CRC initiation and progression2528. Finally,
some cell types in the cancer microenvironment expressed additional ligands compared to their
normal tissue counterparts, for instance, crypt-base-like fibroblasts in the tumor express FGF7 and IGF
(Supplementary Fig. 7B).
We took a detailed look at Wnt signaling, as this pathway drives stem cell maintenance and tumor
initiation in the gut. A signature of Wnt/β-catenin target genes was most active in the TC1-5 and stem
cell compartment of tumor epithelium (Fig. 3B), while activity was lower among the differentiated CRC
cells, similar to the normal colon epithelium. Indeed, it has been shown that Wnt/β-catenin is dynamic
in CRC and can be activated by Wnt ligands29,30, although the pathway is frequently activated by loss
of APC. We detect the highest connectivity for Wnts and the R-Spondin family of Wnt amplifiers
between CAFs, expressing WNT2 and WNT5A, and stem cells and the tumor-specific cell clusters (Fig.
3A).
We next investigated EGFR-RAS-RAF-ERK signaling that plays a central role in CRC development and is
a key target of therapy. ERK-regulated genes31 were most active in TC4 (Fig. 3B). We identified CAFs,
endothelial cells, but also immune cells as potential sources of EGFR family ligands in the CRC
microenvironment (Fig 3A). Monocytes and macrophages that we found consistently enriched in
tumor versus normal tissues, express the ligand-encoding genes AREG, EREG, and HBEGF (Fig 3C and
Supplementary Table 3). These ligands could play roles in the activation of the EGFR-RAS-RAF-ERK
cascade in CRC, particularly in tumors lacking mutations in RAS, RAF or other activating components
of the pathway. Macrophages and CAFs also express HGF, encoding the MET receptor tyrosine kinase
ligand driving cancer progression at the invasive tumor front13. Indeed, we found macrophages to be
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
8
enriched in tumors specifically at the invasive front (Fig. 3C, D), suggesting that these could play roles
in the induction of epithelial-mesenchymal transition via HGF-MET.
We additionally detected TGF-β target gene activity, along with expression of a gene signature of
epithelial-to-mesenchymal transition in the tumor cell cluster TC4, and to a lesser degree in TC2 (Fig.
3B, E). Genes encoding TGF-β ligands were expressed in multiple stromal cell types, including CAFs,
and cognate receptors were present in the TC clusters (Fig. 3E, Supplementary Fig. 7A, B). While the
connectivity of the TC2 cluster was generally low, maybe also due to lower sequencing depth per cell,
the cluster was defined by expression of VIM and S100A4 (see above, Supplementary Fig. 4 and
Supplementary Table 1), supporting the association of TC2 cells with the epithelial-to-mesenchymal
transition. Interestingly, the size of cluster TC2 was highly correlated with the size of the CAF
population in the cancer microenvironment in the eight tumor profiles (Fig. 3F), suggesting a potential
role for CAFs in the support of the tumor cells clustering in TC2.
Cell type composition informs Consensus Molecular Subtypes
Consensus molecular subtypes (CMS) represent a transcriptome-based classification system for CRC
with clinical utility32. We applied the CMS classifier to our single-cell data and found strong inter- and
intratumor heterogeneity. Epithelial cells were exclusively assigned to CMS1-3 (Fig. 4A). The
continuous cluster comprising intestinal stem cells, TA cells, and enterocytes of the normal tissue, as
well as the TC1-5 tumor cell subtypes consists mainly of intermingled CMS1 and CMS2 cells. In our
limited set of CRCs, we find that epithelial cells of the two cancers probably arising via serrated
precursors (P007) or inflammation-induced progression (P008) are scoring predominantly CMS1, while
epithelial cells of the other cancers score mainly as CMS2 (Supplementary Fig. 7). CMS3 appears to be
confined to cells differentiating into the secretory lineage.
CMS subtypes were also unevenly distributed in the tumor microenvironment. Almost all immune cells
were assigned CMS2, and only a minority scored as the “immune” subtype, CMS1 (Fig. 4B). Stromal
cells were mostly CMS2, but a minority population of normal fibroblasts that we assigned previously
as potential crypt base fibroblasts were CMS4 (Fig. 4C). Most interestingly, CAFs in the tumor tissue
provided a strong CMS4 component that in our tumors was most prominent in P008 and P016
(Supplementary Fig. 7). As CMS4 was assigned to fibroblasts exclusively, we suggest that this cell type
also drives the mesenchymal” CMS4 assignment from bulk CRC tissue. It is of note that CMS4 cancers
have a worse clinical prognosis which may, therefore, also be linked to the presence of stromal tumor-
specific fibroblasts. In summary, we found that different cell types of the tumor ecosystem are
preferentially assigned to different CMS, and thus, that bulk tissue CMS assignments report
combinations of cell state and cell type prevalence in the tumor.
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
9
Cell type composition of patient-derived CRC organoids and matched tumor samples
We established organoid lines of the tumor samples P009 and P013, using standard culture conditions
with media containing EGF, FGF, and p38- and TGF-β inhibitors (Fig. 5A)33,34. P009 tumor tissue initially
grew out unevenly, however, formed a uniformly spheroidal organoid culture within three passages.
P013 tumor tissue grew out swiftly and uniformly, forming complexly folded organoids. We used panel
sequencing to confirm the identity of the organoids with the matched tumor tissue on a mutational
level (Supplementary Table 6).
We used cell suspensions of the organoid lines for single-cell RNA sequencing to generate profiles
before the first passage (designated as p0), essentially sequencing the primary tissue after one to two
weeks of expansion in culture. We also sequenced transcriptomes of the established lines P009 and
P013 after two and three passages, respectively. Single-cell profiles of organoids were of a higher
quality than those from epithelial cells of the tumors, as judged by the uniformly low fraction of
mitochondrial reads, despite having used similar conditions for disaggregation at 37°C. To exclude
confounding effects of technical differences in the different single-cell profiles, we anchored the
organoid profiles in the space of the matched epithelial tumor transcriptomes35.
In the resulting integrated data set, organoid cell transcriptomes of the two patients and the different
passage numbers intermingled (Fig. 5B). Next, we re-clustered the transcriptomes and assigned cell
identities per cluster by matching cell types with the previous annotation (Fig. 5C, D; for the previous
annotation, see Fig. 2A, B). We found that profiles corresponding to TA cells, differentiated cells, and
the tumor-specific TC1-TC5 cell types were present in the organoids. Stem cells, tuft cells and
BEST4/OTOP2-positive enterocytes, which are all cell types that were present only in small numbers in
the original primary tumor samples, were not called in any organoid single-cell library. Surprisingly,
despite the different phenotypic appearance of the organoid cultures, cell type distributions were very
comparable. In particular, TC cell types were present at similar fractions regardless of organoid line
and passage number, unlike the dissimilar TC cell type ratios in the matched tumors. In all four
organoid libraries, the highly proliferative TC1 cell type was overrepresented compared to the matched
tumors.
Differentiation trajectories of CRC organoids
As all major epithelial cell types were present in the organoids, we used the transcriptomes to establish
how the tumor cells are related to each other within differentiation trajectories. We used diffusion
maps to take into account a pseudotemporal order of the transcriptomes that is related to cell
differentiation. Profiles of both organoid lines ordered into a structure containing cells expressing stem
cell markers such as LGR5, PROM1 or CD44 merging into an extended projection containing cells
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
10
positive for the absorptive and secretory differentiation markers FABP1 and TFF3 (Supplementary Fig.
9). Markers specific for the TC cell clusters, that is, MMP7, S100A4, and VIM were found overlapping
and adjacent to the stem cell zone defined by LGR5, PROM1 and CD44.
The organoid line P009 showed an enlarged zone inhabited by transcriptomes positive for the TC cell
markers and was, therefore, suitable to infer a more granular picture of cell plasticity by taking into
account RNA velocity, that is, direction of cell differentiation defined by ratios of immature unspliced
versus mature spliced mRNAs (Fig. 5E). We could define a single trajectory origin in the vicinity of cells
expressing the normal intestinal stem cell marker LGR5 and CD44, which is a CRC stem cell marker
highly expressed in TC1. The main differentiation trajectory extended towards the FABP1- and TFF3-
positive differentiated cells. A shorter trajectory stretched towards the area with the highest
expression of the TC4 cell marker MMP7. We conclude that the TC cells can inhabit a zone of cell
plasticity encompassing CRC stem cells and progenitor-like descendants that are, however, distinct
from absorptive or secretory progenitors.
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
11
Discussion
Here, we use droplet-based single-cell sequencing of transcriptomes to characterize patient-derived
matched CRC and normal colon tissue, as well as CRC organoids. We find that CRC, regardless of specific
genetic mutations or clinical parameters, contains a large proportion of tumor-specific
undifferentiated epithelial cells in addition to cells that resemble differentiated cell types of the normal
colon epithelium. We assign the tumor-specific epithelial cells to five clusters, TC1-5, that form a
continuum of cell plasticity. The TC cell types are distinguished by proliferative activity, oncogenic
signaling, and gene expression patterns related to stemness, tissue invasion, and epithelial-to-
mesenchymal transition. We conclude that these cancer traits are unevenly distributed between
tumor cell subpopulations. Furthermore, we identify stromal and immune cell types, including CAFs,
macrophages, monocytes, and subsets of CD8+ T cells as cell types enriched in the tumor
microenvironment that are sources of multiple growth factors initiating or amplifying Wnt-, TGF-β-,
EGFR- and HGF-MET-signaling. This data suggest that paracrine signaling is a defining factor shaping
the CRC ecosystem.
Our single-cell analysis illuminates a couple of clinically relevant features of CRC. Classification of CRC
by bulk cancer transcriptome analysis can be achieved by the consensus molecular subtype system32.
CMS4, also termed as the “mesenchymal” subtype, is notable for its worse relapse-free and overall
survival. We show here on a single-cell level that CMS4 transcriptomes stem specifically from
fibroblasts, in particular, CAFs, while epithelial tumor cells can only assume CMS1 - CMS3. We
conclude, therefore, that CMS4 is assigned to CRCs with a high content of CAFs that could possibly also
contain a large fraction of the correlated TC2 tumor cell subtype. As we find CAFs to produce multiple
pro-oncogenic growth factors including HGF and TGF-β ligands, it appears as a plausible strategy to
target paracrine signaling as a future therapeutic option for the CMS4 CRC subtype. Furthermore, we
could assign epithelial differentiation states to CMS subtypes. Our findings could be used to
incorporate further informative and cell-type-specific genes into the CMS classifier, in order to assign
CRCs with greater specificity, including those CRCs that currently cannot be assigned to a CMS subtype.
Anti-EGFR antibodies serve as first-line targeted therapy for patients with metastatic disease and no
mutations in the EGFR-RAS-RAF-ERK signaling axis. Treatment success in this cohort has been linked to
the production of the EGFR ligands AREG and EREG, and to immune infiltration in separate
publications28,36. Connecting these phenomena, we have identified here AREG and EREG to be
expressed by immune cells in the CRC ecosystems that we investigated here, in addition to epithelial
cells. AREG and EREG were most strongly active in monocytes, and AREG was additionally expressed
in other immune cell types. We hypothesize that AREG-/EREG-expressing immune cells contribute to
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
12
the paracrine signaling loop activating ERK in KRAS-, NRAS- and BRAF-wildtype CRC cells, and possibly
influence anti-EGFR antibody therapy outcome.
Our results imply that CRC cells display considerable cell plasticity and have multilineage differentiation
capacity, in agreement with pioneering single-cell sequencing studies37,38. Stem cells have traditionally
been seen as unique cells driving tissue homeostasis and regeneration of normal tissue, but also
therapy-resistance in cancer. However, recent studies have shown that combinations of oncogenic
mutations and paracrine signals can steer the cell-intrinsic signaling network so that differentiation
trajectories are reversed and more differentiated cells can regain stem cell characteristics3943. In CRC,
stem cells can be maintained and induced by factors such as HGF13 and IL17A20. We found
Macrophages and CAFs to express HGF and EGFR ligands, extending previous observations44, 45. We also
observed an expansion of IL17A-expressing CD8+ T cells. As these cells also showed transcription of
KLRB1, these cells might be mucosal-associated invariant T cells (also known as MAITs)46,47. To verify
this, an analysis of the invariant T cell receptor Valpha7.2 could be performed in future studies.
Some tumor samples in our analysis contained a small proportion of normal epithelial tissue (see
Supplementary Fig. 1). To ascertain the origin of transcriptomes assigned to differentiated epithelial
cell clusters, we calculated probabilities for single-cell transcriptomes to be derived from the tumor,
taking into account RNA reads covering somatic mutations. While this approach successfully assigned
a small fraction of transcriptomes as deriving from normal or tumor cells, respectively, a large majority
of single-cell profiles remained unassigned (Supplementary Fig. 10). We conclude that single-cell
transcriptomes acquired by our droplet-based sequencing platform contain insufficient information
for mutation-based tumor cell assignment. However, several lines of evidence support that CRC
contains many cells similar to normal differentiated cell types: firstly, we find enterocyte- and goblet
cell-like transcriptomes in tumor samples, but tuft cells and BEST4/OTOP2-positive enterocytes are
selectively depleted, arguing against normal tissue contamination. Secondly, we could stain a
substantial proportion of tumor cells using antibodies against the goblet cell and enterocyte
differentiation markers TFF3 and FABP1, respectively. Thirdly, organoids cultured in a medium
supporting the specific outgrowth of tumor cells maintained differentiated cell populations, with the
exception of tuft cells and BEST4/OTOP2-positive enterocytes.
In organoid cultures, the complex in vivo microenvironment is substituted by a uniform extracellular
matrix and only a few growth factors. It has not been examined in detail how organoid culture
conditions affect tumor cell heterogeneity compared to the tissue of origin. We show here that CRC
organoids maintain all main cell types of CRC, including the tumor-specific cell types TC1-TC5, in the
absence of stromal or immune cells. However, organoids were enriched for profiles of the strongly
proliferative TC1 cells in both patient-derived cultures and multiple passages, probably reflecting the
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
13
growth conditions with high concentrations of EGF, but lacking a complex cellular microenvironment
(see Fig. 5D). It is of note that improved experimental procedures are now available incorporating
fibroblasts into organoid cultures to better mimic paracrine interactions present in vivo48. Single-cell
analysis of such models could inform the dissection of cellular interdependences in CRC, and would
also provide a more realistic scenario for preclinical drug tests.
Our workflow for dissociating clinical samples resulted in the acquisition of many types of major
epithelial, immune and stromal cell transcriptomes. In addition, we capture subtle transcriptome
differences within T cells, plasma cells, fibroblasts, and the tumor-specific TC cells that were assigned
to multiple similar clusters. Analysis of single-cell transcriptomes is subject to multiple confounding
factors, including numbers of genes detected, fractions of spliced versus unspliced mRNAs, and the
fraction of mitochondrial reads. Indeed, particularly the epithelial cell transcriptomes that we captured
for the present study varied on these quality parameters within, but also between the clusters.
Reassuringly, the distinction of phenotypic features between clusters was largely independent of the
confounders. However, we note that the discrimination of five tumor-specific epithelial clusters is
purely heuristic and, presently, we consider tumor-specific epithelial cells to have continuous
phenotypic plasticity rather than inhabiting fixed states that can be clustered with confidence.
Improvements to methods for tissue dissociation have recently been published49 and should be
incorporated into future clinical workflows to diminish tissue processing artifacts. However, with
clinical samples, cell-type-specific degradation of transcriptome quality during the ischemic time
window between the restriction of blood flow and completion of the operation probably cannot be
avoided completely.
The extension of single-cell analyses of CRC to multi-omics, taking also in account genetic and
epigenetic heterogeneity50,51, promises to identify cell plasticity and genetic diversity in cancer at a
cellular resolution. We believe that such approaches will aid the future identification and eradication
of CRC cell populations responsible for therapy resistance.
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
14
Methods
Acquisition and processing of clinical specimens
Fresh normal colon and colorectal cancer tissues were acquired during the intraoperational
pathologist´s examination. Tissues (approx. 0.1-0.4g) were minced using scalpels and stored short-
term on ice in Tissue Storage Solution (Miltenyi # 130-100-008) for transport. Next, tissues were
processed using the Miltenyi human Tumor Dissociation Kit (Miltenyi, #130-095-929) and a Miltenyi
gentleMACS Octo Tissue Dissociator with heaters (Miltenyi, #130-096-427), using program
37C_h_TDK_1 for 30-45min. Cell suspensions were filtered using 100µm filters, and all subsequent
steps were performed at 4°C or on ice. Cells were pelleted by centrifugation in BSA-coated low-binding
tubes, and cells were treated with 1ml ACK erythrocyte lysis buffer for 60 seconds and washed with
DMEM. Cells were pelleted, resuspended in PBS, cell suspensions were filtered using 2m filters,
debris was removed using the Debris Removal Solution (Miltenyi #130-109-398), and cells were
counted using a Neubauer chamber. At least 104 cells of all suspensions were analyzed for cell viability
>75% using LIVE/DEAD Fixable Dead Cell Stain Kit (488nm; Thermo Fisher) and a BD Accuri cytometer.
Hematoxylin-and-eosin staining and immunostaining
3-5 µm tissue sections of fresh frozen or formalin-fixed and paraffin-embedded (FFPE) tissue were used
for immunofluorescence, immunohistochemistry and hematoxylin and eosin staining.
Immunohistochemical and immunofluorescence stainings of FFPE tissue sections were performed on
the BenchMark XT immunostainer (Ventana Medical Systems). For antigen retrieval, tissue sections
were incubated in CC1 mild buffer (Ventana Medical Systems) for 30 min at 100°C. Sections were
incubated with primary antibodies for 60 min and with secondary antibodies for 30 minutes at room
temperature diluted in Dako Real Antibody Diluent (Dako, S2022). The following primary antibodies
were used: rabbit anti-TFF3 (1:250, Abcam, ab108599), mouse anti-FABP1 (1:1000, Abcam, ab7366),
rabbit anti-OLFM4 (1:100, Atlas Antibodies, HPA077718), mouse anti-EPCAM (1:100, ThermoScientific,
MS-144-P1), rabbit anti-Ki67 (1:400, Abcam, ab16667), mouse anti-Ki67 (1:50, Dako, M7240), mouse
anti-CD68 (1:100, Dako, M0876). Images were taken using an Axio Vert.A1 fluorescence microscope
(Zeiss) equipped with an Axiocam 506 color camera (Zeiss).
Single-cell RNA sequencing
104 single cells were used for single-cell library production, using the Chromium Single Cell 3´Reagent
Kits v3 and the Chromium Controller (10x Genomics). Libraries were sequenced on a HiSeq 4000
Sequencer (Illumina) at 200-400mio. reads/library to a mean library saturation of 50%. This resulted in
35.000 to 120.000 reads per cell.
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
15
DNA Sequencing
For panel sequencing for frequent oncogenic driver mutations, tumor-enriched areas (> 40 % tumor
cells) were macrodissected from FFPE tissue sections, DNA was extracted using the Maxwell RSC DNA
FFPE Kit (Promega) and sequenced using the Ion AmpliSeq Cancer Hotspot Panel v2, and an IonTorrent
sequencer (ThermoFisher) according to the manufacturer’s instructions. For variant calling the
Sequence Pilot Software (Version 4.4.0, JSI Medical Systems) was used. For the sequencing of exomes
(patients P007, P008, P009), DNA was isolated from fresh frozen tumor tissue after the pathologist´s
examination using the DNeasy Blood and Tissue Kit (Qiagen). Exomes were sequenced using the
AllExon Human SureSelect v7 Kit (Agilent).
Organoid Culture
Cell filtrates from patient-derived tumor tissues that were retained in the 2m filters after
dissociation were washed in Advanced DMEM/F12 medium (Gibco), embedded in Matrigel, and
cultured in 24-well plates, according to published procedures. Rho-kinase inhibitor Y27632 (10µM,
Sigma) was used for the first passage to avoid anoikis. Cells originally embedded in Matrigel (Corning)
were termed passage 0 (p0), and outgrowing organoids were passaged by removal from Matrigel,
washing in PBS, and partial digestion using TrypLE cell dissociation solution (Gibco) at 37°C, washing in
medium and re-embedding in Matrigel. For single-cell sequencing, organoids were dissociated
completely using TrypLE and DNAseI, and filtering via a 20µm filter.
Single-cell RNA-seq data analysis
For each sample, UMIs were quantified using cellranger 3.0.2 with reference transcriptome GRCh38.
Spliced, unspliced and ambiguous UMIs were quantified with velocyto52 (mode: run10x, default
parameters). Quality control filters were set to only include cells with 500 to 5000 genes detected,
1000 to 50000 UMIs counted, fraction of mitochondrial reads ranging between 0 and 0.8, fraction of
spliced reads ranging between 0.3 and 0.9, fraction of unspliced reads ranging between 0.1 and 0.7
and fraction of ambiguous reads ranging between 0 and 0.2. After filtering, UMI counts were variance-
stabilized using scTransform53, while regressing out fractions of mitochondrial reads and differences in
S-Phase and G2M-Phase scored with Seurat v335. Next, main cell types (epithelium, stromal, and
immune cells) were identified by scoring cell type markers across Louvain clusters for each sample
(resolution = 1). In samples where more than 2000 cells were assigned to a certain cell type, a random
sample of 2000 cells was used of this cell type for the given sample. Normalized subsets were merged
for each main cell type of normal and tumor samples without further batch correction. Louvain cluster-
specific marker genes of merged normal and tumor samples were used to identify sub cell types among
epithelial, stromal and immune subsets. Gene expression sets were taken from the hallmark signature
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
16
collection of the Broad institute54, unless otherwise referenced in the main text, and were scored with
Seurat v3. Epithelial subsets of tumor and matched organoid samples were integrated using Seurat v3
with tumor samples as reference35. Consensus molecular subtypes were scored using random forrest
approach included in R package CMSclassifier.
Diffusion map analysis and RNA velocity were performed using scanpy55 and scvelo56. Cells were first
filtered by the number of genes (between 2000 and 5000) and the percent mitochondrial reads
(between 0.075 and 0.2) and normalized, using scvelo standard settings. Cell cycle was scored
according to the scanpy tutorial, and S_score, G2M_score, percent mitochondrial reads and UMI
counts per cell were regressed out. The diffusion map was calculated on the top 10 principal
components, and using a neighborhood graph with 50 neighbors and calculated on all genes. Moments
were calculated on 30 principal components and 30 neighbors, and velocity was calculated using the
stochastic model.
To compute ligand-receptor connectivity between cell type clusters, UMI counts were summed for all
ligands of the same pathway in each stromal or immune cell type of normal or tumor samples. Summed
ligand counts were scaled to range between zero and one for each pathway. The fraction of normal
and tumor proliferative epithelial cells expressing a given receptor was calculated and fractions were
averaged across receptors for each pathway and cell type. Averaged fractions of cells expressing
receptors were likewise scaled to range between zero and one for each pathway. Connectivity
between stroma or immune ligand expression and epithelial receptor expression was calculated as the
product of scaled ligand counts and scaled receptor expression fractions and, accordingly, also ranged
between zero and one.
Tumor cell calling
For tumor-specific single-nucleotide variant (SNV) calling on single-cell data, we employed exome
sequence data of patients P007, P008 and P009. We used Mutect257 to detect SNVs, retaining only
events classified as somatic. We additionally filtered these results by removing variants on non-
canonical chromosomes or within repeat regions (UCSC genome browser RepeatMasker track). We
then used cellSNP 58 to quantify the total number dij of (UMI-collapsed) reads covering variant i in cell
j (reference and variant allele), and the number aij of reads supporting the alternative allele, using all
variants detected with at least one read in any cell, and all cells containing at least 3 variant-covering
reads. Adapting the EM-type approach proposed in McCarthy et al.59 to our situation, we evaluate the
likelihood pT,j that cell j is a tumor cell using a binomial model:
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
17
Here,
θ
T is the "success probability" for the somatic variants, measuring how likely it is to get a read
supporting the variant allele. Similarly, we compute pN,j as the likelihood that cell j is normal, with a
fixed parameter
θ
N=0.01 allowing for sequencing errors and uncertainties in the variant calls. We
calculate pT,j and pN,j in the E-step and estimate the parameter
θ
T in the M-step as weighted sum over
the counts dij and aij:
E- and M-steps are iterated until convergence of the likelihood
Finally, the criterion pT,j > pN,j is used to define likely tumor cells.
Ethics permission
All patients were aware of the planned research and agreed to the use of tissue. Research was
approved by vote EA4/164/19 of the ethic´s commission of Charité Universitätsmedizin Berlin.
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
18
Funding
The work was in part funded by the Berlin Institute of Health (to NB, CS, DH and MM), and the German
Cancer Consortium DKTK (to NB and MM).
Author contributions
PB, FU, MM, AS, ML conducted and analysed experiments; FU, NB, BO, EB performed bioinformatic
analyses; MM, NB, CS, BS, DH conceived, designed, interpreted experiments and/or supervised parts
of the study; PB, MM, DH, CK contributed to clinical sample acquisition and preparation; MM wrote
the manuscript; all authors provided critical feedback and helped shaping the research, analysis, and
manuscript.
Competing Interests
The authors declare no competing interest.
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
19
References
1. Hanahan, D. & Weinberg, R. A. Hallmarks of cancer: the next generation. Cell 144, 646674
(2011).
2. Sever, R. & Brugge, J. S. Signal transduction in cancer. Cold Spring Harb. Perspect. Med. (2015)
doi:10.1101/cshperspect.a006098.
3. Binnewies, M. et al. Understanding the tumor immune microenvironment (TIME) for effective
therapy. Nat. Med. (2018) doi:10.1038/s41591-018-0014-x.
4. Becht, E. et al. Immune and stromal classification of Colorectal cancer is associated with
molecular subtypes and relevant for precision immunotherapy. Clin. Cancer Res. (2016)
doi:10.1158/1078-0432.CCR-15-2879.
5. Kather, J. N. & Halama, N. Harnessing the innate immune system and local immunological
microenvironment to treat colorectal cancer. British Journal of Cancer (2019)
doi:10.1038/s41416-019-0441-6.
6. Fearon, E. R. Molecular genetics of colorectal cancer. Annu. Rev. Pathol. 6, 479507 (2011).
7. The Cancer Genome Atlas Network. Comprehensive molecular characterization of human
colon and rectal cancer. Nature 487, 330337 (2012).
8. Ricci-Vitiani, L. et al. Identification and expansion of human colon-cancer-initiating cells.
Nature 445, 111115 (2007).
9. O’Brien, C. A. et al. A human colon cancer cell capable of initiating tumour growth in
immunodeficient mice. Nature 445, 106110 (2007).
10. Shimokawa, M. et al. Visualization and targeting of LGR5+ human colon cancer stem cells.
Nature 545, 187192 (2017).
11. Brabletz, T., Jung, A., Dag, S., Hlubek, F. & Kirchner, T. β-catenin regulates the expression of
the matrix metalloproteinase-7 in human colorectal cancer. Am. J. Pathol. (1999)
doi:10.1016/S0002-9440(10)65204-2.
12. Spaderna, S. et al. A Transient, EMT-Linked Loss of Basement Membranes Indicates
Metastasis and Poor Survival in Colorectal Cancer. Gastroenterology (2006)
doi:10.1053/j.gastro.2006.06.016.
13. Vermeulen, L. et al. Wnt activity defines colon cancer stem cells and is regulated by the
microenvironment. Nat. Cell Biol. 12, 468476 (2010).
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
20
14. Smillie, C. S. et al. Intra- and Inter-cellular Rewiring of the Human Colon during Ulcerative
Colitis. Cell (2019) doi:10.1016/j.cell.2019.06.029.
15. Parikh, K. et al. Colonic epithelial cell diversity in health and inflammatory bowel disease.
Nature (2019) doi:10.1038/s41586-019-0992-y.
16. Du, L. et al. CD44 is of functional importance for colorectal cancer stem cells. Clin. Cancer Res.
(2008) doi:10.1158/1078-0432.CCR-08-1034.
17. Toiyama, Y. et al. Increased expression of slug and vimentin as novel predictive biomarkers for
lymph node metastasis and poor prognosis in colorectal cancer. Carcinogenesis (2013)
doi:10.1093/carcin/bgt282.
18. Gongoll, S. et al. Prognostic significance of calcium-binding protein S100A4 in colorectal
cancer. Gastroenterology (2002) doi:10.1053/gast.2002.36606.
19. Schewe, M. et al. Secreted Phospholipases A2 Are Intestinal Stem Cell Niche Factors with
Distinct Roles in Homeostasis, Inflammation, and Cancer. Cell Stem Cell 19, 3851 (2016).
20. Lotti, F. et al. Chemotherapy activates cancer-associated fibroblasts to maintain colorectal
cancer-initiating cells by IL-17A. J. Exp. Med. (2013) doi:10.1084/jem.20131195.
21. De Lau, W. et al. Lgr5 homologues associate with Wnt receptors and mediate R-spondin
signalling. Nature (2011) doi:10.1038/nature10337.
22. Yan, K. S. et al. Non-equivalence of Wnt and R-spondin ligands during Lgr5 + intestinal stem-
cell self-renewal. Nature (2017) doi:10.1038/nature22313.
23. Haramis, A.-P. G. et al. De novo crypt formation and juvenile polyposis on BMP inhibition in
mouse intestine. Sci. (New York, NY) 303, 16841686 (2004).
24. Birchmeier, C., Birchmeier, W., Gherardi, E. & Vande Woude, G. F. Met, metastasis, motility
and more. Nat. Rev. Mol. Cell Biol. 4, 915925 (2003).
25. Rohlin, A. et al. GREM1 and POLE variants in hereditary colorectal cancer syndromes. Genes
Chromosom. Cancer (2016) doi:10.1002/gcc.22314.
26. Miyoshi, H., Ajima, R., Luo, C. T., Yamaguchi, T. P. & Stappenbeck, T. S. Wnt5a potentiates
TGF-β signaling to promote colonic crypt regeneration after tissue injury. Sci. (New York, NY)
338, 108113 (2012).
27. Bakker, E. R. M. et al. Wnt5a promotes human colon cancer cell migration and invasion but
does not augment intestinal tumorigenesis in apc1638N mice. Carcinogenesis (2013)
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
21
doi:10.1093/carcin/bgt215.
28. Jacobs, B. et al. Amphiregulin and epiregulin mRNA expression in primary tumors predicts
outcome in metastatic colorectal cancer treated with cetuximab. J. Clin. Oncol. (2009)
doi:10.1200/JCO.2008.21.3744.
29. Brabletz, T. et al. Nuclear overexpression of the oncoprotein β-Catenin in colorectal cancer is
localized predominantly at the invasion front. Pathol. Res. Pract. (1998) doi:10.1016/S0344-
0338(98)80129-5.
30. Voloshanenko, O. et al. Wnt secretion is required to maintain high levels of Wnt activity in
colon cancer cells. Nat. Commun. (2013) doi:10.1038/ncomms3610.
31. Uhlitz, F. et al. An immediate–late gene expression module decodes ERK signal duration. Mol.
Syst. Biol. 13, 928 (2017).
32. Guinney, J. et al. The consensus molecular subtypes of colorectal cancer. Nat. Med. 21, 1350
1356 (2015).
33. Sato, T. et al. Long-term expansion of epithelial organoids from human colon, adenoma,
adenocarcinoma, and Barrett’s epithelium. Gastroenterology 141, 17621772 (2011).
34. Schütte, M. et al. Molecular dissection of colorectal cancer in pre-clinical models identifies
biomarkers predicting sensitivity to EGFR inhibitors. Nat. Commun. 8, 14262 (2017).
35. Stuart, T. et al. Comprehensive Integration of Single-Cell Data. Cell (2019)
doi:10.1016/j.cell.2019.05.031.
36. Pozzi, C. et al. The EGFR-specific antibody cetuximab combined with chemotherapy triggers
immunogenic cell death. Nat. Med. (2016) doi:10.1038/nm.4078.
37. Dalerba, P. et al. Single-cell dissection of transcriptional heterogeneity in human colon
tumors. Nat. Biotechnol. (2011) doi:10.1038/nbt.2038.
38. Li, H. et al. Reference component analysis of single-cell transcriptomes elucidates cellular
heterogeneity in human colorectal tumors. Nat. Genet. 49, 708718 (2017).
39. van Es, J. H. et al. Dll1(+) secretory progenitor cells revert to stem cells upon crypt damage.
Nat. Cell Biol. 14, 10991104 (2012).
40. Schwitalla, S. et al. Intestinal Tumorigenesis Initiated by Dedifferentiation and Acquisition of
Stem-Cell-like Properties. Cell (2012).
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
22
41. Buczacki, S. J. A. et al. Intestinal label-retaining cells are secretory precursors expressing Lgr5.
Nature 495, 6569 (2013).
42. Jadhav, U. et al. Dynamic Reorganization of Chromatin Accessibility Signatures during
Dedifferentiation of Secretory Precursors into Lgr5+ Intestinal Stem Cells. Cell Stem Cell (2017)
doi:10.1016/j.stem.2017.05.001.
43. Tomic, G. et al. Phospho-regulation of ATOH1 Is Required for Plasticity of Secretory
Progenitors and Tissue Regeneration. Cell Stem Cell (2018) doi:10.1016/j.stem.2018.07.002.
44. Nabeshima, A. et al. Tumour-associated macrophages correlate with poor prognosis in myxoid
liposarcoma and promote cell motility and invasion via the HB-EGF-EGFR-PI3K/Akt pathways.
Br. J. Cancer (2015) doi:10.1038/bjc.2014.637.
45. Vlaicu, P. et al. Monocytes/macrophages support mammary tumor invasivity by co-secreting
lineage-specific EGFR ligands and a STAT3 activator. BMC Cancer (2013) doi:10.1186/1471-
2407-13-197.
46. Walker, L. J. et al. Human MAIT and CD8αα cells develop from a pool of type-17
precommitted CD8 + T cells. Blood (2012) doi:10.1182/blood-2011-05-353789.
47. Dusseaux, M. et al. Human MAIT cells are xenobiotic-resistant, tissue-targeted, CD161 hi IL-
17-secreting T cells. Blood (2011) doi:10.1182/blood-2010-08-303339.
48. Karpus, O. N. et al. Colonic CD90+ Crypt Fibroblasts Secrete Semaphorins to Support Epithelial
Growth. Cell Rep. (2019) doi:10.1016/j.celrep.2019.02.101.
49. Adam, M., Potter, A. S. & Potter, S. S. Psychrophilic proteases dramatically reduce single-cell
RNA-seq artifacts: A molecular atlas of kidney development. Dev. (2017)
doi:10.1242/dev.151142.
50. Bian, S. et al. Single-cell multiomics sequencing and analyses of human colorectal cancer.
Science (80-. ). (2018) doi:10.1126/science.aao3791.
51. Roerink, S. F. et al. Intra-tumour diversification in colorectal cancer at the single-cell level.
Nature (2018) doi:10.1038/s41586-018-0024-3.
52. La Manno, G. et al. RNA velocity of single cells. Nature (2018) doi:10.1038/s41586-018-0414-
6.
53. Hafemeister, C. & Satija, R. Normalization and variance stabilization of single-cell RNA-seq
data using regularized negative binomial regression. bioRxiv (2019) doi:10.1101/576827.
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
23
54. Liberzon, A. et al. The Molecular Signatures Database Hallmark Gene Set Collection. Cell Syst.
(2015) doi:10.1016/j.cels.2015.12.004.
55. Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: Large-scale single-cell gene expression data
analysis. Genome Biol. (2018) doi:10.1186/s13059-017-1382-0.
56. Bergen, V., Lange, M., Peidli, S., Wolf, F. A. & Theis, F. J. Generalizing RNA velocity to transient
cell states through dynamical modeling. bioRxiv (2019) doi:10.1101/820936.
57. Cibulskis, K. et al. Sensitive detection of somatic point mutations in impure and
heterogeneous cancer samples. Nat. Biotechnol. (2013) doi:10.1038/nbt.2514.
58. Huang, Y., McCarthy, D. J. & Stegle, O. Vireo: Bayesian demultiplexing of pooled single-cell
RNA-seq data without genotype reference. bioRxiv (2019) doi:10.1101/598748.
59. McCarthy, D. J. et al. Cardelino: Integrating whole exomes and single-cell transcriptomes to
reveal phenotypic impact of somatic variants (under review at Nature Methods). bioRxiv
(2018) doi:10.1101/413047.
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
24
Figure 1: Generation and initial assignment of CRC single-cell RNA sequencing data. A Clinical data
for the eight patients under investigation in this study. For mutational data, see also Supplementary
table 6. Loc (Localisation): C: cecum; S: sigmoid colon; T: transverse colon; A: ascending colon; R:
rectum. Prog (Predicted Progression): S: via serrated precursor; I: inflammatory/colitis-associated; C:
canonical. Tissues used for single-cell RNA sequencing: N: Normal; T: Tumor; O: Organoid. B UMAPs of
epithelial, immune and stromal cells, color-coded for patients. C UMAPs of epithelial, immune and
stromal cells, color-coded by tissue of origin.
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
25
Figure 2: Cell type census in normal colon and CRC. A UMAPs of epithelial, immune and stromal cells,
separated by tissue of origin. Color code for cell type assignment. B Relative fractions of epithelial,
immune and stromal cell types across all patient-derived libraries. For fractions per patient, see
Supplementary Figure 3. C Single cell gene activities of stem cell marker OLFM4, proliferative marker
MKI67, differentiated absorptive cell marker FABP1, and secretory cell marker TFF3. D
Immunofluorescence analysis for OLFM4, MKI67, FABP1, and TFF3 in normal and tumor tissue. All
sections are from patient P009, except the EPCAM/OLFM4 co-staining that was done on tumor tissue
of P016. Scale bars indicate 100µm. E Relative fraction of TC1-5 in the tumor tissues of the patients. F
Transcriptome-inferred cell-cycle distribution in the tumor cell fractions TC1-TC5 (bar graph), and in
the UMAPs of normal and tumor epithelium.
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
26
Figure 3: A connectivity map of potential paracrine interactions in CRC. A Connectivity analysis for
ligand expression in stromal and immune cells, and receptor expression in proliferative epithelial cell
clusters. Connectivity takes into account expression levels of ligands, the prevalence of the ligand-
expressing cell, and fractions of receptor-expressing cells. See methods for details. Circle sizes: Cell
numbers for ligand-expressing cells, as in figure legend. Red: High connectivity; grey: low connectivity.
B Activities of Wnt, ERK and TGFβ target genes in the UMAPs of normal and tumor epithelium. C
Expression of key ligands for CRC progression and therapy response in immune cells. D
Immunohistochemistry for macrophages in normal and tumor tissue of P009, using an antibody against
CD68. In the tumor, macrophages are more prevalent, in line with the single-cell data. Macrophages
are enriched at the invasive front (tumor, to the right). Scale bar indicates 100µm. E Activity of
receptors implicated in the epithelial-to-mesenchymal transition, and activity of genes comprising the
EMT hallmark signature. F Correlation between fractions of CAFs in the stromal cell compartment and
TC2 tumor cells in the epithelial tumor compartment.
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
27
Figure 4: Consensus molecular subtype analysis on the single-cell level. A CMS calling on cells of the
epithelial compartment. B CMS calling on cells of the immune compartment. C CMS calling on cells of
the stromal compartment. In all subfigures, CMS subtypes are mapped to the normal and tumor tissue
UMAPs to the left, and posterior probability scores for the complete tumor and normal cells in the
respective compartments per patient are given to the right.
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
28
Figure 5: Epithelial cell types in matched tumors and organoids. A Phenotypes of organoid lines
established from study patients. Patient and passage number are given to the left. B UMAPs of tumor-
derived epithelial tissue profiles and anchored organoid profiles, color coded for sample type, patient,
organoid passage number and cell cycle phase, as indicated. C UMAPs of tumor epithelial and organoid
cells, color-coded by cell-type assignment. D Cell type distributions in tumor tissue and organoids. E
RNA velocity of organoid line P009. Cells highly expressing the enterocyte marker FABP1, the normal
stem cell markers LGR5 the TC1 stem cell marker CD44, and the TC4 tumor cell marker MMP7 are
highlighted by green, blue and red and rose color, respectively. Arrows indicate two trajectories with
a common root near the LGR5- and CD44-expressing cells, and extending towards the FABP1 and
MMP7-positive cells.
.CC-BY-NC-ND 4.0 International licenseIt is made available under a perpetuity.
this preprint (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in
The copyright holder for. http://dx.doi.org/10.1101/2020.01.10.901579doi: bioRxiv preprint first posted online Jan. 11, 2020;
... In all the cohorts of CRC patients studied, S3 CRCs display a better survival than the two other groups [14]. A more recent study based on the analysis of >50,000 single cells from CRCs and matched normal tissues provided evidence that CRC development is associated in all cases analyzed with changes at the level of epithelial, immune and stromal cell compartments [15]. Interestingly, in the epithelium, five different tumor-specific stem and progenitor-like cell populations were identified [15]. ...
... A more recent study based on the analysis of >50,000 single cells from CRCs and matched normal tissues provided evidence that CRC development is associated in all cases analyzed with changes at the level of epithelial, immune and stromal cell compartments [15]. Interestingly, in the epithelium, five different tumor-specific stem and progenitor-like cell populations were identified [15]. This single-cell analysis showed also that epithelial tumor cells and cancer-associated fibroblasts are fundamental and essential for the assignment of each CRC to a given CMS subtype [15]. ...
... Interestingly, in the epithelium, five different tumor-specific stem and progenitor-like cell populations were identified [15]. This single-cell analysis showed also that epithelial tumor cells and cancer-associated fibroblasts are fundamental and essential for the assignment of each CRC to a given CMS subtype [15]. ...
Article
Full-text available
Genome sequencing studies have characterized the genetic alterations of different tumor types, highlighting the diversity of the molecular processes driving tumor development. Comprehensive sequencing studies have defined molecular subtypes of colorectal cancers (CRCs) through the identification of genetic events associated with microsatellite stability (MSS), microsatellite-instability-high (MSI-H), and hypermutation. Most of these studies characterized primary tumors. Only recent studies have addressed the characterization of the genetic and clinical heterogeneity of metastatic CRC. Metastatic CRC genomes were found to be not fundamentally different from primary CRCs in terms of the mutational landscape or of genes that drive tumorigenesis, and a genomic heterogeneity associated with tumor location of primary tumors helps to define different clinical behaviors of metastatic CRCs. Although CRC metastatic spreading was traditionally seen as a late-occurring event, growing evidence suggests that this process can begin early during tumor development and the clonal architecture of these tumors is consistently influenced by cancer treatment. Although the survival rate of patients with metastatic CRC patients improved in the last years, the response to current treatments and prognosis of many of these patients remain still poor, indicating the need to discover new improvements for therapeutic vulnerabilities and to formulate a rational prospective of personalized therapies.
... For the metabolic labeling of RNA in single cells, we made two major improvements: First, we made the method accessible for high-throughput microfluidics by performing the chemical conversion of 4sU in methanol fixed cells. Since microfluidic platforms such as 10X Chromium are becoming increasingly popular, the combination of both will pave the way for more studies of transcriptome dynamics, for example for studying the effects of drugs or novel compounds on cells and organoids (Uhlitz et al., 2020). For the first time, we applied this method in an in vivo system to study the fate of maternally provided transcripts by injecting 4sU into one-cell stage zebrafish embryos. ...
Thesis
Die Lokalisierung von mRNA ist ein wichtiger regulativer Mechanismus in polarisierten Zellen und in frühen Embryonalstadien. Dort sind räumliche Muster maternaler mRNA für die korrekte Entwicklung der Körperachsen und die Spezifizierung der Keimzellen verantwortlich. Systematische Analysen dieser Prozesse wurden jedoch bisher limitiert durch einen Mangel an räumlicher und zeitlicher Auflösung von Einzelzell- Sequenzierungsdaten. Wir analysierten die Dynamik des räumlichen und zeitlichen Transkriptoms während frühen Embryonalstadien von Zebrafischen. Wir verbesserten Empfindlichkeit und Auflösung von tomo-seq und erfassten damit systematisch räumlich aufgelöste Transkriptome entlang der animal-vegetalen-Achse Embryonen im Einzell-Stadium und fanden 97 vegetal lokalisierte Gene. Außerdem etablierten wir eine Hochdurchsatz kompatible Variante der RNA-Markierungsmethode scSLAM-seq. Wir wendeten diese in Embryonen während der Gastrulation. Von den vegetal lokalisierten Genen waren 22 angereichert in Keimzellen, was eine funktionelle Rolle bei der Spezifizierung von Keimzellen nahelegt. Mit tomo-seq untersuchten wir die evolutionäre Konservierung der RNA-Lokalisierung zwischen Zebrafischen und gereiften Oozyten zweier Xenopus-Arten. Wir verglichen die lokalisierten Gene, suchten nach konservierten 3'UTR-Motiven, und fanden zum Teil überlappende Motive, was auf eine mögliche mechanistische Konservierung der Lokalisierungsmechanismen hinweist. Wir untersuchten auch RNA-Editierung von Adenin zu Inosin während der Embryonalentwicklung und in den Organen erwachsener Fische. In im Gehirn exprimierten Transkripten fanden wir 117 Editierstellen, die hauptsächlich für Ionentransporter kodieren und zum Teil zum Menschen konserviert sind. Die höchsten Editierraten konnten wir in Eierstöcken, Hoden und frühen Embryonen nachweisen, was auf eine mögliche Rolle bei der Regulierung der RNA-Stabilität hindeutet.
... The tumour microenvironment in CRC consists of multiple stromal and immune cell types, including cancer-associated fibroblasts, pericytes, monocytes, macrophages and a subset of T cells [88]. The numbers of CD66b-positive neutrophils are increased in CRC in comparison to adjacent mucosa, and a high density of intratumoural neutrophils is an independent factor for poor prognosis of patients with CRC. ...
Article
Full-text available
In global cancer statistics, colorectal carcinoma (CRC) ranks third by incidence and second by mortality, causing 10.0% of new cancer cases and 9.4% of oncological deaths worldwide. Despite the development of screening programs and preventive measures, there are still high numbers of advanced cases. Multiple problems compromise the treatment of metastatic colorectal cancer, one of these being cancer stem cells—a minor fraction of pluripotent, self-renewing malignant cells capable of maintaining steady, low proliferation and exhibiting an intriguing arsenal of treatment resistance mechanisms. Currently, there is an increasing body of evidence for intricate associations between inflammation, epithelial–mesenchymal transition and cancer stem cells. In this review, we focus on inflammation and its role in CRC stemness development through epithelial–mesenchymal transition.
... Paracrine or microenvironment-derived EGFR stimuli can provide essential amplification of oncogenic MAPK signalling to promote tumour growth, as we also deduced from multiple PDX models of CRC cancer (Fig. 7). Accordingly, transcriptome analysis of human CRCs identified increased levels of paracrine EGFR stimuli in tumour tissue 60 . Autocrine production of EGFR-activating ligands has been demonstrated to predict the response to EGFR inhibition [61][62][63] . ...
Article
Full-text available
Direct targeting of the downstream mitogen-activated protein kinase (MAPK) pathway to suppress extracellular-regulated kinase (ERK) activation in KRAS and BRAF mutant colorectal cancer (CRC) has proven clinically unsuccessful, but promising results have been obtained with combination therapies including epidermal growth factor receptor (EGFR) inhibition. To elucidate the interplay between EGF signalling and ERK activation in tumours, we used patient-derived organoids (PDOs) from KRAS and BRAF mutant CRCs. PDOs resemble in vivo tumours, model treatment response and are compatible with live-cell microscopy. We established real-time, quantitative drug response assessment in PDOs with single-cell resolution, using our improved fluorescence resonance energy transfer (FRET)-based ERK biosensor EKAREN5. We show that oncogene-driven signalling is strikingly limited without EGFR activity and insufficient to sustain full proliferative potential. In PDOs and in vivo, upstream EGFR activity rigorously amplifies signal transduction efficiency in KRAS or BRAF mutant MAPK pathways. Our data provide a mechanistic understanding of the effectivity of EGFR inhibitors within combination therapies against KRAS and BRAF mutant CRC. Ponsioen et al. use a FRET‐based ERK biosensor EKAREN5 in patient‐derived organoids to show that EGFR activity amplifies signal transduction efficiency in KRAS or BRAF mutant MAPK pathways.
Preprint
Full-text available
The introduction of RNA velocity in single cells has opened up new ways of studying cellular differentiation. The originally proposed framework obtains velocities as the deviation of the observed ratio of spliced and unspliced mRNA from an inferred steady state. Errors in velocity estimates arise if the central assumptions of a common splicing rate and the observation of the full splicing dynamics with steady-state mRNA levels are violated. With scVelo (https://scvelo.org), we address these restrictions by solving the full transcriptional dynamics of splicing kinetics using a likelihood-based dynamical model. This generalizes RNA velocity to a wide variety of systems comprising transient cell states, which are common in development and in response to perturbations. We infer gene-specific rates of transcription, splicing and degradation, and recover the latent time of the underlying cellular processes. This latent time represents the cell's internal clock and is based only on its transcriptional dynamics. Moreover, scVelo allows us to identify regimes of regulatory changes such as stages of cell fate commitment and, therein, systematically detects putative driver genes. We demonstrate that scVelo enables disentangling heterogeneous subpopulation kinetics with unprecedented resolution in hippocampal dentate gyrus neurogenesis and pancreatic endocrinogenesis. We anticipate that scVelo will greatly facilitate the study of lineage decisions, gene regulation, and pathway activity identification.
Preprint
Full-text available
Common genetic variants have been used as natural barcodes to demultiplex cells from pooled RNA-seq experiments. Existing demultiplexing strategies rely on access to complete genotype data from the set of pooled samples, which greatly limits the applicability of such methods, in particular when genetic variation is not the primary object of study. To address this, we here present Vireo, a computationally efficient Bayesian model to demultiplex data from pooled experimental designs. Our model can be applied to dataset with partial or without any genotype information of the pooled samples. Using simulations and results on real data, we demonstrate the robustness of our model and illustrate the utility of multi-sample experimental designs for common expression analyses.
Article
Full-text available
Significant progress in the development of new immunotherapies has led to successful clinical trials for malignant melanoma and non-small cell lung cancer; however, for the majority of solid tumours of the gastrointestinal tract, little or no progress has been seen. The efficacy of immunotherapies is limited by the complexities of a diverse set of immune cells, and interactions between the tumour cells and all other cells in the local microenvironment of solid tumours. A large fraction of immune cells present in and around solid tumours derive from the innate arm of the immune system and using these cells against tumours offers an alternative immunotherapeutic option, especially as current strategies largely harness the adaptive arm of the immune system. This option is currently being investigated and attempts at using the innate immune system for gastrointestinal cancers are showing initial results. Several important factors, including cytokines, chemotherapeutics and the microbiome, influence the plasticity and functionality of innate (myeloid) cells in the microenvironment, and this complexity of regulation has limited translation into successful trials so far. In this review, current concepts of the immunobiology of the innate arm in the tumour microenvironment are presented in the context of clinical translation.
Article
Full-text available
Intestinal epithelial cells have a defined hierarchy with stem cells located at the bottom of the crypt and differentiated cells more at the top. Epithelial cell renewal and differentiation are strictly controlled by various regulatory signals provided by epithelial as well as surrounding cells. Although there is evidence that stromal cells contribute to the intestinal stem cell niche, their markers and the soluble signals they produce have been incompletely defined. Using a number of established stromal cell markers, we phenotypically and functionally examined fibroblast populations in the colon. CD90+ fibroblasts located in close proximity to stem cells in vivo support organoid growth in vitro and express crucial stem cell growth factors, such as Grem1, Wnt2b, and R-spondin3. Moreover, we found that CD90+ fibroblasts express a family of proteins—class 3 semaphorins (Sema3)—that are required for the supportive effect of CD90+ fibroblasts on organoid growth. : Stem cell proliferation is dependent on regulatory signals from surrounding cells, including stromal cells. Karpus et al. identify a membrane marker, CD90, that specifically marks stem cell niche fibroblasts in the colon. CD90+ fibroblasts produce class 3 semaphorins that promote stem cell proliferation via the Nrp2 receptor. Keywords: colon, intestine, stem cell, niche, fibroblast, stroma, Gli1, CD90, semaphorin, Nrp2
Preprint
Full-text available
Single-cell RNA-seq (scRNA-seq) data exhibits significant cell-to-cell variation due to technical factors, including the number of detected molecules in each cell, which can confound biological heterogeneity with technical effects. To address this, we present a modeling framework for the normalization and variance stabilization of molecular count data from scRNA-seq experiments. We propose that the Pearson residuals from 'regularized negative binomial regression', where cellular sequencing depth is utilized as a covariate in a generalized linear model, successfully remove the influence of technical characteristics from downstream analyses while preserving biological heterogeneity. Importantly, we show that an unconstrained negative binomial model may overfit scRNA-seq data, and overcome this by pooling information across genes with similar abundances to obtain stable parameter estimates. Our procedure omits the need for heuristic steps including pseudocount addition or log-transformation, and improves common downstream analytical tasks including variable gene selection, dimensional reduction, and differential expression. Our procedure can be applied to any UMI-based scRNA-seq dataset and is freely available as part of the R package sctransform, with a direct interface to our single-cell toolkit Seurat.
Article
Full-text available
The colonic epithelium facilitates host–microorganism interactions to control mucosal immunity, coordinate nutrient recycling and form a mucus barrier. Breakdown of the epithelial barrier underpins inflammatory bowel disease (IBD). However, the specific contributions of each epithelial-cell subtype to this process are unknown. Here we profile single colonic epithelial cells from patients with IBD and unaffected controls. We identify previously unknown cellular subtypes, including gradients of progenitor cells, colonocytes and goblet cells within intestinal crypts. At the top of the crypts, we find a previously unknown absorptive cell, expressing the proton channel OTOP2 and the satiety peptide uroguanylin, that senses pH and is dysregulated in inflammation and cancer. In IBD, we observe a positional remodelling of goblet cells that coincides with downregulation of WFDC2—an antiprotease molecule that we find to be expressed by goblet cells and that inhibits bacterial growth. In vivo, WFDC2 preserves the integrity of tight junctions between epithelial cells and prevents invasion by commensal bacteria and mucosal inflammation. We delineate markers and transcriptional states, identify a colonic epithelial cell and uncover fundamental determinants of barrier breakdown in IBD. Profiling of single epithelial cells in healthy and inflamed colons identifies specialized cellular subpopulations, including a type of goblet cell that secretes the antibacterial protein WFDC2, which preserves the integrity of the epithelial barrier layer.
Article
Full-text available
Although genomic instability, epigenetic abnormality, and gene expression dysregulation are hallmarks of colorectal cancer, these features have not been simultaneously analyzed at single-cell resolution. Using optimized single-cell multiomics sequencing together with multiregional sampling of the primary tumor and lymphatic and distant metastases, we developed insights beyond intratumoral heterogeneity. Genome-wide DNA methylation levels were relatively consistent within a single genetic sublineage. The genome-wide DNA demethylation patterns of cancer cells were consistent in all 10 patients whose DNA we sequenced. The cancer cells’ DNA demethylation degrees clearly correlated with the densities of the heterochromatin-associated histone modification H3K9me3 of normal tissue and those of repetitive element long interspersed nuclear element 1. Our work demonstrates the feasibility of reconstructing genetic lineages and tracing their epigenomic and transcriptomic dynamics with single-cell multiomics sequencing.
Article
Full-text available
RNA abundance is a powerful indicator of the state of individual cells. Single-cell RNA sequencing can reveal RNA abundance with high quantitative accuracy, sensitivity and throughput1. However, this approach captures only a static snapshot at a point in time, posing a challenge for the analysis of time-resolved phenomena such as embryogenesis or tissue regeneration. Here we show that RNA velocity-the time derivative of the gene expression state-can be directly estimated by distinguishing between unspliced and spliced mRNAs in common single-cell RNA sequencing protocols. RNA velocity is a high-dimensional vector that predicts the future state of individual cells on a timescale of hours. We validate its accuracy in the neural crest lineage, demonstrate its use on multiple published datasets and technical platforms, reveal the branching lineage tree of the developing mouse hippocampus, and examine the kinetics of transcription in human embryonic brain. We expect RNA velocity to greatly aid the analysis of developmental lineages and cellular dynamics, particularly in humans.
Article
Single-cell transcriptomics has transformed our ability to characterize cell states, but deep biological understanding requires more than a taxonomic listing of clusters. As new methods arise to measure distinct cellular modalities, a key analytical challenge is to integrate these datasets to better understand cellular identity and function. Here, we develop a strategy to "anchor" diverse datasets together, enabling us to integrate single-cell measurements not only across scRNA-seq technologies, but also across different modalities. After demonstrating improvement over existing methods for integrating scRNA-seq data, we anchor scRNA-seq experiments with scATAC-seq to explore chromatin differences in closely related interneuron subsets and project protein expression measurements onto a bone marrow atlas to characterize lymphocyte populations. Lastly, we harmonize in situ gene expression and scRNA-seq datasets, allowing transcriptome-wide imputation of spatial gene expression patterns. Our work presents a strategy for the assembly of harmonized references and transfer of information across datasets.
Article
Genome-wide association studies (GWAS) have revealed risk alleles for ulcerative colitis (UC). To understand their cell type specificities and pathways of action, we generate an atlas of 366,650 cells from the colon mucosa of 18 UC patients and 12 healthy individuals, revealing 51 epithelial, stromal, and immune cell subsets, including BEST4+ enterocytes, microfold-like cells, and IL13RA2+IL11+ inflammatory fibroblasts, which we associate with resistance to anti-TNF treatment. Inflammatory fibroblasts, inflammatory monocytes, microfold-like cells, and T cells that co-express CD8 and IL-17 expand with disease, forming intercellular interaction hubs. Many UC risk genes are cell type specific and co-regulated within relatively few gene modules, suggesting convergence onto limited sets of cell types and pathways. Using this observation, we nominate and infer functions for specific risk genes across GWAS loci. Our work provides a framework for interrogating complex human diseases and mapping risk variants to cell types and pathways.