open source immgen: network perspective on metabolic ... · 15/07/2020 · open source immgen:...
TRANSCRIPT
Open Source ImmGen: network perspective on metabolic diversity among mononuclearphagocytes
Anastasiia Gainullina1,2, Li-Hao Huang1, Helena Todorov9, Kiwook Kim1, Lim Sheau Yng3,Andrew Kent4, Baosen Jia5, Kumba Seddu6, Karen Krchma1, Jun Wu1, Karine Crozat7, ElenaTomasello7, Vipin Narang8, Regine Dress8, Peter See8, Charlotte Scott9, Sophie Gibbings10,Geetika Bajpai11, Jigar V. Desai12, Barbara Maier13, Sébastien This14, Peter Wang1, StephanieVargas Aguilar7,15,16, Lucie Poupel17, Sébastien Dussaud17, Tyng-An Zhou18, Veronique Angeli3,J. Magarian Blander5, Kyunghee Choi1,19, Marc Dalod7, Ivan Dzhagalov18, Emmanuel L.Gautier17, Claudia Jakubzick10, Kory Lavine11, Michail S. Lionakis12, Helena Paidassi14, MichaelH. Sieweke7,15,16, Florent Ginhoux8, Martin Guilliams9, Christophe Benoist6, Miriam Merad13,Gwendalyn J. Randolph1, Alexey Sergushichev1,2,*, Maxim N. Artyomov1,*, ImmGen Consortium
1Department of Pathology and Immunology, Washington University School of Medicine, St. Louis, MO, USA2Computer Technologies Department, ITMO University, St. Petersburg, Russia3Department of Microbiology and Immunology, Life Science Institute, Yoon Loo Lin School of Medicine, NationalUniversity of Singapore, Singapore4Department of Internal Medicine, University of Colorado Hospital, Aurora, CO, USA5The Jill Roberts Institute for Research in Inflammatory Bowel Disease, Joan and Sanford I. Weill Department ofMedicine, Department of Microbiology and Immunology, Weill Cornell Medicine, Cornell University, New York, NY10021, USA6Department of Immunology, Harvard Medical School, Boston, MA, USA7Aix Marseille University, CNRS, INSERM, CIML, Centre d'Immunologie de Marseille-Luminy, Marseille, France8Singapore Immunology Network, Agency for Science, Technology and Research, Singapore, Singapore9Laboratory of Immunoregulation, Inflammation Research Centre, VIB Ghent University, Ghent, Belgium10Department of Pediatrics, National Jewish Health, Denver, CO, USA11Center for Cardiovascular Research, Division of Cardiology, Department of Medicine, Washington UniversitySchool of Medicine, Saint Louis, MO, USA12Fungal Pathogenesis Section, Laboratory of Clinical Immunology and Microbiology, National Institute of Allergyand Infectious Diseases, National Institutes of Health, Bethesda, Maryland, USA13Immunology Institute and Tisch Cancer Institute, Icahn School of Medicine at Mount Sinai, New York, New York,USA14International Center for Infectiology Research, University of Lyon, Lyon, France15Department of Immunology, Weizmann Institute of Science, Rehovot, Israel16Max-Delbrück-Centrum für Molekulare Medizin in der Helmholtzgemeinschaft (MDC), Berlin, Germany17INSERM UMR-S 1166, Sorbonne Université, Hôpital de la Pitié-Salpêtrière, Paris, France18Institute of Microbiology and Immunology, National Yang-Ming University, Taipei, Taiwan19Graduate School of Biotechnology, Kyung Hee University, Yong In, South Korea*e-mail: [email protected]*e-mail: [email protected]
1
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
Abstract
We dissect metabolic variability of mononuclear phagocyte (MNP) subpopulations across
different tissues through integrative analysis of three large scale datasets. Specifically, we
introduce ImmGen MNP Open Source dataset that profiled 337 samples and extended
previous ImmGen effort which included 202 samples of mononuclear phagocytes and their
progenitors. Next, we analysed Tabula Muris Senis dataset to extract data for 51,364 myeloid
cells from 18 tissues. Taken together, a compendium of data assembled in this work covers
phagocytic populations found across 38 different tissues. To analyse common metabolic
features, we developed novel network-based computational approach for unbiased
identification of key metabolic subnetworks based on cellular transcriptional profiles in large-
scale datasets. Using ImmGen MNP Open Source dataset as baseline, we define 9 metabolic
subnetworks that encapsulate the metabolic differences within mononuclear phagocytes, and
demonstrate that these features are robustly found across all three datasets, including lipid
metabolism, cholesterol biosynthesis, glycolysis, and a set of fatty acid related metabolic
pathways, as well as nucleotide and folate metabolism. We systematically define major
features specific to macrophage and dendritic cell subpopulations. Among other things, we find
that cholesterol synthesis appears particularly active within the migratory dendritic cells. We
demonstrate that interference with this pathway through statins administration diminishes
migratory capacity of the dendritic cells in vivo. This result demonstrates the power of our
approach and highlights importance of metabolic diversity among mononuclear phagocytes.
2
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
Introduction
The diversity of myeloid cells across different tissues is truly astonishing, both in function and
in their developmental trajectory1,2. Additional dimension of this diversity is manifested by the
metabolic characteristics of individual mononuclear phagocytes which can vary significantly
based on the cell type and its location3-5. At present, direct metabolomics profiling of tissue
residing subpopulations is not feasible, as the process of ex vivo sorting can be lengthy and
cause significant metabolic perturbations6,7. However, RNA levels are significantly more stable
to the sorting process and can serve as a reasonably reliable proxy to activities of metabolic
pathways8,9. In this work we focus on understanding metabolic variability across phagocytic
subpopulations through integrated examination of several large-scale datasets that
transcriptionally profiled subsets of myeloid cells (Fig. 1a-c). Specifically, we have assembled
compendium of three datasets, including the first public release of the new dataset generated
by Mononuclear Phagocytes Open Source (MNP OS) ImmGen project10.
ImmGen MNP OS dataset totals 337 samples and provides a unique source of information
about individual cell subpopulations (Fig. 1d, e). It extends previous ImmGen effort that
included 202 samples of various mononuclear phagocytes, also analysed in this study (Fig. 1f,
g). In addition to increased number of mature cell populations from adult mice (monocytes,
macrophages, and dendritic cells), the MNP OS dataset contains macrophages from the yolk
sac (E10.5) and macrophages differentiated in vitro from embryonic stem cells (embryoid body
derived macrophages, E6-E8). Furthermore, we leveraged recently released single-cell RNA-
seq profiling of the multiple murine organs (Tabula Muris Senis11) and reanalysed those data by
focusing only on the phagocytic populations, comprising 36,480 cells across 18 tissues (Fig.
1h, i). Taken together, a compendium of data assembled in this work covers multiple cell
subpopulations found across 38 different tissues (Fig. 1b).
Using these transcriptional data, we sought to identify the major metabolic features
characteristic of the different populations of phagocytic cells, and define how these features
vary across the cell types and their locations. Such computational task has not been
addressed previously for the datasets of such scale. Indeed, we previously described a
computational approach, called GAM12, that uses metabolic networks as the backbone for
analysis of transcriptional data and provides a verifiable and systematic description of the
metabolic differences between two conditions9. However, datasets in question contain
hundreds or even thousands of individual profiles, while GAM approach is designed to analyse
3
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
comparison between two conditions. Therefore, in this work we have developed novel
computational approach, GAM-clustering, which performs unbiased search of a collection of
metabolic subnetworks that jointly define metabolic variability across large datasets (available
on GitHub, see Methods). By doing so, GAM-clustering reveals metabolically similar
subpopulations in a manner that does not require explicit annotation or pair-wise comparison
of individual samples. Our analysis revealed major metabolic features associated with different
cell subpopulations and highlighted a number of metabolic modules that are specific to
individual cell types, tissues of residence, or developmental stages. As an example, GAM-
clustering analysis revealed that cholesterol de novo synthesis pathway might play an
important role in the context of migratory dendritic cells (DCs), which we validated using in vivo
pharmacological inhibition of this pathway followed by tracking of DC migration. Consistent
with the analysis, statins have demonstrated inhibitory effect on DC migratory ability, finding
that has not been reported previously.
Taken together, our work provides both (1) unique data and analysis resource in terms of
studying variability of mononuclear phagocytes, as well as (2) validated computational
approach that can unbiasedly analyse both single-cell RNA-seq data as well as multi-sample
bulk RNA-seq datasets in terms of key underlying metabolic features. Furthermore, we provide
direct interactive access to the data for examination and visualization through both single-cell
RNA-seq and bulk RNA-seq visualization servers including metabolic cluster annotations
obtained in this work (https://artyomovlab.wustl.edu/immgen-met/).
RESULTS
Mononuclear Phagocytes Open Source (MNP OS) and Mononuclear Phagocytes from
ImmGen Phase 1 (MNP P1) Datasets
As a part of the Open Source ImmGen Project, a total of 337 samples were collected and
profiled through the collaborative effort of 16 laboratories (Fig. 1d, e, Supplementary Fig. 1a,
Supplementary Table 1). Each laboratory sorted specific populations of mononuclear
phagocytes from 26 distinct tissues, isolated RNA from these populations and submitted it for
centralized deep RNA-sequencing and subsequent quantitation. Along with their samples of
interest, each laboratory included RNA from locally sorted peritoneal macrophages as a
common control for evaluation/correction of potential batch effects (Methods). Of note, 15
samples from MNP OS dataset were previously used in the study of sexual dimorphism of the
4
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
immune system transcriptome13, while complete dataset is analysed in this work for the first
time.
Overall, the transcriptional data demonstrated high concordance between different collection
sites, and were merged into a final transcriptional master table (Supplementary Fig. 1b, c,
Supplementary Table 2, Methods). Previously established markers of individual myeloid
subpopulations14–18 matched well with the sorted populations (Supplementary Fig. 2),
indicating overall consistency of the dataset across different research groups. As individual
principal component analysis (PCA) plots show (Fig. 1d), samples have clustered in accord
with their broad annotation as macrophages, DCs, monocytes or microglia. Generally,
subpopulation-specific effects were stronger than tissue specific differences within individual
subpopulations as evident by comparing Figures 1d and 1e. To estimate the degree of
metabolic variability in the data, we examined the enrichment of annotated metabolic pathways
in this dataset, revealing coherent transcriptional patterns across individual subpopulations
(Supplementary Fig. 3a). This indicated that systematic evaluation of the metabolic
subnetworks within the data is warranted.
Initial ImmGen Phase 1 data published previously5 include 202 samples of mononuclear
phagocytes with higher contribution of progenitor populations, and smaller number of microglial
samples (Fig. 1f) overall spanning 16 tissues (Fig. 1g) – we will refer to this subset as
ImmGen MNP P1 from here onward. Similar to MNP OS dataset, enrichment in metabolic
pathways across subpopulations in MNP P1 data has demonstrated coordinate variations
across the samples (Supplementary Fig. 3b).
Single-cell Myeloid Tabula Muris Senis (mTMS) dataset
Tabula Muris consortium has performed single-cell RNA-sequencing for a large number of
tissues without explicit sorting into individual cell populations11. These data include myeloid
cells localized in the corresponding tissues which can be computationally separated based on
the expression of common myeloid signatures. Using latest public dataset, Tabula Muris Senis,
we have analysed the data for 235,325 cells to identify 51,364 myeloid cells (mononuclear
phagocytes and neutrophils) that expressed key myeloid markers (Lyz2, H2-Aa, Mki67,
S100a9, Flt3, Emr1, Ccr2, Cx3cr1, Sall1, Clec4f, Lyve1, Itgax, Xcr1, Clec4a4, Siglech, Ccr7,
Fig. 2a,d, Methods). These cells comprised dedicated dataset, further referred to as myeloid
Tabula Muris Senis dataset (mTMS). While single-cell RNA-seq data inevitably detect smaller
number of genes per cell compared to bulk RNA-sequencing (Supplementary Fig. 4), the
5
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
depth of the mTMS dataset was sufficient to resolve classical cell populations. Specifically,
unbiased clustering revealed 15 subpopulations within mTMS dataset (Fig. 2b), which could
be readily identified as well described populations of plasmacytoid DCs, monocytes, Kupffer
cells, microglia etc (Fig. 2c) based on previously described cell-specific markers (Fig. 2d). To
our knowledge, we provide the first large-scale curated annotation for the myeloid cells within
Tabula Muris Senis data. Corresponding annotations are available for hands-on exploration in
the interactive single-cell browser (http://artyomovlab.wustl.edu/immgen-met/, see Tabula
Muris Senis).
GAM-clustering: identification of metabolic subnetworks in datasets with multiple
conditions
Previously, we have shown that metabolic remodelling between two conditions can be
analysed using network-based analysis of their transcriptional profiles9,12. Specifically, the GAM
(‘Genes and Metabolites’) algorithm searches for optimal subnetworks within a global
metabolic network by weighing individual enzymes in accord with differential expression of
their genes and then solving the generalized maximum-weight connected subgraph (GMWCS)
problem12,19. While this approach cannot be directly translated to multi-sample/single-cell
datasets such as ImmGen or Tabula Muris Senis data, we were able to reformulate weighting
scheme in a manner that allows GMWCS subnetwork search without explicit annotation of
individual samples or conditions. Here, we describe novel algorithm called GAM-clustering that
allows the user to obtain metabolic subnetworks enriched within the transcriptional data that
include many samples across multiple conditions.
In brief, GAM-clustering searches for connected metabolic subnetworks that have most
correlated expressions of individual enzymes, resulting in a collection of subnetworks that
follow distinct transcriptional profiles. To achieve that, we first initialize a pattern-generating
profiles by clustering all metabolic genes based on their co-expression patterns (Fig. 3, see
Methods and Supplementary Material for details). For initialization of the multisample bulk
RNA-seq data we use k-medoids clustering with k=32 (see Supplementary Material and
Supplementary Fig. 5a,b for parameters sensitivity), any other gene expression clustering
approach can be used in this step since downstream steps include significant re-grouping and
merging of individual clusters. For initialization of single-cell RNA-seq data, we first cluster the
cells in the dataset to multiple clusters (~100) which provides sufficient balance between fine
resolution of the data and minimal coarse-graining needed to avoid drop-out artefacts (see
6
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
Methods and Supplementary Material for details). Then, genes are clustered using the same
procedure as for multisample bulk datasets.
Next, each enzyme is weighed with respect to the similarity of its transcript’s profile to the
average cluster/pattern profile, resulting in multiple weights per gene that are specific for each
pattern (Fig. 3). For any given pattern, weights of individual enzymes serve as input to the
GMWCS solver, resulting in the individual subnetworks that are associated with each pattern.
Individual subnetworks are then refined in an iterative procedure of updating the gene content
for each pattern (see Supplementary Material). The final output presents a set of specific
subnetworks that reflect metabolic variability within a given transcriptional dataset (Fig. 3).
Major metabolic modules within mononuclear phagocyte subpopulations
The GAM-clustering algorithm was applied to data from all baseline (non-infected) samples
and yielded nine distinct metabolic modules (Fig. 4a, Supplementary Table 3). Hierarchical
clustering of samples based on the Euclidean distance metric in space of these nine metabolic
modules showed that they can be broadly separated based on the cell types: yolk sac
macrophages, dendritic cells, monocytes, and macrophages from adult organism. Broadly
defined mononuclear cell types are further split into several smaller metasamples: dendritic
cells subdivided into plasmacytoid dendritic cells (pDCs), tissue specific dendritic cells, and
migratory dendritic cells (migDCs), and macrophages subdivided into microglia, adipose tissue
macrophages, and a large metasample of tissue residing macrophages, as well as an
additional metasample composed of embryoid body, alveolar, and small peritoneal
macrophages (SPMs) that clustered distinctly from other macrophage subpopulations (Fig.
4a).
While obtained metabolic modules/subnetworks provide a more accurate description of
metabolic diversity compared to canonically annotated pathways, the latter can be useful for
coarse-grained understanding of functionalities associated with each subnetwork (Fig. 4b).
Indeed, pathway enrichment analysis along with subnetworks gene content analysis indicate
that modules 1, 2, 8 and 9 represent various aspects of lipid metabolism, and modules 3, 4 –
two types of fatty acid synthesis pathways. Finally, distinct modules represent cholesterol
synthesis metabolism (module 5), glycolysis (module 6) and nucleotide/folate metabolism
associated subnetworks (module 7).
7
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
The underlying metabolic phenotypes for each metasample can be represented using radar
chart diagrams (Fig. 4c): each metasample is defined by a specific combination of metabolic
features that provides unique insights into metabolic wiring within those populations. Here, the
names to metasamples are given based on the most common sample type inside the cluster.
An alternative view of the samples in the space of metabolic modules can be obtained using
PCA that is built based on only 9 metabolic modules, which shows distinct separation of
individual metasamples (Supplementary Fig. 6a). Consistently, when overlaid with the PCA
representation from Figure 1, individual metabolic modules formed coherent patterns
indicating the groups of metabolically similar samples (Supplementary Fig. 6b). Altogether,
the metabolic modules/subnetworks and corresponding metasamples encapsulate metabolic
variability across both cell types and their tissues of residency. We next turn to examining
robustness of the obtained subnetworks across three considered datasets.
Three independent large-scale datasets show consistent metabolic features
We next considered if metabolic subnetworks derived from ImmGen MNP OS data can be
seen in the other two large scale datasets considered in this work – ImmGen MNP P1 and
mTMS datasets. While overlap in profiled tissues is considerate (Fig. 1b), 3 datasets are not
identical in terms of populations profiled. We, therefore, grouped the samples into 19 general
classes and compared the datasets by looking at the metabolic enrichments across these
classes (Fig. 5a, Supplementary Table 4). To examine robustness of metabolic signatures,
we computed enrichments of individual metabolic modules from Figure 4a in each of the 19
representative classes of ImmGen MNP OS, ImmGen MNP P1 and mTMS. Indeed, all
datasets modules demonstrated extremely similar enrichment profiles (Fig. 5b): for instance,
Module 1 was enriched in microglia, adipose tissue macrophages and Kuppfer cells, module 8
enriched in alveolar macrophages, module 5 – in pDCs and migratory DCs across all
datasets, etc.
Importantly, independent application of the GAM-clustering algorithm to each of the datasets
also revealed very high degree of similarity in obtained modules, highlighting reproducible and
robust nature of the derived metabolic subnetworks (Supplementary Fig. 7, Supplementary
Table 3, Supplementary Material).
Next we examined individual subnetworks from the perspective of metabolic reactions covered
and describe both published evidence of the corresponding metabolic activities as well as
validation data obtained in this project.
8
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
Subnetworks associated with early developmental stages
Module 6 (Fig. 6a,b,d) is one of the modules most distinctly associated with yolk sac,
embryoid body, alveolar, and small peritoneal macrophages. This module, though unbiasedly
derived by our network analysis, closely matches the canonical glycolysis pathway (Fig. 6b),
indicating strong transcriptional co-regulation of these genes across the collected samples.
Enrichment of the glycolysis module in developmental cell types is consistent with previously
published data highlighting the importance of glycolysis for stem-like and progenitor
populations20–23. This is also consistent with the ImmGen MNP P1 and mTMS data (Fig. 6d),
where this module is also most enriched in progenitor populations. Interestingly, mTMS single-
cell RNA-seq data also demonstrate that this module is enriched in neutrophils, in accord with
described high glycolytic rate in these cells24.
Module 7 (Fig. 6a,c,d) represents another set of metabolic activities including folate and serine
metabolism as well as the nucleotide biosynthesis pathway typically associated with the
progenitor populations25–27. In addition to the yolk sac macrophages, this module is also
enriched in some tissue residing dendritic cells and pDCs (but not in migDCs). Indeed, the
importance of some of these pathways (e.g. folate metabolism) has been demonstrated in
dendritic cell functions such as antigen presentation28.
Cholesterol synthesis pathway is enriched in and functionally important for migratory
DCs
Module 5 almost exclusively consists of enzymes from the cholesterol metabolism/mevalonate
synthesis pathway (Fig. 6d,e,f), and is enriched in embryoid body macrophages and some
dendritic cells. Specifically, cholesterol synthesis appears to play a major role in migDCs, while
it is less prominent in pDCs and conventional tissue residing dendritic cells. Additionally, with
respect to potential tissue-specific imprinting, it is worth noting that a small subset of tissue
residing macrophages, comprised of epithelial and dermal macrophages, is enriched in genes
of the mevalonate/cholesterol synthesis pathway (Fig. 6e). Enrichment of cholesterol
metabolism in migratory dendritic cells was consistent with mechanistic data by Hauser and
colleagues who showed that cellular cholesterol levels are directly linked to the ability of
dendritic cells to oligomerize Ccr7 (key marker of migDCs) and acquire a migratory
phenotype29. Given the results of our analysis and these published mechanistic connections,
we evaluated mobilization of DCs to lymph node following epicutaneous application of
fluorescein isothiocyanate (FITC) in either control animals or animals treated i.p. with low-dose
9
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
Simvastatin (0.57 mg/kg/day) for seven days (Fig. 6g). Draining lymph nodes collected 18 h
after FITC application demonstrated significantly fewer migrated FITC+CD11c+ dendritic cells in
the animals treated with Simvastatin, illustrating that in vivo interference with cholesterol
synthesis reduces dendritic cell migration to the lymph node, fitting with the prominent
expression of cholesterol synthesis genes in DCs. This results illustrates general validity of our
analysis and highlights novel features of the systemic metabolic perturbations, such as statin
treatments, that we previously not recognized.
Subnetworks associated with lipid metabolism
Modules 1 and 2 cover various aspects of lipid metabolism and are strongly specific to
macrophages relative to monocytes and dendritic cells (Fig. 7a). Due to general similarity of
their patterns, we merged the subnetworks for modules 1 and 2 in order to make the
interpretation easier (Fig. 7b, Supplementary Fig. 8a,b). The resulting subnetwork is centred
around phospholipid and arachidonic acid metabolism, and includes parts of the glutathione
and cysteine/glutamate/glycine metabolism pathways, as well as the N-acetylglucosamine
pathway. Indeed, arachidonic acid metabolism has been shown to play major roles in
macrophages30,31. Its metabolic flow is associated with utilization of phospholipids to produce
two major classes of the arachidonic acid derivatives: leukotrienes and prostaglandins. Unlike
prostaglandins, leukotriene production (C4 and downstream) requires glutathione as an
intermediate metabolite, thus involving the glycine, cysteine and glutamate pathways32.
Furthermore, our analysis picked up a distinct subnetwork of co-expressed genes from the
glycerophospholipid pathway (Module 9, see Fig. 7a,c,d) that was particularly highly
expressed in the microglial populations (Fig. 7a). This module included enzymes such as Dgkd
and Lpcat2, suggesting that their role in microglia might be of particular interest33,34. As Fig. 7d
shows, these observations were common across all three datasets.
Subnetworks associated with fatty acid synthesis and degradation
Our analysis identified three distinct subnetworks associated with modulation of fatty acids in
terms of both their synthesis (modules 3 and 4) and fatty acid oxidation (module 8).
The structure of Module 3 (Supplementary Fig. 8a,b,d) reflects energetic demands of the
fatty acid synthesis and includes portions of pentose phosphate pathway and TCA cycle,
where citrate synthase (Cs) is one of the most pattern-specific genes within this subnetwork.
Overall, module 3 is highly enriched in dendritic cell populations, but not in
10
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
macrophage/monocyte samples, underscoring another facet of metabolic divergence between
these cell types. The functional importance of this module for dendritic cells is evident from the
fact that a blockade of Fasn-mediated fatty acid synthesis markedly and selectively decreases
dendropoiesis both in mice and in humans35,36.
Interestingly, the pattern of Module 8 (Fig. 7e-g) was directly opposite to Module 3, and was
strongly enriched among various tissue macrophages, particularly in alveolar macrophages.
Metabolic flow encompassed by this network includes enzymes such as Lipa (LAL), which is
responsible for lysosomal lipolysis and initial breakdown of intracellular lipid storage. This
breakdown is followed by mitochondrial import of cytosolic fatty acids via carnitine transport
shuttle (Cpt1a) and their subsequent breakdown via classical FAO steps (Acox1, Hadha,
etc)37,38 (Fig. 7e). The Lipa expression pattern is one of the most specific for module 8,
indicating its potential importance for macrophages. Indeed, there are studies highlighting the
importance of Lipa for macrophage function, especially in the context of anti-inflammatory
polarization39. Furthermore, Lipa is also likely to be important for human macrophages, as
mutations in the LIPA gene of patients with cholesteryl ester storage disease (CESD) cause
aberrant cholesterol accumulation in tissue macrophages40,41. Enrichment of fatty acid
oxidation-related Module 8 in alveolar macrophages is particularly interesting as it is distinctly
reproduced in ImmGen MNP P1 and mTMS data. Importance of this pathway in lungs is
intriguing and warrants further detailed investigations.
DISCUSSION
Here we introduced unique dataset covering multiple subpopulations of dendritic cells,
monocytes and macrophages from diverse tissues – result of ImmGen MNP Open Source
profiling effort. We focused on understanding potential metabolic variability among collected
myeloid cell subpopulations and co-analysed it in the context of two other large-scale profiling
efforts – ImmGen Phase 1 and Tabula Muris Senis. Using new algorithmic approach (GAM-
clustering), we have defined 9 major metabolic subnetworks that encapsulate the major
metabolic differences that were highly reproducible across three studied datasets. Our analysis
demonstrated that specific metabolic features could be attributed to cell populations as well as
specific tissue of residence for distinct populations (e.g. adipose tissue macrophages).
Our analysis suggested that major metabolic differences between baseline (unactivated)
macrophages and dendritic cells are (1) levels of Fasn-mediated fatty acid synthesis enriched
in dendritic cells’ transcriptional profiles, and (2) regulation of arachidonic acid metabolism,
11
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
which is enriched in macrophages. Among various tissue residing cell types, it was apparent
that microglia and CNS macrophages have a very distinct phenotype relative to other
populations: based on their transcriptional profile they appear more metabolically quiescent,
yet a particular lipid-associated module (module 9) was enriched in these cells, with key genes
being Lpcat2, Dgkd, Csd1 that are involved in phospholipid metabolism and the generation of
bioactive lipids from phospholipid precursors.
Indeed, distinct patterns in lipid metabolism, including pathways related to cholesterol, were
also apparent in dendritic cells versus macrophages. Macrophages' capacity to handle
cholesterol and store it in esterified form to generate so-called macrophage foam cells is a
well-established theme in cardiovascular research and inflammatory disease42,43. Our data
reveal that expression of Lipa, an enzyme involved in breaking down cholesterol esters in the
lysosome and whose mutation is associated with lysosomal storage diseases, is a widespread
characteristic of tissue macrophages but not dendritic cells. On the contrary, we observed that
pathways active in cholesterol synthesis are very low in all tissue macrophages but elevated in
monocytes and dendritic cells, especially migratory dendritic cells.
Thus, it appears as though macrophages are oriented toward handling exogenously derived
cholesterol, such as that which may be derived from engulfment of large amounts of
phagocytic cargo, whereas dendritic cells are oppositely programmed to synthesize their own
cholesterol and associated intermediates. As tissue macrophages are especially incapable of
migrating to distal sites like lymph nodes, a major functional distinction from dendritic cells, we
have validated importance of the cholesterol synthesis pathway for the migratory phenotype in
vivo by using pharmacological interventions with Simvastatin.
Altogether, our analysis underscores metabolic variability across cell types and tissues and
highlights the need to understand metabolic wiring, not only in terms of cellular metabolism,
but also at the level of whole-body communication networks (see e.g. Castillo-Armengol and
colleagues44, Droujinine and Perrimon45). Furthermore, since direct metabolic profiling is not
feasible or sufficiently accurate at the moment, the development of ex vivo metabolomics
profiling technologies46–48 suggests that direct insight into metabolism of various myeloid
subpopulations through in vivo metabolomics techniques will be possible in future.
12
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
METHODS
RNA-sequencing
Bulk RNA-sequencing data were collected from 16 labs. All of the mice used in this study were
handled in accordance with IACUC-approved protocols. Each lab, in addition to their own
samples, sorted a standard peritoneal cavity macrophage population (CD115+B220-
F4/80hiMHCII-) for comparability between all labs. Samples were profiled using ImmGen’s ultra
low input (ULI) sequencing pipeline, in batches of 90-96 samples. All samples were sequenced
in two separate NextSeq500 runs and combined for increased depth (expect 8-12 106 reads
per sample). Following sequencing, raw reads were aligned with STAR to the mouse genome
assembly mm10, and assigned to specific genes using the GENCODE vM12 annotation.
Aligned reads were quantified using featureCounts. Samples that did not pass the QC
threshold for read counts (<2 million reads) were dropped for further analysis. Pearson
correlation was calculated between biological replicates to exclude samples that did not pass a
threshold of 0.9 correlation coefficient. For the cell populations with three biological replicates,
of which one did not agree with the other two, the suspect one was removed from the data set.
In case cell populations had only two replicates, both were removed. Samples with
Jchain>1,000 and Ighm>10,000 were set asides as well as samples with high B cell,
erythrocytes and fibroblasts transcripts. Peritoneal cavity samples were downsampled to keep
consistency across samples number in all tissues.
RNA-sequencing data processing
All gene counts were imported into the R/Bioconductor package EdgeR and TMM
normalization size factors were calculated to adjust for differences in library size across all
samples. Feature not expressed in at least three samples above one count-per-million were
excluded from further analysis and TMM size factors were recalculated to create effective TMM
size factors. The effective TMM size factors and the matrix of counts were then imported into
the R/Bioconductor package Limma and weighted likelihoods based on the observed mean-
variance relationship of every gene and sample were then calculated for all samples.
Performance of the samples was assessed with a Pearson correlation matrix and
multidimensional scaling plots.
Single-cell RNA-seq data processing
13
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
Filtered h5ad file for Droplet subset was downloaded from the official Tabula Muris Senis
repository (https://figshare.com/projects/Tabula_Muris_Senis/64982). The data were
processed by the standard Seurat pipeline and resulted in 235,325 cells organised in distinct
clusters detectable on TSNE/UMAP plots. Next, cells annotated with names corresponding to
myeloid populations were picked out. A differential gene expression analysis between these
cells and all others was performed. Top 250 of these differentially expressed genes were used
as a “myeloid signature genes” (Supplementary Table 5) to identify clusters that most express
them and thus correspond to myeloid cells. Cell content of these clusters was used to create a
new subset of 60,844 cells. Obtained dataset was analysed by non-myeloid marker genes to
detect and remove cell doublets with T-cells, B-cells, NK-cells and fibroblasts (Cd3d, Cd3e,
Cd3g, Cd4, Cd8a, Cd19, Cd79a, Tnfrsf17, Cd22, Nkg7, Gnly, Col6a1, Col6a2, Col6a3). Finally,
dataset of 51,364 cells was obtained and used in the further GAM-clustering analysis.
GAM-clustering
The algorithm for multisample metabolic network clustering (hereinafter referred to as GAM-
clustering, see Supplementary Material and Supplementary Fig. 5 for details) identifies
modules describing dynamic regulation of metabolism and is based on the previously
developed GAM method12. GAM-clustering extends the GAM method by setting the task to find
not one but several metabolic modules (connected subnetworks of metabolic network) with the
condition that each of these modules should contain as many metabolic genes with high
pairwise correlation of their expression as possible.
The initial approximation of the final set of modules is carried out by k-medoids clustering of a
gene expression matrix for all metabolic genes of a dataset with some arbitrary k (here we
used k=32). Each cluster forms a corresponding expression pattern which can be determined
as averaged value of its z-normalized gene expression values. The metabolic network used for
further analysis is presented as a graph where vertices are metabolites and edges are KEGG
database reactions which are mapped with catalysing them enzymes and corresponding
genes. For each particular pattern edges of this graph are scored (weighted) based on their
gene expression similarity with this pattern and dissimilarity with other patterns.
For each case of weighted graph a connected subgraph of maximal weight is found by a signal
GMWCS (generalized maximum weight connected subgraph) solver49 (https://github.com/ctlab/
sgmwcs-solver) and is called a metabolic module. This solver uses the IBM ILOG CPLEX
library, which efficiently performs many iterations of this method in a reasonable amount of
14
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
time. Then, each pattern is updated by replacing it with an averaged gene expression of the
module’s edges with a positive score. If the pattern is changed, a new score set is calculated
and a new iteration is performed. Before moving to the next iteration, small graphs are
eliminated from further analysis so that there are no graphs with less than five edges and
diameter less than four in the output solution. The algorithm continues until the pattern content
stops changing.
GAM-clustering method is applicable not to bulk RNA-seq data only but to single-cell RNA-seq
data as well. Single-cell data need an additional step of preprocessing implying transformation
of individual cells into technical samples. This is performed based on averaging gene
expressions of individual cells inside high resolution clusters. In case of single-cell RNAseq
data, among final metabolic modules might occur ones that do not cover all biological replicas
of cell types they are specific for. These modules are eliminated from the final result.
Thus, the final metabolic modules are subnetworks of the overall metabolic network that
contain a set of closely located genes with high correlation of their expression profile across all
samples.
GAM-clustering method is available at https://github.com/artyomovlab/ImmGenOpenSource.
DC migration
Epicutaneous application of Fluorescein isothiocyanate (FITC) to study DC migration was
performed on three areas of each side of the mouse back skin as described previously 50. Both
females and males were studied. Briefly, FITC (8 mg/ml) was dissolved in acetone and dibutyl
phthalate (Sigma-Aldrich, F7250) and applied in 25-μl aliquots to each site. Recovered LNs, 18l aliquots to each site. Recovered LNs, 18
h later, were teased and digested in 2.68 mg/ml collagenase D (Roche) for 25 min at 37°C.
Then, 100 μl aliquots to each site. Recovered LNs, 18l 100 mM EDTA was added for 5 min, and cells were passed through a 100-μl aliquots to each site. Recovered LNs, 18m cell
strainer, washed, counted, and stained for flow cytometry after counterlabeling with PE
conjugated anti-CD11c (Biolegend). Prior to FITC painting, some cohorts of mice were treated
with simvastatin i.p. at 0.57 mg/kg/day for 7 days, as this protocol was previously shown to
significantly block monocyte diapedesis from the bloodstream51. Control mice received vehicle
i.p.
Data availability
15
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
The bulk RNA-sequencing data are deposited in the GEO repository under the accession
codes GSE122108 and GSE15907. Interactive gene expression heatmaps for both ImmGen
and Tabula Muris Senis datasets as well as metabolic modules described in the present
research are available by the provided link https://artyomovlab.wustl.edu/immgen-met/.
Supplementary information
Supplementary Figures 1–9, Supplementary Tables 1–5 and Supplementary Material.
REFERENCES
1. Monticelli, S. & Natoli, G. Transcriptional determination and functional specificity of myeloid cells: Making sense of diversity. Nature Reviews Immunology vol. 17 595–607 (2017).
2. De Kleer, I., Willems, F., Lambrecht, B. & Goriely, S. Ontogeny of myeloid cells. Front. Immunol. 5, (2014).
3. Jung, J., Zeng, H. & Horng, T. Metabolism as a guiding force for immunity. Nature Cell Biology vol. 21 85–93 (2019).
4. Caputa, G., Castoldi, A. & Pearce, E. J. Metabolic adaptations of tissue-resident immunecells. Nature Immunology vol. 20 793–801 (2019).
5. Gautier, E. L. et al. Gene-expression profiles and transcriptional regulatory pathways thatunderlie the identity and diversity of mouse tissue macrophages. Nat. Immunol. 13, 1118–1128 (2012).
6. Llufrio, E. M., Wang, L., Naser, F. J. & Patti, G. J. Sorting cells alters their redox state and cellular metabolome. Redox Biol. 16, 381–387 (2018).
7. Binek, A. et al. Flow Cytometry Has a Significant Impact on the Cellular Metabolome. J. Proteome Res. 18, 169–181 (2019).
8. Richardson, G. M., Lannigan, J. & Macara, I. G. Does FACS perturb gene expression? Cytom. Part A 87, 166–175 (2015).
9. Jha, A. K. et al. Network integration of parallel metabolic and transcriptional data reveals metabolic modules that regulate macrophage polarization. Immunity 42, 419–430 (2015).
10. Benoist, C. Open-source ImmGen: Mononuclear phagocytes. Nature Immunology vol. 17741 (2016).
11. The Tabula Muris Consortium et al. A Single Cell Transcriptomic Atlas Characterizes Aging Tissues in the Mouse. bioRxiv 661728 (2019) doi:10.1101/661728.
16
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
12. Sergushichev, A. A. et al. GAM: a web-service for integrated transcriptional and metabolic network analysis. Nucleic Acids Res. 44, W194–W200 (2016).
13. Gal-Oz, S. T. et al. ImmGen report: sexual dimorphism in the immune system transcriptome. Nat. Commun. 10, (2019).
14. Davies, C. L., Patir, A. & McColl, B. W. Myeloid cell and transcriptome signatures associated with inflammation resolution in a model of self-limiting acute brain inflammation. Front. Immunol. 10, (2019).
15. Lee, M., Lee, Y., Song, J., Lee, J. & Chang, S. Y. Tissue-specific role of CX3CR1 expressing immune cells and their relationships with human disease. Immune Network vol. 18 (2018).
16. Misharin, A. V., Morales-Nebreda, L., Mutlu, G. M., Budinger, G. R. S. & Perlman, H. Flow cytometric analysis of macrophages and dendritic cell subsets in the mouse lung. Am. J. Respir. Cell Mol. Biol. 49, 503–510 (2013).
17. Holtman, I. R., Skola, D. & Glass, C. K. Transcriptional control of microglia phenotypes inhealth and disease. Journal of Clinical Investigation vol. 127 3220–3229 (2017).
18. Merad, M., Sathe, P., Helft, J., Miller, J. & Mortha, A. The Dendritic Cell Lineage: Ontogeny and Function of Dendritic Cells and Their Subsets in the Steady State and the Inflamed Setting. Annu. Rev. Immunol. 31, 563–604 (2013).
19. Loboda, A. A., Artyomov, M. N. & Sergushichev, A. A. Solving generalized maximum-weight connected subgraph problem for network enrichment analysis. Lect. Notes Comput. Sci. (including Subser. Lect. Notes Artif. Intell. Lect. Notes Bioinformatics) 9838 LNCS, 210–221 (2016).
20. Wang, Y. H. et al. Cell-state-specific metabolic dependency in hematopoiesis and leukemogenesis. Cell 158, 1309–1323 (2014).
21. Shyh-Chang, N., Daley, G. Q. & Cantley, L. C. Stem cell metabolism in tissue development and aging. Development (Cambridge) vol. 140 2535–2547 (2013).
22. Folmes, C. D. L. & Terzic, A. Stem cell lineage specification: You become what you eat. Cell Metabolism vol. 20 389–391 (2014).
23. Oburoglu, L., Romano, M., Taylor, N. & Kinet, S. Metabolic regulation of hematopoietic stem cell commitment and erythroid differentiation. Current Opinion in Hematology vol. 23 198–205 (2016).
24. Kumar, S. & Dikshit, M. Metabolic Insight of Neutrophils in Health and Disease. Frontiersin Immunology vol. 10 (2019).
25. Henry, C. J. et al. Folate dietary insufficiency and folic acid supplementation similarly impair metabolism and compromise hematopoiesis. Haematologica 102, 1985–1994 (2017).
17
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
26. Lucas, D. L., Webster, H. K. & Wright, D. G. Purine metabolism in myeloid precursor cells during maturation. Studies with the HL-60 cell line. J. Clin. Invest. 72, 1889–1900 (1983).
27. Locasale, J. W. Serine, glycine and one-carbon units: Cancer metabolism in full circle. Nature Reviews Cancer vol. 13 572–583 (2013).
28. Wu, C. H., Huang, T. C. & Lin, B. F. Folate deficiency affects dendritic cell function and subsequent T helper cell differentiation. J. Nutr. Biochem. 41, 65–72 (2017).
29. Hauser, M. A. et al. Inflammation-Induced CCR7 Oligomers Form Scaffolds to Integrate Distinct Signaling Pathways for Efficient Cell Migration. Immunity 44, 59–72 (2016).
30. Scott, W. A. et al. Regulation of arachidonic acid metabolism by macrophage activation. J. Exp. Med. 155, 1148–1160 (1982).
31. Boraschi, D., Censini, S., Bartalini, M. & Tagliabue, A. Regulation of arachidonic acid metabolism in macrophages by immune and nonimmune interferons. J. Immunol. 135, 502–5 (1985).
32. Rouzer, C. A., Scott, W. A., Griffith, O. W., Hamill, A. L. & Cohn, Z. A. Depletion of glutathione selectively inhibits synthesis of leukotriene C by macrophages. Proc. Natl. Acad. Sci. U. S. A. 78, 2532–2536 (1981).
33. Shindou, H. et al. Relief from neuropathic pain by blocking of the platelet-activating factor–pain loop. FASEB J. 31, 2973–2980 (2017).
34. Nakano, T. et al. Brain trauma induces expression of diacylglycerol kinase ζ in microglia. Neurosci. Lett. 461, 110–115 (2009).
35. Pearce, E. J. & Everts, B. Dendritic cell metabolism. Nature Reviews Immunology vol. 1518–29 (2015).
36. Rehman, A. et al. Role of Fatty-Acid Synthesis in Dendritic Cell Generation and Function. J. Immunol. 190, 4640–4649 (2013).
37. Nomura, M. et al. Fatty acid oxidation in macrophage polarization. Nature Immunology vol. 17 216–217 (2016).
38. Michel Houten, S. & A Wanders, R. J. A general introduction to the biochemistry of mitochondrial fatty acid β-oxidation. doi:10.1007/s10545-010-9061-2.
39. Huang, S. C. C. et al. Cell-intrinsic lysosomal lipolysis is essential for alternative activation of macrophages. Nat. Immunol. 15, 846–855 (2014).
40. Bernstein, D. L., Hülkova, H., Bialer, M. G. & Desnick, R. J. Cholesteryl ester storage disease: Review of the findings in 135 reported patients with an underdiagnosed disease. Journal of Hepatology vol. 58 1230–1243 (2013).
18
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
41. Li, F. & Zhang, H. Lysosomal Acid Lipase in Lipid Metabolism and beyond. Arterioscler. Thromb. Vasc. Biol. 39, 850–856 (2019).
42. Ghosh, S. Macrophage cholesterol homeostasis and metabolic diseases: Critical role of cholesteryl ester mobilization. Expert Review of Cardiovascular Therapy vol. 9 329–340 (2011).
43. Maguire, E. M., Pearce, S. W. A. & Xiao, Q. Foam cell formation: A new target for fightingatherosclerosis and cardiovascular disease. Vascular Pharmacology vol. 112 54–71 (2019).
44. Castillo Armengol, J., Fajas, L. & Lopez Mejia, I. C. Inter organ communication: a ‐Armengol, J., Fajas, L. & Lopez‐Mejia, I. C. Inter‐organ communication: a ‐Armengol, J., Fajas, L. & Lopez‐Mejia, I. C. Inter‐organ communication: a ‐Armengol, J., Fajas, L. & Lopez‐Mejia, I. C. Inter‐organ communication: a gatekeeper for metabolic health. EMBO Rep. 20, (2019).
45. Droujinine, I. A. & Perrimon, N. Defining the interorgan communication network: Systemic coordination of organismal cellular processes under homeostasis and localizedstress. Front. Cell. Infect. Microbiol. 4, (2013).
46. Gianchandani, E. P., Chavali, A. K. & Papin, J. A. The application of flux balance analysisin systems biology. Wiley Interdiscip. Rev. Syst. Biol. Med. 2, 372–382 (2010).
47. Basler, G., Fernie, A. R. & Nikoloski, Z. Advances in metabolic flux analysis toward genome-scale profiling of higher organisms. Bioscience Reports vol. 38 (2018).
48. Hastings, J. et al. Multi-omics and genome-scale modeling reveal a metabolic shift during C. elegans aging. Front. Mol. Biosci. 6, (2019).
49. Ulland, T. K. et al. TREM2 Maintains Microglial Metabolic Fitness in Alzheimer’s Disease.Cell 170, 649-663.e13 (2017).
50. Platt, A. M. et al. Normal Dendritic Cell Mobilization to Lymph Nodes under Conditions of Severe Lymphatic Hypoplasia. J. Immunol. 190, 4608–4620 (2013).
51. Haka, A. S., Potteaux, S., Fraser, H., Randolph, G. J. & Maxfield, F. R. QuantitativeAnalysis of Monocyte Subpopulations in Murine Atherosclerotic Plaques by MultiphotonMicroscopy. PLoS One 7, (2012).
19
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
END NOTES
AcknowledgmentsWe thank Amanda Swain, Monika Bambouskova and Laura Arthur for constructive criticism ofthe manuscript. This work was supported in part by the Division of Intramural Research of theNIAID, NIH. The work was also partly supported by R01-AI125618 (NIAID) to M.N.A.
Author contributionsA.S., M.N.A., A.G. designed the study and interpreted results. A.G. performed all bioinformatic experiments and draw figures.L.-H.H. hold the DC migration experiment. L.S.Y., A.K., B.J, K.Kr., J.W., K.Cr., E.T., R.D., P.S., C.S., S.G., G.B., J.V.D., B.M., S.T., K.Ki.,P.W., S.V.A., V.A., J.M.B., K.Ch., M.D., I.D., E.L.G., C.J., K.L., M.S.L., H.P., M.H.S., L.P., S.D.,T.-A.Z. performed or participated in fluorescence-activated cell sorting as part of ImmGen MNPOS project.C.B. and M.M. oversaw the ImmGen MNP OS logistics and overall data/sample collection.K.S. processed raw ImmGen MNP OS RNA-seq data and produced gene counts table.H.T. performed ImmGen MNP OS gene counts table normalization.M.N.A., A.G., M.G., F.G., M.M., G.J.R., V.N. and A.S. participated in study discussion andprovided critical insights to the study.M.N.A. and A.G. wrote the manuscript and all the authors contributed to editing andsuggestions.
Competing interestsAuthors declare no conflicts of interests.
20
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
FIGURE LEGENDS
Figure 1. General overview of ImmGen Mononuclear Phagocytes Open Source (IG MNPOS), ImmGen Mononuclear Phagocytes Phase 1 (IG MNP P1) and myeloid Tabula MurisSenis (mTMS) datasets. a, Schematic representation of Mus musculus tissues wheresamples were derived from (marked with colored dots depending on the dataset). b, Numberof tissues overlapping across all datasets. c, Cell types distribution across all datasets.Principal component analysis (PCA) based on 12,000 most expressed genes across allsamples colored by the tissue of its origin (e,g) or cell type (d,f). UMAP representation of cellscolored by the tissue of its origin (i) or its type (h). LN – lymph node.
Figure 2. Tabuls Muris Senis single cell RNAseq dataset. a, Dataset preprocessingresulting in myeloid subset derivation. UMAP plot with natural clusters (b) and cell types (c)identified based on cell specific markers (d). NP – neutrophil, Mo – monocyte, prog –progenitor, DC – dendritic cell, MF – macrophage, alv MF – alveolar macrophage, MG –microglia, KC – Kupffer cell, pDC – plasmacytoid dendritic cell, migDC – migratory dendriticcell.
Figure 3. Scheme of analysis approach for multisample metabolic network clustering(GAM-clustering). The dataset’s metabolic genes are initially clustered based on a k-medoidsalgorithm. Averaged gene expression of the obtained clusters is further considered as patterns.For each gene, a score is calculated on the basis of its correlation with each pattern. Thesescores are superimposed on the KEGG metabolic network. Based on these scores, the mostweighted connected subnetwork is found for each parent. After the refinement procedure,metabolic modules as a final version of subnetworks are obtained.
Figure 4. Metabolic modules as a result of multisample metabolic network clustering ofall myeloid cells but not inflammatory conditions from ImmGen MNP OS dataset. a,Heatmap representing samples hierarchically clustered based on averaged gene expression ofeach of obtained module (from lowest as blue to highest as red). Euclidean distance is used asa clustering metric. YS MF – yolk sac macrophage, EB MF – embryoid body macrophage,alvMF – alveolar macrophage, SPM – small peritoneal macrophage, MG – microglia, MF –macrophage, Mo – monocyte, DC – dendritic cell, pDC – plasmacytoid DC, migDC – migratoryDC. b, Annotation of the obtained modules based on gene enrichment in KEGG and Reactomecanonical pathways. Enrichment value is calculated as a percentage of module genescontained in a particular pathway. c, Radar chart representation of metabolic modules withineach metasample. Each individual sample is shown as a grey line while mean of all samplesinside one metasample is shown as a colored line. Nine radii of radar chart are devoted to thecorresponding metabolic modules: 1,2 – Lipid metabolism, 3 – FAS pathway, 4 – mtFASIIpathway, 5 – Cholesterol synthesis, 6 – Glycolysis, 7 – Folate, serine & nucleotide metabolism,8 – FAO & sphingolipid de novo synthesis, 9 – Glycerophospholipid metabolism. Metasamplesof EB MFs + alvMFs and alvMFs + SPMs cells are shown at one chart as they are extremelyclose in their metabolic characteristics.
Figure 5. Cell types shared between ImmGen Mononuclear Phagocytes Open Source (IGMNP OS), ImmGen Mononuclear Phagocytes Phase 1 (IG MNP P1) and myeloid TabulaMuris Senis (mTMS) datasets have similar patters of metabolic modules signatures. a,Population memberships across the datasets: prog – progenitor, SC – stem cell, MLP –multilineage progenitor, MF YS – yolk sac macrophage, MF EB – embryoid body macrophage,MF – macrophage, alvMF – alveolar macrophage, SPM – small peritoneal macrophage, MG –
21
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
microglia, KC – Kupffer cell, Mo – monocyte, pDC – plasmacytoid dendritic cell, DC – dendriticcell, migDC – migratory dendritic cell, NP – neutrophil (Supplementary Table 4). b,Enrichment of individual metabolic modules across all datasets obtained during GAM-clustering analysis of IG MNP OS dataset.
Figure 6. Subnetworks associated with early developmental stages and dendritic cells.Heatmaps of module patterns along with the expression of some of its genes or genes relatedto the same biological subject (from lowest as blue to highest as red). YS MF – yolk sacmacrophage, EB MF – embryoid body macrophage, alvMF – alveolar macrophage, SPM –small peritoneal macrophage, MG – microglia, MF – macrophage, Mo – monocyte, DC –dendritic cell, pDC – plasmacytoid DC, migDC – migratory DC (a,e). Metabolic modules per sewhere edges of modules are attributed with color according to correlation of its enzyme’s geneexpression to this particular module pattern and thickness according to its score (b,c,f). d,Enrichment of modules genes expression (from lowest as blue to highest as red, transparentdots correspond to treated samples) across all three analysed datasets: ImmGen MononuclearPhagocytes Open Source (IG MNP OS), ImmGen Mononuclear Phagocytes Phase 1 (IG MNPP1) and myeloid Tabula Muris Senis (mTMS) datasets. g, Dendritic cell migrations experimentscheme and results.
Figure 7. Subnetworks associated with fatty acid synthesis and degradation. Heatmapsof module patterns along with the expression of some of its genes (from lowest as blue tohighest as red). YS MF – yolk sac macrophage, EB MF – embryoid body macrophage, alvMF– alveolar macrophage, SPM – small peritoneal macrophage, MG – microglia, MF –macrophage, Mo – monocyte, DC – dendritic cell, pDC – plasmacytoid DC, migDC – migratoryDC (a,e). Metabolic modules per se and corresponding schematic diagrams. Edges ofmodules are attributed with color according to correlation of its enzyme’s gene expression tothis particular module pattern and with thickness according to its score. (b,c,f). Enrichment ofmodules genes expression (from lowest as blue to highest as red, transparent dots correspondto treated samples) across all three analysed datasets: ImmGen Mononuclear PhagocytesOpen Source (IG MNP OS), ImmGen Mononuclear Phagocytes Phase 1 (IG MNP P1) andmyeloid Tabula Muris Senis (mTMS) datasets (d,g).
22
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
●●●●●● ●●●
●●
●●●●
●●●
●
●●●●
●●
● ●
●
●●●●
●●●●●●●●
●
●●●●●●●●●
●●
●●●
●●
● ●
●●●
●●
●●
●
●●
●
●●●
●●●
●●●●
●●
●●●●●●● ●
●●
●● ●
●
●●●
●●
●
●
●●●
●●●●
●●●●
●●●
●●
●●●●●● ●●●
●●●●
●
●●●
●●●●
●
●●●
●●
●●
●●●●●●
●●
●●
●●
●●●●
●●●●
●●
●
●
●●
●●
● ●
●●●●●●●
●●●●
●●●●●●
●
●
●
●●●●●●
●
●
●●
●
●
●
●
●●●●
●
●
●
●
●
●
●
●●●
●
●●
●● ●
●● ●
●● ●●●●
●●
●●
●●●●●
●
●●●●●●●●●
●●
●●
●
●
●●●
●
● ●
●●●
●●
● ●●
●
●
●●●●●
●●●
●●●
●●
●●
●●●
●●
●
●●
●●●●
●●●●● ●●
●●
●●●
●●
●●
●●●
●
●
●●●
●●
●●
−100
−50
0
50
100
−100 −50 0 50 100PC1 (17.9%)
PC2
(12.
6%) ●
●
●
●
●
●
Dendritic ell
Yolk sac macrophage
Macrophage
Embryoid body macrophage
Microglia
Monocyte
●●● ●●● ●●●
●●
●●●●
●●●
●
●●●●
●●
● ●
●
●●●●
●●●●●●●●
●
●●●●●●●●●
●●
●●●
●●
● ●
●●●
●●
●●
●
●●
●
●●●
●●●
●●●●
●●
●●●●●●● ●
●●
●● ●
●
●●●
●●
●
●
●●●
●●●●
●●●●
●●●
●●
●●●●●● ●●●
●●●●
●
●●●
●●●●
●
●●●
●●
●●
●●●●●●
●●
●●
●●
●●●●
●●●●
●●
●
●
●●
●●
● ●
●●●●●●●
●●●●
●●●●●●
●
●
●
●●●●●●
●
●
●●
●
●
●
●
●●●●
●
●
●
●
●
●
●
●●●
●●●
●● ●
●● ●
●● ●●●●
●●
●●
●●●●●
●
●●●●●●●●●
●●
●●
●
●
●●●
●
● ●
●●●
●●
● ●●
●
●
●●●●●
●●●
●●●
●●
●●
●●●
●●
●
●●
●●●●
●●●●● ●●
●●
●●●
●●
●●
●●●
●
●
●●●
●●
●●
−100
−50
0
50
100
−100 −50 0 50 100PC1 (17.9%)
PC2
(12.
6%)
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Aorta
Blood
Bone
Brain
Colon
Dermis
Dorsal root ganglia
Epididymal fat
Epithelium
Heart
Inguinal fat
Kidney
Liver
Lung
Mediastinal LN
Mesenteric fat
Mesenteric LN
Mesenteric
Peritoneal
Sciatic nerve
Spinal
Spleen
Thymus
Yolk sac
A B
Cell
Tissue
Figure 1
Embryoid body
marrow
cord
cavity
sheet
Peyer's patch
ImmGen OS
mTMS
IG MNP P1
C
E
Cell
Tissue
IG MNP OS
IG MNP P1
Tissue overlapping across all datasets
IG MNP P1,samples
mTMS,cells
Cell types distribution across all datasets
●
●● Stem cell
●
Multilineage progenitor
●
●
Dendritic ell
Macrophage
Microglia
Monocyte
MNP OS
IG MNP OS
IG MNP OS,samples
G
ImTMS
●
●
●
●
●
Dendritic ell
Progenitors
Macrophage
Microglia
Monocyte
Cell
IG MNP OS
IG MNP P1
TissuemTMS
● Neutrophils
D
F
H
IG MNP OS
IG MNP P1
mTMS
gland
muscle
and aorta
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
Figure 2
C
ADoubletsremoval
Senis
235 325
0 - NP1 - Mo2 - Mo3 - prog4 - DC5 - MF6 - lung Mo7 - alvMF
8 - MF9 - MG10 - KC11 - prog NP12 - pDC13 - NP14 - KC
B Natural clusters:
min
max
Cell types:
60 844 51 364Myeloidsignature
013
1
2
31112
7
9
51410
4
6
8NP
MG
pDC
alvMF
lung MoMo
prog NP
KC
migDC
MF
prog
DC BM MF
adipose MF
D
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
...
...
no
yes
final metabolic modules:
is patterncontent
changed?
...
pattern scoring
superimposing weights
on metabolic network
finding correlated
subnetworks
...
updating gene content
of each pattern
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
12
12.2
12.4
12.6
11.9
12.1
12.3
12.5
12.7
Pattern1
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
11.6
11.8
12
12.2
12.4
11.5
11.7
11.9
12.1
12.3
Pattern2
Figure 3
Suclg1
Gfpt1
Gm ps
Ogdh
Aldh1b1
Ogdh
Got2
Tkt
Taldo1
Adss
Fasn
Acot7
Fasn
Fasn
Pdhb
Acadl
El ovl5
Rpe
Pfkp
Gfpt1
Aldoc
Gpd2
Mdh2
Me2
Pdhb
Pdhb
Ass1
Got2
Ogdh
Cs
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
Aldh9a1
Cpt2
Echs1
Mecr
Hsd17b10
Acaa2
Hmgcl
Ggct
Gss
Glo1
Glo1
Akr1b3
Gpx4
Nfs1
Gpt
Ethe1
Mgst3
Fahd1
Fh1
Sdhb
Acaa2
Acaa2
Acat1
Dlat
Hmgcl
Fahd1
Eci1
Echs1
Hsd17b10
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
Gbgt1
B3galnt1
H exdc
H exa
Glb1
Glb1
Neu1
H exdc
Gba
Acy1
Tha1
Alas1
Lap3
Sardh
Tha1
Aldh7a1
Pon2
Pon3
Ctps2
Ggh
Ggt5
Oplah
Gpt2
Txnrd2
Txndc12
Gstm5
Ltc4s
Adh fe1
Ggt5
Txnrd2
Pc yox1
Cbr2
Fam213b
Aoah
A bhd12
Lpl
Asah2
Pla2g4a
Plcb3Plcd1
Dgkz
Lpcat3
Pla2g15
Pld3
Pla2g6
Cer k
Sm pd1
Cer s2
Npl
Renbp
Neu1
Tbxas1
Gaa
Gaa
B4galt1
Glb1
Sord
Ocrl
Ltc4s
Al ox5
Ptgs1
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met met
met
met
met
met
met
met
met
met
met
met
Cyp51
Lss
Mvd
Fdps
Idi1
Dhcr24
Msmo1
Tm7sf2
Fdf t1
Pmvk
Sqle
Hmgcs1
Aacs Acat2
Nsdhl
Hsd17b7
Mvk
Hmgcr
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
Pgam1
Pgk1
Gapdh
Dera
Pdha1
Ldha
Pkm
Tpi1
Eno1
Tkt
Gpi1
Pgm2
Aldoa
Acp1
Aldoa
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
Aldh18a1
Cad
Cad
Epr s
Ppat
Ctps
Tyms
Dhfr
Fpgs
Psat1
Cad
Cad
Psph
Ppat
Pr ps1l3
Idh3a
Nudt5
Gar t
Shmt2
Gar t
Adsl
Paics
Atic
Dhodh
Dar s
Um ps
Shmt2
Mthfd1
Mthfd1
Enoph1
Adi1
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
met
Cpt1a
Akr1a1
Kdsr
Sptlc2
Hadha
Lipa
Asah1
Ac ox1
Aldh3b1
Asah1
met
met
met
met
met
met
met
met
met
met
met
Cds1
Pn pla7
Lpcat2
Cept1
Dgkd
met
met
met
met
met
met
Table with scores
samples
avera
ged
gene e
xpre
ssio
n
avera
ged
gene e
xpre
ssio
n
samples
Cluster1Cluster2
ClusterM
MDH
MALt3pp
D_LACt2pp
G1PPpp
ACONMT
ACALDtex
ACONIs
PPC
F6Ptex
MALS
FBA3
ETOHtrpp
FUMt2_2pp 22
GLCP
CS
GLGC
GLYCtpp
PGI
GLYCtex
GLYOX
PGK
FUMtex
ETOHtex
H2tpp
PGMT
G6PDH2r
ACS
GLYK
AKGt2rpp
MALt2_2pp 22
PGL
12PPDRtpp
ACKr
GLDBRAN2
RPI
XYLt2pp
G1Ptex
MALtex
FRUptspp
L_LACD2
PYK
L_LACt2rpp
GLYCDx
MALt2_3pp3
3
DHAPT
GLYC3Pabcpp
PPCK
L_LACtex
PDH
ME2
ACALDtpp
FRUtex
XYLabcpp
FRD3
ACt2rpp
ACONTb
CITtex
RIBabcpp
FUM
EDA
PFL
GLCP2
LDH_D
LCARR
PGM
RPE
FRD2
GLYC3Ptex
PPS
2
GLBRAN2
GLCptspp
SUCCt2_2pp
22
SUCCt2_3pp 33
FBP
TPI
12PPDRtex
ICLICDHyrD_LACtex
MDH3
FORtex
ENO
G3PT
GLCt2pp
RBK
PTAr
GAPD
CITt3pp
SUCOAS
H2tex
ACONTa
PFK
LALDO2x
ACtex
PFK_3MGSA
F6Pt6_2pp2
2
CITt7pp
F6PA
MDH2
GLYC3Pt6pp
PYRt2rpp
TKT1
SUCCt3pp
ALCD2x
SUCCtex
ALDD2x2
G3PD2
XYLI1
CITL
FORt2pp
GLCtex_copy1
PYRtex
SUCDi
FORtppi
TALA
TKT2
XYLtex
AKGDH
AKGtex
LGTHL
ALDD2y
2
EDD
FRUK
MOX
ME1
XYLK
G6Pt6_2pp
2
2
GND
HEX1
FUMt2_3pp3
3
FBA
G6PP
RIBtexFRUpts2pp
ACALD
FHL
GLCS1
G6Ptex
OAADC
LDH_D2
GLCptspp
L_LACD3
mthgxl_c
mal__L_c
nad_c
nadh_c
h_c
oaa_c
h_p
h_c
mal__L_p
lac__D_p
h_p
lac__D_c
h_c
g1p_p
h2o_p pi_p
glc__D_p
acon_T_camet_c
ahcys_c
aconm_c
acald_e
acald_p
acon_C_c
pep_c
h2o_c
co2_c
h_c
pi_c
f6p_e f6p_p
h2o_c
accoa_c
glx_c
h_ccoa_c
s17bp_c
e4p_c
dhap_c
etoh_p
etoh_c
h_p
fum_p
h_c
fum_c
pi_cglycogen_c
g1p_c
accoa_ch2o_c
cit_c
h_c
coa_c
atp_c
h_cppi_c
adpglc_c
glyc_c
glyc_p
g6p_c
f6p_c
glyc_e
h2o_c
lgt__S_c
h_c
gthrd_c
atp_c
3pg_c
adp_c
13dpg_c
fum_e
etoh_e
h2_p
h2_c
nadp_cnadph_c
6pgl_c
h_c
ac_c
atp_c
coa_c
amp_c
ppi_c
atp_c
h_c
glyc3p_c
adp_c
h_p
akg_p
h_c
akg_c
h_ph_c
h2o_c
6pgc_c
h_c
12ppd__R_p
12ppd__R_c
atp_c
adp_c
actp_c
bglycogen_c
r5p_c
ru5p__D_c
h_p
xyl__D_p
xyl__D_c
h_c
g1p_e
mal__L_e
pep_c
fru_p
f1p_cpyr_c
lac__L_c
q8_c q8h2_c
adp_c
h_c
atp_c
h_p
lac__L_p
h_c
nad_c
dha_c
nadh_c
h_c
h_ph_c
pep_c
pyr_c
h2o_c
glyc3p_patp_c
h_c
pi_c
adp_c
atp_c
co2_c
adp_c
lac__L_e
nad_c
coa_c
nadh_c
co2_c
nadp_c
co2_c
nadph_c
acald_c
fru_e
h2o_c
atp_c
adp_c
pi_c
h_c
2dmmql8_c
succ_c
2dmmq8_c
h_p
ac_p
h_c
h2o_c icit_c
cit_ecit_p
rib__D_p
h2o_c
atp_c
pi_c
adp_c
rib__D_c
h_c
h2o_c
2ddg6p_c
g3p_c
coa_c for_c
pi_c
nad_c
h_c
nadh_c
h_cnadh_c
lald__D_c
nad_c
2pg_c
xu5p__D_c
mql8_c mqn8_c
glyc3p_e
atp_c
h2o_c
amp_c
pi_c
h_c
pep_c
pyr_c
succ_ph_ph_c
h_ph_c
fdp_c
h2o_c
pi_c
12ppd__R_e
nadp_c
co2_c
nadph_c
lac__D_e
mqn8_c
mql8_c
for_e
for_p
h2o_c
h2o_c
pi_c
h_p
h_c
glc__D_c
atp_c
adp_c
h_c
pi_c
coa_c
pi_c
nad_c
h_c
nadh_c
h_p
h_c
coa_c
atp_cpi_c
succoa_c
adp_c
h2_e
atp_c
adp_c
h_c
nadh_c
h_c
nad_c
ac_e
atp_c
s7p_c
adp_c
h_c
pi_c
pi_c
pi_p
succ_c
succ_p
q8_c
q8h2_c
pi_cpi_p
h_p
pyr_p
h_c
h_p
h_c
nad_ch_cnadh_c
succ_e
nad_c
h2o_c
nadh_c
h_c
nadp_c
h_c
nadph_c
xylu__D_c
ac_c
h_p
h_c
glc__D_e
pyr_e
q8_cq8h2_c
xyl__D_e
nad_c
coa_c
nadh_c
co2_c
akg_e
nadp_c
h2o_cnadph_c
h_c
h2o_c
atp_ch_c
adp_c
o2_c
h2o2_c
nad_c
nadh_c
co2_c
atp_ch_c
adp_c
pi_c
g6p_p
pi_p
nadp_c
nadph_c
co2_c
atp_c
adp_c
h_c
h_ph_c
h2o_c
pi_c
rib__D_e
pep_c
pyr_c
nad_c
coa_c
h_c
nadh_c
h_c
co2_c
h_c
adp_c
g6p_e
h_cco2_c
q8h2_cq8_c
glc__D_p
g6p_c
mqn8_c
pyr_c
mql8_c
pattern production
I. Initialization of patterns
III. Output
II. Iterative subnetwork identification
medoids
technical fine resolution clusters
bulk RNAseq data:
meta
bolic
genes
coexpre
ssio
n c
lust
eri
ng
samples (bulk RNAseq) / technical clusters (single cell RNAseq)
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
A
Figure 4
cell type
tissue
metaSample
B CModule 1Module 2Module 3Module 4Module 5Module 6Module 7Module 8Module 9
0% 50% 100%
migDCEB MF+alvMF+SPMYS MF MG tissue DCpDCMoadipose MFtissue MF
0 0.5 1
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
Figure 5
Module 1:Lipid metabolism - 1
Module 2:Lipid metabolism - 2
Module 3:FAS pathway
Module 4:mtFASII pathway
Module 5:Cholesterol metabolism
Module 7:Folate, serine & nucleotide metabolism
Module 6:Glycolysis
Module 8:FAO & sphingolipid de novo synthesis
Module 9:Glycerophospholipid metabolism
pro
gSC
MLP
MF
YSM
F EB
MF
BM
MF
alvM
FSPM adip
ose
MF
MG
KC Mo
lung M
o pD
CD
Cm
igD
Cpro
g N
PN
P
IG MNP OS
IG MNP P1
mTMS
A
B
IG MNP OS
IG MNP P1
mTMS
IG MNP OS
IG MNP P1
mTMS
IG MNP OS
IG MNP P1
mTMS
IG MNP OS
IG MNP P1
mTMS
IG MNP OS
IG MNP P1
mTMS
IG MNP OS
IG MNP P1
mTMS
IG MNP OS
IG MNP P1
mTMS
IG MNP OS
IG MNP P1
mTMS
IG MNP OS
IG MNP P1
mTMS
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
Module 5
Module 5 - Cholesterol metabolism
score
0 1
-1.5 2.5
corr
Figure 6
E
B D
cell type
tissue
metaSample
HmgcrMvdPmvkSqleCcr7
Hsd17b7
Idi1
Aacs
Nsdhl
Msmo1
Hmgcr
Hmgcs1
Pmvk Mvk
Dhcr24
Acat2
Cyp51
Tm7sf2
Fdps
Mvd
Sqle
Fdft1
Lss
4alpha-Methylzymosterol
Dimethylallyldiphosphate
2-Oxobutanoate
4alpha-Methylzymosterol-4-carboxylate
3-Keto-4-methylzymosterol
(S)-3-Hydroxy-3-methylglutaryl-CoA
(R)-5-Phosphomevalonate
(R)-Mevalonate
4,4-Dimethyl-5alpha-cholesta-8-en-3beta-ol
Acetyl-CoA
Acetoacetyl-CoA
4,4-Dimethyl-5alpha-cholesta-8,14,24-trien-3beta-ol
Isopentenyldiphosphate
Squalene
14-Demethyllanosterol
(R)-5-Diphosphomevalonate
Lanosterol
(S)-2,3-Epoxysqualene
trans,trans-Farnesyldiphosphate
0 0.5 1
Module 6
Module 7
Module 6 - Glycolysis
Module 7 - Folate, serine & nucleotide metabolism
score
0 1
-1.5 2.5
corr
Ppat
Cad
Ctps
EprsAldh18a1
Fpgs
Psat1
Tyms
Dhfr
Cad
CadCad
Idh3a
Gart
Gart
Nudt5
Psph
Shmt2
Ppat
Prps1l3
Umps
Dars
Adsl
Atic
Paics Dhodh
Adi1
Mthfd1
Mthfd1
Shmt2
Enoph1
L-Glutamine
L-Glutamyl-tRNA(Glu)
L-Glutamate
Dihydrofolate
L-Glutamyl5-phosphate
Carbamoylphosphate
N-Carbamoyl-L-aspartate
Isocitrate
5'-Phosphoribosylglycinamide
ADP-ribose
L-Serine
O-Phospho-L-serine
5-Phosphoribosylamine
Glycine
5-Phospho-alpha-D-ribose1-diphosphate
2-Oxoglutarate
D-Ribose5-phosphate
Orotidine5'-phosphate
L-Aspartyl-tRNA(Asp)
1-(5'-Phosphoribosyl)-5-amino-4-imidazolecarboxamide
1-(5'-Phosphoribosyl)-5-amino-4-(N-succinocarboxamide)-imidazole
(S)-DihydroorotateL-Aspartate
Orotate
1-(5'-Phosphoribosyl)-5-formamido-4-imidazolecarboxamide
Formate
Tetrahydrofolate
1,2-Dihydroxy-5-(methylthio)pent-1-en-3-one
5,10-Methylenetetrahydrofolate
10-Formyltetrahydrofolate
2,3-Diketo-5-methylthiopentyl-1-phosphate Ser/Gly
Nucleicacids
Folate
A
B C
cell type
tissue
metaSample
Idh3aNudt5Mthfd1CadFpgs
AldoaLdhaPkmEno1Gapdh
Aldoa
Tkt
Acp1
Pgm2
Gpi1
Ldha
Aldoa
Dera
Tpi1
Eno1
Pgam1
Gapdh
Pgk1
Pdha1
Pkm
D-Erythrose4-phosphate
Glycerone
D-Glucose6-phosphate
(S)-Lactate
Sedoheptulose1,7-bisphosphate
2-Deoxy-D-ribose5-phosphate
D-Glucose1-phosphate
Glyceronephosphate
D-Fructose6-phosphate
2-Phospho-D-glycerate
3-Phospho-D-glyceroylphosphate
D-Glyceraldehyde3-phosphate
Acetyl-CoA
Phosphoenolpyruvate
3-Phospho-D-glycerate
Pyruvate
0 0.5 1
F G
D
●●●●●● ●●●
●●
●●●●
●●●●
●●●●
●●
●●
●
●●●●
●●●●●●●●
●
●●●
●●
●●●
●●●
●
●●●
●●
●
●●●
●●●●
●●●
●●
●●●●●●●
●
●● ●● ●
●●●●
●●
●
●
●●●
●●●●
●●●●
●●●
●●
●●
●●
● ●● ●
●●●●●●
●
●
●
●●
●
●
●
●●●
●●●
●●●●●●
●● ●●●●
●●
●●
●●●●●●●●●
●●●●●
●●
●●
●●
●●●
●
● ●●
●●●
●● ●
●
●
●
●●●●●●●●●●
●
●●
●●●
●●
●●
●●
●●●●●●●●●● ●●
●●
●●●
●●
●●
●●●
●
●
●●●
●●
●●
●●
●●●●
●●
● ●●
●●●
●●●●
●●●●●●●●●●
●●
●
●●●
●●●●●●
●
●●
●●●
●●
●●●
●●
●●●●●●
●●●
●●●
●●●●●●
●●●
●
●●
●●●
●●●
●●●●●●
●●
●●●
●●●
●
● ●●
●●●
●●
●● ●
●●●
●● ●●●●
●● ●
●●
●●●
●●
●●
● ●●
●●●
●●
●●
●●●
●
●●●●●●●
●●
●●
●●●
●●
●●
●●●●●
●
●
●
●
●●
●●●
●●●●●
●●
●●●
●●
●●
●
●
●●●●
●●●
●●●
●●●●●● ●●●
●●
●●●●
●●●●
●●●●
●●
●●
●
●●●●
●●●●●●●●
●
●●●
●●
●●●
●●●
●
●●●
●●
●
●●●
●●●●
●●●
●●
●●●●●●●
●
●● ●● ●
●●●●
●●
●
●
●●●
●●●●
●●●●
●●●
●●
●●
●●
● ●● ●
●●●●●●
●
●
●
●●
●
●
●
●●●
●●●
●●●●●●
●● ●●●●
●●
●●
●●●●●●●●●
●●●●●
●●
●●
●●
●●●
●
● ●●
●●●
●● ●
●
●
●
●●●●●●●●●●
●
●●
●●●
●●
●●
●●
●●●●●●●●●● ●●
●●
●●●
●●
●●
●●●
●
●
●●●
●●
●●
●●
●●●●
●●
● ●●
●●●
●●●●
●●●●●●●●●●
●●
●
●●●
●●●●●●
●
●●
●●●
●●
●●●
●●
●●●●●●
●●●
●●●
●●●●●●
●●●
●
●●
●●●
●●●
●●●●●●
●●
●●●
●●●
●
● ●●
●●●
●●
●● ●
●●●
●● ●●●●
●● ●
●●
●●●
●●
●●
● ●●
●●●
●●
●●
●●●
●
●●●●●●●
●●
●●
●●●
●●
●●
●●●●●
●
●
●
●
●●
●●●
●●●●●
●●
●●●
●●
●●
●
●
●●●●
●●●
●●●
●●●●●● ●●●
●●
●●●●
●●●●
●●●●
●●
●●
●
●●●●
●●●●●●●●
●
●●●
●●
●●●
●●●
●
●●●
●●
●
●●●
●●●●
●●●
●●
●●●●●●●
●
●● ●● ●
●●●●
●●
●
●
●●●
●●●●
●●●●
●●●
●●
●●
●●
● ●● ●
●●●●●●
●
●
●
●●
●
●
●
●●●
●●●
●●●●●●
●● ●●●●
●●
●●
●●●●●●●●●
●●●●●
●●
●●
●●
●●●
●
● ●●
●●●
●● ●
●
●
●
●●●●●●●●●●
●
●●
●●●
●●
●●
●●
●●●●●●●●●● ●●
●●
●●●
●●
●●
●●●
●
●
●●●
●●
●●
●●
●●●●
●●
● ●●
●●●
●●●●
●●●●●●●●●●
●●
●
●●●
●●●●●●
●
●●
●●●
●●
●●●
●●
●●●●●●
●●●
●●●
●●●●●●
●●●
●
●●
●●●
●●●
●●●●●●
●●
●●●
●●●
●
● ●●
●●●
●●
●● ●
●●●
●● ●●●●
●● ●
●●
●●●
●●
●●
● ●●
●●●
●●
●●
●●●
●
●●●●●●●
●●
●●
●●●
●●
●●
●●●●●
●
●
●
●
●●
●●●
●●●●●
●●
●●●
●●
●●
●
●
●●●●
●●●
●●●
PC2
(12.
6%)
PC1 (21.9%)
PC2
(17.
2%)
UMAP_1
UM
AP_2
PC2
(12.
6%)
PC1 (21.9%)
PC2
(17.
2%)
UMAP_1
UM
AP_2
PC2
(12.
6%)
PC1 (21.9%)
PC2
(17.
2%)
UMAP_1
UM
AP_2
IG MNP OS IG MNP P1 mTMS
0 0.5 1 0 0.5 1
migDCEB MF+alvMF+SPMYS MF MG tissue DCpDCMoadipose MFtissue MF
migDCEB MF+alvMF+SPMYS MF MG tissue DCpDCMoadipose MFtissue MF
PC1 (17.9%)
PC1 (17.9%)
PC1 (17.9%)
Module
5M
od
ule
7M
od
ule
6
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint
Module 1Module 2
Module 9
Module 1&2 - Lipid metabolism
Module 9 - Glycerophospholipid metabolism
PG
LT
GlcNAc
GSH
Membr. lipids
ARA
score
0 1
-1.5 2.5
corr
Figure 7
A
B C
cell type
tissue
metaSample
HexaPtgs1SordAlox5Gstm5
Cept1Lpcat2Pnpla7DgkdCds1
Dgkz
Pld3
Adhfe1
Gpt2 Pla2g15Pla2g4aPla2g6
Lpcat3
Ltc4sAlox5OplahCtps2
Ggt5Ggh
Plcb3
Smpd1
Gba
CerkAsah2
Lpl
Abhd12Aoah
Ltc4sGgt5 Gstm5Ptgs1
Renbp
Npl
Tbxas1
Neu1Neu1
Hexdc
Hexa Glb1Glb1
B3galnt1
Cers2
Gbgt1Hexdc
Pon3
Txnrd2
Txnrd2
Fam213b
Pcyox1
Txndc12
Cbr2
Pon2
Glb1
B4galt1
Sord
Gaa
Gaa
Tha1
Tha1
SardhLap3
Alas1
Aldh7a1
Acy1
Ocrl
Plcd1
Phosphatidate2-Oxoglutaratesn-Glycero-3-phosphocholine
PhosphatidylcholineLeukotrieneA4
Pidolicacid
L-Glutamine
L-Glutamate
1-Phosphatidyl-D-myo-inositol
N-Acylsphingosine
1-Acyl-sn-glycero-3-phosphocholine
TriacylglycerolCeramide
1-phosphate
Fatty acid
3-(Acyloxy)acylgroup ofbacterial
toxinLeukotriene
C4R-S-Glutathione
THF-polyglutamate Arachidonate
2-Acylglycerol
N-Acetyl-D-mannosamine
ThromboxaneA2
N-Acetylneuraminate
N-Acetyl-D-galactosaminyl-(N-acetylneuraminyl)-...
N-Acetyl-D-glucosamine
GA2
beta-D-Galactosyl-(1-4)-beta-D-glucosyl-(1-1)-ceramide
alpha-D-Galactosyl-(1-4)-beta-D-galactosyl-(1-4)-...
beta-D-Glucosyl-(1-1)-ceramide
Sphingomyelin
Globoside
N-Acetyl-D-galactosamine
4-Hydroxyphenylacetate
S-Glutathionyl-L-cysteine
ProstaglandinH2
Farnesylcysteine
Glutathione
ProstaglandinE2
ProstaglandinF2alpha
Phenylacetate
L-Cysteine
Glutathionedisulfide
Lactose
Sphingosyl-phosphocholine
D-Sorbitol
Sucrose
N-Acetyl-D-galactosaminyl-N-Acetyl-D-galactosaminyl-...
D-Glucose
D-FructoseD-Galactose
L-Allothreonine
Glycine
Acetaldehyde
R-S-Cysteinylglycine
Sarcosine
S-SubstitutedN-acetyl-L-cysteine
5-Aminolevulinate
Acetate
(R)-2-Hydroxyglutarate
1-Phosphatidyl-1D-myo-inositol4-phosphate
1,2-Diacyl-sn-glycerol
1-Phosphatidyl-D-myo-inositol4,5-bisphosphate
Lpcat2Pnpla7 Cds1DgkdCept11-Acyl-sn-glycero-3-phosphocholine
Fatty acid
CDP-diacylglycerol
Phosphatidate
1,2-Diacyl-sn-glycerol
Phosphatidylcholine
●●●●●● ●●●
●●
●●●●
●●●●
●●●●
●●
●●
●
●●●●
●●●●●●●●
●
●●●
●●
●●●
●●●
●
●●●
●●
●
●●●
●●●●
●●●
●●
●●●●●●●
●
●● ●● ●
●●●●
●●
●
●
●●●
●●●●
●●●●
●●●
●●
●●
●●
● ●● ●
●●●●●●
●
●
●
●●
●
●
●
●●●
●●●
●●●●●●
●● ●●●●
●●
●●
●●●●●●●●●
●●●●●
●●
●●
●●
●●●
●
● ●●
●●●
●● ●
●
●
●
●●●●●●●●●●
●
●●
●●●
●●
●●
●●
●●●●●●●●●● ●●
●●
●●●
●●
●●
●●●
●
●
●●●
●●
●●
PC1 (17.9%)PC
2 (1
2.6%
)
●●
●●●●
●●
● ●●
●●●
●●●●
●●●●●●●●●●
●●
●
●●●
●●●●●●
●
●●
●●●
●●
●●●
●●
●●●●●●
●●●
●●●
●●●●●●
●●●
●
●●
●●●
●●●
●●●●●●
●●
●●●
●●●
●
● ●●
●●●
●●
●● ●
●●●
●● ●●●●
●● ●
●●
●●●
●●
●●
● ●●
●●●
●●
●●
●●●
●
●●●●●●●
●●
●●
●●●
●●
●●
●●●●●
●
●
●
●
●●
●●●
●●●●●
●●
●●●
●●
●●
●
●
●●●●
●●●
●●●
PC1 (21.9%)
PC2
(17.
2%)
UMAP_1
UM
AP_2
●●●●●● ●●●
●●
●●●●
●●●●
●●●●
●●
●●
●
●●●●
●●●●●●●●
●
●●●
●●
●●●
●●●
●
●●●
●●
●
●●●
●●●●
●●●
●●
●●●●●●●
●
●● ●● ●
●●●●
●●
●
●
●●●
●●●●
●●●●
●●●
●●
●●
●●
● ●● ●
●●●●●●
●
●
●
●●
●
●
●
●●●
●●●
●●●●●●
●● ●●●●
●●
●●
●●●●●●●●●
●●●●●
●●
●●
●●
●●●
●
● ●●
●●●
●● ●
●
●
●
●●●●●●●●●●
●
●●
●●●
●●
●●
●●
●●●●●●●●●● ●●
●●
●●●
●●
●●
●●●
●
●
●●●
●●
●●
●●
●●●●
●●
● ●●
●●●
●●●●
●●●●●●●●●●
●●
●
●●●
●●●●●●
●
●●
●●●
●●
●●●
●●
●●●●●●
●●●
●●●
●●●●●●
●●●
●
●●
●●●
●●●
●●●●●●
●●
●●●
●●●
●
● ●●
●●●
●●
●● ●
●●●
●● ●●●●
●● ●
●●
●●●
●●
●●
● ●●
●●●
●●
●●
●●●
●
●●●●●●●
●●
●●
●●●
●●
●●
●●●●●
●
●
●
●
●●
●●●
●●●●●
●●
●●●
●●
●●
●
●
●●●●
●●●
●●●
●●●●●● ●●●
●●
●●●●
●●●●
●●●●
●●
●●
●
●●●●
●●●●●●●●
●
●●●
●●
●●●
●●●
●
●●●
●●
●
●●●
●●●●
●●●
●●
●●●●●●●
●
●● ●● ●
●●●●
●●
●
●
●●●
●●●●
●●●●
●●●
●●
●●
●●
● ●● ●
●●●●●●
●
●
●
●●
●
●
●
●●●
●●●
●●●●●●
●● ●●●●
●●
●●
●●●●●●●●●
●●●●●
●●
●●
●●
●●●
●
● ●●
●●●
●● ●
●
●
●
●●●●●●●●●●
●
●●
●●●
●●
●●
●●
●●●●●●●●●● ●●
●●
●●●
●●
●●
●●●
●
●
●●●
●●
●●
●●
●●●●
●●
● ●●
●●●
●●●●
●●●●●●●●●●
●●
●
●●●
●●●●●●
●
●●
●●●
●●
●●●
●●
●●●●●●
●●●
●●●
●●●●●●
●●●
●
●●
●●●
●●●
●●●●●●
●●
●●●
●●●
●
● ●●
●●●
●●
●● ●
●●●
●● ●●●●
●● ●
●●
●●●
●●
●●
● ●●
●●●
●●
●●
●●●
●
●●●●●●●
●●
●●
●●●
●●
●●
●●●●●
●
●
●
●
●●
●●●
●●●●●
●●
●●●
●●
●●
●
●
●●●●
●●●
●●●
migDCEB MF+alvMF+SPMYS MF
0 0.5 1
D
Module 8 - FAO & sphingolipid de novo synthesis
Module 8
score
0 1
-1.5 2.5
corrSphL
FAO FAdegradation
E
F G
cell type
tissue
metaSample
Asah1LipaHadhaAcox1KdsrCpt1aCpt2Sptlc1Sptlc2
Sptlc2 Kdsr
Akr1a1
Cpt1aAcox1
HadhaAsah1
Lipa
Asah1
Aldh3b1
3-Dehydrosphinganine
Primaryalcohol
L-Palmitoylcarnitinetrans-Hexadec-2-enoyl-CoA
Sphinganine
Sterylester
Dihydroceramide
Palmitoyl-CoA
Fatty acid(S)-3-Hydroxyhexadecanoyl-CoA
Aldehyde
0 0.5 1
●●●●●● ●●●
●●
●●●●
●●●●
●●●●
●●
●●
●
●●●●
●●●●●●●●
●
●●●
●●
●●●
●●●
●
●●●
●●
●
●●●
●●●●
●●●
●●
●●●●●●●
●
●● ●● ●
●●●●
●●
●
●
●●●
●●●●
●●●●
●●●
●●
●●
●●
● ●● ●
●●●●●●
●
●
●
●●
●
●
●
●●●
●●●
●●●●●●
●● ●●●●
●●
●●
●●●●●●●●●
●●●●●
●●
●●
●●
●●●
●
● ●●
●●●
●● ●
●
●
●
●●●●●●●●●●
●
●●
●●●
●●
●●
●●
●●●●●●●●●● ●●
●●
●●●
●●
●●
●●●
●
●
●●●
●●
●●
●●
●●●●
●●
● ●●
●●●
●●●●
●●●●●●●●●●
●●
●
●●●
●●●●●●
●
●●
●●●
●●
●●●
●●
●●●●●●
●●●
●●●
●●●●●●
●●●
●
●●
●●●
●●●
●●●●●●
●●
●●●
●●●
●
● ●●
●●●
●●
●● ●
●●●
●● ●●●●
●● ●
●●
●●●
●●
●●
● ●●
●●●
●●
●●
●●●
●
●●●●●●●
●●
●●
●●●
●●
●●
●●●●●
●
●
●
●
●●
●●●
●●●●●
●●
●●●
●●
●●
●
●
●●●●
●●●
●●●
0 0.5 1 0 0.5 1
MG
PC1 (17.9%)
PC2
(12.
6%)
PC1 (21.9%)PC
2 (1
7.2%
)UMAP_1
UM
AP_2
PC1 (17.9%)
PC2
(12.
6%)
PC1 (21.9%)
PC2
(17.
2%)
UMAP_1
UM
AP_2
IG MNP OS IG MNP P1 mTMS
PC1 (17.9%)
PC2
(12.
6%)
PC1 (21.9%)
PC2
(17.
2%)
UMAP_1
UM
AP_2
IG MNP OS IG MNP P1 mTMS
0 0.5 1 0 0.5 1
tissue DCpDCMoadipose MFtissue MF
migDCEB MF+alvMF+SPMYS MF MG tissue DCpDCMoadipose MFtissue MF
.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted July 16, 2020. . https://doi.org/10.1101/2020.07.15.204388doi: bioRxiv preprint