open source immgen: network perspective on metabolic ... · 15/07/2020 · open source immgen:...

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

mailto:[email protected]

mailto:[email protected]

https://doi.org/10.1101/2020.07.15.204388

http://creativecommons.org/licenses/by-nc-nd/4.0/

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


https://doi.org/10.1101/2020.07.15.204388


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


https://doi.org/10.1101/2020.07.15.204388


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


https://artyomovlab.wustl.edu/immgen-met/

https://doi.org/10.1101/2020.07.15.204388


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


https://doi.org/10.1101/2020.07.15.204388


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


http://artyomovlab.wustl.edu/immgen-met/

https://doi.org/10.1101/2020.07.15.204388


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


https://doi.org/10.1101/2020.07.15.204388


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


https://doi.org/10.1101/2020.07.15.204388


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


https://doi.org/10.1101/2020.07.15.204388


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


https://doi.org/10.1101/2020.07.15.204388


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


https://doi.org/10.1101/2020.07.15.204388


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


https://doi.org/10.1101/2020.07.15.204388


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


https://doi.org/10.1101/2020.07.15.204388


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


https://figshare.com/projects/Tabula_Muris_Senis/64982

https://github.com/ctlab/sgmwcs-solver

https://github.com/ctlab/sgmwcs-solver

https://doi.org/10.1101/2020.07.15.204388


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


https://github.com/artyomovlab/ImmGenOpenSource

https://doi.org/10.1101/2020.07.15.204388


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


https://artyomovlab.wustl.edu/immgen-met/

https://doi.org/10.1101/2020.07.15.204388


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


https://doi.org/10.1101/2020.07.15.204388


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


https://doi.org/10.1101/2020.07.15.204388


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


https://doi.org/10.1101/2020.07.15.204388


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


https://doi.org/10.1101/2020.07.15.204388


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


https://doi.org/10.1101/2020.07.15.204388


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


https://doi.org/10.1101/2020.07.15.204388


●●●●●● ●●●

●●

●●●●

●●●

●

●●●●

●●

● ●

●

●●●●

●●●●●●●●

●

●●●●●●●●●

●●

●●●

●●

● ●

●●●

●●

●●

●

●●

●

●●●

●●●

●●●●

●●

●●●●●●● ●

●●

●● ●

●

●●●

●●

●

●

●●●

●●●●

●●●●

●●●

●●

●●●●●● ●●●

●●●●

●

●●●

●●●●

●

●●●

●●

●●

●●●●●●

●●

●●

●●

●●●●

●●●●

●●

●

●

●●

●●

● ●

●●●●●●●

●●●●

●●●●●●

●

●

●

●●●●●●

●

●

●●

●

●

●

●

●●●●

●

●

●

●

●

●

●

●●●

●

●●

●● ●

●● ●

●● ●●●●

●●

●●

●●●●●

●

●●●●●●●●●

●●

●●

●

●

●●●

●

● ●

●●●

●●

● ●●

●

●

●●●●●

●●●

●●●

●●

●●

●●●

●●

●

●●

●●●●

●●●●● ●●

●●

●●●

●●

●●

●●●

●

●

●●●

●●

●●

−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


https://doi.org/10.1101/2020.07.15.204388


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


https://doi.org/10.1101/2020.07.15.204388


...

...

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)


https://doi.org/10.1101/2020.07.15.204388


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


https://doi.org/10.1101/2020.07.15.204388


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


https://doi.org/10.1101/2020.07.15.204388


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



PC1 (17.9%)

PC1 (17.9%)

PC1 (17.9%)

Module

5M

od

ule

7M

od

ule

6


https://doi.org/10.1101/2020.07.15.204388


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


PC1 (17.9%)

PC2

(12.

6%)

PC1 (21.9%)

PC2

(17.

2%)

UMAP_1

UM

AP_2


0 0.5 1 0 0.5 1

tissue DCpDCMoadipose MFtissue MF



https://doi.org/10.1101/2020.07.15.204388


open source immgen: network perspective on metabolic ... · 15/07/2020 · open source immgen:...

Documents