Reuse of public, genome-wide, murine eosinophil expression data for hypotheses development

Abstract

The eosinophil (Eos) surface phenotype and activation state is altered after recruitment into tissues and after exposure to pro-inflammatory cytokines. In addition, distinct Eos functional subsets have been described, suggesting that tissue-specific responses for Eos contribute to organ homeostasis. Understanding the mechanisms by which Eos subsets achieve their tissue-specific identity is currently an unmet goal for the eosinophil research community. Publicly archived expression data can be used to answer original questions, test and generate new hypotheses, and serve as a launching point for experimental design. With these goals in mind, we investigated the effect of genetic background, culture methods, and tissue residency on murine Eos gene expression using publicly available, genome-wide expression data. Eos differentiated from cultures have a gene expression profile that is distinct from that of native homeostatic Eos; thus, researchers can repurpose published expression data to aid in selecting the appropriate culture method to study their gene of interest. In addition, we identified Eos lung- and gastrointestinal-specific transcriptomes, highlighting the profound effect of local tissue environment on gene expression in a terminally differentiated granulocyte even at homeostasis. Expanding the “toolbox” of Eos researchers to include public-data reuse can reduce redundancy, increase research efficiency, and lead to new biological insights.

Publicly archived eosinophil genome-wide expression data can be used to answer original questions, generate new hypotheses, and serve as a launching point for experimental design.

1 INTRODUCTION

Eosinophils (Eos) are terminally differentiated granulocytes that accumulate at mucosal surfaces, including the gastrointestinal tract (GI), lungs, and uterus, to contribute to tissue homeostasis.1,2 In addition, Eos have long been associated with exacerbation of allergic and other inflammatory disorders.3 It is well established that the Eos surface phenotype is altered during disease,4–7 presumably as a consequence of exposure to activating cytokines. Indeed, IL-5 stimulation results in differential surface phenotype and gene expression for Eos.8–10 In addition, tissue-resident Eos at homeostasis have been shown to have a distinct phenotype,11–13 suggesting that the local environment induces changes in Eos phenotype for tissue-specific functions. Understanding the mechanisms by which Eos subsets achieve their tissue-specific functions and whether these functions can be modulated for therapeutic purposes are identified goals for the eosinophil research community.

Advancements in high-throughput, next-generation sequencing have yielded large amounts of genome-wide gene expression data that are collected in pubic archives, including the Gene Expression Omnibus (GEO).14 Importantly, publicly archived data can be reused to answer questions beyond those posed in the initial study that generated the data and to serve as a launching point for the design of future experiments.15,16 For example, gene expression data from multiple published studies was reused to delineate peripheral blood signatures for respiratory viral infections,17,18 highlighting how the reuse of public data can lead to new biological insight and be used to test and generate new hypotheses. Publicly archived expression data can complement the traditional research model, hypothesis generation followed by experiments designed to test the hypothesis, via the analysis of deposited data to refine (or generate) hypotheses. To demonstrate how this could be used in Eos-focused research, we searched the GEO public archive for available Eos genome-wide RNA sequencing (RNA-seq) data and reused the data to investigate the effect of genetic background, tissue residency, and culture methods on homeostatic Eos gene expression.

2 MATERIALS AND METHODS

2.1 Public archive search

We searched GEO for deposited Eos genome-wide gene expression data.14 At the time of our analysis, almost 2000 GEO DataSets were associated with the key words “eosinophil” or “eosinophils”, but only 6 DataSets contained murine Eos RNA-seq data. Seven samples within these DataSets19,20 were data derived from native Eos purified from tissues at homeostasis from wild-type mice (Table 1). In addition, we identified RNA-seq data from murine Eos differentiated from unselected (GSE55385) whole bone marrow (WBM) cells and low-density (GSE43660) bone marrow (LDBM) cells in culture.21,22

2.2 Mice

BALB/c and C57BL6/J wild-type mice were analyzed at 4 to 8 weeks of age. All mice were housed under specific pathogen-free conditions and handled under approved protocols (#2E09072) of the Institutional Animal Care and Use Committee of Cincinnati Children’s Hospital Medical Center.

2.3 Eosinophil isolation

Native Eos were sorted, as previously reported,19 on a FACSAria II (BD Biosciences, San Jose, CA) maintained by the Research Flow Cytometry Core at Cincinnati Children’s Hospital Medical Center.

2.4 Culture-differentiated eosinophils

For confirmatory expression experiments, Eos were differentiated from unselected WBM or LDBM cells in culture as reported.23,24

2.5 Gene expression analysis

For the native Eos from pooled bone marrow of C57BL6/J mice, total RNA and RNA-seq libraries were prepared as previously described.19 The native C57BL6/J eosinophil RNA-seq data was deposited in GEO database (GSE110299). Sequencing files for Eos RNA from GSE55385, GSE43660, GSE56292, GSE69707, and GSE106213 were downloaded from the GEO database.14 RNA-seq analysis was performed using BioWardrobe,25 as previously described.19 A minimum reads per kilobase per million mapped reads (RPKM) equivalent to 2 was deemed the lower limit of expression. Differentially expressed genes with DESeq.26,Padj < 0.05 were analyzed. For gene set enrichment analysis (GSEA), ranked gene lists were created from DESeq output by removing all genes that did not reach 5 RPKM expression in at least one condition and sorting the remaining genes by DESeq-calculated log fold change. GSEA v.3.0 was run on the pre-ranked list against C5bp (Gene ontology biological process) v6.1 gene set collection with default parameters.27,28 Enrichment plots for gene sets that are overrepresented in the ranked gene lists were generated by GSEA.27 Briefly, enrichment scores are calculated by going through the list of genes ranked by fold change in expression and increasing the cumulative score when a gene is included in a specific set and decreasing when a gene is not included in the set. Enrichment plots provided by GSEA are a graphical view of the enrichment scores, which reflect the degree to which a gene set is overrepresented in the differentially expressed genes. Gene sets with distinct peaks at the beginning of the plot are overrepresented in induced genes, where as those with vallies at the end are overrepresented among silenced genes. Both of these are likely to be interesting to the investigator, as they represent greater enrichment for genes in that specific biological pathway. For confirmatory expression experiments, cDNA was synthesized from pooled total RNA from sorted native Eos (n ≥ 10 mice per group, 2–3 independent sorts) and from total RNA from culture-differentiated Eos (cEos) (n ≥ 9 mice per group, n = 2–3 independent experiments) using Superscript VILO cDNA synthesis kit (Thermo Fisher Scientific, Waltham, MA). Quantitative PCR (qPCR) was performed using PowerUp SYBR Green Master Mix (Thermo Fisher Scientific) and specific primers for Cd300ld (TCAGCCACATTCCCACTCAT [forward], CTGCCTCCTGAGTGTTGAGA [reverse]) and Mpo (CTTTCCTGCCCTTCCCAGTA [forward], CGTGCCATATTGTGCCATCA [reverse]). Data were analyzed for fold expression over Gapdh (2^−dCt) and groups compared using a two-tailed, unpaired t-test (GraphPad Prism). Differences were considered statistically significant when P < 0.05.

3 RESULTS

3.1 Genetic background has minimal effect

Strain-specific genetic variation can result in differential binding of transcription factors to regulatory elements and strain-specific gene expression patterns29; thus, we compared gene expression between native Eos sorted from the bone marrow of BALB/c (2015) and C57BL6/J (GSE110299) to assess the effect of genetic background on gene expression at homeostasis (Table 1). There were 7201 genes that were expressed (RPKM > 2) by either BALB/c or C57BL6/J bone marrow Eos (Fig. 1A, Supplementary Table 1). Approximately 10% (717/7201) of the expressed genes had expression levels that were significantly different (Padj ˂ 0.05), and the differential expression was 2-fold or more between BALB/c and C57BL6/J Eos (Fig. 1A). Expression of a member of the CD300 family of molecules, CLM-5 (Cd300ld), was significantly different between native BALB/c and C57BL6/J Eos from the bone marrow, and we confirmed differential expression with independent samples via qPCR (Fig. 1B). Overall, there was a strong correlation (r = 0.81, P < 0.0001) for gene expression between native BALB/c and C57BL6/J Eos, highlighting the small effect of genetic background variation on homeostatic gene expression in Eos from the bone marrow (Fig. 1A).

3.2 Culture-differentiated Eos differ from native Eos

Robust and highly reproducible Eos differentiation systems have been developed that start with unselected23 or low-density24 murine bone marrow cells. Genome-wide gene expression data were available from cEos from unselected bone marrow (WBM, C57BL6/J, GSE55385, 2015) and from cEos from LDBM cells (BALB/c, 2013). We compared gene expression in cEos-LDBM with that of native Eos sorted from BALB/c bone marrow (Fig. 2A) and in cEos-WBM with that of native Eos sorted from C57BL6/J bone marrow (Fig. 2B). Expression levels for 43% (4163) of the 9716 genes that were expressed (RPKM > 2) by either the cEos-LDBM or the native Eos (BALB/c) were significantly different (Padj < 0.05, Fig. 2A, Supplementary Table 2). In comparison, expression levels were significantly different for 69% (6556/9427) of the genes expressed in the cEos-WBM compared to the native Eos (C57BL6/J) with 62% (4097/6556) of the differentially expressed genes higher in the cEos-WBM (Fig. 2B, Supplementary Table 2). Native Eos expressed higher levels of Mpo and Elane mRNAs than did cEos (Fig. 2C). In contrast, cEos expressed markedly higher levels of mRNAs for the granule proteins Epx (eosinophil peroxidase) and Prg2 (major basic protein) than did native Eos (Fig. 2D). We confirmed the differential expression of Mpo between native Eos and cEos with independent samples (Fig. 2E). Notably, gene expression in native Eos was more comparable to that of cEos-LDBM (r = 0.73, Fig. 2A) than cEos-WBM (r = 0.56, Fig. 2B). In addition, we compared expression for genes known to be associated with Eos in inflammatory environments and noted higher expression in the cEos than native Eos, except for Il13ra1 (Table 2).

Gene ontology analysis of the genes of the differentially expressed genes between cEos-WBM and native Eos showed that this gene set was enriched for pathways associated with protein synthesis and localization (Fig. 3A), with higher expression in the cEos-WBM than native Eos. Notably, translation of select mRNAs is induced in mature human Eos in response to cytokine stimulation30,31; thus, differentiation in the presence of IL-5 may enhance protein translation in murine cEos. In contrast, the differentially expressed gene set between cEos-LDBM and native Eos was enriched for genes associated with sterol synthesis and metabolism (Fig. 3B), emphasizing the transcriptomic differences not only between the cEos and native Eos, but also between the method used to produce the cEos.

3.3 Tissue residency alters Eos transcriptome

As Eos are primarily tissue-residing leukocytes,3,32 we next evaluated the effect of tissue residency on Eos gene expression at homeostasis. We first compared Eos sorted from murine lungs (2014) to Eos sorted from bone marrow (GSE110299) at homeostasis (Fig. 4A). Of the genes expressed (RPKM > 2) either by the lung- or bone marrow–resident Eos (BM Eos, Supplementary Table 4), approximately 57% (4793/8438) were differentially expressed (Padj < 0.05) with higher expression in the lung Eos for 62% (2979/4793) of the differentially expressed genes (Fig. 4A). We noted higher expression of Cd69, Cd9, and Il4 in the lung Eos than in the BM Eos (Fig. 4B), highlighting the phenotypic changes to Eos that occur in response to the local lung environment. Notably, 32% (2451/7591) of the genes expressed by the lung Eos were not expressed by the BM Eos (Fig. 4C), and 19% (1424/7591) of the expressed genes were uniquely expressed by the lung Eos (Fig. 4C). Overall, there was a modest correlation (r = 0.52) for gene expression between lung Eos and BM Eos at homeostasis (Fig. 4A). Gene ontology analysis of the differentially expressed genes between lung Eos and BM Eos revealed an enrichment for genes associated with IFN-γ production, innate immune responses, and chemotaxis in the lung Eos compared to BM Eos (Fig. 4D).

Eos isolated from the colons of murine GIs at homeostasis (GI Eos) and BM Eos expressed fewer genes (5628 and 5579, respectively) than did lung Eos (7591, Fig. 4C). Of the total number of genes expressed by either BM Eos or GI Eos, 61% (4761/7745) were differentially expressed, with 52% of those (2463/4761) having higher expression in the GI Eos (Fig. 5A, Supplementary Table 5). Expression of several chemokines, including Cxcl2, Ccl9, and Ccl6, were higher in the GI Eos than BM Eos (Fig. 5B). In contrast, mRNA for Il4 was lower in GI Eos than BM Eos (Fig. 5B). Notably, 1838 genes were expressed by GI Eos and not BM Eos, and 811 genes were uniquely expressed by GI Eos (Fig. 4C). Overall, the transcriptome of Eos isolated from murine colons at homeostasis was significantly different (r = 0.18) than the transcriptome of Eos isolated from the bone marrow (GI Eos versus BM Eos, Fig. 5A). Gene ontology analysis of the differentially expressed genes revealed enrichment for genes associated with chemotaxis, response to vitamins and response to IL-1 in the GI Eos compared to BM Eos (Fig. 5C).

There was little correlation between the transcriptomes of the tissue-resident Eos (Fig. 5D, lung Eos vs. GI Eos, r 5 = 0.38). Gene ontology analysis of the differentially expressed genes revealed enrichment for genes associated with cytokine production in lung Eos compared to GI Eos (Fig. 5E). Notably, over 1000 genes were expressed in both the tissue-resident Eos that were not expressed by the BM Eos (Fig. 4C), suggesting a common transcriptional program that includes Il4ra, Cd69, Cd14, and Il10ra for Eos when they enter a mucosal environment.

We compared expression of genes recently associated with regulatory Eos found in the lung at homeostasis11 (Table 3) and noted higher expression of Nedd4, Runx3, Ldlr, and Sell in lung Eos than in BM Eos, highlighting consistency of expression data for lung Eos at homeostasis across studies.11,20 Surface expression of CD62L discriminated between regulatory and inflammatory Eos in the lung.11 Lung Eos did have modestly higher expression of mRNA for L-selectin (CD62L, encoded by Sell) than BM or GI Eos (Table 3), suggesting that transcriptional regulation of Sell expression is likely one of several mechanisms to control CD62L surface expression on Eos.

4 DISCUSSION

Publicly available genome-wide gene expression data can be used for hypothesis development and increase research efficiency.15,16 We searched the public archive GEO for deposited RNA-seq data from Eos at homeostasis14 and identified 7 samples for our analysis (3 from GI [C57BL6/J], 2 from lung [C57BL6/J], and 2 from BM [BALB/c]). We added 2 samples from Eos sorted from C57BL6/J BM for the study (GSE110299). Comparatively, murine neutrophils and basophils have 38 and 4 data sets, respectively, that contain RNA-seq data.14 RNA-seq profiling of purified human peripheral blood Eos left untreated or treated with TGF-β has been completed (n = one sample of each), and supplementary data for the published study include RPKM values.33 No human Eos RNA-seq raw data are otherwise currently publicly available. Thus, Eos are underrepresented in the public archives of genome-wide gene expression. As biologist-friendly analysis tools (e.g., Biowardrobe) continue to become available, it will be important for Eos researchers to continue to add expression data to the public archives.

Immune responses vary between inbred laboratory mouse strains due to genetic variation (i.e., polymorphisms).34 This is highlighted by susceptibility differences between strains to experimental food allergy models.35,36 We noted greater expression of Cd300ld, a CD300 family member, in C57BL6/J than BALB/c Eos. The CD300 family of receptors belongs to the immunoglobulin receptor super family37 and have been shown to regulate Eos functional responses.38,39 Currently, there is not much knowledge regarding the expression or function of Cd300ld in Eos. Eos are known to express various CD300 family members including CD300a and CD300f.40 Interestingly, biochemical analyses have shown that CD300ld likely acts as the paired co-activating receptor of CD300f. Differential expression of innate receptors, such as Cd300ld, may contribute to the varied immune responses between strains. At homeostasis, there was strong correlation between the transcriptomes of Eos sorted from the BM of C57BL/6 and BALB/c mice, with ∼10% genes differentially expressed (Fig. 1).

Since its initial description in 2008,23 a liquid culture system that starts with unselected WBM to generate Eos ex vivo has been cited in almost 100 publications, highlighting the importance of this tool for researchers. We described an alternate culture system that starts with LDBM cells for studies focused on development, as the mature cells are depleted via density centrifugation.24,41 Comparing the Eos differentiated in the cultures (cEos) to native Eos sorted from the bone marrow revealed differential expression of a large subset of genes, likely due to the IL-5 needed to expand and differentiate the bone marrow cells to Eos ex vivo. Interestingly, the pathways enriched in the cEos are not the same between the two culture methods. We did not directly compare the two populations of cEos due to the difference in genetic background with the available expression data, so it is unclear whether the culture method–specific transcriptomes are due to differences in starting population or strain-specific differences in response to IL-5. Pathways associated with protein synthesis and localization were enriched in the gene sets upregulated in the cEos derived from the unselected bone marrow (GSE55385, Fig. 3A). This is consistent with higher expression of granule proteins in the cEos than native Eos. The differentially expressed genes in cEos that result from LDBM cells (GSE43660) showed enrichment for pathways associated with sterol synthesis (Fig. 3B), perhaps suggesting increased lipid body formation in response to IL-5.42 In addition, cEos contained higher mRNA expression for genes known to be associated with Eos in inflammatory environments (Table 2). Together, these data suggest that the cEos may represent a unique Eos subset that is produced under IL-5–mediated pressure. Similarly, neutrophil development is differentially regulated depending on the level of G-CSF expression,43 resulting in alternate neutrophil phenotypes. Ideally, expression data comparing cEos and native Eos with matching genetic backgrounds for multiple strains would be publicly available for researchers in the future to compare expression of their gene-of-interest to aid in selecting an appropriate culture system for their experiments.

Prior studies have noted differential gene expression between Eos subsets,11–13 but key questions remain unanswered regarding the Eos role at homeostasis at mucosal surfaces. Understanding the mechanisms by which Eos subsets achieve their tissue-specific functions remains an unmet need. Similar to studies with tissue-specific macrophage identities,44,45 differential expression patterns between tissue-resident Eos subsets and the BM-resident Eos will aid in identifying pathways critical for tissue-specific functional responses. We noted significant differential gene expression between GI-, lung-, and BM-resident Eos at homeostasis. The transcriptome divergence between the tissue-resident Eos suggests a greater transcriptional responsiveness than expected for the terminally differentiated cells.

Importantly, variation in the methods for RNA-seq library construction, including variable levels of DNA contamination and quality of the published datasets, likely contributes to some of the differential expression detected; yet, the pathways enriched make biologic sense, such as chemotaxis pathways in tissue-resident Eos, highlighting the relevancy of the analysis. With data reuse analysis, new research avenues can be developed that focus on the environmental cues, as well as the transcription factors, that are responsible for the tissue-specific transcriptomes. Though the human Eos proteome correlates generally with mRNA abundance,46 there are notable exceptions30; thus, independent confirmation of gene and protein expression is critical as new hypotheses are developed on the basis of genome-wide data reuse. It remains to be determined whether the responses of specific Eos subsets, such as the recently identified regulatory Eos subset residing in the lung,11 can be modulated for therapeutic purposes. Understanding the mechanisms by which the various Eos subsets are developed is an important next step. Expanding the “toolbox” of Eos researchers to include public-data reuse is especially attractive due to the inherent challenges associated with working with this relatively rare and terminally differentiated leukocyte.

ACKNOWLEDGMENTS

This work was supported by the NIH grant R01 AI130033 (P.C.F.) and DP2 GM11913 (A.B.). This project was also supported in part by the NIH grant P30 DK078392 (Gene and Protein Expression and Flow Cytometry Cores) of the Digestive Disease Research Core Center in Cincinnati. We thank Shawna Hottinger for editorial assistance.

AUTHORSHIP

J.O.G, A.Mk., and H.R. performed experiments. J.O.G., A.M., H.R., A.Mz., A.B., and P.C.F. analyzed data. J.O.G., A.Mk., A.B., and P.C.F. wrote the manuscript.

DISCLOSURES

P.C. Fulkerson has received grants from the National Institutes of Health, has served as a consultant for Genentech, Inc., and has received research funding from Knopp Biosciences, LLC. The remaining authors declare no relevant conflicts of interest.

Abbreviations

 
• DESeq

  differential expression sequence

 
• Eos

  eosinophils

 
• GEO

  Gene Expression Omnibus

 
• GSEA

  gene set enrichment analysis

 
• LDBM

  low-density bone marrow

 
• qPCR

  quantitative PCR

 
• RPKM

  reads per kilobase million mapped reads

 
• WBM

  whole bone marrow

REFERENCES