Electronic Sorting of Immune Cell Subpopulations Based on Highly Plastic

This information is current as Pingzhang Wang, Wenling Han and Dalong Ma of October 2, 2021. J Immunol 2016; 197:665-673; Prepublished online 10 June 2016; doi: 10.4049/jimmunol.1502552 http://www.jimmunol.org/content/197/2/665 Downloaded from

Supplementary http://www.jimmunol.org/content/suppl/2016/06/10/jimmunol.150255 Material 2.DCSupplemental

References This article cites 56 articles, 18 of which you can access for free at: http://www.jimmunol.org/ http://www.jimmunol.org/content/197/2/665.full#ref-list-1

Why The JI? Submit online.

• Rapid Reviews! 30 days* from submission to initial decision

• No Triage! Every submission reviewed by practicing scientists by guest on October 2, 2021 • Fast Publication! 4 weeks from acceptance to publication

*average

Subscription Information about subscribing to The Journal of Immunology is online at: http://jimmunol.org/subscription Permissions Submit copyright permission requests at: http://www.aai.org/About/Publications/JI/copyright.html Email Alerts Receive free email-alerts when new articles cite this article. Sign up at: http://jimmunol.org/alerts

The Journal of Immunology is published twice each month by The American Association of Immunologists, Inc., 1451 Rockville Pike, Suite 650, Rockville, MD 20852 Copyright © 2016 by The American Association of Immunologists, Inc. All rights reserved. Print ISSN: 0022-1767 Online ISSN: 1550-6606. The Journal of Immunology

Electronic Sorting of Immune Cell Subpopulations Based on Highly Plastic Genes

Pingzhang Wang, Wenling Han, and Dalong Ma

Immune cells are highly heterogeneous and plastic with regard to expression and cell phenotype. In this study, we categorized genes into those with low and high gene plasticity, and those categories revealed different functions and applications. We proposed that highly plastic genes could be suited for the labeling of immune cell subpopulations; thus, novel immune cell subpopulations could be identified by gene plasticity analysis. For this purpose, we systematically analyzed highly plastic genes in human and mouse immune cells. In total, 1,379 human and 883 mouse genes were identified as being extremely plastic. We also expanded our previous immunoinformatic method, electronic sorting, which surveys big data to perform virtual analysis. This approach used correlation analysis and took dosage changes into account, which allowed us to identify the differentially expressed genes. A test with human CD4+ T cells supported the method’s feasibility, effectiveness, and predictability. For example, with the use of human non- regulatory T cells, we found that FOXP3hiCD4+ T cells were highly expressive of certain known molecules, such as CD25 and Downloaded from CTLA4, and that this process of investigation did not require isolating or inducing these immune cells in vitro. Therefore, the sorting process helped us to discover the potential signature genes or marker molecules and to conduct functional evaluations for immune cell subpopulations. Finally, in human CD4+ T cells, 747 potential immune cell subpopulations and their candidate signature genes were identified, which provides a useful resource for big data–driven knowledge discoveries. The Journal of Immunology, 2016, 197: 665–673. http://www.jimmunol.org/ any phenotypes and immune cell subpopulations have that growing numbers of subpopulations will be discovered. One been identified, either from developmental commit- important question is whether these potential immune cell sub- ments or simply as representatives of certain func- populations and their functions can be predicted. Generally, for M + tional states (1–7). For example, CD4 T cells can be classified wet experiments, immune cell subsets must first be isolated by into distinct subsets based on their signature cytokines (8, 9), such FACS or be induced through in vitro culture and then used in gene as Th1, Th2, Th9, Th17, Th22, regulatory T cells (Tregs), and expression analyses or functional tests. As yet, no immunoinformatic follicular helper T (Tfh) cells. Several studies also revealed that method, using the virtual sorting of immune cells, was reported to different immune cell subsets are capable of conversion (e.g., the function with immune cell subpopulations. transitions from Th17 or Th2 to Th1 or from Tregs or Tfh to Th17), However, the rapid accumulation of publicly available high- by guest on October 2, 2021 which suggests phenotypic flexibility and plasticity (10, 11). throughput omics data, such as data from microarrays and next- Immune cell subpopulations are generally denoted by a set of generation sequencing, allows the direct generation of “big signature genes or marker molecules, especially cytokines and cell data.” These big data imply that a great deal of knowledge will be surface markers. More than 35,000 potential -coding genes discovered through data mining. Previously, we used immuno- exist in the (12), and .3,700 genes potentially logically relevant transcriptome data from microarrays to analyze encode cell surface transmembrane (13). Because gene gene coexpression (15), and we evaluated marker molecules based expression is highly plastic, and because each gene has its char- on gene plasticity (GPL) (14). GPL is the degree of change in the acteristic gene expression profile under various experiments, expression of a gene that happens in response to various envi- treatments, or pathophysiological conditions (14), we can expect ronmental or genetic influences. The GPL score is a quantitative measurement that can be used for the evaluation of biomarkers

Department of Immunology, School of Basic Medical Sciences, Peking University (14). Highly plastic genes often have large GPL scores; thus, they Health Science Center, Beijing 100191, China; Peking University Center for Human are not suitable as lineage-specific biomarkers of immune cells. Disease Genomics, Beijing 100191, China; and Key Laboratory of Medical Immu- In this study, we evaluated immune cell subpopulations based on nology, Ministry of Health, Beijing 100191, China GPL analysis. We expanded the meaning of electronic sorting, which Received for publication December 14, 2015. Accepted for publication May 17, 2016. previously described the retrieval of cell states or experimental conditions based on gene expression intensity from miscellaneous This work was supported by Grant 31270948 from the National Natural Science Foundation of China and by Leading Academic Discipline Project of Beijing Edu- immune cell samples (14). As a method to “virtually sort big data cation Bureau (BMU20110254). samples to do analysis,” we could also call this process “virtual Address correspondence and reprint requests to Associate Prof. Pingzhang Wang, sorting” or “in silico sorting,” because no actual immune cells were Peking University Health Science Center, 38 Xueyuan Road, Beijing 100191, China. sorted as they are in the FACS process. First, we systematically E-mail address: [email protected] analyzed highly plastic genes in human and mouse immune cells and The online version of this article contains supplemental material. found that the genes with high plasticity were well suited for labeling Abbreviations used in this article: APA rate, absent-present and present-absent rate; + ARS, average rank score; DEG, differentially expressed gene; GO, ; immune cell subpopulations. Then, we used human CD4 T cells as a GPL, gene plasticity; KEGG, Kyoto Encyclopedia of Genes and Genomes; PP, model to perform electronic sorting. Our results revealed the feasi- present-present; RNA-seq, RNA-sequencing; Tfh, follicular helper T; Treg, regulatory bility, effectiveness, and predictability of electronic sorting as a Tcell. means to discover the potential signature genes or marker molecules, Copyright Ó 2016 by The American Association of Immunologists, Inc. 0022-1767/16/$30.00 and this method was also suitable for other immune cells. www.jimmunol.org/cgi/doi/10.4049/jimmunol.1502552 666 ELECTRONIC SORTING OF IMMUNE CELLS

Materials and Methods samples were first sorted based on rank score, and three rounds of elec- Data sets tronic sorting were performed based on the quantile values (Q1, Q2, and Q3, respectively) (Fig. 1). During each round, the samples were divided All microarray data and immune cell groups were described in a previous into two groups (or virtual subpopulations). The difference (i.e., d value) in study (14). The 619 samples (Dataset 1) of human CD4+ T cells in the the average rank score (ARS) for each gene at the probe set level in both ImmuSort database (http://immusort.bjmu.edu.cn or http://immu.tipsci. subpopulations was calculated using the ARS in the high-expression com/Account/) were used. To improve reliability by cross-validation, six subpopulation subtracted from that in the low-expression subpopulation. additional data sets were prepared. Dataset 2 (816 samples) was a newly Thus, each round produced a gene list of 41,477 probe sets with unique updated version of Dataset 1, which contained all available human CD4+ annotation. After three rounds of sorting, three gene lists (List1, List2, and T cell samples included in the Gene Expression Omnibus database as of List3) were produced and used for further identification of differentially December 20, 2015 (16). Dataset 3 (361 samples) and Dataset 4 (470 expressed genes (DEGs). samples) were derived from healthy individuals in Datasets 1 and 2, re- For upregulated (or downregulated) candidate DEGs, genes with deltas spectively. Dataset 5 (346 samples) contained all samples derived from greater (or less) than 10 (or 210) in any two of the primary gene lists patients and several samples without disease information in Dataset 2, (List1, List2, and List3) were identified. These candidates were further including patients with various cancers, systemic lupus erythematosus, filtered by a threshold of the Pearson correlation coefficient (r $ +0.5) for rheumatoid arthritis, and asthma. Dataset 6 (474 samples) and Dataset 7 upregulated DEGs and by a mutual repulsion rate $ 0.7 for downregulated (658 samples) were taken from Datasets 1 and 2, respectively, but the DEGs. Datasets 2 through 7 were also analyzed for cross-validation. The samples from patients with hematologic system diseases and lymphomas final DEGS were obtained when they were supported by at least four were removed. Several T cell clones, lymph node samples, and trans- datasets during cross-validation and used for further enrichment analysis. formed cell lines identified during reanalysis of the sample sources were also discarded from Datasets 6 and 7. All samples are listed in Supplemental Correlation analysis Table I. The expressional correlation analysis and present-present (PP) rate cal-

Enrichment analysis culation were described previously by using the ImmuCo data set (15). The Downloaded from Pearson’s r was calculated by using the signal values in human CD4+ Enrichment analysis is a high-throughput strategy that increases the like- T cells. lihood that investigators can identify the biological processes most pertinent to their study. Enrichment can be quantitatively measured with many Plots and statistics statistical methods (17) to determine whether certain biological processes, R (http://www.r-project.org/) language and the software environment were such as gene annotations from gene ontology (GO) (18), or pathways, such used for statistical computing and graphical displays of all data. The as Kyoto Encyclopedia of Genes and Genomes (KEGG) (19), are enriched network analysis was performed using Cytoscape software (http://www. http://www.jimmunol.org/ in the relevant gene sets identified in the user’s study. Functional anno- cytoscape.org/) (21). tations concerning the GO terms and KEGG pathways were downloaded from the DAVID database (version 6.7) (20). The Fisher exact test was adopted. The p values were adjusted using the R function “p.adjust” with Results the Bonferroni correction (“bonferroni”), in which the p values were Identification of highly plastic genes multiplied by the number of comparisons. Annotation terms with Bonferroni-corrected p values # 0.05 were considered statistically sig- High plasticity indicates that a gene easily changes its expression nificant and were used for further analysis. under various experimental conditions (14). In this study, we performed a systematic analysis of GPL scores and found that the Mutual repulsion rate average scores of all genes in human and mouse immune cells For our array platform, a probe set was flagged absent if the relevant transcript were 15.07 and 15.37, respectively. We identified genes with GPL by guest on October 2, 2021 was not considered to be expressed. Otherwise, the probe set was flagged as scores $ 45 (three times greater than average) as extremely plastic present for expression or as marginally present if the expression could not be determined. The mutual repulsion rate or absent-present and present-absent genes and those with GPL scores # 5 (one third of the average rate (APA rate) meant that one of the genes in any two-gene combination GPL score) as extremely low plasticicity genes. In total, 1379 (gene pair) at the probe set level was expressed in one sample (present), but human and 883 mouse genes were identified to be extremely plastic the other gene was not expressed in the same sample (absent), or vice versa (Supplemental Table II). Moreover, 10,889 and 7,655 extremely (absent-present). The APA rate was calculated by dividing the sample counts of the absent-present and the present-absent states by the total sample count. low plasticicity genes were identified (Supplemental Table II). These two gene sets were used for functional enrichment analysis. Electronic sorting analysis As shown in Table I and Supplemental Table III, the molecular During electronic sorting, the expressional intensity of the genes was re- functions relevant to cytokine activity, chemokine activity, or flected by the rank score (14). For each highly plastic gene (label gene), the chemokine receptor binding were highly enriched in the highly

FIGURE 1. The electronic sorting process used in this analysis. See Materials and Methods for details. The Journal of Immunology 667 plastic gene set. The biological processes relevant to the immune genes with low plasticity on the basis of our previous study system, immune response, response to external stimulus, defense (Supplemental Fig. 3) (14). Similar results were observed for response, and inflammatory response were also highly enriched. A mouse samples (Table I, Supplemental Table III). Therefore, our further analysis revealed that cytokines, including chemokines, results suggested that genes with high plasticity play more im- constituted an important gene set in the highly plastic genes. This portant roles in immune responses than do low plasticity genes. finding was confirmed by KEGG pathway analysis (Table I, Genes with high plasticity indicated the presence of certain Supplemental Table III). Moreover, our results also showed that conditions that generated the high and low (or no) expression states highly plastic genes were mainly derived from moderately of these genes. This suggested that highly plastic genes as label expressed genes (Supplemental Figs. 1, 2), which suggests that genes could be suitable to mark immune cell subpopulations, as moderately expressed genes should play important roles in con- well as that potentially novel immune cell subpopulations could tribution to the phenotypic heterogeneity of immune cells. be discovered through GPL analysis (see “Discussion” and Unlike the highly plastic genes, low plasticicity genes were Supplemental Table IV). Moreover, through electronic sorting of generally associated with cellular metabolic processes, protein immune cell samples with different expression intensities of a transport and localization, cell cycles, or RNA processing. Their highly plastic gene, the tightly coexpressed genes could be iden- molecular functions were primarily related to protein binding, tified and further used to evaluate the function of the potential catalytic activity, nucleotide binding, or transferase activity immune cell subpopulation.

(Supplemental Table III). Therefore, low plasticicity genes should hi + play important roles in basic cell life processes, performing Electronic sorting of FOXP3 CD4 T cells functions akin to housekeeping (22). Moreover, immune cell The electronic sorting process is shown in Fig. 1. FOXP3, a lineage-specific biomarkers were expected to be found among the characteristic transcription factor of Tregs, was found to be highly Downloaded from

Table I. Functional enrichment analysis of genes with GPL scores $ 45

Category Species ID Name p Value

GO: molecular function Human GO:0005125 Cytokine activity 4.13E216 http://www.jimmunol.org/ GO:0008009 Chemokine activity 2.63E212 GO:0042379 Chemokine receptor binding 7.03E212 GO:0001664 G-protein-coupled receptor binding 1.47E210 GO:0005102 Receptor binding 2.77E210 GO:0019955 Cytokine binding 1.28E208 GO:0060089 Molecular transducer activity 5.44E206 GO:0004871 Signal transducer activity 5.44E206 GO:0004950 Chemokine receptor activity 3.55E205 GO:0001871 Pattern binding 3.64E205 2

Mouse GO:0030246 Carbohydrate binding 6.16E 15 by guest on October 2, 2021 GO:0001871 Pattern binding 9.65E212 GO:0030247 Polysaccharide binding 9.65E212 GO:0005125 Cytokine activity 3.66E211 GO:0008009 Chemokine activity 1.20E210 GO:0042379 Chemokine receptor binding 1.51E210 GO:0005102 Receptor binding 2.49E210 GO:0005539 Glycosaminoglycan binding 5.62E210 GO:0001664 G protein–coupled receptor binding 2.06E209 GO:0019838 Growth factor binding 1.72E207 GO: biological process Human GO:0002376 Immune system process 2.46E225 GO:0006955 Immune response 1.09E223 GO:0009605 Response to external stimulus 1.72E219 GO:0009611 Response to wounding 1.80E216 GO:0006952 Defense response 3.86E216 GO:0006935 Chemotaxis 1.42E215 GO:0042330 Taxis 1.42E215 GO:0040011 Locomotion 5.61E215 GO:0006954 Inflammatory response 3.53E214 GO:0048731 System development 6.55E214 Mouse GO:0006955 Immune response 2.35E222 GO:0002376 Immune system process 1.01E221 GO:0009605 Response to external stimulus 1.71E221 GO:0050896 Response to stimulus 1.63E220 GO:0006952 Defense response 2.15E217 GO:0009611 Response to wounding 9.96E217 GO:0006954 Inflammatory response 1.27E216 GO:0006950 Response to stress 1.50E213 GO:0048518 Positive regulation of biological process 8.23E212 GO:0048522 Positive regulation of cellular process 1.96E211 KEGG pathway Human hsa04060 Cytokine-cytokine receptor interaction 9.84E220 hsa04640 Hematopoietic cell lineage 2.08E210 hsa04062 Chemokine signaling pathway 3.85E207 Mouse mmu04060 Cytokine-cytokine receptor interaction 1.99E216 mmu04640 Hematopoietic cell lineage 3.62E208 mmu04512 ECM-receptor interaction 1.03E205 For more information, see Supplemental Table III. 668 ELECTRONIC SORTING OF IMMUNE CELLS plastic, with a GPL score of 33.5 (Supplemental Table IV), which Our results showed that most of the upregulated genes in FOXP3hi made it excellent for sorting FOXP3hiCD4+ T cells. The three gene CD4+ T cells were actually verified in Tregs. These genes were lists used during the sorting process are shown in Supplemental significantly upregulated in Tregs according to their ARSs Table V. After cross-validation using additional data sets, a total (Table II). Our results revealed that genes with functional impor- of 20 upregulated and 19 downregulated genes was identified at the tance in Tregs could be successfully identified by electronic sorting probe set level under our strict filter conditions (Table II). (even if the key surface marker was IL2RA/CD25) and that the Among the upregulated genes, IL2RA (also known as CD25) is a process was actually independent of Treg samples. In addition, our well-known cell surface marker of Tregs. CTLA4 is an important results revealed some novel genes in the Tregs’ functions. These inhibitory molecule that mediates the function of Tregs (23). DUSP4 is genes included SLCO4A1, PMCH, FAM126A,andASB2(ankyrin also one of the signature molecules of Tregs (24). LRRC32, also repeat and SOCS box-containing 2). The functional associations known as GARP, was identified as a Treg-specific activation marker between these genes and Tregs have not been established. and was specifically induced in CD4+CD25hiFOXP3hi Tregs (25–28). During the identification of downregulated genes, the Pearson’s r GAPR can bind latent TGF-b1 on the Treg surface, and it can value was not particularly powerful, because with the increase in stimulateTregstoreleasematureTGF-b1 from the GARP-latent negative correlation strength (i.e., the increase of the absolute r TGF-b1 complex. LAPTM4B plays a number of oncogenic roles, value), the correlated genes tended to be coexpressed (or coexist) and high levels of this protein are associated with poor prognosis in in the same cells (Fig. 2). To solve this problem, a mutual re- many types of cancers. LAPTM4B was recently identified as binding pulsion rate (or APA rate) was introduced. Generally, a high APA to GARP and serving as a negative regulator for TGF-b1 production rate indicated low correlation (Fig. 2). As shown in Table II, most in human Tregs (29). Tregs also express high levels of FAS molecules downregulated genes in FOXP3hiCD4+ T cells were rare in Tregs’ (Fas cell surface death receptor, also known as APO-1/CD95), which functions. Among these genes, ADRB2 (also known as B2AR) was Downloaded from play an important role in CD95L-mediated apoptosis (30). CD80 (also recently reported to be downregulated in Tregs (32). known as B7-1) is a costimulatory molecule that can be expressed on Significant enrichments were observed only in the upregulated activated T cells, but the costimulatory activity produced by CD80 gene set. The results revealed that the functions of FOXP3hiCD4+ expressed by APCs has been the major focus of research. CD80 on T cells were primarily associated with regulatory roles in the Tregs may transduce an inhibitory signal into T cells (31). immune system, especially negative regulations, such as negative http://www.jimmunol.org/

Table II. DEGs between FOXP3hi and FOXP3lo/2 CD4+ T cells

Probe Set Symbol Gene ID Average D r APA Rate ARS (Treg) ARS (CD4T) Da p Valueb 219911_s_at SLCO4A1 28231 24.02 0.61 71.18 51.89 19.29 4.13E209 206942_s_at PMCH 5367 23.72 0.52 88.21 43.55 44.66 1.12E229 211269_s_at IL2RA 3559 21.85 0.52 95.18 74.56 20.62 1.10E220 223625_at FAM126A 84668 19.42 0.54 68.33 51.15 17.18 6.93E210 2 206341_at IL2RA 3559 18.77 0.51 93.04 73.79 19.25 3.05E 20 by guest on October 2, 2021 233857_s_at ASB2 51676 16.87 0.56 59.71 47.77 11.94 3.16E204 234964_at TRD@ 6964 16.55 0.58 63.19 50.66 12.53 5.49E205 1558920_at LOC100128590 100128590 16.08 0.61 61.83 32.35 29.48 2.08E211 227915_at ASB2 51676 15.82 0.55 69.9 51.95 17.95 1.05E208 204015_s_at DUSP4 1846 15.52 0.56 91.75 72.07 19.68 6.74E218 221331_x_at CTLA4 1493 15.29 0.59 92.03 74.95 17.08 2.65E222 234362_s_at CTLA4 1493 15.11 0.60 92.46 75.48 16.98 2.54E221 1554679_a_at LAPTM4B 55353 15.04 0.54 67.81 61.89 5.92 1.55E202 244620_at LOC100128590 100128590 14.89 0.63 49.72 20.26 29.46 2.56E207 231794_at CTLA4 1493 13.79 0.59 91.97 72.11 19.86 2.23E228 208767_s_at LAPTM4B 55353 13.24 0.56 80.76 73.47 7.29 2.69E205 216252_x_at FAS 355 12.17 0.63 86.53 74.81 11.72 2.07E209 1554519_at CD80 941 11.72 0.51 74.61 54.01 20.6 2.12E213 215719_x_at FAS 355 11.56 0.62 90.42 79.1 11.32 1.11E209 203835_at LRRC32 2615 11.35 0.69 84.93 40.91 44.02 1.61E234 1564139_at LOC144571 144571 223.73 0.71506 55.33 66.23 210.9 1.87E205 1558569_at LOC100131541 100131541 220.03 0.72051 61.6 76.58 214.98 2.27E204 1557239_at BBX 56987 219.04 0.70236 55.4 59.64 24.24 4.40E201 236966_at ARMC8 25852 218.02 0.70417 65.75 72.21 26.46 9.87E202 229457_at ANKHD1 54882 217.09 0.70054 76.56 72.23 4.33 7.24E202 215754_at SCARB2 950 216.65 0.70417 60.86 66.53 25.67 1.92E204 226344_at ZMAT1 84460 216.31 0.71143 65.49 76.75 211.26 4.30E208 230707_at SORL1 6653 215.24 0.70054 77.74 79.29 21.55 3.49E201 236583_at GIMAP1 170575 215.23 0.70599 73.24 81.22 27.98 2.01E202 239005_at FLJ39739 388685 215.07 0.72777 59.11 77.09 217.98 2.52E210 232346_at LOC388692 388692 215.01 0.70962 58.72 72.33 213.61 3.74E209 1556472_s_at SCML4 256380 214.83 0.7078 47.79 71.61 223.82 1.12E215 238851_at ANKRD13A 88455 214.77 0.70417 56.58 63.35 26.77 4.44E203 206170_at ADRB2 154 214.75 0.71143 40.44 64.2 223.76 3.48E217 229141_at WDR33 55339 214.25 0.74229 70.54 77.88 27.34 7.32E206 236166_at LOC285147 285147 213.99 0.71143 60.72 61.18 20.46 9.41E201 1554638_at ZFYVE16 9765 213.09 0.71143 52.68 57.32 24.64 2.62E202 1559658_at C15orf29 79768 212.38 0.71869 48.6 57.99 29.39 2.73E205 228708_at RAB27B 5874 211.66 0.7078 42.39 73.18 230.79 1.15E227 aD, Difference in ARSs between Tregs and CD4+ T cells. bFor the difference between two ARSs; calculated using the Wilcoxon rank-sum test. The Journal of Immunology 669 regulation of lymphocyte activation (Table III). This finding the filter conditions were loosened. For example, if we did not agreed well with current knowledge about the regulatory roles of consider the size of Pearson’s r value, hundreds of candidate Tregs. These results suggested that electronic sorting could also be DEGs would be produced (Supplemental Table V), including used for functional evaluation. other known cytokines that are highly expressed by Th17 cells (42), such as IL-22 (222974_at, r = +0.027935), CSF2 (210229_s_at, Electronic sorting of cytokine-producing CD4+ T cells r = +0.030727), IL-17F (234408_at, r = +0.42007), CCL20 (205476_at, Some cytokines with a high level of plasticity (Supplemental r = +0.36316), IL-21 (221271_at, r = +0.11259), and IFNG Table IV) were used previously to label Th cell subpopulations. (210354_at, r = +0.14145). Similarly, cytokines, such as IL-17A, These molecules are actually characteristic cytokines for Th1 cells IL-22, IL-26, IL-9, CCL20, and IL-21, were also observed in the (IFNG), Th2 cells (IL-4, IL-5, IL-13), Th17 cells (IL-17A, IL-17F, IL-7F–labeled subpopulation (Supplemental Table V). IL-22, IL-26), Th9 cells (IL-9), Th22 cells (IL-22), Tfh cells We also found that some genes were frequently screened out from (IL-4, IL-21), and Tregs (IL-10, TGFB1) (8). We continued to sort different subpopulations, such as IL-2, IL-3, CSF2, and LIF for + these cytokine-producing CD4 T cells. In this analysis, the cytokines and ZBED2, NFATC1, IRF4, BATF3, and ZBTB32 for + cytokine-producing CD4 T cells were not fully equivalent to the transcription factors. These results suggested that these molecules corresponding Th cells, especially with regard to the unspecific could play important roles in controlling the development and/or cytokines produced by multiple Th subsets, such as IL-10, which the functions of most Th cells. The transcription factor ZBED2 can be produced by Th2 cells, Th9 cells, Th17 cells, and Tregs (9). often obtained a high rank order among the upregulated gene sets of The complete results are shown in Supplemental Table V. We IL4hiCD4+,IL5hiCD4+,IL9hiCD4+,IL13hiCD4+,IL21hiCD4+,and listed some key examples, as follows. IFNGhiCD4+ T cells (Table IV, Supplemental Table V), which IL-12 is a dominant factor driving the development of Th1 cells suggested the functional importance of this factor in Th cells. Downloaded from and in the induction of IFN-g. During Th1 cell development, the Downregulated genes received less attention with regard to key transcription factor TBX21 responds to the initial TCR acti- functions in these cytokine-producing T cells. The candidate vation and promotes expression of IL-12R b2 chain (IL12RB2), transcription factor ZMAT1 was frequently downregulated in the which results in greater IL-12 responsiveness and further pro- subpopulations marked by IL-4, IL-5, IL-9, IL-13, IL-17A, IL-17F, duction of IFN-g (33). In accordance with this pattern, IL12RB2 was IL-21, IL-22, and IL-26 (Supplemental Table V). No functional hi + observed to be upregulated in IFNG CD4 T cells (Supplemental associations have been established between ZMAT1 and Th cells. http://www.jimmunol.org/ Table V). In addition, Th1 cytokines, such as TNF, LTA, and CSF2, ALOX5 (also known as 5-LO) was downregulated in IL-4–, IL-5–, also known as GM-CSF, were observed in the upregulated gene set. and IL-13–labeled subpopulations (Supplemental Table V). It was IL-4, IL-5, and IL-13 are Th2 cytokines; according to our results, revealed that splenic CD4+ T cells from 5-LO–deficient mice + they were tightly coexpressed in CD4 T cells. For example, both produce significantly higher levels of IL-4 and IL-13; moreover, hi + IL-5 and IL-13 occurred in the upregulated gene set of IL4 CD4 immunization of 5-LO–deficient mice results in skewing toward a T cells. Other cytokines, such as IL-3, IL-10, IL-21 (34), CSF2, Th2 immune response (43). This finding is in accord with our LIF, and TNF (35), along with some transcription factors, such as findings in human CD4+ T cells. IRF4 (36), NFATC1 (37), SOCS2 (38), and GFI1 (39, 40), were Because few pure Th cells (Th1, Th2, Th9, Th17, Th22, and hi + by guest on October 2, 2021 also upregulated in IL4 CD4 T cells. These molecules play Tfh), based on sample annotations (Supplemental Table I), were important roles in the development and/or functions of Th2 cells. actually contained in the CD4+ T cell samples, our results further IL-4, IL-10, IL-13, CSF2, LIF, IRF4, and NFATC1 were also showed the feasibility, effectiveness, and predictability of the hi + observed to be upregulated in IL5 CD4 T cells. Moreover, these method for the identification of signature genes. In addition, the hi + molecules (and IL-5) were also upregulated in IL13 CD4 T cells large numbers of upregulated or downregulated candidate genes (Supplemental Table V). (Supplemental Table V) provide new opportunities for studies on For IL-17A, only two upregulated signature molecules the functions and mechanisms of Th cells. (i.e., IL-26 and IL-23R) were obtained under our strict filter + conditions. IL-26 belongs to the Th17 cytokine family; IL-23R is Functional relationships for cytokine-producing CD4 T cells the key receptor for Th17 development, and it defines human Th17 A further analysis revealed the functional relationships among cells (41). However, more upregulated genes would be identified if these cytokine-producing CD4+ T cells when the upregulated

FIGURE 2. Relationships between APA or PP rates and Pearson’s correlation. The x-axes repre- sent correlation coefficients, and the y-axes indicate the PP rate (A and B) or the APA rate (C and D). A total of 20,283 human and 20,963 mouse genes was used for analysis. For each panel, the bottom and top of the box represent the first and third quartiles, and the band inside the box represents the second quartile (the median). The whiskers represent the datum still within the 1.5 interquartile range. 670 ELECTRONIC SORTING OF IMMUNE CELLS

Table III. Functional evaluation of FOXP3hiCD4+ T cells

GO Term Name p Value Adjusted p Value GO:0045619 Regulation of lymphocyte differentiation 9.05E208 3.17E205 GO:0051249 Regulation of lymphocyte activation 2.56E206 8.97E204 GO:0002694 Regulation of leukocyte activation 3.98E206 1.39E203 GO:0050865 Regulation of cell activation 4.92E206 1.72E203 GO:0045580 Regulation of T cell differentiation 5.02E206 1.76E203 GO:0051250 Negative regulation of lymphocyte activation 7.27E206 2.55E203 GO:0006924 Activation-induced cell death of T cells 8.20E206 2.87E203 GO:0070231 T cell apoptosis 8.20E206 2.87E203 GO:0002695 Negative regulation of leukocyte activation 8.41E206 2.94E203 GO:0042129 Regulation of T cell proliferation 9.23E206 3.23E203 GO:0050866 Negative regulation of cell activation 1.06E205 3.70E203 GO:0050670 Regulation of lymphocyte proliferation 2.12E205 7.43E203 GO:0032944 Regulation of mononuclear cell proliferation 2.20E205 7.69E203 GO:0070663 Regulation of leukocyte proliferation 2.20E205 7.69E203 GO:0002683 Negative regulation of immune system 2.59E205 9.08E203 process GO:0070227 Lymphocyte apoptosis 3.04E205 1.06E202 GO:0050863 Regulation of T cell activation 6.43E205 2.25E202 GO:0043029 T cell homeostasis 9.82E205 3.44E202 GO:0002682 Regulation of immune system process 1.14E204 3.97E202 Downloaded from

DEGs of each subpopulation were adjusted to contain similar gene ships. IL-22–producing CD4+ T cells also produced high levels counts by decreasing the correlation strength. As shown in Fig. 3, of IL-13 in patients with atopic dermatitis, and a possible shift some cytokine-producing CD4+ T cells were tightly connected, from Th22 to Th22/Th2 cells in the skin microenvironment was such as those labeled IL-4, IL-5, IL-10, IL-13 and IL-22. IL-4, IL- suggested (45). http://www.jimmunol.org/ 5, and IL-13 belong to the Th2 family cytokines. Th2 cells also In relative terms, IL-9 and IL-17F, as well as IL-17A and IL-26, stably express larger amounts of IL-10 than do other effector were tightly connected (Figs. 3, 4). In addition to expressing IL- T cell subsets (44). Our results showed that these Th2 cytokine– 17A and IL-17F, Th17 cells express IL-9 (46), IL-21, IL-22, and producing CD4+ T cells had closer relationships. IL-22 is not a IL-26 (41, 47). Th9 cells can produce large amounts of IL-9 and specific Th cell cytokine, because Th17 and Th22 cells can IL-10 and small amounts of IL-17, IL-21, IL-22, and IFN-g (47). produce it (9). In our system, IL22hiCD4+ T cells only referred to In addition, IFN-g can be coexpressed with IL-17A (48) and IL-21 CD4+ T cells that highly expressed IL-22; thus, they might be (49). According to our results, these cytokine-producing CD4+ derived from both Th17 and Th22, and/or from other CD4+ T cells shared more close relationships than Th2 cytokine–pro- T cells that can produce IL-22. Our results revealed that IL22hi ducing T cells (Figs. 3, 4). In addition, TGFB1hiCD4+ T cells were by guest on October 2, 2021 CD4+ T cells were actually more similar to Th2 cells. Correla- closer to FOXP3hiCD4+ T cells than to other subpopulations tion analysis also revealed that IL-22, IL-10, and some other Th2 (Figs. 3, 4), which might be expected, because it is well known cytokines were clustered together (Fig. 4), which further sug- that TGF-b can induce Treg differentiation, and Tregs need gested that these Th cells could share close functional relation- TGF-b to exert their regulatory functions (8).

Table IV. The top five genes at the probe set level for several cytokine-producing subpopulations

Label Gene Probe Set Symbol r Mean of Average D of All Data Sets IFNG 219836_at ZBED2 0.70 34.19 207849_at IL2 0.60 35.00 207906_at IL3 0.54 31.56 228066_at C17orf96 0.69 31.24 221271_at IL21 0.57 29.42 IL-4 210229_s_at CSF2 0.96 34.78 207849_at IL2 0.78 31.13 207906_at IL3 0.63 30.27 207952_at IL5 0.93 30.58 219836_at ZBED2 0.56 32.06 IL-9 207906_at IL3 0.51 23.73 219836_at ZBED2 0.55 23.94 228066_at C17orf96 0.67 22.45 204695_at CDC25A 0.62 20.90 204603_at EXO1 0.61 18.59 IL-17A 221111_at IL26 0.65 26.33 1552912_a_at IL23R 0.65 15.60 IL-22 210229_s_at CSF2 0.79 33.65 207906_at IL3 0.69 29.91 207849_at IL2 0.78 29.23 207952_at IL5 0.93 27.99 207539_s_at IL4 0.79 28.15 For more details, see Supplemental Table V. The Journal of Immunology 671 Downloaded from http://www.jimmunol.org/

FIGURE 3. Connections between Th cell subpopulations based on electronic sorting results. For each labeled gene (nodes in red), the white nodes, which are directly connected by edges (green lines between any two nodes), represent the upregulated genes.

Therefore, our results revealed that, through electronic sorting regulation of immune system process (GO:0002682), response to and correlation analyses, the functional relationships among dif- wounding (GO:0009611), external stimulus (GO:0009605), in- by guest on October 2, 2021 ferent immune cell subpopulations could be clearly addressed. flammatory response (GO:0006954), and many other regulatory functions (Supplemental Table VII). Therefore, these results pro- Prediction of CD4+ T cell subpopulations and the evaluation of vide a rich resource for big data–driven experimental verification their functions to discover novel CD4+ T subpopulations and to further identify Because different immune cells (either from developmental line- their signature genes and functions. ages or differential states) are actually committed to distinctive functions, we used the term “subpopulations” to designate all of Discussion the functional states of immune cells, without considering whether With the accumulation of publicly available immunomics data, these subpopulations represent true developmental lineages and integrative computational analysis and effective utilization of these stages. Next, we made a systematic analysis of all of the potential big data for gaining immunological insights are becoming in- CD4+ T cell subpopulations and their functional clues. creasingly important for immunologists. As a great and valuable During screening of label genes, genes with GPL scores $ 30 resource, omics big data should be explored more deeply and used were considered. Moreover, the samples were required to have a to drive big data–based discoveries that can provide key clues and maximum rank score . 90 (or 80 in cases in which the third solutions to resolve immunological problems. In recent years, quartile [Q3] was not , 55) to ensure the inclusion of sufficient immune cells were found to be extremely complex, and growing samples with high and low/no expression. Some genes had mul- numbers of subpopulations have been or will be discovered. High tiple probe sets, and the matched probe sets with the maximum heterogeneity and plasticity among immune cell subsets further ARS were used. In total, we identified 747 novel label genes for increases the complexity. Thus, the use of big data as a new further analysis, and a total of 760 genes (747, plus the 12 technology makes it highly feasible and effective to perform cytokine-encoding genes above, plus FOXP3) was electronically comprehensive and global evaluations of many immunological sorted (Supplemental Table I). issues, as well as to solve immunological problems, with great In total, 127,142 upregulated and 24,775 downregulated genes savings in terms of time, money, and research effort. were identified using our strict filter conditions (Supplemental Analysis of GPL based on big data can be well suited to the Table VI). This meant that an average of 177 (127,142/717) up- evaluation of marker molecules (14). Genes can be divided into regulated and 42 (24,755/583) downregulated genes (at the gene those with low and high plasticity, which have different applica- level but not the probe set level) were screened out for each label tions (Supplemental Fig. 3). Genes with low plasticity are more gene. Functional analyses revealed that the main functions of the likely to maintain relatively stable states with high or low/no potential CD4+ T cell subpopulations were associated with re- expression (Supplemental Figs. 1, 2), and these genes are suit- sponse to immune system process (GO:0002376), immune re- able for immune cell lineage markers, including positive and sponse (GO:0006955), hematopoietic cell lineage (hsa04640), negative markers (14 ) (Supplemental Figs. 4, 5). However, highly 672 ELECTRONIC SORTING OF IMMUNE CELLS

with inflammatory response (GO:0006954) (Supplemental Table VII). Through reverse electronic sorting, we found that, in addition to the FOXP3hiCD4+ T cells, at least the ASB2hiCD4+ T cells were asso- ciated with negative regulatory functions (No. 409 in Supplemental Table VII). ASB2 was highly plastic in CD4+ T cells (GPL score = 54. 5), and it was positively correlated to FOXP3 (Table II, Supplemental Table VI). However, all electronic sorting analyses for highly plastic genes must be completed before reverse electronic sorting. Although microarray data were used for analysis in this study, we believe that the current pipeline is also suitable for RNA-sequencing (RNA-seq) data; however, more complicated data pretreatments are needed. For example, the raw data file from RNA-seq is often very large and is more difficult and time-consuming to process. Moreover, a common gene set must be selected because the detectable gene counts FIGURE 4. Cluster analysis of several label genes based on their ex- often vary among samples. Many genes often have the same raw pressional correlations. Gene expression matrix of all genes in human counts, reads per kilobase of exon model per million mapped reads CD4+ T cells was from the ImmuCo database. All Pearson’s coefficients values, or fragments per kilobase of transcript per million fragments for each label gene to other genes were calculated and used to generate the mapped values. This results in confusion when determining the proper correlation matrix. Euclidean distance was calculated, and the function rank scores using our previous method (14). Most importantly, the “hclust” with default parameter was used to generate the object for sample sizes of RNA-seq for immune cells are still far smaller than Downloaded from graphics plotting in the R software environment. those from microarray in the Gene Expression Omnibus database. It can be expected that increasing numbers of immune cell plastic genes are very suitable to serve as markers for immune cell subpopulations will be identified in the future. Therefore, all of the subpopulations (Supplemental Fig. 3, Supplemental Table IV). potential immune cell subpopulations that can be labeled by various Our results also implied that more subtle subdivisions could be markers or their combinations will constitute a huge repertoire of further identified among the greater subpopulations based on GPL. subpopulations representing various functional states that we can call http://www.jimmunol.org/ For example, Treg subpopulations, such as CXCR3+ Tregs (50), a “populationome.” Correspondingly, the plastic genes used to label HLA-DRA+ Tregs (51), TIM3+ Tregs (52), CCR6+ Tregs (53), a populationome constitute a “plasticitome.” The repertoire of a LAG3+ Tregs (54), and CD8A+ Tregs (55), were reported. All of populationome is huge. However, there remain important questions these label genes are highly plastic in Tregs, as shown in the concerning the conditions in which these subpopulations occur and ImmuSort database. Actually, many studies revealed that Tregs are how their phenotypes can be stabilized. Our research on electronic highly heterogeneous (56, 57). Therefore, GPL directly contrib- sorting can only partially resolve these questions (14), because not utes to phenotypic diversity. all of the experimental conditions were described in detail during In this study, we established a new method, electronic sorting, to the data submission and publication. Determining the nomenclature by guest on October 2, 2021 identify the potential characteristic genes of immune cell sub- for the numerous immune cell subpopulations is another issue. populations. Electronic sorting uses correlation analysis and takes Actually, many cytokines, such as IL-2, CSF2, IL-4, IL-9, IL-10, dosage changes into account. Theoretically, the genes that are IL-17, IL-21, IL-22, and IFNG, are cross-produced by different Th tightly associated with a cytokine of interest, a transcription factor, cell subsets (one explanation for this phenomenon is phenotypic or a cell surface membrane protein can be identified by electronic flexibility and plasticity). Perhaps we should establish a nomen- sorting. Electronic sorting can provide vital information for FACS clature system that is based entirely on single genes and their analysis; however, as a bioinformatics method, it cannot replace combinations and not merely on Th9, Th17, Th22, Tfh, and so on. FACS, which is generally performed at the protein level and is In conclusion, the highly plastic genes identified by big data widely used in immunological studies. analysis have important functions in the immune system, and they Because no or very few characteristic genes were identified for can be used to denote immune cell subpopulations. Electronic some label genes (Supplemental Table VI), the process of elec- sorting provides a novel method for using highly plastic genes to tronic sorting for these genes may require further optimization. identify immune cell subpopulations and their functional studies. Based on our statistics for the 760 label genes (Supplemental Moreover, these methods are suitable for other immune cells, such Table VI), we found that the genes with higher rank orders of- as CD8+ T cells, B cells, dendritic cells, monocytes, macrophages, ten received more support during cross-validation (Supplemental neutrophils, and NK cells, as well as for nonimmune cells and Fig. 6). This finding suggests that genes with larger absolute d cancers for studies of gene expression and regulation. values might have more stability. Actually, a large-magnitude change in the d values generally achieved greater statistical Disclosures significance, as shown in the volcano plots, when FOXP3 was The authors have no financial conflicts of interest. used as an example (Supplemental Fig. 7). In addition, our results show that the identified DEGs for all label genes are generally highly plastic (Supplemental Fig. 8), which suggests that the References method to detect DEGs in the differential expression study might 1. Satpathy, A. T., X. Wu, J. C. Albring, and K. M. Murphy. 2012. Re(de)fining the dendritic cell lineage. Nat. Immunol. 13: 1145–1154. actually be to identify certain highly plastic genes. Finally, it must 2. Serafini, N., C. A. Vosshenrich, and J. P. Di Santo. 2015. Transcriptional regu- be noted that not all label genes with multiple DEGs are expected lation of innate lymphoid cell fate. Nat. Rev. Immunol. 15: 415–428. 3. Shen, P., and S. Fillatreau. 2015. Antibody-independent functions of B cells: a to be observed at the same time under an experimental condition. focus on cytokines. Nat. Rev. Immunol. 15: 441–451. Electronic sorting made it possible to trace immune cell sub- 4. Galli, S. J., N. Borregaard, and T. A. Wynn. 2011. Phenotypic and functional populations according to predefined DEGs and/or functional an- plasticity of cells of innate immunity: macrophages, mast cells and neutrophils. Nat. Immunol. 12: 1035–1044. notations. This was a process of reverse electronic sorting. For 5. Hashimoto, D., J. Miller, and M. Merad. 2011. Dendritic cell and macrophage example, there were $168 potential subpopulations associated heterogeneity in vivo. Immunity 35: 323–335. The Journal of Immunology 673

6. Ziegler-Heitbrock, L., P. Ancuta, S. Crowe, M. Dalod, V. Grau, D. N. Hart, 32. Guereschi, M. G., L. P. Araujo, J. T. Maricato, M. C. Takenaka, P. J. Leenen, Y. J. Liu, G. MacPherson, G. J. Randolph, et al. 2010. Nomen- V. M. Nascimento, B. C. Vivanco, V. O. Reis, A. C. Keller, P. C. Brum, and clature of monocytes and dendritic cells in blood. Blood 116: e74–e80. A. S. Basso. 2013. Beta2-adrenergic receptor signaling in CD4+ Foxp3+ regu- 7. Della Chiesa, M., M. Falco, M. Podesta`, F. Locatelli, L. Moretta, F. Frassoni, and latory T cells enhances their suppressive function in a PKA-dependent manner. A. Moretta. 2012. Phenotypic and functional heterogeneity of human NK cells Eur. J. Immunol. 43: 1001–1012. developing after umbilical cord blood transplantation: a role for human cyto- 33. Amsen, D., C. G. Spilianakis, and R. A. Flavell. 2009. How are T(H)1 and T(H)2 megalovirus? Blood 119: 399–410. effector cells made? Curr. Opin. Immunol. 21: 153–160. 8. Deenick, E. K., C. S. Ma, R. Brink, and S. G. Tangye. 2011. Regulation of T 34. Wurster, A. L., V. L. Rodgers, A. R. Satoskar, M. J. Whitters, D. A. Young, follicular helper cell formation and function by antigen presenting cells. Curr. M. Collins, and M. J. Grusby. 2002. Interleukin 21 is a T helper (Th) cell 2 Opin. Immunol. 23: 111–118. cytokine that specifically inhibits the differentiation of naive Th cells into in- 9. Akdis, M., O. Palomares, W. van de Veen, M. van Splunter, and C. A. Akdis. terferon gamma-producing Th1 cells. J. Exp. Med. 196: 969–977. 2012. TH17 and TH22 cells: a confusion of antimicrobial response with tissue 35. Ullah, U., P. Tripathi, R. Lahesmaa, and K. V. Rao. 2012. Gene set enrichment inflammation versus protection. J. Allergy Clin. Immunol. 129: 1438–1449; quiz analysis identifies LIF as a negative regulator of human Th2 cell differentiation. 1450–1451. doi:10.1016/j.jaci.2012.05.003 Sci. Rep. 2: 464. 10. Murphy, K. M., and B. Stockinger. 2010. Effector T cell plasticity: flexibility in 36. Ahyi, A. N., H. C. Chang, A. L. Dent, S. L. Nutt, and M. H. Kaplan. 2009. IFN the face of changing circumstances. Nat. Immunol. 11: 674–680. regulatory factor 4 regulates the expression of a subset of Th2 cytokines. J. 11. Nakayamada, S., H. Takahashi, Y. Kanno, and J. J. O’Shea. 2012. Helper T cell Immunol. 183: 1598–1606. diversity and plasticity. Curr. Opin. Immunol. 24: 297–302. 37. Brogdon, J. L., D. Leitenberg, and K. Bottomly. 2002. The potency of TCR 12. Takeda, J., C. Yamasaki, K. Murakami, Y. Nagai, M. Sera, Y. Hara, N. Obi, signaling differentially regulates NFATc/p activity and early IL-4 transcription in T. Habara, T. Gojobori, and T. Imanishi. 2013. H-InvDB in 2013: an omics study naive CD4+ T cells. J. Immunol. 168: 3825–3832. platform for human functional gene and transcript discovery. Nucleic Acids Res. 38. Knosp, C. A., H. P. Carroll, J. Elliott, S. P. Saunders, H. J. Nel, S. Amu, 41: D915–D919. J. C. Pratt, S. Spence, E. Doran, N. Cooke, et al. 2011. SOCS2 regulates T helper 13. da Cunha, J. P., P. A. Galante, J. E. de Souza, R. F. de Souza, P. M. Carvalho, type 2 differentiation and the generation of type 2 allergic responses. J. Exp. D. T. Ohara, R. P. Moura, S. M. Oba-Shinja, S. K. Marie, W. A. Silva, Jr., et al. Med. 208: 1523–1531. 2009. Bioinformatics construction of the human cell surfaceome. Proc. Natl. 39. Zhu, J., L. Guo, B. Min, C. J. Watson, J. Hu-Li, H. A. Young, P. N. Tsichlis, and Acad. Sci. USA 106: 16752–16757. W. E. Paul. 2002. Growth factor independent-1 induced by IL-4 regulates Th2 14. Wang, P., Y. Yang, W. Han, and D. Ma. 2015. ImmuSort, a database on gene cell proliferation. Immunity 16: 733–744. Downloaded from plasticity and electronic sorting for immune cells. Sci. Rep. 5: 10370. 40. Zhu, J., D. Jankovic, A. Grinberg, L. Guo, and W. E. Paul. 2006. Gfi-1 plays an 15. Wang, P., H. Qi, S. Song, S. Li, N. Huang, W. Han, and D. Ma. 2015. ImmuCo: a important role in IL-2-mediated Th2 cell expansion. Proc. Natl. Acad. Sci. USA database of gene co-expression in immune cells. Nucleic Acids Res. 43: D1133– 103: 18214–18219. D1139. 41. Wilson, N. J., K. Boniface, J. R. Chan, B. S. McKenzie, W. M. Blumenschein, 16. Barrett, T., S. E. Wilhite, P. Ledoux, C. Evangelista, I. F. Kim, M. Tomashevsky, J. D. Mattson, B. Basham, K. Smith, T. Chen, F. Morel, et al. 2007. Develop- K. A. Marshall, K. H. Phillippy, P. M. Sherman, M. Holko, et al. 2013. NCBI ment, cytokine profile and function of human interleukin 17-producing helper GEO: archive for functional genomics data sets–update. Nucleic Acids Res. 41: T cells. Nat. Immunol. 8: 950–957.

D991–D995. 42. Annunziato, F., L. Cosmi, F. Liotta, E. Maggi, and S. Romagnani. 2012. Defining http://www.jimmunol.org/ 17. Huang, W., B. T. Sherman, and R. A. Lempicki. 2009. Bioinformatics enrich- the human T helper 17 cell phenotype. Trends Immunol. 33: 505–512. ment tools: paths toward the comprehensive functional analysis of large gene 43. DiMeo, D., J. Tian, J. Zhang, S. Narushima, and D. J. Berg. 2008. Increased lists. Nucleic Acids Res. 37: 1–13. interleukin-10 production and Th2 skewing in the absence of 5-lipoxygenase. 18. The Gene Ontology Consortium. 2013. Gene Ontology annotations and re- Immunology 123: 250–262. sources. Nucleic Acids Res. 41: D530–D535. 44. Rutz, S., and W. Ouyang. 2011. Regulation of interleukin-10 and interleukin-22 19. Kanehisa, M., S. Goto, Y. Sato, M. Kawashima, M. Furumichi, and M. Tanabe. expression in T helper cells. Curr. Opin. Immunol. 23: 605–612. 2014. Data, information, knowledge and principle: back to metabolism in 45. Teraki, Y., A. Sakurai, and S. Izaki. 2013. IL-13/IL-22-coproducing T cells, a KEGG. Nucleic Acids Res. 42: D199–D205. novel subset, are increased in atopic dermatitis. J. Allergy Clin. Immunol. 132: 20. Huang, W., B. T. Sherman, and R. A. Lempicki. 2009. Systematic and integrative 971–974. analysis of large gene lists using DAVID bioinformatics resources. Nat. Protoc. 46. Nowak, E. C., C. T. Weaver, H. Turner, S. Begum-Haque, B. Becher, 4: 44–57. B. Schreiner, A. J. Coyle, L. H. Kasper, and R. J. Noelle. 2009. IL-9 as a me-

21. Shannon, P., A. Markiel, O. Ozier, N. S. Baliga, J. T. Wang, D. Ramage, diator of Th17-driven inflammatory disease. J. Exp. Med. 206: 1653–1660. by guest on October 2, 2021 N. Amin, B. Schwikowski, and T. Ideker. 2003. Cytoscape: a software envi- 47. Ja¨ger, A., V. Dardalhon, R. A. Sobel, E. Bettelli, and V. K. Kuchroo. 2009. Th1, ronment for integrated models of biomolecular interaction networks. Genome Th17, and Th9 effector cells induce experimental autoimmune encephalomy- Res. 13: 2498–2504. elitis with different pathological phenotypes. J. Immunol. 183: 7169–7177. 22. Chang, C. W., W. C. Cheng, C. R. Chen, W. Y. Shu, M. L. Tsai, C. L. Huang, and 48. Hirota, K., J. H. Duarte, M. Veldhoen, E. Hornsby, Y. Li, D. J. Cua, H. Ahlfors, I. C. Hsu. 2011. Identification of human housekeeping genes and tissue-selective C. Wilhelm, M. Tolaini, U. Menzel, et al. 2011. Fate mapping of IL-17- genes by microarray meta-analysis. PLoS One 6: e22859. producing T cells in inflammatory responses. Nat. Immunol. 12: 255–263. 23. Shevach, E. M. 2009. Mechanisms of foxp3+ T regulatory cell-mediated sup- 49. Xiao, L., L. Jia, Y. Zhang, S. Yu, X. Wu, B. Yang, H. Li, and C. Wu. 2015. pression. Immunity 30: 636–645. Human IL-21+IFN-g+CD4+ T cells in nasal polyps are regulated by IL-12. Sci. 24. Feuerer, M., J. A. Hill, D. Mathis, and C. Benoist. 2009. Foxp3+ regulatory Rep. 5: 12781. T cells: differentiation, specification, subphenotypes. Nat. Immunol. 10: 689– 50. Hoerning, A., K. Koss, D. Datta, L. Boneschansker, C. N. Jones, I. Y. Wong, 695. D. Irimia, K. Calzadilla, F. Benitez, P. F. Hoyer, et al. 2011. Subsets of human 25. Wang, R., Q. Wan, L. Kozhaya, H. Fujii, and D. Unutmaz. 2008. Identification of CD4(+) regulatory T cells express the peripheral homing receptor CXCR3. Eur. a regulatory T cell specific cell surface molecule that mediates suppressive J. Immunol. 41: 2291–2302. signals and induces Foxp3 expression. PLoS One 3: e2705. 51. Baecher-Allan, C., E. Wolf, and D. A. Hafler. 2006. MHC class II expression iden- 26. Wang, R., L. Kozhaya, F. Mercer, A. Khaitan, H. Fujii, and D. Unutmaz. 2009. tifies functionally distinct human regulatory T cells. J. Immunol. 176: 4622–4631. Expression of GARP selectively identifies activated human FOXP3+ regulatory 52. Gautron, A. S., M. Dominguez-Villar, M. de Marcken, and D. A. Hafler. 2014. T cells. Proc. Natl. Acad. Sci. USA 106: 13439–13444. Enhanced suppressor function of TIM-3+ FoxP3+ regulatory T cells. Eur. J. 27. Tran, D. Q., J. Andersson, R. Wang, H. Ramsey, D. Unutmaz, and Immunol. 44: 2703–2711. E. M. Shevach. 2009. GARP (LRRC32) is essential for the surface expression of 53. Kleinewietfeld, M., F. Puentes, G. Borsellino, L. Battistini, O. Ro¨tzschke, and latent TGF-beta on platelets and activated FOXP3+ regulatory T cells. Proc. K. Falk. 2005. CCR6 expression defines regulatory effector/memory-like cells Natl. Acad. Sci. USA 106: 13445–13450. within the CD25(+)CD4+ T-cell subset. Blood 105: 2877–2886. 28. Stockis, J., D. Colau, P. G. Coulie, and S. Lucas. 2009. Membrane protein GARP 54. Camisaschi, C., C. Casati, F. Rini, M. Perego, A. De Filippo, F. Triebel, is a receptor for latent TGF-beta on the surface of activated human Treg. Eur. J. G. Parmiani, F. Belli, L. Rivoltini, and C. Castelli. 2010. LAG-3 expression Immunol. 39: 3315–3322. defines a subset of CD4(+)CD25(high)Foxp3(+) regulatory T cells that are ex- 29. Huygens, C., S. Lie´nart, O. Dedobbeleer, J. Stockis, E. Gauthy, P. G. Coulie, and panded at tumor sites. J. Immunol. 184: 6545–6551. S. Lucas. 2015. Lysosomal-associated Transmembrane Protein 4B (LAPTM4B) 55. Sarrabayrouse, G., C. Bossard, J. M. Chauvin, A. Jarry, G. Meurette, decreases Transforming Growth Factor b1 (TGF-b1) production in human E. Que´vrain, C. Bridonneau, L. Preisser, K. Asehnoune, N. Labarrie`re, et al. regulatory T cells. J. Biol. Chem. 290: 20105–20116. 2014. CD4CD8aa lymphocytes, a novel human regulatory T cell subset induced 30. Fritzsching, B., N. Oberle, N. Eberhardt, S. Quick, J. Haas, B. Wildemann, by colonic bacteria and deficient in patients with inflammatory bowel disease. P. H. Krammer, and E. Suri-Payer. 2005. In contrast to effector T cells, CD4 PLoS Biol. 12: e1001833. +CD25+FoxP3+ regulatory T cells are highly susceptible to CD95 ligand- but 56. Yuan, X., G. Cheng, and T. R. Malek. 2014. The importance of regulatory T-cell not to TCR-mediated cell death. J. Immunol. 175: 32–36. heterogeneity in maintaining self-tolerance. Immunol. Rev. 259: 103–114. 31. Chen, L., and D. B. Flies. 2013. Molecular mechanisms of T cell co-stimulation 57. Li, L., and V. A. Boussiotis. 2011. Molecular and functional heterogeneity of T and co-inhibition. Nat. Rev. Immunol. 13: 227–242. regulatory cells. Clin. Immunol. 141: 244–252.