A Single-Cell Gene-Expression Profile Reveals Inter-Cellular Heterogeneity within Human Monocyte Subsets

Human monocytes are a heterogeneous cell population classified into three different subsets: Classical CD14++CD16-, intermediate CD14++CD16+, and non-classical CD14+CD16++ monocytes. These subsets are distinguished by their differential expression of CD14 and CD16, and unique gene expression profile. So far, the variation in inter-cellular gene expression within the monocyte subsets is largely unknown. In this study, the cellular variation within each human monocyte subset from a single healthy donor was described by using a novel single-cell PCR gene-expression analysis tool. We investigated 86 different genes mainly encoding cell surface markers, and proteins involved in immune regulation. Within the three human monocyte subsets, our descriptive findings show multimodal expression of key immune response genes, such as CD40, NFⱪB1, RELA, TLR4, TLR8 and TLR9. Furthermore, we discovered one subgroup of cells within the classical monocytes, which showed alterations of 22 genes e.g. IRF8, CD40, CSF1R, NFⱪB1, RELA and TNF. Additionally one subgroup within the intermediate and non-classical monocytes also displayed distinct gene signatures by altered expression of 8 and 6 genes, respectively. Hence the three monocyte subsets can be further subdivided according to activation status and differentiation, independently of the traditional classification based on cell surface markers. Demonstrating the use and the ability to discover cell heterogeneity within defined populations of human monocytes is of great importance, and can be useful in unravelling inter-cellular variation in leukocyte populations, identifying subpopulations involved in disease pathogenesis and help tailor new therapies.


Introduction
Blood monocytes are a heterogeneous population of innate immune leukocytes. They are involved in the innate immune response to pathogens by phagocytosis, the release of reactive oxygen species, cytokines and chemokines and by antigen presentation, thereby modulating and activating cells within the adaptive immune system [1]. The diversity within the human classical monocytes were converted to cDNA and amplified with 85 target genes ( Table 1). The 85 genes analysed were chosen based on differential expression shown in earlier micro-array results [11,15], and by their relation to cell function, as well as biological or immune regulated processes. The genes were plotted into a PCA score plot showing that the single-sorted cells were clustered according to their subset, classical, intermediate or non-classical, verifying that the cells were sorted correctly ( Fig 1B). The expression of each gene was analysed using the SINGuLAR R package. No signal in the Fluidigm qPCR were interpreted as expression below detection limit and referred to as non-expression. We found a differential expression of 80 genes within the 3 subgroups of monocytes ( Fig 1C and Table 2). Five of the genes, CCR7, MMP12, MRC1, ULBP1 and ULBP2 were not expressed by any of the subsets. Four genes, CCR2, CD163, CLEC4E, and SERPINB2 were exclusively not expressed by the non-classical subset. Expression of IL6 was not detectable in the classical subset and very low expression was observed in the intermediate and non-classical subset. MMP9 was weakly expressed by the classical and non-classical monocytes and no expression was found in the intermediate subset. TLR3 was not expressed by the classical and intermediate, but weakly by the non-classical monocytes, whereas TREM2 was weakly expressed by the classical subset but not by the intermediate and non-classical subsets. Expression of CCR5 and RAET1G was found only in the intermediate monocytes. These data show that there is great diversity within the three monocyte subsets and that the subsets have a differential gene expression profile with regard to the genes investigated in this study.

Comparison of gene expression investigated by single-cell PCR gene expression analysis versus micro-array
Previous microarray data have shown that genes within diverse cellular and immunologic process are differentially expressed within the three subsets. We therefore wanted to validate our gene expression data generated by Fluidigm (Table 2) with existing array data. We compared our single cell gene-expression data to current expression profiles published by Wong et al. by looking at the general expression pattern since data have not been derived using the same methods. In agreement with the study conducted by Wong

Functional characteristics of the three subsets
To investigate unique characteristics of each monocyte subset, the genes examined were grouped into categories based on functionality ( Table 3). The classical monocytes were shown  to have the highest expression of CD93, CD209, CLEC4E, which are genes involved in pathogen recognition and phagocytosis. Also genes encoding proteins involved in migration and adhesion such as the chemokine receptors, CCR1, CCR2, CCR9, CX3CR1, ITGAM and SELL were most highly expressed by the classical monocytes. The expression of genes encoding scavenger receptors was mostly expressed by the classical and intermediate monocytes. As shown earlier, the intermediate cells had the highest expression of genes involved in co-stimulation and antigen presentation, but also genes involved in NK cell and CD8 T cell activation, such as MICB, RAET1E, RAET1G, RAET1L and ULBP3 were enriched on intermediate monocytes. Furthermore, we also found that the intermediate monocytes had the highest expression of genes involved in cell differentiation and cell function, e.g. the genes encoding the transcription factors IRF5 and IRF8, the NFⱪB1 and RELA genes involved in heterodimer formation of the central immunologic regulatory transcription factor NFⱪB, and CSF1R where the encoded protein controls cell differentiation. Also, genes encoding the cytokines IL1β, IL12A, IL18 and IL23A, proteases, such as cathepsins and MMPs, and protease inhibitors, CST3, TIMP1 and SERPINA1 were highly expressed by the intermediate monocytes.
In contrast, the non-classical monocytes showed the highest expression of TNF and the metalloprotease ADAM17 gene, which are involved in the processing of TNF from the cell surface. Low expression of genes involved in bacterial phagocytosis were found in the non-classical monocyte subsets, thus they had the highest expression of C1QA, a complement component, and the FcγR3A involved in antibody-mediated phagocytosis. Genes encoding the TLR8 and 9 proteins, which are classical, innate pattern recognition receptors (PRR) were also highly expressed by the non-classical monocytes.

Detection of inter-cellular variation
In contrast to previous gene expression profiling studies [9,11,15], the aim of the current study was to investigate the biological genetic variability within the monocyte subsets by using the Fluidigm single-cell gene expression tool. The probes used for the PCR amplification have been tested with DNA, and were able to amplify and detect transcripts. This technique is therefore able to assess selected genes within the individual cells, and investigate possible multimodal gene expression within cell populations. Using a Hartigans dip test, we identified genes within the three subsets that had multimodal expression (Table 4, and Fig 2). Within the classical, intermediate and non-classical subsets, 37, 39 and 36 genes, respectively showed multimodal expression ( Table 4). Some of the genes showed a general multimodal expression in all three subsets such as the co-stimulatory molecule CD40, the scavenger receptor MARCO, the adhesion molecules CD33, CD93 and CX3CR1, the apoptosis genes BAX and BCL2, the genes encoding the cytokines IL12A, IL15 and IL23A, TLR8 and TLR9 and the proteases, CTSK, MMP1 and MMP3. Apart from the genes with multimodal expression in all three subsets, the classical subset showed multimodal expression of the phagocytosis-associated gene CLEC4E, the scavenger receptor gene CD163 and the FcγR3A gene, adhesion and migration genes   BCL2L1 and SELL, respectively, and the scavenger receptor genes CD36, MARCO and MSR1. Furthermore, it was in particular genes involved in regulating immune responses such as C1QA, CD14, IL1β, IL18, TNF, TLR7, TREM1, and SIRPα, and genes encoding the proteases, CTSL and MMP7 that showed multimodal expression. The non-classical monocytes exhibited, in addition to the genes which also showed multimodal expression by the classical and intermediate monocytes, multimodal expression of genes involved in apoptosis, differentiation and activation such as BCL2L1, CSF1R, and IRF5. Similar to the intermediate subset, the non-classical monocytes also showed multimodal expression in a range of immune responsive genes such as ADAM17, C1QA, CD14, LTB, TREM and SIRPα as well as the genes encoding the proteases, CTSK, CTSL, MMP1 and MMP3. We demonstrate here that single-cell gene-expression analysis is a valuable tool in detecting multi-modality within cell populations. To exclude that the bi-modality we observed is not caused by a so-called "drop out effect" due to technology artefact, we calculated the relationship between log-transformed transcript expression data and Hartigan's dip test p-value. When we include the cells with an expression we found no difference comparing uni-and bi-modal genes (p = 0.54). If genes with low expression were more prone to drop out effects we would have expected an overrepresentation of these amongst the bi-modal genes.  Table 4

Classification of subpopulations
With single-cell gene expression profile we are able to study the heterogeneity between and within cell populations. In addition to the multimodal analysis we also analysed co-expression of genes within the classical monocytes, intermediate monocytes and non-classical monocytes in order investigate potential subgroups with distinct gene expression profiles. Using the PCA of single-cell PCR gene expression analysis data it was possible to identify one subgroup of monocytes that diverged from the main populations (Fig 3 and S1 Table). Within the population of classical monocytes, we found one subgroup that showed differential expression of 22 genes (Fig 3A). Of these genes, only 2 genes showed a higher expression than the main population, namely TNFSF15 and TREM1. The remaining genes were all found to have lower expression than in the main population. Within the intermediate subset, we also identified a subgroup of cells displaying differential gene expression of eight genes (Fig 3B). Compared to the main population of intermediate monocytes, only IL10 was found to have lower expression whereas LTB, PTPRC, HMOX1, CSF1R, FCGR3A, RAETL1 and TNF were all found to be significantly more highly expressed. Likewise, within the non-classical monocytes we found a subgroup of cells with a co-expression of IRF8 and RAET1E at higher levels than the main population but showing lower gene expression of CTSS, IL123A, IRF5 and TNF (Fig 3C).
Comparing the expression of genes within the subgroup of cells identified by the PCA plot, we demonstrate that the classical, intermediate and non-classical human blood monocytes each contain a subgroup of cells characterized by distinct gene signatures. This highlights the great diversity and possible plasticity within the human monocyte sub-populations.

Discussion
Comprehensive genome-wide analyses have shown distinct heterogeneity within human monocytes [9,11,15]. Three subsets have been identified in humans, namely the classical, intermediate and non-classical monocytes. In this study we have used the single-cell PCR gene analysis technique from Fluidigm, and confirmed existing data defining three monocyte subsets and demonstrated differential gene expression among the classical, intermediate and non-classical monocytes. Moreover, the differential expression of genes encoding cell surface molecules identified, for instance, the expression of CD163 and TREM2 in classical monocytes and the expression of CCR5 and RAET1G in intermediate monocytes, could be useful when discriminating the three monocyte subsets by methods such as flow cytometry. Furthermore, we have shown inter-cellular variation of genes within each subset, which highlights the heterogeneity of monocytes as a diverse group of innate immune leukocytes containing possible further functional subclasses. Thus, by our study, performed with one donor, we here demonstrate the possibilities of subgrouping the monocytes using single-cell PCR gene expression analysis.
The single-cell gene expression analysis presented here demonstrates high phagocytic capacity (CD93, CD209, CLEC4E, and SIRPA) of the classical monocytes, which is in agreement with previous studies. Also, a high expression of a broad range of innate sensing receptor genes, pro-inflammatory genes and genes linked to innate immune responses (CD14, TLR2, TLR4 and TREM1) are observed for the classical monocytes [9,11,15]. In addition to these findings we have previously shown that the classical monocytes secrete high levels of IL-1β, IL-subgroup is marked by filled red circles in the PCA score plot. B) Bar graph demonstrating the differentially expressed genes by the subgroup within the intermediate monocyte subset identified on the PCA score plot. The subgroup is marked by filled green triangles in the PCA score plot. C) Bar graph demonstrating the differentially expressed genes by the subgroup within the non-classical monocyte subset identified on the PCA score plot. The subgroup is marked by filled blue pluses in the PCA score plot.
doi:10.1371/journal.pone.0144351.g003 10, and TNFα, and that most IL-6 and MMP1 is produced in response to LPS and immunecomplex activation by the classical subset compared to the intermediate and non-classical subset [12]. In accordance to earlier micro-array studies, we also saw the highest expression of genes involved in migration. These findings, together with our previous results showing the highest migratory capacity of the classical monocytes towards CCL2 [12] may underline the capacity of classical monocytes to support inflammation and mount an immune response towards microbial pathogens.
The intermediate monocytes had high antigen presenting potential, which has also been demonstrated by their property to induce CD4 + T cell proliferation [15]. Moreover, the intermediate monocytes were the only subset that expressed CCR5, a chemokine receptor responsible for recruiting dendritic cell (DC) precursors from blood to the draining lymph nodes [17]. In addition, our data demonstrate higher gene expression of several cytokines (IL1β, IL12A, IL18, and IL23A) which are important in inducing functionally distinct CD4 + T helper (Th) cells. IL12 plays a role in the differentiation of Th1 cells, whereas IL6 and IL23 are important in driving and sustaining the differentiation of Th17 cells [18]. Moreover, IL12, IL15, IL18 and TNFSF15 are involved in the induction of T cell receptor-independent cytokine production by CD4 + T cells [19][20][21]. The intermediate monocytes also showed a higher expression of genes linked to the activation status such as apoptosis regulation (BAX and BCL2), cell differentiation and regulation (CSF1R, IRF5, IRF8, NFⱪB1, RELA and PTPRC). This may suggest that these cells are more activated than the classical monocytes.
The CD16 + monocytes have been shown to adhere to the endothelium and mediate arrest through the CX3CL1-CX1CR3 interaction [22]. Additionally, the non-classical monocytes are thought to patrol the vasculature and selectively respond to viral infected-or damaged cells. In line with previous data, in non-classical monocytes we found considerably higher expression of genes coupled to complement (C1QA) and FcR-mediated phagocytosis (FcRγ3A), adhesion (ITGAL and SIGLEC10) and TLR9. However, we find that the genes encoding TLR7 and TLR8, which sense nucleic acids and viruses, are mostly expressed by the intermediate subset.
Moreover, we find the highest expression of the CX3CR1 gene in the classical subset, albeit with small variation among the three subsets. This latter finding conflicts with previous data showing high cell surface expression of CX3CR1 on non-classical monocytes and their capacity to respond to damaged cells and viral infections. However, these observations in gene expression do not always correspond to the actual protein expression. Thus difficulty in functional interpretation of genes is a limitation to transcriptome analysis and may explain the discrepancies between observed protein expression and gene analysis data.
The advantage of single-cell gene expression analysis compared to micro-array is that it provides the possibility to analyse gene expression within single cells, facilitating the discovery of multimodal expression on single cells and possible new subpopulations in cell populations identified to date [23,24]. Therefore, in addition to comparing our results with previous microarray data, we also investigated the intercellular variation among the monocyte subtypes and analysed data for co-expression of genes within each monocyte subset. Interestingly, we could demonstrate that multimodal gene expression is present in all three subsets. The transcription factors IRF8, NFⱪB1 and RELA are among the genes that show multimodal expression in all three subsets. Also genes involved in apoptosis regulation (BAX and BCL2) and cell adhesion (CD33, CD93, CX3CR1 and ITGAM) are expressed multimodally in all three subsets. This multimodality seen in the monocyte subsets may be a reflection of differential maturation and immune activation status. In addition, the classical monocytes also show multimodal expression of the chemokine receptor genes (CCR1 and CCR2), adhesion genes (ITGAL and ITGAM) and innate immune response genes (TLR4, TLR8-9, IL10, IL12A, IL15, IL18, and IL23A), favouring evidence for potentially immune activated cells. However, further data are needed to establish if this is in fact the case, thus these potentially different activation states could be of importance in light of the monocytes ability to extravasate into tissues and respond to pathogenic stimulation. For example, cells expression high levels of CCR2 together with CLEC4E, TLR4, and TNF-a might be more prone for migration and response to bacteria, in contrast to monocytes expressing low levels of CCR2. Also, cells expressing high levels of FCGR3A, CD36 and CD163 might be more prone for scavenging and phagocytosis. Several genes encoding immune responses also show multimodal expression within the intermediate subset together with genes encoding the proteases. Though our study has been carried out using monocytes from a healthy donor, the data presented here serve as a basis for discovering targets on certain subpopulations of cells that may be implicated in disease pathogenesis. In addition, single-cell PCR analyses have revealed that colon cancer tissues contain cell populations distinct from healthy colon, which were not identified by immunohistochemistry or flow cytometry [25]. Moreover we identified a subgroup of cells within the classical group of monocytes that may be less activated given lower expression of, for example genes within the NFⱪB complex, IRF8, and genes encoding several cytokines such as IL1b, IL12A, IL15, IL23A and TNF. In contrast, we describe here a subgroup of cells among the intermediate monocytes, showing higher gene expression of TNF and FCGR3A and lower expression of IL10, suggesting that the cells of this subgroup are more differentiated towards an immune-activated phenotype. The transcription factor IRF8 is known to be involved in the differentiation of monocytes and DCs [26][27][28]. The subgroup identified in the non-classical cells showed higher gene expression of IRF8 and lower gene expression of IRF5, IL23A and TNF. This could imply that this subgroup of cells is more differentiated towards dendritic cells. Of note, non-classical monocytes are likewise shown to have a higher propensity to become dendritic cells [29].
The advantage of using single-cell gene expression profiling methods compared to microarray is the ability to distinguish potential target cells or subgroups of cells implicated in disease pathogenesis. However, the limitation of the single-cell gene expression technique is its being dependent on a predesigned panel of primers compared to the broad dataset obtained using micro-array, which is not limited to investigating predesigned genes only.

Monocyte Purification
A buffy coat from a healthy donor was purchased from the Clinical Immunology Blood Bank, The State University Hospital, Copenhagen. Peripheral Blood Mononuclear Cells (PBMC) were obtained using Ficoll-paque Plus (GE-Healthcare Bio-sciences AB, Uppsala, SE) density centrifugation. Untouched monocytes were isolated from PBMC by negative selection using antibody-coated magnetic bead separation (EasySep Human Monocyte Enrichment Kit without CD16 depletion, Stemcell Technologies, Vancouver, Canada) according to the manufacturers' instructions. The study was approved by the Regional Ethics Committee in Lund, Sweden.

Single-cell gene expression analysis
The monocytes were sorted into 96-well plates by FACS with the target of one cell in each well containing 5 μl RT Mix solution (mixture of VILO reaction mix, SUPERase-ln and 10% NP40 according to the manufacturer's protocol). Samples were frozen on dry ice and after thawing, synthesis of cDNA was performed with SuperScript VILO (Invitrogen). Specific targets amplification (STA) was done with a mixture of 85 PCR primers (see Table 1) using Taqman preamp master mix (Invitrogen) running 22 cycles. The probes used have all been tested with DNA and were able to amplify and detect transcripts. Residual primers were subsequently removed by treatment with Exonuclease I (New England Biolabs).
After the clean-up step, 6 μl of STA PCR product from each sample was transferred to a new microtiter plate and a standard qPCR (TaqMan 2x Universal PCR Master Mix, Applied Biosystems) with a Taqman assay (Invitrogen) directed against 18S, was performed to identify empty wells. Wells with no 18S Ct value or with an 18S Ct value above 40 were considered empty.
Single-cell qPCR was performed on a Fluidigm BiomarkHD instrument using SSO Fast Eva-Green SUpermix (Bio-Ras Laboratories) according to the manufacturer's protocol. All primer pairs from genes listed in Table 1 were used. The data were analysed using the SINGuLAR R package (Fluidigm) and expression values were normalised to that of ACTB.
Sub-groups for each monocyte sub-type were defined from the initial principal component analysis (PCA) plot (Fig 1B). Differentially regulated genes in each sub-group were identified by comparing gene expression in the subgroup versus the rest of the cells in the sub-type using a Students t-test. Differentially regulated genes were defined as those with a p-value < 0.05 and a log2 fold change of at least 1.
Supporting Information S1 Table. List of genes differentially expressed by the monocyte subgroups. (DOCX)