Preliminary characterisation of the response to LPS in different macrophage populations
We harvested macrophages from twenty-five pigs, five from each breed: Duroc (DU), Piétrain (PIE), Landrace (LR), Hampshire (HAM) and Large White (LW), with an average age of 9 weeks. A major advantage of the pig is that large numbers of cells can be harvested for biochemical studies; 1×109 cells for each compartment (Figure 1A). To optimise the comparison, we first examined a time course of the response to LPS of alveolar macrophages (AM), bone marrow-derived macrophages (BMDM) or monocyte-derived macrophages (MDM) from Large White pigs (Figure 1B). There was a clear distinction between the CSF-1 cultured macrophages (BMDM, MDM) and AM. As noted in human and mouse macrophages, TNF production in BMDM and MDM was transient. After 10 hours, there was no further increase in supernatant TNF. By contrast, in AM, TNF production continued to rise even 54 hours post-stimulation. These preliminary studies suggested that the 7 hour time point, used in previous studies of mouse and human macrophages [5], would also provide representative coverage of the response to LPS responsive mRNAs in all of the porcine macrophage populations.
To enable the study of the macrophages from the twenty-five animals at the same time and under the same conditions, AM, PBMC and BMC were frozen as described previously [16] on the day of the harvest and used a few weeks later. PBMC and BMC were cultured for 5 to 7 days in the presence of rhCSF-1 until differentiation into macrophages [16]. The three types of macrophages were seeded at 1×106 cells/ml, cultivated overnight before removing non-adherent cells, replacing the medium, and cultivating with or without LPS (100 ng/ml). Morphologically, the BMDM and MDM were more spread on the substratum by comparison to AM, where a subpopulation of cells is non-adherent (Figure 1C, D, E). Each of the populations expressed the macrophage markers CD14 and CD16, albeit at varying levels (Figure 1F-K). In order to control for the efficacy of LPS stimulation in each experiment, prior to expression profiling, TNF concentration was measured in the supernatant of the culture at 0 and 7 h (Figure 1L). With the exception of AM from HAM and LW, there were no obvious differences between the breeds in terms of the magnitude of this response. The higher production of TNF by AM from these 2 breeds appeared to be due to a higher percentage of adherent macrophages amongst the cells from the broncho-alveolar lavage, which would not interfere with the microarray analysis.
The mRNA was extracted from 25 pigs (from 5 breeds) from 3 different compartment (AM, BMDM and MDM) and at 2 time points and was profiled using the recently developed Snowball Affymetrix array [15]. Of 150 arrays, 10 failed the quality check. The data from the microarray are available at NCBI GEO (GSE45145). Since the pigs studied were a mixture of male and female, we investigated first whether gender has any global impact on gene expression in pig macrophages. A principal components analysis (PCA) of the data did not show any distinct clusters related to the gender of the animals (Figure 2A). Detailed analysis on the differentially regulated (DR) genes highlighted a small list of 10 genes that differed between males and females. The annotation of these probesets links them to obvious sex-dependent genes, such as USP9 Y-linked or inactive-X-specific transcript (XIST). Hence, the inflammatory response to LPS of macrophages in vitro is independent of the gender and we considered all the data as a single set.
Alveolar macrophages show a distinct expression profile from BMDM and MDM
The recently published pig gene expression atlas [15] included replicates of AM, BMDM and MDM from two individual crossbred pigs, but did not compare the populations in detail. We inferred that AM were distinct from macrophages in the wall of the gut, notably in their expression of C-type lectin receptor genes.. The current dataset permits comparison in much greater depth, with the macrophages isolated from the same animals, and with 25-fold replication of the comparison. PCA analysis of the data based upon cell compartment clearly distinguishes AM from BMDM and MDM which are very similar to each other (Figure 2B).
To identify sets of co-expressed genes, the transcriptomic data were loaded into Biolayout Express3D. Using a Pearson correlation threshold cut-off of R=0.91, we obtained a graph comprising 3,586 nodes (individual probesets) made up of 505 different clusters. The clusters can be segregated broadly into two superclusters as shown in Figure 3. The larger of the two superclusters (1,470 nodes) separates into further groups with related expression profiles. One group (including cluster 03), with elevated expression in BMDM/MDM compared to AM (Figure 3A) includes genes encoding proteins involved in the cell cycle and DNA replication such as Cyclin B1, centromere proteins CENP and histones-related proteins HIST1H1A, HIST1H2AC. This most likely reflects the fact that BMDM/MDM have been stimulated to proliferate with the growth factor, CSF-1. A second group, including cluster 10, includes the genes more highly expressed in AM. As inferred previously based upon a much smaller dataset [15], it includes genes encoding the C type lectins such as MRC1, and the signalling molecules TLR4 and MAP3K2 (Figure 3B). A third group, including Cluster 1 (e.g. CCR5, CXCL10, CXCL11, DDX58, IL15), comprises genes that were up-regulated after LPS stimulation in BMDM/MDM but not at all in AM (Figure 3C). Finally, another cluster exemplified by cluster 30, (Figure 3D) which includes genes such as IL1A, CCL3L1 or TRAF3IP2, was induced in all the macrophage populations studied.
The second supercluster (407 nodes) is made up of clusters of genes down-regulated after LPS stimulation, such as cluster 05 and 23 (Figure 3E-F). These two clusters contain genes encoding proteins linked to intracellular signalling, kinase and phosphatase (PKC, PIK3IP1, TRAK2, DNM3, PLEK). The full list of clusters and the probes within them can be found in the Additional file 1.
To confirm the apparent difference between the macrophage populations, we analysed the data using an ANOVA method. 3,322 probesets (fold change >2 or < -2; p adjusted value < 0.01) distinguished AM from BMDM and 3,058 distinguished AM from MDM. Only 144 probesets (fold change >2 or < -2; p adjusted value < 0.01) distinguished the cultivated macrophages; 69 elevated in MDM and 75 in BMDM, essentially confirming the similarity of the populations indicated by PCA analysis. The genes that distinguished BMDM from MDM were expressed at low levels and are probably due to minor contamination with other cells: lymphocytes within the MDM and fibroblasts within the BMDM. The list included genes such as T-cell receptor gamma chain, alpha chain for the MDM and collagen type I and collagen type IV for the BMDM (list in Additional file 2). As shown in Figure 4A, most probesets differentially expressed between AM-MDM and AM-BMDM are shared (n=2,709) and only 47 probesets are differentially expressed in all 3 types of macrophages. Therefore, we selected the genes (n=49, p<0.01) that were at least 30-fold differentially-expressed between AM and BMDM and created a heat-map in Figure 4B. Most of the BMDM/MDM-specific genes are part of cluster 14 (Figure 4C), and include genes such as cell-adhesion molecule 1 (CADM1), integrin alpha 6 (ITGA6), CD36 and the insulin-like growth factor 1 (IGF1). IGF1 is known to be CSF-1-inducible [18]. The genes that are restricted to AM are part of cluster 02. The genes expressed specifically in AM (Figure 4D) included the alveolar macrophage-derived chemotactic factor-II (AMCF-II), the chemokine (C-C motif) ligand (CCL) 24, indoleamine 2,3-dioxygenase (IDO) 1, IL1 beta, the vascular endothelial growth factor (VEGF) A and its receptor (FLT1).
Differential regulation of LPS-responsive genes in AM
PAMPs are recognised by several classes of receptors, including the toll-like receptors (TLR) and intra-cytoplasmic receptor such as nucleotide-binding oligomerization domain-containing protein (NODs) or retinoic acid-inducible gene 1 (RIG-I, also known as DDX58). As the TLRs and NODs are highly polymorphic in pigs at the protein level [19–21], we considered the possibility that they might also be differentially expressed between breeds. However, there was no evidence for differential gene expression amongst the 25 animals surveyed. There was evidence of selective expression in different macrophage populations (Figure 5A). TLR3, 7, 8 and 9 were more highly-expressed in MDM and BMDM. In contrast, TLR4 and TLR2 were highly expressed in AM. TLR6 and TLR1 proteins can both heterodimerize with TLR2 [22] but have different expression. TLR1 was barely detected, but TLR6 was selectively expressed in AM.
We selected the genes significantly regulated by LPS in AM, BMDM and MDM with a p adj. value <0.01 and a fold change >2 or <-2 (Figure 5 B, C respectively). There was again a substantial overlap between these lists in BMDM and MDM. These gene lists include the up-regulation of IDO1, IRG1 CXCL11, CXCL9 and CXCL20. As expected the 3 types of macrophages share significant up-regulation of genes encoding inflammatory mediators such as TNF, CCL20, IL23A, IL27 and G-CSF. A small set of genes induced only in the AM included GM-CSF (CSF2), LIF or IL19. The former is of interest because of the extensive literature on the specific function of GM-CSF in lung macrophage homeostasis [23]. IL19 has also been implicated in lung injury in septic shock [24]. As already shown in Figure 3, AM have a higher basal expression of inducible genes such TNF suggesting that they are primed for an inflammatory response. Indeed, there were only 91 genes significantly up-regulated in AM after LPS stimulation compare to the 1,400 in BMDM and 942 in MDM. The complete list of DR genes in the 3 populations of macrophage after LPS stimulation is listed in the Additional file 3.
Breed-specific variation in macrophage gene expression
One of our goals was to identify the set of macrophage-expressed genes that have diverged under natural and artificial selection in the pig and/or to identify individual variation that might be exploited in breeding for increased disease resistance or tolerance. In addition to Figure 1L where we showed no significant difference in TNF production between the 5 breeds for BMDM and MDM, we used a PCA on the totality of the microarrays and found no obvious clustering depending on the breeds (Figure 6A). In order to identify any breed-dependent difference in the gene expression during inflammation, we analysed all the samples (AM, BMDM and MDM) after 7 hours of LPS stimulation. We applied the stringent test of a p adjusted value <0.01 and a fold change (FC) >3 or <-3 (Figure 6B). The spectrum of differential responses amongst breeds for alveolar macrophages is shown in (Figure 6C). Only a small number of innate immune-related cytokine/chemokines genes were found differentially expressed between the breeds, including IL12A and CSF2 (GM-CSF) which were more abundantly expressed in HAM than in LW and PIE. Some genes encoding innate immune receptors are also DR amongst the breeds. TLR6 had a higher expression level in LR compared to LW and the C-type lectin CD302 expression was higher in HAM than in PIE. In addition, the IL-8 receptor beta, CXCR2, was weakly expressed in the LR breed compare to 3 other breeds (DU, HAM and PIE). Other interesting genes are DR between the breeds. The eukaryotic translation initiation factor 4H (EIF4H) was highly expressed in the Hampshire and Piétrain breed. Swine Leukocyte Antigen (SLA) (i.e. Major Histocompatibility Complex) genes (DOA, DQB2) were more highly expressed in Duroc and Hampshire in comparison to Landrace, Large White and Piétrain. These 3 breeds show a small number of DR genes, as shown in Figure 6B. In all of the comparisons, HAM and DU clustered separately from LR, LW and PIE, consistent with evidence from genomic comparisons [14]. The full list of genes DR between breeds in AM, BMDM and MDM is presented in Additional file 3. The heat-map of probesets DR between breed in MDM and BMDM are presented in Additional file 4.
Comparison with other species: the pig as a convenient model for human disease
To compare pig macrophages with mice and human, we selected 2,505 LPS-regulated genes identified in a previous study [5]. After removing 447 genes for which no porcine orthologous genes have been annotated on the Snowball microarray, we selected 834 genes that were DR between mouse and human after 6 hours stimulation of LPS, with a p < 0.01, in the earlier study (Figure 7A). We then clustered these genes using Biolayout (R=0.85, MCL 2.2 (Figure 7B). These clusters segregate into 2 superclusters. Cluster 1 contains IDO1 (green line) which is highly up-regulated in human and pig macrophages in response to LPS, whereas Cluster 4 contains NOS2A (grey line) which is LPS-inducible only in mouse macrophages. Cluster 1 include genes such as DDX58 (also known as RIG-I). Cluster 4 (168 nodes) includes various genes playing a role in inflammation such as IRAK3, IL12B, IL18, IL6 (Figure 7C). Amongst the 142 genes sharing the profile of IDO1, 33 had a correlation >0.95 (Figure 7D). This set included GMPR, IL7R, SLC25A28, identified previously using a more limited platform and smaller dataset [16]. The mouse-specific pattern exhibited by NOS2A was shared by 143 other genes, 40 of which, including ARG2 or CD86, had a correlation > 0.95 (Figure 7E). The full dataset of correlated expression across species is provided in Additional file 5.
Investigation on the inter-individual differences between pigs
We examined the array data to identify candidate null mutations. In keeping with previous evidence of variable expression [25], we found that one pig had no expression of swine leukocyte antigen 6 (SLA6), the major histocompatibility complex in pigs (Figure 8A) and two Piétrain pigs had substantially lower expression of the cyclin M3 (CCNM3) gene than the rest of the animals (Figure 8B). To identify genes with highly variable expression between individuals, we calculated the average expression and the SED of all probesets for all 6 conditions (3 types of cells and 2 stimulation timepoints). Then, we analysed the percentage of variance (SED/Mean*100) of each probeset. The full list of probesets with their variance is provided in Additional file 6. The majority of genes have a percentage variance < 10.
The variance data are plotted in Figure 8C and D. The genes with highest percentage variance (between 25–50%) are clearly enriched for immune function. In BMDM and MDM the most variable gene is CXCL10 followed by interferon beta, interferon-induced protein with tetratricopeptide repeats (IFIT) 1, DDX58, IDO1, CXCL11, IL7R and IL1RN. In AM, the top genes (>50%) are linked to the immunoglobulin chain, due to the small contamination of B cell from the alveolar lavage. Outside of this set, genes with a variance between 20–50% included IL33, CCR2, IL23A, IGF1, CXCL9 and CXCL10. Genes with a variance > 20% were analysed using the DAVID functional annotation webtool [26] [http://david.abcc.ncifcrf.gov]. For each compartment and time-point (a total of 6), the biological process clusters were given an enrichment score, ranking the overall importance of the annotation term group (Figure 8E). The score of 1.3 is taken as the threshold for functional importance (red line). Clearly, immune response genes are identified in every comparison.