In the present study we evoked LPS induced inflammation in adipose and liver tissues in vitro in order to mimic IR caused by inflammation in vivo. We aimed to compare the changes in the inflamed transcriptomes and secretomes of both tissues in order to (1) better understand contribution of the inflamed adipose and liver tissues to the development of insulin resistance and (2) to identify candidate biomarkers indicative for tissue specific inflammation/IR.
The gene functional classification analysis revealed that both adipose and liver tissue share common response mechanisms that are activated during inflammation (chemokine signaling, (anti)apoptosis, extracellular matrix remodelling, adhesion and migration of different immune cells involved in inflammatory reactions). Although functional clustering led to identification of the same functional groups, both tissues had a different set of genes within one functional group, suggesting tissue-specific inflammatory signaling. The significantly upregulated adipose tissue transcriptome contained additional gene functional categories belonging to SOCS and several transporters (Additional file 1, Table S1). The SOCS signaling was shown previously to be involved in induction of insulin resistance during acute inflammation in human adipose tissue  and our ex vivo data are in line with these in vivo findings. The analysis of the down regulated functional groups pointed out towards redox/detoxification processes affected in both tissues and mitochondrial functions observed in liver tissue. These processes could contribute to the enhanced reactive oxygen species (ROS) production recognized as one of the mechanisms implicated in the development of IR/T2D . Furthermore, adipose tissue had downregulated genes involved in the extracellular matrix activity which is involved in multiple processes including modulation of immune responses. In liver tissue downregulation of genes involved in amino acid metabolism and polysaccharide binding were observed. There are reports about changed amino acids concentrations in animal models of obesity and obese humans [24, 25], however interpretation of this ex vivo finding in relation to these reports is not unequivocal.
The additional network identification for the common (overlapping) and differential adipose and liver tissue transcriptomes was in line with the data obtained from the gene functional analysis and distinguished the common and differential networks. Moreover, several of these networks were described previously in the literature for their role in induction of IR thereby supporting our model system to study the inflammation related insulin resistance in vivo. For example, in our study we found upregulated chemokine signaling and matrix remodelling in both adipose and liver tissues which were also previously linked to the development of IR in vivo [26, 27]. SOCS signaling is implicated in induction of IR [28–30] and similarly it was found by us to be upregulated by LPS in adipose tissue ex vivo. The decreased PPARγ expression in adipose tissue is recognized as one of the events associated with IR and occurred in our ex vivo studies as well [31, 32]. Similarly upregulated Jak-STAT and NFκB signaling identified previously in IR liver  was present in our experiments.
The GO analysis and gene count revealed that adipose tissue had more LPS-induced upregulated GO terms and genes related primarily to "inflammation", "angiogenesis", and "development". Moreover, the predicted secretome studies showed that the adipose tissue predicted inflammatory secretome is more abundant compared to the liver tissue secretome. This observation indicates that adipose tissue is more active during inflammation, compared to liver tissue, and supports the hypothesis that adipose tissue plays the major role in the development of inflammation-related IR .
The reason for different responses of the adipose and liver tissues could be due to a different expression of TLR4 and other components involved in signal transduction via TLR4 (LBP, CD14, TREM1), but unfortunately in our studies we can not directly compare expression values between the adipose tissue and the liver data (two different DNA microarray platforms were used). Nevertheless, we observed that the expression patterns/ratios of all the TLR4 signaling molecules in both tissues were very similar (data not shown).
The predicted secretome analysis
The microarray data analysis of both tissues revealed that adipose and liver tissues have numerous overlapping LPS-responsive genes which protein products are predicted to be secreted. Among these genes we identified several known markers associated with insulin resistance such as IL-6, IL-1β, IL-8, and PAI 1. Other proteins known to be upregulated during insulin resistance by adipose tissue  such as RANTES, MCP1, PLAUR, CXCL5, were found in our studies to be upregulated in both adipose- and liver tissues. Additionally, in both tissues we found genes, previously shown to be regulated in adipose tissue in relation to insulin resistance: CXCL1, CXCL10, CXCL11, ICAM1, TNFAIP6 , FGF2, IL6 , and ICAM1, IL-1 . Although TNFα is known to be involved in the development of insulin resistance in both adipose tissue and the liver, it was only significantly upregulated in adipose tissue. However, we observed that 3 out of 5 livers had upregulated expression of TNFα and previously we showed that in liver tissue in vitro, TNFα mRNA level was significantly upregulated after 5 hrs while after 24hrs the TNFα mRNA level returned to basal values [15, 37]. In order to explain this phenomenon we hypothesized that the TNFα response after LPS treatment could be related to number of Kupffer cells (assessed by CD68 expression) or to the expression of TLR4. Thereby, we looked at correlations between TNFα expression and both CD 68 and TLR4. There was no correlation between TNFα expression and CD68, R2 = 0.0063 (data not shown). The correlation between TNFα and TLR4 indicated on a good positive correlation (R2 = 0.4) between these genes and it could indeed explain the observed differences (data not shown).
Furthermore, the comparative analysis of adipose and liver tissues secretomes in vitro provides a source of candidate biomarkers related to tissue specific inflammation/insulin resistance. Similarly to Shah et al. , we identified in the inflamed adipose tissue secretome genes such as: SELE, CD274, ORM1, PLA1A, SLAMF1, CX3CL1, OSM, TNF, C19ORF59, PTX3, IER3, CCL8, CXCL2, SERPINE1, BMP2, FAM107A, GPX3. Moreover, we identified genes of yet unknown functions such as: C14ORF162, C20ORF59 or genes implicated in other than insulin resistance inflammatory diseases: epiregulin, IL-19 or sarcoglycan [38–40].
The analysis of the predicted secretome of inflamed liver tissue revealed several significantly changed genes with a known- and an unknown- relationship to insulin resistance. Identification of biomarkers indicative for inflamed liver tissue could be a useful tool in a diagnosis of NAFLD patients, where the only "golden standard" is an invasive liver biopsy . Biomarkers previously associated with liver diseases and identified in our samples were among others: ANGPTL3, IGFBP2, SDC4, IL1RN [7, 42]. Examples of other pro-inflammatory proteins affiliated with inflammation but not liver insulin resistance were cathepsin S  or granzyme A . In future it has to be validated if the other most differentially regulated genes between both tissues such as: SGCD, LCE3D, EREG, NDP and CXCL9, FSTL3, PDZK1IP1 could be used as biomarkers related to insulin resistance of adipose or liver tissues respectively.
Comparison of transcriptomics and proteomics data
Finally, the transcriptome data encoding for the adipose tissue inflammatory secretome was validated and compared with the protein data of the inflamed adipose tissue culture medium. The analysis showed that the transcriptome data were in line with the proteomics data, in respect to observed upwards and downwards fold changes (FC) for genes and their corresponding protein products. However, the FC derived from the proteomics experiment cannot be directly compared with the FC of the transcriptome experiment due to substantial technical differences between both technologies. By combination of the comparative transcriptome analysis and proteomics technology we identified leukemia inhibitory factor (LIF), matrix metalopeptidase-1(MMP-1), pentraxin related gene product (PTX3), fractalkine (CX3CL1), and PAI 1 as the potential set of biomarkers for the inflamed adipose tissue.
Chronic LIF exposure in cardiomyocytes was linked to insulin resistance , however the role of LIF in induction of IR in adipose tissue is not known. Proteins of the matrix metalloproteinase (MMP) family are involved in the breakdown of extracellular matrix in normal physiological processes, such as embryonic development, reproduction, and tissue remodelling, as well as in disease processes, such as arthritis and metastasis [45–47]. MMP1 was not reported earlier as a biomarker of IR/T2D and its role in adipose tissue is not known.
Pentraxin related gene (PTX3) plays a role in innate immunity, inflammation, vascular integrity, fertility, pregnancy, and also in the central nervous system. The PTX3 could influence the development of autoimmune reactions and vascular disorders in humans [48, 49]. Recently pentraxin was also associated with obesity and metabolic syndrome [50, 51] and it was shown to be secreted by adipocytes .
Moreover, very recently CX3CL1 (fractalkine) was proposed as a novel human adipochemokine associated with T2D in humans .
Other proposed by us candidate biomarkers such as TNFα and SERPINE 1 (PAI) are commonly associated with inflammation, IR, and T2D and are known to be secreted by the stromal vascular fraction of adipose tissue .
In summary, based on the obtained data we postulate that during inflammation related to IR the target peripheral tissues (adipose tissue, liver) secret a set of unique proteins which could serve as tissue-specific biomarkers related to the investigated pathology. We believe that our approach of using multiple biomarkers could result in more specific diagnosis for a tissue specific insulin resistance related to inflammation, than the use of single biomarkers.
One of the shortcomings of our study is the use of two different DNA microarray platforms, since the data used here were generated in two different laboratories. However, previous studies comparing human Affymetrix and Illumina platforms show that the obtained results, using the same human material, are highly comparable, especially for genes which are predicted to be differentially expressed . Furthermore, in our studies we compared only genes which were significantly affected and present on both platforms; therefore genes which were not present on both platforms were excluded from the analysis and we did not compare intensities of corresponding genes since they would be different due to the platform specific design. Another possible disadvantage of our studies is application of patients with different gender (the adipose tissue was obtained form females and the liver was derived form males), BMI, age, and other anthropometric and biochemical parameters. However, due to limited access to human tissues we could not control all the parameters according to the proper experimental design. Nevertheless, we are confident that the results presented provide a good basis for future in vivo validation studies.