Over past decade, transcriptome profiling efforts that employ either microarray or RNAseq have already generated enormous amounts of data, with respective data analysis often only scratching the surface [16, 17]. In many cases, high-quality datasets are generated to investigate specific hypothesis, and consequently, these datasets get analyzed in a particular way. At least in theory, the study design of these narrow-set, but technically sound experiments should allow extraction of additional information that could remain unrecovered at the moment that the main manuscript gets sent to the publishers [18].

In their 2009 paper, Pazolli et al. started to investigate the mechanisms of the manner in which senescent BJ fibroblasts stimulate the growth of preneoplastic cells in vitro and in vivo [8]. In their experiments, replicative senescent (RS) and stress-induced premature senescent (SIPS) fibroblasts were equally proficient at inducing the growth of HaCaT cells. Their study of fibroblasts/HaCaT xenografts in vivo arrived essentially at same results [8]. The authors subsequently hypothesized that growth-promoting activities of both types of senescent cells are maintained by a common core of genes. Based on that hypothesis, they embarked on microarray-driven dissection of secreted factors commonly produced by RS and SIPS fibroblasts. After a set of validation experiments in qRT-PCR and in-cell cultures, soluble protein osteopontin was highlighted as the protein of functional importance, and its gene, SPP1, was identified as a master regulator of a cancer niche environment [8]. An objective of the study achieved; however, the microarray dataset never got analyzed in larger context, i.e. for the purpose of direct comparison between RS and SIPS drivers.

In this study, we used the dataset of Pazolli et al., 2009 to extract the differentially expressed genes (DEGs) that differentiate the processes of RS and SIPS, to reconstruct relevant molecular cascades and to gain additional insights into popular cellular model of bleomycin induced senescence. Analysis of signaling events indicated that an involvement of caspase-3/keratin-18 pathway that is indicative of apoptotic rather than necrotic cell death [19] and an evolutionarily conserved serine/threonine kinase Aurora A/ MDM2 pathway essential for mitotic progression [20] was shared between both types of senescence. Observed upregulation of Aurora A is consistent with previously demonstrated increase in a number of aneuploid cells observed in ageing fibroblast cultures [21]. Our analysis also highlighted concerted alteration of glutamate, polyamine and choline metabolisms as well as wnt/β-catenin and SDF-1/CXCL12 cascades. All these findings are generally consistent with previous studies of various ageing fibroblasts both in culture and in human cohorts [22–24]. This consistency prompts us to stress on the high quality of the dataset of Pazolli et al., 2009 being analyzed.

An analysis aimed at identifying master regulator molecules that influence the replicative senescence and bleomycin–induced senescence expression programs, pointed at stromelysin/MMP3 and N-acetylglucosaminyltransferase enzyme MGAT1 that initiates the synthesis of hybrid and complex N-glycans as key orchestrating components in replicative senescence and in bleomycin-induced senescence, respectively (Fig. 2 and Additional file 7). Traditionally, MMP3 is seen as end-point biomarker or effector molecule associated with ageing in fibroblasts. However, in Hutchinson-Gilford progeria syndrome, there is a progressive loss of MMP3 mRNA and protein expression [25]. Another study linked carrier status for MMP3 6A (rs3025058) allele to skin and lung aging [26]. Moreover, an exposure to MMP3 stimulates expression of Rac1b, a tumor-associated protein with cell-transforming properties that aids in bypassing replicative senescence [27] while driving motility and protumorigenic responses of the stroma [28]. Hence, there is an accumulation of evidence that stresses on an importance of MMP3 as a molecule of importance in replicative senescence that deserves additional investigations. An identification of MGAT1 that controls the synthesis the complex N-glycan sugars in the Golgi as the key regulator of SIPS is even more intriguing as there is strong associations between human plasma N-glycans and age [29, 30].

Specific question that we aimed to dissect was on the differences of the senescence programs executed in SIPS and RS. Indeed, our analysis showed that in RS fibroblasts, the list Top deregulated events is populated by fragments of Aurora-B driven cell cycle signaling that are accompanied by the suppression of anabolic branches of the fatty acids and estrogen metabolism. This may be interpreted as an execution of ordered senescence program that proceeds along with shutting down the metabolism on a way to the halt of mitotic progression and apoptosis that is being upregulated in both RS and SIPS. On the other end, in bleomycin exposed fibroblasts, Aurora-B signaling is deprioritized and the synthetic branches of cholesterol metabolism are upregulated, rather than downregulated, while proteasome/ ubiquitin ligase pathways of protein degradation are dominating the regulatory landscape. This picture is indicative that the cells are going down actively fighting overwhelming amounts of stress that is facilitating premature senescence of cells, but fail to completely activate orderly program of replicative senescence. Latter observation is consistent with activation of 26S proteasome and enhanced protein polyubiquitination previously observed in both idiopathic and bleomycin-induced pulmonary fibrosis [31]. Generalized mechanistic depiction of cellular processes common and differentiating RA and SIPS is presented at Fig. 2.

The list of the transcription factors capable of binding within the promoter regions of the genes that change their expression in either RS or SIPS was unusually enriched by the members of homeobox family, with particular emphasis on HMX1, IRX2, HDX and HOXC13. The possibility of an involvement of homeobox genes in ageing has been proposed earlier [32], with many homeobox containing TFs included in manually curated GenAge reference database [33]. Our findings indicate that the senescent program may be orchestrated by transcription factors (TFs) of Homeobox family at least in case of replicative senescence in vitro. On the other hand, promoters of genes that change their expression in bleomycin-induced senescence but not in replicatively senescent fibroblasts were enriched by binding sites for transcription factors Ikaros, RelA, HNF3B, GKLF and MAZ. Both RelA and GKLF are known stress-induced transcription regulators. RelA is the central player in the classical (or canonical) pathway of induction of NF-κB subunits that promotes senescence when activated in human lung fibroblasts exposed to ROS [34]. GKLF-deficient fibroblasts exposed to excessive levels of reactive oxygen species are more prone to become prematurely senescent than normal fibroblasts [35]. Moreover, yet another transcription factor, HNF3B/FOXA2 is epigenetically silenced in peroxide-stressed fibroblasts [36], therefore, an enrichment for binding sites for this factor in transcripts downregulated in bleomycin induced senescence is not surprising.

SPP1-encoded osteopontin, a secreted stromal driver for tumor growth, is overexpressed by both RS and SIPS fibroblasts [8]. Concise network constructed using Shortest Path function in Pathway Studio software (Fig. 3) highlighted Iroquois Homeobox 2 (IRX2) and POU4F1 were highlighted as most likely signaling events to connect the DEGs identified by GeneXPlain-guided microarray analysis and osteopontin. In this network, suppression of Aurora kinases that normally monitor the mitotic checkpoint, centrosome separation and cytokinesis, cause catastrophic consequences and result in increase in apoptosis, thus, being in in agreement with recently published observations of senescent fibroblasts [37]. Apoptosis activated caspase-3 directly or indirectly eliminates POU4F1/Brn-3a, the prediction that is consistent with previous observation of enhanced apoptosis in the neurons derived from Brn-3a knockout mice [38]. Moreover, POU4F1 gene is expressed in fibroblasts where it is required for proliferation, and cooperates with activated RAS/RAF signalling by reducing oncogene-induced senescence, consistent with its caspase-driven downregulation in both RS and SIPS [39]. In our network, POU4F1/Brn-3a suppresses transcription factor IRX2 that repeatedly showed up in lists of TF that recognize bindings sites differentially enriched in promoters of genes associated with fibroblast senescence. Caspase-3-driven removal of POU4F1 allows higher levels of IRX2 biosynthesis that is known for its ability to upregulate VEGF, metalloproteinases and other secreted molecules [40, 41].

An involvement of IRX2 in the transcription of osteopontin-encoding SPP1 gene was never evaluated in wet lab experiments; however, the knowledge-based algorithm identified IRX2 as positive regulator of SPP1 expression by three independent molecular interaction events involving AKT1, VEGFA and INS. Moreover, marker co-expression pattern of IRX2 and SPP1 was observed during hair-cell development in the chick’s cochlea [42]. Two independent studies demonstrated that an expression of IRX2 is commonly suppressed by DNA methylation of its promoter [43, 44], including its differential methylation noted in osteoarthritis and osteoporosis [45], two age-related diseases of the cartilage and the bone charactrized by changes in the levels of osteopontin secretion [46, 47]. As IRX2 is strongly expression in human primary osteoblasts of the skeleton [48], its putative roles in SPP1 regulation in osteoarthritis and osteoporosis are worthy of investigation.

Importanttly, Pazolli and co-authors followed up on their own study that identified osteopontin as driver of tumor cell proliferation supplied by senescent stromal fibroblasts [8] and showed that the treatment with histone deacetylase (HDAC) inhibitors that reverse CpG methylation is sufficient to induce expression of osteopontin [49]. Moreover, an examination of PWM matches in the promoter of SPP1 showed that it contains 25 sites for IRX2 binding within 1100 nucleotides located between positions −1000 to +100 relative to major transcription start site (TSS) for SPP1 gene (Additional file 8).

All this evidence adds up in favor of the hypothesis that SPP1/osteopontin expression may be controlled by IRX2, and that its derepression in senescent fibroblast aids in SIPS-dependent stromal activation that, in turn, stimulate the growth of tumor cells.