Stimulation with TNFα
Regarding stimulation with TNFα, a coordinated down-regulation with of the two TF SMAD7 (inhibitor of TGFβ1/SMAD4 signalling) and ETS1 emerges, as indicated by the rules 33, 114, 135, 144, 157, and 186 (see additional file 8). For example, rule 186:
<90> COL1A1.out.off, EST1.out.off → SMAD7.out.off
has the meaning: in all simulated and observed states characterised by the absence of COL1A1 and ETS1, SMAD7 is also off. <90> stands for the support of the rule, i.e., the number of transitions (90 out of 294) that actually have the attributes of the premise. Rules 114, 135, 144, 157, and 186 indicate: if the TNFα-dependent genes are not induced (ETS1 as mediator), then simultaneously the expression of TGFβ1-dependent genes is enabled (SMAD7 is off). This suggests that TNFα and TGFβ1 may act as antagonists in SFB, as described in [55, 56].
The expression of NFKB1, which is also induced by TNFα, proceeds conversely to that of ETS1 and SMAD7 (rules 34, 45, 70, 71, 134, 144, 154, 157, and 173) reflecting the different targets of NF-κB and SMAD7. The antagonistic pattern of NFKB1 and SMAD7 appears indirectly in rule 33, where the two genes show up in the premise of a rule with high support:
<150> (...) NFKB1.out.on, SMAD7.out.off → EST1.out.off
Regarding this rule, it is interesting that ETS1 always acts in the same direction as NF-κB, according to the network derived from the literature (Figure 1). In the adapted network (Table 4), we assumed a necessary cooperation (i.e., an AND connective) for the positive regulation of ETS1, MMP1, MMP3, MMP9, and TNF, as well as for the inhibition of COL1A1 and COL1A2. Thus, rule 33 further suggests that the coordinated action of NF-κB and ETS1 is turned off in states which are characterised by supplementary conditions as SMAD7.out.off.
The generated rules adequately reflect the major influence of the TF AP-1 in the system: the expression of prominent targets, such as COL1A1, MMP1, and MMP3, depends on JUN (rules 211 and 258) and/or FOS (rule 204), with JUN as the key player. These rules connect input and output states and thus their semantics is directly related to dynamics, as seen in rule 211:
<87> TGFB1.in.on, TIMP1.in.on, ETS1.in.on, JUN.in.on → MMP1.out.on
making this strong statement: if ETS1 and JUN are on, MMP1 will always be up-regulated in the future (at least within the time frame of 12 hours).
Sometimes, MMP1 is expressed simultaneously or before ETS1 and JUN. In the simulation, MMP1 was always on in the output state and from time point 2 h in the data. An exception can be found for the experimental results from OA sample OA3 (Table 5), where MMP1 is off after 12 h. This is the reason for the computation of the auxiliary conditions TGFB1.in.on and TIMP1.in.on in rule 211.
Concerning the regulation of target genes, the expression of MMP1, MMP3, and MMP13 is co-regulated (rules 35, 63, 82, 86, and 176), while MMP9 is expressed independently (rules 24 and 35). There is a contradiction between the simulation and the data: in the observed experimental time series, MMP13 is always off, whereas the Boolean network predicts an up-regulation similar to MMP1 and MMP3. This unexpected absence of predicted MMP13 expression may be an indication for a more complex regulation of MMP13 transcription, exceeding the already known regulatory interrelations. Therefore, the MMP13 promoter and further enhancer/repressor sequences should be targeted for a more pronounced structural and functional analysis. For MMP9, the simulation and the experimental data are in good agreement: the gene is off in most, but not all states. However, since the expression of MMP9 by (S)FB is discussed controversially in the literature (see  and vs. ), the calculated expression of MMP9 by fibroblasts – at least at a limited number of time points – supports the majority of studies, reporting detectable MMP9 mRNA amounts in (S)FB.
Several rules unanimously indicate the co-expression of the ECM-forming genes COL1A1 and COL1A2 (rules 87, 88, and 95), but contradictory rules occur concerning their expression profile in comparison to the MMPs. COL1A1 and COL1A2 seem to be co-expressed with MMP1 (rules 90 and 176), for COL1A2, however, a certain co-expression with MMP9 is calculated as well (rules 76 and 77), which conflicts with the opposing expression of MMP1 and MMP9 (see above). Therefore, the expression of collagens does often, but not necessarily always correlate with the expression of MMPs. This reflects the imbalance between MMP-dependent destruction and collagen-driven regeneration/fibrosis of ECM in the joints in inflammatory RA.
The calculated knowledge base also contains a further unexpected correlation. According to rule 166:
<94> FOS.in.off, TIMP1.in.on, SMAD7.out.off → TGFB1.in.on, MMP1.out.on, TGFB1.out.on
and rule 188, the expression of MMP1 may also be induced in the absence of FOS (e.g., by JUN-containing AP-1 complexes), indicating that the regulation of MMP1 does not predominantly depend on FOS as proposed in the literature [17, 60]. This result may point to the influence of other TFs, e.g., NF-κB, ETS1, or AP-1 complexes containing JUN, which may indeed be able to induce target gene expression in the absence of FOS.
Stimulation with TGFβ1
For the stimulation with TGFβ1, we had a total number of 341 transitions. The SMADs play a major role for the expression of TGFβ1-dependent target genes, as reflected by various classes of rules containing SMAD4 and/or SMAD7 (see additional file 8). For example, SMAD4 can be involved in the expression of COL1A1, see rule 15 (and also rules 21, 26, and 30):
<239> ETS1.out.off → SMAD4.in.on, COL1A1.out.on, SMAD4.out.on
This also suggests an antagonistic behaviour of ETS1 and SMAD4: if ETS1 was off, then SMAD4 was on, as well as in all previous states. Rules 52 and 57 suggest a dependency of MMP1 on SMAD4. However, this seems to be one amongst many other influences (or could be a non-influencing coincidence), since SMAD4 was permanently on during simulation and experimental stimulation with TGFβ1 (exception: sample RA3 at time point 2 h).
The expression of MMP9 is neither induced by SMAD4 (rules 7, 24, and 41) nor by any other TF, indicating that MMP9 is not influenced by TGFβ1. The fact that TGFβ1 obviously does not induce MMP9 (but other MMPs) agrees with findings reported previously  and represents a clear contrast to the MMP expression profiles following TNFα stimulation.
A further case of an antagonistic expression pattern was calculated for MMPs and COL1A1 (rules 21, 30, 36, 41, 54, and 60), for example, in rule 54:
<170> SMAD4.in.on, MMP3.out.off, MMP9.out.off, MMP13.out.off, ... → COL1A1.out.on
Antagonistic expression profiles also can be observed for SMAD4 and other TFs, e.g., JUN and JUNB (rules 12, 39) or ETS1 (rule 15, see above). The variety of TF combinations found, even following the same stimulus, exceeds the possibilities of conventional TF studies because stimulation experiments are generally restricted to a selected set of readout parameters (e.g., the expression of single TFs or target genes) which are not able to reflect the multiplicity of different effects in the cell.
Following stimulation with TGFβ1, interestingly COL1A2 appears to be constitutively expressed since its status is always calculated as on (rule 1). Therefore, for the formation of collagen I, which contains COL1A1 and COL1A2 chains, COL1A1 expression seems to be the critical switch.