Fig 1.
Discovery of novel feedback and crosstalk structures in a BCR-driven signaling network by perturbation data-based modeling of BL-2 cells.
(A) Systematic perturbation data shown as log2 fold changes to solvent control DMSO. Data was generated by pre-treating BL-2 cells with inhibitors targeting key effectors downstream of BCR for 3h with subsequent BCR stimulation using α-IgM for 30 min (upper panel) or no stimulation as consistency check (lower panel). Phosphorylation of indicated signaling proteins (cf. D) was measured using bead-based ELISAs (mean, n = 3). (B) Modeling workflow using the Modular Response Analysis-based method STASNet to derive a semi-quantitative directed network. The model requires systematic perturbation data (depicted in A) and a curated literature network as starting network (cf. D). In order to avoid overfitting, the data was split into two parts: (1) α-IgM stimulated data was used for parameter fitting and network adjustment and (2) unstimulated inhibitor data was used for verifying model consistency. After each network adjustment step the unseen data part was simulated and compared. If the error reduction as compared to the null model was not significantly worse, the new network adjustment was upheld otherwise the next best solution was simulated and tested. (C) Model performance for each modeling step from the literature-derived starting model to the final model: (TOP) goodness of fit as weighted sum squared residuals divided by number of free parameters and (BOTTOM) consistency check step as percentage of error reduction compared to unperturbed control as null model (see S1 Text BL-2_network_model.html: Tab ‘Network derivation BL-2’). (D) Literature network adjusted to the final signaling network for BL-2 cells derived by the modeling pipeline depicted in B. grey line/text—removed links/nodes; green line/text—added links.
Fig 2.
BL-2-derived modeling structure can be transferred to cell line BL-41.
(A) Systematic perturbation data for BL-41 cells generated alike the procedure described in Fig 1A (mean, n = 3). (B) Model development statistics (TOP) goodness of fit as reduced chi-square and (BOTTOM) unseen data consistency check as percentage of error reduction compared to unperturbed control as null model for each modeling step from the literature-derived starting model (black and grey arrows in Fig 1D) to the final model (grey–reduction, green—extension). See S1 Text BL-41_network_model.html: Tab ‘Network derivation BL-41’. (C) Venn diagram indicating the shared and not shared structural adjustments in the development of BL-2 and BL-41 cells starting from the same literature network (cf. S3 Fig). (D) Model fit and consistency check statistics for fitted models on BL-2 and BL-41 perturbation data for three different network structures: literature, cell-specific adjusted network (adjusted) and for the best-found structure of the respective other cell line (transfer). See also S2 and S4 Figs. (E) Network coefficients heatmap from models fitted to the BL-2 learned structure for the indicated cell lines. Comparability was ensured by fixing the inhibitor coefficients to BL-2-learned values as both cells received the same inhibitor doses. Stars denote coefficients that are significantly different (i.e., 95%-point wise confidence intervals do not overlap, see S1 Table). (F) Data excerpt for the model-derived negative crosstalk prediction from p38 to RAF/MEK/ERK pathway in BL-2 and BL-41 cells showing the upregulation of α-IgM-induced activation of pERK and pMEK by the p38 inhibitor SB203580 (mean ± s.e.m., n = 3), but no upregulation by p38 inhibitor alone.
Fig 3.
Increased MEK/ERK-pathway activity in BCR-activated B cell lines after p38 intervention.
(A) Changes in the phosphorylation of MEK and ERK in BL-2 cells after treatment with α-IgM in the presence or absence of the p38 inhibitor SB203580. (B) Phosphorylation of ERK is further increased in α-IgM-treated BL-2 cells after 24h of p38α (MAPK14) knockdown. (C) Phosphorylation of ERK is enriched within the nucleus of α-IgM treated BL-2 which is further enhanced by inhibiting p38. Tubulin and HDAC1 were used as reference for the cytosolic and nuclear fraction, respectively. (D) (TOP) Phosphorylation of c-RAF at serine-residues 289/296/301 is increased after 30 min α-IgM treatment in BL-41 cells but not affected by p38 inhibition. The inhibition of MEK using AZD6244 interrupts the phosphorylation of RAF. Representative Western blot. (BOTTOM) Bar plots quantifying c-RAF phosphorylation measurements for n = 2 replicates. (E) p38 affects the MEK/ERK pathway in a comparable way in different BL cell lines after α-IgM treatment. Shown are phosphorylations of Raf1 at serine 338 (activatory site) and 289/296/301 (ERK feedback sites) as well as of ERK and p38 in Burkitt lymphoma cells BL-2, BL-41 and CA-46.
Fig 4.
Phosphoproteomics analysis supports the BCR-signaling model and reveals a dominant effect of the PI3K pathway inhibition onto BCR-signaling in BL-2 cells.
Analysis of Tandem-Mass-Tag (TMT) Mass spectrometry measurements for BL-2 cells treated with α-IgM and inhibitor solvent DMSO or inhibitors of PI3K (BKM120), MTORC1 (Rapamycin) or p38 (SB203580) (n = 2). (A) Hierarchical clustering of 3000 most varying phosphosites demonstrates a global effect of α-IgM and PI3K inhibitor BKM120 on the phosphoproteome and subtle effects of the remaining inhibitors. (B) Principal component analysis shows that α-IgM effect is governing the principal components 1 and 2. Only PI3Ki treatment is able to partly revert the α-IgM effect. (C) Overlap of differentially regulated phosphosites (limma, FDR≤5%) for indicated selected comparisons. (D) Upstream kinase activity assessment on base of log2 fold changes (vs. α-IgM+DMSO) in PhosphoSitePlus-annotated target sites (Nov 2021) for selected kinases. Significance asserted by two-sided t-test: ns—not significant; * - 0.05; ** - 0.01; *** - 0.001; **** - 0.0001); Average value indicated. (E). Phosphosites significantly regulated by p38 inhibitor SB203580 (limma, FDR≤5%). Left panel denotes which site was found to be significantly regulated (blue—down; red–up) by the indicated comparison. Sites are annotated as follows: ‘HGNC symbol’_’amino acid’_’position’_’number. of phosphosites’; ERK activation sites and known target sites of ERK [55] are indicated by green and orange circles, respectively.
Fig 5.
BL-2-derived network structure sets a veritable starting base to develop networks for DLBCL cell lines HBL-1 and OCI-LY3.
(A) Systematic perturbation data of bead-based ELISA measurements of the DLBCL cell lines HBL-1 and OCI-LY3 quantified as log2 fold changes to solvent control (DMSO); mean of n = 3. (B) Goodness of fit expressed as reduced chi-square statistic Xr on selected network structures for the two DLBCL cell lines HBL-1 and OCI-LY3. literature–network from Fig 1B; BL-2 –BL-2 network derived from Fig 1C; adjusted from BL-2 –BL-2 network locally adjusted to respective DLBCL cell line; OCI-LY3/HBL-1 –final adjusted network of respective other DLBCL cell line (see C). (C) Network structures of BL-2 derived starting network and final DLBC-specific networks trained on HBL-1 and OCI-LY3 data (see A). (D) Side-by-side comparison of the model coefficient (path)s (log scale) with fixed inhibitor strengths set to mean of both cell line models. Empty tiles indicate missing links in one of the cell lines (individual links). Asterisks point to non-overlapping confidence intervals as estimated by STASNet profile likelihood function (see S2 Table).
Fig 6.
Conserved core network of chronic and acute B cell receptor signaling.
Consensus network structure and coefficient quantification of links present in at least 3 of the 4 modelled cell lines with chronic (HBl-1, OCI-LY3) and acute (BL-2, BL-41) BCR signaling. BL-2 and BL-41 information retained from models trained on the BL-2 structure and with inhibitor strength fixed to BL-2 model (Fig 2E and S1 Table), the connection for GSK3 to Btk was retrieved by multiplying the coefficients r_ZAP70_GSK3A.B with r_Btk_ZAP70. HBL-1 and OCI-LY3 information retained from their respective adjusted model when started from the BL-2 network with inhibitor strength fixed to average of both cell line models. (Fig 5D and S2 Table). Note that the coefficients for ERK->RAF, p38-> RAF and Btk->RAF all occur in combination with the RAF->MEK coefficient so that the modelled strength represents the response on MEK instead of RAF.