Based on the existing knowledge on CLL microenvironment and using systems biology approaches, we have identified in silico compounds with a potential to counteract the CLL microenvironment. Then, we screened and validated some of these compounds in CLL cells using in vitro assays. The analytical workflow consisted of three interconnected parts: data sources, in silico analysis and in vitro screening are summarized in Figure 1.
Identification of key molecular enclaves involved in CLL microenvironment
A molecular description of CLL microenvironment was obtained by the identification of effector proteins with a known role on CLL microenvironment according to PubMed publications and a manual review of the literature. A total of 139 unique proteins (Additional file 5) were selected.
The known functional associations of these proteins with other proteins where retrieved from public data bases and we built a network around the known proteins related with CLL microenvironment effect. The contributions of close neighbor proteins on the effector proteins were also taken into consideration. To narrow down the target area of analysis a combination of different systems biology measures based on the human protein functional microenvironment network was used. Firstly, we used the ANN11 based model to identify effectors that connect with most of the other effectors. Proteins that are closer to a higher number of other proteins in the microenvironment motive have more chances to be a good target. Secondly, we used a mathematical modeling strategy based on sampling methods  to select those with higher impact on the whole response when their action in disease stage was reverted. The combination of both analyses has identified a subset of 57 proteins, from now on referred as “key proteins” (Figure 2a and (Additional file 6).
Compound library selection
The same network-based mathematical model of ANN  was used to select compounds with the best target combination to affect the larger number of key proteins or effectors of the CLL microenvironment. Compounds were selected from BindingDB and DrugBank databases based on three criteria: i) compounds with a predictive score> 78 and a p-value lower than 0.05 using the ANN algorithm , ii) linkage to a drug supplier and iii) among the compounds fulfilling the previous criteria with the same target profile, the one with the best reported binding constant was selected. Finally, the number of compounds moving forward to the phenotypic screening was 65 (54 bioactive compounds and 11 drugs) (Additional file 1). The number of targets affected by the selected compounds is represented in Figure 2b.
Compound library screening in CLL cells
The CLL HG3 cell line and the human bone marrow stromal HS5 cell line were used to test the effect of the 65 selected compounds. MTT analysis was performed after incubation of cells for 48 h with 15 µM of the different compounds (Figure 3a). Only 8 compounds were affecting metabolic activity (color dot) taking into account a threshold of 50% determined by a ROC analysis (Additional file 4). Cytotoxicity was also analyzed by Annexin-V/PI staining in HG3 alone or in coculture with the stromal cell line HS5 (Figure 3b). Six compounds were cytotoxic for HG3 cells using a threshold of 20% determined by a ROC analysis (Additional file 4). One of these compounds (A5) was a false positive due to autofluorescence. Similar results were obtained when HG3 cells were cocultured with HS-5 cells (Figure 3b). Two compounds (A1 and A2) exerted a cytostatic effect (MTT analysis) but with no effect on viability (Annexin V/PI staining). These results were also confirmed in the MEC-1 CLL cell line (Additional file 7). Then, a dose response (1 to 15 µM) was performed in the HG3 and HS5 cell lines with the 8 selected compounds. At all concentrations used, the CLL cell line was more sensitive to these compounds in a dose dependent manner by MTT (Figure 3C) or Annexin-V analysis (Figure 3d) than the stromal HS-5 cell line. Compound D1 exerted a high cytotoxic effect in all cell lines even at the lowest doses used.
Compound library screening in primary CLL cells
A dose-response screening of these 8 selected compounds was performed in primary CLL cells with doses ranging from 0.1 to 2.5 µM. Compounds C5, C7, D1 and F1 exerted a significant dose-dependent cytotoxic effect (Figure 4a). Compounds A1, A2, A12 and C11 did not exert any cytotoxic effect regardless of the doses used. Then, these 4 active compounds (C5, C7, D1 and F1) were analyzed in primary CLL cells in the presence of HS-5 in order to mimic the microenvironment. Compound D1 was discarded for not being selective for tumoral CLL cells, as a high cytotoxic effect was observed on HS-5 cells. Compounds C7 and F1 were selective for CLL cells even in the presence of HS-5 cells at all tested doses. In contrast, compound C5 only exerted a significant (p<0.05) and selective effect at the dose of 2.5 µM (Figure 4b).
We also analyzed if these compounds had any effect on CLL proliferation. CFSE-labeled primary CLL cells were induced to proliferate by incubating them with a medium containing the CpG oligonucleotide ODN2006, which triggers growth and cell division in the proliferative centers of CLL patients, and the inflammation-linked cytokine IL-15, which is constitutively produced by stromal cells  for 6 days. As shown in Figure 4c, ODN2006 plus IL15 induced an increase of CFSElow viable CLL cells indicative of increased cell proliferation. All compounds tested (C5, C7 and F1) decreased the percentage of CFSElow viable CLL cells, indicating that these compounds induced a significant reduction on CLL proliferation. Ibrutinib 0.25 µM was used as a positive control to inhibit proliferation of CLL cells under these conditions. In order to analyze if the cytotoxicity effect of these compounds was specific for CLL cells, we incubated PBMCs from healthy donors with these compounds at the same doses used in primary CLL cells. The effect on normal B (CD19+) and T (CD3+) lymphocytes was analyzed by flow cytometry. Compounds C7 and F1 were selective for CLL cells at all doses used. In contrast, compound C5 lost the selectivity at 2.5 µM, the highest dose tested (Figure 4d).
Target validation of selected compounds
Two compounds were selected C7 and F1. One of the main targets of compound C7 was NOD1 (nucleotide-binding oligomerization domain-containing protein 1) (Additional file 8). NOD1 is an innate immune receptor which together with NOD2 recognizes intra-cellular bacterial components . As NOD1 was also a target of other possible effective compounds (A1, A12 and C5) we compared the cytotoxic effect of these compounds with 5 currently available commercial NOD inhibitors (Additional file 2): BDBM62265, BDBM54356, NOD-IN-1 and noditinib (NOD1 inhibitors) and GSK583 (inhibitor of RIP2, a downstream effector of NOD1/2) . Any of these specific inhibitors exerted a cytotoxic effect in the CLL cell lines (HG3, MEC-1) and the stromal HS-5 cell line analyzed by MTT analysis (Figure 5a). We confirmed by AnnexinV/PI staining that these inhibitors were not cytotoxic for HG3 alone or in coculture with HS-5 cells (Figure 5b). Therefore, we considered that NOD1 was not the main target responsible of the effects seen with C7, although we cannot disregard that NOD1 had some pleiotropic effects with other identified targets for this compound. F1 that corresponded to simvastatin was the other compound of interest (Additional file 1). This drug inhibits the synthesis of cholesterol in the liver by the enzyme 3-hydroxy-3-methylglutaryl coenzyme A reductase (HMGCR) . We compared the effect of simvastatin with other commercially available stains (lovastatin, fluvastatin and rosuvastatin) with different IC50 (Additional files 2 and 9). As it has been reported that statins also inhibit the integrin LFA-1 , we also tested two specific LFA-1 inhibitors (lifitegrast and BDBM50199033). We observed that all statins exerted a cytotoxic effect, analyzed by MTT (Figure 5c) and AnnexinV/PI staining (Figure 5d), although simvastatin was the statin with the highest effect. In contrast, LFA-1 inhibitors did not show any cytotoxic effect. Again, we cannot rule out a possible contribution of LFA-1 when combined with HMG-CoA reductase in the effects seen. In this way, we hypothesized that statins through the inhibition cholesterol synthesis pathway, might participate in the decrease of cell survival and proliferation and in addition, they might also influence cellular adhesion by inhibiting LFA-1 (Figure 5e). To validate this hypothesis, first we analyzed the confluence of cell culture of HG3 and HS5 alone and HG3 in coculture with HS5 treated with simvastatin 1 µM for 48 h. We observed that simvastatin induced a dramatic decrease on cell proliferation in HG3 both in the absence and in the presence of HS-5 cells. In contrast, no effect was observed in HS-5 alone (Figure 6a). We next analyzed the effect of statins and LFA-1 inhibitors on ICAM-mediated adhesion and migration of CLL cells triggered by CXCL12 and CXCL13, key chemokines for CLL cell homing to lymphoid tissues . All different statins and specific LFA-1 inhibitors induced a significant reduction (p<0.05) on CLL adhesion/invasion induced by CXCL12 (Figure 6b) and CXCL13 (Figure 6c). We also confirmed by CFSE staining that all statins tested reduced significantly (p<0.05) the proliferation of CLL cells induced by incubation of cells with ODN2006 plus IL15 at 6 days (Figure 6d). Stains decreased the percentage of CFSElow CLL cells as shown in Figure 6e. To further study the effect of statins in CLL cells, we incubated primary CLL cells alone or in coculture with the stromal cell line HS5 with these different drugs. As observed in Figure 7a, all statins tested induced a cytotoxic effect on CLL cells, and this effect was not protected by incubating the cells with the stromal cell line HS-5. Furthermore, this effect is selective for CLL cells, as no effect was observed in the HS-5 cells at the low doses tested (0.1 and 1 µM; Figure 7a). The effect of these statins was also analyzed in PBMCs from healthy donors (Figure 7b). Statins exerted a significant selective cytotoxic effect in CLL cells compared to B (CD19+) and T (CD3+) cells from healthy donors at the doses of 0.1 and 1 µM.
Systems biology analysis of possible combination therapies
To identify if statins could improve CLL current treatments, we looked for the best combination therapies with simvastatin using the same network-based mathematical model screening used for single drug screening . Drugs were selected when the prediction score for the combination was superior to the individual ones and the p-value associated with the prediction score was lower than 0.1. As simvastatin displayed a high prediction score for CLL (75), it was difficult to identify compounds that increased its individual action. We tested 22 drugs commonly used for CLL treatment and 5 of them showed a probability for the combination above threshold, but only for ibrutinib and venetoclax the combination score was higher than the individual ones (Additional file 10).
Validation in vitro of combination therapies with statins
According to the systems biology prediction, we tested the combination of statins with ibrutinib and venetoclax. We incubated CLL cells in the presence of ODN2006 plus IL15 for 6 days, and CLL cell-viability was analyzed in CD19+ CLL cells by AnnexinV staining. We observed that incubation of cells with different statins and ibrutinib (0.1 µM) reduced CLL cell viability significantly (Figure 8a). Furthermore, proliferation of CLL cells decreased after ibrutinib and statins when were used alone and this effect on CLL proliferation was significantly higher when statins were combined with Ibrutinib (Figure 8b). Also, the cytotoxic effect of venetoclax 1 nM increased with the addition of statins 0.1 µM. We observed a significant (p<0.05) decrease in cell viability when venetoclax and statins were incubated together (Figure 8c).