A Systems Biology Analysis of Chronic Lymphocytic Leukemia

Simple Summary

Chronic lymphocytic leukemia (CLL) is a unique and slowly progressing cancer that affects white blood cells, and research on CLL has highlighted the inconsistency of gene mutations across patients. Using a novel approach that merges statistical thermodynamics and systems biology, this research examines the CLL protein–protein interaction networks to pinpoint proteins integral to the onset of the disease. Betti number (a topology of complexity) estimates, which measure the importance of individual proteins when removed from the network, helped identify numerous potential therapeutic targets, notably within the Wnt signaling pathway, a pathway implicated in various cellular processes and known for its defective expression in CLL. The finding advocates for a multi-target inhibition approach, focusing on several key proteins to minimize side effects, thereby laying a foundation for designing more effective therapies for CLL. This paper emphasizes the potential benefits of a comprehensive study, spanning cellular-to-genome-wide scales, to design personalized treatments for CLL patients.

Abstract

Whole-genome sequencing has revealed that TP53, NOTCH1, ATM, SF3B1, BIRC3, ABL, NXF1, BCR, and ZAP70 are often mutated in CLL, but not consistently across all CLL patients. This paper employs a statistical thermodynamics approach in combination with the systems biology of the CLL protein–protein interaction networks to identify the most significant participant proteins in the cancerous transformation. Betti number (a topology of complexity) estimates highlight a protein hierarchy, primarily in the Wnt pathway known for aberrant CLL activation. These individually identified proteins suggest a network-targeted strategy over single-target drug development. The findings advocate for a multi-target inhibition approach, limited to several key proteins to minimize side effects, thereby providing a foundation for designing therapies. This study emphasizes a shift towards a comprehensive, multi-scale analysis to enhance personalized treatment strategies for CLL, which could be experimentally validated using siRNA or small-molecule inhibitors. The result is not just the identification of these proteins but their rank-order, offering a potent signal amplification in the context of the 20,000 proteins produced by the human body, thus providing a strategic basis for therapeutic intervention in CLL, underscoring the necessity for a more holistic, cellular, chromosomal, and genome-wide study to develop tailored treatments for CLL patients.

1. Introduction

Chronic lymphocytic leukemia (CLL) is a type of cancer that affects white blood cells and tends to progress slowly over many years. It is a chronic lymphoproliferative disorder characterized by an increased production of morphologically mature but immunologically dysfunctional B lymphocytes. As a result, these cells are unable to fight infections as well as normal white blood cells do [1].

The disease starts developing in the bone marrow, since here leukemia cells survive longer and eventually outnumber normal cells. Then, cells further grow and may spread to other parts of the body, including the spleen, the lymph nodes, and the liver [1]. Since the growth of leukemia cells is slow, CLL may remain latent for many years before it causes symptoms, and it is usually harder to cure than acute leukemias [1].

From the genetic perspective, CLL is a unique disease with multiple gene signatures. One cohort of patients can exhibit a different gene-signature set than another cohort. Whole-genome sequencing has revealed that TP53, NOTCH1, ATM, SF3B1, BIRC3, ABL, NXF1, BCR, and ZAP70 are often mutated in CLL, but not consistently across all CLL patients [2,3,4]. For example, NOTCH1 is mutated in about 10% of newly diagnosed patients and in about 15% to 20% of progressive ones. Similarly, SF3B1 is mutated in about 10% of newly diagnosed CLL patients and about 17% in late-stage disease [2]. Just because a gene is mutated does not mean that it will be strongly expressed. One of the goals of our study is to show a molecular thermodynamics approach to determine the most energetically significant pathways supporting a given patient’s CLL initiation and progression. This new molecular systems approach may shed light on optimal treatment for each patient—essentially personalized therapy. Before we present this new methodology, we provide an overview of the known biomarkers for CLL and then a survey of the current treatment options, as well as experimental drugs in development.

It is of fundamental importance to obtain information about the patient’s status and prognosis to define the therapeutic strategy. There exist several laboratory-based prognostic markers, such as high levels of serum beta-2 microglobulin (B2M) and the absolute lymphocyte count (ALC). However, chromosomal aberrations detected using Fluorescent In Situ Hybridization (FISH) serve as the main prognostic tools. The most common aberrations detected in CLL patients are as follows [5]:

• Deletions on the long arm of chromosome 13 (del(13q)): In patients with this aberration, the disease progresses slowly.
• Deletions on the long arm of chromosome 11 (del(11q)): This usually occurs among young males and tends to manifest with bulky lymph nodes. It is associated with rapid disease progression and short survival. The 11q chromosome contains the Ataxia–Telangiectasia-mutated gene (ATM), and ATM kinase is responsible for inhibited cell cycle progression in the case of DNA damage. Furthermore, ATM kinase acts on p53 by phosphorylating it in order to induce apoptosis. Therefore, when 11q is deleted, this phosphorylation does not occur, and the cell damage cannot be repaired [6].
• Deletion on the short arm of chromosome 17 (del(17p)): This results in the loss of TP53, which is the most important prognostic marker in CLL. It is associated with rapid disease progression and resistance to fludarabine chemoimmunotherapy. In addition to the role of TP53 as a prognostic marker in CLL, it is also fundamentally a predictive marker for chemo-immunotherapy, guiding treatment decisions and potentially influencing the response to specific therapies, such as fludarabine chemoimmunotherapy [7].
• Immunoglobulin heavy-chain variable region gene (IGHV) mutational status: For prognosis and therapy choice, it is important to detect IGHV mutational status since the unmutated state is correlated with low survival.
• Other markers, which are present in a low percentage of newly diagnosed CLL patients, but whose incidence increases in patients who are refractory to fludarabine chemotherapy, are mutations of NOTCH1, SF3B1, and BIRC3. Finally, combining genetics, clinical parameters, and biochemistry, the CLL International Prognostic Index (CLL-IPI) is a tool to predict the status of the disease [8].

Wnt signaling is a network of interacting protein pathways which control processes such as cell differentiation, cell cycle regulation, proliferation, apoptosis, cytoskeletal rearrangement, cell polarity, adhesion, motility, migration and invasion, and the interaction with the microenvironment [9]. Wnt signaling is correlated with hematopoiesis and is linked with leukemogenesis of cancers such as CLL [9]. Two Wnt signaling pathways are associated with CLL, namely the Wnt/β-catenin-dependent and -independent pathways. The Wnt/β-catenin is associated with cell proliferation, homeostasis, and cell cycle regulation, and thus its malfunction indicated a hallmark of many cancers. Regarding the Wnt/β-catenin independent pathway, the Wnt/PCP (Planar Cell Polarity) is the most important one, and it takes place in the regulation of cell polarity, migration, and invasion. Wnt pathways play a role in CLL pathogenesis and response to treatment. Moreover, the expression of Wnt signaling molecules from Wnt/β-catenin and Wnt/PCP pathways is defective in CLL. For example, ROR1 (receptor tyrosine kinase-like orphan receptor), a Wnt-5 (a Wnt protein)-dedicated receptor in the Wnt/PCP pathway, is expressed on the surface of CLL cells and not on the healthy B cells. Therefore, ROR1 is a sensitive marker of a possible relapse of patients with a more aggressive form of the disease.

The scope of this study extends beyond the traditional single-target silver bullet approach in drug development, acknowledging the intricate network of proteins that drive the pathological transformation of CLL. A systems biology perspective indicates that targeting a manageable group of five or six network nodes could be more effective for combination therapy design, considering the potential for serious side effects due to overlapping off-target interactions. The statistical thermodynamics method applied here aims to identify and hierarchize such targets, which could be inhibited by existing approved or investigational drugs, setting the stage for a more nuanced and personalized treatment approach in CLL.

2. Current Treatment Options and Experimental Drug Candidates

Unfortunately, currently available treatments may relieve CLL patients from their symptoms and extend their survival, but still CLL remains incurable [6]. For patients without “active disease” and who are asymptomatic, or those with early-stage disease, the treatment consists of just a simple observation during which blood counts are performed every three months [6]. For patients with “active disease”, before choosing therapy, the clinical status must be evaluated in terms of general health, characteristics such as TP53 abnormalities or adverse cytogenetics, or relapsed disease [6]. Standard treatment has been chemoimmunotherapy with fludarabine, cyclophosphamide, and rituximab (FCR). However, it has demonstrated a lack of efficacy, and it leads to numerous side effects, especially in patients with TP53 or NOTCH1 mutations, unmutated IGHV, and deletion of 17p or 11q [8].

Target agents are small molecules that have greater efficacy in patients harboring TP53 mutation or del(17p), whose examples include the following [6]:

• Bruton tyrosine kinase inhibitors (ibrutinib and acalabrutinib),
• BCL-2 inhibitor (venetoclax),
• Purine analogs (fludarabine and pentostatin),
• Alkylating agents (cyclophosphamide, chlorambucil, and bendamustine),
• Monoclonal antibodies (rituximab, ofatumumab, and obinutuzumab),
• PI3K inhibitor (idelalisib).

The most common chemotherapy medications used are listed below, together with their main mode of action:

• Fludarabine: a purine analogue and an antineoplastic agent.
• Cyclophosphamide: an alkylating agent.
• Rituximab: a monoclonal antibody which targets the B-lymphocyte antigen CD20 expressed on the surface of B cells.
• These three together (FCR) constitute a chemoimmunotherapy treatment:
• Bendamustine: an alkylating agent used along with Rituximab (BR) to form another combination chemoimmunotherapy treatment.
• Chlorambucil: an alkylating agent.
• Ibrutinib is a Bruton tyrosine kinase (BTK) inhibitor. BTK, an enzyme which works for B-cell survival and growth, helps delay the progression of cancer. It inhibits CLL cell migration, proliferation, and survival [10]. Unfortunately, it presents some side effects, such as pneumonia, upper respiratory tract infection, atrial fibrillation, sinusitis, headaches, nausea, and many more [10].
• Acalabrutinib is a more selective irreversible BTK inhibitor since it acts just like ibrutinib, but without the side effects involving other kinases [10]. Its most common side effects are headaches, tiredness, low red blood cells, low platelets, and low white blood cells [10].
• PI3K (Phosphatidylinositol-3-kinase) inhibitors such as Idelalisib [11], which was FDA-approved in 2014 for use in combination with rituximab for treating relapsed CLL [12]. However, Idelalisib is also toxic with nearly 40% of patients having had to interrupt the therapy due to rash or 3–4-grade transaminitis, and pulmonary infections. A PI3Kδ inhibitor called TGR 1202 has better selectivity compared to idelalisib. It was approved for medical use in the USA in February 2021. TGR 1202 reduces the phosphorylation of AKT in lymphoma and leukemia cells.
• Venetoclax binds and inhibits the antiapoptotic protein B-cell lymphoma 2 (BCL-2) [9]. In CLL, the inhibition of this pathway has been considered an optimal therapeutic strategy [13]. The use of Venetoclax was approved by the FDA in 2016 [13]. Its side effects usually include low levels of white and red blood cells, respiratory infections, diarrhea, nausea, tiredness, and tumor lysis syndrome (TLS).
• Sotorasib (AMG510) is a highly selective and irreversible inhibitor which binds at an allosteric pocket, leading to the trapping of KRAS (Kirsten rat sarcoma virus) in an inactive GDP-bound state. Note that KRAS transmits signals for growth, division, and differentiation to the nucleus of the cell from the outside. KRAS mutations are among the most oncogenic events in carcinomas, including CLL, and a majority of them consist of missense mutation of the 12th codon (glycine). It was approved by the FDA in May 2021. Some of its side effects are diarrhea, nausea, and muscle or bone pain [14].
• Adagrasib (MRTX849) is an irreversible covalent inhibitor of G12C KRAS mutation that makes a covalent bond to cysteine and binds in the switch-II pocket of KRAS in its inactive GDP state. It demonstrated improved antitumor activity when in combination with vistusertib (an mTOR inhibitor). In clinical trials, some patients experienced pneumonitis and heart failure, which led to the interruption of the treatment. Others experienced nausea, fatigue, and anemia. This inhibitor is still in clinical trials, together with numerous other experimental drugs under development [14].
• Experimental drug candidates also include AKT pathway allosteric inhibitors: ARQ092/miransertib, BAY1125976, MK2206, and TAS-117 [15].
• ATP-competitive AKT inhibitors: Capivasertib and ipatasertib showed a favorable safety profile, along with signs of activity in phase I monotherapy trials [14]. Other AKT inhibitors include the following compounds: afuresertib (GSK2110183), uprosertib (GSK2141795 and GSK795), and ordidonin (NSC-250682) [16].

Additionally, various drug candidates are in development with MYC pathway-inhibition profiles [17]:

• Compound 361 (MYCi361, NUCC-0196361) [18],
• Compound 975 (MICi975, NUCC-0200975) [19],
• MYCMI-6 [20],
• KSI-3716 and MYRA-A [20],
• KI-MS2-008 [20],
• L755507 [17].

3. Systems Biology Background

The conceptual framework for understanding the thermodynamics and energetics of the molecular biology of human diseases from a network biology perspective has been developed over the past decade. Various studies were undertaken to quantify different signaling and metabolic pathways in various cancer types and other diseases, using metrics such as network entropy and the Gibbs free energy applied to each specific case [21,22,23,24,25,26]. Here, we give only a brief summary of this approach. The transcriptome and other “-omic” (e.g., proteomic, genomic, etc.) measures can represent the energetic state of a cell. Here, we mean “energetic” from a thermodynamics perspective. There is a chemical potential between interacting molecules in a cell, and the chemical potential of all the proteins that interact with each other can be imagined forming a rugged landscape, not dissimilar to Waddington’s epigenetic landscape [27,28].

The method we propose uses mRNA transcriptome data or RNA-seq data as a surrogate for protein concentration. This assumption is largely valid. Kim et al. [29] and Wihelm et al. [30] have shown an 83% correlation between mass spectrometry-generated proteomic information and transcriptomic information for multiple tissue types. Furthermore, Guo et al. [31] found a Spearman correlation of 0.8 in comparing RNAseq and mRNA transcriptome from TCGA human cancer data [32].

Given a set of transcriptome data, a representative of protein concentration, we overlay that on the human protein–protein interaction network from BioGrid [33]. This means that we assign to each protein on the network the scaled (between 0 and 1) transcriptome value (or RNAseq value). From that, we can compute the Gibbs free energy of each protein–protein interaction using the mapping relation:

$$G_i=c_i\ln {\displaystyle \frac{c_i}{{\displaystyle {\sum }_j{c}_j}}}$$

(1)

where c_i is the “concentration” of the protein i, normalized or rescaled to be between 0 and 1. The sum in the denominator is taken over all protein neighbors of i, including i. Therefore, the denominator can be considered a degree-entropy. Carrying out this mathematical operation essentially transforms the “concentration” value assigned to each protein to a Gibbs free energy. Thus, we replace the scalar value of transcriptome to a scalar function—the Gibbs free energy.

The above equation is derived from a well-known concept in chemical thermodynamics [34]. A biological cell or a group of cells (a tumor) exist in a complex chemical balance produced by a network of interacting molecular species, ranging from small molecules to some very large molecules on the order of hundreds to thousands of Daltons. The molecular concentration balance in this network is the Gibbs free energy, G. This thermodynamic quantity is typically expressed in the context of systems kept at a constant temperature and pressure, where the system can exchange molecules with the environment. For an arbitrary molecular system, the Gibbs function is given as a molar difference [35] in Equation (2):

$$\mathsf{\delta }\mathrm{G}=\mathsf{\mu }\mathsf{\delta }\mathrm{n}.$$

(2)

where μ symbolizes the chemical potential, $\mathsf{\delta }\mathrm{G}$ is the Gibbs energy, and $\mathsf{\delta }\mathrm{n}$ is the molar difference (essentially concentration difference). Typically, one writes the chemical potential as follows:

$${\mathsf{\mu }}_{\mathrm{i}}={\left({\displaystyle \frac{\mathsf{\delta }\mathrm{G}}{\mathsf{\delta }{\mathrm{n}}_{\mathrm{i}}}}\right)}_{\mathrm{P},\mathrm{T},\mathrm{j},\mathrm{k},\mathrm{e}\mathrm{t}\mathrm{c}.}$$

(3)

Equation (3) above assumes that the molar concentrations of other molecular components (other than i) are held constant, along with constant temperature and pressure. Using Equation (1) and given a network of interacting chemical species or proteins, and given their concentration, we can compute the Gibbs free energy for a single protein in the PPI.

The Gibbs free energy is a negative number, so associated with each protein on the network is a negative energy well. This results in a rugged energy landscape represented schematically in Figure 1. If we use what is referred to as a topological filtration on this landscape, we essentially move a filtration plane up from the deepest energy well. As the filtration plane is moved up, larger-and-larger energetic subnetworks are captured. For convenience, we stop the filtration at energy threshold 32—meaning 32 nodes in the energetic subnetwork. We call these subnetworks Gibbs-homology networks.

We now compute the Betti centrality, a topological measure, on the 32-node energetic networks, as described in Benzekry et al. [23]. The concept is easily described. In networks, there are holes or rings of various sizes. In these energetic pathways, protein–protein interaction networks, the proteins form interaction rings. In densely connected, but not fully connected, networks, the rings or holes may consist of triangles and larger rings of interaction. To find the Betti centrality, we ask ourselves the following question: Which protein, when removed from the network, will change the overall total number of rings the most? The total number of rings is called the Betti number. Given a network G consisting of edges, e, and vertices, v, the Betti centrality is given by Equation (4):

$$B(v_i)=B(G)-B(G-\{v_i\})$$

(4)

Hence, the difference from the total Betti number, B(G), and the Betti number of the network after removing node i gives the Betti centrality for node i. We compute this for all nodes in the threshold-32 energetic network. Often, there will be two or more proteins in the network that have equivalent Betti centrality.

4. Methodology and Datasets

We report on a meta-analysis of 1001 samples from CLL patients and cancer cell lines. This study used online data from GEO [36]: GSE10139, GSE28654, GSE31048, GSE39671, GSE49896, GSE50006, and GSE69034. The data were mRNA expression numbers, all collected using Affymetrix Human Genome Array, HG-U133_2. We also used the human protein–protein interaction network from Biogrid [33]. In particular, we used the dataset downloaded from the BIOGRID-ORGANISM: homo_sapiens-3.5.172.2. An integrative analysis, including the t-SNE visualization of samples and subgroups, is detailed in Appendix C (Figure A1).

Reiterating the method, we collected the GSE expression datasets, and then the expression value for each gene was overlaid on the human protein–protein interaction network for each protein or node in the network. For each node in the network, we then applied Equation (1), which resulted in the Gibbs energy for that node. This resulted in a rugged landscape similar to the one shown in Figure 1. Then, the procedure consisted in performing a filtration and computing the Betti number for zero nodes removed and then removing a node and recomputing the Betti number and replacing the node. This removal–computation–replacement procedure resulted in a list of nodes that had the largest impact on complexity of the Gibbs homology network. We finally ranked significant nodes in a Pareto chart for each patient. Pareto charts were prepared at several filtration thresholds: 32, 48, 64, and 96.

5. Results

Our discussion of the results is presented below, and it follows an analysis of the individual datasets and the research publication associated with it (if present) prior to presenting the meta-analysis Pareto chart and the network graphs.

i.

Ref. [37] (GSE10137) “A genomic approach to improve prognosis and predict therapeutic response in chronic lymphocytic leukemia”, by Friedman et al., 2009. This was one of the papers with a large table in the Supplementary section. The table consisted of upregulated and downregulated probes indicative of progressive disease; upregulated and downregulated probes indicating chlorambucil resistance; upregulated and downregulated probes indicative of Pentostatin, Cyclophosphamide, and Rituximab signature. An important quote from the paper states that: “Others have previously noted the prognostic significance of cytoskeletal genes and the tumor necrosis factor in CLL. Notably, probes for ZAP-70 did not constitute this genomic signature, although mean expression for ZAP-70 probes in samples from patients with progressive disease was higher than those from patients with stable disease”. The table of genes was parsed from the PDF document and used in our subsequent analysis (discussed below).

ii.

Ref. [3] (GSE28654) “Gene expression profiling identifies ARSD as a new marker of disease progression and sphingolipid metabolism as a potential novel metabolism in chronic lymphocytic leukemia” (Trojani et al., 2012) [3]. A table in the manuscript lists about 65 genes that were selected as being differentially expressed in two cohorts of CLL patients. Of those genes, the authors selected 19 genes for PCR analysis because of their significance. Those genes are ZAP70, ARSD, LPL, ADAM29, AGPAT2, CRY1, MBOAT1, YPEL1, NRIP1, RIMKLB, P2RX1, EGR3, TGFBR3, APP, DCLK2, FGL2, ZNF667, CHPT1, and FUT8. An important quote from the paper is as follows: “In the literature, lists of differentially expressed genes obtained using high-throughput microarray by different laboratories and research centers have often limited overlap [38,39]. These differences are matters of important scientific discussions and are imputed, among other causes, to dataset dimensions: small number of subjects (some tens) with respect to the number of variables (tens of thousands of genomic probes in human). Notably and reassuringly the gene set list (65 genes) emerged from this study showed a substantial (but not quantitated) overlap with results from previously published microarray studies [40,41,42,43]”. The list of 65 genes was incorporated in our subsequent analysis.

iii.

Ref. [40] (GSE31048) “Somatic mutation as a mechanism of Wnt/$\beta$-Catenin pathway activation in CLL” [40]. In the Supplementary Materials to this paper were two large tables listing genes. One table listed from their own study (Wnt pathway), and the other table listed Wnt genes from the literature and websites. Both tables were combined for the study. A quote from the paper: “… our data demonstrate that altered gene expression is indistinguishable between samples with and without mutations”.

iv.

Ref. [41] (GSE39671) “Subnetwork-based analysis of chronic lymphocytic leukemia identifies pathways that associate with disease progression” (Chuang et al., 2012) [41]. The Supplementary data included only figures and graphs. No table with a gene list. Of note is the following quote: “Furthermore, the marker sets identified by different research groups often share few genes in common. Two landmark studies, Rosenwald and colleagues [42] and Klein and colleagues [43] each identify approximately 100 genes that were expressed differentially by CLL cells that use mutated versus unmutated IGHV genes. However, only 4 marker genes were identified in common between these studies”.

v.

Ref. [44] (GSE49896) “miR-150 influences B-cell receptor signaling in chronic lymphocytic leukemia by regulating expression of GAB1 and FOXP1” (Mraz, et al., 2014) [44]. The following is quoted from their paper: “We identified miR-150 as being the most abundantly expressed miRNA in CLL. However, we observed significant heterogeneity in the expression levels of this miRNA among CLL cells of different patients. Low-level expression of miR-150 associated with unfavorable clinicobiological and prognostic markers, such as expression of ZAP-70 or use of unmutated IGHV (p < 0.005). Additionally, our data suggest that the levels of methylation of the upstream region of 1000 nt proximal to miR-150 associate with its expression. We demonstrated that GAB1 and FOXP1 genes represent newly defined direct targets of miR-150 in CLL cells. We also showed that high-level expression of GAB1 and FOXP1 associates with relatively high sensitivity of CLL cells to surface immunoglobulin ligation. High levels of GAB1/FOXP1 and low levels of miR-150 associate with a greater responsiveness to BCR ligation in CLL cells and more adverse clinical prognosis”.

vi.

GSE50006—no manuscript.

vii.

GSE69034—no manuscript.

We created a master list of all genes cited and/or given in the tables associated with the above manuscripts. This list is in Appendix B. There were 515 genes in total. The list of genes was inputted into the DAVID platform for functional annotation analysis [45], and only 208 genes were found, which indicates that DAVID’s database has annotations for only 208 of those genes. The missing genes might be due to them being less well-characterized, newer discoveries not yet integrated into DAVID’s database, or they might be represented differently in the user’s list compared to DAVID’s nomenclature. To identify genes relevant for a generic condition like “leukemia”, the KEGG [46] and OMIM [47] databases are used to filter and analyze the results such that both are integrated into DAVID. These databases contain curated information about genes related to specific pathways or diseases. By cross-referencing the 208 identified genes with “leukemia” in both KEGG and OMIM, genes whose expression or mutation is linked with the onset, progression, or other aspects of leukemia are pinpointed, aiming at narrowing down potential targets for research, therapeutic development, or further molecular study. Searching that file resulted in the following list: AKT1, CTBP1, CTBP2, CTBPA, SMAD4, HDAC1, LEF1, RARA, TCF3, TCF7, TCF7L1, TCF7L2, and MYC.

Comparing the PublishedGeneList with our CLLnet96 list, only four were found: MYC, HDAC1, CTNNB1, and APP. Two of those, MYC and HDAC1, are known to participate in leukemia. The CLLnet96 list is assembled from all 1001 patients at Gibbs threshold 96. To reiterate the concept of threshold, for any given patient, the deepest well in the landscape is usually the same for all thresholds; but there may be differences based on the expression, and this gives rise to differences in the Gibbs homology network. An energy threshold of, say, 32 will result in a network of 32 nodes that are the largest negative energy values. This is called a topological filtration. Using this technique, we can produce 1 of these 32 threshold networks for each patient. If we do that, and then concatenate the entire list of nodes for each of the patients at this threshold, followed by sorting and discarding redundant nodes in the list, the result will be what we call the CLLnet32 list. By the nature of the filtration, CLLnet32 ⊂ CLLnet48 ⊂ CLLnet64 ⊂ CLLnet96, meaning that CLLnet32 is a proper subset of CLLnet48, etc. So, taking the list CLLnet96 will, by definition, incorporate all others. Comparing our CLLnet96 with the superset of published genes (i.e., PublishedGeneList in the Appendix B), we find only four that were both lists, MYC, HDAC1, CTNNB1, and APP.

After comparing the PublishedGeneList and the CLLnet96 superset, we then used DAVID, an online bioinformatics resource that allows one to submit a list of genes (or other biological components, e.g., proteins), and it returns important information such as the KEGG pathway or OMIM associated with that gene. There are 98 genes in the superlist of CLL96net list. From that analysis, we find the following genes to be associated with leukemia (various types): KRAS, GRB2, HDAC1, NPM1, TP53, and MYC. In that superlist, CUL1, TP53, and CTNNB1 are associated with the Wnt signaling pathway.

The published gene lists consisted of two parts. Keep in mind that, although the papers cited above included GSE expression data, most of them did not include tables of genes they identified from their analysis as being important. Instead, they were looking for prognostic markers for disease progression. So, Part 1 of the published gene list consisted of selections identified by the authors from GSE10137 and GSE28654. The combined list consisted of 320 genes. Of those 320 genes, 22 were found in DAVID. Only CEBPA and MYC were found to be associated with any form of leukemia. And CSNK2A1 and MYC were found to be associated with the Wnt signaling pathway. When we expand the published list to include GSE321048, which was a focused study on the Wnt pathway and CLL [40], the list expands to 515. Naturally, a huge number of genes were flagged by DAVID as being in the Wnt pathway (78 total). And a smaller subset was found to be associated with some form of leukemia: AKT1, CTBP1, CTBP2, CEBPA, SMAD4, HDAC1, LEF1, RARA, TCF3, TCF7, TCF7L1, TCFL2, and MYC. Looking for common genes between the expanded published and our larger list of 96 threshold, we find KRAS, GRB2, HDAC1, NPM1, TP53, MYC, APP, and CTNNB1.

At threshold 32, CTNNB1 is the best Betti target once out of 1001 patients, but it is present in the threshold 32 networks 326 times. Keep in mind that anything found in the 32 threshold is energetically important. So, we find it in 32.5% of the population as a potentially good target for CLL (at 48 threshold 37.9%, at 64 threshold 44.1%, and at 96 threshold 58.9%). CTNNB1 is an important gene involved in CLL. It is also an important node in the Wnt pathway [48].

6. Results and Discussion: Wnt Pathway

It is interesting that so many of the authors of the papers cited above did not find overlap among their gene list and other investigators. There was little overlap between those authors’ lists until we included the dataset from GSE321048, the Wnt pathway. We speculate that the reason our Gibbs analysis of expression data did not overlap well with other expression data is that the Gibbs function includes a measure of network entropy (denominator in Equation (1)). Furthermore, many of the genes that are highly expressed, as reported in the literature, are not necessarily mutated based on whole-genome sequencing.

Figure 2 shows the Wnt pathway from KEGG. After using the online R-script KEGGraph at Bioconductor, it was converted to an edge-list of relevant protein–protein interactions [49].

The resulting edge-list was plotted using Cytoscape 3.7 [50]. The PPI network is shown in Figure 3. Two nodes are highlighted. MYC is highlighted and connected to LEF1, TCF7, TCF7L1, and TCF7L2. MYC, as we will see, is an important player in the Wnt pathway. Also, CTNNB1 has 24 neighboring interactions and has a betweenness of 0.3155, the highest in this network. It also is an important player in the Wnt pathway.

As described above, we computed the Betti centrality for the Gibbs homology networks. Figure 4 shows a Pareto chart for the Betti centrality nodes at threshold-48. In our analysis of the 1001 expression samples, CTNNB1 was present as a key Betti centrality node in three samples. Whereas MYC was not present as a key Betti centrality node at threshold-48, but at threshold-32, MYC was present 24 times; 12 times (50%), it was found in dataset GSE30671, which is associated with the manuscript by Chuang et al. [41]. This again shows the inconsistency in gene expression values from samples of CLL patients. Of key importance is the fact that RPS15 is a Betti centrality node in three patients, and RPS15A is a Betti centrality node in five patients at threshold-48. These are shown in Figure 4.

As shown in Figure 3, MYC is an important node in the Wnt pathway. It is directly connected to LEF1, TCF7, TCF7L1, and TCF7L2. Except for LEF1, which is a lymphoid enhancer-binding factor, the others are transcription factors. Figure 5 shows a Gibbs homology network at threshold-48 for a patient (GSM787065 part of GSE31048 [40]) whose RPS15 is the Betti centrality node. In the network diagram, the nodes are in a degree-sorted order, starting at the bottom, with MYC as the highest degree (48), and going around counterclockwise. RPS15 and MYC are pulled out of the network for easy locating, and MYC and all its first connections are highlighted in yellow.

As we pointed out above, a gene can be mutated, and yet not overexpressed or underexpressed relative to normal. This is likely the main cause for differences in reported transcriptome data from various investigators. What is clear from the literature (e.g., Wang et al.) [40] is that the Wnt pathway is highly important, and overexpressed genes in that pathway are often indicative of cancer. MYC is a regulator of ribosome protein synthesis [51] and has been shown to be a key regulator in supporting and maintaining tumorigenesis [52]. For example, Wu et al. [52] found that inactivation of MYC resulted in some tumors undergoing regression, and mutated RPS15 was identified in almost 20% of CLL patients who relapsed after FCR treatment. These mutations are associated with clinical aggressiveness in CLL, along with the mutant RPS15 displaying defective regulation of endogenous p53, which indicates a novel molecular mechanism underlying CLL pathobiology [52]. RPS15 and RPS15A are often overexpressed in CLL [53], and our results confirm this with all (1001 patients) Gibbs-homology subnetworks at threshold-96, showing RPS15 or RPS15A as being an energetically important node.

7. Conclusions

We showed in this study that the genes BTK, NFkB, JAK/STAT, NOTCH1, BCL2, and EEF2, among others, play a significant role in the support of CLL. Yet, some of them are only rarely studied in the literature because they are not strongly expressed; our research confirms this. Just because a gene or two are mutated does not mean that they will be strongly expressed. As pointed out above, gene expression found inconsistent sets of genes that were highly expressed and that had high Gibbs energy. We found the same inconsistent results when we looked at the Hi-C results for CLL patients. Speedy et al. [54] found that BCL2 was strongly implicated in the disease. They also found a disruption at the NFkB-binding site, but other genes, such as JAK/STAT, BTK, and EEF2, were not mentioned in their manuscript. Beekman et al. [55] found only NOTCH1, Puiggros et al. [56] found only NOTCH1 and SF3B1 as candidates for high risk of mutation, and Kiefer et al. [57] found NOTCH1 for trisomy 12.

Though the actual causal agent of CLL is not well known, we can speculate that if there is some molecular agent (e.g., herbicide) or an energetic EM signal (e.g., X-ray), it will typically impact the cell only during a specific phase of the cell cycle [58]. There are regions of the genome that are more sensitive to alterations due to some specific energy level in the overall molecular network we call a cell. These mutations are driven by the relevant chemical potential, stereochemistry, and Gibbs free energy. We argue that the locations of the relevant genes in the chromosome and the 4D dynamics of the nucleome may suggest a more holistic molecular and cellular approach to understanding CLL and, therefore, new therapeutic strategies [59]. Building on this notion, the insights from the 4D Nucleome Network [59] elucidate the intricacies of genome organization in space and time. The project underlines the critical role of the genome’s three-dimensional organization in gene regulation. In the context of CLL, the spatial dynamics of chromatin can have a profound impact on gene expression patterns, emphasizing the importance of the genome’s spatial and temporal dynamics in understanding and potentially treating the disease [60]. Elucidating this idea further, the work conducted by Sawh et al. in 2022 revealed that the eukaryotic genome is a multilayered entity, exhibiting intricate organization levels that range from nucleosomes to larger chromosomal scales [61]. These layers undergo significant remodeling across different tissues and developmental stages in C. elegans. It is noteworthy that advancements in C. elegans research, both imaging-based and sequencing-based, have unveiled the influence of histone modifications, regulatory elements, and broader chromosome configurations in this 4D organization. Specific revelations, such as the physiological implications of topologically associating domains and compartment variability during initial developmental phases, underscore the depth of genome dynamics. These insights provide compelling evidence that understanding such 4D genome organization nuances is crucial for decoding complex diseases like CLL. Interestingly enough, none of the genes described in Appendix A is in chromosome 13, which often has deletions in about 50% of CLL patients [62]. In Appendix A, we support our argument for a larger view that CLL genes are widely spread throughout the whole genome and different chromosomes.

In conclusion, our study challenges the conventional single-target paradigm in CLL therapy, advocating for a higher-level, network-oriented strategy. The identification and hierarchical ranking of 20–30 significant proteins, amidst the roughly 20,000 synthesized by the human organism, represent a leap in signal detection and amplification [63]. This nuanced profiling, achieved via a statistical thermodynamics approach, underscores the potential of targeting a selective array of five or six network nodes. This selectivity is crucial to mitigate the risk of adverse effects caused by overlapping off-target interactions commonly seen with broader therapeutic targets. The proteins highlighted in our research, notably within the Wnt signaling pathway, are not merely isolated entities but components of a complex network that drives the CLL pathology. Therefore, our proposed method does not end at the identification of these proteins but extends to rank-ordering them in terms of therapeutic relevance. The next step for validating the findings involves experimental assays using siRNA [64] or small-molecule inhibitors, which will provide the empirical backbone for our theoretical model. Such an approach may revolutionize the current treatment regimens by transitioning from a one-size-fits-all model to a more customized, patient-specific strategy. This could be especially beneficial given the genetic variability among CLL patients, as indicated by the inconsistent mutation patterns observed in whole-genome sequencing. By incorporating the principles of systems biology and acknowledging the network dynamics of protein interactions, we can begin to envision a more effective, personalized therapeutic landscape for CLL. This, in turn, may pave the way for similar strategies in other cancers, marking a paradigm shift in oncological treatment towards precision medicine.