Chronic lymphocytic leukemia (CLL), a mature and monoclonal CD5+ CD23+ B cell malignancy, proliferates and accumulates in the bone marrow, blood, and lymphoid nodes (Hallek et al., 2018). It is often asymptomatic in the early stage. It is often found that painless lymphadenopathy or the absolute value of lymphocytes is increased for unknown reasons. Patients have mild fatigue, fatigue, and other non-specific manifestations. Once they enter the advanced stage, they can present with weight loss, repeated infection, bleeding, and anemia in addition to systemic lymph nodes and splenomegaly. CLL cases are fewer in Asia than those in the Western world, and it is reasonable to assume that genetic and environmental factors play roles in pathogenesis (Burger, 2020). During 2014–2018, the rate of new cases of CLL was 4.9 per 100,000 per year and the median age at diagnosis is 72 years, the death rate was 1.1 according to the aforementioned survey [The Surveillance Epidemiology and End Results (SEER) Program of the National Cancer Institute. Cancer fact sheets: chronic lymphocytic leukemia (CLL). https://seer.cancer.gov/statfacts/html/clyl.html (accessed 22 September 2021)].

CLL is widely known as a heterogeneous disease that exhibits variable clinical symptoms, time-to-treatment (TTT), easily progression and difficult prognosis. CLL patients are often diagnosed with incidental findings, and the clinical course ranges from an asymptomatic, indolent disease that requires no treatment to a rapidly progressive and chemotherapy-resistant disease until death within a short period (Burger, 2020). The indications for treatment mainly include the clinical stage and symptoms of patients, and the standard therapy is chemoimmunotherapy. Unfortunately, the majority of CLL patients are too old to tolerate intensive standard chemotherapy; therefore, an effective prognostic model is needed to predict the individual clinical courses and to improve the outcome. Over the past few decades, great advances have been made in figuring out the molecular and genetic biology of CLL to identify the indicators of progression and survival. These indicators include cytogenetics, age, IGHV gene mutation status, β2-microglobulin (β2-MG), clinical-stage (RAI/BINET stage), and so forth (Bosch and Dalla-Favera, 2019). In CLL, 13q14, 11q22-23, trisomy of 12q, and 17p deletions are found in 80% of the cases. 11q22-23 and 17p deletions are associated with poor survival, whereas 13q14 deletions and trisomy of 12q have a longer TTT and survival time (Dohner et al., 2000). TP53 aberrations (Zenz et al., 2010) indicate a more aggressive disease progression and extensive drug-resistant and worse outcome, and the same role applies to IGHV genes (Damle et al., 1999) and ZAP-70 (Crespo et al., 2003). Unmutated IGHV and high-expression of ZAP-70 have a comparatively aggressive disease course too, and other relevant risk markers include expression of CD38 (Rassenti et al., 2008), CD49d (Bulian et al., 2014), lipoprotein lipase (LPL) (Prieto and Oppezzo, 2017), serum concentrations of thymidine kinase (Hallek et al., 1999), and β2-microglobulin (Hallek et al., 1996).

In this article, a SE-associated gene list was used to carry out LASSO-penalized Cox regression analysis, and construct a nine SE-associated gene prognostic model, namely, TCF7, VEGFA, MNT, GMIP, SLAMF1, TNFRSF25, GRWD1, SLC6AC, and LAG3. Meanwhile, this model was verified by testing GEO datasets and the ICGC-CLL dataset, respectively. Univariate and multivariate Cox regression analyses, and the ROC curve were analyzed to evaluate the prognostic accuracy of this nine-gene model. Moreover, the aforementioned validated steps, the role of the nine-gene prognostic model, and the nine hub genes were further explored in CLL genesis and the relationship between this prognostic model and other known risk markers, such as IGHV status, FISH abnormality, and ZAP70 expression level. It was indicated that the model demonstrated predictive power and had an expected relationship with known risk markers. In addition, an interactive nomogram based on the nine-gene risk score and clinical traits was constructed. Finally, paired pre- and post-treatment datasets were used to examine the effects of treatments on the risk score or each of the nine hub genes’ expression, and we predicted 25 clinical drugs that may be more sensitive to high-risk patients. The improved nine-gene prognostic model of this work provided a bright future for the diagnosis, disease stratification, and therapy of patients with CLL.

CLL is considered to have a highly heterogeneous clinical course, with time to first treatment varying from months to years and many patients eventually progressing and requiring chemotherapy, although initially, CLL is reported as an indolent malignancy. A review of the data so far, disease stratification, IGHV mutation status, 17p, and ZAP70 expression are the validated predictors of overall survival. Beyond that, gene expression analysis was carried out on various surrogate markers for genetic features and prognosis. A total of six surface antigens (CD62L, CD54, CD49c, CD49d, CD38, and CD79b) and prognostic risk models were put in place to diagnose and predict the OS for CLL (Zucchetto et al., 2006). Moreover, some large-scale gene expression profiling analyses generate different prognostic factors (Kienle et al., 2010; Herold et al., 2011a; Schweighofer et al., 2011). But the previous studies constructed no prognostic model according to SE-associated genes which regulate the expression of hub genes related to CLL tumorigenesis.

A super-enhancer is a new concept developed in recent years; a growing body of evidence indicates an explicit relationship between increasing tumorigenesis and malignancy of cancer and SEs. SEs drive not only the expression of genes but also non-coding RNA that regulates biological functions directly and indirectly. LASSO-penalized Cox regression has become popular in recent years because it could minimize overfitting (Ma et al., 2019). Hence, in our article, we use this novel bioinformatic strategy and the Cox proportional hazard regression models to screen and optimize hub genes related to survival.

In our research, the LASSO-penalized Cox regression analysis was carried out by filtering out the potential SE-associated genes and yielding a nine-gene prognostic model to foresee the OS of CLL patients. All of the individual markers in the nine-gene model associated with OS of CLL by Cox regression analysis were identical. K–M survival analysis also indicated that the majority of the nine genes correlated to OS. Beyond that, the nine-gene prognostic model was highly significant in the multivariate analysis of patients without treatment. The AUCs and C-index showed that our model performed well in the prediction of survival. The effectiveness of this prognostic model could be validated by an independent patient cohort. Moreover , this risk model was another indicator of TTT. We utilized the nine-gene risk score in the GSE22762 and GSE39671 datasets, and the results also indicated that the nine-gene model could be applied to predict TTT. The high-risk patients had less time-to-treatment than the low-risk patients. These data strongly indicated that the nine-gene prognostic model was a significant and valid risk forecaster.

We not only evaluated the data by a rigorous training and validation design, but also concentrated on the connection between individual genes and selected disease characteristics, such as IGHV mutation status, FISH abnormality, and ZAP70 expression level. The results of three of the markers (TCF7, SLAMF1, and LAG3) detected according to the association with IGHV status were expected. The lack of a public database that included both survival data and mutation information limited further research on a correlation between the nine-gene model and ZAP70, a FISH abnormality. But in the poor prognosis groups, like ZAP70-high and 17q-patients, the nine-gene risk score was significantly higher than that in the low-risk group, and we found that the low expression of SLAMF1 in CLL was associated with ZAP70-high expression. The quantitative relationship between TCF7, LAG3, and SLAMF1 expression and inferior overall survival was an accurate finding and indicated that these genes had a pathogenic role in CLL. Additionally, the nine-gene prognostic model also played an important role in CLL etiopathogenesis. The WGCNA of the GSE50006 dataset revealed that TCF7 and LAG3 belonged to two gene modules, respectively. In addition to this, the expression of GMIP, SLAMF1, and TNFRSF25 were also significantly different in normal and CLL patients. Therefore, the five genes contained in our model were possibly functionally vital in the pathogenesis of CLL. In the present study, SLAMF1, TCF7, TNFRSF25, MNT, and VEGFA were protective factors, whereas GRWD1, SLC6A3, GMIP, and LAG3 appeared to be harmful factors in CLL; we subsequently discussed each gene in the prognostic model.

Transcription factor 7 (TCF7), the T-cell–specific transcription factor required for T-cell development in animal models, suggests that it probably functions as a tumor suppressor (Roose et al., 1999). TCF7 over-expression in mice led to a disease resembling CLL, indicating that it was probably involved in the CLL transformation in a direct way (Bichi et al., 2002). In CLL, TCF7 expression provided a high rate (74%) of correct assignment of patients at genetic risk (IGHV unmutated, V3-21 usage, 11q-, or 17p-) (Kienle et al., 2010). The aforementioned results are consistent with ours, and this indicates TCF7 plays an important role in CLL.

Signaling lymphocytic activation molecule family member 1 (SLAMF1), also known as CD150, regulates hematopoietic stem cell differentiation, leukocyte adhesion and activation, and humoral immune responses. SLAMF1 comparatively over-express in normal peripheral blood B cells according to the meta-analysis of three gene expression profiling studies. Recently, researchers found lower levels of SLAMF1 expression in cases with ZAP70-high (p < 0.001), IGHV-unmutated (p < 0.001), and 17q- (p = 0.003). In past studies, we believed that loss of SLAMF1 expression in CLL modulates genetic pathways regulating chemotaxis and autophagy and that potentially affects drug responses, suggesting that the effects underlie unfavorable clinical outcomes experienced by SLAMF1-low patients (Bologna et al., 2016). Together, SLAMF receptors, the vital modulators of the BCR signaling axis, improve immune control in CLL by potentially interfering with NK cells (von Wenserski et al., 2021). In our research, the univariate and multivariate analyses presented that downregulated SLAMF1 levels had an independent negative prognostic impact on overall survival (p < 0.05). We subsequently discovered that SLAMF1 is relatively overexpressed in IGHV-mutated and ZAP70-low CLL patients. The strict correlation among low levels of it and high-risk genetic features indicated that it probably represented a marker of surrogate genomic complexity; however, the mechanism of this correlation is still unknown.

Lymphocyte activating 3 (LAG3), the immune inhibitory checkpoint receptor, is one of the immunoglobulin superfamily with about 20% amino acid homology with CD4. The expression of it activates and exhausts T, NK cells, B cells, dendritic cells, and regulatory T (Treg) cells. LAG3 high expression in CLL cells correlates with unmutated IGHV (p < 0.0001) and decreased treatment-free survival (p = 0.0087) (Kotaskova et al., 2010). Increased LAG-3 expression on leukemic cells correlates with shorter time-to-treatment and poor outcome in CLL; moreover, treatment with relatlimab, a novel anti-LAG-3 blocking monoclonal antibody currently under clinical trial for different solid and hematological malignancies including CLL, restored, at least in part, NK and T-cell–mediated anti-tumor responses (Sordo-Bahamonde et al., 2021). CART cell generation with the showing of ibrutinib created enhanced cell viability and expansion of CLL patient-derived CART cells. Also, ibrutinib enriched the mentioned cells with the less-differentiated naïve-like phenotype and declined expression of exhaustion markers (PD-1, TIM-3, and LAG-3) (Fan et al., 2021).

Vascular endothelial growth factor A (VEGFA) is a member of the PDGF/VEGF growth factor family. The angiogenesis process makes a significant contribution to the pathogenesis of B-cell chronic lymphocytic leukemia (B-CLL), the levels of VEGFA and bFGF being higher in patients than in healthy people (Ballester et al., 2020). Whereas, in our research, VEGFA has a protective role in CLL. The high expression of VEGFA indicated a good prognosis by the K–M survival analysis, and in normal samples, the level of VEGFA was higher even though it was not statistically significant.

The TNF receptor superfamily member 25 (TNFRSF25), the receptor expressed preferentially in the tissues of lymphocytes, possibly plays functions vital to the regulation of lymphocyte homeostasis. The receptor stimulates sNF-kappa B activity and regulates cell apoptosis. TNFRSF25 was differentially expressed, activating CLL cells and predominantly detected in those with early clinical stage disease (Cavallini et al., 2015) and probably alters the balance between cell proliferation and death, influencing CLL physiopathology and results in the clinic.

A total of three genes (GRWD1, GMIP, and SLC6A3) have not been described in the context of CLL before, and all of them were upregulated in high-risk CLL patients. The results of the univariate and K–M survival curves were not completely consistent with multivariate analysis. Glutamate-rich WD repeat containing 1 (GRWD1) was identified as one of the ribosomal/nucleolar proteins that promote tumorigenesis (Takafuji et al., 2017). Meanwhile, GRWD1 was also viewed as having histone-binding activity and regulating chromatin openness to specific chromatin locations (Sugimoto et al., 2015). Overexpression in colon carcinoma tissues was related to pathological grading, tumor size, N stage, TNM stage, and poor survival; knockdown of GRWD1 function as an inhibitor on cell proliferation and colony formation, and induced cell cycle arrest and more drug susceptibility, and suppressed the migration and invasion (Zhou et al., 2021). GEM interacting protein (GMIP), a RhoA-specific GAP, in a proteomics screen for proteins interacting with Girdin (Girders of actin), an actin-binding protein critical for neuronal migration to the olfactory bulbs, is identified as one of the major regulators of neuronal migration in the postnatal brain (Ota et al., 2014). Solute carrier family 6 member 3 (SLC6A3) involving in the metabolism of dopamine and catecholamine is the potential gene for Parkinson’s disease and alcoholism. The significance of the aforementioned three genes in CLL remains to be further studied.

In GSE14973, the risk score was significantly downregulated after the valproic acid (VPA) treatment in vitro; meantime, protective factors (VEGFA and MNT) were highly expressed, and pathogenic genes (GMIP) were less expressed than in the previous treatment, except TNFRSF25, and these results were almost consistent with our previous conclusion. VPA is a well-tolerated anti-epileptic drug with HDAC inhibitory activity. HDAC1 and HDAC3 inhibition or knockdown results could be figured out in HDAC7 downregulation, which was related to a decline in histone 3 lysine 27 acetylation (H3K27ac) at transcription start sites (TSS) and super-enhancers (SEs) prominently in stem-like BrCa cells. In GSE112953 and GSE15913, the only upregulated gene was LAG3, and it may suggest that combination drug treatment with an anti-LAG3 monoclonal antibody would have a better outcome.

In the present study, a nomogram based on the nine-gene risk score and other clinical traits was constructed, and to determine the predictive effect, we applied the nomogram to a specific patient in the ICGC project; moreover, the predictive model containing the nine-gene risk score was more accurate than the nomogram model containing only four clinical traits. Meanwhile, the risk score was strongly correlated with some known prognostic indicators, such as IGHV mutation state and chromosomal abnormalities. While, a further dissection of the nine-gene risk score on OS in the IGHV mutation state could identify that the nine-gene risk score value was apparent only in the less aggressive M-IGHV subtype, and this predicted trait corresponded to what has been reported in an article which studied the relationship between the ENDOG expression and prognostic study of CLL. The reason why this situation occurred needed further exploration.

The introduction of fludarabine, fludarabine/cyclophosphamide, and either of these combined with rituximab has improved the outcome for younger patients with CLL. Treatment options available for patients in the setting of relapsed disease following receipt of chemoimmunotherapy are limited where most patients have high-risk genomic findings including IgVH un-mutated disease, del (17p13.1) and del (11q22.3) associated with poor treatment response (reviewed in Rassenti et al., 2008). Identifying therapies with novel mechanisms of action for this patient group is important (Johnson et al., 2012). In our research study, all patients were divided into two risk subtypes based on the nine-gene prognostic model, and we endeavored to estimate the drug response of each patient based on IC50 according to the activation of different pathways. ADZ8055 was a dual mTOR kinase inhibitor with inhibition of both mTORC1 and mTORC2 that preferentially decreased cell viability of poor prognostic CLL subsets like with del (11q) or del (17p). One class of drugs that has promise for the treatment of relapsed CLL is the cyclin-dependent kinase (CDK) inhibitors (Seftel et al., 2017). Interestingly, one research study has described that the pan-CDK inhibitor dinaciclib has potent pre-clinical in vitro activity against CLL cells independent of high-risk genomic features (Johnson et al., 2012). In our drug sensitivity prediction, there are three kinds of CDK inhibitors which seemed to be more effective for high-risk CLL patients. The reasons that could account for this difference may include: 1) Different drugs have different mechanisms of action, although they are all one class of inhibitor. 2) The criteria of stratifying patients into “High-risk” and “Low-risk” were not consistent. 3) The most important point is the lack of experimental validation in our research.

To sum up, it was the initial study using the LASSO model to screen prognostic indicators from the profile of SE-associated genes in CLL. A useful prognostic score for OS in untreated CLL patients was presented, and the determination of the score can be achieved via the measurement of the expression levels of nine genes. It also could be done easily in a routine diagnosis. These nine SE-associated genes in this model were not only vital in the development and progression of CLL, but also could assist in guiding the development of alternative treatments.