Pan-cancer evaluation of clinical value of mitotic network activity index (MNAI) and its predictive value for immunotherapy

Increased mitotic activity is associated with the genesis and aggressiveness of many cancers. To assess the clinical value of mitotic activity as prognostic biomarker, we performed a pan-cancer study on the mitotic network activity index (MNAI) constructed based on 54-gene mitotic apparatus network. Our pan-cancer assessment on TCGA (33 tumor types, 10,061 patients) and validation on other publicly available cohorts (23 tumor types, 9,209 patients) confirmed the significant association of MNAI with overall survival, progression-free survival, and other prognostic endpoints in multiple cancer types, including lower-grade gliomas (LGG), breast invasive carcinoma (BRCA), as well as many others. We also showed significant association between MNAI and genetic instability, which provides a biological explanation of its prognostic impact at pan-cancer landscape. Our association analysis revealed that patients with high MNAI benefitted more from anti-PD-1 and Anti-CTLA-4 treatment. In addition, we demonstrated that multimodal integration of MNAI and the AI-empowered Cellular Morphometric Subtypes (CMS) significantly improved the predictive power of prognosis compared to using MNAI and CMS alone. Our results suggest that MNAI can be used as a potential prognostic biomarker for different tumor types toward different clinical endpoints, and multimodal integration of MNAI and CMS exceeds individual biomarker for precision prognosis.


Introduction
A high proliferation rate is one of the hallmarks of cancer (1). Cancer cells acquire the capability to enable chronic proliferation through genetic and epigenetic alterations (2). A few methods have been developed for clinical use to assess proliferative activity, such as the Ki-67 nuclear antigen and mitotic index. Mitotic index, the ratio of the number of cells undergoing mitosis to the total number of cells in a population, has been widely used in clinical practice for tumor grading and staging (3)(4)(5). Moreover, it has been successfully demonstrated to associate with patient survival in many cancer types (3,(6)(7)(8). In clinical practice, mitotic activity is typically evaluated as the number of mitotic figures in the high-power field or per fixed number of tumor cells through immunohistochemical (IHC) staining. With biotechnique advances, we seek a new approach to assess proliferative activity for prognostic impact.
Cell proliferation is tightly regulated by an intricate network of structural proteins, molecular motors, regulatory kinases, and phosphatases for error-free chromosome segregation. The ensemble of such upregulated genes that encode these proteins forms the mitotic apparatus network (9) and shows biological and clinical associations with cancer (6,(10)(11)(12)(13). Furthermore, the association of elevated mitotic activity with the genesis and progression of many cancers encourages the development and evaluation of small molecule inhibitors of mitotic apparatus proteins as anticancer agents (6,14). Despite the clinical trials on cancer treatment, it is also essential to evaluate the prognostic value of mitotic activity in different tumor types for potential clinical implications.
Furthermore, mitotic genes form robust transcriptional network in various tumor types in both human and animal models (15,16), indicating the importance of mitotic network in tumor development. In this study, we define the mitotic network activity index (MNAI) as the sum of the expression of 54 genes pre-identified in the mitotic apparatus network following the same protocol as previously described (9). The use of MNAI can potentially overcome the inter-observer variations due to labor-intensive human counting, and provides an efficient and effective prognostic biomarker with clinical utility. Our previous study has shown that higher MNAI is significantly associated with poor overall survival in breast cancer (9). To further assess the clinical value of MNAI as a prognostic biomarker, we performed a pan-cancer study on MNAI. We demonstrated the association of MNAI with genetic instability at the pan-cancer level; and confirmed its prognostic value across tumor types, which covers the major cancers, including carcinoma, leukemia, lymphoma, melanoma, and tumors related to the central nervous system. Our results suggest that MNAI can be used as a potential prognostic biomarker for different tumor types toward different clinical endpoints.

Biological evaluation of MNAI
The difference in MNAI between tumor and normal samples was evaluated in 25 tumor types using TNMplot (https://tnmplot.com/ analysis/). Alteration frequencies of genetic variants, including mutation, chromosome structural variant, amplification, deep deletion, and multiple alterations, were evaluated in 32 tumor types in TCGA using cBioPortal. The aneuploidy score, mutation count, fraction genome altered, and tumor mutation burden (TMB) were also compared between altered and unaltered groups using cBioPortal.

Clinical evaluation of MNAI
The patients in each cohort were stratified into low-/high-MNAI groups using surv_cutpoint function (with default parameter settings: minprop=0.1) in survminer R package based on overall survival (OS) or progression-free survival (PFS). The prognostic value of MNAI in terms of low and high categories was evaluated using univariate Cox Proportional-Hazards Model (CoxPH) and Kaplan-Meier survival curves towards different prognostic outcomes, including OS, PFS, disease-free survival (DFS), event-free survival (EFS), metastasis-free survival (MFS), relapse-free survival (RFS) and biochemical failure-free survival (BFFS). KMPLOT (http://kmplot.com/analysis/) was used to assess the association of MNAI with beneficiaries from different immunotherapies in pan-cancer patients. Association of MNAI with PD1, PD-L1, CTLA-4 and immune cell infiltrations across tumor types in TCGA was assessed using Spearman's correlation, where the abundance of different cell types were estimated via CIBERSORT in absolute mode (17).

Multimodal integration and evaluation of MNAI and cellular morphometric subtype
The patient cellular morphometric subtype (CMS) was defined by AI-empowered technique based on whole slide images of tissue histology, which has been detaily described in our previous study (18). CMS in LGG and BRCA patients were developed based on cellular morphometric biomarkers mined using machine learning techniques from TCGA-LGG and TCGA-BRCA cohorts. CMSs in these two cohorts have independent clinical values and underlying molecular associations (18,19). Given these pre-established CMSs, the multimodal integration and evaluation were based on multivariate CoxPH model (20). Specifically, concordance index (C-index) was used to evaluate the performance of the CMS, MNAI, and integrated models with 1000 bootstrapping iterations and 80% sampling rate per iteration. Mann-Whitney non-parametric test was used for the comparison across models.

Statistical analysis
Pairwise strategy was used to deal with the missing data in TCGA and other cohorts in our study. For example, the assessment of the association between MNAI and OS was carried out with patients that have both MNAI and OS information, even when the PFS information was missing in some of the patients during OS evaluation. All statistical analyses were performed in R (version 3.6.0). The univariate and multivariate CoxPH model (survival package in R, version 3.2-7) and Kaplan-Meier analysis (survminer package in R, version 0.4.9) were used to assess the prognostic value of MNAI. Mann-Whitney nonparametric twotailed test was used to test the statistical difference for continuous variables with a significant threshold of p<0.05.

MNAI is significantly elevated in tumors compared to normal samples
One of the major characteristics of cancerous tissue is rapid proliferation, where an increased proportion of cells undergo mitosis compared to normal ones. To confirm the power of MNAI in differentiating cancerous tissues from normal tissues, we evaluated the MNAI abundance among 25 different tumor types ( Figure 1). Unsurprisingly, our results indicated significantly elevated MNAI in cancerous tissues among all tumor types (p< 0.0001).

The MNAI significantly associates with genetic instability
Genomic instability is a common characteristic of most cancers, as reflected by the increased tendency of genome alteration during cell division, and is associated with increased mitotic activity (10). Our thorough evaluation based on 32 tumor types in TCGA revealed a dynamic genetic alteration frequency in MNAI genes per tumor type, including mutations, chromosome structural variants, amplifications, deep deletions, and multiple alterations. This ranged from around 80% in lung squamous cell carcinoma (LUSC) to less than 10% in thyroid carcinoma (THCA) (Figure 2A), Not surprisingly, MNAI in altered samples is significantly higher than in unaltered samples (Figure 2A, inserted panel), suggesting genetic alteration is one possible mechanism for elevated MNAI. Furthermore, significantly different aneuploidy scores (p<1.0E-15, Figure 2B), mutation counts (p<1.0E-15, Figure 2C), fraction genome altered (p<1.0E-15, Figure 2D), and tumor mutation burden (TMB, p<1.0E-15, Figure 2E) were detected at the pancancer level between altered and unaltered patient groups with respect to MNAI genes. Such significant difference were also observed between high and low MNAI groups within both altered and unaltered patients (Supplementary Figure 1), where the MNAI high/low stratification was optimized towards OS at pan-cancer level. Many existing studies have demonstrated that alteration of mitotic genes, such as AURKA, PLK1, and BUB1, induce replicative stress and mis-segregation of the chromosome, which cause aneuploidy (21,22). Specially, one previous study showed that dysregulated transcription of MNAI genes happen prior to DNA copy number alterations (23). Together with our findings, we conclude that alteration in MNAI genes is associated with genomic instability in human cancer.

MNAI significantly associates with prognosis
Increased mitotic activity is a hallmark of cancer aggressiveness (9), and therefore associated with the prognosis of cancer patients. The evaluation and validation of MNAI on 33 different tumors confirm its significant prognostic value in the pan-cancer landscape. Specifically, in TCGA pan-cancer cohort (Supplementary Table 1), at pan-cancer level, both genetic alteration within MNAI genes ( Figure 3A) and MNAI ( Figure 3B) are significantly associated with PFS and OS. Importantly, MNAI overrode genetic alteration ( Figure 3C) in pan-cancer prognosis. Therefore, our remaining study focused on MNAI. Consistent with previous studies, genomic instability factors significantly correlated with prognosis ( Supplementary Figure 2), and our multivariate analysis indicates the independent and superior prognostic power of MNAI compared with genomic instability factors ( Figure 3D).

MNAI predicts the beneficial effects of immunotherapy
An increasing number of studies show that tumor mutation burden (TMB) is a predictive biomarker for immunotherapy (25,26). Therefore, we assessed whether MNAI predicts the beneficial effects of immunotherapy. KMPlot analysis showed that cancer patients with high MNAI had significantly prolonged PFS after Association of MNAI with genomic instability. (A) Dynamic genetic alteration frequency in MNAI genes per tumor type in TCGA, including mutation, chromosome structural variant, amplification, deep deletion, and multiple alterations, where patient with any of above changes with at least one MNAI gene was classified into altered group, otherwise the patient was classified into unaltered group; (B-E) Association between patient groups with respect to altered and unaltered MNAI genes with aneuploidy score (B), fraction genome altered (C), mutation count (D) and tumor mutation burden (E). cBioportal was used for the analysis, where Mann-Whitney non-parametric test was used to test the difference between groups. anti-PD-1 (p = 0.00021, Figure 6A) and anti-CTLA-4 (p = 0.00071, Figure 6B) treatment but not after anti-PD-L1 treatment (p = 0.22, Figure 6C). Similar observations were found for OS ( Figures 6D-F).
Utilizing the TCGA cohorts, significant association (p<0.05) was observed between MNAI and PD1, PD-L1, CTLA-4, and immune cell infiltrations found in a majority of tumor types (Supplementary Figure 6), which can explain its predictive powers towards certain immunotherapies. Altogether, these findings suggest that MNAI can potentially serve as a predictive biomarker for selecting patients for certain types of immunotherapies.

Multimodal integration of MNAI and CMS significantly improve the predictive power of prognosis
Tissue histology sections are rich in content, displaying the different cell types and overall tumor microenvironment of a given sample, and remain as the gold standard for the assessment of tissue neoplasm. To capture the cellular heterogeneity and cellular context in tissue histology, we previously defined patient subtypes as CMS using machine learning techniques from whole slide images and demonstrated its significant and independent clinical values in various tumors (18,19). Given the translational potential of CMS and MNAI, we assessed whether the multimodal integration of

Discussion
In this study, we assess the prognostic and predictive value of MNAI across multiple types of human cancer. Our results showed that MNAI is consistently overexpressed in all types of cancer in comparison to their corresponding normal tissues, consistent with one of the most prominent characteristics of a cancer cell: constant proliferation. The previous study in breast cancer has shown that the upregulation of MNAI is controlled by transcription factors, including MYC (9), which may hold true over a broad range of tumors. Additionally, high MNAI values significantly correlate with increase in genome instability (aneuploidy score, fraction genome alteration, mutation count, and tumor mutation burden). It is possible that high MNAI causes mitotic errors in cell divisions, leading to the loss or gain of chromosomes and, consequently, to aneuploidy. Moreover, many MNAI genes are included in the chromosomal instability gene signature (13). These results elucidate mitotic genes' potential role in genome instability and subsequent contribution to cancer development and progression.
We showed that high MNAI value in human tumors predicts poor patient outcomes, including DFS and OS in the majority of cancer types. However, we also found a favorable prognosis associated with high human tumor MNAI values in a few types of cancer. This is probably due to different first-line chemotherapies that were used in different cancer types. Patients with high MNAI only benefit from certain kinds of chemotherapies. For example, aneuploidy can make cells sensitive to certain chemotherapies but also resistant to others (27,28). Thus, we propose that clinical implementation of the MNAI could contribute to precision cancer medicine by choosing more effective therapeutic regimens. In recent years, immunotherapy has raised tremendous attention. Many clinical studies have shown that immune checkpoint blockade therapy can significantly improve the OS rates of patients in many types of cancer, but only in a subset of patients with cancer (29). As a result, predictive biomarkers for checkpoint inhibitor-based immunotherapy have been extensively studied, including tumorinfiltrating lymphocytes, mutational burden, immune gene signatures, and multiplex immunohistochemistry (30). In this study, we found that patients with high MNAI benefit significantly more from anti-PD-1 and Anti-CTLA-4 treatments but not from anti-PD-L1 treatment. The resistance of patients with high MNAI to anti-PD-L1 treatment is possibly due to the contribution of mitosis in cancer cells to the high expression of PD-L1 (31). These patients may potentially benefit from combined anti-mitotic and anti-PD-L1 treatments. Thus, we suggest that MNAI could serve as a novel predictive biomarker for specific checkpoint inhibitor-based therapies. Furthermore, our analysis using CancerDR database (32) revealed significant association between MNAI and many molecular targets, indicating it as a potential biomarker for predicting the response to more cancer treatment targets besides PD-1 and CTLA-4 (Supplementary Figure 8 lower pane, spearman correlation |R|>0.3 and p value< 0.05).
In addition, CMS extracted from whole slide images of tissue histology through advanced machine learning techniques (33) has emerged as a novel imaging biomarker in various tumor types, providing independent clinical values (18,19). In this study, we found that the multimodal integration of MNAI and CMS exceeds the individual biomarker in precision prognosis. Thus, we suggest that MNAI and CMS could serve as a novel multimodal prognostic biomarker for improved patient risk stratification.

Conclusions
In conclusion, through pan-cancer assessment, we demonstrated the prognostic and predictive power of MNAI in patients with different types of cancer and its association with genetic instability. Moreover, we believe that the clinical implication of MNAI and its multimodal integration with CMS will contribute to the precision prognosis and personalized treatment of cancer patients. In addition, the clinical utility of MNAI can be easily developed and implemented with an assay that simultaneously measure all 54 genes via different techniques, such as NanoString. The clinical utility of its multimodal integration can be achieved based on the MNAI assay as well as the adoption of the machine learning pipeline in clinical practice. In our future study, we will focus on validating MNAI as a predictive biomarker using the MNAI assay we developed for immunotherapy or PARP inhibitors and others through clinical collaborations to maximize its clinical impact.

Data availability statement
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding author.