Copy number alteration of the interferon gene cluster in cancer: Individual patient data meta-analysis prospects to personalized immunotherapy

Interferon (IFN) therapy has been the standard of care for a variety of cancers for decades due to the pleiotropic actions of IFNs against malignancies. However, little is known about the role of copy number alteration (CNA) of the IFN gene cluster, located at the 9p21.3, in cancer. This large individual patient data meta-analysis using 9937 patients obtained from cBioportal indicates that CNA of the IFN gene cluster is prevalent among 24 cancer types. Two statistical approaches showed that notably deletion of this cluster is significantly associated with increased mortality in many cancer types particularly uterus (OR = 2.71), kidney (OR = 2.26), and brain (OR = 2.08) cancers. The Cancer Genome Atlas PanCancer analysis also showed that CNA of the IFN gene cluster is significantly associated with decreased overall survival. For instance, the overall survival of patients with brain glioma reduced from 93m (diploidy) to 24m (with the CNA of the IFN gene). In conclusion, the CNA of the IFN gene cluster is associated with increased mortality and decreased overall survival in cancer. Thus, in the prospect of immunotherapy, CNA of IFN gene may be a useful biomarker to predict the prognosis of patients and also as a potential companion diagnostic test to prescribe IFN α/β therapy.


Introduction
The type-I interferons (IFN) are cytokines which play essential roles in inflammation, immunoregulation, tumor cells recognition, and T-cell responses [1] . From the 1980s onward, members of type-I IFN family have been the standard care as immunotherapeutic agents in cancer therapy Abbreviations: CAN, copy number alteration; IFN, interferons. [ 2 , 3 ]. In particular, IFN α has been approved by the US Food and Drug Administration (FDA) for cancer. To date, pharmaceutical companies produce several types of recombinant and pegylated IFN α for clinical use; e.g., IFN α2a (Roferon-A, Roche), IFN α2b (Intron-A, Schering-Plough) and pegylated IFN α2b (Sylatron, Schering Corporation) [2] for treatment of hairy cell leukemia, melanoma, renal cell carcinoma, Kaposi's sarcoma, multiple myeloma, follicular and non-Hodgkin lymphoma, and chronic myelogenous leukemia [ 2 , 4 , 5 ]. Human IFN β (Feron, Toray ltd.) has also been approved in Japan to treat glioblastoma, medulloblastoma, astrocytoma, and melanoma [5] .
Despite the pleiotropic (e.g., antiangiogenic, immunomodulatory, differentiation-inducing, antiproliferative, and proapoptotic) actions of IFN α against malignancies [ 2 , 6 ]. It is still unclear why IFN α treatment is only effective in a subtype of patients (e.g. , melanoma), whilst might promote tumor progression in another subset [7] . Therefore, the administration of IFN α has been later surpassed by more effective and less toxic agents. For example, drugs such as thalidomide, lenalidomide and bortezomib are more recommended for myeloma treatment [ 2 , 6 ], while imatinib and tyrosine kinase inhibitors are more prevalently used for chronic myelogenous leukemia [ 2 , 6 ]. Additionally, vascular-endothelial growth factor (VEGF) and mammalian target of rapamycin (mTOR) inhibitors supersede for treatment of renal cell carcinoma. Nevertheless, no other treatment has shown superior efficacy to IFN α in the adjuvant phase of malignant melanoma yet [2] . Recently, the use of type-I IFNs has attracted attention once more in cancer therapy [ 8 , 9 ]. It has been shown that type-I IFNs have a synergistic effect on checkpoint blockade and adoptive T-cell immunotherapies by increasing the proliferation and cytotoxicity of T-cells, the maturation and cross-priming capacity of dendritic cells and stimulating NK cells to kill tumor cells [ 8 , 9 ]. In addition, advances in immunotherapy led to the emergence of cancer immunoediting which can serve as a framework to re-evaluate the IFN α's immunological role in tumor development and immunotherapy [1] . For example, in 2020, a phase I/II clinical trial showed that adoptive cell therapy with tumor-reactive T cells in combination with a mild IFN α regimen could increase the median overall survival in metastatic refractory melanoma patients from non-responders (7 months) to responders (36 months) [10] .
In the human genome, a cluster of thirteen functional IFN genes is located at the 9p21.3 cytoband over approximately 400 kb including coding genes for IFN α ( IFNA1, IFNA2, IFNA4, IFNA5, IFNA6, IFNA7, IFNA8, IFNA10,  IFNA13, IFNA14, IFNA16, IFNA17 and IFNA21 ), IFN ω ( IFNW1 ), IFN ɛ ( IFNE ), IFN к ( IFNK ) and IFN β ( IFNB1 ), plus 11 IFN pseudogenes [ 2 , 11 ]. Among 19 cancer types, prevalent homozygous deletion of IFN gene cluster has been observed in high frequencies (7-31%) indicating that deletion of type-I IFN genes exacerbates overall or disease-free survival rates [11] . Defects in interferon signaling pathways have been suggested to induce mechanism of resistance to immunotherapy in prostate cancer cell lines [12] i.e., copy number deletion of IFN genes activates oncogenic pathways and repress immune signaling pathways by both promoting tumorigenesis and helping tumor cells to evade immunosurveillance [11] . Copy number deletion of the IFN gene cluster may also leads to a worse prognosis in melanoma patients [13] and could be useful as a prognostic marker to predict resistance to immunotherapy (e.g., anti-CTLA4 treatment) [11] . Consequently, it is suggested that individuals with deletions may benefit from combinations of IFN α with T cell-directed therapies [13] .
In this study, we investigate the association of copy number alteration (CNA) of the IFN gene cluster with the mortality and survival of patients with different cancer types. This individual patient data meta-analysis aims to predict which CNA subtype in different cancers could benefit from interferon α/ β therapy and fills a gap in our understanding of type I interferon gene copy number on cancer progression and treatment.

Methods
The individual patient data meta-analysis is based on data obtained from the cBioportal database ( www.cbioportal.org) containing published and unpublished data mostly from The Cancer Genome Atlas (TCGA) [ 14 , 15 ]. Only individuals without a prior history of cancer, complete information on the CNA of the IFN gene cluster, and 5-y follow-up were included in this study.

Data extraction
Datasets of demographics, clinical information related to cancer (including anatomical location and histological subtype), and cancer genomics have been extracted for all individuals ( www.cbioportal.org / faq.jsp) [ 14 , 15 ]. Data obtained from cBioportal for the CNA categories were computed using Genomic Identification of Significant Targets in Cancer (GISTIC) version 2.0 [16] ; this method has been described in more detail previously [17] .

Data analysis
A meta-analysis of individual patient data was conducted in Stata/MP14.2 (StataCorp, USA). Two meta-analytical approaches were used to determine the stability and consistency of the results. The main outcome was 5-y mortality for each anatomical location of cancer, expressed as odd ratios (OR) and 95 % confidence intervals (CI), using diploidy (normal) as the reference group [18] . Descriptive statistics are shown as the number of individuals and proportions (%). In order to maintain sufficient statistical power, shallow (-1) and deep (-2) deletion was combined for the meta-analyses, as well as gain ( + 1) and amplification ≥+ 2); respectively referred to as "deletion" and "amplification." ORs above 1 imply a higher risk of mortality in the deletion/amplification group, compared to the reference; whilst lower values propose a protective effect. If a 95% CI includes the value of 1 (indicating no difference), differences with the reference group are statistically insignificant.
The first meta-analytic approach was based on random effect modeling using the ipdmetan package in Stata. This two-stage individual patient data meta-analysis pools and visualizes the effect of CNA on the risk of death (yes or no) within 5 y after diagnosis, weighted for the different anatomical locations, and the results are presented as forest plots [19] . I-squared statistics were used to quantify statistical heterogeneity, with values < 50%, 50% to 75%, and > 75% defined as low, moderate, and high heterogeneity, respectively [20] . This approach does not allow for adjustment for confounding or interaction. Therefore, a second meta-analytic method was implemented, multivariable logistic regression analyses (one-step approach) [18] . For each anatomical location, three models were used. Model 1 was crude (unadjusted); Model 2 was adjusted for sex, age, and calendar period; and Model 3 was additionally adjusted for interaction with tumor stage. Analyses were only presented if at least 10 individuals were included in each risk group, and subgroups with zero deaths were omitted. All analyses are based on complete-case analyses.

Survival analysis
Information regarding the association of CNA of IFN gene cluster and patients' survival time in cancer was analyzed using TCGA PanCancer Atlas Studies database, available at cBioportal. Only patients with the CNA of the IFN gene cluster and overall survival data ( n = 10,712) in 32 cancer type were analyzed to observe the prognostic value using Kaplan-Meier Plots. All reported P < 0.05 were considered statistically significant.
Age distribution was similar in all groups, yet individuals with advanced tumor stage (III-IV) were overrepresented in the group with copy number deletion (20% and 8%) and amplification (20% and 10%) compared to patients carrying diploidy (12% and 5%).
In 6 of the 24 anatomical locations, the large majority of the cancers presented either deletion or amplifications, with diploidy only present in 37% (head and neck cancer), 31% (bladder), 32% (ovaries), 30% (esophagus), 27% (skin), and 27% (lung). Table 2 Distribution of deletion, diploidy and amplification by anatomical location, indicating which proportion of each location has normal or abnormal copy number alterations (CNA).

Clinical characteristics
All patients having heterogeneous CNAs within the IFN gene cluster and those with a prior malignancy or incomplete 5-y follow-up information were excluded. In total, 24 different anatomical locations were reported with breast (15%), and brain tumors (9%) being most common.
The one-step approach provided similar results, and all three models were presented for each anatomical location if feasible ( Table 3 ). For deletion, model-2 showed a significantly increased 5-y mortality for cancers of the uterus (OR = 3.35), kidney (OR = 2.82), brain (OR = 2.46), and liver (OR = 1.72). After full adjustment (model 3, in half of the cohort with complete tumor stage), none of the results was confirmed; however, deletion remained significantly associated with a worse prognosis in the total cohort (OR = 2.41) ( Table 3 ). For amplification, model-2 found significant associations for cancers of the brain (OR = 4.12), uterus (OR = 3.11), colorectal (OR = 2.52), and head and neck (OR = 1.82) cancers; again, not confirmed in model 3 ( Table 3 ).

Overall prognosis
At 5 y after diagnosis, 27% of patients have died, 46% were diseasefree, and 11% had a recurrence but were still alive. Recurrence information was missing in 11% of individuals who survived. Of those who died, 20% presented with IFN gene cluster diploidy. Of those who were alive, 51% presented diploidy ( P < 0.00001) ( Table 1 ).
The one-step meta-analysis approach was used to assess if the effects of IFN gene cluster ploidy on mortality remained after adjustment for confounding and interaction using diploidy as a reference. The unadjusted 5-y mortality (model 1, n = 9937) showed similar results as above, with respectively 256% (OR = 2.56, 95% CI 1.74-3.76) and 174% (OR = 1.74, 95% CI 1.24-2.43) increased risks of death for deletion and amplification compared to the diploid group ( Table 3 ). After adjustments for age, sex, calendar period, and clustering by study (model 2, n = 7666), the results remained stable yet lost significance. Since the interaction between IFN cluster ploidy and tumor stage was present ( P = 0.0031). Model 3 ( n = 4859) is additionally adjusted for interaction with tumor stage, resulting in doubled immunotherapy A. Razaghi et al. Table 3 The 5-y mortality associated with copy number alterations (CNA) per anatomical location, calculated by multivariable logistic regression and presented as number (%) and odds ratio's (OR) and 95% confidence intervals (CI).  Testicular  52  100 0  0  59  97  2  3  17  100 0  0  ------------Thymus  6  100 0  0  92  95  5  5  8  89  1  11  ------------Thyroid  18  90  2 Fig. 2. Forest plots assessing the association between IFN gene cluster. (A) Amplification and (B) Deletion. 5-y mortality per cancer type using diploidy as a reference. The numbers in the columns refer to the total number of individuals presenting with each cancer type, and the number who died within 5 y, for the total group of individuals, those with diploidy (reference) and those with amplification (A) or deletion (B). The diamond represents the average of studies. Weights are derived from a random-effects model. CI, confidence interval; OR, odds ratio. risks among those with deletion (OR = 2.41, 95% CI 1.66-3.51), yet no difference for those with amplification (OR = 1.01, 95% CI 0.54-1.89) ( Table 3 ).

Survival
Screening of available data of 10,712 patients from TCGA combined PanCancer Atlas using cBioportal shows 7% CNA of IFN gene cluster in a total of 32 cancer types included in the study. In addition, overall, diseasefree, progression-free, and disease-specific survival in groups with CNA of IFN gene cluster are significantly decreased to 24, 100, 17, and 32 months (median), respectively ( Fig. 3 ).
Furthermore, overall survival is significantly altered in 6 out of 35 cancer types due to the CNA of the IFN gene cluster in patients. The overall survival is decreased in cholangiocarcinoma ( Fig. 4 A), liver hepatocellular carcinoma ( Fig. 4 B), glioblastoma multiforme ( Fig. 4 C), brain low-grade glioma ( Fig. 4 D) and mesothelioma ( Fig. 4 E). In contrast, the overall survival is increased in uterine corpus endometrial carcinoma ( Fig. 4 F).

Discussion
This study shows consistent results using two statistical approaches, and a Pan-Cancer analysis. The CNA of the IFN gene cluster (both in the deletion and amplification forms) was associated with increased mortality of cancer patients. In addition, deletion of the IFN gene cluster was more prevalent than diploidy and amplification. Patients with deletion of the IFN gene cluster show higher mortality in 9 out of 24 cancer types including the brain, breast, cervical, kidney, liver, lung, mesenchyme, mesothelioma, thyroid, and uterus. While patients with amplification of the IFN gene cluster show higher mortality in 6 cancer types including brain, breast, colorectal, head and neck, mesenchyme, and uterus. No association was found for the other cancer types, but power may have been too limited. In a total of 35 cancer types provided by TCGA/cBioportal, survival (overall, diseasefree, progression-free, and disease-specific) in patients with CNA of IFN gene cluster are less in comparison to unaltered individuals. In particular, the overall survival for cholangiocarcinoma, liver hepatocellular carcinoma, glioblastoma multiforme, brain low-grade glioma, and mesothelioma is less, yet higher mortality was found for liver, brain, and lung cancers.
This large individual patient data meta-analysis is based on the cBioportal and TCGA with high quality/valid data. The cBioportal is an open-access resource for interactive exploration of cancer omics data-sets including TCGA, empowering researchers to translate these valuable data-sets into biological insights and clinical applications [14] . However, our study may have limited power for the different histological subtypes, and possible selection bias since some cancer types may be over-represented in the cohort compared to cancer distribution in the total population. We could adjust for age, sex, calendar period and to a certain extent for tumor stage (despite a large amount of missing value) but other "residual" confounders may play a role as well i.e., these results indicate associations, but not causations.
Nevertheless, deletion of the IFN gene cluster might play a more deleterious role in patients due to the lower level of type-I interferon expression leading to less immunosurveillance in the tumor microenvironment. Because type-I IFNs exerts their anti-tumor activity through driving the high maturation status of dendritic cells, impacting cytotoxic T lymphocytes and NK cell activation, inducing tumor cell death and inhibiting angiogenesis [21] , i.e., lower expression of type-I IFNs can impede immunosurveillance.
Among the cancer types which IFN α/ β therapy has been approved for, only mortality of kidney and brain cancer types was associated with CNA of IFN gene cluster. In contrast, the mortality of hematological malignancies (e.g. , multiple myeloma, follicular and non-Hodgkin lymphoma, and chronic myelogenous leukemia) was not associated with CNA of IFN gene cluster and skin cancer had limited power. It has already been shown that subset of melanoma patients showing deletion of IFN gene cluster is resistant to anti-CTLA4 immunotherapy proposing that deletion of IFN gene cluster can be used a prognostic biomarker for immunotherapy resistance [11] . In this regard, phase II clinical trial of combinational therapy of anti-CTLA4 (Ipilimumab) with recombinant IFN α in treating melanoma patients is currently undergoing (ClinicalTrials.gov Identifier: NCT01708941) [22] suggesting that such a regimen might have potential benefit in cancer types with CNA of IFN gene cluster (e.g., kidney and brain cancers). For example, two phase II clinical trials of IFN α with temozolomide in patients with recurrent glioblastoma multiforme demonstrated improvement in six months progression-free survival outcomes [23] .
In 2018, anti-CTLA4 (ipilimumab) plus nivolumab was approved by the FDA for renal cell carcinoma treatment [24] . Our results also show that a subset of patients with deletion of IFN gene cluster (23%) in kidney cancer has a significantly higher mortality. Thus, deletion of the IFN gene cluster might be potentially used as a prognostic biomarker for anti-CTLA4 immunotherapy resistance in renal cell carcinoma as well. Furthermore, in the phase II clinical trial, a combination of IFN α and chemotherapy (Oxaliplatin/Adriamycin/5-Fluorouracil) showed manageable toxicity and improved survival of patients with advanced hepatocellular carcinoma [25] . Therefore, deletion of the IFN gene cluster in liver cancer might also be used as a potential prognostic biomarker for IFN α therapy in liver cancer.
In general, the new strategies for application of interferons are included to the in-vivo use of IFN α as immune adjuvants of cancer vaccines, and the combination of certain chemotherapies with IFN α adjuvanted cancer vaccines [26] . Personalized medicine/ immunotherapy is another emerging area for application of interferons, fostering the development of specialized treatments for each subtype of cancer, based on the measurement and exploitation of patients' omics data (e.g., genomics, transcriptomics, metabolomics, proteomics) [ 27 , 28 ]. In this light, understanding the alteration of IFN gene cluster helps to predict the resistance to cancer therapy outcome as a prognostic biomarker in companion diagnostics. Particularly, for cancer types such as brain, kidney, skin, and hematologic malignancies which IFN α/ β therapy is already in clinical use. Furthermore, the information about alteration of IFN gene helps to expand the clinical application of interferon α/ β therapy in the type of cancers showing higher mortality associated with CNA of the IFN gene cluster e.g., breast, and uterine cancers.

Conclusions
This large individual patient data meta-analysis indicates that CNA of the IFN gene cluster is prevalent in cancer. Two statistical approaches showed that notably amplification and deletion of the IFN gene cluster are significantly associated with increased mortality in at least 6 and 9 cancer types, respectively. PanCancer TCGA analysis using cBioportal also showed CNA of IFN gene cluster is significantly associated with decreased survival in liver, brain, and mesothelioma cancers. Therefore, CNA of the IFN gene cluster can be suggested as a useful biomarker to predict the prognosis of patients (e.g. , liver and renal cancers) and also as a potential companion diagnostic test to prescribe IFN α/ β therapy and predict the outcome of immunotherapy for clinical use.