*Article* **Radiation Type- and Dose-Specific Transcriptional Responses across Healthy and Diseased Mammalian Tissues**

**Eftychia Sagkrioti 1,2, Gökay Mehmet Biz 3,†, I¸sıl Takan 4,5,†, Seyedehsadaf Asfa 4,5, Zacharenia Nikitaki 1, Vassiliki Zanni 1, Rumeysa Hanife Kars 6, Christine E. Hellweg 7, Edouard I. Azzam 8, Stella Logotheti 1, Athanasia Pavlopoulou 4,5 and Alexandros G. Georgakilas 1,\***


**Abstract:** Ionizing radiation (IR) is a genuine genotoxic agent and a major modality in cancer treatment. IR disrupts DNA sequences and exerts mutagenic and/or cytotoxic properties that not only alter critical cellular functions but also impact tissues proximal and distal to the irradiated site. Unveiling the molecular events governing the diverse effects of IR at the cellular and organismal levels is relevant for both radiotherapy and radiation protection. Herein, we address changes in the expression of mammalian genes induced after the exposure of a wide range of tissues to various radiation types with distinct biophysical characteristics. First, we constructed a publicly available database, termed RadBioBase, which will be updated at regular intervals. RadBioBase includes comprehensive transcriptomes of mammalian cells across healthy and diseased tissues that respond to a range of radiation types and doses. Pertinent information was derived from a hybrid analysis based on stringent literature mining and transcriptomic studies. An integrative bioinformatics methodology, including functional enrichment analysis and machine learning techniques, was employed to unveil the characteristic biological pathways related to specific radiation types and their association with various diseases. We found that the effects of high linear energy transfer (LET) radiation on cell transcriptomes significantly differ from those caused by low LET and are consistent with immunomodulation, inflammation, oxidative stress responses and cell death. The transcriptome changes also depend on the dose since low doses up to 0.5 Gy are related with cytokine cascades, while higher doses with ROS metabolism. We additionally identified distinct gene signatures for different types of radiation. Overall, our data suggest that different radiation types and doses can trigger distinct trajectories of cell-intrinsic and cell-extrinsic pathways that hold promise to be manipulated toward improving radiotherapy efficiency and reducing systemic radiotoxicities.

**Keywords:**radiation response; bioinformatics; oxidative stress; transcriptomics; radiobiology database; gene signature

#### **1. Introduction**

Radiation therapy has witnessed unprecedented advances during the last decades, asserting its place as a major part of everyday clinical practice [1]. It contributes to ~40%

**Citation:** Sagkrioti, E.; Biz, G.M.; Takan, I.; Asfa, S.; Nikitaki, Z.; Zanni, V.; Kars, R.H.; Hellweg, C.E.; Azzam, E.I.; Logotheti, S.; et al. Radiation Type- and Dose-Specific Transcriptional Responses across Healthy and Diseased Mammalian Tissues. *Antioxidants* **2022**, *11*, 2286. https://doi.org/10.3390/ antiox11112286

Academic Editors: Elena Obrador Pla and Alegria Montoro

Received: 17 August 2022 Accepted: 15 November 2022 Published: 18 November 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

of curative cancer treatments [2], alone or in combination with chemotherapy [3], and tends to be less morbid than surgery [4]. In addition to its direct cytotoxic effects on the targeted tumors, irradiation often triggers indirect localized and systemic responses. These responses are not only occasionally linked with early or late adverse side effects proximal or distal to the treatment site but can also be beneficial for patient outcomes. Intriguingly, recent studies show that radiotherapy induces *bona fide* immunogenic cell death and engages tumor-targeting immune responses in support of enhancing treatment efficacy. Local irradiation reshapes the tumor microenvironment (TME) by promoting prooxidant and proinflammatory reactions, which may trigger adaptive immune responses [1]. Stressed and dying irradiated cells release numerous bioactive molecules, for example, major histocompatibility complex, cell-adhesion molecules, and proinflammatory cytokines and their receptors, as well as molecules with damage-associated molecular patterns (DAMPs), small metabolites, nucleic acids and lipids. These tumor-associated antigens interact with the immune system to induce immunogenic cell death [5,6] since they are taken up by the dendritic cells and stimulate downstream effector T cells, which subsequently recognize and lyse tumor cells both locally and at distant sites [7]. In several clinical cases, tumors distal to the targeted site regressed in response to irradiation-induced immunogenicity, a phenomenon termed as an abscopal effect [7]. In this respect, the irradiated cells act as in situ vaccines against tumors, sensitizing the immune system to detect cancer cells even long after the completion of radiation treatment. Hence, systemic effects of radiotherapy may act as a "blessing in disguise" due to their potential to ally with the immune system and increase responses that control the growth of micrometastases and malignant tissues at distant sites. However, the effects may also be a "curse" resulting in the suppression of antitumor immunity by mechanisms involving regulatory T cells [8].

The newly-discovered immunomodulatory properties of radiation have been linked with its ability to primarily activate the DNA damage response and repair (DDR/R) machinery. DDR/R is a highly conserved and complex network of signal transduction pathways that preserves the genetic information by repairing a variety of DNA lesions, such as nucleotide alterations, bulky adducts, single-strand breaks (SSBs) and doublestrand breaks (DSBs). These pathways can be lesion-specific, for example, non-homologous end joining (NHEJ) and homologous recombination (HR) repair for DSBs; single-strand break repair (SSBR) for nicked DNA strands; mismatch repair (MMR) for errors that occurred during replication; base excision repair (BER) for oxidative base modifications; and nucleotide excision repair (NER) for helix-distorting lesions [9]. The stimulation of different components of DDR/R, either endogenously or from external sources, such as exposure to ionizing radiation (IR), alerts host immunity at the systemic level and vice versa [10], thereby accounting for the intriguing immunogenic properties of irradiated cells. These novel concepts have rejuvenated clinical interest to exploit this dynamic and bidirectional crosstalk between DDR/R and immune response (ImmR) signaling and manipulate it towards personalized radiotherapeutic solutions.

There are several types of therapeutic modalities, classified according to radiation quality associated mainly with the linear energy transfer (LET), a parameter accounting for the amount of energy deposited per unit length of the irradiating particle's path. Low-LET radiation entails the more frequently used γ- and X-rays, while high-LET refers to protons, carbon ions and α-particles that capitalize on the physical and radiobiological properties of charged particles for an improved dose distribution and increased cell killing efficacy. Carbon ions kill cells twice or three times more effectively than protons and conventional radiation modalities [11]. In general, high-LET types induce more DSBs per dose unit, and more complex and dense lesions than low-LET types because they deposit large amounts of energy within a small distance [12]. The type of initial DNA damage largely determines the repair pathway that is subsequently activated. For example, heavy ions preferentially shift towards DSB repair pathways, such as HR and NHEJ, when compared with sparsely ionizing irradiation [1]. Given that a different type of DNA damage can trigger different DDR components, which in turn are associated with the release of

immunostimulatory neoantigens as "danger signals" (i.e., DAMPs), it is reasonable to envisage that each of these irradiation types governs distinct trajectories of DNA damage type—DDR pathway—immunogenic responses, which, however, to date, have not been identified [1]. In this regard, understanding the major differences of low- and high-LET treatment options is a current challenge of radiotherapy, not only for minimizing side effects, but also for making the most of each modality toward stimulating tumor-targeting adaptive immunity post-irradiation.

The effects of the various radiation types are mediated, at least partly, through changes in the transcriptomes of the irradiated cells. In general, different types of radiation trigger distinct gene transcription programs associated with divergent cellular responses both in cancer and normal cells. Although radiation type-specific transcriptional changes have been examined sporadically [13–18], to our knowledge, there is no systematic effort to characterize the effects of several high-LET or low-LET radiation types and doses of radiation in normal or diseased tissues, which would set a basis to untangle their side effects from their beneficial cytotoxic and immunogenic properties. Simultaneous screening of the transcriptomes across irradiated cancer and normal tissues would require largescale experiments for each radiation type and/or dose. Furthermore, due to the genetic heterogeneity of cells in irradiated tissues, which is a major parameter of the efficacy of radiotherapy, extensive testing on a large variety of tissue contexts is required, transforming this effort to a "Herculean task".

As a "*deus ex machina*", computational approaches have entered the stage of radiobiology to accelerate and complement these efforts. In the present work, we constructed a publicly available, user-friendly database, termed RadBioBase version 1 (http://radbiodb.physics.ntua.gr/), which includes a collection of up-to-date existing data on mammalian genes differentially expressed after exposure to different types (X-rays, γ-rays, protons, carbon ions and α-particles) and doses of radiation in a variety of cell types. This database is a comprehensive tool for correlations of radiation type and/or dose with corresponding transcriptional responses across a variety of tissues. Following an integrated bioinformatics approach that included gene-centric, pathway-oriented and machine learning analyses, we consolidated the IR-induced differential gene expression to biological pathways and human diseases. In addition, we identified gene signatures for different radiation types. Our analyses provide insights into the links between the IR-induced damage and the signal propagation of stress to distant sites, and hold promise for a deeper understanding of the association between DDR and the immune system to a wider context, in a coordinated multiscale manner, which could be translated to more efficient and safer radiotherapy schemes.

#### **2. Materials and Methods**

#### *2.1. Data Hybrid Collection and Transcriptomic Analyses*

A broad collection of genes was initially obtained by rigorous text mining of the bibliographic database MEDLINE/PubMed 2.0 (https://pubmed.ncbi.nlm.nih.gov/, accessed on 15 March 2021), with the use of keywords related to X-ray, γ-ray, proton, carbon ion and α-particle irradiation, i.e., ((gamma radiation) OR (gamma rays) OR (γ rays)) AND gene expression; ((proton(Title/Abstract)) AND (radiation(Title/Abstract))) AND (gene expression(Title/Abstract)); ((carbon(Title/Abstract)) AND (radiation(Title/Abstract))) AND (gene expression(Title/Abstract)) from 1 January 2006 to 30 August 2021. The articles were independently retrieved from the literature by three of the authors (E.S, R.H.K. and A.P.). Relevant data were extracted from the articles and recorded into an Excel worksheet.

For the articles to be considered eligible for inclusion in our study, they had to report the following information: (i) tissue/cell line, (ii) cell type (cancer or normal), (iii) model organism, (iv) type of irradiation, (v) irradiation exposure time, (vi) dose amount, (vii) availability of data regarding genes differentially expressed between irradiated and non-irradiated (control) cells/tissues, or sufficient data to calculate differential gene expression. To minimize investigator biases, compliance of the screened articles with the study eligibility criteria was assessed, independently, by three researchers, E.S, R.H.K. and A.P. and validated by the supervising researcher (A.G.G.). In this way, a total of 39 studies were selected. Gene symbols were assigned to the extracted human, mouse and rat genes according to the HUGO Gene Nomenclature Committee (HGNC) (https://www.genenames.org/, accessed on 20 November 2021).

In cases where differential gene expression data were not provided in the corresponding articles, we searched for the original gene expression data files deposited in NCBI GEO (Gene Expression Omnibus) DataSets [19] according to the selection criteria: (i) gene expression data derived from irradiated and non-irradiated (control) tissue/cell samples, and (ii) inclusion of >5000 genes in the dataset. The following microarray transcriptome datasets were obtained where their respective GEO series and PubMed references are shown in brackets: X-rays (GSE107685 [20], GSE113611 [21], GSE107443 [22], GSE90909 [23], GSE85323 [24], GSE59861 [25], GSE6262 [26]); α particles (GSE12435 [27], GSE21059 [28], GSE18760 [29]); carbon ions (GSE6630 [30]); protons (GSE20629 [31]). The GEO2R interactive web server [19] was employed to detect genes differentially expressed at different conditions.

The differentially expressed genes (DEGs) with an absolute log2 fold-change (FC) greater than 1.5 (|log2FC ≥ 1.5|), or FC > 1.5 and FC < 0.67, and FDR-adjusted *p*-value (q-value) less than 0.05 or *p*-value < 0.001 (for transcriptomic data) and *p*-value < 0.05 (for the text mining data) were retained.

#### *2.2. Functional Enrichment Analysis*

Venn diagrams were constructed using the online tool Draw Venn Diagram (https://bioinformatics.psb.ugent.be/webtools/Venn/, accessed on 20 January 2022) to identify common up and downregulated genes across radiation types, as well as of lowversus high-LET radiation (only for entries where the corresponding LET was provided in the original paper) and deregulated genes of lower versus higher doses in the range of clinical interest (0.3–0.5 Gy vs. 0.6–2.0 Gy). Furthermore, overrepresented biological pathways, along with the corresponding disease pathways, were identified in different sets of genes, related to every type of irradiation, as well as for low and high LET, and low and high clinical doses. Functional enrichment analysis was conducted with WebGestalt (WEB-based GEne SeT AnaLysis Toolkit) 2019, an online tool used for the identification of statistically significant enriched terms in the given gene sets compared to selected reference sets [32]. The WebGestalt parameters chosen were "Organism of Interest": Homo sapiens, "Method of Interest": Over-Representation Analysis (ORA), "Functional database": geneontology/Biological Process noRedundant and pathway/Wikipathway for biological paths, or disease/Disgenet for diseases, "Select gene ID type": gene symbol, "Select Reference set": genome; the default advanced parameters were chosen, and only pathways with false discovery rate (FDR)-corrected *p*-value less than 0.05 were considered in the analysis. Affinity propagation was used for clustering the terms (i.e., biological process and disease) by selecting a subset of representative terms.

#### *2.3. Database Construction*

API-based Directus (https://docs.directus.io/, accessed on 10 April 2022), an open-source data platform, was used for content management, and MySQL (https://dev.mysql.com/, accessed on 10 April 2022), an open-access database management system, was used to store the data on the backend side. Data stored in excel format were imported to the MySQL database using Node.js.

On the front end, the popular VueJS framework, which provides officially maintained support packages for building web UIs, was used to create easily accessible content interfaces. Axios library (https://axios-http.com/, accessed on 12 April 2022), a promise-based HTTP client for the browser and Node.js, was used to obtain the data provided by Directus content management API services. Tailwind CSS framework (https://tailwindcss.com/, accessed on 12 April 2022) was utilized for the styles of the website interface.

#### *2.4. Machine Learning Approach*

Random Forest is a bagging ensemble algorithm, which uses multiple different algorithms to generate a consensus output. It accepts as input a random sample generated from a given dataset with replacement, and then this sample is fed into the tree classifiers. At the end, the class of the sample is determined by voting with the principle of majority rule. During data classification, it can also provide the importance score of each variable (e.g., gene) and evaluate its role in the classification. There are many popular methods for gene selection, including deep gene selection [33], WERFE [34], Based Bayes error Filter [35], etc. The basic principle of all of these methods is to firstly rank the genes on the basis of certain evaluation criteria, and then select an optimal subset of genes. However, these methods cannot capture the relationship between the selected genes and the precision of the classification. Su and colleagues developed an algorithm based on recursive feature elimination (RFE), by taking into account the impact of both the gene numbers and prediction performance [36].

RFE is a greedy algorithm that creates gene sets recursively and then determines an optimal subset from those sets. The goal of RFE is to obtain the smallest possible sets of variables in an iterative way. RFE discards those genes of least importance in an iterative way and performs classification based on the new subsets of genes. All the gene subsets are evaluated based on their classification performance.

In our study, in order to prioritize the genes in the groups (a) irradiated *versus* nonirradiated, and (b) cancer *versus* normal, we first applied the RFE algorithm in Random Forest. All the methods were implemented by using the Python 3.9.7 *scikit-learn* module (https://pypi.org/project/scikit-learn/, accessed on 16 February 2022). To this end, we randomly divided our datasets into 75% training data and 25% testing data for all the models used for classification; the random state was set to 42. We first fit the model, then removed the less relevant genes (listed in the RadBioBase) and calculated the classification performance metric. After that, we removed the least important genes, fitted the model again and calculated the performance. This process was repeated until there were no genes left. The final set of genes was the set that maximized the performance. However, the gene subset selected in this study was the one with the highest accuracy since accuracy is the most common evaluation metric adopted for assessing the robustness and efficiency of algorithms. The final gene subsets of high *versus* low LET demonstrated classification accuracies of 95.54%, respectively.

Finally, to enhance the robustness of our results, robust rank aggregation (RRA) [37] was applied to the output of the previous steps so as to obtain the top-ranking genes. The RRA method uses a noise-robust probabilistic model to aggregate ranked lists, such as lists of genes, and to calculate the statistical significance (*p*-values) for all ranked elements. RRA was performed in the R programming environment (version 4.1.3) (https://www.r-project.org/, accessed on 10 March 2022).

#### *2.5. Functional Network*

The STRING database (version 11.5) (https://string-db.org/, accessed on 15 May 2022) [38] was used to investigate and visualize both known and predicted associations among the protein products of the genes under study.

#### **3. Results and Discussion**

#### *3.1. Development of RadBioBase*

For database construction, we performed text mining in PubMed, using appropriate keywords across studies that experimentally address the overall effect of five high- and low-LET radiation types of interest in a broad range of mammalian cell types, including human, mouse and rat study model systems. The database includes 7436 entries, with a total of 3730 unique genes derived from 14 tissues/cell lines [20–31,39–65]. For each entry, the following information was provided:


The above data are available through RadBioBase (http://radbiodb.physics.ntua.gr/). RadBioBase has a user-friendly interface and can be searched by using several options, such as (a) differentially expressed genes (b) gene expression status (up or downregulated), (c) type of radiation, (d) cell type (normal or cancer), (e) radiation dose, (f) radiation exposure time, (g) as well as a combination of the above options (Figure 1). The search results are displayed in a new window, in a tabular format, and can be downloaded to a CSV file. RadBioBase v1 is maintained by the National Technical University of Athens, Greece, and will be updated at regular intervals.

**Figure 1.** Example output page of RadBioBase. The database was searched using the "X-rays & γ-rays", "Cancer" and "up" options, by selecting the time and dose ranges 0.5–12 h and 0.1–5 Gy, respectively.

#### *3.2. Commonalities among Radiation Types across a Number of Mammalian Tissues*

Using RadBioBase, we performed a comparison among all different types of irradiation (X-rays, γ-rays, protons, carbon ions and α-particles), to unveil basic commonalities across all therapeutic modalities and mammalian tissue types. One important consideration regarding this database is that, since its generation is based on publicly available data, it is inevitably more representative for the types of tissues and irradiation most frequently used across the corresponding studies. To collectively describe the content of this database, we estimated the number of entries for tissue type, radiation type, organism type and normal versus cancer cell type (Figure 2A–D). Overall, the database includes 14 types of cells/tissues (Figure 2A). The highest number of entries are assigned to blood, breast and lung tissue, possibly reflecting the types of cancers where irradiation represents a frequent standard of care treatment. Similarly, 50% of the entries correspond to X-rays, which have been in research and clinical use for longer periods than the more recent radiation types (Figure 2B). Moreover, 74% of the entries represent normal and 26% cancer cells (Figure 2C). The percentages of entries in human versus rodent cells are similar, leading to a ratio of 1.06 (Figure 2D). This information facilitates the design of downstream analyses, interpretations of the results and inferences about disease pathways, especially in cases where the data are combined to generate universal signatures.

As shown in Figure 2E, among all up and downregulated genes (included in the current version of RadBioBase), we identified five genes that are commonly activated in all radiation groups (*GDF15*, *GADD45A*, *SESN1*, *CDKN1A* and *TP53INP1*). These genes are downstream effectors/targets of p53, a major tumor suppressor gene that encodes a transcription factor with a central role in preserving cell homeostasis and is one of the most important targets for translational cancer research. The physiologically low levels of mature p53 increase upon cellular stresses and, together with post-translational modifications, lead to the formation of oligomers that bind to specific p53 responsive elements on target gene promoters. Upon limited DNA damage, p53 induces cell cycle arrest and DNA repair genes, whereas upon extended and severe damage, it induces genes mediating senescence or cell death so as to isolate damaged cells from the intact cellular population [73]. The p53 pathways control five different kinds of cell death: (i) apoptosis, (ii) ferroptosis, (iii) TNF ligand- or (iv) FAS ligand-mediated necroptosis and (v) cellular senescence followed by the secretion of cytokines that attract immune system cells [74]. Our results are consistent with studies suggesting that the p53 pathway is a universally-induced sensitizer of cells to any type of irradiation [74]. They also suggest that p53-targeting molecules hold potential to be combined with any type of radiotherapeutic modality to increase treatment efficacy across a number of tissues.

As shown in Figure 2E, the number of non-overlapping genes for each radiation type tends to be higher than the genes that are in common in two or more radiation types. The fact that transcriptional responses tend to be radiation type-specific strongly indicates that along with the p53 cascades, each radiation modality can activate distinct biological pathways to exert its effects on cells. In an analogous manner, radiation-specific transcripts might be associated with different disease pathways, which can be predictors of specific side effects. To shed light on these aspects, we performed a detailed analysis of the overrepresented biological and disease pathways related to each type of radiation separately, along with the corresponding genes.

**Figure 2.** *Cont*.

**Figure 2.** A description of the contents of the database and commonalities among the several radiation types. Database statistics. (**A**) Number of entries corresponding to tissues and cell lines are shown on top of the bars; the height of the bars is proportional to the number of entries. Percentage of entries related to (**B**) radiation types, (**C**) cancer and normal, (**D**) human and rodent tissues/cells across different types of radiation. (**E**) Venn diagram illustrating the overlapping of differentially expressed genes (both up and downregulated) between all radiation groups. The five common genes in all radiation groups are *GDF15*, *GADD45A*, *SESN1*, *CDKN1A* and *TP53INP1*.3.3. Each radiation type is linked to distinct biological functions.

We found that each radiation type, in general, exhibits a unique set of biological processes, beyond the expected pathways of response to stress and cell death. In particular, X-rays are related to metabolic processes, including the "fatty acid metabolic process", "small molecule catabolic process" and "sulfur compound metabolic process" (Figure 3A and Table S2). This is consistent with several studies showing that IR can cause metabolic changes, oxidative stress and cell death [75,76] and that sulfur-related enzymes play a major role in the radiation-induced oxidative stress response and detoxification [77]. Upon irradiation, where the levels of oxygen-free radicals are increased, sulfur-related metabolism acts as an antioxidative stress defense pathway. These processes are particularly prominent in the liver since its function is critical in the protection against induced stress, rendering the liver extremely sensitive to radiation. X-ray irradiation was also found to be associated with fatty acid (FA) metabolism. Interestingly, recent studies suggest that FA metabolism represents the link between X-ray irradiation and ferroptosis, a novel type of programmed cell death that depends on iron and is characterized by the accumulation of lipid peroxides [78]. This FA-related type of cell death is genetically and biochemically distinct from other forms of regulated cell death. In agreement, ferroptosis-inducing agents can sensitize cancer cells to X-ray irradiation [79], while pro-ferroptotic FA metabolism renders cancer cells immunogenic [80]. In light of these data, it would be interesting to investigate whether X-rays initiate an FA metabolism–ferroptosis axis, which subsequently modulates the immunogenic properties of irradiated cells towards enhancing therapeutic responses to immunotherapy.

Additionally, we found associations of X-ray-induced transcriptomes with zinc and copper homeostasis (Figure 3A and Table S2). On the one hand, zinc homeostasis is indirectly related to post-irradiation effects through increases in oxidative stress [81–83]. Zinc exhibits protective effects against irradiation by activating antioxidant enzymes, which in turn reduce reactive oxygen species (ROS) levels and oxidative stress [81,83,84]. In addition, zinc acts as an intracellular signaling molecule, activating apoptotic pathways, immunodeficiency and inflammation suppression [81,83,85]. On the other hand, copper ions contribute to radiation- and stress-resistance [86], tumor growth, inflammation and angiogenesis [87–90].

Among higher LET radiation types, protons are strongly related to apoptosis and oxidative stress (Figure 3B and Table S2), while carbon ion and alpha particles with enhanced proinflammatory signaling. However, while carbon ions exhibit overrepresented interleukin-18 (IL-18) signaling pathways (Figure 3B and Table S2), α-particles appear to be linked with photodynamic therapy (PDT)-induced NF-κB survival signaling (Figure 3B and Table S2). IL-18 is a proinflammatory cytokine of the interleukin-1 family, expressed in several cell types, including, but not limited to, macrophages, dendritic cells and epithelial cells. It is also involved in the regulation of immunomodulatory cytokine networks that mediate host defense, inflammation and tissue regeneration [91]. Regarding the transcription factor NF-κB, it integrates several stress signals and can regulate DNA transcription, cell survival, as well as immune system and inflammatory responses in a pleiotropic manner. NF-κB pathways are triggered by PDT and regulate the interplay between the immune system and an anti-cell death response through the release of cytokines and chemokines and the control of apoptosis or necrosis [92]. Intriguingly, IL-18 can also activate NF-κB; therefore, it is possible that the effects of carbon ions and alpha particles revolve around a complex inflammatory and immunomodulatory network, where NF-κB occupies a central hub position suggested also by Hellweg (2015) [93]. Taking into account that higher LET radiation can cause a higher level of DSBs and DNA damage clusters [94], it would be interesting to further investigate if these pathways may stand at the crossroads of high LET-specific DNA damage and the immune response [95,96].

**Figure 3.** Gene set enrichment analysis. Overrepresented biological pathways (affinity propagation) for the DEGs in different types of radiation. (**A**) Low-LET radiation (X- and γ-rays) and (**B**) higher LET radiation (protons, carbon ions, α-particles). (**C**) Overrepresented disease pathways (affinity propagation) for all DEGs genes in different radiation types. The *x*-axis corresponds to the enrichment ratio, i.e., the ratio of the number of observed genes to the number of expected genes from each category in the input gene list.

We also observed that carbon ions activate transcripts involved in axon guidance and cell migration [96]. This finding is consistent with studies suggesting that cell migration and apoptosis in normal and tumorigenic tissues is regulated by many axon guidance molecules [97]. Notably, tumor-intrinsic activation of genes indispensable for neuronal development and neurological function is a nearly universal phenomenon in cancer, which, depending on the cancer type, can have either a negative or a positive effect in disease

initiation and progression [98,99]. To date, it remains a *terra incognita* as to whether some radiotherapeutic modalities also trigger this phenomenon. Another hypothesis is that the axon guidance processes identified in Figure 3B reflect associations between IL-18 and neuroinflammation and neurodegeneration (which are conditions further related to high-LET irradiation [100]). IL-18 is constitutively expressed in resident cells of the central nervous system (CNS), supporting a local IL-18-dependent immune response that can influence neural tissue homeostasis [101,102]. Investigating how carbon ion beams, compared to other radiation types, may activate neuronal pathways, and how this reflects to the tissue microenvironment and the crosstalk of the irradiated cells with surrounding neuronal and immune cells, remains a subject of fruitful research. Dissecting the connection between therapeutic radiation and the co-option of neuronal programs in the irradiated cells could provide invaluable insights for increasing the therapeutic efficacy of radiation and ameliorating any side effects on healthy tissues.

#### *3.3. Radiation Type-Specific Disease Pathways Inferred from Transcriptomes of Irradiated Cells*

An analogous analysis of the overrepresented human disease pathways that are associated with irradiation-responsive transcripts indicated relatively distinct disease profiles across radiation types (Figure 3C and Table S3). In detail, liver dysfunction pathways are dominant upon X-ray irradiation (Figure 3C and Table S3), perhaps as a sequalae of the critical function of this organ in the protection against induced stress, hence indicating a sensitivity of the liver upon radiotherapy. Another vital organ that might be affected is the heart since carbon ion irradiation was found to be associated with atherosclerotic disease (Figure 3C and Table S3), in agreement with clinical reports that patients who have undergone radiotherapy are at increased risk for cardiovascular diseases (CVDs) [103]. Since IL-18 participates in atherogenesis [104], the increased incidence of CVDs might reflect the activation of IL-18 signaling pathways that are associated with this radiation type. These findings suggest that increased monitoring, further investigation and timely treatment might be required in order to prevent these unwanted effects. Proton-based therapy appears to be related with inflammation and fever (Figure 3C and Table S3), two mild side effects that are amenable to clinical management. Energetic carbon and alpha particles are associated with reperfusion injury (Figure 3C and Table S3), a type of ROS-induced tissue damage occurring when blood supply returns to tissue after a period of ischemia or hypoxia. Interestingly, single-dose radiotherapy coupled with early tumor ischemia/reperfusion can lead to tumor lethality via the inactivation of homologous recombination [105]. Hence, occurrence of this side effect in patients undergoing radiotherapy might be an indicator of selective tumor radiosensitization and increased therapeutic efficiency. In conclusion, our analysis reveals radiation type-specific side effects and possible comorbidities that call for increased surveillance for relevant patient complaints after radiation treatment.

#### *3.4. Machine Learning-Generated Gene Signatures of Cell Sensitivity to High- Versus Low-LET Radiation Types*

One issue in the clinic is the selection of individual patients for high- or low-LET radiation treatment, which is in turn dependent on the radiobiological properties of the tumor [106]. In this regard, transcriptomics data of irradiated cells can infer radiosensitivity predictors, whereby differentially expressed genes are ranked on the basis of certain evaluation criteria, and then an optimal subset of genes is selected [107]. Although insightful, previously described methods may pose limitations in gene selection, as the produced gene signatures may not accurately capture the relationship between the selected genes and the precision of the classification. To bypass these limitations, we applied machine learning, a robust computational method that holds promise to reduce the complexity of whole genome gene expression patterns and produce manageable signatures of response while simultaneously taking into account several important selection criteria [108]. We used a recently-developed algorithm based on recursive feature elimination (RFE), which creates gene sets recursively and then determines an optimal subset, aiming to obtain

the smallest possible sets of variables in an iterative way while discarding those genes of least importance [36]. To verify the ability of the algorithm to generate gene signatures linked to the features of interest, we initially ran a control test in DEGs of cancer versus normal tissues that are included in the RadBioBase. The algorithm predicted correctly a number of markers of tumor initiation and progression, such as CD44 [109], MMP9 [110], CDC20 [111], FOS [112] and WNT5A [113] (Figure S1). Several of these genes are also associated with sensitivity to radiation, as confirmed by further comparisons versus our previously published comprehensive lists of molecular determinants of radiation response in cancer tissues [114]. Having assured the accuracy of the algorithm in our datasets, we proceeded to generate a gene signature (Figure 4A) for high- *versus* low-LET radiations, using clinically relevant criteria such as post-irradiation time and dose on the data of RadBioBase. The five types of radiation were grouped into two groups because the larger the dataset, the more information the machine learning algorithm can capture, thereby enhancing its predictive performance. This led to the identification of a 22-gene signature that is characteristic for the response to high-LET as opposed to low-LET irradiation. GSEA analysis showed that the most significantly enriched (FDR < 0.05) processes of those genes are cell cycle, cell division and inflammation.

**Figure 4.** Gene signature of high- *versus* low-LET radiation types. (**A**) Twenty-two-gene signature characteristic of response to high-LET vs. low-LET irradiation. (**B**) Network depicting the associations (edges) of twelve signature genes/gene products (nodes) in cell cycle (red), cell division (blue) and inflammation (green).

A further STRING analysis of this signature revealed that twelve of those genes/proteins appear to interact (Figure 4B) and mediate cell cycle, cell division and/or inflammation (Figure 4B, genes with red, blue and green color-coding), thereby accurately reflecting the main processes known to be induced by LET. Among these genes, we were able to identify several recently-characterized effectors of radiosensitivity, such as RAD51-associated protein 1 (RAD51AP1), which plays an integral role in homologous recombination by activating RAD51 recombinase, and its knockout is shown to induce radiosensitivity [115]; TTK protein kinase, the inhibition of which radiosensitizes basal-like breast cancer cells through impaired homologous recombination [116]; the DNA methyltransferase 3B (DNMT3B), an epigenetic modifier that protects centromere integrity by restricting R-loop-mediated DNA damage [117], and its silencing can restore the p53/p21 signaling pathway via DNA demethylation [118]; and TRAIP, a novel RAP80-interacting protein that is necessary for translocation of RAP80 to DNA lesions and promotes homologous recombination in response to DNA damage [119]. This signature also revealed novel genes that are associated with the response to radiation, for example, the Spindle And Kinetochore Associated

Complex Subunit 3 (SKA3) and the Rac GTPase Activating Protein 1 (RACGAP1). Future clinical validation of this signature in tissues from patients that have undergone high-LET radiation therapy can define indicators of responsiveness in this therapeutic modality towards improving patient selection.

#### *3.5. Low-Dose Irradiation Is Associated with Cytokine Cascades, While High with ROS Metabolism*

Thus far, the implementation of new technologies in radiotherapeutic treatment has been largely empirical and driven by the belief that increasing doses will increase cure [1]. Consequently, in a large number of studies, high doses have been preferentially used to address the effects of irradiation on tissues. However, increased doses pose clinical risks for acute and/or chronic toxicities, without substantially enhancing the therapeutic benefits. Moreover, there is emerging evidence that low doses can be beneficial against several pathological entities. For example, in cancer, low-dose irradiation reverses resistance to immunotherapy by reprogramming the TME of immune-cold tumors [120], while in COVID-19-induced pneumonia, it induces antiinflammatory responses [121]. To further explore whether low doses could have therapeutic potential, we mined the RadBioBase database for differences in transcriptomes induced at different doses. The database contains entries from 37 studies using doses over 5 Gy, 2 studies using less than 0.5 Gy and 3 studies using both. For our analysis, we particularly considered the entries with a value of 0.3–0.5 Gy as "low" and those with a value of 0.6–2.0 Gy as "high" since these correspond to the clinically relevant low/high dose ranges. A GSEA analysis for the 445 genes found commonly deregulated at the 0.3–0.5 Gy range, underscored a profound overrepresentation of cytokine and inflammatory response pathways, implying that low doses are capable of inducing inflammation-related cascades. This is distinct from effects at high doses, where the 668 genes commonly responding to the 0.6–2.0 Gy range are associated with ROS metabolism (Figure 5 and Table S4). Following a gene-centric approach, we found many up or/and downregulated cytokines and interleukins, as well as other inflammation-related genes, deregulated at "low" doses. These include, but are not limited to, the upregulation of antiinflammatory genes *IL4* and *TNFA1P3*, and downregulation of the proinflammatory genes *IL12B* and *CDK5R2*. Nevertheless, genes that can exert both anti and proinflammatory activity depending on cell content, for example, *IL1A*, *IL1B*, *IL6* and *CXCL3*, appear to be upregulated at low doses in the original dataset, implying a complex cytokine profile at this range. The transcription of cytokines and other secreted molecules mediating intercellular communication (e.g., *CCL3*, *CCL4*, *CXCL2*, *IL22*, *TNF*, *IL18R1*, *IL7R* and *IL13RA2*, *IL13* and *IL10)* were also deregulated at high doses. Hence, low doses alter the transcription of secreted factors, but the composition of these factors is distinct compared to that of the high dose. In support, it was recently shown that high and low doses of irradiation induce different secretome profiles [122]. Given that our analyses are inevitably based on a relatively small number of available studies at low doses, further comprehensive characterization of the secretomes of low-dose irradiated cells is required to confirm these findings and decipher the inflammatory molecules with *bona fide* effects from those related to toxicities and radioresistance [123]. Considering that different doses/types induce different kinds of DNA damage, future high-throughput identification and functional characterization of the secretomes of cells irradiated with several types and/or doses holds promise to unveil links between intrinsic cell damage and the effects on adjacent and remote tissues, which can be translated into improved clinical patient management.

**Figure 5.** Overrepresented biological pathways (affinity propagation) for all 445 DEGs in low-dose (0.3–0.5 Gy) radiation group and all 668 DEGs in high-dose (0.6–2.0 Gy) radiation group considering all types of irradiation. The *x*-axis corresponds to the enrichment ratio.

#### **4. Conclusions**

The effects of irradiation are cell-intrinsic and cell-extrinsic, with the ability to reprogram the microenvironment both proximal and distal to the irradiated sites. Each irradiation type is suspected to cause different initial DNA lesions and activate distinct DDR/R components, inducing cell–cell interactions that ultimately lead to distinct immunogenic effects on cancer cells and on remote normal tissues. Predicting and characterizing the track of localized and systemic effects for each radiation type and dose can help fine-tune radiotherapy used alone or in combination with chemo- or immunotherapies, in a way that is less empirically-based and more guided by solid clinical and radiobiological data. To this end, comprehensive comparisons of changes in gene expression across normal and cancer cells for the several types and/or doses of radiation have a high clinical value for informing and improving decisions for radiotherapy. To address these novel challenges, we developed a database, termed RadBioBase, that can provide systemic insights into the attributes of irradiation relative to gene transcription in mammalian tissues. Further extending and updating this database to include additional tissue types in the future is anticipated to provide a cornerstone for the in silico prediction of the beneficial and toxic effects of radiation locally and systemically, which can be translated to more efficient and safer radiotherapy schemes. On the one hand, analyses of transcriptome changes in cancer cells can reveal novel pathways that enhance the response to radiation and/or awaken the immune system against the tumor cells. On the other hand, analyses of normal tissues can indicate genes associated with radiation type-specific side effects.

Notably, our database is designed to provide correlations between irradiation and the full-length transcripts of genes. At this point, it should be mentioned that several genes can synthesize isoforms or mutant forms with distinct or even opposing functions. Members of the TP53 family constitute such representative cases. For example, while wild-type TP53 induces radiosensitivity, expression of its missense mutants correlates with radioresistance [124]. Similarly, TP73, a sibling of TP53, synthesizes not only full-length TAp73 isoforms, which sensitize cells to irradiation, but also N-terminal truncated isoforms that are generated via aberrant splicing or alternative promoter usage at the 5'end and act as dominant negative inhibitors of their TAp73 counterparts, favoring resistance to radiation [125,126]. In cases of such genes, where their various protein products exert divergent effects on DDR and radiosensitivity, our database detects general associations with irradiation, without deciphering among the functionally divergent isoforms. The involvement of alternative forms or gain-of-function mutants needs to be subsequently

addressed in a more detailed, gene-centric manner, using complementary targeted next generation sequencing approaches.

Last but not least, the COVID-19 pandemic has changed our world by accelerating new digital and virtual reality megatrends in healthcare and setting in motion a dynamic that is expected to last and reform society and science at several levels. These changes are now more than ever before extrapolated to radiotherapy, a field that has historically evolved by taking advantage of contemporary technological trends. An important lesson taught is that central databases that share and disseminate information can improve global digital healthcare at several levels [127]. In line with this trend, our initiative to collect and systemically organize all available molecular information on the responses of mammalian tissues to irradiation can become a useful means for driving radiation oncology towards a new exciting digital health era.

**Supplementary Materials:** The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/antiox11112286/s1, Figure S1: Markers of tumor initiation and progression; Table S1: MCDS simulation results for double-strand breaks (DSBs) and total clusters of DNA damage (DSBs and non-DSB oxidative lesions) per Gy per Gbp for radiation types used in this work; Table S2: Overrepresented biological pathways of the X-ray, γ-ray, protons, carbon ions and αparticle types of radiation and the corresponding retrieved genes; Table S3: Overrepresented disease pathways of the X-ray, protons, carbon ions and α-particle types of radiation and the corresponding genes; Table S4: Overrepresented pathways and related genes of the high (0.6–2.0 Gy) and low (0.3–0.5 Gy) radiation doses for all types of irradiation; Figure S1: Gene signature of cancer *versus* normal cells/tissues. Those genes implicated in radioresistance are underlined.

**Author Contributions:** Conceptualization, A.G.G.; methodology, E.S., S.A., S.L. and A.P.; software, S.A., G.M.B. and I.T.; validation, E.S., R.H.K., V.Z. and Z.N.; formal analysis, E.S., S.A., G.M.B., I.T., R.H.K. and A.P.; investigation, A.G.G., E.S., S.L. and A.P.; data curation, E.S., S.A., G.M.B., I.T., C.E.H., E.I.A. and A.P.; writing—original draft preparation, A.G.G., E.S., S.L. and A.P.; writing—review and editing, all authors; project administration, A.G.G. All authors have read and agreed to the published version of the manuscript.

**Funding:** AGG was funded from the project 21GRD02 BIOSPHERE, which has received funding from the European Partnership on Metrology, co-financed by the European Union's Horizon Europe Research and Innovation Programme and by the Participating States. Funder ID: 10.13039/100019599. Grant number: 21GRD02 BIOSPHERE.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** All data and analysis methodologies are contained in the manuscript. Any additional data requests can be addressed to the corresponding author.

**Acknowledgments:** The authors thank Konstantinos Anagnostopoulos (NTUA) for his help with the online version of the database.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


## *Article* **Serum Proteomic and Oxidative Modification Profiling in Mice Exposed to Total Body X-Irradiation**

**Masaru Yamaguchi 1,†, Yota Tatara 2,†, Eka Djatnika Nugraha 3, Yoshiaki Sato 1, Tomisato Miura 4, Masahiro Hosoda 1,4, Mukh Syaifudin 5, Shinji Tokonami <sup>4</sup> and Ikuo Kashiwakura 1,\***


**Abstract:** The details of the dose-dependent response of serum proteins exposed to ionizing radiation, especially the oxidative modification response in amino acid sequences of albumin, the most abundant protein, are unknown. Thus, a proteomic analysis of the serum components from mice exposed to total body X-irradiation (TBI) ranging from 0.5 Gy to 3.0 Gy was conducted using LC-MS/MS. The analysis of oxidative modification sequences of albumin (mOMSA) in TBI mouse serum revealed significant moderate or strong correlations between the X-irradiation exposure dose and modification of 11 mOMSAs (especially the 97th, 267th and 499th lysine residues, 159th methionine residue and 287th tyrosine residues). In the case of X-irradiation of serum alone, significant correlations were also found in the 14 mOMSAs. In addition, a dose-dependent variation in six proteins (Angiotensinogen, Odorant-binding protein 1a, Serine protease inhibitor A3K, Serum paraoxonase/arylesterase 1, Prothrombin and Epidermal growth factor receptor) was detected in the serum of mice exposed to TBI. These findings suggest the possibility that the protein variation and serum albumin oxidative modification responses found in exposed individuals are important indicators for considering the effects of radiation on living organisms, along with DNA damage, and suggests their possible application as biomarkers of radiation dose estimation.

**Keywords:** total body irradiation; proteomic analysis; oxidative modification profiling; serum albumin; amino acid sequences; radiation biomarker

#### **1. Introduction**

Very recently we reported on a proteomic analysis and oxidative modification profiling of serum collected from residents of a newly discovered high-level natural background radiation area (HBRA, annual effective dose of approximately 50 mSv y<sup>−</sup>1) and normal-level background radiation area (NBRA, 1.22 mSv y−1) in Mamuju, Indonesia [1]. A proteomic analysis showed that the apolipoprotein B-100 and hemoglobin subunit α1 expression of residents in the HBRA was significantly lower than that of residents in the NBRA. In addition, a total of 270 oxidation-mediated modification sites were identified in the amino acid sequence of human serum albumin (HSA) by liquid chromatography-tandem mass spectrometry (LC-MS/MS). Among these, four specific amino acid sequences of HSA showed a dose-dependent oxidative modifications. Notably, the 162nd and 356th tyrosine

**Citation:** Yamaguchi, M.; Tatara, Y.; Nugraha, E.D.; Sato, Y.; Miura, T.; Hosoda, M.; Syaifudin, M.; Tokonami, S.; Kashiwakura, I. Serum Proteomic and Oxidative Modification Profiling in Mice Exposed to Total Body X-Irradiation. *Antioxidants* **2022**, *11*, 1710. https://doi.org/10.3390/ antiox11091710

Academic Editors: Elena Obrador Pla and Alegria Montoro

Received: 22 July 2022 Accepted: 27 August 2022 Published: 30 August 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

residues and 111th and 470th methionine residues were found. None of these findings have been reported in humans exposed to chronic low-dose radiation. This can be used as a biomarker not only for the assessment of the presence or absence of radiation exposure but also for dose prediction of chronic radiation exposure in living organisms. These results suggest that traces of radiation exposure are recorded in serum albumin and that there is a possibility of a new methodology that can evaluate biological responses below 100 mSv.

Regarding the health effects of chronic low-dose radiation exposure, epidemiological studies of human populations, such as occupational studies of nuclear workers, are not as clear regarding whether low-dose-rate exposure results in lower risks than seen among Japanese atom bomb survivors who were acutely exposed to radiation [2]. In addition, the UNSCEAR report showed that epidemiological studies in several regions of the world (Ramsar, Yangjiang, Kerala and Guarapari) reported no correlation between radiation exposure and cancer rate or mortality in areas with high natural background radiation [3], indicating that the effect of low dose rates on health and the cancer risk after exposure to ionizing radiation is still unclear. Tang et al. also reported that the mechanisms of low-dose ionizing radiation (≤100 mSv) or low-dose-rate ionizing radiation (<6 mSv/h)—induced health effects are poorly understood [4]. Issues related to the health effects of low doses require further research in the future.

The annual effective dose shown in the previous report was estimated as the accumulation of the dose from external exposure (environmental gamma radiation) and internal exposure (mainly through breathing of indoor radon) based on our previous reports. This radioactivity was mainly derived from uranium (238U), thorium (232Th), radon (222Rn), thoron (220Rn) and their progeny contained in soil [5,6]. However, do the proteomic changes observed in residents living under chronic long-term low-dose radiation exposure also occur with a single acute high-dose radiation exposure, such as a radiation exposure accident? Furthermore, the details of whether there is a dose-dependent response are unknown. In particular, prodromal symptoms seen in patients within 1 to 2 days after acute radiation exposure of ≥1 Gy may include symptoms such as loss of appetite, nausea, vomiting (>2 Gy), and diarrhea, making it easy to confirm the biological response to radiation exposure [7]. As pointed out by Shin et al., the effects of low-dose radiation, which many experimental studies consider to be defined as <0.5 Gy, are subtle, and the absence of reliable biological markers has been an obstacle [8,9]. With the rapid progress of analytical techniques in recent years, an increasing number of studies have reported on the search of the proteome of exposed individuals [9–12], and it is expected to be utilized as a biomarker for dose estimation in triage in the event of nuclear or radiological disasters [13–15]. However, the details of the relationship between the radiation dose and oxidative modification of serum albumin are unknown. Furthermore, considering its application to radiation accidents and nuclear disasters, it is necessary to verify it in animal experimental models, as it cannot be verified in humans.

In the present study, we analyzed the proteome and oxidative modification profile by LC-MS/MS using mouse serum after 24 h of single total body X-irradiation (0.5–3 Gy), assuming a nuclear disaster/radiation accident.

#### **2. Materials and Methods**

#### *2.1. Animal Experiments*

Seven-week-old female C57BL/6JJcl mice were delivered from the breeding facilities of CLEA Japan (Tokyo, Japan). All mice were housed in a conventional clean room at an ambient temperature of 23 ◦C with 50% relative humidity, and a 12-h light/dark cycle. The mice had ad libitum access to sterilized standard laboratory mouse chow (CLEA Rodent Diet CE-2, CLEA, Tokyo, Japan) and drinking water. After obtaining approval from the animal experiment committee (approval number: G17001), all experiments were conducted according to the legal regulations in Japan and the Guidelines for Animal Experiments, and all efforts were made to minimize the number of animals used and their suffering in this study. After a week of acclimatization, 8-week-old mice were randomly divided into

4 groups with more than 8 mice per group and subjected to varying TBI doses of 0, 0.5, 1 or 3 Gy from X-rays (150 kVp, 20 mA, 0.5-mm aluminum and 0.3-mm copper filters) at a dose rate of 1.0 Gy/min using an MBR-1520R X-ray generator (Hitachi Medical, Tokyo, Japan) with a distance of 450 mm between the focus and the target. The air kerma was monitored with a thimble ionization chamber, which integrated the radiation dose and blocked X-rays when it reached a present dose value. Peripheral blood was harvested by capillary tube 24 h after TBI from the orbital venous plexus of mice after they were anesthetized using isoflurane (Powerful Isoful; Zoetis, London, UK). Samples were placed at room temperature for at least 30 min to allow blood clotting. Serum was collected by centrifugation at 1200× *g* for 10 min and stored at −80 ◦C until use. In addition, serum collected from non-irradiated mice was subjected to varying TBI doses of 0, 0.5, 1 or 3 Gy from X-rays; incubated at 37 ◦C for 24 h; and stored at −80 ◦C until use.

#### *2.2. Liquid Chromatography-Tandem Mass Spectrometry (LC-MS/MS) and High-Resolution Multiple Reaction Monitoring (MRM-HR)*

The measurement was performed according to a previous report [1]. Briefly, serum proteins were precipitated with acetone and the precipitates were dissolved and denatured with 50% trifluoroethanol. The proteins were reduced and alkylated before trypsinization. Tryptic peptides were analyzed using a TripleTOF6600 mass spectrometer (AB Sciex). A non-labeled quantitative method (SWATH) was used for a serum proteome analysis. The peak areas of peptides were normalized to the sum of the total peak area sum of all measured peptides. The principal component analysis (PCA) and orthogonal partial least square-discriminant analysis (OPLS-DA) were performed using the Simca software program (Infocom Corp, Tokyo, Japan).

The high-resolution multiple reaction monitoring (MRM-HR) method was used to profile oxidative modification of serum albumin. On the basis of the information-dependent acquisition results (data not shown), an assay for MRM-HR experiments was developed using the Skyline software program (MacCoss Lab, University of Washington, Seattle, Washington, DC, USA). The transitions of MRM-HR are shown in Supplementary Table S1. All peak pickings were manually checked after automated matching. The peak areas obtained were normalized by calculating the relative abundance of each modified peptide using the corresponding non-modified peptide.

#### *2.3. Statistical Analysis*

We used the Origin Pro 2020b software program (Northampton, MA, USA) for Windows to perform the linear and polynomial correlation analysis. Furthermore, the data were analyzed with a one-way ANOVA and Tukey-Kramer or Bonferroni/Dunn multiple comparison tests. Statistical significance in the analysis was all tested using a two-sided *p* value of 0.05. The oxidation modification patterns of amino acids were drawn using BKChem, a freely available chemical drawing program.

#### **3. Results**

#### *3.1. Multivariate Analysis of Serum Proteome of Mice with Different Irradiation Doses*

Eight-week-old female C57BL/6JJcl mice were subjected to varying TBI doses of 0.5, 1 or 3 Gy from X-rays, and peripheral blood was harvested 24 h after TBI for serum collection (in vivo model). Regarding animal conditions after 24 h of TBI, as shown in Figure 1A, the body weights of TBI (1 Gy and 3 Gy) mice were significantly decreased in comparison to those of the non-irradiated mice. However, haematocrit values, which indicate the ratio of the total volume of red blood cells to the total blood, did not differ to a statistically significant extent among all groups. To elucidate the effects of each irradiation exposure, LC-MS/MS was used to examine the expression of proteins in the serum in each treatment. Finally, 161 types of protein were identified. The full dataset from all serum samples was subjected to PCA to obtain an overview of the data. The first and second principal component scores were 16.8% and 8.81%, respectively, as shown in Figure 1B (the ellipse represents a 95% tolerance region for the scores based on Hotelling's T2). There was no evidence of separation among the four classes along the first and second principal components. There were no major outliers. The score scatter plots of the OPLS-DA model in Figure 1C demonstrated satisfactory separation between non-irradiated mice and mice exposed to TBI (0.5 Gy, 1 Gy, or 3 Gy) using one predictive component and one orthogonal component. The above groups were completely separated along the first predictive component. These results indicate that the serum proteome profile can be used to distinguish mice with TBI doses of 0.5 Gy, 1 Gy or 3 Gy from X-rays from non-irradiated mice.

**Figure 1. Proteomic analysis of serum from mice exposed to single TBI in in vivo model.** [**A**] Eightweek-old female C57BL/6JJcl mice were randomly divided into 4 groups with more than 8 mice per group and subjected to varying TBI doses of 0, 0.5, 1 or 3 Gy of X-rays at a dose rate of 1.0 Gy/min. Peripheral blood was harvested 24 h after TBI from the orbital venous plexus of mice and placed at room temperature for at least 30 min to allow blood clotting for serum collection (in vivo model). Body weight changes and haematocrit values at the time of serum collection. Statistically significant differences were evaluated by a one-way ANOVA and the multiple comparison tests; *p* < 0.05 (\*). [**B**] PCA score scatter plot of the serum proteome. Each dose treatment group of the samples are represented as shown in the (**left**) figure, respectively. Uncharacterized samples are plotted at the center, and those with features are plotted at a distance from the center (right) figure. Similar features are plotted at close positions. [**C**] The OPLS-DA model to discriminate the serum proteome of each irradiated mouse. The score scatter plot and S-plot are represented. The ellipse in the score scatter plot indicates the Hotelling T2 (0.95) range for each model.

The resultant S-plots using the OPLS-DA model revealed a significant increase in serine protease inhibitor A3K (Serpin A3K) in TBI (1Gy and 3Gy) mice (Table 1) and a further weak dose-dependent correlation was observed when the proteins expressed in non-TBI and TBI mice were compared (Supplemental Figure S1). Angiotensinogen (Serpin A8) and Odorant-binding protein 1a (Odorant-binding protein 1A) were decreased in TBI (0.5 Gy) mice and TBI (1 Gy) mice, respectively. Further serum paraoxonase/arylesterase 1 (PON1), prothrombin, and epidermal growth factor receptor (EGFR) were identified by TBI (3Gy) in addition to Serpin A3K (Table 1). At this time, only PON1 was decreasing.


**Table 1.** Proteins that varied significantly according to the dose of X-irradiation.

<sup>1</sup> FC, Fold change in comparison to non-irradiated samples. <sup>2</sup> Probability represents the *p* value determined by a *t*-test.

#### *3.2. Oxidative Modification of Serum Albumin (OMSA) under Acute Single Radiation Exposure*

Next, we analyzed the oxidative modification of the chemical and spatial structure of albumin that occurred because of acute single radiation exposure. The amino acid sequence of mouse albumin and the identified modifications are shown in Figure 2. The albumin structural region was also totally glycated and oxidatively modified. In addition, nitration of the tyrosine residue and oxidation of the arginine residue, proline residue, methionine residue, and lysine residue were observed. The sequence information for 48 mouse OMSA (mOMSAs) is listed in Supplemental Table 1. For MRM-HR profiling of mOMSA, peptides containing each oxidatively modified amino acid residue were standardized against the peak area value of the corresponding unmodified peptide and analyzed for correlation with the radiation exposure dose as a relative peak area ratio. The fitting of quadratic equations was investigated for each mOMSA; in fact, most human genes show quadratic dose response to radiation [16,17]. In the in vivo model, eleven sequences showed significant dose-dependent correlations (*r* value) of >0.5 by linear or curve fitting. Especially, significant moderate or strong correlations were found between the individual acute high radiation exposure dose and five mOMSAs: mOMSA3 (Linear *r* = −0.51, *p* < 0.01, Polynomial *r* = −0.65, *p* < 0.001), mOMSA9 (Linear *r* = 0.53, *p* < 0.01, Polynomial *r* = 0.54, *p* < 0.01), mOMSA14 (Linear *r* = 0.6, *p* < 0.001, Polynomial *r* = 0.63, *p* < 0.001), mOMSA20 (Linear *r* = 0.55, *p* < 0.001, Polynomial *r* = 0.60, *p* < 0.01), and mOMSA41 (Linear *r* = 0.5, *p* < 0.01, Polynomial *r* = 0.5, *p* < 0.05) (Figure 3). Furthermore, serum samples collected from non-irradiated mice were subjected to varying TBI doses of 0, 0.5, 1, or 3 Gy of X-rays and incubated at 37 ◦C for 24 h as an in vitro model (Figure 4A). Fourteen sequences showed significant dose-dependent correlations (*r* value) of >0.5 by linear or curve fitting. Especially, significant moderate or strong correlations were found between the individual acute high radiation exposure dose and seven mOMSAs: mOMSA9 (Linear *r* = −0.7, *p* < 0.00001, Polynomial *r* = −0.74, *p* < 0.001), mOMSA13 (Linear *r* = −0.5, *p* < 0.01, Polynomial *r* = −0.55, *p* < 0.01), mOMSA16 (Linear *r* = 0.5, *p* < 0.01, Polynomial *r* = 0.5, *p* < 0.05), mOMSA23 (Linear *r* = −0.62, *p* < 0.001, Polynomial *r* = −0.65, *p* < 0.001), mOMSA25 (Linear *r* = −0.56, *p* < 0.001, Polynomial *r* = −0.66, *p* < 0.001), mOMSA33 (Linear *r* = 0.7, *p* < 0.00001, Polynomial *r* = 0.7, *p* < 0.0001), and mOMSA36 (Linear *r* = −0.55, *p* < 0.001, Polynomial *r* = −0.63, *p* < 0.001) (Figure 4B).

**Figure 2. Oxidative modification of SA obtained from single TBI exposure.** Identified MSA sequences and their oxidative modification by LS-MS/MS. The modification sites are marked as follows: pink, aminoadipic acid; light blue, oxidation; blue, dioxidation; yellow, γ-glutamyl semialdehyde; purple, allysine; and grey, nitrotyrosine. Glycated or glycosylated amino acids are indicated with asterisks. The peptides targeted by an MRM-HR are underlined.

**Figure 3. Correlation between the oxidative modification sequence of MSA with single TBI and the radiation dose in vivo model.** [**A**] Eight-week-old female C57BL/6JJcl mice were exposed to varying TBI doses of 0, 0.5, 1 or 3 Gy of X-rays. Peripheral blood was harvested 24 h after TBI from the orbital venous plexus of mice (in vivo model). [**B**] Eleven sequences that showed significant dosedependent correlations by linear or curve fitting are shown. Five sequences (mOMSA3, mOMSA9, mOMSA14, mOMSA20 and mOMSA41) showed a correlation coefficient (*r* value) of >0.5. *p* values of <0.05 were considered statistically significant.

**Figure 4. Correlation between oxidative modification sequence of MSA irradiated in vitro.** [**A**] Serum samples collected from non-irradiated mice were subjecting to varying TBI doses of 0, 0.5, 1, or 3 Gy of X-rays and incubated at 37 ◦C for 24 h (in vitro model). [**B**] Fourteen sequences that showed significant dose-dependent correlations by linear or curve fitting are shown. Seven sequences (mOMSA9, mOMSA13, mOMSA16, mOMSA23, mOMSA25, mOMSA33 and mOMSA36) showed a correlation coefficient (*r* value) of >0.5. *p* values of <0.05 were considered statistically significant.

Based on the results of our previous report [1], the oxidative modification sites of MSA obtained in this study were compared with the results of humans with chronic low-dose radiation exposure (Figure 5). Among the identified amino acid sequences of mouse albumin, lysine, methionine and tyrosine underwent dose-dependent oxidative modification. In particular, half of the oxidative modifications occurred at lysine, unlike in the case of human albumin (Table below in Figure 5). These results indicate that the profile of OMSA induced by radiation exposure is quite different between mice and humans.


**Figure 5. Comparison of oxidative modification sequences of human and mouse serum albumin.** Based on the results of our previous report [1], the oxidative modification sites of the serum albumin obtained in this study were compared. Among the identified amino acid sequences of mouse albumin, lysine, methionine and tyrosine underwent dose-dependent oxidative modification. Oxidationmodified amino acids in chronic low-dose radiation exposure in humans, acute single radiation exposure in mice (in vivo), and acute single radiation exposure in mouse serum (in vitro).

#### **4. Discussion**

In the present study, a proteomic analysis of serum components from mice exposed to 0.5 to 3.0 Gy single TBI revealed significant, dose-dependent variation in six proteins (Table 1). Among these proteins, Serpin A8, Odorant-binding protein 1A, and PON1 were decreased, while other proteins were increased. In particular, the expression of Serpin A3K was found to increase in a dose-dependent manner (Supplemental Figure S1). Serpin A3K is a member of the serine protease inhibitor family and is also known as kallikreinbinding protein, with anti-inflammatory and anti-angiogenic activities [18]. This is a new finding, as no previous reports have shown an association between radiation and the expression of Serpin A3K. Similarly, Serpin A8, which is involved in blood pressure [19], and Odorant-binding protein 1A, which is involved in the sense of smell [20], have never been reported to be related to radiation, and this point was clarified for the first time in this study. Numerous reports on the association with radiation have been made for

EGFR, which is a receptor of tyrosine kinase involved in cell survival/growth signaling that is overexpressed in several cancers [21,22]. In particular, EGFR is expressed in more than 90% of squamous cell carcinomas of the head and neck and is one of the most important therapeutic targets [23]. Following radiation, the activation of EGFR has been reported, leading to downstream signaling that contributes to cancer cell survival [24]. Further, EGFR has been shown to be involved in mediating DNA repair after irradiation, leading to the repair of damaged DNA [25]. There are also several reports on PON1 and prothrombin. Paraoxonase (PON-1) is an antioxidant enzyme that belongs to a family of calcium-dependent esterases that includes PON-1, PON-2 and PON-3 [26]. Serhatlioglu et al. examined the levels of malondialdehyde (an end-product of lipid peroxidation) and PON-1 activity/phenotypes in people, radiology workers, who were exposed to ionizing radiation for different time periods and doses [27]. They showed that PON-1 activity was reduced by 25–35% in subjects exposed to high-dose radiation (>3.5 mSv y<sup>−</sup>1) and in people with long−term exposure (>5 years) to radiation in comparison to the controls. Moustafa et al. evaluated the role of various enzymes in irradiated rats (6 Gy), demonstrating that the PON activity was significantly declined (*p* < 0.05) in comparison to the control group in both serum and the liver [26]. Similarly in this study, the PON1 value was reduced to 38% (Table 1). In addition, prothrombin, a glycoprotein (carbohydrate-protein compound) occurring in blood plasma and an essential component of the blood-clotting mechanism, is transformed into thrombin by a clotting factor known as factor X or prothrombinase [28]. Rithidech et al. reported an increase in prothrombin precursors in the plasma of irradiated (3 Gy) mice on day 2, suggesting an association with radiation [29]. As shown above, significant fluctuations in six serum proteins were observed 24 h after TBI (0.5–3.0 Gy) mice, suggesting that these molecules may be an effective biomarker in this exposure dose range.

In our previous study, we developed an MRM-HR method targeting the 38 patterns of hOMSA using LS-MS/MS and performed the assay on serum samples collected from the residents of a newly discovered HBRA (annual effective dose approximately 50 mSv y<sup>−</sup>1) [1]. As a result, we reported a dose-dependent oxidative change in a specific sequence of human serum albumin. Dose-dependent oxidative modification of mouse serum albumin was observed in single total-body-irradiated mice as well as in the residents with chronic lowdose radiation exposure. In this study, four sequences (mOMSA9, mOMSA14, mOMSA20, and mOSMA41) in the in vivo model and two sequences (mOMSA16 and mOMSA33) in the in vitro model showed positive dose-dependent correlations, but one sequence (mOMSA3) in the in vivo model and five sequences (mOMSA9, mOMSA13, mOMSA23, mOMSA25, and mOMSA36) in the in vitro model showed negative dose-dependent correlations (r value) of >0.5 with *p* values of <0.05 by linear and curve fitting, suggesting that the correlation between the oxidative modification of MSA and the response to radiation differed between the in vivo and in vitro models (Figures 3 and 4). It is well known that proteins circulating in the blood are one of the main targets of reactive oxygen species (ROS) produced by the interaction of ionizing radiation and water molecules. In addition, two prime suspects in the intracellular generation of ROS are also the membrane-bound NADPH oxidase complex and the mitochondrial electron-transport chain (ETC) in vivo model. Differences in the sources of ROS production in both models may contribute to the responsiveness to radiation exposure. Under the action of ROS, proteins undergo oxidative modification, leading to disruption of their structures and functions. Oxidatively damaged proteins accumulate during the course of ageing and under various pathological conditions [30]. In particular, serum albumin, which is present in mouse blood at a concentration of approximately 27 mg/mL, is the most abundant protein, accounting for approximately 54% of the plasma protein weight [31], and since it is exposed to various active chemical species at high frequency, it provides information on oxidative stress in systemic circulation. Of these amino acid residues, the amino acid lysine in proteins is subject to the largest variety of physiological post-translational modifications and is also among the most frequently carbonylated amino acids [32] (Figure 6). In this study as well, lysine carbonylation accounted for half of the identified serum albumin oxidation-modified

sequence OMSA (Figure 5). Peroxynitrite (ONOO−) binds to the phenolic ring of tyrosine residues to produce nitrotyrosine. In addition, the methyl thioether group of methionine changes to a sulfoxide structure in response to increased levels of intracellular oxidative stress (Figure 6). Oxidative modification of serum albumin was also observed in mice after a single exposure to TBI, as it was in humans with chronic low-dose radiation exposure. Interestingly, mOMSA14, an oxidation-mediated modification site of the 159th methionine residue in the mouse albumin amino acid sequence, and an oxidation-mediated modification site of the 162nd tyrosine residue in the human albumin amino acid sequence, is located in domain IB of the albumin molecule (Figure 5). Although the profile of oxidative modification of albumin as a whole differs between humans and mice, it is noteworthy that the amino acid residues Met-159 in MSA and Tyr-162 in HSA, which correlate with exposure, are both located in domain IB of albumin [33]. The common oxidative modification response of domain IB of serum albumin to radiation suggests a possible link between the steric structure of the protein and the biological response to radiation and can be used as a biomarker of both acute and chronic radiation exposure for living organisms. Persistent non-physiological protein modifications, such as non-reparable oxidative protein carbonylation, are irreversible and mostly deleterious to protein activity, and expectedly to their interactions with partner molecules. However, the functional changes of oxidatively modified albumin, the subsequent effects on individual health conditions, diseases, and longevity, as well as the relationship with radiation damage, are issues to be addressed in the future.

**Figure 6. Hypothesized mechanism of oxidative modification of mouse serum albumin after single TBI.** When tyrosine reacts with peroxynitrite or hydroxyl radicals, it becomes nitrotyrosine or hydroxytyrosine, but the aromatic substituent pattern cannot be identified by mass spectrometry. When methionine reacts with hydroxyl radicals, it becomes methionine sulfoxide and then methionine sulfone. When lysine reacts with hydroxyl radicals, it becomes allysine.

Regarding the biological effects of ionizing radiation, initially, it was dominated by target theory, which quantifies the damage caused by traversal of cellular targets such as DNA by ionizing tracks [34,35]. Genomic DNA is the primary target, and double-strand breaks (DSBs) were found to be the most important radiation-induced DNA damage. DSBs differ from base excision repair in SSBs in that their repair pathway is more complex and requires more proteins. Later, the importance of "non-target" or "bystander" effects became recognized with the discovery that mutagenesis, death and/or altered behavior sometimes occur in cells that were not themselves traversed by any radiation tracks but which merely interacted with traversed cells [35,36]. A variety of short- and long-range cell-to-cell propagating signals have been reported, including small molecules capable of moving through gap junctions (e.g., lipid peroxide products, nucleotides), diffusible long-range signals such as proinflammatory cytokines (e.g., tumor necrosis factor-α) [37], and potentially micro RNAs [35] and exosomes [38]. Thus, various proteins are involved in radiation-induced damage and repair. On the other hand, the results of this study demonstrated the occurrence of dose-dependent protein oxidative modification by single TBI, revealing molecular damage to important proteins in targeting theory and non-targeted effects. Radman et al. reported that the first bottleneck in cell recovery from radiation damage is functional (proteome) rather than informational (DNA) [39,40]. They also indicated that although proteins and DNA are equally important for long-term survival, residual proteome activity after radiation stress can repair inactive damaged DNA and make it fit again for transcription and replication, but DNA cannot restore the proteome without pre-existing relevant protein activity. The results of this study–that single TBI causes many oxidative modifications to serum albumin in a dose-dependent manner– suggest that other proteins in the body undergo similar oxidative modifications. This means that, as Rithidech et al. postulated, changes in the expression levels of proteins may potentially be associated with late-occurring adverse effects [29]. Thus, protein variation and serum albumin oxidative modification responses found in exposed individuals are important indicators for considering the effects of radiation on living organisms along with DNA damage, in addition to their use as biomarkers estimation of the radiation dose.

#### **5. Conclusions**

Our previous report suggested that biological responses to chronic low-dose radiation in humans can be assessed by fluctuations in certain blood proteins and oxidative modification of HSA. The present results revealed significant increases or decreases in the serum levels of six proteins and demonstrated a dose-dependent oxidative modified region in serum albumin prepared from acute single TBI mice. Although the dose-dependent profiles of OMSA differed between acute single TBI in mice and chronic low-dose exposure in humans, the amino acid residues that correlate with exposure are all located in domain IB of albumin. It is interesting to note that the domains of albumin that are sensitive to oxidative reactions are consistent. These radiation responses are expected to have the potential to be used as biomarkers of acute and chronic radiation exposure in living organisms. DNA, the genetic material that holds all of the information of life phenomena, is an important biological target of ionizing radiation. Protein damage caused by ionizing radiation also needs to be considered in more detail.

**Supplementary Materials:** The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/antiox11091710/s1, Figure S1: Correlation between the Serpin A3K expression levels with single TBI and the radiation dose.; Table S1: Oxidative modification pattern of peptide sequence consisting serum albumin in mice exposed to acute high dose/dose-rate radiation.

**Author Contributions:** Conceptualization, M.Y., S.T. and I.K.; methodology, M.Y., Y.T., E.D.N., Y.S. and T.M.; software, M.Y. and Y.T.; dose estimation, E.D.N. and M.H.; data curation, M.Y. and Y.T.; writing—original draft preparation, M.Y. and I.K.; writing—review and editing, M.Y., Y.T. and I.K.; visualization, Y.T.; project administration, M.S., S.T. and I.K.; funding acquisition, M.Y. and I.K. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was partially supported by the Japan Society for the Promotion of Science (JSPS) KAKENHI Grant Nos. JP16K15368, JP18KK0261, JP18K10023, JP18K18190, JP20H00556, JP21H02860, and Hirosaki University Institutional Research Grant.

**Institutional Review Board Statement:** The study was approved by the animal experiment committee of the CLEA Japan (Tokyo, Japan) (approval number: G17001).

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** All raw data for proteomics experiments are provided in supplementary information Table S1. The proteomics data are also available online using accession numbers "PXD025948, PXD025949, PXD025950" for Proteome Xchange [41] and accession numbers "JPST001166, JPST001167, JPST001168" for jPOST Repository [42]. Any additional data that support the findings of this study are available from the corresponding author upon reasonable request.

**Acknowledgments:** The authors are grateful to Miyu Miyazaki at the Center for Scientific Equipment Management, Hirosaki University Graduate School of Medicine, for help with the LC-MS/MS analysis.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


## *Review* **Targeting Mitochondrial Metabolism to Reverse Radioresistance: An Alternative to Glucose Metabolism**

**Chenbin Bian 1,2,3, Zhuangzhuang Zheng 1,2,3, Jing Su 1,2,3, Huanhuan Wang 1,2,3, Sitong Chang 1,2,3, Ying Xin 4,\* and Xin Jiang 1,2,3,\***


**Abstract:** Radiotherapy failure and poor tumor prognosis are primarily attributed to radioresistance. Improving the curative effect of radiotherapy and delaying cancer progression have become difficult problems for clinicians. Glucose metabolism has long been regarded as the main metabolic process by which tumor cells meet their bioenergetic and anabolic needs, with the complex interactions between the mitochondria and tumors being ignored. This misconception was not dispelled until the early 2000s; however, the cellular molecules and signaling pathways involved in radioresistance remain incompletely defined. In addition to being a key metabolic site that regulates tumorigenesis, mitochondria can influence the radiation effects of malignancies by controlling redox reactions, participating in oxidative phosphorylation, producing oncometabolites, and triggering apoptosis. Therefore, the mitochondria are promising targets for the development of novel anticancer drugs. In this review, we summarize the internal relationship and related mechanisms between mitochondrial metabolism and cancer radioresistance, thus exploring the possibility of targeting mitochondrial signaling pathways to reverse radiation insensitivity. We suggest that attention should be paid to the potential value of mitochondria in prolonging the survival of cancer patients.

**Keywords:** radioresistance; reactive oxygen species; oxidative phosphorylation; oncometabolites; apoptosis

#### **1. Introduction**

Cancer is a serious problem that threatens human life, and the number of cancerrelated deaths and incidences are increasing annually. According to the 2020 World Cancer Report, 4.57 million new cancer cases and 3 million cancer-related deaths have occurred in China, ranking it first in the world for cases and deaths [1]. As a traditional cancer treatment, radiotherapy causes nuclear DNA damage directly via ionizing radiation (IR) or indirectly via the production of reactive oxygen species (ROS), pushing cancer cells with high levels of DNA damage over the threshold for cell death [2]. As some tumors, such as malignant lymphoma, testicular seminoma, and nephroblastoma, are highly sensitive to IR, an explosion of interest in the role of radiotherapy in eradicating tumor cells has been observed in recent decades [3–5]. Mitochondria exist in most cells and are the main sites of cellular aerobic respiration, adapting to rapid tumor growth demands by regulating the process of energy production [6]. It is worth noting that mitochondria are the most important targets of IR damage aside from the nucleus [7]. Radiation-induced mitochondrial DNA mutations and electron transport chain (ETC) disruption activate oxidative stress and eventually trigger the mitochondrial apoptosis pathway, which seriously affects the survival of tumor cells [8]. However, tumor cell resistance to IR remains an important obstacle that hinders the clinical application of radiotherapy, potentially leading to poor

**Citation:** Bian, C.; Zheng, Z.; Su, J.; Wang, H.; Chang, S.; Xin, Y.; Jiang, X. Targeting Mitochondrial Metabolism to Reverse Radioresistance: An Alternative to Glucose Metabolism. *Antioxidants* **2022**, *11*, 2202. https://doi.org/10.3390/ antiox11112202

Academic Editors: Elena Obrador, Alegria Montoro and Stanley Omaye

Received: 20 September 2022 Accepted: 2 November 2022 Published: 7 November 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

prognosis, tumor recurrence, and metastasis [9,10]. In addition, radioresistance increases the incidence of radiation-induced damage to normal tissue cells surrounding the tumor and the disruption of homeostasis, mainly manifesting as radiation pneumonitis, intestinal dysbiosis, hemorrhage, and cardiac-related complications [11,12]. Fractionated treatment regimens have been established for radiotherapy. As fractionation is the process of dividing a radiation dose into multiple fractions, fractionated radiotherapy ensures as much tumor cell death as possible while reducing normal tissue complications. However, IR has been shown to activate epithelial–mesenchymal transition transcription factors such as Snail, Slug, Twist, ZEB1/2, hypoxia-inducible factor 1 (HIF1), and signal transducer and transcriptional activator 3 (STAT3); interfere with glucose and mitochondrial metabolism; promote metastatic potential; and increase the likelihood of radioresistance [13–15].

Metabolic disorders have long been recognized as carcinogenic factors [16]. Metabolic reprogramming, the alteration in metabolic pathways by which cancer cells can proliferate rapidly, survive under conditions of nutrient depletion and hypoxia, and evade the immune system, is considered a hallmark of cancer [17].

Glucose is the primary energy source that drives the rapid proliferation of cancer cells, and cancer starvation therapy based on glucose deprivation to induce oxidative stress has become an effective method for inhibiting tumor growth and survival [18]. 2-Deoxy-D-glucose (2DG), a glucose analog, targets glucose metabolism to deplete energy in cancer cells [19]. For most cancer cells, 2DG treatment alone does not significantly induce cell death, but renders cells more vulnerable to the oxidative stress induced by radio- or chemotherapy [20]. For example, 2DG combined with cisplatin or radiation enhances the cytotoxicity of head and neck squamous cell carcinoma through metabolic oxidative stress [21]. Furthermore, inhibition of glycolysis (2DG) and intracellular redox metabolism (glutathione/thioredoxin) improves the radiation response of radioresistant cervical cancers [22]. Unexpectedly, glucose deprivation promotes the death of malignant cells and induces colorectal cancer migration, invasion, and epithelial–mesenchymal transition (EMT). Knockdown of thioredoxin-1 can decrease G6PD protein expression and activity thereby reducing NADPH production, increasing ROS levels, enhancing glucosestarvation-induced cell death, and reversing aggressive or metastatic potential during cancer progression [23]. Rapidly proliferating cells tend to have high G6PD activity, while the pentose phosphate pathway (PPP) is the main pathway for glucose catabolism, and its reductant NADPH can be used to detoxify intracellular ROS, thus acting as an antioxidant defense [24]. During oxidative stress, cancer cells selectively shut down the glycolytic pathway, thereby increasing the glucose flux through PPP to meet the need for NADPH synthesis [25]. Snail, a key transcriptional repressor of EMT, regulates the glucose flux between glycolysis and PPP by inhibiting the platelet isoform of phosphofructokinase (PFKP) expression, which plays an important role in cancer cell survival [26]. Thus, interfering with the PPP to disrupt NADPH homeostasis not only enhances radiotherapy-induced immunogenic cell death but also overcomes cisplatin resistance [27,28].

Because of these classical conclusions, it was erroneously believed that malignant cells met their bioenergetic and anabolic needs primarily through glucose metabolism, and the role of mitochondrial metabolism in all steps of tumorigenesis was ignored [29]. However, the latest research has indicated that malignant transformation, tumor progression, and evasion of exogenous stress are influenced by mitochondria metabolism [30,31]. In addition, although the effect of radiation therapy is primarily dependent on glucose metabolism, there is growing awareness that changes in mitochondrial metabolism, such as mitochondrial function associated with antiradiation effects, also contribute to the development of radioresistance in head and neck squamous cell carcinomas and gliomas (Figure 1) [32,33]. Changes in mitochondrial size and shape or mutations in mitochondrial DNA interfere with the normal physiological function of mitochondria, thereby enhancing their adaptability to radiation [34]. Therefore, it is necessary to understand the molecular mechanisms underlying these changes caused by mitochondria to improve the efficacy of radiotherapy. Here, we briefly review the research progress on the relationship between mitochondrial metabolism and radioresistance from four aspects: regulating oxidative stress, participating in oxidative phosphorylation (OXPHOS), producing oncometabolites, and triggering apoptosis (Figure 1), with a focus on the possibility of targeting mitochondrial metabolism for cancer therapy.

**Figure 1.** A schematic model illustrating the effects of two major metabolisms on radioresistance. Six targets in glucose metabolism have the most significant impact on radiation resistance, regulating their corresponding molecules or processes to intervene in the therapeutic effect. In general, smallmolecule inhibitors can be used to help IR restore the expected efficacy, and there are several approved drugs currently available for clinical treatment. The effects of mitochondrial metabolism on radiation resistance can be summarized in four aspects, the details of which are presented below. Abbreviations: GLUT1/4—glucose transporter 1/4, PFK—phosphofructokinase, HK1/2—hexokinase 1/2, PKM2—pyruvate kinase M2, HIF1 α—hypoxia-inducible factor 1 α, OXPHOS—oxidative phosphorylation, ROS—reactive oxygen species.

#### **2. ROS and Radioresistance**

Reactive oxygen species are products of normal cellular metabolism and mainly include superoxide anions (O2 −), hydrogen peroxide (H2O2), and hydroxyl radicals ( −OH) [35]. Multiple lines of evidence indicate that there are three main sources of ROS in vivo, namely macrophages, the mitochondrial respiratory chain, and mitochondrial polyunsaturated membrane lipid peroxidation, a process during which ROS from mitochondrial polyunsaturated membranes pose the greatest threat to cells [36]. Under normal physiological conditions, cells tend to maintain redox homeostasis, that is, the balance between the production of free radicals and reactive metabolites (oxidants, ROS, or reactive nitrogen species) and their elimination through protective mechanisms (antioxidant systems) [37]. When the balance between ROS and antioxidants is disrupted, the body is in a state of oxidative stress, resulting in damage to important biomolecules and cells with potential effects on the entire organism [38]. It is worth noting that oxidative stress is generally present in tumor cells, and data show that the concentration of ROS is usually 10 times higher than that in normal cells, which may further lead to DNA mutation, genomic instability, and tumor cell proliferation [39].

Most ROS in mammalian cells are generated by the mitochondrial oxidative respiratory chain [40]. Furthermore, an inextricable relationship exists between ROS production and radioresistance. Mitochondrial H2O2 can trigger the accumulation of potential oncogenic DNA or activation of potential oncogenic signaling pathways, including the mitogenactivated protein kinase (MAPK) and epidermal growth factor receptor (EGFR) signaling

pathways, thereby promoting cell proliferation and malignant transformation [41,42]. Experiments have demonstrated that activation of the MAPK and EGFR signaling pathways can increase the resistance of cervical and lung cancer cells to radiation, respectively, and knockout of thyroid hormone receptor interactor 4 (TRIP4) promotes the inactivation of MAPK signaling, which effectively improves the sensitivity of the former to radiation [43,44]. It has also been reported that activated O2 −- and H2O2-mediated cell survival in non-small-cell lung cancer (NSCLC) occurs via the c-Met-PI3K-Akt and c-Met-Grb2/SOS-Ras-p38 pathways [45]. Interestingly, acidosis is a common characteristic of the tumor microenvironment. Under such acidic conditions, the specific mitogen reaction of cancer cells reduces extracellular acidification and increases O2 − production by switching from glycolysis to OXPHOS, which promotes tumor invasiveness and insensitivity to radiation therapy [31]. However, the mechanism appears to be different in endothelial cells, where mitochondrial ROS (mtROS) can stimulate the activity of NAD(P)H oxidases (NOXs), resulting in a positive feedback loop of ROS-induced ROS generation [46]. Recent studies have also shown that NOXs significantly contribute to H2O2 and O2 − production in gastrointestinal and pancreatic cancers [47,48]. Kim etal. reported that the novel PPARGligand PPZ023 can lead to NOX4-derived mtROS generation to induce death of radioresistant NSCLC cells via exosomal endoplasmic reticulum stress [49].

The reason for these diametrically opposite results may be the dual role of ROS, in which the difference in the level of ROS is dominant (Figure 2). In normal cells, ROS are produced at low concentrations and are effectively neutralized by the potent antioxidant systems of the cells. A moderate increase in ROS levels as in states of chronic oxidative stress induces random mutations in cells and promotes tumor cell proliferation, metastasis, and radioresistance. If ROS levels continue to increase beyond the antioxidant capacity of cells, this will cause apoptosis, ferroptosis, or cuproptosis, thereby significantly improving the efficacy of radiotherapy [38,50]. Sublethal levels of ROS stimulate tumor cell proliferation by inhibiting tumor suppressors such as redox-sensitive phosphatase and tensin homologues (PTEN), thereby promoting the PI3K-Akt signaling pathway or stabilizing HIF1 α, and are associated with chemotherapy resistance and prevention of tumor cell death [51]. In addition, a slight increase in superoxide can activate signal transduction pathways related to metastasis, including the mtROS-Src-SMAD-Pyk2 signaling pathway; in particular, Src can also promote radiation resistance in glioblastoma (GBM) [52,53]. Combining the novel Src inhibitor Si306 with radiotherapy represents a promising approach to increasing the therapeutic effect on GBM [54]. Importantly, moderately elevated ROS levels increase the resistance of cancer cells to radiotherapy by triggering an adaptive hormetic response and promoting autophagy activation [55,56]. Conversely, in the case of severe oxidative stress, ROS cause regulated cell death (RCD) or trigger apoptosis independently of DNA damage, thereby increasing the sensitivity to radiotherapy (Figure 2) [57]. For example, elesclomol (STA-4785) targets tumor ROS, which can further increase ROS levels in tumor cells, induce cytotoxicity in tumor cells, and selectively induce apoptosis in melanoma cells [58]. Unexpectedly, elesclomol did not show a significant radiosensitization effect on prostate cancer cells, indicating that there was no clear linear relationship between the specific ROS dose and radioresistance [59]. Of course, we should not ignore the fact that early and late ROS accumulation can lead to opposite carcinogenic effects. Radiationinduced early ROS signaling is responsible for the activation of Jak3-Erk-STAT3, which leads to a cell survival response, whereas late ROS production is different [60].

In cells, ROS production is counterbalanced by cellular antioxidant defense systems. Superoxide dismutases (SODs), the most potent antioxidant enzymes in mitochondria, can catalyze O2 − to H2O2 [61]. SOD-produced H2O2 can be subsequently reduced to H2O by catalases (CATs), glutathione peroxidases (GPXs), and peroxiredoxins (Prxs) [31]. To date, an increasing amount of evidence has suggested that the antioxidant stress system is responsible for radio- and chemoresistance [38]. Furthermore, ROS induced by chemoradiotherapy activate the Keap1-Nrf2 and PI3K-AKT pathways, which regulate several antioxidant enzymes in downstream signaling, ultimately triggering both radio- and chemoresistance [62,63]. The

inhibitors of these two signaling pathways, trigonelline and delicaflavone, can significantly reverse radioresistance and enhance radiosensitivity, further demonstrating the detrimental effects of the antioxidant stress system on cancer therapy [64,65]. Studies have shown an important relationship between an increase in the survival rate of pancreatic cancer cells after γ-ray irradiation and enhancement of the activity of manganese superoxide dismutase (Mn-SOD), the main antioxidant enzyme in the body, which also indicates that MnSOD significantly increases the resistance of pancreatic cancer to radiotherapy [66]. CuZn-SOD overexpression confers radioresistance on human glioma cells by suppressing irradiation-induced late ROS accumulation (superoxide) [67]. GPX4 inhibition promotes lipid peroxidation and re-sensitizes radioresistant cancer cells to IR-induced ferroptosis, resulting in radiosensitization [68]. In addition, redox-active metal ions are involved in antioxidant reactions, such as O2 −- and H2O2-mediated disruption of Fe metabolism, sensitizing NSCLC and GBM to pharmacological ascorbate [69]. However, recent studies have yielded conflicting results that antioxidant supplementation is detrimental to patients with adequate antioxidant status (lung, gastrointestinal tract, head and neck, and esophagus), whereas individuals with deficient antioxidant systems respond positively [47].

Based on the above studies, it can be concluded that ROS is a double-edged sword; it is one of the ways in which radiotherapy can eradicate tumor cells, yet even moderate intracellular concentrations may lead to radioresistance. Although the dual role of ROS is a major challenge in cancer therapy, it presents a promising strategy to differentiate normal cells from cancer cells using specific cellular signals to target tumor killing. An in-depth understanding of the dynamic balance between ROS and antioxidant levels and the role of ROS in different stages of the disease will help researchers to develop personalized therapies for different tumor types. Both disabling cellular antioxidants and adding specific ROS inducers provide new ideas for the precise treatment of tumors and the improvement in radiosensitivity.

#### **3. OXPHOS and Radioresistance**

Carbohydrates are the main source of cellular energy and are involved in the oxidative breakdown of glucose including glycolysis and OXPHOS [70]. Normally, cells favor the application of the mitochondrial OXPHOS, which is more efficient at producing ATP; however, the rate of glucose metabolism by aerobic glycolysis is 10–100 times faster than that of the complete oxidation of glucose in the mitochondria [71]. Therefore, Warburg initially believed that cancer cells could have an active glycolytic phenotype even in the presence of adequate oxygen supply and completely functioning mitochondria [72]. Warburg later proposed that this phenomenon was due to a developmental defect in the mitochondria of tumor cells that resulted in impaired aerobic respiration and reliance on glycolysis, hypothesizing that this event was the primary cause of cancer [73,74]. In contrast, Koppenol et al. offered a more plausible explanation, emphasizing the impairment of glycolytic regulation rather than mitochondrial respiration. There are clear indications that glycolysis is upregulated in most tumors without mitochondrial dysfunction. In these cancers, OXPHOS continues normally, even producing as much ATP as normal tissue at the same partial pressure of oxygen [75]. Notably, Weinhouse also strongly criticized the Warburg effect for his finding that well-differentiated Morris hepatomas do not produce lactic acid in aerobiosis [76]. At the same time, metabolic changes and adaptations occurring in tumors have been demonstrated to extend well beyond the Warburg effect and are seen as a secondary effect of tumorigenesis [77]. Subsequent studies have further revealed that certain types of tumors such as ovarian cancer and acute myeloid leukemia can also rely on mitochondria-specific OXPHOS to maintain biosynthesis and bioenergetics in addition to glycolysis [78]. Furthermore, in B16 melanoma the Warburg effect has been shown to be dispensable owing to the upregulation of mitochondrial metabolism [79].

Based on the above findings, we can conclude that metabolic reprogramming endows cancer with the ability to utilize multiple metabolic modalities to rapidly progress in vivo [80]. Some studies indicate that metabolic plasticity allows cells to efficiently produce energy through multiple metabolic pathways, thereby conferring on cancer cells a high degree of adaptability to a wide range of stresses and harsh tumor microenvironments [31]. In other words, tumor tissues are less sensitive to conventional chemoradiotherapy. In cancer cells that rely on glycolytic metabolism, OXPHOS can promote resistance to therapy through both the cancer-cell-intrinsic and -extrinsic pathways. In contrast, tumor cells that primarily utilize OXPHOS for energy production can become resistant to ETC inhibitors because they gain partial glycolytic metabolism (Figure 3) [81]. For example, recent studies have shown that acquired radioresistance is associated with a switch from glycolytic to oxidative metabolism in laryngeal squamous cell carcinoma cancer cells [32]. Similarly, a switch from glycolysis to OXPHOS was observed in glioma cells that developed acquired resistance to PI3K inhibitors [82]. In addition, glycolytic-dependent BRAF-mutant melanoma cells are more sensitive to the BRAF inhibitor vemurafenib, while resistant cells display upregulation of the mitochondrial biogenesis co-activator PGC1α through the melanocyte master regulator microphthalmia-associated transcription factor (MITF), leading to resistance to the original treatment and sensitivity to OXPHOS inhibitors [83]. Interestingly, glucose deprivation significantly promotes mitochondrial elongation, thereby inducing a metabolic shift from glycolysis to OXPHOS during energy stress in tumor cells, which is critical for hepatocellular carcinoma (HCC) survival [84]. Additionally, Dynaminrelated protein 1 (DRP1) is necessary for mitochondrial elongation in HCC cells. Elongated mitochondria amplify OXPHOS through facilitating cristae formation and assembly of respiratory complexes and in turn, exerting a feedback inhibitory effect on glycolysis through NAD-dependent SIRT1 activation [84]. Consistent with this, nutrient-deprivation-related OXPHOS/glycolysis interconversion has also been observed in glioma cell lines, although the role of mitochondrial dynamics has not been investigated [85]. The only exception is that dichloroacetate, which activates OXPHOS by reversing aerobic glycolysis, improves the radiosensitivity of high-grade gliomas [86]. In conclusion, the vast majority of malignant cells can switch freely between the two metabolic modes, simply inhibiting glycolysis or OXPHOS as a reasonable therapeutic candidate. The combination of a glycolysis inhibitor (2-DG) with an OXPHOS inhibitor (metformin) significantly enhances the radiosensitization of neuroblastoma and glioma cells, suggesting that dual metabolic targeting may be a good strategy to control tumor progression and eliminate radioresistance [87]. Unfortunately, the cytotoxic effect of this combination on normal tissue remains the biggest obstacle to its clinical application.

**Figure 3.** A schematic summary of glycolysis, TCA cycle, and OXPHOS in regulating radioresistance. (1) In the cytoplasm, pyruvate produced by glycolysis crosses the outer mitochondrial membrane and participates in the TCA cycle, and the subsequently produced NADH and FADH2 are oxidized by a stepwise, continuous enzymatic reaction on the ETC located in the inner mitochondrial membrane, thereby releasing energy for the body to utilize. (2) Inactivation of FH, SDH, D2HGDH, and L2HGDH, mutation of IDH1/2, or promiscuous activity of MDH/LDH can induce accumulation of oncometabolites. On the one hand, oncometabolites promote tumorigenesis; on the other hand, they also amplify the benefits of radiotherapy. (3) Enhancing OCR or favoring the reprogramming of tumor cells' metabolic pathways induces radioresistance. Abbreviations: TCA cycle—tricarboxylic acid cycle, ETC—electron transport chain, OCR—oxygen consumption rate, FH —fumarate hydratase, SDH—succinate dehydrogenase, D2HGDH—D-2-hydroxyglutarate dehydrogenase, L2HGDH—L-2-hydroxyglutarate dehydrogenase, IDH1/2—isocitrate dehydrogenase-1/-2, LDH/MDH—lactate dehydrogenase/malate dehydrogenase, α-KG—α-ketoglutarate, D-2HG—D-2-hydroxyglutarate, L-2HG—L-2-hydroxyglutarate.

Another mitochondrial condition of interest is hypoxia, which poses a problem for radiation therapists because the scarcity of oxygen induces radioresistance [88]. Efforts to increase oxygen delivery to tumors have not shown positive clinical effects because of poor tumor vascularization [89]. This implies that attempts to target tumor hypoxia should focus on normalizing oxygen levels in remote tumor regions by reducing the oxygen consumption rate (OCR). Therefore, an attractive strategy is to achieve this by inhibiting mitochondrial OXPHOS as it reduces the OCR, increases oxygenation, and thus improves the radiation response [90,91]. Several clinical trials are underway to repurpose FDA-approved drugs to curb mitochondrial function and reverse radioresistance. The antidiabetic drugs metformin and phenformin have been shown to increase the partial pressure of oxygen (pO2) in local tumors by inhibiting mitochondrial complex I, thereby significantly improving the effect of radiotherapy on colorectal cancer cells [92]. Another complex I inhibitory molecule, arsenic trioxide (As2O3), has shown strong superiority in the treatment of acute promyelocytic leukemia. However, in recent years, more attention has been paid to the potential of As2O3 to overcome radioresistance in solid tumors [93]. Both OXPHOS levels and the OCR are impaired by As2O3 to varying degrees in liver and lung cancer cells, with enhanced radiosensitivity [94]. Papaverine, a smooth muscle relaxant used as a vasospasm and erectile dysfunction agent, not only leads to reduced hypoxia and an increased response to radiotherapy in NSCLC and breast cancer by blocking complex I, but also has significantly fewer side effects than other OXPHOS inhibitors [95]. Atovaquone was originally used to treat and prevent parasitic infections; however, in hypopharyngeal, colorectal, and lung cancer cell lines, it significantly increased oxygenation and sensitized tumors to radiotherapy by inhibiting electron transport complex III (Table 1) [96].


#### **Table 1.** List of OXPHOS inhibitors under clinical trials or potential.

\* indicates that the drug has not been validated by clinical trials but has been confirmed in in vitro cell experiments and in vivo xenograft models.

In addition to the above-mentioned drugs that have been investigated in clinical trials as hypoxia regulators, there has also been an explosion of interest in small molecules that have the potential to overcome radioresistance. For instance, since annonacin is a natural lipophilic inhibitor of complex I, in addition to its known ability to promote selective cancer cell death through NKA- and SERCA-dependent pathways, it is reasonable to speculate that annonacin may also act as a radiosensitizer through its potential ability to target OXPHOS [103,105]. In addition, experiments have shown that the anthelmintic pyrvinium pamoate inhibits the proliferation of myeloma, erythroleukemia, and pancreatic cancer cells by targeting mitochondrial respiratory complex I [98,106]. Considering this, IR combined with pyrvinium pamoate is a promising future direction for addressing the unsatisfactory effects of radiotherapy on radioresistant pancreatic cancer cells (Table 1). However, further clinical research is needed to clarify various issues that may be overlooked by new treatments, such as balancing the relationship between therapeutic effects and toxic side effects.

Given that OXPHOS involves two distinct modalities that interfere with the radiation response along with active metabolic reprogramming activity or persistent local hypoxia in some tumors, targeting mitochondrial respiration to overcome radioresistance has attracted attention. Indeed, as the therapeutic index is the decisive factor for the utility of any therapy, those targeting OXPHOS are often limited by side effects rather than a lack of efficacy; therefore, there is an urgent need to find novel and more specific radiosensitizers.

#### **4. Oncometabolites and Radioresistance**

Oncometabolites, defined as metabolites that accumulate abnormally from distorted metabolic pathways, play pivotal roles in tumor transformation, cancer progression, invasiveness, and therapy resistance [107]. Mutations in the genes encoding isocitrate dehydrogenase 1/2 (IDH1/2) or promiscuous activity of lactate dehydrogenase/malate dehydrogenase (LDH/MDH) leads to the synthesis of D-2-hydroxyglutarate (D-2HG) and L-2-hydroxyglutarate (L-2HG), respectively [108]. Furthermore, loss of function of the tricarboxylic acid cycle enzymes succinate dehydrogenase (SDH) and fumarate hydratase (FH) results in the accumulation of succinate and fumarate (Figure 3) [109]. Oncometabolites act as structural mimics of α-ketoglutarate (α-KG) and thus competitively interfere with α-KG-dependent dioxygenases, which are involved in regulating the demethylation status of histones, RNA, and DNA, and targeting HIF-α degradation [110,111]. For example, oncometabolites lead to extensive hypermethylation of histone 3 lysine 9 (H3K9me3) by inhibiting histone lysine demethylase (KDM), which hinders the recruitment of DNA repair factors, leading to genomic instability that promotes tumor growth [112]. Thus, fumarate, succinate, D-2HG, and L-2HG have been characterized as bona fide tumor metabolites and have become pathognomonic hallmarks of a growing number of cancers (Figure 3), including neuroendocrine tumors, gliomas, leukemia, renal cell carcinomas, and head and neck squamous cell carcinomas [113–116]. In recent years, there has been great interest in the possible role of oncometabolites in cancer cell resistance to radiation, and numerous clinical trials have been conducted.

Furthermore, IDH1/2 mutations have been predicted in clinical trials and retrospective analyses to improve the response to radiotherapy in low-grade gliomas, showing significantly prolonged progression-free survival and overall survival [117]. However, mutated IDH1, when co-expressed with inactivating TP53 and alpha thalassemia/mental retardation syndrome Xlinked gene mutations in gliomas, induces genome stability and enhances the DNA damage response, triggering resistance to IR [118]. Thus, pharmacological inhibition of the DNA repair pathway is necessary if radiotherapy demonstrates superior therapeutic advantages in IDH1/2-mutated glioma cells. Studies have shown that FH expression in gastric cancer cells is significantly higher than that in nearby normal cells and is negatively correlated with patient prognosis. In addition, cisplatin is the first-line treatment for gastric cancer, and FH can significantly inhibit the cytotoxicity of cisplatin. Recent experiments have concluded that miconazole nitrate enhances the effects of cisplatin in vitro and in vivo by inhibiting FH activity [119]. Given that activated FH restrains sensitivity to traditional chemotherapy drugs, it could have the same adverse effects on radiation therapy. Patients with hereditary leiomyomatosis, renal cell cancer (HLRCC), and a substantial accumulation of fumarate are susceptible to kidney cancer with type 2 papillary morphology, which is refractory to current radiotherapy [120,121]. Interestingly, fumarate can covalently modify GPX4 and inhibit its activity, thereby activating ferroptosisselective HLRCC cell death [122]. Furthermore, SDH5 is required for the activity of the SDH complex, and its rapid depletion inhibits p53 degradation through the ubiquitin/proteasome pathway, thereby promoting apoptosis and enhancing NSCLC radiosensitivity [123].

Taken together, it is not difficult to see this as an interesting phenomenon, and while oncometabolites are beneficial for cancer progression, they also appear to significantly reverse the resistance of tumor tissue to IR, likely because tumors that accumulate high levels of oncometabolites are more vulnerable to therapies that cause DNA damage [124]. Moreover, it has been demonstrated that KDM induces transforming growth factor (TGF) β2 transcriptional activation by downregulating the enrichment of H3K9me3 at its promoter region. Activated TGF-β2 further enhances Smad/ATM/Chk2 signaling, which confers radioresistance in lung cancer [125]. Therefore, oncometabolites may be suitable signals

indicative of radiosensitivity, providing new insights into possible methods for predicting radiotherapy responses in patients who cannot tolerate biopsy.

#### **5. Apoptosis and Radioresistance**

Apoptosis is a tightly controlled mode of programmed cell death that plays an essential role in development, tissue homeostasis, and defense against unwanted, redundant, and potentially dangerous cells, particularly in the regulation of tumorigenesis [126,127]. The mitochondrial apoptotic pathway initiated by caspases and regulated by members of the Bcl-2 family of proteins or inhibitors of apoptotic proteins may have particularly relevant roles in radiation signal transduction. In differential expression analysis of related genes in cervical cancer cell lines, 33 genes have shown changes in expression after radiation induction, which may have potential effects on the apoptosis of cervical cancer cells after radiotherapy [128]. These findings suggest that IR can lead to significant changes in the expression of apoptosis-related genes, thereby inducing radioresistance. Some genetic dysregulation, as commonly observed in apoptotic signaling pathways in aggressive cancer cells, greatly limits the efficacy of anticancer treatments such as radiotherapy, which relies on these pathways to eradicate tumors [129]. Enhancing apoptosis to improve the therapeutic effect in cancer can be accomplished in two ways: by upregulating pro-apoptotic genes or by interfering with anti-apoptotic protein function [130].

We have summarized the apoptotic molecules associated with radiation resistance, and insights into the mechanisms involved can guide subsequent therapeutic approaches as follows. (1) NF-κB is increasingly recognized as a key player in many steps from cancer initiation to progression, with some degree of activation in various tumors, such as gastric, colorectal, lung, nasopharyngeal, and prostate carcinomas [131–134]. The activity of NF-κB is often enhanced by radiation and plays a central role in the resistance of cancer cells to radiation through the activation of the pro-survival proteins Bcl-2 and Bcl-XL in downstream signaling pathways [135]. Curcumin, one of the most important inhibitors of NF-κB, significantly delays tumor regeneration in irradiated mice [136]. (2) p53, a tumor suppressor gene, also has a diametrically opposite significance in the development of radioresistance. Downregulation of p53-induced death-domain-containing protein expression and inhibition of ataxia–telangiectasia-mutated protein (ATM) directly silence NF-κB, which inhibits DNA damage repair and ultimately increases the radiosensitivity of tumor cells [129]. Conversely, radiation-induced DNA damage can also activate the downstream effector kinase Chk2 of ATM, which contributes to further activation of p53 and pro-apoptotic proteins PUMA and BAX to induce apoptosis [137]. A study showed that the loss of components in the ATM/Chk2/p53 pathway was associated with radioresistance in a glioma mouse model [138]. Radiotherapy, the standard treatment for patients with nasopharyngeal carcinoma (NPC), induces DNA methyltransferase 3B, which greatly contributes to radioresistance in NPC by methylating p53 and p21 [139]. (3) Apoptosisrelated proteins in the TGF-β signaling pathway (ARTS) are alternative spliceosomes of the Sept4 gene located in the outer mitochondrial membrane [140]. As the only dual proapoptotic protein in vivo, ARTS directly bind to and restrain XIAP and Bcl-2 and assist p53 in inhibiting Bcl-XL [141]. Therefore, targeting the ARTS-mediated degradation of antiapoptotic proteins may represent an effective way of sensitizing tumor cells to radiotherapy. (4) Most patients with breast cancer treated with radiotherapy are completely cured, but in partial IR-induced triple-negative breast cancer, activated STAT3 and Bcl-2 and reduced ROS promote cell proliferation, reduce apoptosis, increase angiogenesis, and increase immune evasion, thus severely compromising the effectiveness of radiotherapy [142]. Niclosamide, a small-molecule STAT3 inhibitor, leads to a significant decrease in the protein levels of downstream anti-apoptotic target genes (such as Bcl-XL and survivin) by inhibiting Tyr-705 phosphorylation and nuclear translocation of STAT3, thereby improving the survival of patients with radioresistant breast cancer [142,143]. (5) Recent studies have shown that amplification of the cancer-associated gene YWHAZ is an indicator of poor prognosis in patients with urothelial carcinoma of the bladder (UCUB) [144]. Upregulation of YWHAZ

resulted in insufficient expression of pro-apoptotic proteins (BAK and BAX) and several caspases (CASP 3, 7, and 10) involved in mitochondrial apoptotic cascade reactions, with an emphasis on radiation insensitivity [145]. Notably, gene knockdown using a specific shRNA triggered a significant increase in cell death after radiation therapy, providing a new therapeutic target for YWHAZ-overexpressing UCUB [145].

In addition, Jumonji C domain histone lysine demethylases (JmjC-KDMs), Wnt1 inducible-signaling protein 1 (WISP1), and Caveolin-1 can also interfere with apoptosis and further induce radioresistance [146–148] (Figure 4). Therefore, elucidating the precise mechanism underlying the interaction between mitochondrial apoptosis and radioresistance would benefit the development of novel radiosensitizers. Although drugs developed based on this principle still require more clinical experiments to verify their indications and safety, they undoubtedly provide a promising starting point for the treatment of tumors with high target gene expression levels.

**Figure 4.** A schematic representation of the signaling mechanism of apoptosis-related molecules. IR induces DNA damage, leading to abnormal expression of mitochondria-related proteins, or directly regulates apoptosis-related genes, thereby promoting DNA damage repair and inhibiting mitochondrial apoptosis, ultimately causing the occurrence of radioresistance. Abbreviations: SSB/DSB—single-strand breaks/double-strand breaks, ATM—ataxia–telangiectasia-mutated protein, PIDD—p53-induced death-domain-containing protein, Chk2—checkpoint kinase 2, NF-κB nuclear factor-κB, RAIDD—receptor-interacting protein (RIP)-associated ICH-1/CED-3-homologous protein with a death domain, ARTS—apoptosis-related proteins in the TGF-β signaling pathway, STAT3—signal transducer and transcriptional activator 3, WISP1—Wnt1-inducible-signaling protein 1, JmjC-KDMs—Jumonji C domain histone lysine demethylases.

#### **6. Conclusions and Perspectives**

Acquired radioresistance is the main clinical obstacle for patients with tumors receiving radiotherapy and is affected by several factors. Since the Warburg effect was proposed, glucose metabolism has received unprecedented attention. However, a considerable number of studies point to the development of radioresistance closely related to mitochondrial metabolism, not only because mitochondria predominate in the tolerance of malignant

cells to radiation-induced RCD, but also because it underlies metabolic reprogramming. Maintaining the normal physiological function of the mitochondria is an important factor that improves the effect of radiotherapy. To date, many small-molecule inhibitors have been developed against ROS and oncometabolites or to regulate OXPHOS and apoptosis, which can target specific receptors and enhance the radiation response of tumor tissue. However, owing to the lack of high specificity, the indiscriminate attack of radiosensitizers on non-tumor cells can have unwanted effects, which also hinders their generalization. Due to different tumor types and specific metabolic processes or molecules, we need to individualize the treatment of tumors, so the development of more effective and specific sensitizers has become an irreplaceable solution. Despite these challenges, with a deeper understanding of the mechanism of radioresistance, targeting mitochondrial metabolism to reverse radiation insensitivity may be a safe and efficient radiosensitizing method in the future and thus deserves more attention.

**Author Contributions:** Conceptualization, X.J. and Y.X.; methodology, C.B., Z.Z., J.S., and H.W.; software, C.B., J.S., H.W., and S.C.; formal analysis, Z.Z.; data curation, C.B., J.S., and S.C.; writing—original draft preparation, C.B. and Z.Z.; writing—review and editing, X.J. and Y.X.; visualization, S.C.; supervision, Y.X.; funding acquisition, X.J. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was supported by the Jilin Provincial Science and Technology Foundations (20210509003RQ and 20210402002GH), Health Talents Special Project of the Jilin Provincial Finance Department (JLSWSRCZX2021-065), Changchun Science and Technology Development Plan (21ZY29), and Achievement Transformation Guiding Foundations of the First Hospital of Jilin University (CGZHYD202012-029).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


#### MDPI

St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Antioxidants* Editorial Office E-mail: antioxidants@mdpi.com www.mdpi.com/journal/antioxidants

Academic Open Access Publishing

www.mdpi.com ISBN 978-3-0365-8189-7