Abstract

   Chemoradiotherapy is the standard of care for many locally advanced
   cancers, including cervical and head and neck cancers, but many
   patients cannot tolerate chemotherapy. Clinical trials have shown that
   radiotherapy combined with hyperthermia (thermoradiotherapy) may be
   equally effective, yet it yields a suboptimal overall survival of
   patients, emphasizing the need for improvement. MicroRNAs (miRNAs),
   short non-coding RNA sequences, are often dysregulated in cancer and
   exhibit significant potential as radiosensitizers by targeting genes
   associated with the DNA damage response. In this study, high-throughput
   miRNA screening of four cervical cancer cell lines identified 55 miRNAs
   with significant sensitizing potential, with 18 validated across 10
   additional cancer cell lines (6 cervical and 4 head and neck).
   Functional studies of 6 miRNAs, including miR-16, miR-27a, miR-181c,
   miR-221, miR-224, and miR-1293, showed that they reduced DNA damage
   repair by downregulating ATM, DNA-PKcs, Ku70/80, and RAD51.
   Additionally, differential expression of miR-27a, miR-221, and miR-224
   in treatment-sensitive versus treatment-resistant patients indicated
   their predictive biomarker potential for treatment response of cervical
   cancer patients. Conclusively, this study has identified 18 promising
   miRNAs for the development of sensitizers for thermoradiotherapy and
   may provide potential biomarkers for predicting treatment response in
   locally advanced cancers.

   Keywords: MT: Non-coding RNAs, miRNA, thermoradiotherapy, DNA damage
   repair, 3D cell model, locally advanced cancer

Graphical abstract

   graphic file with name fx1.jpg
   [39]Open in a new tab
     __________________________________________________________________

   Snoek and colleagues identify 18 microRNAs (miRNAs) with significant
   potential to enhance thermoradiotherapy in cervical and head and neck
   cancers. By targeting DNA repair pathways, these miRNAs act as
   radiosensitizers and hold promise for improving therapeutic outcomes.
   Additionally, miR-27a, miR-221, and miR-224 show potential as
   biomarkers for predicting treatment response.

Introduction

   Chemoradiotherapy (radiation combined with chemotherapy) is the
   standard treatment for locally advanced cervical cancer and head and
   neck cancers. However, many patients cannot tolerate chemotherapy
   because of frailty and side effects (neurotoxicity, nephrotoxicity,
   bone marrow depression); thermoradiotherapy (radiation therapy combined
   with hyperthermia) serves as a well-established alternative treatment
   strategy.[40]^1 Preclinical and clinical studies have demonstrated a
   clear benefit of thermoradiotherapy over radiotherapy alone, and
   non-inferior to chemoradiotherapy with reduced
   toxicity.[41]^2^,[42]^3^,[43]^4^,[44]^5

   The clinical use of hyperthermia as a sensitizer of radiotherapy or
   chemotherapy is increasingly gaining recognition. Currently,
   thermoradiotherapy is applied for the treatment of recurrent breast
   cancer and advanced cervical cancer, and numerous studies have
   demonstrated the clinical efficacy of thermoradiotherapy for treatment
   of bladder cancer, esophageal cancer, head and neck cancers, and
   melanoma.[45]^1^,[46]^6^,[47]^7

   Despite these encouraging findings, the clinical outcome of patients
   with locally advanced cancer remains poor. For locally advanced head
   and neck cancer patients, the 5-year overall survival rate ranges
   between 50% and 78%, depending on the specific tumor site.[48]^8 The
   5-year overall survival rate for locally advanced cervical cancer
   patients treated with thermoradiotherapy ranges between 40% and
   60%.[49]^9^,[50]^10^,[51]^11 Bladder cancer patients treated with
   thermoradiotherapy show a reported 3-year overall survival rate of only
   28%.[52]^3 These observations underscore the clinical need for
   improving treatment outcomes.

   Hyperthermia is employed as a sensitizer of radiotherapy. During
   hyperthermia, the tumor is locally heated to mild temperatures ranging
   between 39°C and 42°C for approximately 1 h. This process has been
   demonstrated to enhance cell sensitivity to radiation through various
   mechanisms, such as (1) inhibition of radiation-induced DNA damage
   repair due to temporarily inhibiting the homologous recombination (HR)
   DNA damage repair pathway, (2) sensitization of S phase cells, and (3)
   targeting tumor cells in hypoxic and nutrient-deprived
   areas.[53]^1^,[54]^12^,[55]^13 In addition to its radiosensitizing
   properties, hyperthermia has been found to trigger systemic anti-tumor
   immune responses.[56]^14^,[57]^15^,[58]^16

   Previously, we have demonstrated that the high specificity of
   hyperthermia against human papillomavirus (HPV)^+ cervical cancers
   results from its ability to temporarily restore p53 function by
   disrupting the binding between the HPV oncoprotein E6 and p53.[59]^17
   This p53 restoration leads to induced tumor cell apoptosis following
   sublethal DNA damage caused by irradiation. In addition, Wang et al.
   found that hyperthermia leads to the downregulation of HPV oncoprotein
   E7, effectively reversing HPV-associated carcinogenesis in vitro and
   in vivo.[60]^18 Furthermore, Jiang et al. showed that phosphorylation
   of phospholipid scramblase 3, a protein often associated with cancer,
   is involved in the signal transduction mechanism underlying
   hyperthermia-induced apoptosis in tongue squamous cell
   carcinoma.[61]^19

   MicroRNAs (miRNAs) are a class of small non-coding RNAs approximately
   20–25 nt in length, which play a pivotal role in post-transcriptional
   gene regulation and are often dysregulated in cancer.[62]^20^,[63]^21
   miRNAs have been implicated in tumor radiosensitivity by targeting key
   components of the DNA damage response.[64]^22 However, most studies
   have focused on individual miRNAs. No studies have investigated the
   effects of miRNA modulation on thermoradiotherapy.

   In the present study, we aim to discover miRNAs that are modifiers of
   thermoradiotherapy and are potential agents for the development of
   sensitizers of thermoradiotherapy. Here, we use a high-throughput miRNA
   screen in three-dimensional (3D) tumor cell spheroids, a model that we
   recently developed and further optimized to assess the effectiveness of
   thermoradiotherapy.[65]^23 Comprehensive target prediction, combining
   computational and experimental approaches, were used to investigate the
   association of identified miRNAs with DNA damage repair. The function
   of miRNAs was subsequently studied with clonogenic assays, γ-H2AX foci
   staining, and automated western blots. The biomarker potential was
   assessed using miRNA expression profiles of treatment-sensitive and
   treatment-resistant patients obtained from The Cancer Genome Atlas
   (TCGA) cervical cancer cohort.[66]^24

Results

Discovery screen identifies 55 miRNAs as potent sensitizers for
thermoradiotherapy

   To identify potential sensitizers for thermoradiotherapy, we developed
   a high-throughput miRNA screening method using 3D spheroid models and
   cell viability as readout. The discovery screen entailed transfection
   of cervical cancer cells with 378 miRNA mimics (see [67]materials and
   methods for selection criteria and [68]Table S1) in 384-well plates
   followed by irradiation at 2 Gy and hyperthermia at 42°C for 60 min.
   Three days post-treatment, the cells were assayed for viability. The
   strategy of this study is outlined in [69]Figure 1A.

Figure 1.

   [70]Figure 1
   [71]Open in a new tab

   Overview of screen strategy and results of the discovery screen across
   four cervical cancer cell lines

   (A) Schematic representation of the study outline. (B) Relative effect
   sizes obtained for each miRNA by comparison to the modal effect (across
   all miRNAs). miRNAs with a significant treatment effect (FDR <0.1; n =
   55) are highlighted in blue. Additionally, miR-140-5p, with a relative
   effect size of approximately 0.00 is indicated in green. (C) Cell
   viability results normalized to negative control (NC; gray dotted line)
   for each cell line (C-4I, CaSki, HeLa, and SiHa), with the modal effect
   (across all miRNAs; orange dotted line), miR-140-5p as representative
   “control miRNA,” and the top 5 identified miRNAs indicated. Means ± SDs
   are shown. Untreated refers to miRNA transfected cells only, while 2
   Gy+HT refers to cells that received thermoradiotherapy combined with
   miRNA transfection.

   Given that thermoradiotherapy is clinically applied for the treatment
   of locally advanced cervical cancer, we conducted our discovery screen
   across four cervical cancer cell lines (C-4 I, CaSki, HeLa, and SiHa).
   We performed a comprehensive analysis of the obtained data from all
   cell lines combined to identify miRNAs with a broad sensitizing effect.
   The miRNA-specific treatment effects, measured by tumor cell viability
   after treatment, were compared to the modal effect (across all miRNAs)
   and are indicated as relative effect size, as shown in [72]Figure 1B.
   Due to our specific focus on the sensitization effect, our statistic
   model was established to detect a larger treatment effect, resulting in
   a one-side volcano plot. A false discovery rate (FDR) <0.1 indicates
   miRNAs with statistically significant treatment effects. Our discovery
   screen identified 55 miRNAs with significant sensitizing effect to
   thermoradiotherapy ([73]Figure 1B; [74]Table S1). While the modal
   effect shows a relative effect size of 0.00, miRNAs with significant
   sensitizing potential exhibit relative effect sizes ranging between
   −0.11 and −0.57 ([75]Figure 1B). Consistent effects were observed
   across all four cervical cancer cell lines ([76]Figure 1C). Notably,
   C-4 I and HeLa cells were more sensitive to thermoradiotherapy compared
   to CaSki and SiHa cells. The modal effect observed for each cell line
   aligns with the cell viability results of the negative control (NC),
   underscoring the robustness of the statistical model.

Validation screen reveals 18 miRNAs with significant sensitizing potential

   The 55 candidate miRNAs identified in the discovery screen were
   subsequently validated in an independent set of 10 cancer cell lines,
   including 6 cervical cancer cell lines (778, 879, C-33A, DoTc2 4510,
   HT-3, and MS751) and 4 head and neck cancer cell lines (UM-SCC-47,
   VU-SCC-040, VU-SCC-120, and VU-SCC-147). Head and neck cancer was
   chosen because of the demonstrated efficacy of hyperthermia in a number
   of clinical trials.[77]^1^,[78]^25 Additionally, we aimed to explore
   whether the identified miRNAs exhibit sensitizing effects beyond
   cervical cancer.

   Similar as in the discovery screen, we used 3D tumor spheroids for the
   validation screen and evaluated cell viability 3 days post-treatment
   (2 Gy irradiation and hyperthermia at 42°C for 60 min). In addition to
   the 55 candidate miRNAs, we selected 13 miRNAs from the discovery
   screen as “control” miRNAs, based on their relative effect size of
   approximately 0.00 ([79]Table S1). These control miRNAs were used to
   calculate the effect size of the 55 candidate miRNAs. The analysis was
   conducted on the combined data of all cell lines and identified 18
   miRNAs (miR-16, miR-22, miR-27a, miR-92a-1, miR-106b, miR-181a,
   miR-181c, miR-181d, miR-193a, miR-195, miR-221, miR-224, miR-330,
   miR-375, miR-455, miR-497, miR-1293, and miR-3158) with a significant
   sensitizing effect to thermoradiotherapy, demonstrating effect sizes
   ranging between −0.30 and −1.49 ([80]Figure 2A). Importantly, the
   thermoradiosensitizing effect mediated by the 18 identified miRNAs
   demonstrated consistency across most cell lines, as illustrated for the
   top 5 identified miRNAs in [81]Figure 2B. However, we observed
   variations in cell line sensitivity to thermoradiotherapy, with cell
   lines 879, VU-SCC-040 and VU-SCC-120 showing the strongest sensitivity.
   Notably, miR-224 demonstrated the strongest sensitizing effect in the
   majority of cell lines.

Figure 2.

   [82]Figure 2
   [83]Open in a new tab

   Results of the validation screen across 10 independent cervical and
   head and neck cancer cell lines

   (A) Log2 effect sizes obtained for each miRNA by comparison to the
   control miRNAs (n = 13), with only significant miRNAs shown (18 out of
   55; ∗p < 0.05; ∗∗p < 0.001; ∗∗∗p < 0.00001). The top 5 miRNAs with the
   strongest sensitizing effect are highlighted in red. (B) Cell viability
   results for the top 5 identified miRNAs normalized to NCs (controls) on
   each cell line (untreated as gray dotted lines; 2 Gy+HT as orange
   dotted lines). Means ± SDs are shown. Statistical analysis of
   individual cell lines did not reveal significant differences, likely
   due to the limited number of samples per group (3 vs. 3). Untreated
   refers to miRNA transfected cells only, while 2 Gy+HT refers to cells
   that received thermoradiotherapy 1 day after miRNA transfection.

Identified miRNAs are associated with DNA repair pathways

   Previous studies have shown that miRNAs can enhance sensitivity to
   ionizing radiation by targeting genes involved in DNA damage
   response.[84]^26 To explore potential gene targets for the 18 miRNAs
   identified from our screens, we used the ENCORI or miRDB platform for
   miRNA target prediction,[85]^27 followed by pathway analysis using the
   Database for Annotation, Visualization, and Integrated Discovery
   (DAVID) ([86]Figure 3A).[87]^28 Interestingly, pathway enrichment
   analysis using the predicted targets showed that the ataxia
   telangiectasia mutated (ATM)-dependent DNA damage response was one of
   the most significantly enriched pathways for 6 of the 18 miRNAs
   ([88]Figure 3B; miR-16, miR-181c, miR-193a, miR-224, miR-375, and
   miR-497), supporting their involvement in inhibiting
   irradiation-induced double-strain break. For three other miRNAs
   (miR-27a, miR-330, and miR-455), we also found enrichment of the DNA
   damage response; however, this was not significant ([89]Figures 3B and
   [90]S1). Notably, miR-221 showed broad targeting ability on the DNA
   damage response, as demonstrated by the enrichment of pathways
   associated with the DNA damage response related to both ATM and ataxia
   telangiectasia and Rad3-related (ATR) kinases ([91]Figure 3C). An
   overview of significantly enriched pathways for each miRNA is provided
   in [92]Figure S1. Most of the predicted miRNA targets were associated
   with the two main DNA repair pathways: HR and non-homologous end
   joining (NHEJ) ([93]Table 1). All other predicted targets linked to the
   DNA damage response are listed in [94]Table S2.

Figure 3.

   [95]Figure 3
   [96]Open in a new tab

   Overview of miRNA target prediction strategy, pathway enrichment, and
   correlation heatmap

   (A) Flowchart of miRNA target prediction using the ENCORI (13 miRNAs)
   or miRDB (5 miRNAs) platform. (B) miRNAs that show significant
   enrichment for the DNA damage response. Gene ratio indicates the
   percentage of significant genes over the total genes associated with
   the DNA damage response. (C) The top 5 enriched pathways for miR-221
   predicted targets. (D) Correlation heatmap illustrating miRNA-target
   co-expression in TCGA cervical cancer dataset for genes associated with
   DNA repair pathways HR and NHEJ. miRNAs selected for functional
   validation are indicated in bold. ∗p < 0.05; ∗∗p < 0.01. HITS-CLIP,
   high-throughput sequencing of RNA isolated by crosslinking
   immunoprecipitation.

Table 1.

   Overview of predicted miRNA targets associated with DNA damage repair
   pathways HR and NHEJ
   miR-16-5p miR-22-5p miR-27a-5p miR-92a-1-5p miR-106b-3p miR-181a-5p
   miR-181c-5p miR-181d-5p miR-193a-3p miR-195-5p miR-221-3p miR-224-3p
   miR-330-5p miR-375 miR-455-5p miR-497-5p miR-1293 miR-3158-3p
     __________________________________________________________________

   HR
     __________________________________________________________________

   ATM X X X X X
   BRCA1 X X
   MRE11 X X
   NBN X X X X X
   PAXIP1 X
   POLD3 X X X
   RAD50 X X X X
   RAD51 X X
   RAD52 X
   RAD54B X
     __________________________________________________________________

   NHEJ
     __________________________________________________________________

   LIG4 X
   NHEJ1 X
   POLL X
   POLM X X
   PRKDC X X X
   WRN X X
   XRCC5 X X X X
   XRCC6 X
     __________________________________________________________________

   Literature on radiosensitizer
     __________________________________________________________________

   yes yes no no no yes no no no yes no no no yes no yes no no
   CxCa[97]^29 NSCLC[98]^30 PCa[99]^31 HCC[100]^32 CRC[101]^33
   CxCa[102]^34 GBM[103]^35 BC[104]^36
   CRC[105]^37
   LC[106]^38 NSCLC[107]^39 CxCa[108]^41
   NB[109]^42 NSCLC[110]^43 OSCC[111]^44 BC[112]^45 CxCa[113]^46
   ESCC[114]^47 NSCLC[115]^48
   [116]Open in a new tab

   Predicted targets of each miRNA (n = 18) are marked with an X, along
   with relevant literature supporting their role as radiosensitizers in
   cancer. BC, breast cancer; CRC, colorectal cancer; CxCa, cervical
   cancer; ESCC, esophageal squamous cell carcinoma; GBM, glioblastoma;
   HCC, hepatocellular cancer; HR, homologous recombination; LC, laryngeal
   cancer; NB, neuroblastoma; NHEJ, non-homologous end joining; NSCLC,
   non-small cell lung cancer; OSCC, oral squamous cell carcinoma; PCa,
   prostate cancer.

   Given that miRNAs operate by downregulating their gene targets, we
   computed correlations between the 18 identified miRNAs and genes
   associated with DNA repair pathways HR and NHEJ using publicly
   available genome-wide miRNA and messenger RNA (mRNA) expression
   profiles from cervical and head and neck cancers.[117]^24^,[118]^40
   While each miRNA displayed at least one negative correlation, miR-22,
   miR-27a, miR-193a, miR-221, and miR-455 exhibited negative correlations
   with the majority of their predicted targets ([119]Figure 3D).

   Furthermore, a literature search was conducted that revealed supporting
   evidence for 7 of the 18 identified miRNAs in their role as
   radiosensitizers in cancer ([120]Table 1). The 6 miRNAs
   (miR-16,[121]^29^,[122]^30^,[123]^31 miR-22,[124]^32
   miR-181a,[125]^33^,[126]^34^,[127]^35
   miR-195,[128]^36^,[129]^37^,[130]^38^,[131]^39
   miR-375,[132]^41^,[133]^42^,[134]^43^,[135]^44 and
   miR-497)[136]^45^,[137]^46^,[138]^47^,[139]^48 have been linked to
   radiosensitivity of cervical cancer and head and neck cancer, but also
   other cancer types such as breast cancer, colorectal cancer, esophageal
   and gastric cancer, glioblastoma, hepatocellular cancer, neuroblastoma,
   lung cancer, prostate cancer, and thyroid cancer ([140]Table 1).

Thermoradiotherapy in combination with miRNAs results in reduced DNA damage
repair

   From the 18 identified miRNAs from the validation screen, 6 miRNAs were
   prioritized for subsequent functional validation. We selected miRNAs
   with the strongest sensitizing effect observed during the validation
   screen (effect size of <-0.5; miR-16, miR-27a, miR-181c, miR-224, and
   miR-1293). Additionally, miR-221 (effect size of −0.38) was included
   due to its numerous predicted targets associated with DNA repair
   pathways HR and NHEJ ([141]Table 1).

   We conducted clonogenic assays in SiHa cells to study cell survival
   following different treatment combinations, including miRNA mimic alone
   (untreated), irradiation (2 Gy) combined with miRNA mimics, and
   thermoradiotherapy (2 Gy and 42°C for 60 min) combined with miRNA
   mimics. We observed a significantly lower survival fraction when
   thermoradiotherapy was combined with miR-221, miR-224, or miR-1293
   mimics compared to thermoradiotherapy combined with a NC
   ([142]Figure 4A; representative images of the colony-forming assay are
   included in [143]Figure S2). Upregulated expression of miRNA after
   transfection was measured using quantitative reverse-transcription PCR
   (RT-qPCR) ([144]Figure S3B), with a fold change relative to the NC
   ranging from 10^2 to 10^8.

Figure 4.

   [145]Figure 4
   [146]Open in a new tab

   Response to irradiation (2 Gy), thermoradiotherapy (2 Gy+HT), and miRNA
   mimics in SiHa cells

   (A) Clonogenic survival fractions for each miRNA are shown and compared
   to NC. (B) Representative images of y-H2AX foci staining (in green) for
   each condition. (C) The number of y-H2AX foci (the number of
   double-stranded breaks) 24 h after thermoradiotherapy combined with
   miRNA mimics or NC. Each graph represents the γ-H2AX foci in each cell,
   with a minimum of 70 cells per experiment. (D) Normalized protein
   expression against NC. Means ± SDs are shown.
   ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001

   Furthermore, we investigated residual (unrepaired) DNA double-stranded
   breaks (DSBs) 24 h post-treatment using γ-H2AX foci staining.
   Thermoradiotherapy combined with either miR-16, miR-27a, miR-181c,
   miR-221, miR-224, or miR-1293 demonstrated a significant increase in
   DNA DSBs compared to the control, with the most pronounced increase
   observed for miR-221 ([147]Figures 4B and 4C).

   We subsequently performed western blot analysis to study the impact of
   miRNA modulation on the protein expression of several predicted targets
   associated with DNA repair pathways, specifically HR and NHEJ
   ([148]Figures 4D and [149]S4). Transfection of SiHa cells with miRNA
   mimics miR-16, miR-27a, miR-181c, miR-221, miR-224, or miR-1293 led to
   a reduction in ATM, a key player in HR. Regarding the regulation of
   NHEJ, we found strong effects for all six miRNAs on the protein
   expression of DNA-dependent protein kinase catalytic subunit
   (DNA-PKcs). In addition, we analyzed the protein expression of two
   other key players in NHEJ: Ku70 (gene name XRCC6) and Ku80 (gene name
   XRCC5). We observed that miR-1293 strongly reduced the protein
   expression of Ku80, while Ku70 expression was affected by miR-27a,
   miR-221, and miR-224. Additionally, these three miRNAs also reduced the
   expression of RAD51, a pivotal player in HR.

miR-27a, miR-221, and miR-224 exhibit predictive biomarker potential for
treatment response

   Next, we explored the predictive biomarker potential of the identified
   miRNAs. To accomplish this, we utilized genome-wide miRNA expression
   profiles from advanced stage cervical cancer patients obtained from
   TCGA.[150]^24 Unfortunately, TCGA dataset lacks information on cervical
   cancer patients who have undergone thermoradiotherapy, requiring us to
   focus on chemoradiotherapy instead. Using the clinical data from TCGA,
   we defined two groups of cancer patients based on their response to
   treatment: treatment-sensitive (n = 37), including those with complete
   and partial responses, and treatment-resistant (n = 18), including
   those with stable and progressive disease. Interestingly, we found that
   miR-27a, miR-221, and miR-224 are significantly downregulated in
   treatment-resistant cervical cancer patients ([151]Figure 5),
   indicating their potential as predictive biomarkers for treatment
   response in cervical cancer patients.

Figure 5.

   [152]Figure 5
   [153]Open in a new tab

   Expression profiles of miR-16, miR-27a, miR-181c, miR-221, and miR-224
   in TCGA cervical cancer dataset

   Cancer patients were divided into two groups based on their response to
   chemoradiotherapy: treatment sensitive (n = 37), including those with
   complete and partial responses, and treatment resistant (n = 18),
   including those with stable and progressive disease. ∗p < 0.05;
   ∗∗p < 0.01. RPM, reads per million.

Discussion

   Chemotherapy-based radiosensitization is the standard of care for
   treating locally advanced cervical and head and neck cancers. However,
   many patients cannot tolerate chemotherapy, and clinical trials have
   shown that thermoradiotherapy may be an equally effective alternative.
   miRNAs have previously been studied as potential targets for
   radiosensitization, but none of these studies have studied their
   potential role in thermoradiotherapy sensitization. In the present
   study, we identified and validated 18 miRNAs with significant
   sensitizing potential for thermoradiotherapy in 14 cervical and head
   and neck cancer cell lines. Functional validation showed that
   miRNA-mediated sensitization was achieved by targeting genes associated
   with DNA repair pathways, including RAD51 (HR) and Ku80 (NHEJ).

   An important aspect of this study is that we used a high-throughput
   miRNA screen on 3D spheroid models with cell viability as the readout,
   rather than conventional clonogenic assays to assess cell survival
   after treatment. Our approach overcomes the practical limitations of
   the clonogenic assay, including the long experiment duration, low
   sample throughput and time-consuming colony counting. The validity of
   our approach is supported by several findings. First, using the
   cellular ATP content as a measure of cell viability, we found a similar
   variation in sensitivity among cervical cancer cell lines and head and
   neck cell lines as reported by others using clonogenic assays.[154]^49
   This underscores the reliability of exploring treatment sensitization
   by measuring cell viability in a 3D model, while overcoming the
   difficulty when certain cell lines do not form colonies. Second, we
   observed similar sensitizing effects for most miRNAs when comparing
   results from clonogenic survival assays with cell viability results
   ([155]Figure 4B). This is consistent with previous studies showing
   similar results between these two types of assays.[156]^50^,[157]^51
   Third, among the discovered thermoradiosensitizing miRNAs, we
   identified several miRNAs previously described as potent
   radiosensitizers in cancer, including miR-16-5p, miR-181a-5p,
   miR-195-5p, miR-221-3p, miR-375-3p, and miR-497-5p ([158]Table 1).
   Notably, several of these miRNAs have been described as
   radiosensitizers in cervical and/or head and neck cancers. For example,
   miR-16-5p has been shown to modulate the radiosensitivity of cervical
   cancer cells by regulating coactivator-associated arginine
   methyltransferase 1 (CARM1).[159]^29 CARM1 plays a critical role in the
   DNA damage response by methylating p300, which enhances the interaction
   between p300 and BRCA1 (key player in HR), thereby promoting BRCA1
   transcriptional activity.[160]^52 Furthermore, miR-181a-5p was found to
   target Raf kinase inhibitor protein in nasopharyngeal cancer cells
   thereby enhancing radiotherapy resistance.[161]^53 In addition, Ke and
   colleagues found that miR-181a-5p can confer radiotherapy resistance in
   cervical cancer cells by targeting pro-apoptotic protein kinase C delta
   (PRKCD) and that its expression may serve as a prognostic biomarker for
   radiotherapy response.[162]^34

   On top we observed mostly consistent sensitizing effects in different
   cancer types (cervical and head and neck cancer), highlighting the
   potential broad applicability of these miRNAs in improving therapeutic
   outcomes. Some observed differences between cell lines are most likely
   due to inherent differences in radiation sensitivity and/or differences
   in endogenous miRNA levels between cell lines
   ([163]Figure S3A).[164]^54

   Interestingly, Western blot analysis revealed downregulation of ATM and
   DNA-PKcs (gene name PRKDC) after transfection of all six miRNAs
   ([165]Figure 4D). As early sensors of DNA DSBs, ATM and DNA-PKcs
   activate a series of kinase reactions essential for DNA DSB
   repair.[166]^55^,[167]^56 Therefore, the downregulation of ATM and
   DNA-PKcs could inhibit DNA damage repair, which is consistent with our
   findings from γ-H2AX stainings. However, the types of interactions may
   vary among the miRNAs. The negative correlation in expression levels
   between three miRNAs (miR-27a, miR-181c, and miR-221) and ATM was
   confirmed by TCGA data ([168]Figure 3D). Target prediction using a
   combination of computational and experimental strategies
   (high-throughput sequencing of RNA isolated by crosslinking
   immunoprecipitation [HITS-CLIP] data) also supports the interaction
   between these miRNAs and their targets, suggesting that ATM is directly
   targeted by miR-27a, miR-181c, and miR-221. For DNA-PKcs, although TCGA
   data show a negative correlation between the miRNAs and PRKDC, target
   prediction did not provide evidence for a direct interaction between
   the miRNAs and PRKDC, suggesting an indirect effect of the miRNAs.

   Repair of radiation-induced DNA DSBs is typically accomplished by two
   pathways: NHEJ and HR.[169]^57 During classical NHEJ, DNA DSBs are
   initially recognized by the Ku70-Ku80 heterodimers, which recruit and
   promote the binding of other NHEJ proteins, including DNA-PKcs, ligase
   4, and XRCC4.[170]^58 Thus, downregulation of Ku70 (observed with
   miR-16, miR-181c, miR-221, miR-224, and miR-1293) and Ku80 (observed
   with miR-16, miR-27a, miR-221, miR-224, and miR-1293) could inhibit
   NHEJ. While miR-16[171]^29 and miR-1293[172]^59 have been reported to
   target DNA repair genes, their effects on Ku70 and Ku80 have not been
   previously demonstrated. Target prediction supports miR-16 targeting of
   Ku80 (XRCC5), but the mechanism by which miR-1293 inhibits DNA damage
   repair requires further investigation.

   Alternatively, DNA DSBs can also be repaired by HR. Following
   initiation resection by the MRN complex (MRE11-RAD51-NBS1) and
   activation of ATM, HR is catalyzed by several enzymes, in particular by
   recombinase RAD51.[173]^60 RAD51 plays an important role in HR by
   assembling filament on DNA, interacting with other proteins, and then
   facilitating HR.[174]^61 In our validation, miR-221 mimic transfection
   resulted in a pronounced downregulation of RAD51 protein levels. This
   observation is further supported by computational and experimental
   target prediction, suggesting that miR-221 may inhibit HR by directly
   affecting RAD51 levels ([175]Figure 3D). These results are summarized
   in a network model shown in [176]Figure 6, which illustrates our
   proposed mechanism of miRNA-mediated sensitization of
   thermoradiotherapy.

Figure 6.

   [177]Figure 6
   [178]Open in a new tab

   Network model illustrating our proposed mechanism of miRNA-mediated
   sensitization to thermoradiotherapy based on the results of this study

   Direct regulation is indicated by black solid line, and indirect
   regulation is indicated by gray dotted line.

   One limitation of our study is the use of single-dose irradiation (2
   Gy) instead of applying a clinically relevant treatment regimen. For
   locally advanced cervical cancer, the clinical treatment regimen
   typically involves fractionated radiotherapy with 20–25 fractions and a
   total dose of 45–50 Gy.[179]^10 However, single-dose irradiation is
   often used in in vitro studies to simplify experimental conditions and
   allow for controlled analysis of cellular responses to
   radiation.[180]^49 Future studies using fractionated radiation may
   provide more clinically relevant insights and potentially confirm our
   findings. Additionally, our functional validation was conducted
   exclusively in cervical cancer cell lines. Expanding validation to
   other cancer cell lines, particularly head and neck cancer cells, as
   well as non-malignant cells, may strengthen our findings and provide
   valuable insights into the toxicity profile of the miRNAs. miRNA
   toxicity is a critical issue that could significantly impact
   therapeutic efficiency, making it essential to evaluate the toxicity
   profile of each miRNA. In this study, we used a 2-nM miRNA mimic
   concentration for each cell line, which was selected based on prior
   optimization efforts.[181]^23^,[182]^62 While miR-224 exhibited the
   strongest sensitizing effects across cell lines, it also notably
   reduced cell viability before treatment ([183]Figure 2B), suggesting
   general toxicity. To address this, we tested a lower concentration of
   miR-224 and found that it reduced its impact on cell viability
   ([184]Figure S5), particularly in HeLa and SiHa cells, demonstrating a
   dose-dependent effect. A similar trend was observed for miR-16,
   miR-27a, and miR-1293. These findings highlight the need for further
   optimization of miRNA mimic concentrations in future studies to
   minimize potential toxicity while maintaining therapeutic efficacy.

   While this study primarily focused on miRNA-mediated modulation of the
   two main DNA repair pathways, HR and NHEJ, other DNA damage
   response-associated genes remain largely unexplored in this context.
   Additionally, our pathway analysis revealed several other significantly
   enriched pathways, including those related to angiogenesis, apoptosis,
   cell-cycle regulation and immune responses, as shown in [185]Figure S1.
   Further investigation into these pathways and other DNA damage
   response-associated genes may provide additional leads for therapeutic
   intervention and could further advance our understanding of
   miRNA-mediated treatment sensitization.

   The field of miRNA-based therapeutics has gained significant interest
   due to the central role of miRNAs in gene regulation and their
   involvement in various cancers. Several miRNA mimics and inhibitors,
   such as miR-10b, miR-16, miR-34a, miR-155, and miR-193a, have been or
   are currently being investigated in clinical trials for the treatment
   of cancer and other diseases.[186]^63^,[187]^64 Unlike single-gene
   therapies, miRNA-based therapeutics offer a unique advantage due to
   their ability to simultaneously target multiple pathways (i.e., HR and
   NHEJ). However, despite promising preclinical results, the translation
   of miRNA therapeutics into clinical practice remains highly
   challenging.[188]^65

   One major hurdle of miRNA therapeutics is the difficulty in achieving
   efficient and specific delivery. miRNA mimics and inhibitors require
   stable delivery systems to prevent degradation and ensure targeted
   cellular uptake, yet existing strategies often lead to off-target
   effects, immune activation, or insufficient bioavailability. Early
   clinical trials, such as a phase 1 trial for MRX34 (a miR-34a mimic;
   this study was registered at [189]ClinicalTrials.gov:
   [190]NCT01829971),[191]^66 a phase 2 trial for lademirsen (a miR-21
   inhibitor; this study was registered at [192]ClinicalTrials.gov:
   [193]NCT02855268),[194]^67 and phase 2 trials for Miravirsen (a miR-122
   inhibitor; these studies were registered at [195]ClinicalTrials.gov:
   [196]NCT01872936 and [197]NCT02031133)[198]^68 encountered setbacks due
   to severe immune-mediated toxicities or limited efficacy, ultimately
   leading to trial discontinuation. These examples highlight the
   complexities of miRNA-based therapeutics, particularly regarding safety
   and specificity.

   Nanoparticle-based delivery systems have shown promise in overcoming
   some of these challenges by enhancing stability and targeting
   efficiency.[199]^64 In the context of cervical cancer, localized
   delivery approaches could mitigate systemic toxicity.[200]^69 However,
   while the anatomical accessibility of the cervix may facilitate
   localized administration, this does not inherently resolve broader
   delivery challenges, including endosomal sequestration and
   biodistribution concerns. Furthermore, tumor heterogeneity, hypoxic
   environments, and variable miRNA expression profiles within cervical
   cancer further complicate therapeutic efficacy and patient
   stratification.

   Despite these challenges, miRNA therapeutics continue to hold
   potential, particularly when integrated into multimodal treatment
   strategies. For example, in thermoradiotherapy, hyperthermia enhances
   tumor vasculature permeability, which could improve nanoparticle uptake
   and facilitate miRNA-based radiosensitization.[201]^70 However,
   preclinical and clinical studies are needed to determine the safety and
   efficacy of such approaches.

   In addition to their role as therapeutic agents, this study has
   identified several miRNAs that show potential as predictors of
   treatment response in cervical cancer patients. Specifically, miR-27a,
   miR-221, and miR-224 show significant downregulation in
   treatment-resistant patients, suggesting their utility as predictive
   markers of treatment response. In support of these findings, Ren and
   colleagues demonstrated that miR-27a plays a critical role in
   modulating the radiosensitivity of triple-negative breast cancer cells,
   suggesting that miR-27a expression could serve as a valuable marker to
   identify patients who will benefit from radiotherapy treatment.[202]^71
   In addition, Upraity et al. found that increased miR-224 expression
   increased the radiosensitivity of glioblastoma cells.[203]^72 Notably,
   TCGA cervical cancer dataset used in our study only includes patients
   treated with radiotherapy or chemoradiotherapy. While our study focuses
   on thermoradiotherapy, the potential biomarker utility of the
   identified miRNAs in the context of chemoradiotherapy underscores their
   broad applicability. This suggests that our findings could be
   extrapolated to other combined-modality treatments that include
   radiotherapy. Further validation studies in larger patient cohorts and
   in other cancer types are warranted to confirm the predictive value of
   miR-27a, miR-221, and miR-224 and to explore their applicability in
   clinical practice.

   In conclusion, this study highlights the potential role of miRNAs as
   sensitizers of thermoradiotherapy and improves our understanding of the
   molecular mechanisms underlying miRNA-mediated sensitization of
   thermoradiotherapy. Further research on the clinical translation of
   these findings, including prospective validation studies and
   mechanistic investigations, is warranted to fully exploit the
   therapeutic potential of miRNA-based approaches in combination with
   thermoradiotherapy for the treatment of locally advanced solid tumors.

Materials and methods

Cell lines and cell culture

   Human cervical cancer cell lines C-33A, C-4 I, CaSki, DoTc2 4510, HeLa,
   MS751, and SiHa were obtained from the American Type Culture Collection
   (ATCC, Manassas, VA) and cultured in Dulbecco’s modified Eagle’s medium
   (DMEM) formulated with high glucose and l-glutamine (Thermo Fisher
   Scientific, Waltham, MA) and supplemented with 10% fetal bovine serum
   (FBS), 100 U/mL penicillin, and 100 μg/mL streptomycin (all from Thermo
   Fisher Scientific). Human cervical cancer cell line HT-3 was obtained
   from the ATCC and cultured in McCoy’s 5a medium supplemented with 10%
   FBS, 100 U/mL penicillin, and 100 μg/mL streptomycin (all from Thermo
   Fisher Scientific). Two cervical cancer cell lines, 778 and 879, were
   previously provided by Dr. P.L. Stern (Paterson Institute for Cancer
   Research, Manchester, UK) and cultured in RPMI 1640 medium (Thermo
   Fisher Scientific) supplemented with 10% FBS, penicillin at 100 U/mL,
   and streptomycin at 100 μg/mL (all from Thermo Fisher
   Scientific).[204]^73 Head and neck squamous cell carcinoma (HNSCC) cell
   lines UM-SCC-47, VU-SCC-040, VU-SCC-120, and VU-SCC-147 were obtained
   as described previously and cultured with DMEM supplemented with 10%
   FBS, 1 mM sodium pyruvate, 100 U/mL penicillin, and 100 μg/mL
   streptomycin (all from Thermo Fisher
   Scientific).[205]^74^,[206]^75^,[207]^76 Cell cultures were maintained
   at 37°C in a humidified atmosphere with 5% CO[2]. Cell lines were
   authenticated by short-tandem repeat (STR) testing, monthly tested for
   the presence of mycoplasma and HPV status was confirmed using the
   QIAscreen HPV PCR assay (Qiagen, Hilden, Germany).

Irradiation and hyperthermia

   Cells were irradiated with 2 Gy of γ-radiation at room temperature
   using the Gammacell 220 Cobalt irradiator with a ^131Cs source (MDS
   Nordion, Ottawa, Canada). Hyperthermia treatment of cells started
   within 30 min after irradiation. Incubation of cells was performed by
   partially submerging the plates in a thermostatically controlled water
   bath (VWR International, Amsterdam, the Netherlands) for 60 min at 42°C
   (±0.1°C). The temperature was measured in parallel plates with a
   thermometer, and the desired temperature was reached in approximately
   8 min for 384-well plates and 5 min for 6-well plates.

High-throughput miRNA screens

   For the discovery screen, four cervical cancer cell lines (C-4 I,
   CaSki, HeLa, and SiHa) were subjected to high-throughput reverse
   transfection using 384-well ultra-low attachment plates (Corning, New
   York, NY) to allow for 3D cell cultures (spheroids). A total of 378
   miRNAs ([208]Table S1) were selected based on in-house and publicly
   available genome-wide miRNA expression data from cancer types relevant
   to our study (i.e., cervical cancer, head and neck cancers) as well as
   literature supporting their role as a radiosensitizer in
   cancer.[209]^22^,[210]^24^,[211]^40^,[212]^77^,[213]^78 Cells were
   transfected using an optimized reverse transfection protocol, in which
   efficiency was carefully assessed through miRNA qPCR, miRNA target
   western blot analysis, and cellular viability measurements, as
   described previously.[214]^23^,[215]^62 In brief, on day 0 transfection
   complexes were prepared by mixing 0.04 μL DharmaFECT4 transfection
   reagent (Horizon Discoveries, Cambridge, UK) and 2 nM (end
   concentration) of miRNA mimic (miRIDIAN miRNA mimic library version 19
   [Horizon Discoveries]) with 20 μL Opti-MEM serum-free medium (Thermo
   Fisher Scientific). Cells were harvested, resuspended in culture
   medium, and plated in a 384-well format (500 cells/20 μL/well)
   simultaneously with the addition of transfection complexes. Plates were
   sealed with breathable film to prevent evaporation. On day 1, cells
   were treated with thermoradiotherapy or kept in culture as a control.
   Cell viability was assessed on day 4 by adding 15 μL Cell-Titer Glo 3D
   (Promega, Madison, WI) per well, after which luminescence was measured
   using an Infinite F200 microplate reader (Tecan Group, Männedorf,
   Switzerland). The discovery screen was performed in triplicate.

   For the validation screen, 10 additional cancer cell lines were
   utilized, including 6 cervical cancer cell lines (778, 879, C-33A,
   DoTc2 4510, HT-3, and MS751) and 4 HNSCC cell lines (UM-SCC-47,
   VU-SCC-040, VU-SCC-120, and VU-SCC-147). From the discovery screen, 55
   candidate miRNAs along with 13 control miRNAs were selected for
   validation. The setup and readout of the validation screen was similar
   to that of the discovery screen, including the amount of cells,
   transfection reagent, and miRNA mimic concentration. The validation
   screen was performed in triplicate.

miRNA transfection for functional assays

   SiHa cells were reverse transfected with miRNA mimics for
   hsa-miR-16-5p, hsa-miR-27a-5p, hsa-miR-181c-5p, hsa-miR-221-3p,
   hsa-miR-224-3p, hsa-miR-330-5p, hsa-miR-1293, or NC #2 (C-300483-03,
   C-301028-01, C-300556-03, C-300578-05, C-301304-00, C-301082-01,
   C-301347-00, CN-002000-01 [Horizon Discoveries]) in ultra-low
   attachment plates following the same protocol outlined for the miRNA
   screens. In brief, transfection complexes were prepared by mixing 2 μL
   (end volume 0.04 μL) DharmaFECT4 transfection reagent (Horizon
   Discoveries) and 2 nM (end concentration) of miRNA or control mimic
   with 200 μL Opti-MEM serum-free medium (Thermo Fisher Scientific).
   Harvested cells were plated in 6-well ultra-low attachment plates
   (Corning) at a density of 85,000 cells/1.8 mL/well.

RNA isolation

   Total RNA was isolated from cells using the miRNeasy Tissue/Cells
   Advanced kit, including on-column DNase treatment, according to the
   manufacturer’s instructions (Qiagen). RNA concentrations were measured
   using NanoDrop One (Thermo Fisher Scientific).

RT-qPCR

   This study was performed in compliance with the minimum information for
   publication of qPCR experiments guidelines, where applicable.[216]^79
   Reverse transcription was performed using the TaqMan microRNA Reverse
   Transcription kit (4366596; Thermo Fisher Scientific). The
   manufacturer’s protocol was adapted to multiplex reverse transcription
   for all miRNAs and the reference gene, as described
   previously.[217]^77^,[218]^80 In brief, 50 ng total RNA was reverse
   transcribed in 16.5-μL reactions containing 9 μL primer pool, 0.45 μL
   deoxynucleotide triphosphate (100 mM), 2.25 μL 10× RT buffer, 0.3 μL
   RNase inhibitor (20 U/μL), and 4.5 μL MultiScribe Reverse
   Transcriptase. Expression of hsa-miR-16-5p, hsa-miR-27a-5p,
   hsa-miR-181c-5p, hsa-miR-221-3p, hsa-miR-224-3p, hsa-miR-330-5p, and
   hsa-miR-1293 was determined using TaqMan microRNA assays (000391,
   002445, 000482, 000542, 121210_mat, 002230, 002905; Thermo Fisher
   Scientific). U75 was included as a reference gene (001219; Thermo
   Fisher Scientific). qPCR was performed in 10-μL reactions containing
   5 μL TaqMan Universal Master Mix II, no UNG (4440043; Thermo Fisher
   Scientific), 0.5 μL miRNA-specific TaqMan microRNA assays (catalog
   numbers for each miRNA assay are listed above), 3.5 μL H[2]O, and 1 μL
   cDNA. qPCR reactions were performed on the ABI 7500 Fast Real-Time PCR
   System (4351107; Thermo Fisher Scientific) using the following
   conditions; 5 min at 95°C followed by 40 cycles of 15 s at 95°C and
   1 min at 60°C. H[2]O was tested as a negative PCR control. qPCR
   analysis and determination of Ct values were conducted using Applied
   Biosystem 7500 Software version 2.3 (4397808; Thermo Fisher
   Scientific). The miRNA expression levels were normalized to U75 by
   applying the 2^−ΔCt method.[219]^81

Clonogenic assay

   Clonogenic assays were performed as described
   previously.[220]^82^,[221]^83 Cells were collected 2 days post-miRNA
   transfection (to harvest enough cells for the subsequent protocol) and
   plated at different densities depending on the treatment group (50,
   100, 200, 500, 1,000 or 2,000 cells/6-well) 4–8 h prior to treatment.
   Subsequently, cells were treated with either irradiation (2 Gy) alone
   or thermoradiotherapy (2 Gy and 42°C for 60 min) and incubated for
   10–12 days to allow for colony formation. The plates were stained with
   a methanol-crystal violet solution (5 mg/mL crystal violet in 10%
   ethanol, 50% methanol, and 40% water) followed by scanning and counting
   with GelCount (Oxford Optronix, Abington, UK). Plating efficiencies
   were defined by dividing the amount of counted colonies by the amount
   of seeded cells. Survival fractions were calculated by normalization to
   the plating efficiency of untreated cells.

γ-H2AX immunofluorescence assay

   SiHa cells were reverse transfected in the 8-well Nunc Lab-Tek II
   Chamber Slide system (Thermo Fisher Scientific) by mixing 0.6 μL (end
   volume 0.04 μL) DharmaFECT4 transfection reagent (Horizon Discoveries)
   and 2 nM miRNA or control mimic with 80 μL opti-MEM serum-free medium
   (Thermo Fisher Scientific). Harvested cells were seeded at 9,000
   cells/520 μL/well. Subsequently, cells were treated with ionizing
   radiation (2 Gy) and hyperthermia (42°C for 60 min) and 24 h
   post-treatment the cells were fixed with 2% paraformaldehyde and
   permeabilized with 0.1% Triton X-100 in phosphate-buffered saline
   (PBS). Slides were then incubated with 1:200 Alexa Fluor 488 anti-H2A.X
   Phospho (Ser139) antibody (catalog no. 613405; BioLegend, San Diego,
   CA) and counterstained with 1 μg/mL DAPI. Slides were then visualized
   with a Nikon A1R HD25 confocal microscope (Nikon Europe, Amstelveen,
   the Netherlands). H2AX foci numbers were counted manually and at least
   70 cells per condition per experiment were scored in three independent
   experiments.

Automated western blot

   SiHa cells were collected 24 h post-miRNA transfection and plated in
   6-well plates at a density of 85,000 cells/well 4–8 h prior to
   treatment. Cells were treated with irradiation (2 Gy) and hyperthermia
   (42°C for 60 min), and 4 h post-treatment cells were harvested.
   Whole-cell lysates were prepared on ice using the RIPA Lysis Buffer
   System (Santa Cruz Biotechnology, Dallas, TX), according to the
   manufacturer’s instructions. Protein concentrations were determined
   using Pierce bicinchoninic acid protein assay following the
   manufacturer’s instructions. The automated western blots were conducted
   using JESS Simple Western instrument (ProteinSimple, Bio-Techne,
   Minneapolis, MN). Proteins (3 μg) were diluted in 0.1× sample buffer
   and 1× fluorescent master mix (EZ standard pack; Bio-Techne) to a total
   volume of 6 μL/well and separated using a 12- to 230-kDa or a 66- to
   440-kDa separation module (Bio-Techne). Primary antibodies were diluted
   in antibody diluent 2 (Bio-Techne) and included rabbit anti-ATM (1:50,
   catalog no. 2873; Cell Signaling Technologies, Danvers, MA), mouse
   anti-β-actin (1:12.5, catalog no. 3700s; Cell Signaling Technologies),
   rabbit anti-DNA-PKcs (1:100, bsm-52493R, Bioss, Woburn, MA), rabbit
   anti-Ku70 (1:500, catalog no. PA5-27538, Thermo Fisher Scientific),
   rabbit anti-Ku70/Ku80 (1:500, catalog no. PA-5-27538; Thermo Fisher
   Scientific), and mouse anti-RAD51 (1:50, catalog no. MS-988-P1; Thermo
   Fisher Scientific). Secondary antibody detection modules anti-mouse and
   anti-rabbit were used according to the manufacturer’s instructions. The
   RePlex module (Bio-Techne) was used to allow sequential detection of
   multiple proteins. Data were analyzed using the software Compass for
   Simple Western version 6.3.

miRNA target prediction

   Target prediction of selected miRNAs from the validation screen (n =
   18) was performed using the ENCORI platform version 2.0[222]^27 or
   miRDB ([223]https://mirdb.org).[224]^84 The ENCORI platform was
   employed for 13 miRNAs (miR-16-5p, miR-106b-5p, miR-181a-5p,
   miR-181c-5p, miR-181d-5p, miR-193a-5p, miR-195-5p, miR-221-3p,
   miR-224-3p, miR-330-5p, miR-375, miR-455-5p, and miR-497-5p). No
   predicted targets were identified for 5 miRNAs (miR-22-5p, miR-27a-5p,
   miR-92a-1-5p, miR-1293, and miR-3158-3p) using the ENCORI platform,
   leading us to utilize miRDB for target prediction. Targets predicted by
   ENCORI with a HITS-CLIP number ≥4 and targets by miRDB with a
   prediction score ≥60 were subjected to pathway analysis using DAVID
   version 2024q2,[225]^28 with WikiPathways version 20241110 as
   reference. Pathways associated with the DNA damage response were
   selected for further investigation.

External expression data of cervical tissue specimens

   Genome-wide miRNA and mRNA expression profiles along with clinical data
   from cervical cancer patients[226]^24 and head and neck cancer
   patients[227]^40 was obtained from TCGA. Pearson correlation
   coefficients (r) were computed between miRNAs and their predicted miRNA
   targets and visualized in a correlation heatmap. Patients with advanced
   cervical cancer were classified into two groups according to their
   response after chemoradiotherapy: treatment sensitive (n = 37),
   including complete and partial response, or treatment resistant (n =
   18), including stable and progressive disease.

Statistical analysis

Analysis of high-throughput screens

   As the number of cell lines was relatively small in the discovery
   screen, we opted for a simple analysis model per miRNA: a linear mixed
   effect model for the log[2](response), with thermoradiotherapy
   treatment and cell lines as fixed factors and assay (biological
   replicates) as a random effect.[228]^85 miRNA-specific treatment
   effects were contrasted with the modal effect (across all miRNAs) to
   filter out effects that are mostly due to the treatment alone. Hence,
   resulting miRNAs have, in conjunction with the treatment, a
   significantly larger effect than the modal effect. An FDR of ≤0.10 was
   used as threshold for significance. The discovery screen data was
   analyzed using the R-package ShrinkBayes version 2.13.7, which enables
   efficient estimation for large screens and small sample sizes.[229]^85
   From the results of the discovery screen, 13 miRNAs were designated as
   control miRNAs based on their relative effect size (around 0.0;
   [230]Table S2).

   For the validation screen, the utilization of a larger number of cell
   lines allowed for a slightly more complex model, also to disentangle
   the treatment and miRNA effect. For each miRNA, a linear mixed-effects
   model was used on the log[2](response) using miRNA presence (yes/no),
   thermoradiotherapy treatment (yes/no), cancer type (cervical
   cancer/HNSCC), and HPV status (negative/positive) as fixed effects (the
   latter two as potential confounders), plus cell line, assay (biological
   replicates nested within cell line), and pair (nested within assay) as
   random effects. The first accounts for variation across cell lines (now
   modeled as random effects as we have 11 levels), the second for
   correlation between technical repeats of screens, and the last accounts
   for correlation between responses of the parental (no miRNA) and
   non-parental settings within one screen. This model was fit for both
   control miRNAs and the candidate miRNAs were corrected for the mean
   effect of the control miRNAs (by subtracting the latter from the
   responses of the candidate miRNAs) to ensure that selected miRNAs have
   significantly stronger effects than the control miRNAs. Significance of
   the miRNA was tested using a likelihood ratio test (chi-squared; df =
   1) of the model against the null model, which contains all factors
   except for the miRNA presence covariate. Likewise, treatment-miRNA
   interactions were tested by adding such an interaction to the model and
   testing it against the model with main effects only. Resulting p values
   were Bonferroni corrected for multiple testing, and effect size
   estimates were back-transformed to the original scale, rendering fold
   changes. The R-package lme4 version 1.1–35 was used to fit the mixed
   models.[231]^86

Analysis of functional assays

   All data are expressed as mean ± SD. The statistical significance of
   clonogenic assays, γ-H2AX immunofluorescence assays, and immunoblotting
   was tested using the Kruskal-Wallis test with the post hoc Dunn’s test
   for comparison to NC within the same treatment group (untreated, 2 Gy
   or 2 Gy+HT). Comparison of miRNA expression profiles between
   radio-sensitive and radio-resistant groups was made using the
   Mann-Whitney test (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001;
   ∗∗∗∗p < 0.0001). Correlation between miRNA and mRNA expression was
   tested using Spearman correlation (∗p < 0.05; ∗∗p < 0.01). Analyses of
   functional experiments were performed with GraphPad Prism version 10
   (GraphPad Software, La Jolla, CA). Statistical analyses of publicly
   available data were performed with R (version 4.2.1,
   [232]https://www.R-project.org/). Data processing and figures were
   generated using the readr (version 2.1.5), dplyr (version 1.1.4) and
   ggplot2 (version 3.4.4) packages. Study outline and network model were
   illustrated using BioRender ([233]https://www.biorender.com/) and Adobe
   Illustrator 2023.

Data availability

   mRNA and miRNA expressions and treatment response information are
   available from TCGA dataset. All other data supporting the findings of
   this study are available from the corresponding author upon reasonable
   request.

Acknowledgments