Chemoradiation Reprograms Tumor Cells and the Immune Microenvironment in Cervical Cancer

Abstract

Despite advances in screening and prevention, cervical cancer remains a leading cause of cancer-related deaths worldwide, underscoring the need for better treatments. Here, we conducted a multi-cohort longitudinal study of human cervical tumors and the tumor microenvironment during chemoradiation therapy (CRT) and integrated RNA sequencing and single-cell transcriptomics to define the cellular and molecular programs shaping cell interactions and how CRT alters them. The analysis identified multiple therapeutic targets in CRT-resistant tumors, notably including MDM2, a key mediator of radiation responses in tumor and immune cells. MDM2 inhibition enhanced the effects of radiotherapy in HPV-positive, TP53 wild-type cervical cancer cells, improved radiation response, and reshaped the immune landscape in preclinical models. These findings highlight the potential of combining MDM2 inhibition with CRT to overcome resistance and improve patient outcomes. The insights into therapy-induced changes in tumor and immune compartments could guide improved strategies against treatment-resistant HPV-positive cancers.

Significance

Mapping of the impact of chemoradiation on cellular interactions in cervical cancer reveals how treatment reshapes the tumor microenvironment and highlights targets for developing future immunotherapeutic approaches.

INTRODUCTION

Cervical cancer remains one of the leading causes of cancer-related deaths among women worldwide, with over 662,301 new cases and 348,874 deaths reported in 2022¹. High-risk subtypes of the human papillomavirus (HPV) are the primary drivers of the disease². Tumors are genetically heterogenous, with common oncogenic genetic aberrations each occurring in only a relatively small proportion of tumors. Unlike many solid tumors that harbor common oncogenic mutations offering clear molecular targets, cervical cancer is primarily driven by HPV infection³. The viral oncoproteins E6 and E7 disrupt critical tumor suppressor pathways by inactivating p53 and Rb, respectively⁴. This unique etiology results in a distinct molecular landscape and poses challenges for targeted therapy, necessitating alternative approaches that differ from those used in other solid tumors.

Although great strides have been made in prevention strategies and early diagnosis, many patients continue to present with advanced stages of cervical cancer, where definitive chemoradiation therapy (CRT) has been the standard-of-care for decades⁵. While advancements in radiotherapy technology have improved local disease control, relapse remains a significant challenge, with one-third of patients with locally advanced disease experiencing recurrence within 2-3 years post-treatment^6–8. Anti-PD1 immune checkpoint inhibitor (ICI) therapy, designed to activate the immune system, was recently shown to improve recurrence free and overall survival in combination with CRT, not only in cervical cancer but in other cancer types. However, treatment with ICI is not without toxicity and many patients do not benefit from the combination of ICI + RT. Biomarkers predictive of the benefit of ICI have not yet been identified, and concurrent trials testing anti-PDL1 therapy with CRT in cervical and head and neck cancers failed to demonstrate similar benefits^9,10. This underscores a critical unmet need for novel strategies that can enhance or complement CRT and, in addition, biomarkers to better select patients for ICI or other immunotherapies to improve long term patient outcomes.

The tumor microenvironment (TME) in cervical cancer comprises a dynamic and complex interplay among tumor, stromal and immune cells, and the nature of these interactions significantly influences the progression of the disease and therapeutic resistance¹¹. Advancements in single-cell RNA sequencing (scRNA-seq) have facilitated a deeper understanding of cellular composition and intercellular communications within complex tumor milieus, as demonstrated in pretreatment studies of pancreatic¹², breast¹³ and ovarian¹⁴ tumors. While pretreatment scRNA-seq analyses have been conducted in cervical cancer^15–17, the impact of therapeutic interventions, such as chemoradiotherapy, on the cervix TME has not been comprehensively examined using scRNAseq and longitudinally collected tumor samples which are difficult to obtain. Given recent conflicting clinical evidence regarding the benefit of the addition of immune checkpoint blockade to chemoradiation, longitudinal study of patient materials collected during therapy has the potential to identify important new mechanisms and biomarkers that can be used to improve patient outcomes.

In this study, we analyzed the TME of human cervical cancer using two independent longitudinally sampled patient cohorts. Our results examined the impact of standard-of-care CRT on the tumor and the immune cell compartments at the single cell level and revealed intricate novel cell-to-cell communication networks. Notably, we identified MDM2 as a potential therapeutic target induced by CRT in tumor cells. Utilizing two clinically relevant mouse tumor models of HPV-positive cervical cancer, including novel CRT-naïve and CRT-edited patient derived xenografts (PDXs), we demonstrated the feasibility of pharmacologically targeting MDM2 together with radiation therapy as a potential means to improve therapeutic response during CRT and in the setting of CRT resistant tumors. This study provides a comprehensive, longitudinal analysis of CRT’s impact across distinct compartments of the cervical cancer tumor microenvironment, uncovering novel CRT-induced targets that may boost conventional therapies and enhance emerging immunotherapeutic combinations.

MATERIALS AND METHODS

Patient cohorts and clinical data annotation

Studies were conducted in accordance with the Declaration of Helsinki and approved by the Washington University Institutional Review Board.

Patients in Cohorts 1 and 2 were prospectively enrolled in an institutional longitudinal tumor-banking protocol (IRB #201105374) and provided written informed consent. Eligible participants had FIGO 2018 stage I-IVB cervical cancers of any histology and were ≥ 18 years old (32-79 years; Table 1). Institutional radiation-therapy practices have been described previously¹⁸. All patients received pelvic EBRT to 50.4 Gy in 28 fractions with concurrent interdigitated brachytherapy boost. Total tumor dose and EQD2 calculations were derived from accumulated EBRT and brachytherapy plans using institutional software. An extended version of this section can be found in the Extended Materials and Methods file.

Table 1:

Demographic and clinical characteristics of patients in the bulk-sequenced (Cohort 1, n = 15) and single cell-sequenced (Cohort 2, n = 10) cohorts. Age is reported as the mean with range, while categorical variables (Race, FIGO 2018 Stage, Node Positive, Histology, Therapy, and Status) are presented as numbers with corresponding percentages.

	Cohort 1		Cohort 2
Enrollment period	2016-2018		2019-2021
Analytical Platform	Bulk RNA sequence		Single-cell RNA sequence
Number of patients	15		10
Age (mean, range)	55.7 years, 32-79 years		48.1 years, 32-77 years
Race
Black	4 (26.7%)		0 (0%)
Caucasian	11 (73.3%)		10 (100%)
FIGO 2018 Stage
I	3 (20%)		2 (20%)
II	2 (13.3%)		3 (30%)
III	10 (66.7%)		3 (30%)
IV	0		2 (20%)
Node Positive *
Yes	11 (73.3%)		4 (40%)
No	4 (26.7%)		6 (60%)
Histology
Adenocarcinoma	3 (20%)		2 (20%)
Clear Cell	0 (0%)		1 (10%)
Poorly Differentiated	1 (6.7%)		0 (0%)
Squamous	11 (73.3%)		7 (70%)
Therapy
Curative	15 (100%)		8 (80%)
Palliative	0 (0%)		2 (20%)
Status
AWD	1 (6.7%)		5 (50%)
DOD	4 (26.7%)		1 (10%)
NED	10 (66.7%)		4 (40%)
Time points	Pre-RT	On-RT	Pre-RT	On-RT#1	On-RT#2
Number of samples	15	15	10	4	4

Abbreviations: FIGO, International Federation of Gynecology and Obstetrics; AWD, Alive with Disease; DOD, Died of Disease; NED, No Evidence of Disease.

^*Node Positive indicates the presence of lymph node involvement.

Sample processing for genomic DNA and total RNA extraction

Tumor tissues were obtained, and after a piece was removed for fresh single-cell preparation the remaining sample was snap-frozen in liquid nitrogen and stored at −80 °C. Before bulk extraction, samples were cryo-pulverized (Covaris LE220, RRID:SCR_026895) and aliquoted. Genomic DNA was extracted from tumor tissue using either the DNeasy Blood and Tissue Kit (Qiagen, 69504) or the QIAamp DNA Mini Kit (Qiagen, 51304). Total RNA was isolated with TRI Reagent (Millipore Sigma, T9424), treated with DNase I (Qiagen, 79254), and purified using the RNeasy MinElute Cleanup Kit (Qiagen, 74204). RNA integrity was assessed with a Bioanalyzer or TapeStation (Agilent Technologies). Germline DNA was purified from cryopreserved PBMCs with the QIAamp DNA Mini Kit (Qiagen, 51304) and quantified using the Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific, Q32854).

Whole exome sequencing

Genomic DNA (100-250 ng) was fragmented on a Covaris LE220 (RRID:SCR_026895) targeting 250-bp inserts. Dual-indexed libraries were prepared using the KAPA Hyper Library Prep Kit (Roche) on the SciClone NGS platform (Perkin Elmer). Up to ten equimolar libraries (5 μg total) were pooled for hybrid capture with the xGen Exome Research Panel v1.0 (IDT Technologies, RRID:SCR_025813), covering a 39-Mb target region (19,396 genes). Hybridization was performed for 16-18 h at 65 °C, followed by stringent washes and PCR cycle optimization to avoid over-amplification. Enriched libraries were amplified using KAPA HiFi Master Mix (Roche). Library concentrations were quantified by qPCR with the KAPA Library Quantification Kit (Roche) to ensure optimal cluster density for sequencing on an Illumina NovaSeq 6000 (RRID:SCR_016387). Paired-end 2 × 150 bp reads were generated, targeting ~12 Gb per library to achieve ~100× coverage.

RNA sequencing

RNA integrity was assessed using the Agilent Bioanalyzer 4200 TapeStation (RRID:SCR_018435). Libraries were prepared from 500 ng-1 μg of total RNA. Ribosomal RNA was depleted with FastSelect reagents (Qiagen) during cDNA synthesis. mRNA was fragmented and reverse-transcribed with SuperScript III Reverse Transcriptase (Life Technologies) and random hexamers, followed by second-strand synthesis to generate double-stranded cDNA. A second strand reaction was performed to yield double-stranded cDNA (ds-cDNA). Resulting cDNA fragments were end-repaired, A-tailed, ligated to Illumina adapters, and amplified for 15 cycles with primers containing unique dual-index tags. Sequencing was performed on an Illumina NovaSeq 6000 S4 (RRID:SCR_016387), generating ~30 million paired-end 2 × 150 bp reads per library.

DNA and RNA sample quality control

Bulk sequencing data quality metrics (adaptor content, mapping quality, coverage and swaps/mislabeling) were determined for DNA and RNA bams using our in-house pipeline SeqQEst. The inclusion criteria for paired DNA and RNA bams with sufficient coverage was >50× coding region coverage in WES or >50 Mb mapped depth in RNA-seq data.

Single-cell suspension preparation

Fresh tumor samples were transported directly for processing. Approximately 10-20 mg of tissue per tumor were minced and enzymatically dissociated using the Human Tumor Dissociation Kit (Miltenyi Biotec, 130-095-929) in DMEM. The suspension was transferred to gentleMACS C-tubes (Miltenyi Biotec, 130-093-237) and processed on the gentleMACS Octo Dissociator with Heaters (Miltenyi Biotec, 130-096-427; RRID:SCR_025922) using the 37h_TDK_1 program for 30-60 min. Dissociated cells were filtered through 40-μm Mini-Strainers (PluriSelect, 43-10040-60), centrifuged at 400 g for 5 min at 4 °C, and subjected to ACK lysis (Thermo Fisher, A1049201) when erythrocyte contamination was evident. Cells were washed, counted, and resuspended in PBS with 0.5% BSA for Chromium loading. An extended version of this section can be found in the Extended Materials and Methods file.

10x library preparation and sequencing of scRNA-seq

Single-cell libraries were prepared using the Chromium Next GEM Single Cell 3′ GEM, Library & Gel Bead Kit v3.1 (10x Genomics, 1000269; RRID:SCR_024537) and the Chromium Controller. Approximately 17,500-25,000 cells were partitioned into nanoliter droplets, capturing 10,000-15,000 cells per sample. During reverse transcription, cDNA molecules were barcoded with 16-nt cell identifiers and 10-nt unique molecular identifiers (UMIs). Full-length poly-A cDNA was fragmented and size-selected (~400 bp) for library construction. Library concentrations were determined by qPCR (Kapa Biosystems) to ensure optimal clustering on the Illumina HiSeq 4000 (RRID:SCR_016386) or NovaSeq 6000 (RRID:SCR_016387). Sequencing generated 26 × 98 bp paired-end reads, targeting ~50,000 read pairs per cell to produce single-cell expression profiles.

Genome alignment and genomic analyses

All WES data first underwent adaptor trimming with Trim Galore 0.6.4. The trimmed fastq data were then aligned to human reference genomes (GRCh38), sorted, and deduplicated. The resulting bam format output was used for downstream analysis. Ancestry prediction, Somatic mutation calling and mutational signature analysis were performed, see details in the extended version of the Materials and Methods.

RNA quantification

RNA quantification was done using our internal pipeline (https://github.com/ding-lab/cptac_rna_expression). Briefly, raw fastq files were aligned to the human reference genome GRCh38 using STAR (2.7.6a); raw read count was then quantified by featureCounts (1.6.4). Gencode v22 was used for gene annotation. After obtaining raw read count, Fragments Per Kilobase of transcript per Million mapped reads (FPKM) and FPKM Upper Quartile (FPKM-UQ) values were calculated.

Differential expression analysis

Gene expression was used to perform pairwise differential analysis between groups of samples. A Wilcoxon rank-sum test was performed to determine the differential abundance of gene expression. At least four samples in both groups were required to have non-missing values, and the p-value was adjusted using the Benjamini-Hochberg procedure, and features were considered significant with an adjusted p-value < 0.05, and the log2-scale fold change greater than 0.8.

scRNA-seq Data Processing and Analysis

Sequencing reads were aligned to the GRCh38 reference genome using Cell Ranger (v3.1.0, RRID:SCR_017344), and downstream processing was performed in Seurat (v3.1.2, RRID:SCR_016341). After stringent quality filtering, normalization, clustering, and cell-type annotation based on canonical markers, differential expression analyses were conducted to compare treatment groups. Pathway and enrichment analyses were performed using ssGSEA, ConsensusPathDB, and Ingenuity Pathway Analysis (RRID:SCR_008653). Cell-cell communication networks were inferred with CellChat (RRID:SCR_021946), and changes in cellular composition were assessed using miloR (RRID:SCR_025630). HPV gene expression was evaluated across cell types and time points using HPV-EM. An extended version of this analysis can be found in the Extended Materials and Methods file.

Cell culture and reagents

Cervical cancer cell lines, SiHa (ATCC Cat# HTB-35, RRID:CVCL_0032), CaSki (ATCC Cat# CRL-1550, RRID:CVCL_1100) and C33A (ATCC Cat# HTB-31, RRID:CVCL_1094) were obtained from the American Type Culture Collection (ATCC) and maintained in IMDM media (Life Technologies) with 10% heat-inactivated FBS. Experiments were performed on cell lines under passage 30. Short Tandem Repeat (STR) profiling was performed in February 2025 (for TC-1) and September 2023 (For SiHa, CaSki and C33a), which confirmed positive match and full cell line authentication per ATCC reference standards. Mycoplasma testing was performed every 3 months to verify no contamination. Protease inhibitor, phosphatase inhibitor cocktails, were purchased from Sigma. AMG-232 was purchased from MedChemExpress (RRID:SCR_025062).

Clonogenic survival assays

Cells were plated at a density of 0.05 × 10⁶ cells/well in 24-well plates and allowed to adhere. After incubation for 24 hours, cells were pretreated with vehicle control or AMG-232 at the indicated concentrations for 48 hours, then irradiated using an RS2000 160kV X-ray Irradiator using a 0.3 mm copper filter (Rad Source Technologies). Cells were harvested 48 hours after treatment by trypsinization including all floating cells, and 500 cells each were plated in a 6-well dish. The colonies were counted 10 days after staining with crystal violet. Clonogenic survival data were fitted using the Linear-Quadratic formula.

Western blotting

Cells or tumor lysates were prepared in ice-cold lysis buffer (Cell Signaling Technology) supplemented with 1 mM PMSF and protease inhibitors, sonicated, and centrifuged at 12,000 rpm for 30 min. Equal protein amounts were resolved on 4–20% SDS-PAGE gels and transferred to PVDF membranes (Bio-Rad). After blocking with 5% nonfat milk, membranes were probed with antibodies against p53 (Santa Cruz Biotechnology, sc-126; RRID:AB_628082), p21 (sc-471; RRID:AB_632123), MDM2 (sc-965; RRID:AB_627920), and β-actin (sc-58673; RRID:AB_2223345). HRP-conjugated secondary antibodies (Santa Cruz Biotechnology) were used for detection with Amersham ECL Select (GE Healthcare). Images were captured on a ChemiDoc Imaging System (Bio-Rad; RRID:SCR_019037).

RNA isolation and Real-time Quantitative RT-PCR

Total RNA was isolated from cells using Trizol reagent (Life Technologies) as previously described¹⁹. Complementary DNA (cDNA) was synthesized using random hexamers and a Superscript III reverse transcriptase kit (Invitrogen, CA). Real-time PCR was performed with complementary DNA for quantitation using the TaqMan gene expression PCR master mix and 6-carboxyfluorescein-minor groove binder probes (FAM). Reactions were performed at 50°C for 2 minutes and 95°C for 10 minutes, cycled at 95°C for 15 seconds and 60⁸C for 1 minute for 40 cycles, and held at 25°C for 2 minutes in a 7500 real-time PCR system (Applied Biosystems, RRID:SCR_018051). Relative gene expression levels were normalized to the GAPDH level using the formula 2^−ΔΔCT (DDCT 5 DCT sample 2 DCT untreated control). Reactions were performed in triplicate.

Cell Death Assays

Cells were trypsinized and washed three times with ice-cold PBS prior to staining with annexin V-PI according to the manufacturer’s protocol (BD Pharmingen). Flow cytometry was performed using a FACS flow cytometer (Miltenyi Biotec, RRID:SCR_008984).

In vivo experiments

All animal studies were approved by the Washington University Division of Comparative Medicine and Institutional Animal Care and Use Committee. TC-1 cells (HPV-E6/E7-transformed mouse lung epithelial cells; gift from Dr. TC Wu, Johns Hopkins University) were implanted subcutaneously into the left flank of 6-8-week-old female C57BL/6J mice. Tumor-bearing mice were treated with combinations of radiotherapy (4 Gy), cisplatin (0.5 mg/kg, intraperitoneal), and the MDM2 inhibitor AMG-232 (25 mg/kg, oral gavage) to assess therapeutic interactions. Image-guided radiotherapy was delivered using the Xstrahl Small Animal Radiation Research Platform (SARRP-200, RRID:SCR_026901). Tumor growth was monitored by caliper measurement, and excised tumors were analyzed by flow cytometry (LSRFortessa X-20, BD Biosciences; RRID:SCR_025285). To validate findings in a patient-derived xenograft (PDX) model, experiments were conducted in collaboration with the NCI PDXNet Consortium using tumors derived from patient 323 collected before and after radiotherapy. NRG mice bearing pre-RT and on-RT PDX tumors were treated with vehicle, AMG-232 (15 mg/kg), radiotherapy (4 Gy), or the combination, and tumor response and p53/MDM2 expression were evaluated. An extended version of this section can be found in the Extended Materials and Methods file.

Flow Cytometry

Analyses were conducted using fluorochrome-conjugated antibodies purchased from BioLegend, unless stated otherwise. Cells were washed with PBS, Fc-gamma receptor-blocked using TruStain FcX (anti-mouse CD16/32, 93, RRID:AB_2783137), and then stained for and stained LIVE/DEAD Fixable blue Dead Cell Stain (Invitrogen) surface markers at 4°C in the dark for 30 minutes using the following antibodies: anti-CD45 (30-F11, 1:200, BioLegend Cat# 103128, RRID:AB_493715), anti-CD3 (17A2, 1:200, BioLegend Cat# 100222, RRID:AB_2242784), anti-CD4 (RM4-5, 1:200, BioLegend Cat# 100548, RRID:AB_2563054), anti-CD8α (53-6.7, 1:200, BioLegend Cat# 104012, RRID: AB_3695728), anti-CD11c (N418, 1:200, BioLegend Cat# 117310, RRID:AB_313779), anti-I-A/I-E (M5/114.15.2, 1:400; Tonbo Biosciences, Tonbo Biosciences Cat# 35-5321, RRID:AB_2621715), anti-CD11b (M1/70, 1:200, BioLegend Cat# 101259, RRID:AB_2566568), anti-F4/80 (BM8, 1:200, BioLegend Cat# 123110, RRID:AB_893486), anti-Ly6g (1A81 1:200, BioLegend Cat# 127633, RRID:AB_2562937), anti-Ly6c (HK1.4, 1:200, BioLegend Cat# 128041, RRID:AB_2565852), anti-NK1.1 (PK136, 1:200, BioLegend Cat# 108745, RRID:AB_2563286), anti-CD19 (6D5, 1:200, BioLegend Cat# 115554, RRID:AB_2564001). Cells were then fix/perm and washed. Flow cytometry was performed on a LSRII or a LSRFortessa X-20 instruments (BD Biosciences) and flow cytometry data were analyzed using FlowJo v.10 (TreeStar).

RESULTS

Comprehensive molecular and cellular characterization of cervical cancer highlights the impact of HPV genotypes

Tumor biopsies were collected from cervical cancer patients undergoing definitive chemoradiation treatment at the Siteman Cancer Center, as part of an Institutional Review Board (IRB)-approved prospective tumor banking study. This investigation encompassed two distinct chronological cohorts, although the samples were analyzed using the molecular techniques available at the time of processing, the enrollment criteria and treatment approaches were consistent across both cohorts. In Cohort 1, we performed whole-exome sequencing (WES) and bulk RNA sequencing of longitudinally collected cervical tumor biopsies from 15 patients treated with definitive chemoradiation (Fig. 1A, Fig. S1A and Table 1). In Cohort 2, samples from 10 patients were obtained before chemoradiation (pre-RT), and at 1-2 weeks (on-RT#1) and 2-3 weeks (on-RT#2) after the initiation of chemoradiation. We applied scRNA-seq to these samples to elucidate gene expression changes within tumor cells, characterize tumor microenvironment interactions, and identify activated gene programs at the single-cell level (Fig. 1A and Table 1).

Figure 1:

Comprehensive molecular and cellular profiling of cervical cancer. (A) Study design. Created in BioRender. Sandoval, T. (2025) https://BioRender.com/esr0ns9 (B) Description of patients in Cohort 1, ancestry, HPV status, time points collected. (C-D) WES analysis of Cohort 1 pretreatment tumor biopsies: (C) Analysis of mutations of genes commonly altered in cervical cancer, number of samples with alterations (on top), mutation types and HPV status (D) Single-base substitution mutational signatures (SBS) associated with HPV-driven mutagenesis (E) Sample description in Cohort 2, case identification, HPV status, time points collected; singlets are samples with only one time of collection, triplets are samples collected at the three time points. (F) scRNA-seq dimensionality reduction UMAP for samples in Cohort 2 showing tumor, immune, and stroma clusters. (G) HPV Enrichment Mapping-based transcriptomic profiling, identifying HPV16, HPV18, and HPV45 expression. (H) Tumor cell sub-clustering colored by case identification (Left) and histology (Right) (I) Enrichment map network of pre-ranked GSEA pathways derived from tumor-cell DEGs. Tumor-cell DEGs were ranked by log₂ fold change and subjected to Gene Set Enrichment Analysis (GSEA), and significantly enriched pathways (FDR < 0.25) were visualized in Cytoscape using the EnrichmentMap plugin. In the network, each node represents a pathway (node size ∝ gene-set size; node color ∝ normalized enrichment score), and edges connect pathways sharing common genes, See Table S1. (J) CD274 (PD-L1) expression and percentage of CD274-expressing cells.

Within Cohort 1, ancestry prediction indicated that 10 patients were of European ancestry (EUR), 4 of African ancestry (AFR), and 1 of admixed American ancestry (AMR) (Fig. 1B). Clinical HPV testing in 12 patients revealed HPV16 positivity (N=5), HPV18/45 positivity (N=3), and clinically HPV negative for HPV 16/18/45 (N=4) (Fig. 1B). Further HPV genotyping using HPV-EM algorithm²⁰ identified additional HPV genotypes beyond HPV16/18/45 in 4 cases initially reported as HPV-negative (Fig. 1B).

Whole exome sequencing of pretreatment tumor biopsies in Cohort 1 showed mutational patterns consistent with previous reports in cervical cancer. Notably, PTEN (33%), FAT1 (27%), NOTCH1 (27%), ARID1A (27%), KMT2D (20%), PIK3CA (13%), and EP300 (20%) were among the most frequently mutated genes. The mean number of somatic mutations per sample was 235, and the tumor mutational burden (TMB) closely aligned with published cervical cancer cohorts²¹. Mutational signature analyses revealed base-pair alterations associated with HPV-driven mutagenesis (e.g., APOBEC) and DNA mismatch repair deficiencies (Fig. 1C–D). These results highlight the significant role of HPV-mutagenic processes in shaping the genomic landscape of cervical cancer.

In Cohort 2 (Table 1), clinical HPV testing classified samples as HPV16-positive (N=2), HPV18/45-positive (N=5), unspecified HPV genotype (N=1), and HPV-negative (N=2). Subsequent HPV-EM profiling provided enhanced specificity, confirming HPV45 as a single genotype in two of the HPV18/45-positive cases and verifying the absence of HPV-related gene expression in the 2 HPV-negative samples (Fig. 1E). These results highlight the heterogeneity of HPV genotypes detected in our patient population.

To investigate the cellular composition in Cohort 2, we performed scRNA-seq, yielding a total of 80,828 cells after quality control. These cells were grouped into three major compartments: 34,484 tumor cells, 38,392 leukocytes, and 7,952 stromal cells, which were mainly comprised of cancer-associated fibroblasts (CAFs) and endothelial cells (Fig. 1F and Fig. S1B). HPV EM-based transcriptomic profiling identified HPV16, HPV18, and HPV45 transcripts within distinct tumor cell clusters (Fig. 1G). Notably, when we focused exclusively on the tumor cell fraction, clustering reflected patient-specific profiles rather than conventional histopathological features (Fig. 1H). Further examination of HPV early gene expression revealed that both the presence and the specific subtype of HPV correlate with distinct tumor-specific gene expression patterns (Fig. S1C).

Pathway enrichment analyses of differentially expressed genes (DEGs) upregulated in tumor cells relative to all other cell types, highlighted key biological processes associated with aggressive tumor behavior. These included elevated metabolic activity (“electron triphosphate transport,” 40 nodes), robust cell division (“anaphase mitotic sister,” 34 nodes and “chromosome mitotic alignment,” 10 nodes), and active DNA damage response pathways (“multiple patch ATR,” 10 nodes). We also observed upregulation of pathways involved in nucleotide and amino acid biosynthesis (“purine monophosphate process,” 7 nodes and “acid amino aspartate,” 6 nodes), highlighting the high proliferative capacity and metabolic demands of tumor cells (Fig. 1I and Table S1). Taken together, our findings highlight the diversity of HPV genotypes expressed in cervical cancer and reveal a complex genomic and molecular landscape of the disease prior to therapeutic intervention, highlighting the potential for more personalized treatment strategies.

Importantly, our results also highlight low pretreatment expression of PD-L1 in cervix tumor cells and the importance of multiple potential alternative mechanisms cervix tumor cells utilize to support rapid growth and evade immune surveillance prior to treatment. Specifically, expression of the immune checkpoint gene PD-L1 (CD274) was consistently low pretreatment across all tumor cell clusters, with only 0-5% of tumor cells expressing PD-L1 at this time point (Fig. 1J). In contrast, pretreatment tumor cells exhibited significant upregulation of multiple alternative and targetable proliferation- and immune-evasion-related genes, including CTNNB1, RELA, STAT3, CD47, TAP1, and TAP2. At the pretreatment time point, immune cells showed elevated expression of JAK1, JAK2, CD274, TGFB1, HLA-A, HLA-C, NLRC5, and TAPBP. In contrast, stromal cells displayed minimal expression of immune related genes (Fig. S1D).

Pretreatment immunological landscape and cell communication in the cervical cancer tumor immune microenvironment

To further characterize the pretreatment tumor microenvironment in more detail, we analyzed pre-chemoradiation therapy samples from Cohort 2 (Fig. 1E). We focused on the immune cell compartment by custom sub-clustering and identifying immune cell subsets based on their canonical marker gene expression (Fig. 2A, B; Table S2). This approach revealed 15 distinct immune cell populations, including three neutrophil subsets characterized by high expression of CCL3/4, HCAR2/3, and CXCR2, respectively; plasmacytoid dendritic cells (pDCs); conventional dendritic cells (cDCs); monocytes; two macrophage subsets distinguished by TREM2 or SPP1 expression; natural killer (NK) cells; CD8⁺ T-cells; CD4⁺ T-cells; regulatory T (T_reg) cells; plasma cells; B cells; and mast cells. As expected, pronounced inter-patient heterogeneity was evident in the pretreatment immune TME (Fig. 2C). However lymphoid cells, especially NK cells and T-cells, consistently comprised a substantial fraction of pretreatment immune infiltrates (Fig. 2D). Within the myeloid compartment, monocytes emerged as the predominant cell type (Fig. 2D). We further corroborated these findings in Cohort 1 through CIBERSORT analysis, which revealed a similar distribution of immune cell types (Fig. S2A), demonstrating consistency among cohorts.

Figure 2:

Characterization of the Immune Landscape and Cell-Cell Communication in the Cervical Cancer TME. (A) UMAP visualization of immune cell populations. (B) Heatmap displaying the top 50 differentially expressed genes for each identified immune cell subset, see Table S1. (C) Pie chart showing the distribution of immune cell populations across individual patients, color coding according to the cell clusters in (A). (D) Violin plots of the proportion of each cell type; each dot represents an individual patient. (E-G) Cell-cell communication using CellChat: (E) Number of interactions visualized as circle plots (left) and as a heatmap (right), line width and color gradient indicate interaction count, respectively. (F) Interaction strength visualized similarly: circle plots (left) and heatmap (right), line width and color gradient indicate signaling strength, respectively. (G) Heatmap summarizing all outgoing and incoming signaling pathways for each cell type.

Using the scRNA-seq data that included stromal and tumor cell compartments, we next inferred the global cell-cell communication networks within the cervical cancer TME. We identified 4,120 putative interactions among 17 distinct cell types (Fig. 2E–G), Performing the same analysis on each individual samples revealed that the top core interactions were conserved across most patients (Fig. S2B). Overall, stromal cells, SPP1⁺ macrophages, and tumor cells displayed the highest number of interactions and acted as both key signal senders and receivers (Fig. 2E and Fig. S2C). Notably, SPP1⁺ macrophages engaged in strong autocrine signaling and influenced monocytes, TREM2⁺ macrophages, and CXCR2⁺ neutrophils via SPP1-CD44 ligand-receptor pairs (Fig. 2E–F; Fig. S2D). Stromal cells were the second most influential senders, affecting similar targets primarily through collagen-CD44 interactions (Fig. 2E–F; Fig. S2D). Strikingly, CellChat failed to detect any immune checkpoint-related interactions (such as those involving PD-1/PD-L1, TIGIT/PVR, CTLA-4/Ligand axes) globally or by-sample (Fig. 2G).

By examining all outgoing and incoming signaling pathways (Fig. 2G), we identified distinct communication patterns linking cell clusters to signaling axes. Seven outgoing signaling patterns emerged, with some key observations: SPP1⁺ macrophages and HCAR2/3⁺ neutrophils formed a group in pattern out-1; stromal cells dominated pattern out-2; and tumor cells characterized pattern out-3, with each engaging in numerous signaling pathways (Fig. S2E). Conversely, incoming signaling patterns were grouped into four categories: pattern inc-1 included myeloid cells; pattern inc-2 encompassed tumor and B cells; pattern inc-3 consisted of stromal cells; and pattern inc-4 contained lymphoid cells, pDCs, and mast cells (Fig. S2E). These patterns suggest that cells of the same lineage exhibit similar responses to signals emanating from the TME, reflecting a level of coordination and specialization in the cellular crosstalk that shapes the cervical cancer milieu.

Standard-of-care chemoradiation reprograms the immune microenvironment in cervical tumors

To investigate treatment-related changes within the immune cell compartment, we performed k-nearest neighbors (k-NN) differential abundance analysis on the scRNA-seq data from Cohort 2, comparing samples collected before chemoradiotherapy (pre-RT) and three weeks post-initiation (on-RT#2) (Fig. 3A). While immune cell changes differed among individual samples before and after RT (Fig. S3A), we detected significant overall shifts in cellular proportions across these time points. Specifically, lymphoid populations, including T- and NK cells, were predominantly represented in the pre-RT samples. In contrast, myeloid populations, such as monocytes and both macrophage populations, were enriched in the on-RT#2 samples. Notably, neutrophils expressing CCL3/4 and CXCR2 were enriched after CRT, while HCAR2/3+ neutrophils were more abundant before therapy. The proportions of pDCs, cDCs, B cells, and mast cells remained relatively unchanged throughout the treatment course (Fig. 3B). Most of these shifts were comparable to those found by CIBERSORT-based predictions, which revealed similar patterns in Cohort 1 (Fig. S3B) and in the independent dataset from Cosper et al., 2020²² (Fig. S3C).

Figure 3:

Overall Reprogramming of the Immune Microenvironment in Cervical Cancer Patients upon Chemoradiotherapy. (A) UMAP visualization highlighting immune cell populations before (pre-RT) and after CRT (on-RT#2), only patients with both time points were included (CE322, CE323, CE326, CE331). (B) K-nearest neighbors (k-NN) differential abundance analysis comparing pre-RT and on-RT#2 samples from Cohort 2 and the results were overlapped using the UMAP (left) and visualized as bee swarm plot (right) to illustrates the main differences. (C) Differential gene expression analysis identifying CRT-induced changes across immune cell types. (D) Top 5 gene ontology terms enriched upon CRT across immune cell types, see Table S2. (E-K) Network plot of top significant upstream regulators (P<10⁻⁴) of immune cells pre-RT vs on-RT#2 time points, the inner nodes illustrate groups of regulators sharing genetic influences and the outer nodes represent contributing marker genes, color indicate the cell type; heatmaps for members of all nodes are presented to illustrate direction. (E) Neutrophils. (F) pDCs, DCs and monocytes. (G) Macrophages. (H) NK cells (I) CD4⁺ and CD8⁺ T-cells. (J) T_regs (K) Plasma and B cells.

Differential gene expression analysis revealed that standard-of-care CRT significantly altered expression of several important genes in each cell type (Fig. 3C). Using these data, we identified the primary gene ontology pathways that were enriched in each cell type upon exposure to chemoradiation. Cytoplasmic translation (GO:0002181) was the most common, found in 9 of the 12 clusters. This may indicate alterations in the production of proteins like cytokines or activation in response to the treatment (Fig. 3D and Table S3).

To investigate systemic signals that may be contributing to CRT-induced changes in the TME, we analyzed serial serum samples collected from these patients at time points corresponding to tumor sampling using mass spec proteomics and cytokine profiling (Fig. S3D–F). Notably, factors associated with macrophage recruitment, such as the soluble form of Folate Receptor Beta (FOLR2) (Fig. S3A) and Macrophage Colony-Stimulating Factor (M-CSF) (Fig. S3D–E), were increased with CRT in the serum of these patients. This is consistent with previous reports of induction of myeloid stimulating cytokines and growth factors by CRT in preclinical models^23,24. Interestingly, certain factors related to immune response stimulation, like MCP-2, IL-18 and Granzyme-B, were found to be decreased after initiation of therapy, suggesting potential systemic immunosuppressive effects or shifts in immune dynamics during treatment, consistent with our previous findings²².

To better understand the programs governing the response to CRT in the cervical cancer immune infiltrate, we performed upstream regulator analysis, integrating the results into enrichment maps to identify common regulatory nodes in cells of the same type. For neutrophil populations, the main four common nodes were related to mTOR (Node A) and neutrophil regulation elements (IRGM, TREX1), which were found to be activated upon CRT, while elements related to type-I interferon were downregulated (Nodes B and C). Interestingly, Node D reflects the proliferation activity of neutrophils recruited after CRT (Fig. 3E). These results suggest that CRT may significantly alter and enhance specific neutrophil subtypes in the immune TME, an observation which may be critically important as neutrophils have been previously shown to be associated with CRT resistance ²⁵.

Next, we analyzed pDCs, cDCs, and monocytes. Common upstream regulator nodes showed proliferation activity (Node A), inflammatory responses (Node B), metabolic processes (Node C), and type-I IFN signaling (Node D). Interestingly, both pDCs and cDCs showed activation of all nodes, while monocytes exhibited the opposite behavior (Fig. 3F). Activation of these nodes in cDCs and pDCs highlights their role in supporting immune responses in the TME of cervical cancer patients, while the opposite behavior in monocytes may indicate a tendency to differentiate into immunosuppressive phenotypes within the TME. These results suggest that enhancing the activity of pDCs and cDCs (e.g., with adjuvants or checkpoint inhibitors) might amplify anti-tumor immunity. Similarly, reprogramming monocytes towards a pro-inflammatory phenotype or preventing their immunosuppressive transformation within the TME (e.g., with M-CSF or IL-10 inhibitors) could improve therapeutic outcomes.

For macrophages, the four main nodes were related to immune response (Node A), inflammatory responses (Node B), TGFB1 signaling (Node C), and proliferation (Node D) (Fig. 3G). Our results highlight opposite behaviors between SPP1⁺ and TREM2⁺ macrophages in most of the regulators of Node A, Node B, and TP53 in Node D, while TGFB1 in Node C was downregulated in both cell types (Fig. 3G). These findings suggest distinct roles for these cells in the TME of cervical cancer upon CRT. Specifically, TREM2⁺ macrophages may play a pro-inflammatory role that could favor tumor progression^26,27, while the suppression of SPP1+ macrophages may reflect tissue repair ²⁸. Targeting both macrophage populations through tailored therapeutic strategies may offer significant benefits for improving patient outcomes.

In our analysis of NK cells, we identified regulatory nodes associated with activation (Node A), IL-15 signaling (Node B), inflammatory responses (Node C), and NK differentiation (Node D). Notably, the predicted downregulation of STAT5A, NFKB1, IL-15, and IL-2 suggests impaired NK cell activation or proliferation (Fig. 3H). This is significant, as STAT5A is essential for NK cell development and survival, primarily mediating signals from cytokines like IL-15 and IL-2²⁹. Similarly, NFKB1 plays a crucial role in immune responses driven by NK cells³⁰, and its downregulation could further contribute to defective NK cell function.

For CD4⁺ and CD8⁺ T-cells, we identified regulatory nodes associated with cell growth regulation (Node A), antiviral response (Node B), T-cell growth (Node C), and T-cell differentiation (Node D) (Fig. 3I). Both CD4+ and CD8+ T-cells exhibited similar responses to CRT, indicating a coordinated adaptation within the T-cell compartment. While this suggests a robust, shared response to CRT, it also underscores the potential for simultaneous activation and functional exhaustion in these populations. We analyzed regulatory T-cells (T_regs) independently and identified nodes corresponding to Treg differentiation (Node A), T-cell activation (Node B), CD40L signaling (Node C), and ID2 regulation (Node D) (Fig. 3J). The prominence of these nodes highlights the multifaceted role of T_regs in balancing immune activation and suppression. Notably, ID2 and CD40L are critical for maintaining Treg homeostasis^31,32. Understanding these regulatory networks is crucial, as T-cell/T_reg dynamics modulate immune responses and influence tumor microenvironment adaptation during CRT. Targeting these pathways could enhance the efficacy of CRT by rebalancing the immune response to favor anti-tumor immunity.

Finally, in plasma and B cells, we identified four regulatory nodes: transcription and metabolic regulation (Node A), immune response (Node B), proliferation (Node C), and immune activation and signal transduction (Node D). Among upstream regulators, only RICTOR, FMR1, TP53, CST5, TP63, and STAT1 were upregulated in both cell types, suggesting an adaptive response to maintain cellular functions and immune responses, while others, including MYCN, MLXIPL, CD40, IFNG, STAT3, EGF, and GLI1, were downregulated, potentially reflecting modulation of proliferative and activation pathways (Fig. 3K).

Tumor-intrinsic upregulation of DNA damage repair, p53/MDM2 and interferon signaling pathways in cervical tumors

We next turned our attention to identifying biologically targetable pathways that were specifically upregulated in tumor cells that persisted during CRT. Bulk RNA sequencing analysis performed in patients from Cohort 1 (Fig. 1A) before and during treatment, revealed that CRT enhanced the expression of genes associated with DNA repair pathways, most notably MDM2 and MAP4K4, while simultaneously downregulating histone-related genes (Fig. 4A). These findings are consistent with the induction of DNA damage responses as expected following radiotherapy. Among the differentially expressed genes, MDM2 displayed the most pronounced increase after CRT, implicating potential activation of p53, p63, or other DNA damage response mechanisms (Fig. 4B and Fig. S4A). Pathway enrichment analyses revealed modulation of extracellular matrix organization, direct p53 signaling, MAP kinase cascades, and immune response activation in response to CRT (Fig. 4C). Furthermore, analysis of our previously published paired bulk RNA-seq dataset²² revealed significant MDM2 upregulation across all patients (Fig. 4D), with a significant increase in MDM2 expression following CRT when comparing individually paired pretreatment samples (Fig. 4E).

Figure 4.

Effects of Chemoradiotherapy (CRT) on Tumors in Cervical Cancer Patients. (A) Heatmap showing differentially expressed genes in 17 paired tumor samples from Cohort 1, before and after CRT. (B) Volcano plot of deferentially expressed genes using −log₁₀P value (Y-axis) and log₂ fold change (X-axis), the red circles highlight genes with highest significance. (C) Pathway enrichment analysis of pre-RT vs on-RT DEGs, nGene Ratio and −log₁₀ (q-value) are shown (D) Volcano plot of differentially expressed genes comparing Pre-RT and Mid-RT (third week of CRT) from Cosper et al., 2020, with −log₁₀ P-value on the Y-axis and log₂ fold change on the X-axis. The red circles highlight the most significantly altered gene. (E) Paired comparison of MDM2 transcript levels in cervical cancer patients before (Pre) and mid-treatment (Mid), from Cosper et al. (2020). Each line connects matched samples; the exact P-value from a two-tailed paired t-test is shown. (F) Pie Charts depicting the changes in tumor and immune cell populations over three time points in samples from Cohort 2. (G) Expression of viral proteins following CRT. (H) Ingenuity pathway prediction for MDM2 activation, according to the expression of its canonical associated transcription factors and target genes prominently expressed in tumor cells prior CRT. (I) Enrichment score of curated MDM2 pathways for each cell type across three different time points from Cohort 2. (J) Canonical pathway analysis of tumor cells pre-RT vs on-RT#2 time points (K) Network plot of the top significant regulators (P < 10⁻⁴) in tumor cells comparing pre-RT and on-RT#2 time points. The inner nodes represent groups of regulators with shared genetic influences, while the outer nodes indicate contributing marker genes. Symbols denote the following: - differentially expressed without a clear predicted direction, ▲ predicted upregulation (orange), and ▼ predicted downregulation (blue).

To further expand upon these results over time at single-cell level, we evaluated samples from four patients in Cohort 2 at three distinct time points, before CRT, on-RT#1 and on-RT#2 (Fig. 1A). In most cases, CRT led to marked decreases in tumor cell numbers and significant changes the proportion of immune cells favoring increases myeloid derived cell populations (Fig. 4F). This is evidenced by reduced HPV-related gene expression in tumor cells from three of the four patients (Fig. 4G). Notably, during this same time course, MDM2 expression is upregulated in persisting cervix tumor cells across all patients, with some patient’s tumor cells (i.e., 323) demonstrating remarkable resistance to CRT. With respect to MDM2 expression, we found that tumor cells showed elevated MDM2 pathway signatures even prior to therapy, as reflected by the high expression of MDM2 related transcription factors and canonical target genes (Fig. 4H). After exposure to CRT, tumor and stromal cells exhibited significantly increased MDM2 pathway activation at all time points, while only mild (Normalized Enrichment Score, NES close to 0) MDM2 activation was observed in myeloid immune cell subsets following CRT (Fig. 4I). CRT-induced DNA damage can trigger type-I interferon signaling and other immunostimulatory pathways³³. In line with this, we observed pronounced type-I IFN activation in tumor cells after CRT exposure (Fig. 4J). Upstream regulator analysis predicted TP53 and TP63, key MDM2 targets, as significant mediators of this process (Fig. 4K). In contrast to these significant changes in p53/MDM2 and IFN-related gene expression in tumor cells after CRT exposure, we observed only mild CRT related changes in the expression of immune checkpoint genes including PD-L1 (CD274), PD-1 (PDCD1), CTL-4, LAG3 and TIGIT (Fig. S4B) across cell types, indicating that additional investigation is needed to understand how CRT modulates immune regulation in the cervix tumor microenvironment.

MDM2 inhibition enhances radiation response and reshapes the tumor microenvironment in preclinical tumor models

While we failed to observe significant overexpression or induction of immune checkpoint pathways by CRT, we did observe that CRT consistently and reproducibly activates DNA damage response pathways, leading to increased immune infiltration within the tumor microenvironment and upregulation of MDM2, a key regulator in the cellular response to therapy. Indeed, in vitro exposure of cervical cancer cell lines to CRT rapidly elevated both p53 and MDM2 protein levels Fig. 5A). Recent studies have identified AMG-232, a selective MDM2-p53 inhibitor³⁴, as a promising agents³⁵ and clinical trials have reported encouraging outcomes for myeloma and leukemia^36,37. For this reason, we wanted to test whether treatment with AMG-232 could improve radiation response in preclinical models of treatment resistant cervical cancer. Our data reveal that AMG-232, when combined with radiation, significantly reduces the proliferative capacity of HPV-positive, TP53 wild-type cervical cancer cell lines (SiHa and CaSki), while exhibiting no effect on HPV-negative, TP53-mutant C33A cells (Fig. 5B). Furthermore, this combination robustly induces the expression of p53 target genes (Fig. 5C) and increases apoptosis compared to radiation alone in HPV-positive cell lines (Fig. 5D). These findings suggest that MDM2 activation is crucial for the survival of TP53-proficient cervical cancer cells following radiation exposure.

Figure 5.

MDM2 inhibitor enhances standard-of-care in preclinical models of cervical cancer. (A-D) Cervical cancer tumor cells were exposed to radiation in the presence of an MDM2 inhibitor. (A) SiHa cells received RT (2 Gy per fraction, three times) administered from day 1 to day 5. Western blot analysis to assess p53, MDM2, and β-Actin levels, with densitometry values quantified relative to the control (day 0). (B) Colony formation assay results. (C) Heatmap showing the relative expression of MDM2 target genes, measured by RT-qPCR, in SiHa, CasKi, and C33A cell lines. (D) Percentage of apoptotic cells in SiHa, CasKi, and C33A cell lines. (E-I) TC-1 cells were implanted subcutaneously into the flank of C57BL/6 mice. Once tumors were established, mice received the MDM2 inhibitor AMG232 orally, in combination with standard-of-care therapy (radiation and cisplatin) or as monotherapy. Created in BioRender. Sandoval, T. (2025) https://BioRender.com/kup7w13 (E) groups: Vehicle, radiotherapy (RT), Cisplatin, standard-of-care radiotherapy combined with cisplatin (SOC), MDM2 inhibitor (MDM2_inh), standard-of-care combined with MDM2 inhibitor (SOC+ MDM2_inh) (F) t-SNE plot of flow cytometry data showing identified populations and (G) density plots for each treatment group. (H) Percentage of immune cells across treatments. One-way ANOVA with multiple comparisons; exact significant p-values are shown. (I) Tumor growth kinetics, with each line representing an individual mouse. (J) Area under the curve (AUC) analysis, with each dot representing an individual mouse. One-way ANOVA with multiple comparisons; exact significant p-values are shown (D, H and J).

Previous research has indicated that AMG-232 can sensitize tumor cells to T-cell-mediated killing in vitro and that MDM2 inhibitors exhibit favorable synergy with checkpoint blockade therapies^38–40. However, the in vivo effects of MDM2 inhibition on the TME in cervical cancer have not been tested. We hypothesized that MDM2 inhibition could not only enhance the efficacy of radiation therapy in tumor cells but also modulate the TME. To test this, we utilized TC-1 cells, a well-established model for HPV-related cervical cancer, as they express HPV-16 E6 and E7 oncoproteins⁴¹. TC-1 cells were implanted subcutaneously in the flanks of C57BL/6 mice. Eleven days post-implantation, mice received an oral dose of AMG-232, followed the next day by a single dose of tumor-directed radiotherapy or intraperitoneal cisplatin, administered alone or combined as standard-of-care (SOC), as shown in Fig. 5E. We identified eight major immune populations in TC-1 tumors using multicolor flow cytometry and a t-SNE dimensional reduction approach, along with conventional gating (Fig. S5). The treatments induced notable alterations in the immune composition (Fig. 5F–G). Interestingly, only the group receiving both SOC and the MDM2 inhibitor demonstrated a pronounced decrease in myeloid derived infiltration (Fig. 5H). These findings suggest that, beyond improving radiation response in tumor cells, MDM2 inhibition may favorably modulate the TME. (Fig. 5I–J). Notably, adding AMG-232 to radiotherapy alone in the TC-1 model further extended tumor growth delay and reduced average tumor sizes, though this decrease did not reach statistical significance compared to radiotherapy alone.

MDM2 Inhibition as a Radiosensitizing Strategy for CRT-Resistant Cervical Tumors.

To further investigate the therapeutic potential of MDM2 inhibition on tumor cells in a more clinically relevant context, we utilized patient-derived xenografts (PDXs) generated from cervical tumor tissue collected from patient 323 at two distinct time points: prior to treatment (pre-RT) and two weeks after the initiation of standard-of-care chemoradiation therapy (on-RT#1) (see Fig. 6A). This CRT resistant human tumor model enabled us to compare the effects of MDM2 inhibition in both treatment-naïve and CRT-edited (treatment-resistant) tumor settings. At the end of the study, tumors were harvested, and protein expression levels of MDM2, p53, and its downstream target p21 were analyzed via western blot. In the CRT-naïve PDX, only mild activation of p53 was observed after RT, with no detectable p21 induction, and treatment with the MDM2 inhibitor did not further increase p53 or MDM2 expression (Fig. 6B). In contrast, in the CRT-edited PDX, the combination of RT and MDM2 inhibition resulted in robust accumulation of p53 and p21, along with elevated MDM2 levels (Fig. 6C), consistent with our results using longitudinally collected human tumor biopsies (Fig. 4A–C).

Figure 6.

MDM2 inhibition enhances CRT-edited patient derived xenografts (PDX). PDX from patient 323 of Cohort 2, were collected at two different time points and implanted to NRG mice then the mice were treated as (A) groups: Control vehicle, MDM2 inhibitor (MDM2_inh), radiotherapy (RT) and the combination (RT+ MDM2_inh). End-point was two weeks after treatment. Created in BioRender. Sandoval, T. (2025) https://BioRender.com/h9tqczo (B-C) Western blot analysis of p53, p21, MDM2 and β-actin in (B) pre-RT PDX and (C) on-RT #1 PDX across all four treatment groups. (D-E) Representative images of tumors from each treatment group for (D) pre-RT PDX and (E) on-RT #1 PDX. (F-G) Tumor growth kinetics shown as spaghetti plots (each line = one mouse) alongside area-under-the-curve (AUC) analysis (each dot = one mouse) for (F) pre-RT PDX and (G) on-RT #1 PDX. Data were analyzed by one-way ANOVA with multiple comparisons; exact p-values are indicated.

At the end of this experiment, tumor size measurements revealed that the combination of RT and MDM2 inhibition significantly reduced tumor burden specifically in the CRT-edited PDX model compared to the CRT-naïve PDX derived from the same patient (Fig. 6D–E). Consistent with these findings, AMG-232 monotherapy did not substantially affect tumor growth in the pretreatment PDX, and RT alone induced only a modest delay in tumor progression (Fig. 6F). While the combination of RT and AMG-232 showed improved tumor control in the CRT-naïve PDX, the differences were not statistically significant. In contrast, in the CRT-edited PDX, the impact of MDM2 inhibition was significantly enhanced, both as a single agent and, more notably, in combination with RT (Fig. 6G). These findings support the notion that MDM2 inhibition may have the greatest therapeutic benefit in the CRT-resistant and or previously treated patient populations.

DISCUSSION

Standard-of-care (SOC) CRT results in an unacceptably high rate of relapse in cervical cancer, and the addition of traditional immune checkpoint blockade to SOC CRT has had mixed results in clinical trials, with improved outcomes for only a small proportion of patients^10,42,43. The present work represents the first multi-cohort translational study that offers a longitudinal, single-cell resolution view of human cervical cancer, both prior to and while receiving SOC CRT. Through holistic integration of WES, bulk RNA-seq, and scRNA-seq across two independent cohorts of patients, our study captures the patient and HPV-related genomic variabilities as well as the fine transcriptional heterogeneities that drive patient specific differences in tumor biology and the immune microenvironment. Importantly, in our study we failed to detect robust expression of traditional immune checkpoint proteins including the PD-1 axis in cervix tumor cells and T cells within the microenvironment, suggesting that alternative approaches to treatment may be needed.

Traditional clinical tests most commonly identify HPV16 and 18/45 in cervical cancer specimens⁴⁴. These genotypes were detected in our samples but further analysis using HPV-EM identified additional rare genotypes, co-infections, and confirmed truly HPV-undetected tumors. Our data illustrate the high genotypic and gene expression diversity of HPV-related genes across patient samples as evidenced by patient-specific clustering in our dimensional reduction approach. Using this strategy, we demonstrated that HPV genotype was a more significant driver of tumor cell clusters than histopathology alone. As HPV infections remain central to cervical cancer biology, refining detection methodologies to accurately capture genotype-specific expression signatures is critical for improving patient stratification and developing personalized treatment strategies²⁰.

Our results indicate that CRT leads to a notable decrease in expression of HPV related genes within tumor cells. This observation suggests that HPV-positive cells might be more susceptible to radiation-induced damage, as reported in other cancer types⁴⁵ or that therapy renders these cells more vulnerable to immune-mediated clearance, a phenomenon seen in virus-driven head and neck cancers⁴⁶. Regardless of the mechanism, a lower viral burden has been associated with decreased immunosuppression and better long-term outcomes^47,48. HPV-driven oncogenic transformation in cervical cancer leads to metabolic reprogramming^49,50. In line with this, our pathway enrichment analyses reveal that tumor cells exhibited a high metabolic need, increased proliferative advantage, and strong DNA damage response capabilities, which could offer an adaptive advantage in the chemoradiation setting. Thus, blocking metabolic pathways or DNA repair pathways could enhance standard treatments and present additional opportunities to bypass therapy immune resistance in cervical cancer patients.

Our comprehensive scRNA-seq analysis revealed a highly heterogeneous immune landscape. Multiple immune subsets emerged, including diverse neutrophil, macrophage, and T-cell populations, corroborating previous reports of robust NK cell, T-cell, and macrophage infiltration in human cervical cancer^11,16,51. Notably, we identified two distinct macrophage populations: TREM2⁺ macrophages, which are strongly immunosuppressive in several solid tumors⁵² and may also exhibit a pro-inflammatory role that could promote tumor progression^26,27, and SPP1⁺ macrophages, implicated in poor clinical outcomes⁵³. Monocytes were the most abundant myeloid cell type, consistent with their known roles in shaping inflammatory and immunosuppressive tumor states. Crucially, stromal cells and SPP1⁺ macrophages emerged as key mediators of cell-cell communication, both emitting and receiving extensive signaling cues. SPP1⁺ macrophages have also been linked to therapy resistance in various cancers^54,55, and some reports showed that SPP1-CD44 axis modulates tumor/T-cell interactions, potentially suppressing T-cell proliferation⁵⁶, and also promoting cancer progression via tumor-associated macrophage-cancer cell interactions⁵⁷. Likewise, an active collagen-CD44 axis, typically associated with wound-healing processes, has been identified as a marker of poor prognosis in gastric tumors⁵⁸ and has also been shown to drive malignant progression by enhancing adhesion, invasion, and metastasis⁵⁹. Our findings support and extend these observations, detailing the extensive autocrine signaling and influence on neighboring immune cells, particularly through SPP1-CD44 and collagen-CD44 interactions, that could serve as promising therapeutic targets.

Notably, we observed a shift from lymphoid to myeloid cell predominance of the immune TME during CRT, with increased proportions of neutrophils, monocytes and TREM2⁺ macrophages during treatment⁶⁰. CCL3/4⁺ and CXCR2⁺ neutrophil populations have been implicated in both pro- and anti-tumor roles across various cancers, while the functions of HCAR2/3⁺ neutrophils are less understood⁶¹. In head and neck squamous cell carcinoma, TREM2⁺ multinucleated giant macrophages are associated with favorable prognosis⁶². Conversely, TREM2⁺ macrophages contribute to an immunosuppressive environment in other tumor types, promoting tumor progression⁶³. Additionally, circulating monocytes increase upon CRT in cervical cancer patients and the consequences of this shift have been related to immunosuppression⁶⁴. The changes we observe present opportunities to optimize immune responses as a strategy to enhance CRT-induced anti-tumor immunity in cervical cancer. For instance, the enrichment of TREM2⁺ macrophages and the observed defective phenotype of NK cells highlight potential new therapeutic avenues²⁶.

Despite the ability of CRT to induce a robust type I IFN signaling in tumor cells, our analysis revealed that this activation did not extend to surrounding immune or stromal cells within the TME, suggesting a limited paracrine IFN signaling necessary for effective immune activation. Upstream regulator analysis indicated modulation of immune response, inflammation, and proliferation pathways across multiple immune cell subsets. Type-I IFNs are critical for anti-tumor immunity, as they enhance antigen presentation and promote cytotoxic T-cell responses which may influence the efficacy of immune checkpoint blockade and other immunotherapies⁶⁵. This selective, cell type-specific modulation of IFN signaling suggests that CRT shift the balance between immune activation and suppression within the TME. However, new strategies are required to extend these signals to surrounding immune cells in the cervical cancer microenvironment, where type-I IFN responses could drive both direct cytostatic/cytotoxic effects on malignant cells and broad immunomodulatory functions⁶⁶.

Moreover, checkpoint profiling revealed distinct, cell-specific layers of immunoregulation. PD-L1 (CD274) was largely confined to SPP1⁺ macrophages, consistent with their documented immunosuppressive role^67,68. As expected, PD-1 (PDCD1) was abundant on NK cells, CD8⁺ T cells, and Tregs, while LAG-3 was enriched in NK cells, collectively illustrating how the tumor microenvironment dampens cytotoxic immunity. A second myeloid checkpoint axis emerged with TIGIT, which was highly expressed by both SPP1⁺ and TREM2⁺ macrophage subsets. In all cases, even in the tumor compartment, CRT induced only mild changes in these patterns. Collectively, these findings indicate that CRT alone does not elicit robust, targetable checkpoint expression, and that additional strategies are needed to boost immunotherapeutic efficacy as part of standard of care treatment in cervical cancer.

While results for expression of immune checkpoint genes was variable across cell types, our analysis of the tumor cell compartment revealed that CRT significantly and consistently upregulates DNA damage response. Furthermore, using an independent cohort from a previously published study²², we validated this finding by showing that MDM2 expression increases following radiation treatment. Notably, results from Cohort 2 confirmed that MDM2 pathway activation was predominantly observed at the tumor cell level rather than the immune component, where no significant changes were observed upon treatment. These findings suggest that CRT induces specific molecular alterations within tumor cells, highlighting MDM2’s potential role in mediating treatment responses. MDM2 can influence cell cycle progression and apoptosis by controlling p53 levels, while MAP4K4 participates in various stress-related signaling pathways. Downregulation of histone-related genes may alter chromatin structure, potentially reshaping gene expression profiles in response to CRT. The elevated MDM2 activity in tumor cells observed prior to therapy, and the activation of this factor during therapy in tumor and stromal cells, suggests that DNA damage response pathways are central to resistance mechanisms and may influence cellular outcomes during treatment.

Based upon these results, we targeted MDM2 in vitro and in vivo using preclinical models of viral induced tumors and novel patient-derived xenografts derived from CRT-naïve and CRT-edited cervix tumor biopsies. We observed that the pharmacological inhibition of MDM2 enhanced the efficacy of radiation in HPV-positive, TP53 wild-type tumor cells and favorably altered the immune microenvironment. These findings demonstrate that combining MDM2 inhibition with SOC CRT may overcome resistance mechanisms and improve clinical outcomes in cervical cancer. Furthermore, using our novel PDXs, we demonstrate that the efficacy of the addition of MDM2 inhibition with radiation is enhanced in CRT exposed tumors. Yet, the prognostic significance of treatment-related MDM2 upregulation in relation to clinical outcomes remains to be determined, as currently available datasets lack sufficient numbers of longitudinally collected samples.

By combining advanced single-cell analyses with functional validation in preclinical models, our study underscores a direct translational route to potentially improve treatment responses in HPV positive TP53-proficient cervical cancer. This work sets the stage for future clinical trials aimed at defining new biomarkers, refining patient selection and maximizing clinical benefit of immunotherapy combined with chemoradiation.

Data Availability

RNA-seq data generated in this study are available in the GEO database under the accession numbers GSE297038 (Chemoradiation Reprograms Tumor Cells and Immune Microenvironment In Cervical Cancer [bulk RNA-seq]) and GSE297041 (Chemoradiation Reprograms Tumor Cells and Immune Microenvironment In Cervical Cancer [scRNA-seq]). The whole exome sequencing data generated in this study are publicly available in the database of Genotypes and Phenotypes (dbGaP) under accession number phs004441.v1.p1. All requests for further information, raw data, or resources and reagents should be directed to and will be fulfilled by the corresponding author, Dr. Julie Schwarz (jschwarz@wustl.edu).