Effects of HOX family regulator-mediated modification patterns and immunity characteristics on tumor-associated cell type in endometrial cancer

Abstract

Endometrial cancer (UCEC) is one of three major malignant tumors in women. The HOX gene regulates tumor development. However, the potential roles of HOX in the expression mechanism of multiple cell types and in the development and progression of tumor microenvironment (TME) cell infiltration in UCEC remain unknown. In this study, we utilized both the The Cancer Genome Atlas (TCGA) database and International Cancer Genome Consortium (ICGC) database to analyze transcriptome data of 529 patients with UCEC based on 39 HOX genes, combing clinical information, we discovered HOX gene were a pivotal factor in the development and progression of UCEC and in the formation of TME diversity and complexity. Here, a new scoring system was developed to quantify individual HOX patterns in UCEC. Our study found that patients in the low HOX score group had abundant anti-tumor immune cell infiltration, good tumor differentiation, and better prognoses. In contrast, a high HOX score was associated with blockade of immune checkpoints, which enhances the response to immunotherapy. The Real-Time quantitative PCR (RT-qPCR) and Immunohistochemistry (IHC) exhibited a higher expression of the HOX gene in the tumor patients. We revealed that the significant upregulation of the HOX gene in the epithelial cells can activate signaling pathway associated with tumour invasion and metastasis through single-cell RNA sequencing (scRNA-seq), such as nucleotide metabolic proce and so on. Finally, a risk prognostic model established by the positive relationship between HOX scores and cancer-associated fibroblasts (CAFs) can predict the prognosis of individual patients by scRNA-seq and transcriptome data sets. In sum, HOX gene may serve as a potential biomarker for the diagnosis and prediction of UCEC and to develop more effective therapeutic strategies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s43556-024-00196-w.

Introduction

UCEC, comprising a spectrum of malignant epithelial tumors originating from the uterine lining, was among the principal gynecological malignancies, constituting approximately 20% to 30% of all cancers within the female reproductive system. Notably, a study conducted in the United Kingdom documented a twofold increase in incidence between 1993 and 2014, with projections indicating a continued rise through 2035 [1]. The HOX genes, a family of highly conserved DNA sequences found across eukaryotes [2], played pivotal roles in developmental regulation, orchestrating not only growth and development but also the establishment of the immune system [3]. In humans, these genes are organized into four clusters—HOXA, HOXB, HOXC, and HOXD—residing on chromosomes 7, 17, 12, and 2, respectively. The transcription factors encoded by HOX genes exhibit dual functions, acting as both tumor suppressors and oncogenes [4]. MicroRNAs (miRNAs) embedded within the HOX loci can modulate HOX expression, impacting apoptosis and thus influencing carcinogenesis [5]. Research has identified specific members of the HOX family with tumor-suppressive capabilities; for instance, HOXB13 has been characterized as a tumor suppressor [6–8]. Similarly, the ectopic expression of HOXA 5 in lung adenocarcinoma cell lines has been shown to inhibit cell migration, invasion, and filopodia formation in vitro, while concurrently reducing metastatic potential in vivo [9]. Furthermore, a significant downregulation of HOXD 10 mRNA expression was observed in gastric cancer tissues, correlating with enhanced tumor cell proliferation, migration, and invasiveness [10].

The tumor immune response was a sophisticated and dynamic interplay that exerts dual influences on the genesis and evolution of cancer. This intricate process unfolded in three distinct phases: elimination, equilibrium, and escape. In the initial phase of tumor development, immune effector cells, including T lymphocytes (TLs), mounted a defense against cancerous cells. TLs could produce inflammatory cytokines and differentiate into effector cytotoxic T lymphocytes (CTLs) capable of directly killing tumor cells [11, 12]. Furthermore, CTLs activation can modulate immune signaling within the TME, potentially enhancing antitumor immunity [13]. As the tumor advances, it adopted strategies to evade immune surveillance. Tumor cells secreted immunosuppressive cytokines such as interleukin-10 (IL-10), which dampened the activity of activated CTLs, facilitating tumor expansion and immune escape [14]. Despite this, research exploring the impact of HOX genes on immunity within the TME remained sparse. One study elucidated that HOX antisense transcript expression correlates with genes involved in immune cell infiltration, distinct immune subtypes, and the immunological response to cancer [15]. In the context of colorectal cancer, elevated expression of HOXC 6 had been linked to the activation of cytokine pathways, upregulation of chemokines that attracted T cells, infiltration of immune cells, and the expression of immune checkpoint molecules [16]. Aberrant expression of specific HOX genes, notably HOXA 10, HOXB 9, and HOXD 10, had been observed in UCEC [17–20]. Dysregulated HOX gene expression activated several pathways, including PI3K/Akt, which fostered endometrial carcinoma cell proliferation and simultaneously hinder apoptosis [21].

CAFs represented a prominent cellular component within the TME. Activated CAFs exerted their influence by secreting a plethora of cytokines and chemokines, including CXC ligand 2 (CXCL2), which facilitated tumor cell proliferation and metastasis [22]. Recent investigations had unveiled the capacity of CAFs to induce angiogenesis in endometrial malignancies via IL-10 secretion, thereby accelerating tumor advancement [23]. Despite the growing understanding of CAFs' contributions to cancer, the specific impact of HOX genes on immunity, the modulation of the TME, and their interaction with CAFs remained largely uncharted territories. Whether HOX genes can potentiate the onset and progression of UCEC through these mechanisms was an open question that warrants further scrutiny.

In this study, aiming to develop a more effective and innovative methodology for evaluating the clinical relevance of UCEC and predicting immunotherapy efficacy and prognosis with greater precision, we integrated HOX-related genomic information from 529 endometrial cancer samples and evaluated the role of HOX-related genes in endometrial cancer, combined with associated immune and TME cell infiltration features. We also constructed a scoring system to quantify HOX expression patterns in individual patients and combined the scores with immune, pharmacotherapy, and related clinical traits to further guide clinical treatment. Moreover, we evaluated expression of related genes and immune cell infiltration on the prognosis of patients with UCEC. Finally, we developed a prognostic risk model to further assess patient outcomes. Our study provides a new approach to guide more effective clinical practice and therapy for patients with endometrial cancer.

Results

Identification of HOX family-related differential expression genes (DEGs) with prognostic significance and genomic variance analysis in the UCEC cohort

To investigate the clinical relevance of HOX gene family members in endometrial carcinoma, we examined the effects of 39 HOX genes pan-cancer based on 4 aspects: disease-free interval (DFI), disease-specific survival (DSS), overall survival (OS), and progression-free survival (PFS). The results showed that the expression of 28 different HOX genes was associated with significant differences in DFI, DSS, OS, and PFS in UCEC (Fig. S1a). There were differences in the regulation of tumor grade between different HOX genes (Fig. S1b). And HOX genes can activate or inhibit certain signaling pathways, such as, receptor tyrosine kinase, ras-mitogen-activated protein kinase, cellcycle and so on (Fig. S1c). These results indicated that the HOX gene family plays a crucial role in UCEC. There was the whole process of our study (Fig. S2).

To determine whether expression of HOX genes is involved in tumor formation, we studied its mutation frequency, copy number variation (CNV), and expression status. Among the 529 samples, 124 showed HOX gene mutations, with a frequency of 23.44%, including mostly missense, nonsense, and multiple hit mutations. HOXA 1 and HOXA 2 showed the highest mutation frequencies, follow by HOXA 11, HOXC 10, and HOXC 6. The only HOX gene without a mutation was HOXB 4 (Fig. S3a). We further analyzed the CNV frequency in HOX gene, mostly consisting of copy number amplifications (Fig. S3b). The locations of CNV alterations in HOX genes on chromosomes were shown (Fig. S3c). HOXA 1–7 are mostly concentrated on chromosome 7; HOXB 1–3 are located on chromosome 17; HOXC 6 and 10–13 are located on chromosome 12; and HOXD 1, 3, 4, and 8–13 are located on chromosome 2, and they all presented as deletions. To explore the functional differences caused by CNV alterations in the HOX genes in UCEC, the Kyoto Encyclopedia of Genes and Genomes (KEGG) the Gene Ontology (GO) functional enrichment were performed. The results exhibited those genes were enriched in signaling pathways regulating pluripotency of stem cells, DNA-binding transcription activator activity, RNA polymerase II-specific and DNA-binding transcription activator activity (Figs. S3d and S3e). In addition, we compared the expression of the HOX gene between UCEC and normal tissues. There were obvious differences in the expression levels of HOX genes in the tumor and normal groups (Fig. 1a). These results suggest that the deletion status of HOX gene regulates its upregulation and downregulation in tumor tissues, and HOX gene plays a vital role in the occurrence and development of UCEC. Next, further analysis of the interactions between HOX genes and tumor formation revealed a strongly significant positive correlation between most of HOX genes, with the correlation coefficient between HOXB 6 and HOXB 5 being the highest (0.97) (Fig. S3f). These results indicate that HOX genes do not affect the development of tumors alone, but also have certain synergistic or antagonistic effects. In summary, this analysis showed the high heterogeneity of genetic and expression landscapes in HOX gene between normal and UCEC samples, suggesting that HOX gene expression imbalance plays a pivotal role in UCEC occurrence and progression.

Fig. 1.

Differential expression of HOX genes. a Differences in the expression levels of HOX genes between normal and UCEC samples. The upper and lower ends of the boxes represent the interquartile range of values. The lines in the boxes represent the median value, whereas dots show the outliers. Asterisks represent the statistical P-value (*P < 0.05; **P < 0.01; ***P < 0.001). b Principal component analysis of the three patterns of HOX expression, showing a remarkable difference on gene profiles between different patterns. c Kaplan–Meier curves of overall survival (OS) for 575 patients with UCEC in the TCGA and ICGC cohorts with different HOX clusters. The numbers of patients in A, B, and C phenotypes were 238, 124, and 39, respectively (P = 0.012. Log-rank test). d GSVA enrichment analysis showing the state of metabolic pathways in distinct patterns between A and C clusters. The heatmap was used to visualize these biological processes. Red indicates activated pathways, whereas blue represents inhibited pathways. e 12 HOX-related DEGs between three HOX clusters are shown in the Venn diagram. f The survival curves of the HOX cluster-related gene signatures were estimated using the Kaplan–Meier plotter. (P < 0.001, Log-rank test). g Unsupervised clustering of overlapping HOX cluster-related DEGs for classifying patients into different genomic subtypes. Red indicates high expression, whereas blue represents low expression. Values represent the expression strength

Identification of HOX phenotype-related DEGs and regulation patterns

To further ascertain the influence of HOX on UCEC, we obtained gene expression data and full clinical annotations for 401 UCEC patients. We used unsupervised clustering analysis and principal component analysis (PCA) to classify 401 patients with UCEC and identified three distinct modification patterns, including 238 cases of pattern A, 124 cases of pattern B, and 39 cases of pattern C (Figs. S4a and 1b). These clusters were named HOX clusters A, B, and C, respectively. Survival analyses indicated that these HOX clusters were significantly related to prognosis of patients with UCEC, with HOX cluster A exhibiting a prominent survival advantage (Fig. 1c).

Next, we examined characteristics of the three HOX clusters according to different clinical traits. We conducted unsupervised clustering of 39 HOX genes in the cohort with clinical annotations, including age, sex, tumor grade, and survival status. The heatmap not only revealed clinical characteristics of the three HOX clusters but also their correlation with HOX gene expression. There were significant differences in HOX transcriptional profiles among the three HOX clusters (Fig. S4b). HOXB sets exhibited a downward trend from HOX cluster A to B to C and showed notably low expression in HOX cluster C. We also found that the proportions of patients with higher stages and death outcomes were significantly higher in HOX cluster C.

Subsequently, we performed gene set variation analysis (GSVA) for the all DEGs to explore the functions of the three HOX clusters. As shown in the heatmap, between HOX clusters A and B, genes associated with DNA replication, cell cycle, RNA degradation, base excision repair, homologous recombination, mismatch repair, and nucleotide excision repair were more highly expressed in HOX cluster B than in HOX clusters A (Fig. S4c). There were some pathways with higher expression in HOX cluster C than in HOX cluster A, including those associated with DNA replication, cell cycle, P53 signaling, base excision repair, homologous recombination, and mismatch repair (Fig. 1d).

These results illustrate that aberrant expression of the HOX gene may affect cell biological behavior and cell signaling and DNA repair pathways, which are closely linked to tumor formation and development. These pathways include the cell cycle, DNA replication, P53 signaling, base excision repair, homologous recombination, mismatch repair, and nucleotide excision repair. These results were further supported by the fact that HOX gene expression was associated with tumor formation and progression.

Construction of a novel risk score for regulation patterns mediated by HOX regulators in UCEC

To investigate the potential biological behavior associated with each HOX gene pattern, we analyzed all of the DEGs in the three HOX clusters and determined 12 HOX cluster-related DEGs (Fig. 1e), including HOXA 7, HOXA 9, HOXB 2, HOXB 3, HOXB 4, HOXB 5, HOXB 6, HOXB 7, HOXB 8, HOXC 4, SEMA3E and ASRGL1, and then performed unsupervised clustering and PCA based on the these HOX cluster-related DEGs to classify patients into two different genomic subtypes named gene clusters A and B (Figs. S4d and S4e). Survival analysis indicated that gene clusters were significantly related to prognosis in patients with UCEC, and gene cluster A exhibited a prominent survival advantage (Fig. 1f). Next, we explored the characteristics based on different clinical traits using unsupervised clustering in the cohort. The heatmap revealed a significant difference in HOX cluster-related DEGs between the two gene clusters: most of the HOX cluster-related DEGs showed a low expression status in gene cluster B, and patients in HOX cluster C were mainly concentrated in gene cluster B (Fig. 1g). These results indicate that two distinct gene patterns are present in UCEC and are closely related to clinicopathological characteristics. Moreover, a prominent difference was noted in the expression of HOX between the two gene clusters, and HOX gene was the source of prominent differences in the two genomic phenotypes (Fig. 2a). To explore the characteristics of these HOX patterns in different clinical traits, we focused on the groups of tumor and normal. The unsupervised clustering by the DEGs revealed distinct patterns of HOX gene expression. To investigate the concrete regulating function of HOX gene in UCEC, further analysis was performed to elucidate the molecular mechanisms of the HOX gene and their interacting functions in three HOX clusters from RNA sequencing datasets (Fig. S5). HOX cluster A contained HOXB 2 and HOXB 3. Compare to the normal samples, the major functions were downregulation in the tumor samples, including DNA-templated DNA replication, DNA replication, double-strand break repair, chromosome segregation, nuclear chromosome segregation, organelle fission. HOX cluster B contained HOXB 6, HOXB 7 and HOXB 13; Major enriched functions included mitochondrial gene expression, mitochondrial translation, oxidative phosphorylation, aerobic respiration, proton motive force-driven ATP synthesis, purine nucleoside triphosphate biosynthetic process. These functions were downregulation in the tumor samples. HOX cluster C contained HOXA 3, HOXA 4, HOXA 13, HOXC 4, HOXC 6 and HOXD 13. The major enriched functions included cell-substrate adhesion, small GTPase mediated signal transduction, cell–matrix adhesion, muscle organ development, regulation of supramolecular fiber organization, response to transforming growth factor beta. The functions above were upregulation in the tumor samples. To explore the relationship between biological behaviors and HOX gene expression in UCEC, we performed GO enrichment analyses of the 12 HOX cluster-related DEGs. GO enrichment analysis showed that HOX-related genes were enriched in multiple pathways, namely DNA-binding transcription repressor activity, DNA-binding transcription activator activity, RNA polymerase II-specific, and transcription regulator complex (Fig. 2b). These pathways play a regulatory role in the formation and development of tumors, further illustrating that HOX-related DEGs play a crucial role in the regulation of tumor cells.

Fig. 2.

HOX scores combined with clinical features and immune cells. a Differences in the expression levels of HOX genes between two different genomic subtypes. The upper and lower ends of the boxes represent the interquartile range of values. The lines in the boxes represent the median value, whereas dots show the outliers. Asterisks represent the statistical P-value (*P < 0.05; **P < 0.01; ***P < 0.001). b GO functional enrichment analysis of 12 overlapping HOX cluster-related DEGs in molecular function (MF). Red indicates high expression, whereas blue represents low expression. The length of the bar chart indicates the count of enriched genes. c Survival analyses of low- (360 cases) and high- (41 cases) HOX score patient groups using Kaplan–Meier curves (P < 0.001, Log-rank test). d Differences in HOX score among distinct tumor grade groups. e Differences in HOX score among distinct survival state groups. f Correlations between HOX score and TME infiltrating cell types using Spearman analysis. Blue indicates negative correlation, whereas red represents positive correlation. Values represent the correlation strength. g Differences in the abundance of immune cells and pathways between low and high HOX score groups. The upper and lower ends of the boxes represent the inter-quartile range of values. The lines in the boxes represent the median value, whereas dots show the outliers. Asterisks represent the statistical P value (*P < 0.05; **P < 0.01; ***P < 0.001)

As the above analyses were based on the entire cohort, they did not accurately predict the patterns of HOX expression in individual tumors. Therefore, we constructed a set of scoring models based on these cluster-related genes to quantify HOX gene expression in individual patients with UCEC and predict treatment response and prognosis of patients with UCEC. We termed this the HOX score. Based on the correlation of the HOX score with different HOX patterns and gene phenotypes, we determined the optimal cutoff value 3.6953 by “survminer” package, and divided 401 patients with UCEC into high (n = 41) and low (n = 360) HOX score groups. Next, we conducted survival analyses to assess the value of the HOX score in predicting patient outcomes. Kaplan–Meier curves showed that patients with low HOX scores had significantly better survival than those with high HOX scores (Fig. 2c). In this cohort, HOX cluster A and HOX gene cluster A were associated with lower HOX scores and better prognosis, verifying the robust and independent prognostic value of HOX gene signatures.

Immune landscape was significantly associated with the expression of HOX score

To better validate the assessment potential of the HOX score in patients with UCEC, we combined this score with associated clinical features including age, tumor grade, and survival. We observed the proportions of patients with different clinical shapes in the high- and low-scoring groups. The histogram shows that in the high group, the proportion of patients with tumor grade 3 as well as dead was almost twice that of the low group (Figs. S6a-b); The proportion was similar in age between the different groups (Fig. S6c). In addition, box plots more intuitively showed the difference in HOX scores between the different clinical shape groups; patients in the G3 group had significantly higher HOX scores than those in the G1 and G2 groups (Fig. 2d); the scores were also higher in the death group than in the survival group (Fig. 2e). There were no statistical differences in HOX scores between the different age groups (Fig. S6d). The changes in the attributes of individual patients with UCEC were visualized using an alluvial diagram (Figs. S6e-g), and a close association was observed between HOX cluster C, gene cluster B, high HOX scores, tumor grade 3, and death. These results demonstrate that the HOX score can predict individual characteristics and has the potential to serve as a biomarker for assessing clinical characteristics and predicting prognosis in patients with UCEC.

To investigate the role of HOX in immune cell infiltration in the TME, TME cell-infiltrating characteristics were compared among the three HOX clusters. Notably, HOX cluster A was remarkably rich in innate immune cells, including activated macrophages, monocytes and plasmacytoid dendritic cell (Fig. S7a). Next, we investigated the relationship between HOX scores and different infiltrating immune cell characteristics. The HOX score was negatively correlated with numbers of activated dendritic cells, eosinophils, macrophages, monocytes, plasmacytoid dendritic cells, and T follicular helper cells, and positively correlated with activated CD 4 cells and type 2 T helper cells (Fig. 2f). Next, we compared the immune cell and immune function scores between high and low HOX score groups and noted remarkable differences. The low HOX score group generally had higher expression of immune cells, including B cells, CD8⁺ T cells, cytolytic activity, human leukocyte antigen (HLA), immature dendritic cells (iDCs), inflammation-promoting cells, natural killer (NK) cells, neutrophils, macrophages, T cell co-stimulation, T helper cells, T helper 1/T helper 2 cells, and type I/II interferon (IFN) responses (Fig. 2g). The heatmap showed that the expression of immune-infiltrating cells was lower in the high-score group than in the low-score group (Fig. S7b). Finally, we further investigated the relationship between immune scores and immune-infiltrating cells, revealing that the following cells had a positive correlation with immune scores: M1/M2 macrophages, NK resting cells, regulatory T cells, activated CD 4 memory T cells, and CD8⁺ T cells. Several other cells, such as naive B cells, plasma cells, activated dendritic cells, and eosinophils, were negatively correlated with immune scores (Fig. S7c).

Some soluble cytokine encoded by the IFNG gene are involved in tumor clearance, dormancy, escape, and progression [24, 25]. To explore the relationship between the IFNG gene and HOX scores, we examined the expression of the IFNG gene, which was found to be higher in the low HOX score group than in the high HOX score group (Fig. S7d). Myeloid-derived suppressor cells (MDSCs) are one of the main suppressive cell populations in the immune system. Some immature myeloid cells become MDSCs and exert immunosuppressive effects on CD8⁺ T cells [26]. To understand the differences in MDSCs and CD8⁺ T cells among the different groups, we observed their expression in the high- and low-scoring groups (Figs. S7e-f). Expression of MDSCs were lower in the low HOX score group than in the high HOX score group, whereas CD8⁺ T cell expression in low HOX score grouping was higher, illustrating that immune cell infiltration was more abundant in the low-score grouping. These results indicate that HOX scores reflect inhibition of immune cell infiltration in TMEs, leading to proliferation and metastasis of tumor cells.

The role of HOX score in predicting tumor mutations and exacerbates tumor microsatellite instability

To understand the mutation status of the HOX gene in tumors, determine the effect of microsatellite instability, and predict the immunotherapy effect for patients in different HOX score groups, we divided the patients into two categories: L-TMB and H-TMB. Kaplan–Meier curves revealed that H-TMB patients had longer overall survival (Fig. 3a). Next, we evaluated overall survival among patients with UCEC combined with tumor mutation burden (TMB) and HOX scores. We observed that patients with both a high HOX score and high TMB showed a significant survival advantage (Fig. 3b). To explore the relationship between HOX scores and TMB, we performed a correlation analysis, which revealed a positive correlation between the two measures (Fig. 3c). We then analyzed somatic mutations between low and high HOX score groups in TCGA database and ICGC database cohorts using the “maftools” package and found higher mutation rates in genes associated with TMB in the high HOX score group (Figs. 3d-e). A box plot shows that the TMB was higher in the high-score group than in the low-score group (Fig. 3f). These results indicate that HOX promotes gene mutations in tumor cells. Next, we investigated the relationship between HOX score and tumor microsatellite instability (MSI). The proportion of patients with different MSIs in the low and high HOX score groups was analyzed, and we found that the proportion of MSI-H in the high HOX score group was nearly twice that in the low HOX score group (Fig. S7g). The HOX score was higher in the MSI group than in the microsatellite stability (MSS) group (Fig. S7h). These results indicate that factors related to high HOX scores may exacerbate tumor MSI. In addition, induced pluripotent stem cells can play a role in cancer therapy by inducing the proliferation of different immune cells [27]; therefore, we explored stem cell differences between the different HOX score groupings and found that the expression of immune cell proportion score (IPS) did not differ between them (Figs. S8a-d).

Fig. 3.

Association of HOX genes with TMB. a Kaplan–Meier curves of overall survival (OS) of patients with two TMB clusters. The numbers of patients in low and high TMB groups were 328 and 58, respectively (P = 0.023, Log-rank test). b Survival analyses of patients with both HOX score and TMB using Kaplan–Meier curves. H, high; L, Low; TMB (tumor mutation burden) (P < 0.001, Log-rank test). c The correlation between TMB and the HOX score in two gene clusters. d-e The waterfall plot of genetic alterations in patients with low (d) and high (e) HOX score. Each column represents individual patients. The upper barplot shows the TMB, while the number on the right indicates the mutation frequency in each gene. The right barplot shows the proportion of each variant type. f Differences in the TMB between low and high HOX score group

The above results show that HOX can promote gene mutations in tumor cells and exacerbate tumor MSI; therefore, patients in high HOX score groups appear to be better able to produce neoantigens, which are sensitive to immunotherapy.

Exploration of effect on immune checkpoints and the association with HOX scores

We next explored the relationship between immune checkpoints and high and low HOX score groups. The high HOX score group presented a lower expression state for most immune checkpoints such as CD44, CD27, CD86, CTLA4, TIGIT and so on (Fig. 4a). Spearman’s analysis showed that checkpoints with the strongest correlation with HOX scores were CD244, CD96, and TIGIT; the correlation was negative (Fig. 4b). Next, we further verified the relationship between the three immune checkpoint genes and HOX scores (Figs. S8e-g). In addition, we verified the relationship between other checkpoint genes and HOX scores, and found that several checkpoints, namely CSF1R, BTLA, CTLA4, HAVCR2 and PDCD1, were negatively correlated with HOX scores (Figs. S8h-l). The results showed that a high HOX score was associated with a block in immune checkpoints, improved immune activation in patients with tumors, and enhanced immune response to immunosuppressants.

Fig. 4.

Combination of HOX genes with immune checkpoints and anti-tumor drugs. a Differences in the abundance of immune checkpoints between low and high HOX score groups. The upper and lower ends of the boxes represent the interquartile range of values. The lines in the boxes represent the median value, whereas black dots show the outliers. Asterisks represent the statistical P value (*P < 0.05; **P < 0.01; ***P < 0.001). b Correlations between HOX score and immune checkpoints using Spearman analysis. Blue indicates negative correlation, whereas red represents positive correlation. Values represent the correlation strength. c Comparison of the relative distribution of immune cells including TIDE, between high and low HOX score groups. d Kaplan–Meier curves of overall survival (OS) of patients. The numbers of patients in low- and high-TIDE groups were 97 and 304, respectively (P = 0.005, Log-rank test). e Survival analyses of patients with both HOX score and TIDE using Kaplan–Meier curves. (P < 0.001, Log-rank test). f Kaplan–Meier curves of overall survival (OS) of patients. The numbers of patients in the low- and high-immune dysfunction groups were 97 and 304, respectively (P = 0.005, Log-rank test). g Survival analyses of patients with both HOX score and immune dysfunction using Kaplan–Meier curves. (P < 0.001, Log-rank test). h-i Differences in the IC₅₀ differences of anti-tumor drugs between different HOX score groups. h: cyclopamine. i: foretinib. j-k Correlation between anti-tumor drugs and HOX scores. j: cyclopamine. k: foretinib

Subsequently, we performed GSVA enrichment analysis to compare differences in the activation states of immune cells and function between patients in distinct HOX score groups. As shown in the heatmap, the high HOX score group showed suppression of various immune cells and pathways such as CD8⁺ T cells, NK cells, T helper cells, and type I/II IFN responses (Fig. S9a).

To explore the correlation between HOX scores and the proportion of immune and stromal cells in TMEs and further examine differences in survival between the two HOX score groups, we analyzed Estimation of STromal and Immune cells in MAlignant Tumour tissues using Expression data (ESTIMATE) scores, immune scores, stromal scores, and tumor purity (Figs. S9b-e). The Violin plot shows that ESTIMATE, immune, and stromal scores were higher in the low-score group than in the high-score group, whereas tumor purity was lower in the low-score group. Therefore, compared with UCEC patients with a high HOX score, those with a low HOX score had tumors with more abundant immune and stromal components, and therefore had stronger immune function and better prognosis.

The use of immune checkpoint inhibitors (ICI) such as programmed death 1 (PD-1) and cytotoxic T lymphocyte-associated antigen-4 (CTLA-4) has become one of the most promising approaches in the field of cancer therapy. To verify how the HOX score relates to responses to anticancer drugs, we used Tumor Immune Dysfunction and Exclusion (TIDE) to predict the therapeutic effect of ICI based on pretreatment tumor profiles [28]. The TIDE score was significantly higher in the low HOX score group than in the high HOX score group, indicating that tumors in patients in the low HOX score group were more likely to experience immune escape and show less therapeutic effects in response to ICI (Fig. 4c). Consistent with this result, TIDE scores were also associated with prognosis. Patients with a high TIDE score had a distinctly better prognosis than those with low TIDE scores (Fig. 4d). We then determined overall survival among UCEC patients combined with TIDE and HOX scores, and found that patients with a combination of high TIDE and low HOX score exhibited a significant prognostic advantage; Patients with a combination of low TIDE and high HOX score had worst prognosis (Fig. 4e). We further analyzed targeted immune dysfunction. Kaplan–Meier curves revealed longer overall survival in the H-dysfunction + HOX score group (Fig. 4f). Survival analysis combining immune dysfunction with HOX scores showed the highest overall survival in the H-dysfunction and low HOX score groups, and lowest overall survival in the L-dysfunction and high HOX score groups (Fig. 4g). Therefore, regardless of the TIDE and immune dysfunction scores, patients in the low HOX score group consistently had better survival than those in the high HOX score group, indicating the value of the HOX score in predicting the therapeutic effect of ICI.

HOX score can be used as a predictor of the therapeutic effect of anti-tumor drugs

Anti-tumor drugs are the main treatment for advanced tumors and after surgery. They have a killing effect on cells in multiple division stages (including the G0 stage), damaging DNA [29] or inhibiting enzymes that facilitate DNA replication [30]. To explore the differences in the treatment effects of anti-tumor agents between the different HOX score groups, we first compared the half maximal inhibitory concentration (IC₅₀) differences of four anti-tumor drugs, namely cyclopamine, foretinib, fedratinib and pazopanib, between different HOX score groups; we found that the IC₅₀ was higher in the low-score group than in the high-score group (Figs. 4h-i and S9f-g). Further, the correlation between IC₅₀ values and HOX scores was negative (Figs. 4j-k and S9h-i). Next, we compared IC₅₀ values of several other drugs, including rapamycin, doxorubicin, thapsigargin and nilotinib, with HOX scores. The IC₅₀ values of drugs in the low HOX-score group were higher than of those in the high HOX-score group (Figs. S9j-l and S10a). These results indicate that patients with UCEC with a high HOX score have a better therapeutic response to anti-tumor drugs.

The cell-type specific regulation of HOX gene in epithelial cell of UCEC

To further explore cellular diversity and HOX gene expression in endometrial tissue, the six subjects enrolled were divided into a UCEC group (three patients with UCEC receiving surgery or other treatments) and a non-UCEC group (three subjects who underwent hysterectomy) (Table 1). Age distribution, gestation, and pregnancy were not different between the groups. We employed scRNA-seq technology. This analysis covered one normal and one tumorous tissue samples. Our results, obtained through clustering analysis, identified six main cell types within the tissue, with epithelial cells being the most dominant (Fig. 5a). We then investigated the differences in HOX gene expression among these cell types. Notably, several genes, including HOXA 3, HOXA 4, HOXA 7, HOXA 9, and HOXA 10, showed high expression levels in epithelial cells (Fig. S10b). The most significant expression differences were observed for HOXA 3, HOXA 4, HOXA 7, HOXA 9, and HOXA 10 (Fig. 5b). Subsequently, we analyzed the expression of these five HOX genes in epithelial cells between the normal and tumor tissue. The UMAP scatter diagram revealed that in epithelial cells, the expression of HOXA 3, HOXA 4, HOXA 7, HOXA 9, and HOXA 10 were expressed more in tumor tissues (Figs. 5c-f and S10c) than in normal tissues (Figs. S10d-h). Further analysis of the HOX score and gene expression differences for HOX genes in epithelial cells indicated that both were significantly higher in the tumor tissue compared to that in normal tissue (Figs. 5g-j and S10i-j). Finally, we explored the relationship between these four HOX genes and various biological pathways. We discovered a positive correlation between HOXA 3 and the Gene Ontology Biological Process (GOBP) for the transforming growth factor β product, between HOXA 4 and the GOBP for the positive regulation of oxidative stress induced cell death, between HOXA 7 and the GOBP for the negative regulation of nucleotide metabolic process, between HOXA 9 and the GOBP for the positive regulation of canonical Wnt signaling pathway, and between HOXA 10 and the GOBP for the mesenchymal cell apoptotic process (Figs. 5k-n and S10k). These findings suggest that tumor occurrence and development may be linked to cellular signaling pathway dysfunctions caused by HOX gene expression.

Table 1.

Comparison of basic data from eligible selected UCEC patients and controls. Including age, gestation, pregnancy, tumor size, histology, histological grading, Bokhman typing, FIGO staging, lymph node metastasis and treatment method

Patient		Ca1	Ca2	Ca3	N1	N2	N3
Characteristic	Age	58	56	67	46	54	61
	Gestation	4	3	1	0	5	3
	Pregnancy	3	1	1	0	3	2
	Tumor size (cm)	2.5*2.1*1.3	1.5*1.0*0.5	3.0*2.0*1.6	-	-	-
	Histology	Adenocarcinoma	Adenocarcinoma	Adenocarcinoma	-	-	-
	Histological grading	G1	G1	G2	-	-	-
	Bokhman typing	I	I	I	-	-	-
	FIGO staging	II	IA	IIIC1	-	-	-
	Lymph node metastasis	negative	negative	positive	-	-	-
	Treatment method	Surgery Radiotherapy	surgery	Surgery Chemotherapy Radiotherapy	-	-	-

a: Patients in the experimental group were annotated as Ga1, Ca2 and Ca3, respectively. Patients in the control group were labeled as N1, N2 and N3, respectively

b: Histological grading is divided into four levels, G1, G2, G3 and G4

c: Bokhman Classification divides UCEC into two types according to the clinical pathological characteristics and prognosis. Type I is oestrogen-dependent, and the type II is non-oestrogen-dependent

d: International Federation of Gynecology and Obstetrics (FIGO)

Fig. 5.

scRNA-seq of endometrial tissues. a UMAP graph showing the expression of the six cell types, including epithelial cells, endothelial cells, macrophages, lymphocytes, fibroblasts, and smooth muscle cells between normal and tumor samples. b Differences in the expression of HOX genes in the six cell types. The size of the dots represents the percentage of cells expressing the gene. The color of the dots represents the average expression level; red indicates high expression, whereas blue indicates low expression. c-f UMAP scatter diagram exhibiting the expression of HOXA 3, HOXA 7, HOXA 9, and HOXA 10 in epithelial cells in tumor samples. g Differences of HOX score in epithelial cells between normal and tumor tissues (P < 0.05, Log-rank test) h-j Differences in the expression of HOXA 3, HOXA9 and HOXA 10 in epithelial cells between normal and tumor tissues (P < 0.05, Log-rank test). k-n Correlation between the HOXA 3, HOXA 7, HOXA 9, HOXA 10, and Gene Ontology Biological Process

Effect of HOX gene on CAFs in UCEC and construction of a prognostic model

CAFs are one of the most abundant cell types in the TMEs. To further understand the role of CAFs in UCEC, the fibroblasts were examined in normal endometrial tissue and tumor tissue by scRNA-seq technology and found the higher HOX score in tumor tissue than in the normal group (Fig. 6a). And The COL1A2 gene, as the marker gene for fibroblasts, was presented more higher expression in the tumor tissue (Fig. 6b). The Correlation analysis showed a positive association between HOX score and COL1A2 gene (Fig. S10l). Next, we divided the samples derived from the TCGA database into high and low CAF score groups and found that patients in the high score group had significantly better survival than those in the low score group (Fig. 6c). Therefore, we used immune-related genes concentrated in UCEC to identify the key modules associated with CAFs using Weighted correlation network analysis (WGCNA). After constructing a similar clustering module through dynamic hybrid cutting, nine modules were identified. The Pearson’s correlation coefficients between ME and CAFs of each module were calculated. The brown module showed the highest positive correlation with CAFs (Fig. 6d). Subsequently, 44 genes intersecting HOX-related genes and Brown module-related genes were examined (Fig. 6e). To infer biologically interpretable results, using Metascape's functional enrichment analysis capability, the several most significantly enriched ontology terms were combined to annotate the putative biological roles of the 44 intersecting genes, such as angiogenesis, Protein Interaction Database integrin β1 pathway (PID integrin 1 pathway), cytokine signaling in lmmune system, regulation of synapse organization and calmodulin induced events (Fig. S11a). To explore the relationship between biological behaviors and intersecting gene expression in UCEC, we performed GO and KEGG enrichment analyses of the 44 intersecting genes. KEGG analysis revealed that these genes were enriched in multiple pathways, including substrate-dependent cell migration, regulation of cell substrate adhesion, and negative regulation of cell matrix adhesion (Fig. 6f). GO enrichment analysis showed that the intersecting genes were enriched in extracellular matrix structural constituents, basement membrane, wound healing, and blood coagulation (Fig. S11b). These pathways play a regulatory role in tumors, further illustrating the role of the 44 intersecting genes in the formation and development of tumors.

Fig. 6.

Combination of CAFs with HOX scores to establish a risk prognostic model. a Differences in the HOX score in fibroblasts between normal and tumor tissues (P < 0.05, Log-rank test). b Differences in the expression of COL1A2 gene in fibroblasts between normal and tumor tissues (P < 0.05, Log-rank test). c The survival curves of different CAF groups were estimated using the Kaplan–Meier plotter. (P < 0.001, Log-rank test). d The heatmap profiling the correlations between module genes, CAFs, and stromal score. e In the Venn diagram, 44 intersecting genes are shown between the HOX-related and brown module genes. f KEGG functional enrichment analysis of 44 intersecting genes. g ROC curve based on the prognostic risk model. h The correlation between the HOX score and CAF&HOX Score. i Analysis of the correlation of CAF scores

Next, based on results of the Cox proportional hazards model (Cox) analysis, 7 of the 44 intersecting genes were screened and found to be significantly associated with OS of patients with endometrial carcinoma (Fig. S11c). A prognostic risk scoring model was established using Least absolute shrinkage and selection operator (LASSO) regression analysis, and 10-fold cross-validation was used to train the model. According to lambda 1se, 0.01 was determined as an appropriate λ value (Fig. S11d). Finally, six non-zero coefficient UCEC genes (IL17RE, MAPRE2, DCLK2, FIBIN, FBXO17, and NCAM1) were obtained and used in survival analysis to verify their prognostic value and to further guide clinical treatment (Fig. S11e). To validate the predictive power of this prognostic risk model, we used receiver operating characteristic (ROC) curves. The results showed an Area Under Curve (AUC) of 0.699 for the risk score (Fig. 6g). The AUCs for 1-, 3-, and 5-year OS in the UCEC cohort were 0.699, 0.684, and 0.652, respectively (Fig. S11f). These data revealed the accuracy of the CAF/HOX score as a risk model for UCEC prognosis.

To verify the relationship between HOX scores and CAFs, we performed a correlation analysis. Scatter plots showed positive correlations between HOX scores and CAF expression (Fig. 6h). Finally, to assess the potential relationship between HOX scores and scores representing six functional states, including the CAF score, TIDE, stromal score, immune score, ESTIMATE score, and tumor purity, we conducted a series of Pearson correlation analyses. The results showed that the HOX score was positively associated with TIDE and tumor purity, but a negative correlation was observed between HOX scores and stromal, immune, and ESTIMATE scores (Fig. 6i). The results showed that HOX scores were related to the formation of CAFs in some ways and created a favorable environment for tumor development.

Overall, the prognostic risk model established by combining HOX score with CAFs appears to be a more complete and accurate scheme that can better guide the clinical treatment of individual patients in the future.

HOX gene upregulated in endometrial tumor tissues compared to normal

To further validate whether expression of the HOX gene set is generalized at the molecular level in the tissues of patients with UCEC, hematoxylin and eosin (H&E) staining, RT-qPCR and IHC were performed on normal endometrial tissues and tumor tissues. We performed H&E staining and RT-qPCR for three patients with UCEC (including endometrial tumor and paracancerous tissues) and three healthy individuals (normal endometrial tissues) (Table 1). H&E staining for the normal endometrial tissue confirmed endometrial histologic characteristics (Figs. S11g-i). H&E staining revealed an increased cell number, disordered arrangement, and increased mitotic figures in the tumor tissues, which was consistent with the diagnosis of UCEC (Figs. S12a-c). IHC revealed PTEN expression in normal endometrial cells (Fig. S12d). While the absence of PTEN expression in the tumor cells (Figs. S12e). These experiments confirmed histological differences between normal endometrial and endometrial tumor samples. Next, we performed RT-qPCR on the six samples. In the experimental group, expression of HOXB 5, HOXB 7, HOXB 9, and HOXB 13 genes was generally higher than that in the normal endometrial tissue (P = 0.0024, P = 0.002, P = 0.0284, and P = 0.0029, respectively) (Figs. 7a-d). We also observed that the expression of HOXB 8, HOXB 9, and HOXB 13 genes in UCEC tissues was generally higher than that in normal tissues (P = 0.0106, P = 0.0041, and P = 0.0373, respectively) (Figs. 7e-g). Additionally, HOXB 8 expression was approximately 20 times higher in tumor tissues than in paracancerous tissues (P = 0.0017) (Fig. 7h). Next, IHC for fifteen normal endometrial tissues and fifteen tumor tissues was also performed using different HOX gene antibodies. The number of samples used to test for each gene was six samples (three normal endometrial tissues and three endometrial tumor tissues). All subjects underwent hysterectomy. Age distribution was not different between the groups (Table S1-5). We found that compare to the normal tissues, HOXA 3, HOXA 4, HOXA 7, HOXA 9, and HOXA 10 showed stronger expression on the cytoplasm and nucleus in tumor tissues (Figs. 7i-l and S12f-k). The bar chart showed that in the tumor tissues, the percentage contribution of positive for HOX genes were higher than that in the normal group (Figs. 7m-p and S12l). These results showed that HOX expression levels gradually increased from normal tissues to tumor tissues, which further verified that HOX can act as a pro-oncogenic gene to regulate the occurrence and development of UCEC.

Fig. 7.

The expression of HOX genes was verified using RT-qPCR and IHC. a-h Levels of mRNA expression of HOX genes. The mRNA expression levels of HOXB 5, HOXB 7, HOXB 8, HOXB 9, and HOXB 13 in endometrial, paracancerous, and normal endometrial tissues were measured using RT-qPCR. GAPDH was used as a loading control. Data are presented as the mean ± SD of triplicate independent experiments. P values were obtained using the Student’s t-test (a-d: control group: normal endometrial tissue; experimental group: paracancerous tissue. e–g control group: normal endometrial tissue; experimental group: endometrial tumor tissue. h control group: paracancerous tissue; experimental group: endometrial tumor tissue. *, P < 0.05; **, P < 0.01; ***, P < 0.001). i-l IHC for detecting the expression of HOX genes in tumor tissues (bar = 50 μm). Each experiment was performed in triplicate. m-p Percentage of positive staining for HOX gene (control group: normal endometrial tissue; experimental group: endometrial tumor tissue. *, P < 0.05; **, P < 0.01; ***, P < 0.001)

Discussion

In this study, we analyzed the correlation of all HOX genes with UCEC. We then utilized unsupervised clustering and principal component analysis to construct HOX scores and combined them with clinical features and multiple biological pathways to identify a positive correlation between HOX score and UCEC. Using scRNA-seq, we identified six major cell types in our endometrial tumor samples, namely epithelial cells, endothelial cells, macrophages, lymphocytes, fibroblasts, and smooth muscle cells. We selected epithelial cells and fibroblasts with the higher HOX expression. Next, we examined the differential expression of HOX within epithelial cells between normal and tumor tissues, and explored the potential relationship between HOX genes and signaling pathways within epithelial cells. In addition, we compared the difference in the HOX score in CAFs between normal and tumor tissues. We explored the correlation of HOX scores and CAFs, and established a risk prognosis model based on the interaction between HOX genes and CAFs. We also found that the HOX gene can regulate some tumor-associated signaling pathways in CAFs. Finally, we collected thirty-six samples (eighteen normal and eighteen endometrial tumor tissues), and tested the expression of the HOX gene using RT-qPCR and IHC.

HOX genes, central regulators in living organisms, yielded transcription factors that exerted a paradoxical effect on tumorigenesis [4]. Their expression profiles had been implicated in both restraining the metastatic potential of cancer cells [9] and fueling tumor proliferation, migration, and invasion [10]. These genes were widely expressed across various solid tumor types, encompassing colorectal, lung, and squamous cell carcinomas [9, 16, 31]. Yet, the investigation into the interplay between HOX gene expression and immune cell infiltration within the TME remained under-explored. A precedent study highlighted a significant correlation between elevated HOXC6 expression and the expression of chemokines that recruit T cells, the degree of immune cell infiltration, and the presence of immune checkpoint markers in colorectal cancer [16]. Such findings underscored the indispensable role of HOX expression in inflammation, immunity, and tumor progression. Nevertheless, the literature on the association between HOX expression and UCEC was scant. The influence of the TME on tumor dynamics had been firmly established, with evidence suggesting that infiltrating immune cells can identify and eliminate neoplastic cells within the TME [11]. For instance, the activation of natural killer T (NKT) cells by α-galactosylceramide was shown to alter the composition of M1 macrophages and effector Th1 cells in secondary lymphoid organs and the TME, effectively curbing tumor growth [32]. However, the majority of contemporary studies concentrate on singular cell types within the TME, leaving the collective impact of multiple HOX genes on TME infiltration patterns largely unappreciated. Given the ambiguity surrounding the role of HOX expression in the etiology and evolution of UCEC, coupled with the dearth of understanding regarding HOX-mediated immune infiltration, a holistic evaluation of the prognostic utility and functional annotations of HOX genes in UCEC was warranted. Deciphering the function of HOX genes in immune cell infiltration provided a foundational framework for advancing our comprehension of the dialogue between HOX genes and antitumor immune responses, paving the way for tailored therapeutic interventions for individuals diagnosed with UCEC.

In this comprehensive analysis, we scrutinized the clinical data and transcriptomic profiles of 401 UCEC patients, focusing on the expression patterns of 39 HOX genes. Our findings delineated three distinct immune phenotypes within the cohort. Cluster A emerged as an immune-inflamed subtype, characterized by robust adaptive immune activation. Conversely, cluster B was typified by immune suppression, whereas cluster C presented as an immune-excluded phenotype, marked by heightened innate immune responses and stromal activation. With regard to the immune inflammatory phenotypes, often denoted as "hot" tumors, we observed that CD8 + T cells could infiltrate the tumor parenchyma, effectively engaging with neoplastic cells at the core. In contrast, the immune-excluded and immune-desert phenotypes, categorized as "non-inflamed" tumors, displayed a higher density of CD8 + T cells confined to the tumor periphery, rather than penetrating the tumor's substance. Particularly in immune-desert phenotypes, CD8 + T cell infiltration was virtually absent. Our data further revealed that Cluster C exhibited a pronounced stromal activation profile, encompassing upregulated pathways involved in angiogenesis, epithelial-mesenchymal transition (EMT), and transforming growth factor β (TGF-β) signaling. By integrating the TME cell-infiltrating features and survival outcomes across the clusters, we validated the accuracy of our immune phenotype classification, which was contingent upon distinct HOX expression signatures. The analysis of TME cellular infiltration patterns associated with varying HOX expression revealed that Cluster C was associated with a poorer prognosis, attributed to the suppression of immune cell activity. Moreover, a comparative assessment of HOX expression levels between endometrial carcinoma tissues and healthy counterparts showcased a degree of heterogeneity. This suggested that an imbalance in HOX gene expression may serve as a critical determinant in the etiology and progression of UCEC, and the intricate relationship between HOX gene expression, immune phenotyping and the tumor microenvironment in UCEC underscored the potential of HOX genes as biomarkers and therapeutic targets for personalized cancer care.

In our investigation, we uncovered that the disparities in mRNA transcriptomes, delineated by distinct HOX expression patterns, were profoundly intertwined with HOX genes themselves and immune-related biological processes. These DEGs, deemed as HOX-related signature genes, played a pivotal role in defining two genomic subtypes that bore significant relevance to matrix activation and immune reactions. These findings resonated with the clustering outcomes based on HOX expression, reinforcing the notion that HOX expression profoundly impacts the configuration of the TME. From these observations, we deduced that an exhaustive appraisal of HOX expression patterns could significantly enrich our insights into the TME's cellular infiltration traits. However, recognizing the limitations of our current analytical approach in precisely forecasting individual patient expression patterns, and taking into account the intrinsic heterogeneity of HOX gene expression, the imperative need for quantifying HOX patterns within individual tumors became evident. To address this exigency, we devised a scoring system dedicated to evaluating the HOX expression profiles in patients diagnosed with endometrial carcinoma. This system revealed that the immune-excluded phenotype was associated with a higher HOX score, while the immune-inflamed phenotype was characterized by a lower HOX score. This disparity suggested that the HOX score serves as a dependable and potent instrument for comprehensively gauging HOX expression patterns within individual tumors. Furthermore, it could facilitate the determination of TME infiltration patterns, also recognized as tumor immunophenotypes. Comparative analyses of mRNA transcriptomes amongst patients with variant HOX scores unveiled substantial correlations with biological pathways intimately linked to HOX genes and immunity. Importantly, the HOX score demonstrated superior evaluative capabilities concerning patient clinical attributes, encompassing tumor differentiation grade, mutational load, age, survival status, and overall clinical prognosis. The correlation between the HOX score, tumor stage, and prognosis was strikingly significant. Integrated analyses corroborated the potential of the HOX score to operate as an autonomous prognostic biomarker for endometrial carcinoma. This finding underscored its utility in clinical decision-making, offering a promising avenue for personalized medicine in UCEC management.

TMB and MSI are increasingly recognized as pivotal biomarkers for predicting the efficacy of immunotherapy in cancer patients [33, 34]. Our analysis disclosed a striking positive correlation between the HOX score and TMB, suggesting a potential synergistic effect on patient outcomes. When we cross-referenced the HOX score with TMB, we observed that patients bearing high HOX and TMB scores enjoyed the most favorable survival outcomes, whereas those with high HOX scores paired with low TMB experienced the least survival benefit. Contrary to initial expectations, patients exhibiting low HOX scores in conjunction with high TMB levels did not demonstrate the most significant survival advantage. This unexpected result could potentially be attributed to the limited sample size in our study. Nevertheless, our data consistently revealed that patients in the low HOX score cohort fared better prognostically, irrespective of their tumor burden status. The integration of HOX scores with TMB in survival analysis offered more detailed and precise prognostic insights. Moreover, we integrated the HOX score with TIDE and physical fitness levels, demonstrating that the HOX score could serve as a novel predictive marker for the effectiveness of ICIs. Our findings underscored the significant role of HOX expression in the development of distinct matrix environments and immune landscapes within the TME, alongside its impact on immune cell infiltration. Immunomodulatory molecules like Programmed Death Receptor-1 (PD-1) are key regulators of immune responses, typically expressed on activated T cells, NK cells, B-cells, and subsets of myeloid cells [35]. Dysregulation of checkpoint gene expression and function is a cardinal mechanism implicated in various diseases [36, 37]. Our results indicated that the HOX score was intricately linked to disrupted immune checkpoints, potentially enhancing responsiveness to immunotherapeutic interventions. The amalgamation of a HOX gene signature with a repertoire of integrated biomarkers, encompassing mutation load, neoantigen load, PD-L1 expression, stromal and immune TME characteristics, and MSI status, emerges as a potent predictive tool for immunotherapy efficacy. This integrative approach holds considerable promise for guiding personalized therapeutic strategies in future clinical practice, optimizing patient selection and treatment outcomes in the realm of immunotherapy.

The aberrant expression of HOX genes can lead to dysregulation in certain tumor-related signaling pathways, which is one of the decisive factors contributing to the development of UCEC. Among different HOX patterns, repair pathways such as nucleotide excision repair, homologous recombination, and mismatch repair are significantly upregulated in cluster C compared to cluster A and B. These DNA damage repair pathways can be persistently activated when nucleotide sequences mutate, enhancing repair capacity and regulating the cell cycle to maintain genomic stability and integrity [38, 39]. The p53 signaling pathway, as a protective response, can be triggered by DNA damage, oxidative stress, and other stimuli. It activates genes involved in DNA repair, such as XRCC1 and APEX1, which encode proteins that help repair DNA damage and prevent further genomic mutations [40, 41]. Through scRNA-seq, we find that HOX genes are highly expressed in UCEC epithelial cells and fibroblasts, and their expression regulates typical pathways associated with tumor invasion and metastasis in these two cell types. In epithelial cells, the expression of HOX genes is significantly positively correlated with processes such as TGF-β, mesenchymal cell apoptosis, and Wnt signaling pathways. Studies have shown that due to metabolic imbalance in tumor cells, reactive oxygen species can activate TGF-β, inducing EMT and increasing malignancy [42]. Mesenchymal cells can induce tumor cell death and inhibit the migratory activity of tumor cells [43]. However, under the regulation of HOX-mediated mesenchymal cell apoptosis, the number of tumor cells can increase, weakening their inhibitory effect on invasion. Abnormal activation of the Wnt signaling pathway may lead to an expansion of the stem cell pool and the generation of cancer stem cells, promoting tumor formation and recurrence. In fibroblasts, HOX-induced modulation of cell–matrix adhesion pathways can alter the mobility and invasiveness of tumor cells. For example, reduced expression of E-cadherin in tumor cells decreases intercellular adhesion, promoting EMT and conferring stronger migratory capabilities [44]. The interaction between tumor cells and vascular endothelial cell adhesion molecules induces angiogenesis, supplying necessary nutrients and stimulating proliferation and invasion of tumor cells [45]. During the activation of wound healing pathways, certain protein factors such as vascular endothelial growth factor (VEGF) and TGF-β can be secreted [46, 47]. These factors may increase the malignancy of tumors by inducing EMT and other mechanisms [42]. Collectively, these results posited that HOX genes may serve as target genes within specific signaling pathways resident in epithelial cells and fibroblasts, culminating in the development of UCEC. This insight underscores the potential of targeting HOX genes as a therapeutic strategy to disrupt the signaling pathways that underpin UCEC progression.

CAFs represent a crucial element within the TME, playing a multifaceted role in cancer progression. These cells secrete a plethora of growth factors, inflammatory mediators, and extracellular matrix components [48], which collectively foster tumor cell proliferation, confer resistance to therapy, and orchestrate immune evasion [49]. In our investigative approach, leveraging scRNA-seq technology, we elucidated the prominent expression of HOX genes within CAFs. This finding was further substantiated by a positive correlation between the HOX scores and the presence of CAFs. To decipher the genetic architecture underlying this association, we employed WGCNA to pinpoint modules of co-expressed genes that were intimately tied to CAF activity. By integrating HOX-related genes with these module genes, we narrowed down to a set of intersecting, prognostic-relevant genes. The development of a risk model based on these genes demonstrated a commendable predictive capability for estimating the 1-, 3-, and 5-year overall survival rates of patients. This validation not only underscored the precision and clinical significance of the HOX score as an independent biomarker for survival prediction but also hinted at its potential as a prognostic tool in diverse cancer types.

To complement the theoretical and computational analyses conducted in our study, we reinforced our findings with empirical validation through RT-qPCR and IHC. The RT-qPCR results unequivocally demonstrated that the expression levels of HOX genes were significantly elevated in endometrial tumor tissues compared to the corresponding normal tissue controls. This quantitative molecular evidence provided a robust confirmation of our bioinformatic predictions, substantiating the differential expression of HOX genes in cancerous versus healthy endometrial tissue. Furthermore, the application of IHC corroborated our molecular findings again. IHC staining revealed a markedly increased positive rate of HOX expression within tumor tissues, visually depicting the enhanced presence of HOX proteins in neoplastic cells. This qualitative data, in conjunction with the quantitative RT-qPCR results, served to strengthen the validity of our theoretical studies. Together, these experimental approaches—RT-qPCR for gene expression quantification and IHC for protein localization—yielded convergent evidence supporting the central role of HOX genes in UCEC. The integration of these methodologies allowed us to establish a comprehensive understanding of HOX gene expression patterns, confirming their upregulation in tumor samples and reinforcing the clinical relevance of our computational discoveries. These findings underscored the importance of HOX genes in the pathogenesis of UCEC, creating the way for potential therapeutic strategies targeting these critical regulatory molecules. The combination of theoretical studies with experimental validation provided a solid foundation for future translational research aimed at developing targeted therapies for patients with UCEC.

Our research substantiated that the landscape of HOX gene expression orchestrates extensive mechanisms within the UCEC TME, and a comprehensive appraisal of HOX expression patterns within individual tumors enhanced our understanding of cellular infiltration characteristics within the TME. Concurrently, through the development of a HOX scoring system to elucidate the role of HOX genes in cancer, we offered a holistic evaluation of HOX profiles and TME cellular infiltration features in individual patients, facilitating the identification of tumor immune phenotypes and guiding clinical practice more effectively. Furthermore, within the epithelial cells and CAFs of UCEC, HOX genes served as target genes for a multitude of tumor-related signaling pathways. This positioning potentially made them critical molecular markers for the advancement of future precision medicine approaches.

In summary, the HOX score demonstrated independent prognostic power for UCEC, potentially serving as a predictive biomarker for patient survival. Our comprehensive analysis of HOX provided a novel and insightful perspective to the field of UCEC research, contributing to a better understanding of its epigenetics and immunotherapeutic landscape.

Methods

Dataset source and preprocessing

Public gene expression data and complete clinical annotation were retrieved from TCGA and ICGC. We selected patients with available survival information for further evaluation and excluded the others. In total, 401 eligible tumor samples from the TCGA-UCEC cohort and 128 eligible tumor samples from the ICGC-UCEC cohort were included for subsequent analysis. All samples were used in the mutation analysis. The samples from the TCGA-UCEC cohort were used in unsupervised clustering and consensus clustering analyses. RNA sequencing data, expressed as Fragments Per Kilobase of exon model per Million mapped fragments (FPKM) values of gene expression were downloaded from the Genomic Data Commons (GDC, https://portal.gdc.cancer.gov/) using the R packages TCGA and ICGC biolinks. These packages were specifically developed for integrative analysis with GDC data [50]. Additionally, somatic mutation data were acquired from the TCGA database. The dataset from TCGA-UCEC was downloaded for CNV analysis. Data were analyzed using R (version 4.2.3) and R Bio conductor packages.

Unsupervised clustering and consensus clustering analysis

We extracted a total of 39 HOX related genes from the TCGA and ICGC datasets to identify distinct modification patterns mediated by these genes. The analyzed genes included HOXA 1, HOXA 2, HOXA 3, HOXA 4, HOXA 5, HOXA 6, HOXA 7, HOXA 9, HOXA 10, HOXA 11, HOXA 13, HOXB 1, HOXB 2, HOXB 3, HOXB 4, HOXB 5, HOXB 6, HOXB 7, HOXB 8, HOXB 9, HOXB 13, HOXC 4, HOXC 5, HOXC 6, HOXC 8, HOXC 9, HOXC 10, HOXC 11, HOXC 12, HOXC 13, HOXD 1, HOXD 3, HOXD 4, HOXD 8, HOXD 9, HOXD 10, HOXD 11, HOXD 12, and HOXD 13. Unsupervised clustering analysis was applied to identify distinct patterns of HOX-related genes based on their expression. This analysis helped classify patients for further analysis. The number of clusters and their stability were determined using a consensus clustering algorithm [51]. The ConsensuClusterPlus package was used to perform these analyses, with 1,000 repetitions conducted to ensure the stability of the classifications [52].

Enrichment gene set variation analysis with functional annotation

We conducted GSVA enrichment analysis using the “GSVA” R package to explore differences in biological processes among HOX patterns. GSVA is a non-parametric and unsupervised method widely used to estimate variation in pathway activity and biological processes across samples in an expression dataset [53]. For this analysis, gene sets titled “c2.cp.kegg.v6.2.-symbols” were sourced from the MSigDB database. Statistical significance was determined with an adjusted P-value of less than 0.05. The clusterProfiler R package was used to perform functional annotation for HOX-related genes, with the cutoff value of false discovery rate (FDR) set at less than 0.05.

TME cell infiltration estimation

To quantify the relative abundance of each cell type infiltrating the UCEC TME, we employed the single-sample gene-set enrichment analysis (ssGSEA) algorithm. The gene set used to mark each type of TME infiltrating immune cell was obtained from the study by Charoentong, which includes various human immune cell subtypes including activated CD8⁺ T cells, activated dendritic cells, macrophages, natural killer T cells, and regulatory T cells, among others [54, 55]. The ssGSEA-derived enrichment scores were used to represent the relative abundance of each TME infiltrating cell in each sample. Additionally, we calculated tumor purity, stromal, immune, and ESTIMATE scores for each sample using the ESTIMATE algorithm. This algorithm assesses the tumor component in each sample [56]. Subgroup comparisons were analyzed using the “limma” R package, allowing for a detailed examination of score variations.

Identification of DEGs between distinct HOX gene phenotypes

To explore the diversity among HOX genes-related phenotypes, we classified patients into three distinct groups based on HOX gene expression. We used the empirical Bayesian approach provided by the “limma” R package to identify DEGs between different patterns [57]. The criteria for determining significant DEGs were set with an adjusted P-value threshold of less than 0.001. Further analysis on the pathways associated with these DEGs was conducted using GO and KEGG databases.

Design and generation of HOX gene score

To evaluate individual patient HOX patterns in UCEC, we developed a scoring system called the HOX score. The process to establish the HOX gene signature involved several key steps:

Initially, DEGs identified from various HOX clusters were normalized across all UCEC samples. Overlapping genes were then extracted. We used an unsupervised clustering method to analyze these overlapping DEGs, classifying the patients into several groups for more detailed analysis. A consensus clustering algorithm was employed to determine the number of gene clusters and assess their stability.

Next, each gene within the signature underwent a prognostic evaluation using a univariate Cox regression model. Genes showing significant prognostic impact were selected for further analysis. We then applied PCA to construct the HOX-relevant gene signature. The first two principal components (PC1 and PC2) were used as the signature scores as follows:

$$HOX\mathrm{score}=\mathrm{\Sigma }(\mathrm{PC}{1}_{\mathrm{i}}+\mathrm{PC}{2}_{\mathrm{i}})$$

where i is the expression of HOX cluster-related genes.

Correlation between the HOX gene score and other related biological processes or clinical features

We performed a correlation analysis to further reveal the association between the HOX gene signature and some related clinical features, including (1) immune-checkpoints such as CD 44, CD 96, TIGIIT and others; (2) microsatellite status, such as MSS, MSI-L, and MSI-L; (3) TMB; (4) TIDE; (5) age; (6) tumor grade; (7) several drugs including cyclopamine, fedratinib, pazopanib, foretinib, and others; (8) ESTIMATES score; (9) immune score; (10) stromal score; (11) tumor purity; and (12) IPS.

Weighted gene co-expression network analysis and prognostic model construction

WGCNA is a method established for studying biological networks and disease correlations. We used the R package “WGCNA 16” to perform WGCNA, aiming to identify key modules related to CAFs using immuno-related genes within the TCGA-UCEC dataset. To identify the genes linked to prognosis, we employed forest plots for dichotomous variables and LASSO regression. The predictive value of the constructed prognostic model was evaluated using ROC curves.

Tissue selection

Endometrial tissues were collected from eighteen patients with endometrial tumors and eighteen with normal endometrial conditions at the first hospital of Shanxi Medical University. All patients underwent hysterectomy. The selection of the tissues was based on histopathological analyses performed by a pathologist with over two years of clinical experience. Each tissue sample measured 1.0 cm in length, 0.5 cm in width, and 0.5 cm in thickness. Paracancerous tissues were excised 1 cm from the tumor margin. Six tissues were subjected to hematoxylin and eosin staining and RT-qPCR. IHC was performed on paraffin-embedded thirty tissue samples. One tumor tissue and one normal tissue were selected from the collected samples for scRNA-seq.

scRNA-seq

Raw reads from human endometrial cell fastq files were processed using the Cell Ranger Software Suite (10 × Genomics Cell Ranger 4.0.0), with refdata-gex-GRCh38-2020-A as the reference for mapping reads to the human genome (GRCh38/hg38). This processing generated unique molecular identifier (UMI) matrices [58]. The outputs from Cell Ranger were then imported into Seurat using the ‘Read10X’ function [59]. Quality control measures included removing cells with UMI counts in the upper 10% for each sample to eliminate potential outliers. Additionally, cells with fewer than 500 UMIs detected or those with more than 40% mitochondrial UMIs were filtered out. Genes expressed in fewer than 1 cell were also removed. Quality control was managed using the ‘Seurat’ package (version 4.0). For normalization, a global scaling method called LogNormalize was applied to normalize gene expression measurements across cells, considering both characteristic and total expression levels. To reduce the computational load and minimize noise, PCA was employed for initial dimensionality reduction. Visualization of the annotated cellular information within the dataset was achieved using uniform manifold approximation and projection.

Immunohistochemistry

Proportion of antibody dilution: HOXA 3 (1: 200) (ab230879, Abcam lnc, Cambridge, MA, USA), HOXA 4 (1: 200) (ab131049, Abcam lnc, Cambridge, MA, USA), HOXA 7 (1: 200) (ab211521, Abcam lnc, Cambridge, MA, USA), HOXA 9 (1: 500) (ab140631, Abcam lnc, Cambridge, MA, USA), HOXA 10 (1: 200) (ab191470, Abcam lnc, Cambridge, MA, USA). 4-μm-thick tissue sections were dewaxed and dehydrated by xylene and a graded ethanol series, respectively. Then, endogenous peroxidase activity was blocked by 3% H₂O₂. Sections were incubated with primary antibodies for 4 h at room temperature, followed by incubation with the corresponding second antibody. Subsequently, tissue sections were stained with 3,3’-diaminobenzidine and hematoxylin. The positive rate of IHC was analyzed by Image J.

RT-qPCR

Total RNA was extracted from tissue samples using Trizol reagent (Ambion, USA). The mRNA was then reverse transcribed into cDNA using the cDNA Synthesis SuperMix (TransGen Biotech, Beijing). PCR amplification was conducted using a PCR machine (Bioer Technology, GeneMax Tc-s-B, China) using the following cycling parameters: an initial denaturation at 95 ℃ for 5 min, followed by 40 cycles of denaturation at 95 ℃ for 10 s, and annealing at 60 ℃ for 30 s. Gene expression levels were quantified using the ΔΔCt method. The experiments were performed in triplicates to ensure reproducibility. The primers used for RT-qPCR were as follows: GAPDH forward 5'-GCTCTCTGCTCCTCCTGTTC-3', reverse 5'-ACGACCAAATCCGTTGACTC-3'; HOXB 5 forward 5'-CTCTGAGCGGCTCTTACAGG-3', reverse 5'-GCCCGGTCATATCATGGCT-3'; HOXB 6 forward 5'-AGGACTGCAGCCCGATACTA-3', reverse 5'-CGAGATTGGGTTTTAGCTTTGC-3'; HOXB 7 forward 5'-TGCGAAGCTCAGGAACTGAC-3', reverse 5'-CCAAAATTTCTCCTTTCTCCCTCC-3'; HOXB 8 forward 5'-GCTGCCATGCAAGCTTAGAC-3', reverse 5'-GTCCGGCGGCTGCTTG-3'; HOXB 9 forward 5'-GAGAGGCCGGATCAAACCAA-3', reverse 5'-GGGAGGACTGGGGGTAATCT-3'; HOXB 13 forward 5'-TAAAACGCTTTGGATTCCCCC-3', reverse 5'-CCGCTGGAGTCTGCAAAT-3'.

Statistical analysis

The correlation between TME infiltrating immune cells and the expression of HOX regulators was assessed using Spearman and distance correlation analyses. Cut-off points were established based on the correlation of the HOX score with patient survival within each dataset subgroup, using the surf R package. Survival curves for the prognostic analysis were generated using the Kaplan–Meier method. The significance of the differences was determined using a log-rank test. All statistical tests were two-sided, with significance set at P < 0.05. All data processing was performed using the R4.2.3 software.

Abbreviations

AUC

Area Under Curve

CAFs

Cancer-associated fibroblasts

CNV

Copy number variation

Cox

Cox proportional hazards model

CTL

Cytotoxic T lymphocytes

CTLA-4

Cytotoxic T lymphocyte-associated antigen-4

DEGs

Differentially expressed genes

DFI

Disease-free interval

DSS

Disease-specific survival

ESTIMATE

Estimation of STromal and Immune cells in MAlignant Tumour tissues using Expression data

FDR

False discovery rate

FPKM

Fragments Per Kilobase of exon model per Million mapped fragments

GDC

The Genomic Data Commons

GO

Gene Ontology

GSVA

Gene set variation analysis

H&E

Hematoxylin and eosin

HLA

Human leukocyte antigen

IC₅₀

Half maximal inhibitory concentration

ICGC

International Cancer Genome Consortium database

ICI

Immune checkpoint inhibitors

iDCs

Immature dendritic cells

IFN

Interferon

IHC

Immunohistochemistry

IL-1

Interleukin-10

KEGG

Kyoto Encyclopedia of Genes and Genomes

LASSO

Least absolute shrinkage and selection operator

MDSCs

Myeloid-derived suppressor cells

miRNAs

MicroRNAs

MSI

Tumor microsatellite instability

MSS

Microsatellite stability

NK

Natural killer

OS

Overall survival

PCA

Principal component analysis

PD-1

Programmed death 1

PFS

Progression-free survival

ROC

Receiver operating characteristic

PID

Protein Interaction Database

RT-qPCR

Real-time quantitative PCR

scRNA-seq

Single-cell RNA sequencing

ssGSEA

Single-sample gene-set enrichment analysis

TCGA

The Cancer Genome Atlas database

TIDE

Tumor Immune Dysfunction and Exclusion

TLs

T lymphocytes

TMB

Tumor mutation burden

TME

Tumor microenvironment

UCEC

Endometrial cancer

VEGF

Vascular endothelial growth factor

WGCNA

Weighted correlation network analysis

Authors’ contributions

SanYuan Zhang and Zhe Wang designed and supervised the study, Zhe Wang, JinPeng Li, et al., collected the data, performed all data analysis and drafted the manuscript. Zhe Wang, JiaoLin Yang, JinPeng Li, et al., performed original draft preparation, writing, review and editing. Zhe Wang provided analytical technical support. Zhe Wang, JiaoLin Yang and JinPeng Li performed experiments, including hematoxylin and eosin staining, RT-qPCR and scRNA-seq. SanYuan Zhang and Zhe Wang were responsible for the data acquisition and critical reading of the manuscript. JiaoLin Yang, JinPeng Li, YuTong Yang, HuanCheng Su, HongRui Guo, Jing Lei, SuFen Li, YaLin Wang, KaiTing Wen, Xia Li, Zhe Wang and SanYuan Zhang reviewed the manuscript. All authors have read and approved the final manuscript.

Funding

China Postdoctoral Science Foundation (2021M691995 to Z.W.), The Natural Science Foundation of Shanxi Province, China (202103021224394 to Z.W.), The Shanxi Provincial Key Research and Development Project (201803D31111 to SY.Z.).

Availability of data and materials

All data used in our study can be acquired from the cancer fenome Atlas (TCGA; https://www.genome.gov/) and the ICGC data portal (https://dcc.icgc.org/). Single-cell RNA sequencing gene expression data generated in this study has been deposited in the Annotare < EMBL-EBI database (https://www.ebi.ac.uk/fg/annotare/edit/18643/#DESIGN:PROTOCOLS). Any other data are available from the corresponding author on reasonable request. Software and resources used for analysis and plotting are described in each method section.

Declarations

Ethics approval and consent to participate

This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Ethics Committee of the First Hospital of Shanxi Medical University (No. K-K221). Informed consent was obtained from all individual participants included in the study.

Consent for publication

Patients signed informed consent regarding publishing their data, including Table 1 and Table S1-5.

Competing interests

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.