Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2019 Jul 26;11(8):1057.
doi: 10.3390/cancers11081057.

Mining of Self-Organizing Map Gene-Expression Portraits Reveals Prognostic Stratification of HPV-Positive Head and Neck Squamous Cell Carcinoma

Laura D Locati  1 , Mara S Serafini  2 , Maria F Iannò  2 , Andrea Carenzo  2 , Ester Orlandi  3 , Carlo Resteghin  1 , Stefano Cavalieri  1 , Paolo Bossi  1 , Silvana Canevari  2 , Lisa Licitra  1   4 , Loris De Cecco  5
Affiliations

Mining of Self-Organizing Map Gene-Expression Portraits Reveals Prognostic Stratification of HPV-Positive Head and Neck Squamous Cell Carcinoma

Laura D Locati et al. Cancers (Basel). .

Abstract

Patients (pts) with head and neck squamous cell carcinoma (HNSCC) have different epidemiologic, clinical, and outcome behaviors in relation to human papillomavirus (HPV) infection status, with HPV-positive patients having a 70% reduction in their risk of death. Little is known about the molecular heterogeneity in HPV-related cases. In the present study, we aim to disclose the molecular subtypes with potential biological and clinical relevance. Through a literature review, 11 studies were retrieved with a total of 346 gene-expression data points from HPV-positive HNSCC pts. Meta-analysis and self-organizing map (SOM) approaches were used to disclose relevant meta-gene portraits. Unsupervised consensus clustering provided evidence of three biological subtypes in HPV-positive HNSCC: Cl1, immune-related; Cl2, epithelial-mesenchymal transition-related; Cl3, proliferation-related. This stratification has a prognostic relevance, with Cl1 having the best outcome, Cl2 the worst, and Cl3 an intermediate survival rate. Compared to recent literature, which identified immune and keratinocyte subtypes in HPV-related HNSCC, we confirmed the former and we separated the latter into two clusters with different biological and prognostic characteristics. At present, this paper reports the largest meta-analysis of HPV-positive HNSCC studies and offers a promising molecular subtype classification. Upon further validation, this stratification could improve patient selection and pave the way for the development of a precision medicine therapeutic approach.

Keywords: HP; head and neck cancer; molecular subtypes; self-organizing map; treatment de-escalation; tumor microenvironment.

PubMed Disclaimer

Conflict of interest statement

The authors declare no conflicts of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, and in the decision to publish the results.

Figures

Figure 1
Figure 1
Human papillomavirus (HPV)-positive head and neck squamous cell carcinoma (HNSCC) tumor clusters: first-level self-organizing map (SOM) and unsupervised clustering analysis. (a) Consensus matrix heatmap imposing three clusters: Cl1 (n = 134; 39%), Cl2 (n = 104; 30%), and Cl3 (n = 108; 31%). The consensus values are reported in a range from 0 (white, samples that never cluster together) to 1 (blue, samples showing the highest clustering affinity). (b) Silhouette plot analysis. The samples are ranked based on silhouette values (S) in each cluster. The heights indicate a strong similarity of the samples within their clusters compared with the samples belonging to other clusters. The colors in the lower bar show the predicted membership by silhouette analysis; the colors correspond to the consensus clustering assignment for all samples with the exception of the seven samples with a negative number but close to 0. (c) First level of the SOM gallery of the three clusters with cluster-specific tiles highlighted. The expression patterns are translated into a color code indicating over- and under-expression in a range from red to blue spots, respectively.
Figure 2
Figure 2
Alluvial diagram. In the diagram, each of the blocks corresponds to the number of features, and the stream fields between the blocks represent changes in the composition of the different blocks. The sizes of the blocks are proportional to the number of samples. We explored the cluster membership taking into account (i) the study of origin of each sample (11 strata); (ii) the different technology platforms used for expression profiling (five strata). Study of the origin: χ2 test = 12.08, p-value = 0.913; Platform χ2 test = 5.93, p-value = 0.655.
Figure 3
Figure 3
HPV-positive HNSCC cluster similarity relationships: second-level SOM. (a) Independent component analysis of meta-gene data. Samples were distributed along the three leading independent components; the plots show the three-dimensional distribution and the projections into the component 1/component 2 (lower panel) and component 1/component 3 (upper panel) dimensions. (b) Sample correlation network. The samples are visualized by nodes connected by edges with a backbone structure linking samples with the highest correlation. The similarity between samples is represented by their reciprocal distance; closer nodes have higher similarity and distant nodes have lower similarity. (c) Neighbor-joining analysis. The sample similarities are summarized in a phylogenetic tree structure computed using Euclidean distance. The neighbor-joining (NJ) analysis visualizes “bush-like” groups of similar samples by assessing their mutual dissimilarity.
Figure 4
Figure 4
Subtype characterization by group overexpression maps. (a) The 18 × 18 map of meta-genes summarizes the expression landscapes over the three subtypes; according to this analysis, co-regulated meta-genes are located in the opposite corners of the map. (b) Detailed analysis of metagenes overexpressed in each subtype: map location (left panels) and bar plot of expression intensity (right panels). The bar plot represents the average meta-gene expression of each sample for the selected tiles.
Figure 5
Figure 5
Tumor microenvironment landscape. (a) Visualization of the immune and “other cell” infiltrates assessed by xCell. Individual patients are summarized based on two-dimensional coordinates from the t-distributed stochastic neighbor embedding (t-SNE) method. The notched boxplots show the ImmuneScores (p-value = 9.9 × 10−29) (b), keratinocytes scores (p-value = 2.03 × 10−32) (c), and stromal cell infiltrates (p-value = 6.3 × 10−18) (d) split into the three different subtypes.
Figure 6
Figure 6
Visualization of the Gene Set Enrichment Analysis (GSEA) functional analysis for each of the three clusters. The boxplots show how the gene set Z score (GSZ) values (depicted in y-axis) are distributed within each of the three clusters (Cl1, green; Cl2, blue; Cl3 red). In each row, comparisons of the GSZ score values for the two most enriched hallmark gene sets are shown: for Cl1, over-expression is shown for the “immune response” hallmark (p-value 1.09 × 10−40) and “interferon (IFN)-gamma response” hallmark (p-value = 9.32 × 10−14); for Cl2, enrichment is shown in the “epithelial–mesenchymal transition (EMT)” hallmark (p-value = 4.30 × 10−33) and “myogenesis” hallmark (p-value = 9.68 × 10−19); for Cl3, over-expression is shown in the “E2F targets” hallmark (p-value = 2.68 × 10−18) and “G2M checkpoint” (p-value 2.10 × 10−13). The p-values were obtained by means of Kruskal–Wallis tests.
Figure 7
Figure 7
Prognostic evaluation of the three-subtype stratification. (a) Survival analysis on the meta-analysis dataset (MetaHPVpos). The 197 cases, entered into the three subtypes (75/134 Cl1 patients; 56/108 Cl2 patients; 66/104 Cl3 patients), were used for the Kaplan–Meier analysis, yielding a log-rank score of p-value = 4.76 × 10−9. The endpoint was overall survival. (b) Gene-signature. Two models were evaluated: (i) radiosensitivity index (RSI), (ii) the 172-gene prognostic model. RSI is directly proportional to radioresistance (high index = radioresistance), while the 172-gene model is directly proportional to the risk of recurrence. Stratification by both signatures reached p-value = 8.76 × 10−13 and p-value = 7.98 × 10−22 for the RSI and 172-gene model, respectively. (c) Validation on GSE112026. The 47 cases belonging to GSE112026 were stratified based on our three subtypes: 18, 18, and 11 cases were predicted as belonging to Cl1, Cl2, and Cl3, respectively. The cases, entered into the three identified subtypes, were used for the Kaplan–Meier analysis, yielding a p-value = 0.0152 (log-rank test).
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

See all "Cited by" articles

References

    1. Fitzmaurice C., Allen C., Barber R.M., Barregard L., Bhutta Z.A., Brenner H., Dicker D.J., Chimed-Orchir O., Dandona R., Dandona L., et al. Global, regional, and national cancer incidence, mortality, years of life lost, years lived with disability, and disability-adjusted life-years for 32 cancer groups, 1990 to 2015: A systematic analysis for the global burden of disease study. JAMA Oncol. 2017;3:524–548. - PMC - PubMed
    1. Chaturvedi A.K., Engels E.A., Anderson W.F., Gillison M.L. Incidence trends for human papillomavirus-related and -unrelated oral squamous cell carcinomas in the United States. J. Clin. Oncol. 2008;26:612–619. doi: 10.1200/JCO.2007.14.1713. - DOI - PubMed
    1. Rettig E.M., D’Souza G. Epidemiology of head and neck cancer. Surg. Oncol. Clin. N. Am. 2015;24:379–396. doi: 10.1016/j.soc.2015.03.001. - DOI - PubMed
    1. Gillison M.L., Castellsagué X., Chaturvedi A., Goodman M.T., Snijders P., Tommasino M., Arbyn M., Franceschi S. Eurogin Roadmap: Comparative epidemiology of HPV infection and associated cancers of the head and neck and cervix. Int. J. Cancer. 2014;134:497–507. doi: 10.1002/ijc.28201. - DOI - PubMed
    1. Marur S., Forastiere A.A. Head and neck cancer: Changing epidemiology, diagnosis, and treatment. Mayo Clin. Proc. 2008;83:489–501. doi: 10.4065/83.4.489. - DOI - PubMed