A Natural History of Actinic Keratosis and Cutaneous Squamous Cell Carcinoma Microbiomes

Abstract

Cutaneous squamous cell carcinoma (SCC) is the second-most-common cancer in Australia. The majority of SCCs progress from premalignant actinic keratosis (AK) lesions that form on chronically sun-exposed skin. The role of skin microbiota in this progression is not well understood; therefore, we performed a longitudinal microbiome analysis of AKs and SCCs using a cohort of 13 SCC-prone immunocompetent men. The majority of variability in microbial profiles was attributable to subject, followed by time and lesion type. Propionibacterium and Malassezia organisms were relatively more abundant in nonlesional photodamaged skin than in AKs and SCCs. Staphylococcus was most commonly associated with lesional skin, in particular, sequences most closely related to Staphylococcus aureus Of 11 S. aureus-like operational taxonomic units (OTUs), six were significantly associated with SCC lesions across seven subjects, suggesting their specific involvement with AK-to-SCC progression. If a causative link exists between certain S. aureus-like OTUs and SCC etiology, therapeutic approaches specifically targeting these bacteria could be used to reduce SCC.IMPORTANCE Actinic keratosis (AK) and cutaneous squamous cell carcinoma (SCC) are two of the most common dermatologic conditions in Western countries and cause substantial morbidity worldwide. The role of human papillomaviruses under these conditions has been well studied yet remains inconclusive. One PCR-based study has investigated bacteria in the etiology of these conditions; however, no study has investigated the microbiomes of AK and SCC more broadly. We longitudinally profiled the microbiomes of 112 AK lesions, profiled cross sections of 32 spontaneously arising SCC lesions, and compared these to matching nonlesional photodamaged control skin sites. We identified commonly occurring strains of Propionibacterium and Malassezia at higher relative abundances on nonlesional skin than in AK and SCC lesions, and strains of Staphylococcus aureus were relatively more abundant in lesional than nonlesional skin. These findings may aid in the prevention of SCC.

Keywords: 16S RNA; Malassezia; Staphylococcus aureus; actinic keratosis; microbiome; skin; squamous cell carcinoma.

FIG 1

Sampling method used throughout the study. Up to six AK lesions were identified on each arm of each subject (orange points) and sampled once per month for 5 months and then once again no later than 9 months after the fifth visit. Sampling was performed by firmly applying and rotating a swab dipped in a sterile saline buffer to the skin site. For comparison to AKs, three nonlesional control (NLC) sites were selected on the photodamaged extensor surface of each forearm and spaced 10 cm apart, starting at the wrist and extending to the antecubital fossa. SCCs arising on any body site were also swabbed prior to excision, and the skin immediately adjacent to the SCC was swabbed for comparison (SCC_PL). Swab- and buffer-only controls were taken at each sampling session. Samples were briefly stored on ice in the clinic prior to –80°C storage in preparation for DNA extraction and rRNA gene amplicon profiling.

FIG 2

Differences in skin microbiome profiles across subjects and time. (a) Bar plots of microbial relative abundances for the seven most abundant bacterial classes (prefixed with c) and the two most abundant fungal phyla (prefixed with p) aggregated across five monthly time points for each lesion (AK, orange dots at the base of each plot) and for photodamaged nonlesional controls (NLCs) from the extensor surface of the lower arm (blue dots) and nasal samples (mustard yellow dots). (b) PCA ordination indicates that samples separate primarily by subject. PC1 and PC2, principal components 1 and 2, respectively. (c) Community differences between months measured using the Yue-Clayton theta distance (18). As theta values approach 1.00, the communities are more similar. “months apart” indicates the similarity of the skin microbiome profiles for the same lesion over time, taken from 1 to 5 months apart. Profiles at 0 months apart are microbiome profiles of separate AK and NLC samples taken at the same time as a frame of reference for temporal differences. AK community profiles were more similar to those of other AKs on the subjects’ arms during the same time point than they were to the profiles from the same AK across time. AK community profiles were on average significantly more stable than NLC profiles in the first 3 months; however, the magnitude of difference was small. ns, not significantly different. (d) AK-only theta distances indicate a small but significant decrease in AK community similarity over time. (e) NLC community similarity did not significantly decrease over the same period.

FIG 3

OTU associations with AK and NLC samples. (a) Boxplots showing the log₁₀ read counts for 30 differentially abundant OTUs (adjusted log ratio test P value < 0.1), which were calculated with DESeq2 between AKs and NLCs for each subject. Staphylococcus OTUs were encountered frequently and were frequently higher in relative abundance in AK samples than in other samples. (b) Multivariate analysis using sPLS-DA identified sets of OTUs that maximally discriminate between AK and NLC samples for each subject. An OTU’s contribution to variation separating samples on both components of the model was used to determine sample type association. This method identified 707 OTUs associated with either AKs or NLCs across all subjects. Shown here are all OTUs associated with a sample type from five selected genera. Each circle represents an OTU tested in each subject. The color represents the sample type association, and the size of the circle is proportional to the sPLS-DA “importance” metric. Small open gray circles indicate that the OTU was testable in the subject but that no sample type association was identified. If no circle is present, the OTU did not have sufficient reads in that sample for testing. Black dots indicate that the OTU was differentially abundant in the univariate analysis. Propionibacterium was significantly enriched in NLC-associated OTUs (adjusted P value, 1.02e^–8, chi-square test). Malassezia and Micrococcus were also enriched in NLC-associated OTUs; however, differences between these genera did not reach significance. OTUs highlighted with black dots were also identified as differentially abundant in the DESeq2 univariate analysis.

FIG 4

Abundance correlation network analysis of AK- and NLC-associated OTUs. (a) Histogram of all pairwise OTU comparisons noting that the vast majority of values are around zero and are not significantly correlated. (b) All significant correlations (P < 0.05) note a difference in y axis range from that of the previous panel. (c) Significant correlations involving two OTUs associated with the same sample type (AK::AK and NLC::NLC) were positive, whereas correlations involving OTUs from different sample types (AK::NLC) were generally negative (the red line indicates mean correlation across all pairwise OTUs). (d) Significant correlations by genus from most positively correlated to most negatively correlated. Intragenus OTU correlations were generally positively correlated (73 out of 97). (e) Correlation network graphically showing major inferred relationships between frequently observed genera. Correlations between Malassezia and Propionibacterium were generally positive, whereas correlations between Staphylococcus and Propionibacterium were generally negative, suggesting competitive exclusion between the latter two genera. (f) The network for subject MS002 shows two highly interconnected modules, one NLC associated (containing predominantly Propionibacterium and Malassezia OTUs) and one AK associated (containing predominantly Staphylococcus OTUs), that are negatively correlated with each other, suggesting a preference for lesional or nonlesional environments.

FIG 5

Identification of OTUs associated with SCC or perilesional skin (SCC_PL). (a) Fourteen OTUs found to be differentially abundant between SCC and SCC_PL samples according to univariate analysis, of which 13 were relatively more abundant in SCC microbiomes. (b) Multivariate analysis highlights a microbial signature differentiating SCC and SCC_PL samples. SCCs from subjects MS002, MS008, MS0011, MS0012, MS0013, and MS0014 were notable for their high relative abundances of several Staphylococcus aureus OTUs, whereas Propionibacterium and Malassezia OTUs were typically of higher relative abundance in SCC_PL samples. (c) Nasal microbiomes indicate potential carriage of several OTUs associated with SCCs, including a number of the S. aureus OTUs.