Characteristics of patient cohort and clinical specimens
To investigate the microbiome on MF lesions, we analysed metagenomic samples from patches (n = 19), plaques (n = 15) and nonlesional skin (n = 31) of 20 MF patients (8 females, 12 males) recruited from two skin cancer centres in Germany (Mainz, n = 14 patients and Frankfurt, n = 6 patients). Nonlesional skin from the contralateral body site of the same patients served as controls. Clinical samples were collected from patients with MF stages IA – IIB, with 15 patients displaying early-stage MF (stages IA – IIA), and 5 patients suffering from advanced-stage MF (IIB). We did not include tumour-stage lesions because these often ulcerate and might thereby offer a unique microbial habitat due to their moist wound characteristic [1, 32]. The mean age at sampling was 66.4 ± 11.7 years, and the mean age at first diagnosis of MF was 60 ± 11.2 years. Patients were treated with common therapy (Table 1). In addition to metagenomic samples, that were collected for the entire patient cohort, we obtained skin punch biopsies from lesional and adjacent skin, as well as peripheral blood in a subset of patients. RT-qPCR (N = 24) was performed from the skin punch biopsies to study expression levels of antimicrobial peptides (AMPs). T cell receptor sequencing (TCRseq) was carried out on a total of 16 skin samples and 5 blood samples, obtained from 10 patients. Living bacterial isolates were picked from plaque and nonlesional skin of one patient (Pat1) and from healthy subjects. The patients clinical course including events (defined as death, start of new therapy and disease progression) and the time from observation start to event (TTE) were assessed. A summary of the patient characteristics is given in Table 1.
Table 1
Patient characteristics, therapy regimen and clinical outcomes of the study cohort. f = female; m = male, MZ = Mainz, Germany; FFM = Frankfurt am Main, Germany; PUVA = psoralen and UVA; MTX = methotrexate, PEG-IFNα = pegylated interferon alpha; RTx = radiation therapy, NA = not available, LFU = lost to follow-up; TTE = Time To Event
Patient ID | Study Site | Sex | Age at Sampling (years) | Clinical Stage at Sampling | Metagenomics: Stage of Lesions (Body Site) | RT-qPCR: Stage of Lesions (Body Site) | TCRseq: Tissues included | ΔSA- subgroup | Therapy | Event | TTE (months) | Follow-up Time (months) |
Pat1 | MZ | f | 81 | IA | Plaque (hip), Patch (forearm) | Plaque (hip) | Skin (Plaque), Blood | positive | PUVA, Mechlorethamine | Yes | 2.60 | 17.56 |
Pat2 | MZ | f | 57 | IB | Plaque (thigh), Plaque (abdomen) | NA | NA | neutral | PUVA, UVB, topical steroids, Pimecrolimus, MTX, Bexarotene, Mogamulizumab | Yes | 18.41 | 18.41 |
Pat3 | MZ | f | 47 | IB | Patch (thigh) | Patch (thigh) | NA | neutral | topical steroids | Censored | Censored | 17.95 |
Pat4 | MZ | m | 70 | IIB | Plaque (upper back), Plaque (flank) | Plaque (upper back) | Skin (Plaque and nonlesional), Blood | positive | topical steroids, Bexarotene, Mogamulizumab, RTx | Yes | 0.92 | 9.99 |
Pat5 | MZ | f | 73 | IB | Patch (thigh), Patch (forearm) | Patch (thigh) | Skin (Patch and nonlesional) | neutral | topical steroids | Yes | 13.87 | 17.10 |
Pat6 | MZ | f | 81 | IA | Patch (breast) | Patch (breast) | Skin (Patch and nonlesional) | neutral | topical steroids, Mechlorethamine | Censored | Censored | 15.55 |
Pat7 | MZ | m | 54 | IB | Patch (forearm), Plaque (flank) | Plaque (flank) | Skin (Plaque and nonlesional) | neutral | PUVA, PEG-IFNα | Censored | Censored | 16.11 |
Pat8 | MZ | m | 82 | IIB | Plaque (head, parietal), Patch (upper back) | Patch (upper back) | Skin (Patch), Blood | positive | topical steroids, MTX, RTx | Yes | 2.53 | 2.53 |
Pat9 | MZ | m | 63 | IB | Plaque (forearm) | Plaque (forearm) | Skin (Plaque) | neutral | Topical steroids, PUVA, RTx | Yes | 6.41 | 14.27 |
Pat10 | MZ | m | 49 | IB | Patch (gluteus right), Patch (gluteus left) | Patch (gluteus left) | Skin (Patch) | neutral | NA | LFU | LFU | LFU |
Pat11 | MZ | m | 61 | IIB | Patch (wrist) | NA | NA | neutral | NA | LFU | LFU | LFU |
Pat12 | MZ | m | 88 | IA | Patch (abdomen), Patch (thigh) | Patch (thigh) | Skin (Patch and nonlesional), Blood | neutral | no special therapy, only moisturizing cremes | Censored | Censored | 11.51 |
Pat13 | MZ | m | 66 | IB | Patch (lower leg, front), Patch (lower leg, back) | Patch (lower leg, back) | NA | neutral | topical steroids | Censored | Censored | 11.51 |
Pat14 | MZ | m | 62 | IB | Plaque (forearm), Plaque (lower leg, front) | Plaque (lower leg, front) | Skin (Plaque and nonlesional), Blood | neutral | topical steroids, Silver iodide | Censored | Censored | 11.05 |
Pat15 | FFM | f | 73 | IB | 2x Plaque (lower leg, front and gluteus) | NA | NA | positive | MTX, RTx | Yes | 2.83 | 11.57 |
Pat16 | FFM | f | 58 | IA | Plaque (scapula) | NA | NA | positive | MTX, Bexarotene | Yes | 4.87 | 11.51 |
Pat17 | FFM | m | 83 | IIB | 2x Patch (hip and rips) | NA | NA | neutral | Bexarotene | Yes | 7.00 | 7.00 |
Pat18 | FFM | f | 61 | IB | 2x Patch (thigh dorsal & abdominal) | NA | NA | neutral | no special therapy | Censored | Censored | 9.99 |
Pat19 | FFM | m | 54 | IIB | Patch (shoulder), Plaque (thigh) | NA | NA | neutral | Bexarotene | Censored | Censored | 9.67 |
Pat20 | FFM | m | 65 | IB | Plaque (forearm) | NA | NA | neutral | Bexarotene | Censored | Censored | 11.11 |
Analysis of rarefied microbial reads shows a dysbiosis on lesional MF skin
In a first analysis we clustered taxonomic profiles that were generated from rarefied metagenomic samples (Fig. 1a). As expected, clustering showed a strong grouping of specimens derived from the same patient, which justified our approach of using intra-patient controls [21]. By contrast, we observed no significant differences in the microbiome composition between the two study centres, and therefore did not consider these in subsequent analysis (see also Suppl. Figure 1d). Overall, lesional samples appeared to have a lower microbial diversity compared to nonlesional controls. Remarkably, some patch and plaque samples from different patients clustered together, showing that the microbiome on lesional skin was altered in a uniform manner. To test whether the composition of the microbial community on patches and plaques differed from that of nonlesional skin, we calculated Shannon indices and Bray-Curtis dissimilarities from rarefied microbial reads. The α-diversity was reduced on MF lesions, with patch stage showing the strongest effect, but significance was not reached (Fig. 1b). However, β-diversity revealed a significant instability of the microbiome on MF lesions that increased with exacerbation of MF lesions (Fig. 1c). Moreover, α- and β-diversity of plaque skin exhibited a bimodal distribution. Several samples showed a strongly decreased Shannon Index and an affected microbiome stability, whereas others resembled the diversity-metrics of nonlesional skin. Notably, no significant associations with other demographics of the study cohort were found (Suppl. Figure 1). Together, these results showed that dysbiosis is a common feature of MF lesions, which increased with exacerbation.
S. aureus colonization stratifies the MF patient cohort into two subgroups with distinct microbiome patterns
To understand which species accounted for the observed microbiome patterns on lesional skin, we set up a generalized linear mixed model (GLMM) using MaAsLin 2 [33]. We tested for differences in the abundance of microbial species on patches and plaques compared to nonlesional skin while adjusting for sequencing depth and the individuality of the patient’s microbiome. Using this approach, we found that S. aureus was highly enriched on plaque while all other significantly associated microbial species were extremely reduced on patch and plaque (Fig. 2a, Suppl. Material 2). Among them were S. hominis, S. epidermidis and Cutibacterium acnes. These commensals can control S. aureus growth in other inflammatory skin conditions [34–37] and confer decreased release of inflammatory cytokines as well as recruitment of leukocytes in skin wounding healing [38].
When plotting the relative abundance of S. aureus patient-wise and differentiating between lesional stages, we noted a patient stratification into two subgroups: In one subgroup, the relative abundance of S. aureus did not change from nonlesional skin to plaque stage. In the other subgroup however S. aureus abundance substantially increased from nonlesional skin to plaque stage. We therefore referred to the first group as Δ S. aureus-neutral (ΔSA-neutral) and to the other group as Δ S. aureus-positive (ΔSA-positive) (Fig. 2b). Strikingly, when stratifying the α-diversity (Shannon Index) to the defined patient subgroups, the observed bimodality resolved: ΔSA-positive patients exhibited a significantly reduced α-diversity on plaques, and ΔSA-neutral patients in turn had a plaque microbiome which was as diverse as that on nonlesional skin (Fig. 2c). Likewise, the three commensals with anti-S. aureus activity were more prominent on plaques of ΔSA-neutral patients, albeit this association was above statistical significance (p = 0.05–0.14). This might be attributed to the fact that every plaque lesion of ΔSA-neutral patients was dominated by only one of the three commensals with anti-S. aureus action while the other two were minor constituents of the skin flora (Fig. 2d-f). When compiling S. hominis, S. epidermidis and C. acnes, S. aureus-inhibiting microbes were significantly more abundant on plaque than on nonlesional skin of ΔSA-neutral patients (Fig. 2g).
Increased and sustained expression of cutaneous antimicrobial peptides could lead to skin dysbiosis
The skin expresses a diverse repertoire of antimicrobial peptides (AMPs) to control microbial colonization and ensure epithelial integrity [39, 40]. Under steady-state conditions, AMPs are constitutively produced at low rates but increase upon injury or inflammation [39, 40]. MF is characterized by an inflammatory microenvironment and lesions might persist over long time periods [41]. In order to evaluate whether cutaneous AMPs may have contributed to the skin dysbiosis we obtained skin punch biopsies from the same MF lesions sampled for metagenomic profiling and analysed AMP expressions levels using RT-qPCR. Adjacent nonlesional skin served as control. RNA expression levels of the AMPs hBD2, hBD3, S100A7, and the calprotectin forming S100A8 and S100A9 were significantly increased in MF lesions but did not significantly differ between patch and plaque stage (Fig. 3a). This observation is consistent with the fact that these AMPs are known to be regulated by microbial recognition mechanisms [40, 42]. In contrast, hBD1 and LL-37 are known to be constitutively expressed rather than inducible [40, 42], which is in agreement with our results (Fig. 3a).
Next we examined whether ectopic AMP levels affect growth of cutaneous bacterial species found on MF lesions. Clinical isolates of S. aureus and S. hominis were picked from plaque skin of Pat1 (termed S. aureus MFMZ1 and S. hominis MFMZ1, respectively). S. aureus (Emo1) and S. epidermidis (MV01) picked from 2 healthy subjects served as negative controls. Another clinical S. aureus isolate historically picked from pleural fluid back in 1884 (S. aureus Rosenbach 1884, strain DSM11823) served as positive control [43]. All bacterial samples were exposed to increasing concentrations of hBD1, hBD3, LL-37 and calprotectin (a heterodimer consisting of S100A8 and S100A9).
Clinical isolates S. aureus MFMZ1, S. hominis MFMZ1 and S. aureus DSM 11823 survived at low concentrations of hBD3 (1 µg), while healthy subject-derived S. epidermidis MV01 was eradicated. All strains survived at low concentrations of LL-37 (1 µg, Fig. 3b). This was to be expected, because LL-37 was constitutively expressed in nonlesional skin and was also not induced by the disease (Fig. 3a). Hence, bacterial skin colonization requires resistance to LL-37 at least at low concentrations. Interestingly, S. aureus MFMZ1 showed the least reduction in survival at high concentrations of hBD3 and LL-37 compared to all other strains tested. (Fig. 3c). Furthermore, exposure to calprotectin resulted in decreased survival of only S. hominis MFMZ1, a bacterium with potential anti-S. aureus properties [34–38]. Surprisingly, hBD1 had a paradoxical effect on S. aureus MFMZ1, even augmenting its survival, while the survival of all other tested strains remained unchanged or decreased (Fig. 3c).
These data demonstrate that clinical isolates of S. aureus, and particularly that obtained from plaque of an MF patient, have considerable survival advantages under ectopic AMP application. As neoplastic T cells and reactive leukocyte infiltrate accumulate in MF lesions over time, plaques can be expected to persist longer than patches and, accordingly, the microbiome on plaques is exposed longer to high AMP levels than the microbiome on patches (and nonlesional skin). Given the fact that skin dysbiosis is most accentuated in patch stage (Fig. 2a), our results collectively indicate that most of the physiological skin flora is eradicated upon onset of increased AMP production. With ongoing persistence of MF lesions and high AMP levels, MF skin commensals eventually adopt to the new environmental conditions and (re-) colonize MF lesions.
MF skin lesions are colonized by distinct S. aureus strains that outgrow other MF skin commensals
We next asked why S. aureus outgrows only in the ΔSA-positive subgroup. Given the substantial survival advantages of S. aureus under high AMP expression levels and the observation of skin dysbiosis exclusively in the ΔSA-positive subgroup (Fig. 2c), we first reasoned that cutaneous AMP levels of the ΔSA-neutral subgroup would be lower compared to the ΔSA-positive subgroup. Surprisingly, the AMP expression levels did not significantly differ between lesions of ΔSA-neutral and ΔSA-positive patients (Suppl. Figure 2). There must hence be another factor or event allowing S. aureus to accumulate and outgrow competitive commensals. A potential explanation might be that S. aureus strains colonizing MF lesions differ from their nonlesional counterparts, since S. aureus is a common commensal of human skin flora but also a frequent pathogen in many diseases [21, 32]. To test this hypothesis, we used the tool PanPhlAn [44] to profile bacterial strains in the MF microbiome by the presence and absence of genes in the respective species’ pangenomes. Dimensionality reduction via principal component analysis (PCA) revealed that strains of S. hominis, S. epidermidis and C. acnes did not differ between MF lesions and nonlesional skin. In contrast, S. aureus strains present on plaques had a unique gene repertoire, clearly demonstrating that lesional S. aureus were of a different strain than their nonlesional counterparts (Fig. 4).
In order to validate these observations, we examined whether strain differences seen in the computational analysis could be transferred to the same living isolates tested before. In a disk diffusion assay, S. aureus MFMZ1 was considerably more resistant towards antibiotics typically used in the clinic than all other tested strains (Fig. 5). While Pat1-derived S. epidermidis MFMZ1 and healthy subject-derived S. epidermidis MV01 showed comparable susceptibilities, S. aureus MFMZ1 was notably more resistant than its counterpart from healthy skin (S. aureus EM01). Remarkably, the other clinical S. aureus DSM11823 strain was sensitive towards both β-lactam antibiotics, ampicillin and carbenicillin, as well as gentamycin, whereas S. aureus MFMZ1 was not (Fig. 5, Suppl. Table 1). We could further determine that S. aureus MFMZ1 was a methicillin resistant S. aureus (MRSA) strain, while S. aureus EM01 was not (Fig. 5b, c).
S. aureus MFMZ1 not only demonstrated reduced susceptibility to antibiotics but also exhibited the strongest adaptation to the AMP conditions found on MF skin (Fig. 3b, c). In particular, this strain was not only resistant to hBD1, but its growth was even promoted by the AMP. Given that the other skin bacteria we tested in this assay did not show this effect, and were less resistant to other AMPs, likely provided S. aureus with an advantage over other skin commensals on plaques.
Taken together, S. aureus present on MF lesions was of a distinct strain that differed from its nonlesional counterpart, probably outcompeting other skin commensals by adapting more effectively to the unique environmental conditions of MF skin.
S. aureus strains on plaque of ΔSA-positive patients are highly virulent
Because S. aureus often acts as a pathogen [32], we next evaluated the virulent properties of the microbiome present on MF lesions. Utilizing ShortBRED [46], we profiled the whole metagenome sequencing (WMS) reads against the Virulence Factor Database (VFDB) [47] to detect the presence of virulence factors. Differences in virulence factor abundance between nonlesional skin and MF lesions were evaluated using MaAsLin 2.
Here, we found numerous S. aureus-associated virulence factors that were significantly more prevalent on plaques (Fig. 6a). Given that S. aureus was the only microbe displaying an increased abundance on plaques within the ΔSA-positive subgroup (Fig. 2a, b), and since all of the virulence factors identified are usually associated with this pathogen, it is highly likely that these virulence factors originated from S. aureus of ΔSA-positive patients.
Surprisingly, we did not find any superantigens, which are typically associated with disease progression in S. aureus infected CTCL patients [17, 48–50]. However, we found an enrichment of α-hemolysin (hly/hla), another virulence factor of S. aureus, which is linked to MF pathogenesis [51, 52]. hly/hla can form pores in human T cells, causing cellular damage [53]. In the context of MF, it was demonstrated that hly/hla preferentially induces cell death in benign T cells over malignant T cells [51] and inhibits cytotoxic T cell mediated killing of malignant T cells [52].
Furthermore, we identified an enrichment of several virulence factors that have not been associated with MF pathogenesis before (Fig. 6a). Most of the S. aureus-associated genes are components of larger complexes that orchestrate nutrition, immune evasion, spread of infection and secretion of virulence factors (Suppl. Material 3 provides a comprehensive overview of these virulence factors and their functional properties). In particular, we found a differential enrichment of Immunoglubin G binding protein A (spa), along with iron-regulated surface determinant protein A (isdA) and type VII secretion system protein D (esaD).
Since isdA was reported to confer resistance to the AMP hBD2 [54], this virulence factor might have enabled S. aureus outgrowth on plaque despite the increased RNA-levels of hBD2 in these lesions (Fig. 3a). An additional factor for S. aureus outgrowth may be esaD, which encodes a strong bactericidal for other bacteria than S. aureus, suppressing growth of competing commensals [55].
Of particular significance was the upregulation of spa on plaques: As previously reported, this virulence factor can activate the NF-κB (Nuclear factor kappa-light-chain-enhancer of activated B cells) pathway via tumour-necrosis factor-α (TNF-α) receptor 1 (TNFR1) through conserved IgG binding domains [56, 57]. Several studies demonstrated that the TNF-α/NF-κB pathway is dysregulated in a subset of MF patients with poor clinical outcome [29–31], and spa could hence be the activating factor. We therefore next investigated spa in Pat1-derived isolate S. aureus MFMZ1 (spa_MFMZ1) and not only confirmed its presence within the genome (Fig. 6b), but also identified mutations via sequencing: Compared to spa of the S. aureus Newman strain serving as a reference (spa_NM, UniProt ID: A0A0H3K686), spa_MFMZ1 had an additional octapeptide repeat (Ins373PGKEDNNK) in the conserved IgG binding domain, resulting in a total of 12 repeats (Fig. 6c). It is known that the spa IgG binding domain is composed of a variable number of octapeptide repeats and activates inflammatory responses via interaction with the Fc fragment of IgG [58, 59]. Strains of S. aureus with more than 7 repeats are generally considered more virulent as they can bind more precisely to the Fc fragment [60, 61]. Notably, the IgG binding domain of spa also activates TNFR1 signalling through the same octapeptide repeats [56, 57] where we detected the mutational insertion. Hence, the mutation-dependently increased number of octapeptide repeats of spa in the strain derived from MFMZ1 may enhance spa binding efficacy to TNFR1 and in turn pronouncedly activate the TNF-α/NF-κB axis.
In summary, these results show that S. aureus strains on plaque of ΔSA-positive patients were highly virulent and shaped environmental conditions to evade the host immune response, spread infection and may fuel disease progression, as deduced from the potential gain-of-function mutation of spa in the MFMZ1 strain.
S. aureus assaults the T cell receptor repertoire and might promote malignancy and dissemination via spa
To assess a potential impact of S. aureus and its identified virulence factors on the malignant and benign T cell infiltrate, we further performed T cell receptor (TCR) sequencing in the skin and blood of MF patients. Surprisingly, although one might expect a skewed TCR repertoire in MF lesions due to the expansion of malignant T cells, instead, abundance and diversity increased with lesional stage (Fig. 7a, b). This augmentation might have been caused by tumour infiltrating lymphocytes (TILs) [62] and the expansion of more than just a single malignant clone resulting in oligoclonality [63–65]. Indeed, oligoclonality was detected in most MF lesions (Fig. 7d-k).
Intriguingly, the TCR repertoire in plaques of the ΔSA-positive group was reduced (Fig. 7a, b). In addition, gene usage analysis of the T cell receptor β variable (TRBV) revealed that TRBV5-1, which another investigation linked to TILs in MF [66], was strongly expanded in plaque of the ΔSA-negative group but not in plaque of the ΔSA-positive group (Suppl. Figure 3). We assumed that the reduction of both total TCR repertoire as well as TRBV5-1 in plaque of ΔSA-positive patients could be due to the virulence factor hla. Hla is known to kill benign T cells in MF [51, 52], and we found this virulence factor enriched in the ΔSA-positive subgroup (Fig. 6a). Collectively, our data indicate that S. aureus affected the balance between malignant T cells and benign tumour infiltrating T cells.
We next examined whether T cells of MF patients were directed against the S. aureus virulence factors identified as enriched on plaque (Fig. 6a). Corresponding epitopes were obtained from the Immune Epitope Database (IEDB) [67], or, when not available in the IEDB, computationally predicted for both Major Histocompatibility Complex class I (MHC-I) and II (MHC-II). Binding Scores were calculated for the most abundant TCRs of a given sample and each of the epitopes (obtained and predicted) of S. aureus virulence factors. As expected, TCRs in patch, plaque and blood showed a significantly higher affinity for S. aureus virulence factors compared to TCRs in nonlesional skin (Fig. 7c). Notably, dominant T cell clones of MF lesions were also detected in nonlesional skin and the blood (Fig. 7d-k). In all tissues tested, the MHC-II epitope of spa was among the epitopes that was recognized most often by TCRs (Suppl. Figure 4). MHC-II interacts with the TCR of CD4 + T cells, which is the T cell subset that undergoes malignant transformation in MF [68]. As already outlined above, spa is known to activate NF-κB and to trigger inflammation [56, 57], typical characteristics of progressive MF [2].
Collectively, our results indicate that the plaque-enriched S. aureus virulence factors, and of these especially spa, were recognized by T cells in MF lesions and that those clones might disseminate into other tissue compartments, thus driving pathogenesis.
ΔSA-positive patients have an inferior event-free survival
We demonstrated that S. aureus strains on MF lesions differed from nonlesional counterparts being highly virulent, and dominated the plaque microbiome of ΔSA-positive patients, which had a comprised TCR repertoire that specifically recognized S. aureus virulence factors. Hence, the ΔSA-positive subgroup might undergo a more severe course of disease. Therefore, the clinical response of the MF study cohort was monitored while patients were treated with approved MF therapy adequate to the patients’ individual situation. Event-free survival (EFS) was evaluated using Kaplan-Meier analysis. Events were defined as death, start of new therapy and disease progression. Two out of 20 patients were lost to follow-up (LFU) for clinical assessment and therefore excluded from analysis. 9 out of 18 patients experienced an event during the observation period (2.5–18.4 months, median 11.5). Metrics are summarized in Table 1.
Patients in the ΔSA-positive group exhibited a strikingly inferior EFS with a hazard ratio of 11.91 (95% confidence interval 1.44 to 98.36) (Fig. 8). Median EFS was 2.6 months in the ΔSA-positive group and not reached in the ΔSA-neutral group. No EFS associations were found for age and gender, however, EFS was significantly lower for patients treated systemically and when undergoing radiotherapy, respectively (Supplementary Fig. 5a, b). This is to be expected as these lines of therapy are chosen in advanced-stage or refractory early-stage MF ([69], and are thus associated with poor clinical course [4]. The cohort analysed consists of 14 patients in early-stage (IA-IIA) MF and 4 patients in advanced-stage (IIB) MF. Since advanced-stage MF tends to progress faster than early-stages [4], joining clinical stages should be considered with caution. Notably, the four 4 stage IIB MF patients were distributed equally between the ΔSA-subgroups (two each) and independent Kaplan-Meier analysis of early- and advanced stage MF patients yielded similar results (Supplementary Fig. 5d, e) as the joint analysis depicted in Fig. 10.
This demonstrates that the skin-microbiome stratifies MF patients into two subgroups with distinct clinical outcomes, opening-up the potential of new treatment regimens.