Diagnostic Utility and Clinicopathologic Associations of Histone H3 Lysine 27 Trimethylation (H3K27me3) Immunohistochemistry for Merkel Cell Carcinoma

Abstract

Merkel cell carcinoma (MCC) is an aggressive cutaneous tumor that must be distinguished from other cutaneous tumors and metastatic small cell carcinoma (SmCC). Additional diagnostic markers are limited for MCC with immunophenotypic aberrancy. MCC can display immunohistochemical loss of the epigenetic marker histone H3 lysine 27 trimethylation (H3K27me3), but to our knowledge, the diagnostic utility of this observation has not been evaluated. In this study, we investigate H3K27me3 labeling in MCC (n = 195), cutaneous epithelial tumors (n = 48), noncutaneous SmCCs (n = 56), and olfactory neuroblastoma (n = 11), comparing with diagnostic markers, cytokeratin-20, neurofilament, SATB2, and POU4F3. H3K27me3 patterns in MCC included global loss, variable/mosaic labeling, and diffuse labeling. Global loss significantly associated with polyomavirus negativity, squamous atypia, and sarcomatoid change. Tumors with global loss displayed EZHIP expression (9 cases) and SUZ12 mutation (1 case). Low but retained H3K27me3 labeling was associated with longer overall and MCC-specific survival. Diagnostically, H3K27me3 labeling in MCC was significantly lower than potential mimics, and global loss of H3K27me3 was highly specific for MCC; stronger H3K27me3 labeling was not informative. Considering reduced/absent H3K27me3 as favoring MCC, diagnostic performance was similar to SATB2. However, H3K27me3 displayed consistent performance in MCC with challenging immunophenotypes, unlike SATB2. In summary, we expand upon descriptions of H3K27me3 labeling in MCC and characterize patterns of H3K27me3 in other tumor types including SmCCs and olfactory neuroblastoma. Our findings support diagnostic utility for the widely available marker H3K27me3 in MCC, with weaker labeling favoring MCC over mimics in challenging cases.

Introduction

Merkel cell carcinoma (MCC) is an aggressive cutaneous neuroendocrine carcinoma with significant risk for metastatic spread and mortality.^1,2 Although rare, MCC has been increasing in incidence. MCC presents on the skin of older individuals and displays increased incidence and aggressiveness in the immunosuppressed. A significant minority of cases present in lymph nodes in the absence of a skin lesion (metastatic MCC of unknown primary).³ Current evidence supports 2 distinct molecular subclasses of MCC: Merkel cell polyomavirus (MCPyV)—positive MCC (virus-positive MCC [VP-MCC]) associated with the oncogenic MCPyV⁴ and virus-negative MCC (VN-MCC) associated with UV genomic damage and inactivating mutations in TP53 and RB1.^1,5–7

MCC typically presents as a mitotically active small round cell proliferation, usually centered in the dermis, with involvement of the epidermis or subcutis in some cases. A subset of MCC displays significant variations from classic morphology that can raise diagnostic challenges.^1,8 Specifically, VN-MCC may display increased pleomorphism in some cases that might raise consideration for other poorly differentiated carcinomas, especially in limited samplings.⁸ Sarcomatoid changes or ductal differentiation can be observed in rare cases.⁸ More commonly, VN-MCC can be associated with areas of squamous atypia similar to squamous cell carcinoma (SCC), which are genetically related to the small cell component.^8–13 Although the majority of MCC lack an identifiable precursor, evidence suggests that SCC in situ (SCCIS) may represent a precursor lesion for VN-MCC in at least some cases.^11–13 By contrast, VP-MCC is not associated with squamous atypia and usually lacks evidence of a precursor lesion, although rare cases have been described arising in association with benign hair follicle tumors (trichoblastomas).^14–16

The differential diagnosis for MCC depends in part on the site of presentation as well as the presence of variant morphology.^1,8 When MCC presents in the skin, other small cell cutaneous malignancies (such as melanoma or lymphoma) can often be excluded by morphologic assessment and lineage markers.⁸ MCC might also be confused with basal cell carcinoma (BCC) or basaloid/poorly differentiated SCC, especially in limited samplings.⁸ Olfactory neuroblastoma (ONB) typically presents in the deeper nasal tissues; however, there is immunophenotypic overlap with MCC with regard to neuroendocrine markers, SATB2, and (in some cases) cytokeratins (CKs).^17,18 However, the most challenging distinction is with noncutaneous small cell carcinomas (SmCCs), which are often morphologically identical to MCC.^1,2,8

For these reasons, immunohistochemistry is essential for MCC diagnosis, especially for distinction from other SmCCs.⁸ CK20 is relatively specific for MCC, especially when diffuse; however, there can be substantial CK20 expression in certain other types of SmCC (such as parotid, cervical, and prostate).^1,8,19 TTF1 expression is more typical of small cell lung carcinoma but can be observed in unusual cases of MCC.⁸ Neurofilament is highly specific for MCC, with reasonable sensitivity.^8,19,20 MCPyV immunohistochemistry or in situ hybridization (ISH) is highly specific for MCC but has more limited sensitivity.^1,8 SATB2 expression has also been reported to favor MCC, especially when diffuse.^20–23 However, no marker is universally sensitive or specific.⁸ In particular, VN-MCC has an increased likelihood of immunophenotypic aberrancies that can render definitive diagnosis challenging.^24–26

Given the limitations of established markers, effective diagnostic markers for MCC remain an active area of investigation. SOX2 is highly sensitive for MCC; however, specificity relative to other SmCCs is lower (70%) than established diagnostic markers, such as CK20 or neurofilament.^23,27 Additional candidate markers for MCC (including POU4F3, ATOH1, and TFAP2B)^20,28–32 are not likely to be routinely available in most clinical diagnostic laboratories. Thus, there remains a need for effective and routinely available markers for MCC diagnosis in challenging cases.

Histone H3 lysine 27 trimethylation (H3K27me3) is a repressive epigenetic marker mediated by the polycomb repressor complex 2 (PRC2), of which EZH2 is the catalytic methyltransferase subunit.^33–35 H3K27me3 dysregulation can contribute to tumor biology in complex and context-dependent ways, including loss in some tumor types, or increased labeling associated with global chromatin condensation and invasiveness in other tumors.^{33,34,36–38} H3K27me3 is an established diagnostic marker for PRC2-mutant malignant peripheral nerve sheath tumor, for which diffuse H3K27me3 loss favors this diagnosis over other spindle cell tumors.³⁹ Loss of H3K27me3 has also been described in MCC.^12,40–45 However, reports on patterns of H3K27me3 labeling are very limited for other SmCCs. To our knowledge, the diagnostic utility of H3K27me3 has not been rigorously evaluated for MCC and most of its mimics.

To define the diagnostic and prognostic utility of H3K27me3 immunohistochemistry for MCC, we examined H3K27me3 immunostaining in a large cohort of MCC and diagnostic mimics, with comparison to established diagnostic markers, MCPyV, CK20, neurofilament, and SATB2, and the new diagnostic marker POU4F3.

Materials and Methods

Case Organization

The study was conducted under a previously approved protocol by the Institutional Review Board. Cases for inclusion were identified by retrospective search of the University of Michigan, Department of Pathology laboratory information system and re-reviewed by a board-certified dermatopathologist (P.W.H.) for diagnostic confirmation. A total of 5 tissue microarrays (TMAs) were used, including several previously described,⁴⁶ consisting of tumor cases represented by duplicate 1.0-mm cores. Cases were evaluated by whole sections if not amenable to TMA or to verify TMA results as detailed in the following sections. Case-level details are provided in Supplementary Table S1.

Immunohistochemistry and In Situ Hybridization

For H3K27me3 immunohistochemistry, sections of 4-μm thickness were deparaffinized, and heat-induced epitope retrieval was performed on the Ventana Benchmark Ultra immunostainer using cell conditioning 1 buffer from Ventana Medical Systems. After blocking endogenous peroxidase activity, the slides were incubated with an H3K27me3 antibody (Cell Signaling; catalog number: 9733S, clone C36B11) diluted at 1:25 for 32 minutes at 37 °C. Immunoreactivity was detected by using the UltraView Universal DAB Detection kit (Ventana Medical Systems). Antibody and protocol details for other immunohistochemical stains (SATB2, H3K27M, neurofilament, POU4F3, and CK20) are listed in Supplementary Table S2. All antibodies used in this study have been rigorously validated for clinical use. H3K27me3 results have been previously reported for 4 cases in this study (Supplementary Table S1).¹² CK20 and neurofilament were predominantly obtained for diagnostic purposes and/or previous studies.^19,24

Chromogenic RNA ISH was performed on a Leica Bond-III autostainer (Leica Biosystems) with mRNA probes purchased from ACDBio, which we validated for clinical use. Freshly cut 4-μm formalin-fixed, paraffin-embedded tissue sections were deparaffinized, and heat-induced epitope retrieval was performed on a Leica Bond-III immunostainer using Bond Epitope Retrieval Buffer 2 from Leica Biosystems for 15 minutes at 95 °C, followed by incubation with Bond Protease Solution for 15 minutes at 40 °C. The Leica Bond RNAScope Detection kit (Leica Biosystems; catalog no.: DS9790) was used to detect single-molecule ISH of MCPyV and human papillomavirus (HPV)—high-risk RNA according to the manufacturer’s protocol. Sections including external control tissue were incubated with Hs-MCPyV (ACDBio; catalog no.: 431008) and Hs-HPV-HR18 (ACDBio; catalog no.: 312598) for 2 hours at 42 °C, followed by step-wise amplification of the probe with the AMP1-AMP6 re-agents supplied with the kit. Slides were subsequently counterstained with hematoxylin (5 minutes) and blued (2 minutes), all at room temperature on the Leica Bond-III immunostainer. RNA ISH results have been previously reported for some cases.^47,48

MCPyV classification was based on either ISH (given the high performance of this stain relative to other assays)⁴⁷ or the previously described combinatorial classification (ISH or real-time PCR).⁴⁷

Scoring

H-scoring was performed by a board-certified dermatopathologist (P.W.H.) for SATB2, EZHIP, and H3K27me3, with an intensity scale from 0 (negative) to 3 (strong). For cases stained on both whole sections and TMA cores, the whole section score was used for subsequent analyses. Punctate staining corresponding to Barr bodies could be observed in many tumors from female patients but did not significantly impact H-scoring. Global loss was defined as H-score = 0 (expression in <1% tumor cells). All cases with global loss of H3K27me3 were confirmed by evaluation of an immunostained whole tumor section. For survival analyses, low vs high H3K27me3 labeling was defined by median labeling among primary tumors (H = 120) with “low” expression defined as less than the median, and “high” expression defined as equal to or greater than the median; tumors with global loss were not included in P value calculations due to the small size of this group.

For neurofilaments, cases were scored as either negative or positive (minimum 5 positive cells) as previously described.¹⁹ Previously reported EZH2 H-scores were used for comparison with H3K27me3.⁴⁶ CK20 and MCPyV stains were scored in a binary (positive or negative) manner. POU4F3 was scored based on percentage positivity as previously described³²: 3+, >80%; 2+, 40% to 80%; 1+, 5% to <40%; and 0, <5% (complete results are provided in Supplementary Table S3).

Statistical analysis was performed using GraphPad Prism software (Dotmatics), with statistical significance defined as P < .05 (the word “significant” is used exclusively for statistical comparisons that meet this threshold). Pairwise group comparisons were determined by t test. Significance for Kaplan-Meier analysis was determined by the log-rank (Mantel-Cox) test. Receiver-operating characteristic was calculated by the Wilson-Brown method.

Results

Cohort Characteristics

Our cohort included MCC (n = 195 tumors from 171 unique patients); extracutaneous SmCCs (n = 54, including 32 small cell lung carcinomas); ONB (n = 11); HPV-associated SCC, including basaloid and poorly differentiated cases (n = 6); and non-MCC cutaneous tumors with basaloid or poorly differentiated features (n = 42), including BCC, poorly differentiated SCC, sebaceous carcinoma, and trichoblastoma (Table 1, Supplementary Table S1).

Table 1.

Summary of H3K27me3 staining in MCC and other tumors

Diagnosis	n	Average H-score	Fraction with global loss
MCC―all cases	195	145.4	0.08
MCPyV status
Virus-negative MCC	98	147.0	0.14
Virus-positive MCC	96	142.2	0
Phenotypic diversity
Squamous atypia present (combined)	34	119.1	0.29
Squamous atypia absent (pure)	144	154.2	0.04
Not assessed	17	124.0	0
CK20 status
CK20-negative MCC	23	141.9	0.17
CK20-positive MCC	167	145.7	0.07
Non-MCC tumors
Noncutaneous SmCC-all	54	205.1	0
SmCC, lung	32	221.4	0
SmCC, bladder	10	193.0	0
SmCC, cervix	3	235.0	0
SmCC, sinonasal	2	168.8	0
SmCC, GI	2	205.0	0
SmCC, ovary	2	125.0	0
SmCC, breast	1	180.0	0
SmCC, laryngeal	1	70.0	0
SmCC, prostate	1	280.0	0
SmCC, thymus	1	20.0	0
Non-MCC cutaneous tumors―all	43	184.5	0
Basal cell carcinoma	26	187.7	0
Cutaneous SCC, poorly differentiated	7	200.0	0
Sebaceous carcinoma	6	164.6	0
Trichoblastoma	4	166.3	0
Olfactory neuroblastoma	11	288.9	0
SCC, HPV associated	6	166.0	0.17

H3K27me3, histone H3 lysine 27 trimethylation; HPV, human papillomavirus; MCC, Merkel cell carcinoma; MCPyV, Merkel cell polyomavirus; SCC, squamous cell carcinoma; SmCC, small cell carcinoma.

Of MCC tumors with relevant stains, 23 (12%) were CK20 negative (with some cases intentionally included for this feature), and 98 (51%) were MCPyV negative (Table 1). MCC cases included primary tumors (n = 113), metastases to locoregional skin and lymph nodes (n = 76), 1 distant metastasis, and 1 local recurrence. Paired tumors (related tumors from the same patient) included 16 primary-metastasis pairs, 1 primary-recurrence pair, and 7 metastasis-metastasis pairs. For primary tumors, MCC survival and recurrence follow-up was available for 95 cases with a median follow-up time of 41 months (range, 2–182).

Patterns and Clinicopathologic Correlations of Histone H3 Lysine 27 Trimethylation Labeling in Merkel Cell Carcinoma

H3K27me3 displayed 3 overall patterns of staining in MCC: global loss (analogous to the pattern seen in malignant peripheral nerve sheath tumor), mixed positive/negative labeling (ranging from rare positive cells to majority of positive cells), and diffuse labeling (Fig. 1A). MCC overall displayed a mean H-score of 145 (median, 133; standard deviation, 95). A subset of cases (n = 21) across the range of H-scores (including all cases with global loss) was evaluated by both TMA and whole slide section, demonstrating strong correlation in H-score between these approaches (Pearson R² = 0.95). There was no significant difference in mean H-score for H3K27me3 labeling between VP-MCC and VN-MCC (Fig. 1B), although VN-MCC displayed a broader range of H-scores as indicated by higher standard deviation and significantly greater variance (P = .03). Tumors with squamous atypia (combined tumors with SCCIS and/or intratumoral squamous differentiation) had a strong trend toward lower H3K27me3 labeling (mean H-score = 119 vs 154, P = .06) (Fig. 1C). Metastases displayed higher H3K27me3 than primary tumors (mean H-score = 164 vs 131, P = .02) (Fig. 1D). Furthermore, in matched primary-metastasis pairs, the metastasis displayed relatively increased H3K27me3 in 12 of 16 pairs, with relatively decreased labeling in only 2 of 16 pairs. The primary-recurrence pair also displayed relatively increased H3K27me3 in the recurrent tumor.

Figure 1.

Histone H3 lysine 27 trimethylation (H3K27me3)-labeling patterns and associations in MCC. (A) Range of H3K27me3 labeling in MCC, from strong on the left to globally lost on the right. Background lymphocytes and vessels are strongly positive. Panels are ×200 magnification, DAB with hematoxylin counterstain. (B-D) Comparison of H3K27me3 scores in MCC subgroups: (B) Merkel cell polyomavirus (virus negative or virus positive), (C) presence or absence of squamous atypia (squamous cell carcinoma in situ [SCCIS] or dermal squamous differentiation), (D) primary vs metastatic. Each circle represents an individual tumor. Box plots show 10th to 90th percentile with whiskers showing minimum to maximum. *P < .05. MCC, Merkel cell carcinoma.

MCC tumors with global H3K27me3 loss (n = 15 from 14 unique patients) represented an uncommon but distinctive group (Figs. 1 and 2). Considering cases from unique patients, significant associations with global loss in MCC included MCPyV negativity (100%, 14/14, P < .001), squamous atypia (71%, 10/14, P < .001), and sarcomatoid change (14%, 2/14, P = .006) (Fig. 2A–F). Both MCC tumors with sarcomatoid change in our cohort (100%, 2/2) displayed global H3K27me3 loss that affected both neuroendocrine and spindle cell components (Fig. 2A–F). There was a trend toward increased rate of CK20 negativity in this group (29% vs 13% in tumors with retained H3K27me3, P = .11) (Fig. 2H). A primary-metastasis pair displayed concordant global loss of H3K27me3 in both tumors; there were no primary-metastasis pairs with discordant global loss in only 1 tumor of the pair.

Figure 2.

Representative cases of MCC with global histone H3 lysine 27 trimethylation (H3K27me3) loss. (A-F) MCC with squamous and sarcomatoid differentiation. (A, D) Small round cell tumor with associated clusters of atypical squamous cells and mitotically active spindle cell proliferation. (B, E) CK20 labels small cell component, with less definitive labeling of squamous component, and no labeling in spindle cell component. (C, F) H3K27me3 displays diffuse loss in all tumor components, with retained labeling in vessels. (G-L) Virus-negative MCC case with immunophenotypic aberrancy, including negativity for CK20 (H) and neurofilament (I, with nerve included as positive control), as well as patchy labeling for SATB2 with areas of strong (left) and negative (right) expression (J). H3K27me3 displays global loss, with retained labeling of vessels (K). EZHIP is expressed (L). Magnification: ×65 (A-C), ×200 (D-K), ×400 (L). Panels A, D, and G are hematoxylin and eosin (H&E); the remainder are DAB chromogen with the indicated immunohistochemistry, and hematoxylin counterstain. MCC, Merkel cell carcinoma.

We next evaluated cases for potential mechanisms of H3K27me3 dysregulation. Immunohistochemistry for H3K27M (histone 3 mutation that causes global H3K27me3 reduction)^49,50 was negative in all 88 MCC cases examined for this marker, including 13 cases with global loss (Supplementary Table S1). The PRC2 inhibitor EZHIP displayed substantial expression (H-score ≥100) in most tumors with global H3K27me3 loss that had staining results (9/12 [75%]), strikingly higher than MCC tumors with retained H3K27me3 (mean H-score = 193.0 vs 1.5, P < .0001) (Supplementary Fig. S1). In cases with concurrent SCCIS, EZHIP upregulation was also observed in the SCCIS component (Supplementary Fig. S2). Based on re-review of previously reported sequencing results (Supplementary Table S1),^5,12,47 1 case with global loss had a truncating mutation of SUZ12 that might result in H3K27me3 loss.⁵ Remaining tumors with global loss did not have sequencing that comprehensively covered PRC2 or histone 3 genes. Variants of uncertain significance in KDM6B and RBBP7 were detected in tumors with high H3K27Me3 (1 tumor each, Supplementary Table S1).⁵ However, for most tumors, there was no comprehensive sequencing across all relevant genes (Supplementary Table S1). Activating mutations in oncogenic kinases47 were significantly more frequent in tumors with H3K27me3 loss compared with VN-MCC tumors without loss (71% [5/7] vs 8% [3/37], P = .001); although the sample size was small, these tumors were otherwise generally similar to other VN-MCC with regard to TP53 and RB1 mutations (Supplementary Fig. S3).

We examined prognostic significance of H3K27me3 labeling, by dividing primary MCC tumors based on median H3K27me3 H-score in primary MCC tumors (H =120), with “low/retained” expression defined as less than the median and “high/retained” expression defined as equal to or greater than the median. We further separated tumors with global loss given the distinct features of this group. Kaplan-Meier analysis demonstrated significantly better outcome in the retained/low H3K27me3 subgroup compared with the high H3K27me3 subgroup, with regard to overall survival and MCC-specific survival (log-rank P < .01 and P = .03, respectively), with a weak trend for improved recurrence-free survival (P = .16) (Fig. 3). This pattern persisted in subgroup analysis for VP-MCC, with more mixed results for VN-MCC; P values were not significant for these subgroup analyses with the exception of overall survival for VP-MCC (Supplementary Fig. S4). Tumors with global H3K27me3 loss displayed high rates of recurrence and MCC-specific death at 3 years postdiagnosis (100% and 67%, respectively); however, this was a very small group, limiting definitive conclusions.

Figure 3.

Kaplan-Meier analysis of survival outcomes. (A) Overall survival (OS). (B) Disease-specific survival (DSS). (C) Recurrence-free survival (RFS; time to first recurrence or last follow-up). Histone H3 lysine 27 trimethylation (H3K27me3) was defined as globally lost, low (retained but less than median), or high (equal to or greater than to median). Log-rank P values are based upon comparison between low/retained and high expressing groups.

Utility of Histone H3 Lysine 27 Trimethylation for Merkel Cell Carcinoma Diagnosis

We next compared H3K27me3 in MCC with other tumor types. H3K27me3 labeling was significantly lower in MCC than noncutaneous SmCC, ONB, and cutaneous epithelial tumors overall (P < .05 for all comparisons), with a trend toward decreased labeling relative to HPV-associated SCC (Fig. 4). Both primary MCC and metastatic MCC subgroups had significantly lower H3K27me3 than noncutaneous SmCCs (P < .001 and P = .009, respectively). Noncutaneous SmCCs tended to diffuse strong labeling, although some tumors displayed mixed patterning (Fig. 4A, C). ONB had consistently high labeling, significantly greater than all other tumor types in this series (P < .05 for all comparisons) (Fig. 4A, D). All types of cutaneous epithelial tumor displayed higher average H3K27me3 labeling than MCC (Fig. 4A, G-J); however, this only reached statistical significance for BCC, possibly related to smaller sizes of the other groups (Table 1, Fig. 4, and data not shown). Global loss of H3K27me3 was rare in other tumor types within our study set, with only 1 HPV-associated SCC demonstrating this pattern (Fig. 4F). No noncutaneous SmCC displayed global loss.

Figure 4.

Histone H3 lysine 27 trimethylation (H3K27me3) labeling in MCC and diagnostic mimics. (A) Relative H3K27me3 scores in MCC and potential histologic mimics. Each dot represents an individual tumor. Boxplot displays 10th-90th percentile and min to max. (B) Representative example of H3K27me3 labeling in MCC (repeated from Fig. 1). (C-J) Representative examples of H3K27me3 labeling in non-MCC tumors. Background lymphocytes are strongly positive. Panels are at ×200 magnification with DAB chromogen and hematoxylin counterstain. *P < .05. BCC, basal cell carcinoma; MCC, Merkel cell carcinoma; ONB, olfactory neuroblastoma; SCC-HPV, HPV-associated squamous cell carcinoma; SCC-poor diff, poorly differentiated primary cutaneous squamous cell carcinoma; SebCA, sebaceous carcinoma; SmCC, noncutaneous small cell carcinoma; TB, trichoblastoma.

The specificity and sensitivity of H3K27me3 for MCC varied by the threshold used, considering labeling below the cutoff to favor MCC (Table 2^8,19,51). Overall diagnostic performance of H3K27me3 was highly similar to SATB2 (receiver-operating characteristic area under the curve [AUC] 0.68 and 0.65, respectively) (Table 2 and data not shown). CK20, neurofilament, and POU4F3 were superior to both markers considering sensitivity and specificity (Table 2).

Table 2.

Sensitivity and specificity of CK20, POU4F3, neurofilament, SATB2, and H3K27me3 for MCC

Marker	Pattern	Sensitivity (%)	Specificity vs other small cell carcinomas (%)	Specificity vs cutaneous tumors (%)	Comment
CK20	Positive	87a	94a	100a	Based on Tetzlaff and Harms8 and Yang et al51
MCPyV	Positive	Variablea	100	100	Positivity varies by population
POU4F3b	Positive (≥5%)	87	96	100	See Supplementary Table S3
Neurofilament	At least focal (5+ positive cells)	76	97	100	Also see Stanoszek et al19
SATB2	Any expression	73	35	90	Decreased sensitivity in virus-negative MCC
SATB2c	Diffuse moderate/strong expression	22	94	100	Decreased sensitivity in virus-negative MCC
H3K27me3	Negative to intermediate (ie, not diffuse)	67	64	38	Similar sensitivity in virus-negative and virus-positive MCC
H3K27me3	Global loss	8	100	100	Only in virus-negative MCC

H3K27me3, histone H3 lysine 27 trimethylation; MCC, Merkel cell carcinoma; MCPyV, Merkel cell polyomavirus.

^aDue to intentional inclusion of CK20-negative cases in the current study, some percentages are either not reported or based on previous studies, as referenced in the Comment column.

^bDue to intentional inclusion of CK20-negative cases in the current study, percentages for this stain may be lower than in actual practice.

^cFor SATB2, diffuse moderate/strong expression was defined as H-score = ≥200.

The real-world need for additional diagnostic markers in MCC arises when classic markers, such as CK20 or MCPyV, are negative. H3K27me3 displayed similar or better performance relative to SATB2 in CK20-negative MCC cases (Table 3). Notably, both VN-MCC and CK20-negative MCC displayed significantly lower H3K27me3 labeling than SmCC (P < .01 for both comparisons), and diagnostic sensitivity of global H3K27me3 loss was higher for both groups than for MCC overall. By contrast, there was no significant difference for SATB2 expression between VN-MCC (mean H-score = 53) and CK20-negative MCC (mean H-score = 54) relative to SmCC (mean H-score = 48) (P = .70 and .76, respectively). For CK20-negative MCC, receiver-operating characteristic AUC was similar to all MCCs for H3K27me3 (AUC = 0.69) but diminished for SATB2 (AUC = 0.52). In 1 case, global loss of H3K27me3 was observed, but CK20, MCPyV, and neurofilament were negative, and SATB2 expression was patchy (Fig. 2G–K and not shown).

Table 3.

Diagnostic Marker Sensitivity in CK20-negative MCC

Marker	Pattern	Sensitivity (%)
MCPyV	Positive	9.1
Neurofilament	Positive	81.0
POU4F3	Positive	58.8
SATB2	Any expression	54.5
SATB2a	Diffuse expression	13.6
H3K27me3	Any nondiffuse expression	100.0
H3K27me3	Global loss	17.4

H3K27me3, histone H3 lysine 27 trimethylation; MCC, Merkel cell carcinoma; MCPyV, Merkel cell polyomavirus.

^aFor SATB2, diffuse moderate/strong expression was defined as H-score = ≥200.

We next compared H3K27me3 to the recently described diagnostic marker POU4F3.³² POU4F3 displayed strong sensitivity and specificity for MCC overall (Table 2, Supplementary Table S3). However, POU4F3 demonstrated significantly lower sensitivity in CK20-negative MCC relative to CK20-positive MCC (59% vs 90%, P = .003; Table 2, Supplementary Table S3), similar to the previously reported trend.³² POU4F3-negative MCC and POU4F3-positive MCC cases displayed similar H3K27me3 labeling (mean H-score = 143 for both groups). Most tumors with global H3K27me3 loss were positive for POU4F3 expression (8/9, 89%) (Supplementary Table S1).

Finally, we observed that significant EZHIP expression (H-score ≥100) was restricted to MCC in comparison to SmCC and other tumor types in this study, with the exception of some cases with ONB (Supplementary Fig. S1). However, EZHIP expression coincided with global H3K27me3 loss in almost all cases (Supplementary Table S1, Supplementary Fig. S1), thus EZHIP expression was rarely additive after accounting for H3K27me3 results.

Discussion

Epigenetic regulation plays important roles in embryogenesis, tissue homeostasis, and tumor biology.^34,35,38,52 Repressive H3K27me3 mediated by PRC2 (with EZH2 as the active methyltransferase component) suppresses the development of normal Merkel cells from epidermal progenitors during embryogenesis.⁵³ Concordantly, decreasing H3K27me3 labeling has been described in association with progression from SCCIS to VN-MCC, suggesting that decreased PRC2 activity promotes MCC formation,¹² analogous to the role for PRC2 in differentiation of normal Merkel cells. In line with this, EZH2 gain-of-function mutations are rare in MCC.¹ However, higher expression of EZH2 has been described as a negative prognostic marker for MCC.^46,54 EZH2 inhibitors have in vitro activity against cultured MCC cell lines,^41,45,55,56 although data are mixed regarding whether this effect is dependent on PRC2. Furthermore, reports have varied regarding patterns of H3K27me3 labeling in MCC.^40–45 Thus, biological roles and clinical significance of H3K27me3 in MCC remain incompletely characterized.

Our findings suggest that biological role(s) of H3K27me3 in MCC may be complex and variable across cases. A subset of cases displayed global loss similar to that described in other tumor types with mutations in histone 3 or PRC2 genes.^37,49,50 In MCC, such tumors displayed increased phenotypic plasticity with regard to squamous atypia and sarcomatoid transformation. The mechanism for H3K27me3 loss in MCC could be proposed for the majority of cases, with EZHIP overexpression in the majority, and mutation involving a PRC2 complex gene in an additional case. All evaluated cases were negative for the H3K27M mutation. Notably, there was significant association between global H3K27me3 loss and activating mutations involving oncogenes such as PIK3CA, suggesting a potential role for inhibitory phosphorylation of EZH2 by Akt⁵⁷ in a subset of tumors with such mutations. EZHIP expression has been previously reported in association with H3K27me3 loss in MCC,⁴¹ possibly due to promoter hypomethylation.⁵⁸ In contrast to the extremes of labeling displayed by many VN-MCC tumors, VP-MCC tumors tended to show more consistently mixed positive/negative labeling. Interestingly, H3K27me3 was significantly increased in MCC metastases compared with primary tumors. Hence, roles for H3K27me3-mediated epigenetic repression in MCC may vary depending on context. In contrast to a previous report,⁴¹ we did not observe a definitive association between EZH2 and H3K27me3 levels (data not shown) but cannot exclude the possibility of PRC2-independent activity of EZH2⁵⁶ and/or contribution of PRC1 to total H3K27me3 levels.

We investigated the potential diagnostic utility of H3K27me3 for MCC and observed significantly lower levels of H3K27me3 in MCC relative to extracutaneous SmCCs and other potential diagnostic mimics. Global loss was highly specific for MCC, with only 1 HPV-associated SCC displaying global loss among the non-MCC tumors examined in our study. A notable strength of H3K27me3 as a diagnostic marker was consistent sensitivity and specificity in challenging contexts, such as CK20 negativity and MCPyV negativity. This is in contrast to other widely available diagnostic markers for MCC that display reduced sensitivity in that context.^1,24–26 We observed that H3K27me3 labeling was similar in POU4F3-negative MCC tumors relative to MCC overall, indicating that H3K27me3 remains informative for that subset of MCC. The etiology for generally lower H3K27me3 levels in MCC relative to other tumors is not clear; potential mechanisms include oncogene mutations in some tumors, mutational inactivation of PRC2, or increased activity of MYC or MYCL.^56,59–61

There are several practical considerations regarding diagnostic use of H3K27me3. Interpretation of relative levels of staining can be more challenging than a binary (positive/negative) marker. Although reduced H3K27me3 labeling favors MCC, strong labeling might best be considered noncontributory as this pattern can be seen in both MCC and other tumor types. Finally, this marker will not be useful for detecting small foci of disease, such as metastases in sentinel lymph nodes. On balance, we predict H3K27me3 will have greatest utility in an expanded immunohistochemistry panel for challenging cases, rather than as a routine diagnostic marker for MCC. Of note, EZHIP immunohistochemistry is generally not additive after accounting for H3K27me3 results but could be considered as an additional marker if a positive stain is desired to complement or help confirm the finding of global H3K27me3 loss.

We also observed prognostic associations with regard to H3K27me3 labeling and outcomes for MCC. A low (but not lost) level of labeling was associated with improved outcomes as compared with high or lost labeling. Notably, some comparison groups were small. Moreover, clinical implementation of H3K27me3 as a prognostic marker may be practically challenging due to interobserver differences in H-scoring relative to a predefined threshold, and the related complexity of distinguishing among 3 potential results (lost, low, or high). Thus, these findings might have greater relevance to tumor biology rather than real-world prognostication.

Our findings confirm and expand upon previous investigations into H3K27me3 labeling in MCC. Previous smaller studies have reported conflicting results on the associations of H3K27me3 levels relative to viral status or squamous atypia (combined tumors) in MCC.^40–42,44 Notably, our observation that VN-MCC tend to display extremes of H3K27me3 labeling (strong diffuse labeling vs low/no labeling) suggests that previous discordance between studies might reflect differences in scoring approach combined with sampling variation in smaller cohorts. However, our findings that MCC cases with global loss are uniformly MCPyV-negative tumors and associated with squamous atypia are in agreement with previous reports.^41,42 We observed that sarcomatoid change is another form of phenotypic plasticity observed with H3K27me3 loss. EZHIP overexpression has been previously associated with H3K27me3 loss in MCC,⁴¹ possibly related to EZHIP promoter hypomethylation.⁵⁸ Unlike our findings, other studies relying on smaller cohorts have reported that H3K27me3 levels lack apparent prognostic significance in MCC.^40,42 Larger studies that consider MCPyV status⁴⁷ or include a sufficient number of cases with H3K27me3 global loss might be helpful in further investigating this question. Overall, despite some conflicting results, the balance of evidence supports distinct roles for H3K27me3 in the biology of VP-MCC and VN-MCC.^12,40,41,62

Our findings on H3K27me3 patterns in non-MCC cutaneous tumors are relatively similar to those reported by Durand et al.⁴¹ Durand et al also described global loss of H3K27me3 in some cases of poorly differentiated cutaneous SCC (IS), which we did not identify in our cohort. However, the HPV-associated SCC with global H3K27me3 loss among our cases was also poorly differentiated, suggesting that global H3K27me3 loss may be related to differentiation rather than HPV in such cases.

To our knowledge, this is the first systematic evaluation of H3K27me3 patterns in certain other tumor groups in our study: specifically, extracutaneous SmCCs and ONB. H3K27me3 labeling in ONB was among the highest of all tumor types in our series; we did not identify previous publications in the English literature on H3K27me3 patterns in this tumor type. Strong H3K27me3 labeling has previously been reported in small cell lung carcinoma tumors and cell lines, although these studies did not provide details on scoring or sample sizes.^63,64 Interestingly, inhibition of H3K27me3 has been shown to induce a neuroendocrine phenotype in the context of prostate carcinoma⁶⁵; however, the clinical significance of this observation remains unclear, as the single case of prostatic SmCC in our series displayed strong H3K27me3 labeling, and to our knowledge, this has not been rigorously examined in prostatic SmCC tumors by other studies.

Our study has several limitations. An inherent limitation of tissue-based biomarker studies is a tendency toward relatively larger tumors; thus, outcomes in our cohort may not reflect those in MCC overall. Furthermore, we intentionally included certain MCC subtypes (such as CK20-negative MCC) that might also influence cohort features, such as rate of MCPyV positivity and clinical outcomes,^1,24,47 relative to MCC overall. An associated limitation is that our study design does not allow for direct comparison of the diagnostic utility of H3K27me3 with MCPyV and CK20. We evaluated retrospective outcomes from a pre-immunotherapy time frame; however, immunotherapy has a major role in the management of an advanced disease; thus, it is not clear how this would impact time to first recurrence (the significant outcome in our study). Although our overall cohort size is robust, some groups (such as subtypes of SmCC or cutaneous carcinoma) are relatively small. Conclusions from TMAs may be hindered by tumor heterogeneity; however, we observed overall good correlation between cores and whole tissue sections, and all cases with global loss were consistently confirmed by whole sections. Our quantitation approach (H-score) is not feasible to directly implement clinically but can be easily generalized to strong/diffuse, weak/variable, and null staining patterns. Finally, as noted above, we have limited molecular data to confirm the mechanism for differing H3K27me3 patterns in most tumors.

In conclusion, we find that MCC tumors display distinct patterns of H3K27me3 labeling ranging from diffuse strong positivity to global loss. Overall, lower H3K27me3 expression favors MCC relative to other tumor types in this study, and global loss is uncommon but highly specific for MCC. Lower H3K27me3 labeling remains a consistent finding in immunophenotypically aberrant cases of VN-MCC, supporting the diagnostic utility of this marker in challenging cases.

Supplementary Material

The online version contains supplementary material available at https://doi.org/10.1016/j.modpat.2025.100945.