Article Text

other Versions

Stromal keratin expression in phyllodes tumours of the breast: a comparison with other spindle cell breast lesions
  1. Yifoong Chia1,
  2. Aye Aye Thike1,
  3. Poh Yian Cheok1,
  4. Luke Yong-Zheng Chong1,
  5. Gary Man-Kit Tse2,
  6. Puay Hoon Tan1
  1. 1Department of Pathology, Singapore General Hospital, Singapore
  2. 2Department of Anatomical and Cellular Pathology, Chinese University of Hong Kong, Hong Kong, SAR China
  1. Correspondence to Dr Puay Hoon Tan, Department of Pathology, Singapore General Hospital, Outram Road, Singapore 169608, Singapore; tan.puay.hoon{at}


Aim To determine the frequency, pattern and distribution of stromal keratin expression in phyllodes tumours if any, which may impact diagnostic approaches.

Methods The clinicopathological features of 109 phyllodes tumours comprising 70 (64.2%) benign, 30 (27.5%) borderline and nine (8.3%) malignant grades were evaluated, and the immunohistochemical expression of a keratin panel (MNF116, 34βE12, CK7, CK14, AE1/3, Cam5.2), p63 and CD34 in their stromal component was assessed.

Results There was focal and patchy cytoplasmic keratin staining in 1–5% of stromal cells in 13 (11.9%), 24 (22%), 31 (28.4%), 2 (1.8%), 9 (8.3%) and 2 (1.8%) cases for MNF116, 34βE12, CK7, CK14, AE1/3, Cam5.2, respectively. CD34 was expressed in 79 (72.5%) cases. There was no stromal staining for p63. Stromal MNF116, 34βE12 and Cam5.2 reactivity was significantly associated with phyllodes tumour grade (p=0.027, p=0.034, p=0.009 respectively), while MNF116 stromal staining was observed in tumours with increasing cellularity (p=0.036), necrosis (p=0.015) and cystic change (p=0.048). Contrary to common understanding, these findings confirm that stromal cells in phyllodes tumours can sometimes express keratins, albeit focal and in a patchy distribution. In comparison, fibromatosis and dermatofibrosarcoma were uniformly negative for the same keratin panel, while spindle cell components of eight metaplastic carcinomas expressed at least two or more keratins in a wider distribution of up to 90% of positively stained spindle cells. All eight spindle cell sarcomas were negative for keratins.

Conclusion The use of keratins as an adjunctive immunohistochemical diagnostic tool in the differential work-up of spindle cell tumours of the breast has to be interpreted with caution especially on limited core biopsy material.

  • Breast cancer
  • breast pathology
  • broad spectrum keratins
  • immunohistochemistry
  • phyllodes tumours
  • prostate
  • spindle cell tumours
  • stromal cells

Statistics from

Phyllodes tumours are uncommon fibroepithelial neoplasms of the breast, representing approximately 1% of all breast tumours and classified into benign, borderline and malignant categories based on a constellation of histological characteristics that include the degree of stromal hypercellularity, stromal atypia, stromal overgrowth, mitotic activity and the nature of the lesional border.1

The epithelial component of phyllodes tumours is typically observed as elongated epithelium-lined clefts. Some degree of epithelial hyperplasia may be encountered. The stromal component can be variably cellular and accompanied sometimes by prominent stromal myxoid change. In malignant phyllodes tumours, besides stromal hypercellularity, atypia, mitoses, overgrowth and permeative margins, stromal alterations can incorporate malignant heterologous elements.2

At the ultrastructural level, stromal cells of phyllodes tumours have features of myofibroblasts and fibroblasts.3 4 Cytokeratin expression is usually not seen in the stromal component of phyllodes tumours in contrast to metaplastic carcinoma/spindle cell carcinoma, in which keratins are demonstrated in the abnormal spindle cells, and this difference has been used in the adjunctive diagnostic armamentarium for distinguishing these biologically distinct lesions.4–7

While it is generally accepted that stromal cells in phyllodes tumours are negative for keratins, there have been few reports that formally investigate their expression for a wide spectrum of keratins in a substantial number of phyllodes tumours. In this study, we apply an immunohistochemical panel of keratins that include those of broad spectrum to a series of phyllodes tumours of different grades to determine the frequency, pattern and distribution of keratin expression if any, in order to attain a better understanding that can assist future diagnostic approaches. A comparison is made with other morphological spindle cell tumours of the breast—low-grade spindle cell lesions, spindle cell sarcoma and spindle cell elements of metaplastic carcinoma.

Materials and methods

Phyllodes tumours

One hundred and nine phyllodes tumours of the breast diagnosed at the Department of Pathology, Singapore General Hospital between January 2007 and December 2009 were included in the study, which was approved by the Institutional Review Board. Patient details and tumour laterality were determined from accession forms. Histological slides were retrieved and reviewed for morphological parameters including microscopic borders, stromal hypercellularity, stromal cell atypia, stromal overgrowth, mitotic activity, myxoid changes, microscopic haemorrhage, necrosis, cystic changes and stromal metaplasia. Criteria were as previously described.2 Briefly, a circumscribed border was defined as a pushing margin that bulged against the surrounding breast tissue; a permeative border was one with irregular extensions of tumour into adjacent tissue. Stromal hypercellularity and cytological atypia were categorised as mild, moderate and severe. Stromal mitotic activity was quantified per 10 high power fields (HPF) of the microscope objective (×40 objective and ×10 eyepiece, 0.196 mm2) in the most mitotically active areas of the stroma. Stromal overgrowth was determined on a low power field (×4 microscope objective and ×10 eyepiece, 22.902 mm2) and was present when there was only stroma without epithelial elements. A benign phyllodes tumour was one with pushing margins, mild or moderate stromal hypercellularity, mild or moderate stromal cytological atypia, occasional mitoses that numbered up to four per 10 HPF and no stromal overgrowth. A malignant tumour showed marked stromal hypercellularity and cytological atypia, stromal overgrowth, brisk mitotic activity (10 or more per 10 HPF) and permeative margins; the finding of a malignant heterologous element placed the tumour into a malignant category. While the presence of low-grade myxoliposarcoma areas have been accorded borderline/low-grade status by some authors,8 our experience with liposarcomatous zones in phyllodes tumours has been those that harbour distinctively pleomorphic lipoblasts, which we classified as malignant. The borderline group showed some but not all the characteristics observed in malignant lesions.

Low-grade spindle cell lesions

Four cases of dermatofibrosarcoma, eight cases of fibromatosis and one myofibroblastoma were subjected to the same panel of immunohistochemical markers. The cases of dermatofibrosarcoma were included as they presented as breast masses for which there was clinical suspicion of breast malignancy.

Spindle cell sarcoma

Eight cases classified as primary breast sarcoma were identified from our pathology files and used in the study. One of these was an angiosarcoma, while the remainder were spindle cell sarcomas, not otherwise specified (NOS). The latter tumours did not demonstrate any epithelium lined clefts on detailed histological examination.

Metaplastic carcinoma (spindle cell component)

Eight cases of metaplastic carcinomas with spindle cell components were included. Only the spindled zones were assessed for immunohistochemical expression.


Sections (4 μm) were cut from representative blocks and fished onto coated slides (POLYSINE; Menzel-glaser, Braunschweig Germany) in a similar orientation to facilitate faster evaluation. The slides were then baked in the oven overnight at 55°C to enhance adhesion of the sections to the slides. Antibodies to cytokeratins (MNF116, 34βE12, CK7, CK14, AE1/3, Cam5.2), CD34 and p63 were applied to the sections. Details of antibodies are shown in table 1. The detection system used was Dako Real EnVision, K5007 on the Dako Autostainer (Dako, Glostrup, Denmark). After that, sections were counterstained with haematoxylin, dehydrated and mounted in depex.

Table 1

Details of antibodies

Only the stromal component of the phyllodes tumours was assessed. Any unequivocal cytoplasmic staining of at least 1% of stromal cells for MNF116, 34βE12, CK7, AE1/3, Cam 5.2, and CK14 was considered positive. For p63, any nuclear staining of at least 1% of stromal cells was regarded as positive. The intensity of staining was categorised as nil, mild, moderate and marked (0, 1+, 2+, 3+), respectively. The proportion of stromal cells stained was noted. The pattern of stromal cell staining was deemed subepithelial if the location of positively stained cells occurred in the stroma adjacent to epithelium; within the frond when positive cells were located in the central stromal portion of the frond away from the epithelium; and peripheral when positive staining was found in stromal cells at the edge of the tumour. Benign epitheliun and myoepithelium in the sections served as internal positive controls. For CD34, the immunoreactive score (H-score) was calculated as: (3× % strong staining) + (2× % moderate staining) + (1× % weak staining), with positive biomarker expression defined as an immunoreactive score of 50 and above;9 vessel walls were used as internal positive controls.

Statistical analysis

The findings were analysed using statistical software SPSS for Windows, V.16. The relationship between clinicopathological parameters, p63, CD34 and keratin immunohistochemical detection was tested using the χ2 and Fisher's exact tests. A p value of less than 0.05 was considered a significant result.


Phyllodes tumours

Clinicopathological parameters of 109 women with phyllodes tumours of the breast are shown in table 2. The study group comprised 70 (64.2%) benign, 30 (27.5%) borderline and nine (8.3%) malignant tumours classified according to stated criteria. All but one tumour represented primary diagnoses of phyllodes tumours in the breast. One malignant tumour was a left chest wall recurrence in a woman with a previous diagnosis of phyllodes tumour. The right breast was affected in 63 (57.8%) cases and the left breast (including the chest wall recurrence) in 46 (42.2%) cases.

Table 2

Clinicopathological characteristics of 109 cases of phyllodes tumours of the breast

Two cases with stromal metaplastic changes were malignant phyllodes tumours with heterologous chondrosarcoma and liposarcoma. When positive, MNF116, 34βE12, CK7, CK14, AE1/3, Cam5.2 immunohistochemical expression was focal and patchy in nature, and observed in 1–5% of stromal cells of 13 (11.9%), 24 (22.0%), 31 (28.4%), two (1.8%), nine (8.3%) and two (1.8%) tumours, respectively (figures 1–3). No immunohistochemical staining for p63 was seen in stromal cells of any case, while CD34 staining was noted in 79 (72.5%) phyllodes tumours (table 3). Stromal MNF116 and 34βE12 immunohistochemical expression, including staining intensity, was associated with tumour grade (p=0.049, p=0.02). Although there was a statistically significant association of Cam5.2 with tumour grade, only one case each in the borderline and malignant categories showed stromal Cam5.2 staining, reducing the strength of this correlation. Table 4 shows details of keratin immunohistochemical staining compared with phyllodes tumour grade. In addition, MNF116 stromal staining was observed in tumours with increasing stromal cellularity (p=0.036), necrosis (p=0.015) and those showing cystic changes (p=0.048). Immunohistochemical interpretation was not performed on necrotic or cystic areas. No other correlations were observed between pathological parameters and stromal keratin reactivity.

Figure 1

Low and medium magnification of a malignant phyllodes tumour with stromal hypercellularity and overgrowth (A, B). Positive immunohistochemical stromal staining for MNF116, CK7 and 34βE12 (C, D, E) was observed only at the periphery of the tumour. Other parts of the tumour were negative. Staining for 34βE12 was noted in only very few stromal cells, arrowed. (F) Shows slightly higher magnification of 34βE12 staining of stromal cells.

Figure 2

Low and medium magnification of a borderline phyllodes tumour with areas of haemorrhage and scattered plump stromal cells within the central portions of the fronds (A, B). Elsewhere, there was increased stromal cellularity and mitotic activity. Positive immunohistochemical stromal staining for MNF116, CK7 and 34βE12 (C, D, E) was observed in the plump stromal cells.

Figure 3

Subepithelial keratin immunohistochemical staining in a case of borderline phyllodes tumour. (A) Haematoxylin and eosin stained section showing a leafy frond. MNF116, CK7 and 34βE12 (B, C, D) are focally observed decorating stromal cells near the benign epithelial component (arrows, circles). (E) Shows higher magnification of CK7 positivity in stromal cells of another tumour, which demonstrates negative staining in endothelial cells of small vessels observed in the same field.

Table 3

Immunoreactivity of biological markers in phyllodes tumour of the breast (N=109)

Table 4

Association of keratin immunohistochemical staining of stromal cells with grade of phyllodes tumour

For cases that demonstrated stromal keratin staining, the number of keratins reacting per case showed an increasing trend with higher tumour grades, p=0.082 (table 5).

Table 5

Association of keratin expression with grade of phyllodes tumour

When the location of keratin-positive stromal cells was assessed, there appeared to be a fairly similar distribution in subepithelial and peripheral zones with those reacting within the cores of stromal fronds of phyllodes tumours, without any statistically significant correlations (table 6).

Table 6

Pattern of distribution of keratin immunohistochemical expression in phyllodes tumours

CD34 stromal staining in phyllodes tumours was seen in 79 (72.5%) cases overall, with 55 (78.6%) of 70 benign, 20 (66.7%) of 30 borderline and four (44.4%) of nine malignant grades revealing positive stromal reactivity, a statistically significant association (p=0.023).

There was no stromal staining observed for p63 in any case.

Low-grade spindle cell lesions

Of a total of 13 tumours within this category, the majority of eight cases were fibromatosis, with four cases of dermatofibrosarcoma and one case of myofibroblastoma. There was no staining for the entire cytokeratin panel, nor was there reactivity for p63 in any case. For CD34, all four dermatofibrosarcomas, the single myofibroblastoma, and two of eight fibromatoses disclosed reactivity.

Spindle cell sarcoma NOS

The histological diagnoses of these tumours were based primarily on exclusion. The absence of an accompanying malignant epithelial component (which would otherwise be categorised as a metaplastic carcinoma) and the lack of benign epithelium-lined fronds (which would implicate a phyllodes tumour) were important microscopic criteria.

Of eight spindle cell sarcomas, none showed staining for any keratins. There were two cases with positive p63 staining and one case with CD34 reactivity.

Spindle cell component of metaplastic carcinomas

Spindle cell components of eight metaplastic carcinomas expressed AE1/3 in five (62.5%), Cam5.2 in six (75%), CK7 in three (37.5%), 34βE12 in two (25%) and MNF116 in seven (87.5%) cases. CK14 immunoreactivity was observed in the spindle cell component of all eight metaplastic carcinomas. p63 immunoexpression was seen in two (25%) cases (figure 4). CD34 was not expressed. Table 7 details the immunohistochemical results in the spindle cell component of all eight cases of metaplastic breast carcinoma.

Figure 4

Spindle cell component of metaplastic carcinoma. (A) Haematoxylin and eosin stained section showing sheets and fascicles of spindled cells. (B) Higher magnification reveals plump spindled cells with vesicular nuclei and occasional mitoses. Immunohistochemical reactivity is noted for (C) CK7 and (D) p63. (E) Metastasis to an axillary lymph node is composed also of spindle cells.

Table 7

Immunohistochemical staining results of the spindle cell component of metaplastic breast carcinomas

Tables 8–10 summarise the immunohistochemical expression of keratins, p63 and CD34, respectively, in phyllodes tumours, spindle cell sarcomas NOS, spindle cell components of metaplastic carcinoma and low-grade spindle cell lesions.

Table 8

Immunohistochemical expression of keratins in phyllodes tumours, spindle cell sarcoma NOS, spindle cell components of metaplastic carcinoma and low-grade spindle cell lesions of the breast

Table 9

Immunohistochemical expression of p63 in phyllodes tumours, spindle cell sarcoma NOS, spindle cell components of metaplastic carcinoma and low-grade spindle cell lesions of the breast

Table 10

Immunohistochemical expression of CD34 in phyllodes tumours, spindle cell sarcoma NOS, spindle cell components of metaplastic carcinoma and low-grade spindle cell lesions of the breast


Phyllodes tumours of the breast are biphasic neoplasms, in which the stromal component is considered the neoplastic element while the epithelium is benign and polyclonal.10 The stroma of phyllodes tumours is therefore able to undergo malignant progression into sarcoma.10–14 While there has been much work invested into interrogating biological markers in stromal cells of phyllodes tumours,15–22 scant data on keratin expression have suggested that it is in general absent in phyllodal stromal cells, and can thus be used as a reliable marker for distinguishing spindle cell and sarcomatoid metaplastic breast carcinomas from malignant phyllodes tumours and primary breast sarcomas.3 23 24 Santini et al24 found keratin expression in carcinomatous and pseudosarcomatous components of metaplastic carcinomas and epithelial but not stromal components of phyllodes tumours, and concluded that a combination of intermediate filaments—keratin, vimentin and desmin, could differentiate metaplastic carcinomas from stromal sarcomas and phyllodes tumours. Aranda et al3 evaluated 28 phyllodes tumours and confirmed that keratins (AE1/3 and Cam 5.2) and epithelial membrane antigen (EMA) were uniformly negative in stromal cells. Dunne et al23 applied a wide panel of keratins (AE1/AE3, 34βE12, CK5 and CK14, Cam5.2, CK7 and CK19, EMA) for which all were negative in stromal components of 26 phyllodes tumours. To the best of our knowledge, there has been only one paper by Auger et al25 that documented the presence of stromal keratin expression in a malignant phyllodes tumour, and the authors suggested that this was related to myoepithelial differentiation in stromal cells of that case. Those studies have incorporated small numbers, and it is with the aim of validating their observations that we embarked on the current work to investigate a larger cohort of phyllodes tumours using a wide panel of commonly used, commercially available, keratins MNF116, 34βE12, CK7, CK14, Cam5.2 and AE1/3.

Contrary to these studies, we discovered expression of keratins MNF116, 34βE12, CK7, CK14, AE1/3, Cam5.2 in stromal cells of 11.9%, 22%, 28.4%, 1.8%, 8.3% and 1.8% of phyllodes tumours, respectively, albeit the positivity was focal and patchy and found in only 1–5% of stromal cells in these cases. The explanation for keratin expression is uncertain. While myoepithelial differentiation was suggested as the reason for keratin being observed in stromal cells of a single case of malignant phyllodes tumour by Auger et al,25 this is unlikely to be the cause in our study because p63, a marker of myoepithelial cells, was negative in all cases. Cytokeratin intermediate filaments characterise cells of epithelial origin.26 It has been reported, nevertheless, that non-epithelial cells can synthesise cytokeratins under special circumstances.26 27 Guo et al28 described the co-expression of α-smooth muscle actin and cytokeratin Cam5.2 in myofibroblasts in the vicinity of gastroduodenal ulcers, and proposed that these were ‘novel’ myofibroblasts. Jahn et al29 observed cytokeratins 8 and 18 in a subset of vascular smooth muscle cells in developing atherosclerotic plaques, and suggested that these were ‘dedifferentiated’ smooth muscle cells. In another study on peritoneal fibroblasts, cytokeratin expression was found in those belonging to patients with continuous peritoneal dialysis, which the authors attributed to mesothelial to myofibroblastic conversion, and commented that the immunophenotype of fibroblasts may vary during the process of fibrosis.30 Among neoplasms, aberrant expression of cytokeratin and EMA has been well documented in leiomyosarcomas, with diffuse reactivity being a diagnostic pitfall.31 It would therefore appear that cytokeratin expression in fibroblasts, myofibroblasts and vascular smooth muscle cells can occur in processes of injury and repair, and that it can also aberrantly manifest in leiomyosarcomas. Our finding of an association between stromal MNF116 reactivity with phyllodes tumours harbouring necrosis and cystic change may suggest a milieu of reparative changes in these tumours that can perhaps explain the acquisition of keratin expression in their stromal fibroblasts and myofibroblasts. The increasing expression of keratins with higher tumour grade may reflect aberrant expression by abnormal stromal cells as they progress towards sarcoma.

It may be argued that subepithelial staining of apparent stromal fibroblasts for cytokeratins may be an observational anomaly, and that these cytokeratin-positive cells may in fact be outpouchings of the adjacent epithelial/myoepithelial component. We specifically excluded any such cases of epithelial undulations, focusing only on cells with spindled appearances. The accompanying absence of p63 positivity in these same cells precludes the consideration of myoepithelial ‘artifact’.

We used a relatively wide panel of commercially available keratin antibodies in this study, to reflect the range of antibodies that may be used in daily practice: AE1/3 (detecting a broad spectrum of keratins); MNF116 (incorporating low and intermediate weight keratins), both of which reliably stain formalin-fixed paraffin embedded tissues;32 34βE12 (a cocktail of high molecular weight keratins); CK7, which marks breast ductal epithelial cells; CK14 (a high molecular weight keratin) and Cam5.2 (decorating mainly low molecular weight keratin 8). CK7 is known also to potentially decorate endothelial cells, and we have been diligent in ensuring that we were not mistaking endothelial cell reactivity for stromal positivity in this study. The differential diagnoses of a spindle cell tumour of the breast include the major categories of spindle cell metaplastic carcinoma, stroma dominant phyllodes tumour and primary stromal sarcoma. Spindle cell metaplastic cancers will be decorated not only by MNF116 but their tendency to be basal like render their reacting with 34βE12 and CK14 highly probable.33 34 In our study, the spindle cell components of all eight metaplastic cancers reacted with CK14. In a comprehensive analysis of 29 cases of spindle cell sarcomatoid carcinomas of the breast, Carter et al35 recommended the use of MNF116, CK14 and 34βE12 to identify these tumours.

An aim in examining the expression of keratins in stromal cells of phyllodes tumours is to understand the extent to which we can rely on keratin reactivity of stromal-like cells to point towards a diagnosis of metaplastic carcinoma and away from that of phyllodes tumour. Although a benign phyllodes tumour should not ordinarily cause diagnostic difficulty in distinguishing from spindle cell metaplastic carcinoma, hypercellular stroma dominant phyllodes tumours can mimic the latter, and not infrequently pose a histological dilemma. Our findings suggest that focal keratin reactivity of stromal cells can be found in phyllodes tumours of all grades, with increasing expression in more sinister lesions. As such, focal keratin detection in a worrisome spindle cell breast tumour has to be interpreted with caution, and a traditional approach of wider sampling of the lesion to identify the leafy fronds of a phyllodes tumour or the malignant epithelial counterpart of a metaplastic carcinoma has to be reiterated.

Core biopsies, however, do not have the advantage of the entire breast tumour being available for additional sampling, and it is in this setting that even more restraint needs to be exercised in interpreting focal keratin staining of stromal cells in a spindle cell neoplasm. Unless the elongated benign epithelial clefts of a phyllodes tumour, or the malignant unequivocally epithelial elements of a metaplastic carcinoma are present, it will be prudent to convey the various underlying diagnostic possibilities on core biopsy material without further definitive qualification, because a more specific conclusion if erroneous can potentially lead to divergent therapeutic approaches.

An unusual and rare situation of ductal carcinoma in situ can be found in malignant phyllodes tumours,36 and it becomes an academic question as to whether keratin-positive spindle cells in these lesions represent those of a metaplastic carcinoma, especially if the positive staining is in ‘subepithelial’ zones. This argument may extend to our current cases of phyllodes tumour with focal stromal keratin expression, as the possibility of this being a metaplastic phenomenon also cannot be completely ruled out. If so, should such lesions be considered as a collision of a metaplastic spindled epithelial process, that is, being ‘hosted’ within a phyllodes tumour?

Regardless of philosophy, our study discloses focal and patchy keratin expression in some phyllodes tumours of all grades. The presence of keratin positivity, especially if focal, to conclude a metaplastic carcinoma in small biopsies of a spindle cell lesion, may not therefore be dependable. In adequate material, diffuse keratin positivity of spindle tumour cells, together with p63 positivity and CD34 negativity can support a metaplastic epithelial origin.

Take-home messages

  • Phyllodes tumours of the breast may express keratins in stromal cells on immunohistochemistry, albeit focal and patchy in distribution.

  • CD34 stromal staining is associated with phyllodes tumour grade, with benign tumours showing a higher proportion of positivity.

  • None of the phyllodes tumours display stromal p63 reactivity.

  • On limited material such as core biopsies, focal keratin expression of a spindle cell breast tumour needs to be interpreted with caution, and should not be immediately concluded as a tumour of metaplastic origin.


View Abstract


  • Funding This study was supported by a grant from the Singapore Cancer Syndicate MS04R.

  • Competing interests None.

  • Ethics approval Ethics approval was provided by the Centralised Institutional Review Board, SingHealth.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement The data in the current paper have not been shared with any other party.

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.