Basal-Like Phenotype in a Breast Carcinoma Case Series from Sudan: Prevalence and Clinical/Pathological Correlations

Basal-like breast cancer, an aggressive subtype associated with high grade, poor prognosis, and younger age, is reported frequently in Africa. We analyzed the expression of the basal cytokeratins (CKs) 5/6 and 17 in a case series from Central Sudan and investigated correlations among basal CK status, ER, PgR, and Her-2/neu, and individual/clinicopathological data. Of 113 primary breast cancers 26 (23%), 38 (34%), and 46 (41%) were, respectively, positive for CK5/6, CK17, and combined basal CKs (CK5/6 and/or CK17). Combined basal CK+ status was associated with higher grade (P < .03) and inversely correlated with ER (P < .002), PgR (P = .004) and combined ER and/or PgR (P < .0002). Two clusters based on all tested markers were generated by hierarchical cluster analysis and k-mean clustering: I: designated “hormone receptors positive/luminal-like” and II: designated “hormone receptors negative”, including both basal-like and Her-2/neu+ tumors. The most important factors for dataset variance were ER status, followed by PgR, CK17, and CK5/6 statuses. Overall basal CKs were expressed in a fraction of cases comparable to that reported for East and West African case series. Lack of associations with age and tumor size may represent a special feature of basal-like breast cancer in Sudan.

The basal-like subtype overlaps, but is not synonymous, with the triple negative subset, which includes BCs that do not express ER, PgR, and Her-2/neu and tend to occur at a younger age and in patients with pathogenetic BRCA1 mutations [18][19][20][21]. Approximately 85% of the ER-/Her-2/neu-BCs are of basal-like phenotype [9]. Most importantly, although most basal-like BCs do not express ER, PgR, or Her-2/neu, in case series of different origin 14% to 45% of the cases were reported to express at least one of these markers [7,9,14].
Basal-like BCs show common as well as heterogeneous morphologic, genetic, and immunophenotypic features, and, up to date, there is no international consensus regarding their exact definition [5][6][7][8][9][10][11][12]20]. Basal CKs, which have been shown to be independently associated with poor outcome [7,9,[24][25][26], are expressed in most, but not all, BCs classified as basal-like by immunohistochemical (IHC) or gene microarray analysis [3,7,20,[27][28][29]. Furthermore in a subset of BCs basal CKs are coexpressed with other markers, including EGFR, P-cadherin, c-KIT, caveolin 1, and p63, although consideration of such markers does not appear to improve the identification of the cases with poor outcome compared to basal CKs alone [20]. Therefore Rakha et al. [20] suggested to rely on basal CK expression alone to define basal-like BC, remarking that, in spite of shared clinicopathologic and IHC features, basal CK-positive BCs and basal-like BCs are not strictly the same entity [7,29].
Genetic, ethnic, and racial factors influence BC phenotypes, possibly by determining intrinsic differences in tumor biology [6,30,31]. In this regard, it is remarkable that basal-like/triple negative BC appears to be more common in African American women [6,12,32] and in BC case series from West and East Africa (range: 22%-34%), where it seems to be also associated with features indicative of poor prognosis [33][34][35][36].
In a previous study we found that a BC case series from Khartoum, Central Sudan, was comparable to one from Milan, Northern Italy, in combined hormone receptors status and BC subtypes [37]. Relative to the Italian patients, the Sudanese patients were younger and their tumors were larger, of higher grade and more advanced in stage [37].
We address here the question of the BC subtypes identified by clustering analyses within the Sudanese BC case series. To this end, we re-evaluated, using more sophisticated statistical analyses, the expression of the basal CKs 5/6 (CK5/6) and 17 (CK17) in relation to estrogen/progesterone receptors (ER/PgR), human epidermal growth factor receptor 2 (Her-2/neu), and the available clinicopathological and individual  data. We refer in this paper to two designations of BCs with basal subtype: (i) basal CK+, defining BCs that express basal CKs regardless of the expression of other markers [20] and (ii) basal-like, identified by the triple-negative CK-positive profile (ER−/PgR−/Her-2/neu−/basal CK+).    [39][40][41]. Lack of data on lymph node status and follow up precluded correlations with stage and prognosis [37]. According to data from the Sudan Federal Ministry of Health, 78% of the Sudanese BC patients have stage III or IV disease [42,43].

Immunohistochemistry.
Whole consecutive sections were immunostained for ER (clone 1D5, Dako), PgR (clone PgR 636, Dako), Her-2/neu (polyclonal, Dako), CK5/6 (clone D5/16 B4, Dako), CK17 (clone E3, Dako) and, as quality controls of antigenic preservation, for the CK pool (clones AE1-AE3, Dako) and vimentin (clone V9, Dako). IHC  : Scree plot of the eigenvalues. The adopted extraction methods were the Kaiser criterion, that is, the sum of squared factor loadings (eigenvalue) >1, and the scree test, that is, the place where the smooth decrease of eigenvalues appears to level off to the right of the plot of the eigenvalues. results were recorded as percentages of immunostained cells in ≥2000 neoplastic cells. Only nuclear reactivity was taken into account for ER and PR, which were classified as negative, when absent or present in <5% of the neoplastic cells, or positive, when present in ≥5% of the neoplastic cells. Only intense and complete cell membrane immunoreactivity in ≥10% of the cells was taken as evidence of Her-2/neu overexpression (score 3+) [44]. Borderline Her2/neu cases (score 2+) were reassessed by fluorescence in situ hybridization (FISH), as previously described [37]. Basal CKs 5/6 and 17 were regarded as positive when any cytoplasmic and/or cell membrane staining was seen [6,9,37].

Statistical Analyses
Unsupervised hierarchical cluster analysis (CA) was done for hormone receptors (ER, PgR), Her-2/neu and basal CK (CK5/6 and/or CK17) statuses to determine the natural clustering of the BCs according to the studied IHC markers. CA was performed using squared Euclidean distance measurements to obtain a dissimilarity matrix. Ward's method was then applied to this matrix to build a tree [45]. This method uses analysis of variance to evaluate distances between clusters, minimizing the sum of squares of any two hypothetical clusters that can be formed at each step. CA was done using SPSS statistical package version 15.0 (SPSS Inc., Chicago, IL).
Unsupervised k-mean clustering algorithm, performed with STATISTICA 7.0 (StatSoft, Inc., Tulsa, Ok), was applied to confirm and explore better the generated cluster(s). The kmean clustering used the Euclidean distance as the similarity metric [46]. Data reduction was done by factor analysis, applying principal components analysis (PCA) to the selected variables (ER, PgR, Her-2/neu, CK5/6, and CK17) to determine  [49,50]. All cut-off values were determined before the statistical procedures. Correlations between different variables were calculated using χ 2 test or t-test. Significance was set at <.05.

Cluster Distribution and Factor
Analysis. Two major clusters of patients were generated using hierarchical cluster analysis (Figure 1(a)): cluster I with 65/113 (57.5%) patients   (Figure 1(b)). Hence, cluster I could be designated as "hormone receptors positive/luminal-like," whereas cluster II as "hormone receptors negative," including both the basallike and the Her-2/neu+ subtypes [5,22,26]. Comparable results were obtained through k-mean clustering, with 72/113 (63.7%) patients joining cluster I and 41/113 (36.3%) cluster II (Figure 1(c)). In addition kmean clustering revealed that hormone receptors (ER, PgR) and basal CKs (CK 5/6, CK 17) played a major role in identifying clusters I and II, respectively (Figure 1(d)). On the other hand, Her-2/neu played quite similar roles in the determination of the two clusters, with slightly higher weight in cluster II (Figure 1(d)).
Factor analysis showed that three factors explained 80.3% of the dataset variance (Figures 2(a) and 2(b)). The first factor (eigenvalue = 2.1) accounted for the largest proportion of variance (42.3%) and corresponded to hormone receptor status (loads: ER: 0.80; PgR: 0.78), while basal CKs (loads: CK17: −0.6: CK5/6, −0.59) and Her-2/neu (load: −0. 39) statuses were negatively loaded on this factor. The second factor (eigenvalue = 1.2) explained 23.4% of variance and corresponded to basal CK status (loads: CK17, 0.55; CK5/6, 0.54), while Her-2/neu status (load: −0.68) loaded negatively on this factor. The third factor, corresponding to Her-2/neu status (eigenvalue = 0.7, with a load of 0.6), explained 14.6% of the variance (Figure 2(a)). Individual factor scores of the extracted factors are shown in Figure 2(b). Other two factors needed to be extracted to explain the complete dataset variance, that is, factor 4, corresponding to CK17 status (eigenvalue = 0.6, load: 0.56), that explained 11.7% of the variance, while CK5/6 (load: −0.49) loaded negatively on this factor and factor 5, corresponding to ER status (eigenvalue = 0.4, load: 0.45), that explained 8% of the variance, while PgR (load: −0.43) loaded negatively on this factor. The Scree plot of the eigenvalues is shown in Figure 3. The component matrix of these five factors is shown in Table 5. Of note, these analyses are in support of the proposal of Rakha et al. [20] who suggested to rely on basal CK expression alone (basal CK+ subtype) to define basal-like BC, regardless of the status of the other markers. In fact, our analyses assigned all the BCs that expressed basal CKs, regardless of the other markers, to cluster II. Furthermore, the basal-like subtype (BCs with triple-negative phenotype that express basal CKs: ER−/PgR−/Her-2/meu−/basal CKs+) was also included in cluster II. It is worth mentioning that the adoption of the latter criterion only for the definition of basal BC would miss many cases, as the basal-like subset accounted for only 10% of the cases versus 41% for the basal CKs+ subset.

Discussion
The expression of basal CKs is a negative prognostic marker, implying resistance to therapy and poor prognosis, particularly in the context of BCs with triple-negative status [12,25,26,35,51]. Basal-like BC, which largely overlaps with triple-negative BC, is a well-recognized BC subtype with the above-mentioned clinically-relevant implications [12,25,26,35,51]. Basal-like/triple-negative BC appears to occur more frequently in African American women and in breast cancer case series from East and West Africa, which could reflect intrinsic differences in tumor biology related to racial/ethnic factors [6,12,21,30,32].
A better understanding of the impact of basal-like/triple negative BC in BC series from native African women would contribute to the assessment of the influence of race on this particularly relevant BC subtype. It is important to develop BC prevention and treatment policies in African populations, that, with increased life expectancy, are predicted to face marked increases in BC rates [12,14,28,35,52,53].
Consideration of two basal subtypes, that is, basal CK+, defined by expression of basal CKs regardless of other markers [20], and basal-like, defined by the triple-negative CK+ phenotype (ER−/PgR−/Her-2/neu−/basal CK+), may explain these discrepancies. In fact, in our BC series from Central Sudan, the frequency of basal-like BC is 10%, as previously reported [37], but that of basal CK+ BC is 41%. This reflects the presence of an excess of cases that express basal CKs together with ER/PgR and/or Her-2/neu.
In the present Sudanese BC series the frequency of basal CK+ status (41%) appears to be much higher than those reported for Western Caucasian and also for African American BC series (13-20% and 26%, resp.), but results quite comparable to the 34% frequency found in a BC series from Kyadondo County in Uganda and to the 33% frequency reported from West Africa (Nigeria and Senegal) [20,25,29,33,35,51]. In the study of Adebamowo et al., basal CKs were not investigated and the basal-like subtype was defined by triple-negative phenotype only (ER−, PR−, and Her-2/neu−) as one category [54]. In this regard it is notable that the Nigerian and the Sudanese case series yield almost the same frequencies of basal-like BCs defined by triple negative phenotype only: 15.8%, that is, 24/152, in the Nigerian series and 15.9%, that is, 18/113, in the Sudanese series [37,54].
In our Sudanese series, basal CK expression was associated with higher histologic grade and with hormone receptor negative status. This is in agreement with well-established evidence that the expression of basal markers occurs in poorly differentiated hormone receptors-negative BCs, as reported for Caucasian and African American series and also for the Ugandan series [25,26,35,51,55]. As in other studies, CK17 was more frequently positive than CK5/6 [25].
It is well established that in both African-American and Caucasian BC series the expression of basal CKs is significantly related to younger age at BC onset [26]. In our Sudanese series basal CK status was not associated with age at disease diagnosis, as also reported for the series from Kyadondo County in Uganda [35]. However, although not significant, the mean age and the mean tumor size were lower in the basal CK+ group than in the basal CK− one. The lack of significance for the difference in age may be due to the fact that the patients were mostly young, reflecting the young age at disease diagnosis typical of the institutional BC series from the Sudan [37,[56][57][58].
Indeed, the higher frequency of basal-like phenotype in African case series could be partially explained by the younger age of the patients [33][34][35][36]. However, socioeconomic, genetic, ethnic, and lifestyle/reproductive factors are also likely to be involved [30,37]. In particular, emerging data reported that certain reproductive factors (i.e., extended breast-feeding/lactation, high parity, and early menarche) may have a greater impact on risk of certain molecular BC subtypes compared to others [59,60]. Furthermore, other confounding factors, like antigen degradation of archival formalin fixed, paraffin-embedded tissue blocks, should also be considered for the reportedly high frequency of hormone receptor negativity, with subsequently higher frequencies of both basal-like BC identified by the triplenegative CK+ profile (ER−/PgR−/Her-2/neu−/basal CK+) and unclassified triple-negative types [33,36,37,54,61].
The lack of association between basal CK+ status and larger tumor size is quite unexpected [51]. This unusual finding might reflect the fact that large size at presentation, due to late disease diagnosis, is one of the main features of BC in Sudanese patients, when compared to BC in patients from Europe and North America [9,37,62,63]. Due to longer survival, this could result in a relative enrichment of less aggressive subtypes among the BCs of larger size [37,64], a hypothesis that requires to be further investigated in larger and prognostically well-characterized BC series from Sudan.
Except one mucinous carcinoma, all the basal CK+ tumors were invasive duct carcinomas, consistent with the literature data [51]. The fact that all the invasive lobular tumors were basal CK− could be relevant but could also reflect a bias due to the relatively low frequency of this histotype in the study series and needs further evaluation on a larger number of cases.
In concordance with the gene expression-based IHC subtypes defined in Western BC case series [5,22,26], clustering based on the five tested IHC markers outlined a hormone receptors-positive/luminal-like cluster and a hormone receptors-negative cluster with basal CKs (CK5/6, CK17) and Her-2/neu. As expected, factor analysis showed that hormone receptor status was the factor that most influenced dataset variance among the other tested factors, being negatively affected by both basal CK and Her-2/neu statuses. Basal CK status was in second position, with Her-2/neu status loaded negatively on this factor, although this was not supported by a direct negative correlation. Her-2/neu status was in the third place. The other two extracted factors (factor 4: CK17 status, and factor 5: ER status) had minimum effects as extracted factors on the dataset variance. Collectively, this demonstrates that the most important factors in the dataset were ER status, followed by PgR, CK17, and CK5/6 statuses.
Her-2/neu status played a complex role in the dataset variance, as it negatively affected both hormone receptor status (which was consistent with statistical correlations) and basal CK status (as demonstrated only by factor analysis). As previously reported, the basal-like phenotype and the Her-2/neu expression are inversely correlated [9,14,65,66], and it is likely that the nonbasal-like tumors include a high prevalence of Her-2/neu amplified tumors [65]. In this regard, it should be considered that the effects of Her-2/neu on the determination of the two clusters were quite similar, being only slightly in favour of cluster II (Figure 1(d)). Interestingly, Harris et al. reported that the expression of basal markers was strongly associated with Her-2/neu+ BCs not responding to preoperative therapy based on 8 Pathology Research International trastuzumab plus vinorelbine [53]. This underlines the need to better verify the BC subsets in which basal CKs, Her-2/neu and hormone receptors could interact, in African and non-African case series.

Conclusion
In the presently studied BC series from Central Sudan the frequency of the tumors expressing basal CKs was much higher than the frequencies reported for Caucasian and African-American BC series, but it was comparable to that found in BC series from East and West Africa [20,25,29,33,35,51]. This suggests that the impact of the tumors expressing basal CKs could be higher in sub-Saharan African patients, a possibility that needs to be confirmed by additional studies in different African populations. In Sudan a higher impact of the tumors expressing basal CKs could be ascribed to a variety of factors, including racial/genetic factors, environmental and reproductive factors, population structure, and sampling/referral bias. However, while an early age of onset is one of the clinical characteristics associated with BC expressing basal CKs, in our case series basal CK-positive status was associated with higher grade and hormone receptor-negative status, but not with age at disease diagnosis and tumor size. This quite unexpected lack of association might reflect a selective effect of late disease diagnosis. The most important factors for clusterization in distinct BC subsets were ER status, followed by PgR, CK17, and CK5/6 statuses. As in West Africa, the identified clusters were in concordance with the gene expressionbased immunohistochemical subtypes defined in Western BC case series [5,22,26,33], despite the difference in patient population. However, the overall frequency of basallike subtype (ER−/PgR−/Her-2/neu−/basal CK+) was low (10%, in Sudanese; 15.8%, in Nigerian), which was mainly due to the reported markedly higher frequency of hormone receptor positivity (ER: 64%; PgR: 67%; ER and/or ER: 75% in Sudanese and ER+: 65.1%; PgR: 54.7% in Nigerian) as compared to the other studies from Africa [33][34][35][36][37]54].