Statistics from Altmetric.com
We have read with interest the recent paper by Jubb et al on the expression of hypoxia inducible factor 1α (HIF-1α) in human tumours.1 We note that they report only 5% of ductal adenocarcinomas of the breast to be HIF-1α positive. This proportion is unusually low compared with our own data and those of other workers. In our various studies, the proportion of HIF-1α positive breast cancers varied from 44% to 80%.2–4 In studies from other groups, these percentages varied from 56% to 76%.5,6 We believe that this discrepancy may be caused by the use of tissue microarrays. In breast cancer, HIF-1α often shows pronounced intratumour heterogeneity because of focal perinecrotic staining, which is clinically highly relevant. Even patients with only 5% of cells overexpressing HIF-1α have a much worse prognosis.3 Thus, tissue arrays probably underestimate the true frequency of HIF-1α overexpression in breast cancer. Hence, data from studies on HIF-1α derived from tissue arrays are probably less reliable with regard to associations between HIF-1α and other biomarkers and prognosis for invasive breast cancer. Therefore, we believe that conventional tumour sections are superior for the assessment of HIF-1α overexpression in this type of cancer. For other cancers with a more diffuse type of staining this may be different.
We thank Professor van Diest and colleagues for their letter raising some interesting issues about hypoxia inducible factor 1α (HIF-1α) in breast cancer. Our study examined vascular endothelial growth factor D (VEGF-D) expression in breast cancer and correlated expression with clinicopathological variables, with an emphasis on hypoxia markers. Van Diest et al suggest that tissue microarrays (TMAs) are unsuitable for the analysis of HIF-1α because they may miss a particular pattern of HIF-1α staining that has prognostic relevance. However, the clinical impact of these patterns is unclear because in their study HIF-1α was highly associated with necrosis and grade and the survival analysis was not multivariate. Nevertheless, we agree that we identified fewer HIF-1α positive tumours (17%) than van Diest et al, who reported 75%,1 54%,2 and 44%3 positivity in three studies using a 1% cutoff value, or 34% in another study when using a 5% cutoff value.1 However, although van Diest et al suggest that others have found a similar range of HIF-1α positivity (56%4 and 76%5), appraisal of the published data shows that if only strong HIF-1α staining is used, then these studies show 23% HIF-1α positivity. Moreover, Zhong et al using whole tissue sections of breast tumours reported strong HIF-1α expression in only 12%.6 We used a semiquantitive score of negative, weak staining, or strong staining and considered strong staining as positive (the proportion of cells was also noted (0, 0%; 1, 1–10%; 2, 11–50%; 3, 51–80%; and 4, 81–100%), but in practice when strong staining was present all the tumour cells were positive). If we re-stratify our cases using the criteria of van Diest and colleagues—nuclei that are “completely and darkly stained” and a > 1% cutoff point2—we have a positivity rate of 78%, which is within the range reported by van Diest et al. Indeed, because each core within a TMA can hold up to 1000 cells, we may be analysing similar amounts of tumour, but the methodology in the van Diest group’s papers with regard to the number of cells, the ranges, medians, etc is unclear, so it is not possible to assess this. Thus, we think that although we may miss some HIF-1α positive cases, it is probably only a small proportion, and may be compensated for by the absence of variability in staining that is common in studying large series by whole tissue sections. Furthermore, because we specifically wish to analyse the relation between molecular markers in the same region of the tumour (in the individual core/exemplar), we view the use of TMAs as an advantage.
Van Diest and colleagues also raise the issue of the subcellular localisation of HIF-1α. Cytoplasmic expression of HIF-1α is well recognised.6,7 To ignore it appears to us premature just because it does not fit in with current models. Many so called cytoplasmic or nuclear only proteins have now been shown to shuttle to and fro between the two compartments. Indeed, it has been shown that a nuclear location is not required for HIF-1α stabilisation and that HIF-1α undergoes oxygen dependent proteasomal degradation in both the nucleus and the cytoplasm.8
Lastly, van Diest et al suggest that the lack of an association between VEGF-D and prognosis may be explained by the absence of lymphangiogenesis in breast cancer.9 The notion that VEGF-D enhances lymphangiogenesis and thereby influences nodal status and prognosis is somewhat one dimensional because VEGF-D has other functions in addition to lymphangiogenesis, including lymphatic growth, lymphatic maintenance, and angiogenesis.
Thus, we think that the lack of an association is more likely the result of several factors leading to nodal metastasis, including VEGF-C, with differences in processing resulting in changes in receptor affinity. Furthermore, it is also likely that a combination of factors rather than a single factor will be clinically useful.10