Abstract BACKGROUND Glioblastomas are highly vascular tumours that overexpress vascular endothelial growth factor (VEGF), however the anti-VEGF antibody bevacizumab has not demonstrated significant overall survival benefit for patients. Vasculogenic mimicry (VM), the ability of tumour cells to form functional microvasculature lacking an endothelial lining, may compensate for insufficient or inhibited angiogenesis. VM has been reported in primary glioblastoma tissue, but not investigated at progression in recurrent tumours. Prostate specific membrane antigen (PSMA) is expressed in abnormal vessels within primary glioblastoma tumours. We aimed to determine whether VM occurs in primary and recurrent glioblastoma tumours and whether VM vessels also express PSMA. MATERIAL AND METHODS Thirty-five matched pairs of primary and recurrent formalin-fixed paraffin-embedded glioblastoma tissue were immunohistochemically labelled for the endothelial cell marker CD34 and PSMA, and counterstained with periodic acid-Schiff (PAS). Vascular structures were categorised as endothelial vessels (CD34+/PAS+) or VM (CD34-/PAS+). Vessels were counted in ten regions within each tumour and a vessel density/mm2 for each vessel category was determined. RESULTS Data are expressed as median values. Endothelial vessels accounted for most blood vessels observed. There was a significant decrease in endothelial vessel density in recurrent (41.34 vessels/mm2) compared to primary (86.98 vessels/mm2) glioblastomas (p <0.001). VM was observed in 15/35 primary tumours (42.86%) and 10/35 recurrent tumours (28.57%). VM accounted for only a small proportion of the overall vasculature and VM density was not different between the primary and recurrent groups (p = 0.266). Kaplan-Meier analyses with log rank tests indicated that VM-tumours had longer median OS time than VM+ tumours, regardless of whether VM status was determined based on the primary or recurrent tumour. However, there was no significant difference in survival distribution between VM+ and VM-tumours (primary χ2(1) = 0.381, p = 0.537; recurrent χ2(1) = 3.552, p = 0.059). PSMA is expressed in endothelia lined vessels based on serial section labelling for PSMA and CD34. PSMA expression significantly decreased in recurrent (median H-score = 0.22) compared to primary (median H-score = 0.44) tumours (one-tailed p = 0.031; n = 32) mirroring the CD34+ vessel density. CONCLUSION VM contributes to a relatively small proportion of the overall vasculature in glioblastoma tumours and was not associated with OS nor the expression of PMSA. Since VM- tumours had longer OS than VM+ tumours, assessment of VM in recurrent tumours in a larger cohort may determine if there is a clinical significance for VM in recurrent glioblastoma.