Polyamines are required for cellular viability and elevated levels are found in many tumour types, including breast cancer , making polyamine synthesis an attractive target for chemotherapy. PG-11047 is a second generation polyamine analogue specifically designed as a therapeutic agent . While it is reported to effectively inhibit cell growth in lung, breast and colon cancer cell lines [7, 12, 24], only a limited number of cell lines had previously been studied. Since breast cancer is now known to be comprised of multiple genomic and transcriptional subsets [10, 14] that progress and respond to therapy differently, we analysed quantitative responses to PG-11047 in a collection of 42 breast cancer and six non-malignant breast cell lines in order to identify biological and molecular features associated with response. Figure 1 shows that the basal subtype is most strongly inhibited by treatment with PG-11047, based on GI50 response profiles. Basal subtype breast cell lines mirror many molecular features of basal-like primary breast tumours [15, 16] including low expression of oestrogen receptor and ERBB2 and high expression of keratin 5/6/14 and EGFR (Spellman et al., personal communication). This suggests that, PG-11047 may be preferentially effective against this more aggressive breast cancer subtype. This is consistent with observations made in a study of the polyamine analogue, N
11-diethylnorspermine (DENSPM) . An analysis of BrdUrd incorporation and apoptosis induction in sensitive and resistant cell lines suggests that responses to micromolar concentrations of PG-11047 mainly involve reduced cell cycle traverse rather than induction of apoptosis. Holst et al.  also reported inhibition of growth in the breast cancer cell lines with PG-11047, although they observed a stronger apoptotic response. This discrepancy might be explained by differences in assay conditions, since they assessed apoptosis as an increase in the fraction of cells showing sub-G1 DNA content, while our study was based on analysis of apoptosis related caspases
The concentration of drug needed to inhibit growth measured as GI50 or TGI ranged over four orders of magnitude which meant that molecular features associated with response could be identified with confidence. The correlation of transcription profiles of the cell lines with their GI50 sensitivity identified 250 genes whose expression levels were associated with response to PG-11047. Network and pathway analyses of these genes are summarized in Figure 2 and Table 1. Increased interferon signaling was implicated with increased sensitivity to PG-11047 by both pathway and network analyses (Network 2). This is consistent with the observation that interferon inhibits the activity of ornithine decarboxylase (ODC) . We speculate that cells with high interferon activity are preferentially sensitive to PG-11047 because interferon-induced down regulation of ODC reduces the endogenous pool of the polyamines with which PG-11047 must compete to affect its inhibitory functions. Other signalling pathways and networks implicated in Table 1 and Figure 2 have been reported to be differentially active in basal and luminal tumours and cell lines. For example, pathway activities associated with basal subtype tumours involve Ephrin receptor [14, 27], BRCA1  and integrins . The strong basal subtype specificity of PG-11047 probably explains the associations with these pathway activities. Differential sensitivity of basal and luminal cells to PG-11047 also probably explains the structure of Network 1 in Figure 2, since this network included several genes with strong subtype specific expression .
A further analysis of the 250 genes associated with response to PG-11047 identified 13 genes whose levels were strongly and independently associated with response (Table 2). We propose that a clinical response predictor could be generated by assaying the levels of expression of these genes. To this end, we evaluated the applicability of the 13 gene set in stratifying breast tumour transcription datasets. We used the model to predict the sensitivities of the tumour samples in the Chin et al.  tumour panel (total = 118 samples). We found that there is a tendency toward separation between the predicted sensitivities of basal versus non-basal tumours (P = 0.08). The stratification improved if we restricted the analysis to high confidence predictions of the model (89/118 samples): P = 0.01. This analysis supports the utility of PG-11047 in the treatment of basal-subtype tumours.
Assessments of the contributions of these 13 response-associated-genes to breast cancer pathophysiology may provide insights into breast cancer responses to PG-11047 that were not directly tested in this study. Table 2 shows that increased sensitivity to PG-11047 is associated with lower expression of WASL, CST3, DEAF1 and ACSL3 and higher expression of GCLM, LAMA3, SSRP1, ACYP1, CYLD, PRPF18, AMFR, PPP1R2 and LOH11CR2A.
Several of these associations suggest that PG-11047 may act to inhibit aspects of migration/metastasis. For example, the protein encoded by CST3 is an antagonist of TGF-β signaling and has been shown to be significantly down regulated in many breast cancers . Thus, PG-11047 may be preferentially effective in cells with active TGF-β signaling, a phenotype that has been associated with aggressive cancer behavior including increased migration/metastasis  Reduced expression of WASL and increased expression of AMFR also have been reported in breast cancer tissue [32, 33] and associated with cellular migration [33–35]. Increased AMFR expression also is associated with increased phospho-AKT levels in primary human breast cancers . Consistent with this, we observed that a higher protein level of phospho-AKT is a significant predictor for PG-11047 sensitive cells (Additional File 4). The implication of PG-11047 as an inhibitor of motility/metastasis is further supported by the fact that aspects of motility are known to be regulated by polyamines . The associations of gains of the chromosomal regions near AMFR (16q21) and loss near CST3 (20p11.2) suggests that the associations of gene expression changes with motility/metastasis and that response to PG-11047 may be driven by genomic abnormalities.
Gene function assessments also suggest that PG-11047 may moderate aspects of epithelial to mesenchymal transition that are associated with aggressive clinical behaviour. For example, Carpenter et al.  have recently shown that laminin 5 may contribute to increased tumour aggressiveness resulting from epithelial to mesenchymal transition (EMT) in breast tumours. Increased expression of LAMA3, a subunit of laminin 5, is associated with response to PG-11047 suggesting the possibility that tumours responding to the drug also may experience reduced EMT. Among the other genes associated with response, GCLM and SSRP1 have been implicated in stress response [39, 40], PRPF18 and SSRP1 are involved with transcription/translation [41, 42], CYLD and PPP1R2 are involved with cell cycle regulation [43, 44] and CYLD and LOH11CR2A have been reported as anti-oncogene/tumour suppressor genes [45, 46].