One of the key challenges facing the development of new cancer therapies is the identification and validation of targets that are tractable to new therapies. In this study, we performed a comprehensive analysis of the plasma membranome of the MDA-MB-231 ‘triple negative’ breast carcinoma cell line with the overall aim of identifying cell surface markers tractable to antibody therapy. From a technical viewpoint, the intention was to explore the usefulness of combining a global proteomic analysis of cell surface proteins with the phenotypic screening of antibodies generated to those cell surface antigens and furthermore to compare the effectiveness of phage display and hybridoma technology for generating pools of antibodies for phenotypic screening.
Comparing the antigens identified by each method, a number were identified by multiple approaches, namely integrin-α2, integrin-α3, EGFR, CD73 and CD44, the latter being identified by all three methods. All of these antigens have been previously cited as biomarkers, potential therapeutic cancer targets or fully validated therapeutic targets, such as EGFR. Hence, this multi-approach model gives confidence to the validity of target antigens when identified by multiple approaches. Others were identified by a single method only including: BCAM, HER2, CD71, galectin-3 and the 31 antigens identified using the proteomic method. Hence, the combination of methods was effective in highlighting antibody targets relevant to tumour therapy, ranging from those already exploited in the clinic to those which could be of great interest as novel antibody targets. For this purpose, the benefit of using a phenotypic screening approach, which provides an antibody reagent for further validation work and potentially a therapeutic lead, was clearly illustrated by the use of the phage display-derived anti-CD73 antibody to inhibit MDA-MB-231 tumour growth in vivo, as discussed later.
Looking in more detail at each method and first considering the proteomic approach, it was shown that the global analysis of membrane protein content identified 37 plasma membrane proteins. A further 74 proteins were also believed to be derived from internal membranes such as those from ER/golgi/ribosome, mitochondria, and nucleus. In order to assign these proteins to their cellular location, we were reliant on prior classifications in public databases with varying levels of experimental validation and therefore may have overlooked the possibility that some proteins had relocated to the plasma membrane as part of the oncogenic process. Nevertheless, this list of 37 plasma membrane proteins provided a useful survey of the membrane protein content and contained some well validated targets, such as EGFR. It also included others that were not identified by the phenotypic antibody screening approaches that have been implicated in cancer progression, metastasis and invasion including: CD151 antigen, basigin, ephrin type-A receptor 2 and myoferlin. However, it does contain a large number of targets for which disease linkage is not well established. As such, these remaining targets would require significant experimental work to validate their utility for antibody intervention. The total number of membrane proteins identified was also relatively low, bearing in mind the total estimated number of human membrane proteins present; however, this may reflect the processing steps required to enrich for pure membrane fractions.
An alternative approach to focus on membrane proteins upregulated in cancerous cells would have required a comparative analysis of matched, normal cell lines. However, it can be extremely challenging to procure sufficient quantities of well-matched normal material for such studies and this would ultimately not lead to functional validation of targets but would remove those targets that are common between the two cell types. Further optimisation of the proteomic method presented here may also be beneficial. This type of ‘shot-gun’ proteomics method often leads to a highly complex mixture of peptides at varying concentrations, thus placing high demands on the liquid chromatography-tandem mass spectrometry (LC-MS/MS) step for identifying the peptides. Longer separation times, or the use of two-dimensional LC, in which the sample is first divided into fractions using a strong cation exchange column, followed by standard reverse phase separation, could potentially enable the discovery of additional targets amenable to antibody therapy.
To explore phenotypic screening approaches for target identification, both phage display and hybridoma technologies were investigated in this study. Both approaches used two main criteria for prioritising antibodies tailored to antibody characteristics we were interested in. These were (i) preferential binding to MDA-MB-231 cells over normal, immortalised breast cell lines, and (ii) internalisation in MDA-MB-231 cells. Importantly, both the binding and internalisation screens could be performed in high-throughput in order to assay thousands of antibodies for function. Using the hybridoma method, we observed that four target antigens could be identified; namely integrin-α2, integrin-α3, CD44 and galectin-3. The last of these was not a true plasma membrane protein but rather a secreted protein which can associate with the cell membrane as a result of its carbohydrate recognition domains. One of the advantages of the hybridoma approach was that the antibodies were highly effective immunoprecipitation reagents, with 16 of the final 20 hybridoma antibodies able to immunoprecipitate a specific target, presumably on the basis that they have undergone affinity maturation in vivo. This high affinity also gives the potential that the antibodies could be directly used as a therapeutic, without further optimisation, if relevant criteria are fulfilled. However, this affinity maturation could also be a disadvantage because several of the antibodies in the final 20 were minor sequence variants of each other. This limited the number of target antigens identified and any future immunisation process would require optimisation to avoid such clonal dominance, either by extensive sequencing early on in the screening process or by shortening the immunisation procedure.
Using the phage display approach on the MDA-MB-231 cell line, the initial screen was able to isolate nine different scFv antibodies that fulfilled the screening criteria. However, due to the speed and ease with which further selections and screening could be performed, we were able to identify an additional 109 functional scFvs from other cell lines. In total, this approach identified six targets: EGFR, B-CAM, CD44, CD73, CD71 and HER2. All of these have previously been implicated in cancer progression, with the least validation associated with CD73[13, 14]. It must be noted that the MDA-MB-231 cell line is described as HER2 negative, and it does not amplify or overexpress HER2. However, it does have a basal level of HER2 expression that could be sufficient to illicit a functional response. Hence, direct cell panning on MDA-MB-231 cells failed to identify anti-HER2 antibodies; however, when antibodies were screened, which had been isolated against other cell types; internalising anti-HER2 antibodies were identified. This highlights the advantages of not restricting a screen to those antibodies that have only been isolated against the test cell type.
The success rate of immunoprecipitation using the phage-display antibodies was low, with only 1 of the final 9 antibodies able to immunoprecipitate a specific target, CD73. This is most probably due to the relatively low affinity as these antibodies were isolated from a naïve library. Target identification success rate was increased by performing ELISAs against known targets and could be improved further by incorporating cDNA-based target identification methods. However, even without affinity maturation and further development this study also demonstrates that an antibody derived from this phage display-based approach can be used to validate an identified target. The anti-CD73 antibody isolated by phage display demonstrated anti-tumour activity in a MDA-MB-231 xenograft model and highlights the advantage that isolating a target and an antibody together can accelerate the early validation of that target in a disease relevant setting. This can be illustrated further by using CD73 as an example. In 1991, Kruger et al. demonstrated expression of CD73 in breast carcinoma and a further 19 years of target biology exploration and experimental validation ensued before Stagg et al. demonstrated that an anti-CD73 antibody could inhibit breast tumour growth and metastasis. This also required the de novo generation of antibody tool reagents in order to test the inhibition hypothesis in a disease model. In the phenotypic screening approach described here, in which antibody generation was an integral part, the whole process from initial screen to in vivo target validation took approximately 12 months. In addition to this in vivo validation these antibodies can also be used to establish disease association via immunohistochemistry of patient tissue samples and in mechanistic studies to understand the optimal mode of action.
This study focused on a single ‘triple negative’ breast cancer cell type. Focusing on a single cell type has two implications. The first is that this kind of analysis is ideal for a personalised healthcare approach if a suitable target cell type can be identified and isolated. Here were have focused on a ‘triple negative’ breast cancer cell type; however, this could easily be substituted for another cell type such as a KRAS-mutant non small cell lung carcinoma cell type. The second implication is the potential to identify many more targets by looking at other key disease-promoting cell types and not just restricting this approach to tumour cell types.