ARTICLES NATURE CELL BIOLOGY VOLUME 6 | NUMBER 6 | JUNE 2004 507 Functional proteomic screens reveal an essential extracellular role for hsp90α in cancer cell invasiveness Brenda K. Eustace 1 , Takashi Sakurai 1,2 , Jean K. Stewart 1 , Dean Yimlamai 1 , Christine Unger 3 , Carol Zehetmeier 3 , Blanca Lain 3 , Claudia Torella 3 , Stefan W. Henning 3 , Gerald Beste 3 , Bradley T. Scroggins 4 , Len Neckers 4 , Leodevico L. Ilag 3 and Daniel G. Jay 1,5 Tumour cell invasiveness is crucial for cancer metastasis and is not yet understood. Here we describe two functional screens for proteins required for the invasion of fibrosarcoma cells that identified the molecular chaperone heat shock protein 90 (hsp90). The hsp90α isoform, but not hsp90β, is expressed extracellularly where it interacts with the matrix metalloproteinase 2 (MMP2). Inhibition of extracellular hsp90α decreases both MMP2 activity and invasiveness. This role for extracellular hsp90α in MMP2 activation indicates that cell-impermeant anti-hsp90 drugs might decrease invasiveness without the concerns inherent in inhibiting intracellular hsp90. Cancer invasiveness is a complex process with three prominent stages: adhesion to the extracellular matrix, digestion of the matrix to release cells from the tumour mass, and migration of the tumour cell to sec- ondary targets 1 . Current post-genomic strategies are being applied to identify genes and their encoded proteins that function in tumour invasion. For example, microarray analyses have identified differen- tially expressed genes from many classes that might contribute to inva- siveness 2–6 , but these expression changes only suggest functional importance. RNA-mediated interference (RNAi) screens in mam- malian cells can functionally assess the genome in high throughput, and these screens for cancer phenotypes have identified new gene tar- gets 7,8 . However, such screens are limited for the following reasons: RNAi knockdown occurs over many hours, thus the screens might be susceptible to compensation mechanisms, and proteins that turn over slowly cannot be efficiently depleted. A high-throughput approach that directly addresses protein function and is temporally restricted could identify new and different targets for tumour invasion or other disease-relevant processes. Towards this end, we developed and performed two proteomic screens to identify and validate proteins required for cancer invasion. These screens used fluorophore-assisted light inactivation (FALI) 9 for acute protein knockdown directed by specific binders to compo- nents of the surface proteome with monoclonal antibody libraries or recombinant phage display antibodies 10 . FALI is a high-throughput method of damaging proteins in situ by using fluorescein-labelled binders and 490-nm diffuse light. FALI-treated cells were then assayed for invasiveness. Antibodies that recognize proteins critical for invasion were then used for immunoprecipitation and mass spec- trometry to identify the target protein. We focused our screens on the surface proteome as an accessible subset of the proteome that would provide good targets for drug discovery. Both screens identified the molecular chaperone hsp90 as an important extracellular mediator of invasion. We further showed that the hsp90α isoform, but not hsp90β, is expressed extracellularly on fibrosarcoma and breast cancer cells. Hsp90α interacts with MMP2 outside the cell and promotes MMP2 activation, which is critical for tumour invasiveness. RESULTS Screening for tumour invasion proteins We performed two independent screens for surface proteins that func- tion in tumour cell invasion. In the first, we used a library of 1,152 monoclonal antibodies (mAbs) raised against HT-1080 fibrosarcoma cells, and selected 19 that bound to the cell surface (Fig. 1). The second screen used a recombinant phage display single-chain variable frag- ment (scFv) binder library (10 7 phages) prepared from spleen mRNA derived from mice immunized with HT-1080 cells. This library is com- posed of phages that express scFv recombinant proteins representing the antigen-binding repertoire of the immunized mice. We screened this library for scFvs that bound preferentially to HT-1080 surface antigens in comparison with a control Hs27 fibroblast cell line, and selected 13 scFvs for further functional analysis (Fig. 1). We performed 1 Department of Physiology, Tufts University School of Medicine, Boston, MA 02111, USA. 2 Present address: RIKEN Brain Science Institute, 2-1 Hirosawa, Wako, Saitama 351-0198, Japan. 3 Xerion-Pharmaceuticals AG, Sauerbruchstrasse 50, D-81377, Munich, Germany. 4 Cell and Cancer Biology Branch, National Cancer Institute, Rockville, MD 20850, USA. 5 Correspondence should be addressed to D.G.J. (e-mail: daniel.jay@tufts.edu). Published online: 16 May 2004; DOI:10.1038/ncb1131 ©2004 Nature Publishing Group