| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
|
FJ
EXPRESS SUMMARY ARTICLE The Full-length version of this article is also available, published online June 18, 2004 as doi:10.1096/fj.04-1538fje. |
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
* BioSeek, Inc., Burlingame, California, USA; and
Laboratory of Immunology and Vascular Biology, Department of Pathology, Stanford University School of Medicine, Stanford, California, USA
1Correspondence: BioSeek, Inc., 863-C Mitten Rd., Burlingame, CA 94010, USA. E-mail: eberg{at}bioseekinc.com
SPECIFIC AIMS
Validation of compounds for activities relevant to human disease biology is a major bottleneck in drug development. Here we apply statistical analyses to protein expression data sets from primary human cell-based systems in multiple complex environments relevant to vascular inflammation to detect and distinguish compound mechanisms. We show that this approach, termed biologically multiplexed activity profiling (BioMAP profiling), is able to discriminate the action of diverse compound classes, identify off-target or secondary activities, and lead to insights into clinical activities of compounds.
PRINCIPAL FINDINGS
1. Complex primary human cell-based BioMAP assays are robust and reproducible
Endothelial cells (EC) express adhesion receptors and chemokines that modulate inflammatory responses by regulating leukocyte traffic. In chronically inflamed tissues, EC are exposed to multiple proinflammatory cytokines including IL-1ß, TNF-
, and IFN-
. As a starting point, we stimulated primary human EC with this combination of cytokines (the "3C" system) for 24 h in the presence of selected compounds. A set of protein readouts (ICAM-1, VCAM-1, E-selectin, CD31, MIG, IL-8, HLA-DR, and MCP-1) selected for their robust modulation in response to one or more cytokine combinations or to specific compounds and for their potential or known relevance to vascular inflammation was measured by cell-based ELISA.
Treatments such as an antibody inhibitor of TNF- and PD169316, a p38 MAPK inhibitor (Fig. 1
a), produced characteristic profiles, indicating that each induce a unique and reproducible response in the 3C system. Dose responses of each agent (e.g., PD169316 in Fig. 1b
) illustrate how the shape of the profile remains similar at different compound concentrations, a common feature of well-behaved competitive inhibitors. The ratio of parameter values with drug to solvent control is more consistent than measurements of the absolute parameter values (illustrated for MCP-1 in Fig. 1c
). The absolute value of MCP-1 in the presence of solvent control (filled circles; 25% CV) or PD169316 (open circles; 31% CV) is much more variable than their ratios (open triangles; 14% CV).
|
2. Effective classification and discrimination of compound mechanisms can be obtained by parallel analysis using multiple BioMAP models
To enhance our ability to discriminate compounds (including cell type-specific effects), we established multicellular systems comprising peripheral blood mononuclear cells and EC, stimulating either the T cell receptor complex with superantigen (the "SAg system") or toll receptor signaling with lipopolysaccharide (the "LPS system"). Eight readout parameters were selected for the SAg system (CD3, CD38, CD69, CD40, IL-8, MCP-1, MIG, and E-selectin) and 10 for the LPS system (CD14, CD69, CD40, E-selectin, VCAM-1, tissue factor/CD142, IL-1, M-CSF, IL-8, and MCP-1). In these multicellular systems, cells respond directly to the initiating stimuli and/or to each other, resulting in a complex cascade of events.
The SAg and LPS systems responded robustly and reproducibly to a number of compounds that were inactive or only weakly active in the 3C system (Fig. 2
a). For example, FK-506 and cyclosporin A, two inhibitors of calcineurin-mediated T cell receptor signaling, were strongly active, inducing strongly correlated responses. Other compounds active in the SAg system (but not the 3C system) included IL-10, the phosphodiesterase 4 inhibitors Ro-20-1274 and rolipram, the immunosuppressant rapamycin, and src family kinase inhibitors PP1 and PP2. Many compounds active in the endothelial inflammation system retained activity in the more complex multicellular systems (e.g., the p38 inhibitors PD169316 and SB220025 and anti-TNF-).
|
To allow objective evaluation of the significance of all relationships between compound activities, profile data from all three systems were combined and the multisystem data for each compound (at a given concentration) were compared by pairwise Pearson correlation (Fig. 2b
). The relationships implied by these correlations were then visualized by multidimensional scaling to represent them in two dimensions. In this graph (Fig. 2c
), the distance between compounds is a reflection of the degree of difference in functional profiles. A statistical permutation method based on minimizing false discovery rate (FDR) was used to identify statistically significant correlations, indicated by lines connecting the drugs.
Examination of the resulting function similarity map (Fig. 2c
) illustrates how the majority of compounds known to share a common mechanism are significantly related to each other. These include glucocorticoids, p38 inhibitors, hsp90 inhibitors, HMG-CoA reductase inhibitors (statins), calcineurin inhibitors, and TNF- antagonists. Compounds known to have poor target specificity such as the general tyrosine kinase inhibitors AG126 and genistein; the JAK inhibitors ZM39923, WHI-P131, and AG490; or the 5-lipoxygenase inhibitors AA861 and NGDA show little functional similarity, but instead are related more closely to compounds from divergent classes, reflecting the unique biological consequences of their inhibition of multiple molecular targets. That the mTOR antagonist rapamycin exhibits homology of function to the general PI-3 kinase inhibitors LY294002 and wortmannin is consistent with the known regulation of p70S6K (an mTOR target) by PI-3 kinase. Thus multisystem analysis allows identification of compounds with related functions across the systems tested.
3. BioMAP analysis produces insights into clinically relevant activities of HMG-CoA reductase inhibitors (statins) and mycophenolic acid
The BioMAP approach can also lead to new biological insights for well-studied compounds. Examination of the profile for statins shows that one of the dominant features of these profiles is the significant reduction in CD69, a T cell activation antigen, in the complex SAg system. Of the HMG-CoA reductase inhibitors tested (pravastatin, rosuvastatin, simvastatin, mevastatin, lovastatin, atorvastatin, and cerivastatin), all seven demonstrated this activity, though with differing potencies. This activity is consistent with reported activities of statins on immune function distinct from their effects on cholesterol synthesis. Because of the association of CD69 with T cell activation, this activity may be related to the reported beneficial effects of statins in autoimmune diseases such as rheumatoid arthritis and lupus.
A novel and distinctive multisystem profile was obtained for mycophenolic acid (MPA), the active form of mycophenolate mofetil, a prodrug approved as an immunosuppressant for use in kidney transplantation. Key features of the MPA profile include strong inhibition of MCP-1 in all three complex systems. MCP-1 is a monocyte chemoattractant that plays a role in monocyte recruitment in many chronic inflammatory disease settings. Increased levels of MCP-1 are found in patients with coronary artery disease and diabetes, and are associated with increased risk of cardiovascular mortality. No other immunosuppressant tested had such a selective effect on MCP-1, and none had any effect on MCP-1 in the 3C system. Mycophenolic acid was recently found to decrease the risk of post-transplant diabetes, in contrast to cyclosporin and FK-506, which are associated with increased incidence.
CONCLUSIONS AND SIGNIFICANCE
In the present study, we evaluated a set of human cell-based model systems that incorporate increased levels of complexity, with relevance for use in rapid identification of effective new therapeutics. The responses measured in these complex systems were surprisingly robust and reproducible, and could be employed for efficient classification of compounds according to their functional activities.
The systems described here (Fig. 3
) have multiple applications to drug discovery. Protein kinases are an important drug development target family, and a number of kinase inhibitors are active in the assays presented here. Due to the large size of the kinome and the greater number of ATP binding proteins, however, specificity is problematic. Our results show that multiplexed profiling can efficiently identify compounds specific for a common kinase target (as seen with the well-characterized p38 inhibitors used in our study), but also identify drugs with off-target activities (such as the functional clustering of the c-Raf antagonist ZM336372 with p38 inhibitors) or poor specificity (such as the JAK inhibitors that cluster away from each other or the general tyrosine kinase inhibitors genistein and AG126, whose functional responses are likely related to the specific set of kinases they inhibit). Thus, similarity of function comparisons can rapidly identify drug specificity problems early in preclinical development.
|
In conclusion, multiplexed activity profiling in scalable complex cellular systems has the potential to rapidly characterize pathways (and mechanisms of action) of novel molecules. The power of this systems biology approach is illustrated by functional classification of a wide variety of anti-inflammatory drug classes. The strength of the approach derives from the complex, combinatorially determined system responses and is enhanced by parallel interrogation of systems in which multiple different pathways are stimulated. Applications include characterization and optimization of lead compounds, structure-activity relationship studies, in-depth mechanism of action studies, large-scale gene function screening, and pathway mapping. The integration of human disease biology and pathophysiology into early stages of the drug discovery process can help improve the efficiency and success of drug development programs.
FOOTNOTES
To read the full text of this article, go to http://www.fasebj.org/cgi/doi/10.1096/fj.04-1538fje;
This article has been cited by other articles:
|
|
 |
|
  C. Kunsch, J. Luchoomun, X.-l. Chen, G. L. Dodd, K. S. Karu, C. Q. Meng, E. M. Marino, L. K. Olliff, J. D. Piper, F.-H. Qiu, et al. J. Pharmacol. Exp. Ther., May 1, 2005; 313(2): 492 - 501. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
0); yellow indicates negative correlations. a, b) The list of compounds was ordered automatically by scaling and pivoting to move high correlations to the diagonal. c) A function similarity map is generated by subjecting pairwise correlation data to multidimensional scaling. Significant correlations (FDR set at 5%) are shown by lines. Distance between compounds is inversely related to the similarity of the compound profiles.