Clin Genet 2014 Printed in Singapore. All rights reserved
© 2014 John Wiley & Sons A/S. Published by John Wiley & Sons Ltd CLINICAL GENETICS doi: 10.1111/cge.12462
Review
Use of genetic technologies to compare medicines Kolitz S.E., Towfic F., Grossman I., Hayden M.R., Zeskind B. Use of genetic technologies to compare medicines. Clin Genet 2014. © John Wiley & Sons A/S. Published by John Wiley & Sons Ltd, 2014 In order to ensure that patients receive the safest and most effective medicines possible, it is often necessary to compare medicines and assess the extent to which they are similar in their clinical impact. Full clinical trials with appropriate endpoints remain the only method to compare the clinical impact of two medicines with absolute certainty. Other available methods (including physicochemical analysis, genomics, and transcriptomics) can provide partial information about certain aspects of a medicine’s biological impact, with possible clinical implications. Especially for biologics and non-biological complex drugs, which are more difficult to characterize by physicochemical means than small molecules, genomics and transciptomic studies can yield valuable insights for physicians, regulators, and drug developers. In this review, we cite and summarize a variety of studies that exemplify the emerging science of applying genomics and transcriptomics technologies to compare medicines. We discuss key aspects of experimental design, conduct of genetic assays, and advanced data analysis, all of which are critical for the successful execution of such studies. Finally, we propose new areas for which such studies can be applied to maximize patient benefit and reduce safety issues. Conflict of interest
The authors are employees of either Teva Pharmaceutical Industries or Immuneering Corporation. Teva is the maker of Copaxone, a therapy discussed in this review. Immuneering Corporation receives funding through a partnership with Teva.
The explosion in the use of genetic technologies over the past decade has created countless opportunities to compare medicines, and mechanisms underlying response to medicines, in new ways. These technologies range from microarrays for measuring the expression levels of tens of thousands of genes, to single nucleotide polymorphism (SNP) chips that detect hundreds of thousands or millions of specific genetic variants, to sequencing and next generation sequencing technologies that provide extensive information on genetic variants. The initial wave of applications for these technologies focused on a few key areas, such as identifying genetic biomarkers of response to medicines, identifying pharmacodynamic markers, and seeking to understand the biological mechanisms underlying specific diseases
S.E. Kolitza,† , F. Towfica,† , I. Grossmanb , M.R. Haydenb and B. Zeskinda a Immuneering Corporation, Cambridge, MA, USA and b Teva Pharmaceutical Industries, Petach Tikva, Israel † These
authors contributed equally.
Key words: biosimilars – computational biology – gene expression – generics – microarray Corresponding author: Benjamin Zeskind, Immuneering Corporation, Cambridge, MA 02142, USA. Tel.: +1 617 5008080x121; fax: +1 617 4499588; e-mail: [email protected] Received 13 June 2014, revised and accepted for publication 16 July 2014
and response to therapies. In parallel, a newer yet equally useful set of applications has been emerging, in which the extent to which two medicines are similar is assessed. Such comparisons are less useful for small molecule drugs, which can be established to be equivalent by physicochemical analytical methods, but more useful for biosimilars and purported generic versions of non-biological complex drugs (NBCDs), both of which are intended to be similar, yet difficult to compare via conventional analytical methods. Examples in which genomics has been applied to make such comparisons are increasingly appearing in the literature, representing an extremely powerful use of such technologies. In this review, we summarize exciting new studies applying genomic technologies to compare Copaxone
1
Kolitz et al. with differently manufactured glatiramoids (NBCDs), finding differences with important implications for safety. We describe studies comparing a biosimilar for Gaucher’s disease to its reference medicine, as well as studies comparing distinct medicines with similar targets or purposes. After thoroughly surveying the landscape of this emerging application for genomics, we conclude with recommendations for the proper design of such studies. Although the information provided by these genomic studies is extremely valuable, it is important to clearly state at the outset that there is no substitute for well-controlled and adequately designed clinical trials for determining that a medicine is safe and effective. No assay technology or analysis method has been determined to be capable of fully characterizing the complexity and scope of human biology, nor predicting with 100% certainty how a medicine will perform in the clinic, and data from each method should be looked at in concert with data from other characterization methods. Nevertheless, the studies described below yield highly useful insights into the biological impact of various medicines. Safety and efficacy of biologics and NBCDs
For traditional small-molecule-based medicines, the exact chemical structure and the physicochemical properties are known, and therefore it can be unequivocally established that a medicine produced by a different manufacturing process (e.g. a purported generic) is the same as the reference listed drug, and should induce the same gene expression signature. However, a substantial number of the most effective new medicines today are not small molecules, but rather biologics or NBCDs. For these advanced medicines, it is much more challenging to characterize the extent to which a differently manufactured product is similar. The FDA recently released draft guidance describing a framework for considering biosimilars, follow-on products for complex biological drugs including antibodies such as anti-TNF-alpha and proteins such as filgrastim (G-CSF), and recognized the utility of genomic data by stating that ‘Using broader panels of biomarkers (e.g. by conducting a protein or mRNA microarray analysis) that capture multiple pharmacological effects of the product may be of additional value’ (1). A similar challenge faces regulators in assessing NBCDs such as low-molecular weight heparins, liposomal drugs, iron-carbohydrate drugs, and glatiramoids (2, 3). For both biosimilars and NBCDs, investigators have hypothesized that genomic technologies can provide valuable information. By first conducting proof-ofconcept studies in a relevant biological system, using robust statistics at the pathway and gene, investigators are able produce a hypothesis regarding similarities and differences that can be subsequently investigated in independent, human-relevant model systems, confirmed by polymerase chain reaction (PCR), etc. Aspects of this approach are illustrated in the examples below.
2
Comparing Copaxone with differently manufactured glatiramoids: background and experimental design
With approximately 2,000,000 patient-years of experience in 57 countries since the 1990s, Copaxone has proven to be a safe and effective treatment option for patients with relapsing forms of multiple sclerosis. Glatiramer acetate (GA), the active substance of Copaxone, is a synthetic polypeptide mixture produced by copolymerization of l-glutamic acid, l-alanine, l-tyrosine, and l-lysine with an average molar fraction of 0.141, 0.427, 0.095, and 0.338, respectively. Copaxone forms a complex colloidal suspension that has proven extremely challenging to fully characterize, despite extensive efforts using the latest technologies (4). Owing to the difficulties in characterizing Copaxone molecularly (5), standard physicochemical and biochemical methods have also proven ineffective for comparing Copaxone to glatiramoids produced in different manufacturing conditions (6). This is due to the fact that most molecular assays for comparing generics are designed for comparing small molecules where the atomic structure of the molecule can be fully characterized. Despite the difficulties characterizing the molecular structure of the drug, decades of research into the mechanism of action of Copaxone (7–10) have shown progress in characterizing the function of the drug. Specifically, published research indicates that Copaxone interacts with the immune system to cause a shift toward a type 2 (Th2) helper T-cell cytokine profile (9, 11), that antigen presenting cells are a prime target for the immunomodulatory effects of GA and essential for the GA mediated shift toward a Th2 profile (12), and that GA induces FOXP3+ regulatory T cells (Tregs) in patients (13, 14). Each of these mechanisms is believed to be important, although specific gene signatures and the mechanism by which the drug causes the shift are not fully elucidated. Two (15, 16) recent publications have proposed new methods of comparing Copaxone with differently manufactured glatiramoids. For these studies, isolated splenocytes from female mice treated with GA were exposed ex vivo to a variety of differently manufactured glatiramoids. Gene expression of 45,000+ transcripts was measured. The sample/chip assignments were randomized to avoid batch effects. The findings of these studies are summarized below. Comparing Copaxone with differently manufactured glatiramoids: summary of initial analyses as described by Bakshi et al. (15)
GA was found to modulate 1474 genes relative to medium (by t-test with fold change (FC) ≥ 1.3 and false discovery rate (FDR)-adjusted p < 0.05), and functional analysis of this list indicated that it included many relevant pathways including T-helper cell differentiation (16). GA drug product was also compared to a differently manufactured glatiramoid purported to be a generic for GA (Glatimer® , Natco Pharma, Ltd., Hyderabad, India). Comparing eight samples from three manufacturing batches of purported generic to Copaxone, 98
Comparing medicines genes were found to be differentially expressed (one-way anova, FDR-adjusted p < 0.05 and FC > =1.3). Pathway analysis on these 98 genes revealed significantly upregulated relevant pathways by the purported generic including inflammatory response (p < 8.67 × 10−19 ), and immune cell adhesion (p < 7.57 × 10−17 ), suggesting that the purported generic may upregulate inflammatory pathways, possibly increasing the risk of adverse events and/or reducing the efficacy of treatment. Comparing Copaxone with differently manufactured glatiramoids and detecting batch-to-batch variability: summary of advanced analyses as described by Towfic et al. (16)
A follow-up analysis focused first on assessing batch-to-batch manufacturing consistency. It was found that fourfold more probesets had significantly higher variability across the 5 batches of purported generic than across the 30 Copaxone batches (F-test), indicating that Copaxone is very consistent from batch to batch, whereas the purported generic has a less consistent impact on expression across batches. In ranking all probesets by the ratio of their variance in purported generic to their variance in Copaxone, the highest ranked was for FOXP3. This finding may have significant implications for the efficacy of the purported generic because FOXP3 is foxhead box P3, a transcription factor that is a key marker of immunosuppressive regulatory T cells (Tregs). Tregs play an important role in preventing autoimmune damage – particularly in the context of multiple sclerosis (MS) (17), where patients have significantly reduced FOXP3 expression (18). One of the mechanisms by which Copaxone induces its beneficial effects is thought to be by mediating immune-tolerance through the induction of Tregs (7). The possibility that Copaxone induces immunosuppressive Treg cells more effectively than the purported generic was further elucidated through the analysis of differentially expressed genes. More than 700 probesets were found to have significantly different expression levels between Copaxone and the purported generic with adj p < 0.05. Among these, FOXP3 had higher expression in Copaxone than purported generic (adj p < 1.37 × 10−3 by anova). Moreover, using gene set enrichment analysis (GSEA) revealed that FOXP3 targets are indeed upregulated to a greater extent by Copaxone than by the purported generic. Other genes and pathways were found to be upregulated to a greater extent by the purported generic than by Copaxone. For instance, CD14, a key marker of inflammatory monocytes, was expressed at higher levels following activation by purported generic (adj p < 0.0203 by anova). CD14 is involved in the inflammatory response to lipopolysaccharide (LPS), and the LPS response pathway was also enriched among the genes upregulated by purported generic relative to Copaxone (adj. p < 4.96 × 10−6 ). These findings are concerning from a safety perspective, suggesting the possibility that the purported generic could induce inflammatory responses that are harmful to MS patients.
In summary, these two published studies utilized gene expression data and advanced analyses, finding differences between Copaxone and a differently manufactured glatiramoid purported to be a generic for GA that warrant further investigation from a safety and efficacy perspective. Comparing biosimilars – imig vs vela for Gaucher disease – summary of studies described by Dasgupta et al. (19)
Biosimilars are follow-on products for complex biological drugs such as antibodies and proteins. In May 2014, the FDA release draft guidance on demonstrating biosimilarity to a reference product (1). Biosimilars bear many similarities to purported generic versions of NBCDs, and comparing each to its respective reference product is challenging. A recent study (19) compared two enzyme replacement therapies for the treatment of Gaucher disease type 1, imiglucerase (imig) and velaglucerase alfa (vela). In this genetic, autosomal recessive disease, mutations in GBA1 cause dysfunction in the lysosomal enzyme glucocerebrosidase, leading to accumulation of glucosylceramide and glucosylsphingosine in macrophages which drives severe organ damage. Gaucher disease type 1 is rare, with an incidence of approximately 1:40,000 (20). Imig was developed first, and approved by the FDA in 1994. Vela was developed later as a biosimilar, and its safety and effectiveness was assessed in three clinical studies before it received FDA approval in 2010. Vela was shown to be non-inferior to imig with no significant differences in secondary endpoints, but differences in immunogenicity were observed, with a higher number of imig-treated patients developing anti-drug antibodies (21). In the gene expression study (19), mice with a mutation in the glucocerebrosidase gene, as well as wild-type mice, were injected weekly with either imig, vela, or saline control for 8 weeks, then killed and their livers, lungs, and spleens were isolated and the RNA extracted. Multiple biological replicates (n = 4–8) were used for each treatment and tissue type. Gene expression was measured using both microarray technology (Affymetrix Mouse Gene 1.0 ST chip, Affymetrix, Inc. Santa Clara, CA, USA) and RNA-Seq (Illumina Hi-Seq2000, Illumina, Inc. San Diego, CA, USA). Outliers were identified and removed using principle component analysis (PCA). Differentially expressed genes (FDR-adjusted p < 0.05, fold change ≥ 1.5) were identified by direct comparison between imig and vela using mixed model anova for the microarray data, and both edgeR and DESeq for the RNA-Seq data. Ninety-seven such genes were identified by microarray, whereas 290 were identified by RNA-seq. This demonstrates the greater potential sensitivity of RNA-seq as a means for comparing medicines. Forty-seven of the differentially expressed genes were identified by both RNA-seq and microarray. These genes represented pathways and processes including cell division/proliferation, the hematopoietic system, and inflammatory/macrophage-related pathways. The inflammatory/macrophage pathway was specifically
3
Kolitz et al. higher in imig than vela (>1.5 FC, p = 5.53 × 10−12 to 5.94 × 10−3 ) which could potentially explain the higher incidence of anti-drug antibodies in imig patients. Comparing distinct medicines with similar targets or purposes Vaccine adjuvants
In one recent study (22), investigators compared the biological impact of various vaccine adjuvants using gene expression methods. Adjuvants are used to enhance the immune response to the antigens within a vaccine. In this study, female Balb/c mice (3 per condition per experiment, in two experiments) were injected with adjuvants including the mineral salt aluminum hydroxide (alum), glucopyranosyl lipid A (GLA), a squalene oil emulsion known as stable emulsion (SE), glucopyranosyl lipid A in a stable emulsion formulation (GLA-SE), or a control (phosphate buffered saline, PBS). Muscle, lymph, blood, and serum samples were obtained at timepoints of 6, 24, 48, and 96 h following treatment. The RNA was extracted, and transcript levels measured using the Affymetrix Mouse Genome 430 2.0 GeneChip (Affymetrix, Inc. Santa Clara, CA, USA). The investigators found that alum induced a very small number of genes (no more than 60) at any timepoint. In contrast, hundreds to thousands of genes were induced by GLA and GLA-SE. Nearly all (97%) of the genes induced by GLA were also induced by GLA-SE, including a variety of immunologically relevant genes such as chemokines and cytokines. GLA-SE also induced many other genes that GLA did not, some of which were common with the genes induced by SE alone. The timepoints at which the greatest number of genes were induced differed between adjuvants. The investigators conclude that GLA-SE induces a greater magnitude and duration of transcriptional activation than GLA, and that GLA-SE has a mechanism of action involving the induction of Th1 responses including chemokines that attract lymphocytes and antigen presenting cells (APCs). GLA-SE was concluded to induce the strongest innate immune response of all the adjuvants tested. The investigators also point out that this study has shed new light on the mechanism of action of oil-in-water emulsion adjuvants such as SE. Anti-TNF-alpha therapies
Since 1998, anti-TNF-alpha therapies have been in clinical use for the treatment of rheumatoid arthritis (RA). Etanercept is a soluble TNF-alpha receptor that first received FDA approval in 1998, whereas adalimumab is an anti-TNF-alpha antibody that first received FDA approval in 2002. As both target the same pathway, their precise mechanisms (and similarities/differences) remain to be elucidated. In one small study (23), peripheral blood mononuclear cells (PBMCs) were drawn from RA patients before receiving either etanercept (N = 4) or adalimumab (N = 4), and 12 weeks after treatment. RNA was extracted, and gene expression measured using human pangenomic oligonucleotide microarrays
4
produced by the French Genopole Network (RNG) to measure 25,341 genes. The study found no statistically significant differences between the gene expression profiles induced by etanercept and adalimumab. The most likely explanation for this result, however, is low statistical power. Only eight total patients samples were used, four from each medicine. With such a small number of samples, the study is only powered to detect extremely large fold changes (differences were reported with fold changes >1.5 in either direction), not the types of subtle yet important differences expected to be affected differentially by medicines of somewhat similar mechanism of action (MoA). Despite the inability to distinguish between etanercept and adalimumab, the study nevertheless found valuable information regarding the mechanism of anti-TNF-alpha therapies by comparing the baseline and 12-month timepoints. Both medicines were found to significantly reduce expression of CD14 and other inflammation-related genes, as well as additional pathways of relevance for elucidating mechanism of action. As the study found that no major differences occurred, more subtle yet important differences remain to be investigated in larger datasets. New comparisons from existing data
One resource for pursuing comparisons of gene expression profiles induced by drugs is the Connectivity Map (CMAP), a resource produced by the Broad Institute of MIT and Harvard. CMAP contains 7000+ gene expression profiles representing the impact of over 1300+ compounds in a standardized set of cell lines (24, 25). Recently, Aramadhaka et al. (26) conducted a study profiling cells treated with Heloderma suspectum venom. The authors utilized CMAP to compare the transcription profiles of cells exposed to venom against gene expression profiles from known small molecules in CMAP. The study found over 2000 correlated profiles of expression changes in response to drugs in CMAP. Interestingly, when Aramadhaka et al. intersected the drug profiles from venom-treated cells to another CMAP analysis conducted using expression profiles from cells exposed to exendin-4 (component of the Heloderma suspectum venom – a synthetic form of exendin-4 is currently used to treat type 2 diabetes), they found several anti-diabetic drugs alongside other small molecules that could potentially have anti-diabetic characteristics. This study provides an excellent example of utilizing CMAP to indirectly compare molecules and perhaps assign new functionality to already existing molecules based on correlation with existing drugs. In two studies focusing on drug repositioning – the idea of determining new disease applications for existing drugs – the Butte group (27, 28) showed that using publicly available gene expression data from the Gene Expression Omnibus (GEO) to obtain disease-specific gene expression signatures and comparing these signatures to drug treatment gene expression signatures found in CMAP enabled predictions about which drugs might be effective in diseases beyond those for which they had been developed. The authors experimentally
Comparing medicines validated several such predictions in mouse models. This approach also enabled clustering of drugs by their gene expression profiles, yielding interesting potential predictions about similarities in effect between drugs not previously thought to be of similar classes (e.g. salicylate drugs and the calcium channel blocker verapamil). The growing quantities of gene expression data available in such compendia, as well as new tools, are increasingly enabling these new types of analyses. For example, the DvD pipeline, usable via R or Cytoscape, allows users to compare such publicly available expression profiles from drugs and disease (29).
Generalizing: data analysis considerations
As shown in the examples above, performing a thorough and advanced analysis is also critical to properly identify robust differences between medicines. Proper normalization and batch correction are critical in order to achieve an appropriate basis for comparison between medicines. The subsequent application of appropriate methods to identify differentially expressed genes and elucidate the biological mechanisms of those genes is also crucial to better understand its clinical consequences. Each of these aspects of experimental design and data analysis contribute along with the data collection itself to producing meaningful results.
Studies examining subsets of genes
In addition to the microarray-based studies described above, other studies have identified differences in medicines by specifically analyzing particular subsets of genes. For instance, one recent study used qPCR to show that different mood stabilizing medicines varied in their impact on the expression of specific housekeeping genes in human cell lines derived from mood disorder patients (30). Drug-specific effects of lithium and sodium valproate were seen on the expression of housekeeping genes; in addition, acute (24 h) vs chronic (1 week) lithium treatment produced differing effects on housekeeping gene expression. These results are particularly important because such genes are often used as reference genes for normalization in qPCR-based studies, and choosing a housekeeping gene that is actually affected by a medicine under study could result in erroneous results from an entire dataset. Generalizing: experimental design considerations
For investigators seeking to design studies to assess the degree of similarity between two medicines, proper experimental design is critical. Perhaps the most fundamental aspect of experimental design involves selecting the appropriate model systems reflecting the disease in humans in order to capture the most relevant biology and its clinical consequences. Second only to this is randomization in terms of the order of handling of samples in the laboratory. Samples of the two medicines must be handled identically and interspersed, not separated into groups or run on different chips. Power is also critical. A sufficient number of samples must be run in order to detect the expected fold change levels and achieve statistical significance. The appropriate number of samples to achieve the desired level of power can be calculated for a particular experimental context using estimates of treatment effect size and standard deviation (31). Finally, selecting an appropriate gene chip (e.g. taking into account considerations such as coverage, relevance of splice isoforms in the case of gene expression, and comparability to other studies) is also a key. All of these aspects of experimental design contribute to the quality of the resulting data and thus the clarity of the results that can be drawn, and the ability to discriminate accurately between medicines.
Generalizing: limitations of model systems
It should be noted that the no model system can fully capture the activity of a medicine in humans. For instance, the microbiome has been shown to play a role in drug metabolism (32, 33) that is difficult to model. Along the same lines, the impact of specific medicines on the microbiome is not captured by traditional model systems. Similarly, the activity of a drug can modulate a broader network of genes or proteins in a way that is difficult to capture. Each model system provides a small window into a medicine’s activity, but it is important to bear in mind that there is substantial added complexity to the drug’s mechanism in humans. Conclusions
The emerging science of applying genomic technologies to compare medicines has many potential applications for ensuring that patients receive the safest and most effective medicines possible, and for facilitating the successful development of new medicines. This also provides a powerful tool for repositioning medications to additional indications. As the field is nascent and constantly evolving, there are already many thoughtful examples of the use of these approaches, and their use will only continue to grow. The authors believe that high-resolution comparative gene profiling studies will become a mainstay in the characterization of NBCDs and biologics, as well as in early R&D experiments seeking to outperform existing drugs. It is becoming widely accepted that in-depth characterization of a drug’s mechanism of action and safety profile is essential for assessment of long-term efficacy and safe use in the market, and that genome-wide gene expression studies conducted across a variety of cell types and model systems can form one helpful piece of the greater puzzle. While such studies yield useful information, they at best provide only a small window into the complexity of a patient, which can only be fully captured through a clinical trial. Acknowledgements We acknowledge funding from Teva Pharmaceutical Industries. We acknowledge David Ladkani, Rivka Schwartz, and Pippa Loupe for helpful feedback.
5
Kolitz et al. References 1. U.S. Food and Drug Administration. Guidance for Industry: Clinical Pharmacology Data to Support a Demonstration of Biosimilarity to a Reference Product. 2014, from http://www.fda.gov/downloads/ Drugs/GuidanceComplianceRegulatoryInformation/Guidances/UCM3 97017.pdf. Accessed June 12, 2014. 2. Holloway C et al. Scientific considerations for complex drugs in light of established and emerging regulatory guidance. Ann N Y Acad Sci 2012: 1276: 26–36. 3. Schellekens H et al. The therapeutic equivalence of complex drugs. Regul Toxicol Pharmacol 2011: 59: 176–183. 4. Conner JB, Bawa R, Michael Nicholas J, Weinstein V. Handbook of clinical nanomedicine – bench bedside. Singapore: Pan Stanford Publishing Pte Ltd, 2015. 5. Varkony H et al. The glatiramoid class of immunomodulator drugs. 2009, from http://informahealthcare.com/doi/abs/10.1517/1465656090 2802877 6. Josephson SA, Johnston SC, Hauser SL. Blockbusters for cheap? The difficulties surrounding generic biologics. Ann Neurol 2013: 74: A7–A8. 7. Putheti P, Soderstrom M, Link H, Huang Y-M. Effect of glatiramer acetate (Copaxone) on CD4+CD25high T regulatory cells and their IL-10 production in multiple sclerosis. J Neuroimmunol 2003: 144: 125–131. 8. Farina C, Weber MS, Meinl E, Wekerle H, Hohlfeld R. Glatiramer acetate in multiple sclerosis: update on potential mechanisms of action. Lancet Neurol 2005: 4: 567–575. 9. Neuhaus O, Farina C, Wekerle H, Hohlfeld R. Mechanisms of action of glatiramer acetate in multiple sclerosis. Neurology 2001: 56: 702–708. 10. Miller A et al. Treatment of multiple sclerosis with Copolymer-1 (Copaxone® ): implicating mechanisms of Th1 to Th2/Th3 immunedeviation. J Neuroimmunol 1998: 92: 113–121. 11. Duda PW, Schmied MC, Cook SL, Krieger JI, Hafler DA. Glatiramer acetate (Copaxone® ) induces degenerate, Th2-polarized immune responses in patients with multiple sclerosis. J Clin Invest 2000: 105: 967–976. 12. Vieira PL, Heystek HC, Wormmeester J, Wierenga EA, Kapsenberg ML. Glatiramer acetate (Copolymer-1, Copaxone) Promotes Th2 cell development and increased IL-10 production through modulation of dendritic cells. J Immunol 2003: 170: 4483–4488. 13. Haas J et al. Glatiramer acetate improves regulatory T-cell function by expansion of naive CD4(+)CD25(+)FOXP3(+)CD31(+) T-cells in patients with multiple sclerosis. J Neuroimmunol 2009: 216: 113–117. 14. Hong J, Li N, Zhang X, Zheng B, Zhang JZ. Induction of CD4+CD25+ regulatory T cells by copolymer-I through activation of transcription factor Foxp3. Proc Natl Acad Sci USA 2005: 102: 6449–6454. 15. Bakshi S et al. Gene expression analysis reveals functional pathways of glatiramer acetate activation. Expert Opin Ther Targets 2013: 17: 351–362. 16. Towfic F et al. Comparing the biological impact of glatiramer acetate with the biological impact of a generic. PLoS One 2014: 9: e83757.
6
17. Fletcher JM et al. CD39+Foxp3+ regulatory T cells suppress pathogenic Th17 cells and are impaired in multiple sclerosis. J Immunol 2009: 183: 7602–7610. 18. Huan J et al. Decreased FOXP3 levels in multiple sclerosis patients. J Neurosci Res 2005: 81: 45–52. 19. Dasgupta N et al. Gaucher disease: transcriptome analyses using microarray or mRNA sequencing in a Gba1 mutant mouse model treated with velaglucerase alfa or imiglucerase. PLoS One 2013: 8: e74912. 20. Mehta A. Epidemiology and natural history of Gaucher’s disease. Eur J Intern Med 2006: 17 (Suppl): S2–S5. 21. Turkia HB et al. Velaglucerase alfa enzyme replacement therapy compared with imiglucerase in patients with Gaucher disease. Am J Hematol 2013: 88: 179–184. 22. Lambert SL et al. Molecular and cellular response profiles induced by the TLR4 agonist-based adjuvant glucopyranosyl lipid A. PLoS One 2012: 7: e51618. 23. Meugnier E et al. Gene expression profiling in peripheral blood cells of patients with rheumatoid arthritis in response to anti-TNF-alpha treatments. Physiol Genomics 2011: 43: 365–371. 24. Lamb J et al. The Connectivity Map: using gene-expression signatures to connect small molecules, genes, and disease. Science 2006: 313: 1929–1935. 25. Lamb J. The Connectivity Map: a new tool for biomedical research. Nat Rev Cancer 2007: 7: 54–60. 26. Aramadhaka LR, Prorock A, Dragulev B, Bao Y, Fox JW. Connectivity maps for biosimilar drug discovery in venoms: the case of Gila monster venom and the anti-diabetes drug Byetta® . Toxicon 2013: 69: 160–167. 27. Dudley JT et al. Computational repositioning of the anticonvulsant topiramate for inflammatory bowel disease. Sci Transl Med 2011: 3: 96ra76. 28. Sirota M et al. Discovery and preclinical validation of drug indications using compendia of public gene expression data. Sci Transl Med 2011: 3: 96ra77. 29. Pacini C et al. DvD: an R/Cytoscape pipeline for drug repurposing using public repositories of gene expression data. Bioinformatics 2013: 29: 132–134. 30. Powell TR et al. Mood-stabilizers differentially affect housekeeping gene expression in human cells. Int J Methods Psychiatr Res 2014: 23(2): 279–288, DOI: 10.1002/mpr.1435. 31. Orr M, Liu P. Sample size estimation while controlling false discovery rate for microarray experiments using the ssize.fdr package. R J 2009: 1: 47–53. 32. Clayton TA et al. Pharmaco-metabonomic phenotyping and personalized drug treatment. Nature 2006: 440: 1073–1077. 33. Clayton TA, Baker D, Lindon JC, Everett JR, Nicholson JK. Pharmacometabonomic identification of a significant host-microbiome metabolic interaction affecting human drug metabolism. Proc Natl Acad Sci USA 2009: 106: 14728–14733.
© 2014 John Wiley & Sons A/S. Published by John Wiley & Sons Ltd CLINICAL GENETICS doi: 10.1111/cge.12462
Review
Use of genetic technologies to compare medicines Kolitz S.E., Towfic F., Grossman I., Hayden M.R., Zeskind B. Use of genetic technologies to compare medicines. Clin Genet 2014. © John Wiley & Sons A/S. Published by John Wiley & Sons Ltd, 2014 In order to ensure that patients receive the safest and most effective medicines possible, it is often necessary to compare medicines and assess the extent to which they are similar in their clinical impact. Full clinical trials with appropriate endpoints remain the only method to compare the clinical impact of two medicines with absolute certainty. Other available methods (including physicochemical analysis, genomics, and transcriptomics) can provide partial information about certain aspects of a medicine’s biological impact, with possible clinical implications. Especially for biologics and non-biological complex drugs, which are more difficult to characterize by physicochemical means than small molecules, genomics and transciptomic studies can yield valuable insights for physicians, regulators, and drug developers. In this review, we cite and summarize a variety of studies that exemplify the emerging science of applying genomics and transcriptomics technologies to compare medicines. We discuss key aspects of experimental design, conduct of genetic assays, and advanced data analysis, all of which are critical for the successful execution of such studies. Finally, we propose new areas for which such studies can be applied to maximize patient benefit and reduce safety issues. Conflict of interest
The authors are employees of either Teva Pharmaceutical Industries or Immuneering Corporation. Teva is the maker of Copaxone, a therapy discussed in this review. Immuneering Corporation receives funding through a partnership with Teva.
The explosion in the use of genetic technologies over the past decade has created countless opportunities to compare medicines, and mechanisms underlying response to medicines, in new ways. These technologies range from microarrays for measuring the expression levels of tens of thousands of genes, to single nucleotide polymorphism (SNP) chips that detect hundreds of thousands or millions of specific genetic variants, to sequencing and next generation sequencing technologies that provide extensive information on genetic variants. The initial wave of applications for these technologies focused on a few key areas, such as identifying genetic biomarkers of response to medicines, identifying pharmacodynamic markers, and seeking to understand the biological mechanisms underlying specific diseases
S.E. Kolitza,† , F. Towfica,† , I. Grossmanb , M.R. Haydenb and B. Zeskinda a Immuneering Corporation, Cambridge, MA, USA and b Teva Pharmaceutical Industries, Petach Tikva, Israel † These
authors contributed equally.
Key words: biosimilars – computational biology – gene expression – generics – microarray Corresponding author: Benjamin Zeskind, Immuneering Corporation, Cambridge, MA 02142, USA. Tel.: +1 617 5008080x121; fax: +1 617 4499588; e-mail: [email protected] Received 13 June 2014, revised and accepted for publication 16 July 2014
and response to therapies. In parallel, a newer yet equally useful set of applications has been emerging, in which the extent to which two medicines are similar is assessed. Such comparisons are less useful for small molecule drugs, which can be established to be equivalent by physicochemical analytical methods, but more useful for biosimilars and purported generic versions of non-biological complex drugs (NBCDs), both of which are intended to be similar, yet difficult to compare via conventional analytical methods. Examples in which genomics has been applied to make such comparisons are increasingly appearing in the literature, representing an extremely powerful use of such technologies. In this review, we summarize exciting new studies applying genomic technologies to compare Copaxone
1
Kolitz et al. with differently manufactured glatiramoids (NBCDs), finding differences with important implications for safety. We describe studies comparing a biosimilar for Gaucher’s disease to its reference medicine, as well as studies comparing distinct medicines with similar targets or purposes. After thoroughly surveying the landscape of this emerging application for genomics, we conclude with recommendations for the proper design of such studies. Although the information provided by these genomic studies is extremely valuable, it is important to clearly state at the outset that there is no substitute for well-controlled and adequately designed clinical trials for determining that a medicine is safe and effective. No assay technology or analysis method has been determined to be capable of fully characterizing the complexity and scope of human biology, nor predicting with 100% certainty how a medicine will perform in the clinic, and data from each method should be looked at in concert with data from other characterization methods. Nevertheless, the studies described below yield highly useful insights into the biological impact of various medicines. Safety and efficacy of biologics and NBCDs
For traditional small-molecule-based medicines, the exact chemical structure and the physicochemical properties are known, and therefore it can be unequivocally established that a medicine produced by a different manufacturing process (e.g. a purported generic) is the same as the reference listed drug, and should induce the same gene expression signature. However, a substantial number of the most effective new medicines today are not small molecules, but rather biologics or NBCDs. For these advanced medicines, it is much more challenging to characterize the extent to which a differently manufactured product is similar. The FDA recently released draft guidance describing a framework for considering biosimilars, follow-on products for complex biological drugs including antibodies such as anti-TNF-alpha and proteins such as filgrastim (G-CSF), and recognized the utility of genomic data by stating that ‘Using broader panels of biomarkers (e.g. by conducting a protein or mRNA microarray analysis) that capture multiple pharmacological effects of the product may be of additional value’ (1). A similar challenge faces regulators in assessing NBCDs such as low-molecular weight heparins, liposomal drugs, iron-carbohydrate drugs, and glatiramoids (2, 3). For both biosimilars and NBCDs, investigators have hypothesized that genomic technologies can provide valuable information. By first conducting proof-ofconcept studies in a relevant biological system, using robust statistics at the pathway and gene, investigators are able produce a hypothesis regarding similarities and differences that can be subsequently investigated in independent, human-relevant model systems, confirmed by polymerase chain reaction (PCR), etc. Aspects of this approach are illustrated in the examples below.
2
Comparing Copaxone with differently manufactured glatiramoids: background and experimental design
With approximately 2,000,000 patient-years of experience in 57 countries since the 1990s, Copaxone has proven to be a safe and effective treatment option for patients with relapsing forms of multiple sclerosis. Glatiramer acetate (GA), the active substance of Copaxone, is a synthetic polypeptide mixture produced by copolymerization of l-glutamic acid, l-alanine, l-tyrosine, and l-lysine with an average molar fraction of 0.141, 0.427, 0.095, and 0.338, respectively. Copaxone forms a complex colloidal suspension that has proven extremely challenging to fully characterize, despite extensive efforts using the latest technologies (4). Owing to the difficulties in characterizing Copaxone molecularly (5), standard physicochemical and biochemical methods have also proven ineffective for comparing Copaxone to glatiramoids produced in different manufacturing conditions (6). This is due to the fact that most molecular assays for comparing generics are designed for comparing small molecules where the atomic structure of the molecule can be fully characterized. Despite the difficulties characterizing the molecular structure of the drug, decades of research into the mechanism of action of Copaxone (7–10) have shown progress in characterizing the function of the drug. Specifically, published research indicates that Copaxone interacts with the immune system to cause a shift toward a type 2 (Th2) helper T-cell cytokine profile (9, 11), that antigen presenting cells are a prime target for the immunomodulatory effects of GA and essential for the GA mediated shift toward a Th2 profile (12), and that GA induces FOXP3+ regulatory T cells (Tregs) in patients (13, 14). Each of these mechanisms is believed to be important, although specific gene signatures and the mechanism by which the drug causes the shift are not fully elucidated. Two (15, 16) recent publications have proposed new methods of comparing Copaxone with differently manufactured glatiramoids. For these studies, isolated splenocytes from female mice treated with GA were exposed ex vivo to a variety of differently manufactured glatiramoids. Gene expression of 45,000+ transcripts was measured. The sample/chip assignments were randomized to avoid batch effects. The findings of these studies are summarized below. Comparing Copaxone with differently manufactured glatiramoids: summary of initial analyses as described by Bakshi et al. (15)
GA was found to modulate 1474 genes relative to medium (by t-test with fold change (FC) ≥ 1.3 and false discovery rate (FDR)-adjusted p < 0.05), and functional analysis of this list indicated that it included many relevant pathways including T-helper cell differentiation (16). GA drug product was also compared to a differently manufactured glatiramoid purported to be a generic for GA (Glatimer® , Natco Pharma, Ltd., Hyderabad, India). Comparing eight samples from three manufacturing batches of purported generic to Copaxone, 98
Comparing medicines genes were found to be differentially expressed (one-way anova, FDR-adjusted p < 0.05 and FC > =1.3). Pathway analysis on these 98 genes revealed significantly upregulated relevant pathways by the purported generic including inflammatory response (p < 8.67 × 10−19 ), and immune cell adhesion (p < 7.57 × 10−17 ), suggesting that the purported generic may upregulate inflammatory pathways, possibly increasing the risk of adverse events and/or reducing the efficacy of treatment. Comparing Copaxone with differently manufactured glatiramoids and detecting batch-to-batch variability: summary of advanced analyses as described by Towfic et al. (16)
A follow-up analysis focused first on assessing batch-to-batch manufacturing consistency. It was found that fourfold more probesets had significantly higher variability across the 5 batches of purported generic than across the 30 Copaxone batches (F-test), indicating that Copaxone is very consistent from batch to batch, whereas the purported generic has a less consistent impact on expression across batches. In ranking all probesets by the ratio of their variance in purported generic to their variance in Copaxone, the highest ranked was for FOXP3. This finding may have significant implications for the efficacy of the purported generic because FOXP3 is foxhead box P3, a transcription factor that is a key marker of immunosuppressive regulatory T cells (Tregs). Tregs play an important role in preventing autoimmune damage – particularly in the context of multiple sclerosis (MS) (17), where patients have significantly reduced FOXP3 expression (18). One of the mechanisms by which Copaxone induces its beneficial effects is thought to be by mediating immune-tolerance through the induction of Tregs (7). The possibility that Copaxone induces immunosuppressive Treg cells more effectively than the purported generic was further elucidated through the analysis of differentially expressed genes. More than 700 probesets were found to have significantly different expression levels between Copaxone and the purported generic with adj p < 0.05. Among these, FOXP3 had higher expression in Copaxone than purported generic (adj p < 1.37 × 10−3 by anova). Moreover, using gene set enrichment analysis (GSEA) revealed that FOXP3 targets are indeed upregulated to a greater extent by Copaxone than by the purported generic. Other genes and pathways were found to be upregulated to a greater extent by the purported generic than by Copaxone. For instance, CD14, a key marker of inflammatory monocytes, was expressed at higher levels following activation by purported generic (adj p < 0.0203 by anova). CD14 is involved in the inflammatory response to lipopolysaccharide (LPS), and the LPS response pathway was also enriched among the genes upregulated by purported generic relative to Copaxone (adj. p < 4.96 × 10−6 ). These findings are concerning from a safety perspective, suggesting the possibility that the purported generic could induce inflammatory responses that are harmful to MS patients.
In summary, these two published studies utilized gene expression data and advanced analyses, finding differences between Copaxone and a differently manufactured glatiramoid purported to be a generic for GA that warrant further investigation from a safety and efficacy perspective. Comparing biosimilars – imig vs vela for Gaucher disease – summary of studies described by Dasgupta et al. (19)
Biosimilars are follow-on products for complex biological drugs such as antibodies and proteins. In May 2014, the FDA release draft guidance on demonstrating biosimilarity to a reference product (1). Biosimilars bear many similarities to purported generic versions of NBCDs, and comparing each to its respective reference product is challenging. A recent study (19) compared two enzyme replacement therapies for the treatment of Gaucher disease type 1, imiglucerase (imig) and velaglucerase alfa (vela). In this genetic, autosomal recessive disease, mutations in GBA1 cause dysfunction in the lysosomal enzyme glucocerebrosidase, leading to accumulation of glucosylceramide and glucosylsphingosine in macrophages which drives severe organ damage. Gaucher disease type 1 is rare, with an incidence of approximately 1:40,000 (20). Imig was developed first, and approved by the FDA in 1994. Vela was developed later as a biosimilar, and its safety and effectiveness was assessed in three clinical studies before it received FDA approval in 2010. Vela was shown to be non-inferior to imig with no significant differences in secondary endpoints, but differences in immunogenicity were observed, with a higher number of imig-treated patients developing anti-drug antibodies (21). In the gene expression study (19), mice with a mutation in the glucocerebrosidase gene, as well as wild-type mice, were injected weekly with either imig, vela, or saline control for 8 weeks, then killed and their livers, lungs, and spleens were isolated and the RNA extracted. Multiple biological replicates (n = 4–8) were used for each treatment and tissue type. Gene expression was measured using both microarray technology (Affymetrix Mouse Gene 1.0 ST chip, Affymetrix, Inc. Santa Clara, CA, USA) and RNA-Seq (Illumina Hi-Seq2000, Illumina, Inc. San Diego, CA, USA). Outliers were identified and removed using principle component analysis (PCA). Differentially expressed genes (FDR-adjusted p < 0.05, fold change ≥ 1.5) were identified by direct comparison between imig and vela using mixed model anova for the microarray data, and both edgeR and DESeq for the RNA-Seq data. Ninety-seven such genes were identified by microarray, whereas 290 were identified by RNA-seq. This demonstrates the greater potential sensitivity of RNA-seq as a means for comparing medicines. Forty-seven of the differentially expressed genes were identified by both RNA-seq and microarray. These genes represented pathways and processes including cell division/proliferation, the hematopoietic system, and inflammatory/macrophage-related pathways. The inflammatory/macrophage pathway was specifically
3
Kolitz et al. higher in imig than vela (>1.5 FC, p = 5.53 × 10−12 to 5.94 × 10−3 ) which could potentially explain the higher incidence of anti-drug antibodies in imig patients. Comparing distinct medicines with similar targets or purposes Vaccine adjuvants
In one recent study (22), investigators compared the biological impact of various vaccine adjuvants using gene expression methods. Adjuvants are used to enhance the immune response to the antigens within a vaccine. In this study, female Balb/c mice (3 per condition per experiment, in two experiments) were injected with adjuvants including the mineral salt aluminum hydroxide (alum), glucopyranosyl lipid A (GLA), a squalene oil emulsion known as stable emulsion (SE), glucopyranosyl lipid A in a stable emulsion formulation (GLA-SE), or a control (phosphate buffered saline, PBS). Muscle, lymph, blood, and serum samples were obtained at timepoints of 6, 24, 48, and 96 h following treatment. The RNA was extracted, and transcript levels measured using the Affymetrix Mouse Genome 430 2.0 GeneChip (Affymetrix, Inc. Santa Clara, CA, USA). The investigators found that alum induced a very small number of genes (no more than 60) at any timepoint. In contrast, hundreds to thousands of genes were induced by GLA and GLA-SE. Nearly all (97%) of the genes induced by GLA were also induced by GLA-SE, including a variety of immunologically relevant genes such as chemokines and cytokines. GLA-SE also induced many other genes that GLA did not, some of which were common with the genes induced by SE alone. The timepoints at which the greatest number of genes were induced differed between adjuvants. The investigators conclude that GLA-SE induces a greater magnitude and duration of transcriptional activation than GLA, and that GLA-SE has a mechanism of action involving the induction of Th1 responses including chemokines that attract lymphocytes and antigen presenting cells (APCs). GLA-SE was concluded to induce the strongest innate immune response of all the adjuvants tested. The investigators also point out that this study has shed new light on the mechanism of action of oil-in-water emulsion adjuvants such as SE. Anti-TNF-alpha therapies
Since 1998, anti-TNF-alpha therapies have been in clinical use for the treatment of rheumatoid arthritis (RA). Etanercept is a soluble TNF-alpha receptor that first received FDA approval in 1998, whereas adalimumab is an anti-TNF-alpha antibody that first received FDA approval in 2002. As both target the same pathway, their precise mechanisms (and similarities/differences) remain to be elucidated. In one small study (23), peripheral blood mononuclear cells (PBMCs) were drawn from RA patients before receiving either etanercept (N = 4) or adalimumab (N = 4), and 12 weeks after treatment. RNA was extracted, and gene expression measured using human pangenomic oligonucleotide microarrays
4
produced by the French Genopole Network (RNG) to measure 25,341 genes. The study found no statistically significant differences between the gene expression profiles induced by etanercept and adalimumab. The most likely explanation for this result, however, is low statistical power. Only eight total patients samples were used, four from each medicine. With such a small number of samples, the study is only powered to detect extremely large fold changes (differences were reported with fold changes >1.5 in either direction), not the types of subtle yet important differences expected to be affected differentially by medicines of somewhat similar mechanism of action (MoA). Despite the inability to distinguish between etanercept and adalimumab, the study nevertheless found valuable information regarding the mechanism of anti-TNF-alpha therapies by comparing the baseline and 12-month timepoints. Both medicines were found to significantly reduce expression of CD14 and other inflammation-related genes, as well as additional pathways of relevance for elucidating mechanism of action. As the study found that no major differences occurred, more subtle yet important differences remain to be investigated in larger datasets. New comparisons from existing data
One resource for pursuing comparisons of gene expression profiles induced by drugs is the Connectivity Map (CMAP), a resource produced by the Broad Institute of MIT and Harvard. CMAP contains 7000+ gene expression profiles representing the impact of over 1300+ compounds in a standardized set of cell lines (24, 25). Recently, Aramadhaka et al. (26) conducted a study profiling cells treated with Heloderma suspectum venom. The authors utilized CMAP to compare the transcription profiles of cells exposed to venom against gene expression profiles from known small molecules in CMAP. The study found over 2000 correlated profiles of expression changes in response to drugs in CMAP. Interestingly, when Aramadhaka et al. intersected the drug profiles from venom-treated cells to another CMAP analysis conducted using expression profiles from cells exposed to exendin-4 (component of the Heloderma suspectum venom – a synthetic form of exendin-4 is currently used to treat type 2 diabetes), they found several anti-diabetic drugs alongside other small molecules that could potentially have anti-diabetic characteristics. This study provides an excellent example of utilizing CMAP to indirectly compare molecules and perhaps assign new functionality to already existing molecules based on correlation with existing drugs. In two studies focusing on drug repositioning – the idea of determining new disease applications for existing drugs – the Butte group (27, 28) showed that using publicly available gene expression data from the Gene Expression Omnibus (GEO) to obtain disease-specific gene expression signatures and comparing these signatures to drug treatment gene expression signatures found in CMAP enabled predictions about which drugs might be effective in diseases beyond those for which they had been developed. The authors experimentally
Comparing medicines validated several such predictions in mouse models. This approach also enabled clustering of drugs by their gene expression profiles, yielding interesting potential predictions about similarities in effect between drugs not previously thought to be of similar classes (e.g. salicylate drugs and the calcium channel blocker verapamil). The growing quantities of gene expression data available in such compendia, as well as new tools, are increasingly enabling these new types of analyses. For example, the DvD pipeline, usable via R or Cytoscape, allows users to compare such publicly available expression profiles from drugs and disease (29).
Generalizing: data analysis considerations
As shown in the examples above, performing a thorough and advanced analysis is also critical to properly identify robust differences between medicines. Proper normalization and batch correction are critical in order to achieve an appropriate basis for comparison between medicines. The subsequent application of appropriate methods to identify differentially expressed genes and elucidate the biological mechanisms of those genes is also crucial to better understand its clinical consequences. Each of these aspects of experimental design and data analysis contribute along with the data collection itself to producing meaningful results.
Studies examining subsets of genes
In addition to the microarray-based studies described above, other studies have identified differences in medicines by specifically analyzing particular subsets of genes. For instance, one recent study used qPCR to show that different mood stabilizing medicines varied in their impact on the expression of specific housekeeping genes in human cell lines derived from mood disorder patients (30). Drug-specific effects of lithium and sodium valproate were seen on the expression of housekeeping genes; in addition, acute (24 h) vs chronic (1 week) lithium treatment produced differing effects on housekeeping gene expression. These results are particularly important because such genes are often used as reference genes for normalization in qPCR-based studies, and choosing a housekeeping gene that is actually affected by a medicine under study could result in erroneous results from an entire dataset. Generalizing: experimental design considerations
For investigators seeking to design studies to assess the degree of similarity between two medicines, proper experimental design is critical. Perhaps the most fundamental aspect of experimental design involves selecting the appropriate model systems reflecting the disease in humans in order to capture the most relevant biology and its clinical consequences. Second only to this is randomization in terms of the order of handling of samples in the laboratory. Samples of the two medicines must be handled identically and interspersed, not separated into groups or run on different chips. Power is also critical. A sufficient number of samples must be run in order to detect the expected fold change levels and achieve statistical significance. The appropriate number of samples to achieve the desired level of power can be calculated for a particular experimental context using estimates of treatment effect size and standard deviation (31). Finally, selecting an appropriate gene chip (e.g. taking into account considerations such as coverage, relevance of splice isoforms in the case of gene expression, and comparability to other studies) is also a key. All of these aspects of experimental design contribute to the quality of the resulting data and thus the clarity of the results that can be drawn, and the ability to discriminate accurately between medicines.
Generalizing: limitations of model systems
It should be noted that the no model system can fully capture the activity of a medicine in humans. For instance, the microbiome has been shown to play a role in drug metabolism (32, 33) that is difficult to model. Along the same lines, the impact of specific medicines on the microbiome is not captured by traditional model systems. Similarly, the activity of a drug can modulate a broader network of genes or proteins in a way that is difficult to capture. Each model system provides a small window into a medicine’s activity, but it is important to bear in mind that there is substantial added complexity to the drug’s mechanism in humans. Conclusions
The emerging science of applying genomic technologies to compare medicines has many potential applications for ensuring that patients receive the safest and most effective medicines possible, and for facilitating the successful development of new medicines. This also provides a powerful tool for repositioning medications to additional indications. As the field is nascent and constantly evolving, there are already many thoughtful examples of the use of these approaches, and their use will only continue to grow. The authors believe that high-resolution comparative gene profiling studies will become a mainstay in the characterization of NBCDs and biologics, as well as in early R&D experiments seeking to outperform existing drugs. It is becoming widely accepted that in-depth characterization of a drug’s mechanism of action and safety profile is essential for assessment of long-term efficacy and safe use in the market, and that genome-wide gene expression studies conducted across a variety of cell types and model systems can form one helpful piece of the greater puzzle. While such studies yield useful information, they at best provide only a small window into the complexity of a patient, which can only be fully captured through a clinical trial. Acknowledgements We acknowledge funding from Teva Pharmaceutical Industries. We acknowledge David Ladkani, Rivka Schwartz, and Pippa Loupe for helpful feedback.
5
Kolitz et al. References 1. U.S. Food and Drug Administration. Guidance for Industry: Clinical Pharmacology Data to Support a Demonstration of Biosimilarity to a Reference Product. 2014, from http://www.fda.gov/downloads/ Drugs/GuidanceComplianceRegulatoryInformation/Guidances/UCM3 97017.pdf. Accessed June 12, 2014. 2. Holloway C et al. Scientific considerations for complex drugs in light of established and emerging regulatory guidance. Ann N Y Acad Sci 2012: 1276: 26–36. 3. Schellekens H et al. The therapeutic equivalence of complex drugs. Regul Toxicol Pharmacol 2011: 59: 176–183. 4. Conner JB, Bawa R, Michael Nicholas J, Weinstein V. Handbook of clinical nanomedicine – bench bedside. Singapore: Pan Stanford Publishing Pte Ltd, 2015. 5. Varkony H et al. The glatiramoid class of immunomodulator drugs. 2009, from http://informahealthcare.com/doi/abs/10.1517/1465656090 2802877 6. Josephson SA, Johnston SC, Hauser SL. Blockbusters for cheap? The difficulties surrounding generic biologics. Ann Neurol 2013: 74: A7–A8. 7. Putheti P, Soderstrom M, Link H, Huang Y-M. Effect of glatiramer acetate (Copaxone) on CD4+CD25high T regulatory cells and their IL-10 production in multiple sclerosis. J Neuroimmunol 2003: 144: 125–131. 8. Farina C, Weber MS, Meinl E, Wekerle H, Hohlfeld R. Glatiramer acetate in multiple sclerosis: update on potential mechanisms of action. Lancet Neurol 2005: 4: 567–575. 9. Neuhaus O, Farina C, Wekerle H, Hohlfeld R. Mechanisms of action of glatiramer acetate in multiple sclerosis. Neurology 2001: 56: 702–708. 10. Miller A et al. Treatment of multiple sclerosis with Copolymer-1 (Copaxone® ): implicating mechanisms of Th1 to Th2/Th3 immunedeviation. J Neuroimmunol 1998: 92: 113–121. 11. Duda PW, Schmied MC, Cook SL, Krieger JI, Hafler DA. Glatiramer acetate (Copaxone® ) induces degenerate, Th2-polarized immune responses in patients with multiple sclerosis. J Clin Invest 2000: 105: 967–976. 12. Vieira PL, Heystek HC, Wormmeester J, Wierenga EA, Kapsenberg ML. Glatiramer acetate (Copolymer-1, Copaxone) Promotes Th2 cell development and increased IL-10 production through modulation of dendritic cells. J Immunol 2003: 170: 4483–4488. 13. Haas J et al. Glatiramer acetate improves regulatory T-cell function by expansion of naive CD4(+)CD25(+)FOXP3(+)CD31(+) T-cells in patients with multiple sclerosis. J Neuroimmunol 2009: 216: 113–117. 14. Hong J, Li N, Zhang X, Zheng B, Zhang JZ. Induction of CD4+CD25+ regulatory T cells by copolymer-I through activation of transcription factor Foxp3. Proc Natl Acad Sci USA 2005: 102: 6449–6454. 15. Bakshi S et al. Gene expression analysis reveals functional pathways of glatiramer acetate activation. Expert Opin Ther Targets 2013: 17: 351–362. 16. Towfic F et al. Comparing the biological impact of glatiramer acetate with the biological impact of a generic. PLoS One 2014: 9: e83757.
6
17. Fletcher JM et al. CD39+Foxp3+ regulatory T cells suppress pathogenic Th17 cells and are impaired in multiple sclerosis. J Immunol 2009: 183: 7602–7610. 18. Huan J et al. Decreased FOXP3 levels in multiple sclerosis patients. J Neurosci Res 2005: 81: 45–52. 19. Dasgupta N et al. Gaucher disease: transcriptome analyses using microarray or mRNA sequencing in a Gba1 mutant mouse model treated with velaglucerase alfa or imiglucerase. PLoS One 2013: 8: e74912. 20. Mehta A. Epidemiology and natural history of Gaucher’s disease. Eur J Intern Med 2006: 17 (Suppl): S2–S5. 21. Turkia HB et al. Velaglucerase alfa enzyme replacement therapy compared with imiglucerase in patients with Gaucher disease. Am J Hematol 2013: 88: 179–184. 22. Lambert SL et al. Molecular and cellular response profiles induced by the TLR4 agonist-based adjuvant glucopyranosyl lipid A. PLoS One 2012: 7: e51618. 23. Meugnier E et al. Gene expression profiling in peripheral blood cells of patients with rheumatoid arthritis in response to anti-TNF-alpha treatments. Physiol Genomics 2011: 43: 365–371. 24. Lamb J et al. The Connectivity Map: using gene-expression signatures to connect small molecules, genes, and disease. Science 2006: 313: 1929–1935. 25. Lamb J. The Connectivity Map: a new tool for biomedical research. Nat Rev Cancer 2007: 7: 54–60. 26. Aramadhaka LR, Prorock A, Dragulev B, Bao Y, Fox JW. Connectivity maps for biosimilar drug discovery in venoms: the case of Gila monster venom and the anti-diabetes drug Byetta® . Toxicon 2013: 69: 160–167. 27. Dudley JT et al. Computational repositioning of the anticonvulsant topiramate for inflammatory bowel disease. Sci Transl Med 2011: 3: 96ra76. 28. Sirota M et al. Discovery and preclinical validation of drug indications using compendia of public gene expression data. Sci Transl Med 2011: 3: 96ra77. 29. Pacini C et al. DvD: an R/Cytoscape pipeline for drug repurposing using public repositories of gene expression data. Bioinformatics 2013: 29: 132–134. 30. Powell TR et al. Mood-stabilizers differentially affect housekeeping gene expression in human cells. Int J Methods Psychiatr Res 2014: 23(2): 279–288, DOI: 10.1002/mpr.1435. 31. Orr M, Liu P. Sample size estimation while controlling false discovery rate for microarray experiments using the ssize.fdr package. R J 2009: 1: 47–53. 32. Clayton TA et al. Pharmaco-metabonomic phenotyping and personalized drug treatment. Nature 2006: 440: 1073–1077. 33. Clayton TA, Baker D, Lindon JC, Everett JR, Nicholson JK. Pharmacometabonomic identification of a significant host-microbiome metabolic interaction affecting human drug metabolism. Proc Natl Acad Sci USA 2009: 106: 14728–14733.
