Emerging Safety Science: Workshop Summary (2008)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

7 Qualifying Biomarkers M any of the biomarkers discussed at the workshop are obser- vational or exploratory in nature. Such biomarkers are useful for screening compounds for toxicity but have not been quali- fied or validated for use in regulatory decision making. Dr. Vonderscher discussed what is involved in transforming observational or exploratory biomarkers into valid biomarkers that can be used in making regulatory decisions. The ideal biomarker Vonderscher outlined the characteristics of an ideal biomarker for kidney toxicity: • It should be visible early, prior to histopathological changes, and should be indicative after active damage. • It should be sensitive, but it should also correlate with the severity of damage. • It should be accessible in the peripheral tissue; in the case of the kidney, for example, it should be measurable in either the blood or the urine. • It should be analytically stable in tissue so it can be measured after This chapter is based on the presentation of Jacky Vonderscher, Vice President, Head of Exploratory Development in Europe, Novartis. 65 

66 EMERGING SAFETY SCIENCE some time has passed, for example, after a biopsy has been taken or a necropsy performed. • It should be translational; that is, it should bridge across species. • It should be associated with a known mechanism. Many current biomarkers are identified through statistical analysis of gene expression, as discussed in Chapter 4, but one should be able to understand the biomarker and what is really going on in a biomolecular sense when it appears. • A biomarker should be able to localize damage. For example, it should pinpoint the particular area of the kidney that has been damaged rather than just indicating kidney toxicity in general. Given this extensive list of characteristics, a panel of biomarkers rather than any single ideal biomarker will likely be needed to character- ize nephrotoxicity. Qualification of Nephrotoxicity biomarkers Before attempting to establish pathways for clinical qualification of biomarkers, Novartis qualified a set of nephrotoxicity biomarkers in ani- mals. The qualification study was performed with 10 compounds: 8 nephro­ toxicants plus 2 hepatotoxicants as negative controls. For each compound, the researchers used 96 rats: four dose levels, including the control, which was a zero dose; four termination time points; and six animals per group. The duration of each exposure was 2–3 weeks. In addition to the traditional toxicology analysis, the researchers performed gene expression analysis on kidney and liver tissue and also multiplex ELISA (enzyme-linked immuno­ sorbent assay) on kidney, liver, urine, and plasma. The nephrotoxicants were chosen to have a variety of modes of toxic- ity, including oxidative stress and damage to podocytes. The hepatotoxi- cants were known to cause cholangitis and liver cancer. The team chose 15 biomarkers, representing 85 percent of the markers being used by the Predictive Safety Testing Consortium (PSTC), from various sources and publications, including some early gene expression work and some known proteomics work. The researchers attempted not to be selective about the source of the markers and to cover most of those that were inter- esting. Before running the studies, they performed a series of prestudies on the nephrotoxicants, in which they determined the correct doses to create lesions between grades 1 and 3. The PSTC public–private partnership, comprising members from industry, academia, and government, was established to identify and clinically qualify safety biomarkers. Novartis is a participant in the PSTC’s efforts to identify and qualify nephrotoxicity biomarkers. 

QUALIFYING BIOMARKERS 67 One of the key aspects of the process was settling on a lexicon of his- topathology. After extensive discussion among the PSTC members, a list that included 12 primary kidney lesions and a larger number of secondary lesions was assembled. Tubular cell degeneration was one of the primary lesion types, for instance; it was subdivided into two secondary types, necrosis and apoptosis. Each of these lesion types was further classified according to where the lesion was localized in the kidney: the proximal convoluted tubule, the thick descending tubule, the loop of Henle, etc. One category was “no precise localization possible.” Example: Establishing Biomarkers to Predict Cisplatin Toxicity After dosing the rats and examining them at various time points, the team identified 79 different types of localized lesions in the kidney. They then tried to determine how each biomarker correlated with the histo- pathological findings. Rats dosed with cisplatin evidenced tubular necro- sis and apoptosis. The team tried to identify biomarkers that predicted the damage and, in particular, that showed a quantitative relationship between the level of biomarker and the amount of damage. The researchers found that serum creatinine was not a particularly useful biomarker (see Figure 7-1) because, although some of the middle- dose animals had histopathology grades 1 and 2 (the highest), only the animals in the high-dose group had serum creatinine above the thresh- old. The results were similar for blood urea nitrogen (BUN): only the high-dose group showed BUN levels above the threshold, while a num- ber of the animals in the middle-dose group had pathology grades of 1 and 2. In contrast, Kim-1 (kidney injury molecule-1) was a much more effective biomarker for tubular necrosis and apoptosis (see Figure 7-2). Unlike creatinine and BUN, it was elevated not only in the high-dose group but also in the middle-dose group—but only in those animals that showed histopathology grades of 1 or 2. Furthermore, there was a clear correspondence between Kim-1 levels and histopathology grades, with the higher Kim-1 levels correlating with the higher histopathology grades. Only a few animals deviated from that pattern: one with a histo- pathology grade of 1 with Kim-1 levels slightly below the threshold, and four with a grade of 0 that fell somewhat above the Kim-1 threshold. The marker urinary clusterin exhibited properties similar to those of Kim-1, but it had more false negatives—that is, animals with levels below the threshold but with histopathology grades of 1 or 2. To obtain a quantitative measure of how well the various biomarkers predicted lesions, the team performed an ROC (receiver operating charac- teristic) analysis on the data. The animals were divided into two groups: control animals that had no lesions (histopathology grade of 0) and were 

68 1. 9 7 1. 8 1. 7 1. 6 Histopathology 1. 5 Grade: 0 1. 4 1 7 2 Creatinine 1. 3 7 1. 2 7 7 3 7 1. 1 7 3 Label: 14 3 14 3 7 1414 3 33 7 14 3 33 14 14 3 3 3 3 1414 1. 0 3 7 7 14 3 3 3 1414 3 14 3 7 14 Necropsy 3 7 7 14 14 3 14 7 7 14 7 7 14 0. 9 3 7 7 14 7 7 14 14 Day 7 14 14 0. 8 Controls 0.5m g/kg 1m g/kg 3m g/kg FIGURE 7-1  Serum creatinine as a biomarker for cisplatin-induced tubular necrosis/apoptosis. The data shown represent the fold change in serum creatinine at increasing dose levels. Results indicate that serum creatinine was not a particularly useful biomarker because, although some of the middle-dose animals had histopathology grades 1 and 2, they were comparable to control animals and only the animals in the high-dose group had serum creatinine above the control threshold. SOURCE: Vonderscher, 2007. Landscape view fig 7-1 

7 3 7 7 7 77 100 14 33 14 14 14 7 3 7 3 7 3 3 Histopathology 14 Grade: 10 0 3 3 1 7 14 2 14 14 UC_Kim1 3 14 14 7 77 33 77 7 7 14 14 14 14 33 14 14 3 14 7 14 Label: 3 3 33 14 7 3 77 Necropsy 14 14 14 3 7 14 3 7 Day 14 3 1 Cont ro ls 0.5 m g/kg 1 m g/kg 3 m g/kg FIGURE 7-2  Kim-1 as a biomarker for cisplatin-induced tubular necrosis/apoptosis. The data shown represent the fold change in urinary Kim-1 at increasing dose levels. Unlike creatinine, Kim-1 may be a useful biomarker for cisplatin-induced tubular necrosis/ apoptosis. Kim-1 was elevated not only in the high-dose group but also in the middle-dose group (in animals that showed histo- pathology grades of 1 or 2). There was a clear correspondence between Kim-1 levels and histopathology grades, with the higher fig 7-2 Kim-1 levels correlating with the higher histopathology grades. (Note that the concentration of Kim-1 in this figure is exhibited on a logarithmic scale, while the concentration of serum creatinine in Figure 7-1 is shown on a linear scale.) SOURCE: Vonderscher, 2007. 69 Landscape 

70 EMERGING SAFETY SCIENCE either nondosed or hepatotoxicant-dosed; and animals with lesions (histo- pathology grade of 1 or 2), regardless of their dose status. To generate an ROC curve, the true positive rate was graphed against the false positive rate as the threshold was varied continuously. The area under the ROC curve gives a quantitative measure of how good the predictions are: in the case of a perfect predictor, with a threshold that has all the positives above and all the negatives below, the area under the curve will be 1.0; in the case of a random predictor, the area under the curve will be 0.5. Vonderscher displayed a graph with the ROC curves for the four markers mentioned above: serum creatinine, BUN, Kim-1, and clusterin (see Figure 7-3). In the case of creatinine, the area under the curve was 0.53—just better than random. BUN was somewhat better, with an area under the curve of 0.62. Clusterin yielded an extremely good result, with an area under the curve of 0.93. But Kim-1 was nearly perfect, with an area under the curve of 0.99. 1 Area Under Curve: 0.9 Random = 0.5 0.8 Creatinine = 0.53 0.7 BUN = 0.62 0.6 Kim -1 = 0.99 Sensitivity 0.5 Clusterin = 0.93 0.4 0.3 0.2 27 Diseased 0.1 17 Controls 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1-Specificity FIGURE 7-3  ROC (receiver operating characteristic) analysis to compare bio- markers for cisplatin-induced tubular necrosis/apoptosis. The area under the curve for a biomarker that perfectly predicts cisplatin-induced tubular Color necrosis/apoptosis would be 1.0. In this experiment, creatinine had an area Figure 7 -3 under the curve of 0.53—just better than random; BUN was somewhat better, with an area under the curve of 0.62; clusterin yielded an extremely good re- sult, with an area under the curve of 0.93; and Kim-1 was nearly perfect, with an area under the curve of 0.99. SOURCE: Vonderscher, 2007. 

QUALIFYING BIOMARKERS 71 Analysis of the Remaining Nephrotoxicants Similar analyses were performed for the remaining nephrotoxicants and two hepatotoxicants. Using ROC curves, the team summarized how well the markers predicted various types of lesions caused by the com- pounds. In one analysis, for example, the team looked at proximal and nonlocalized tubular necrosis (see Figure 7-4). In this case, Kim-1 was still the best-performing biomarker, but its lead over clusterin was reduced, to 0.95 versus 0.93. The researchers found that creatinine and BUN per- formed much better when all of the compounds were included in the analysis rather than just cisplatin. The area under the ROC curve for cre- atinine was 0.83 and for BUN was 0.81. Part of the reason that the ROC 1 Area Under Curve: 0.9 Random = 0.5 0.8 Creatinine = 0.83 0.7 BUN = 0.81 0.6 Kim -1 = 0.95 Sensitivity 0.5 Clusterin = 0.93 0.4 0.3 0.2 78 Diseased 0.1 291 Controls 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1-Specificity FIGURE 7-4  ROC (receiver operating characteristic) analysis to compare biomark- ers for tubular necrosis mostly proximal (but sometimes not clearly localized) in 10 studies with different nephrotoxicants. The area under the curve (AUC) for a Color biomarker that perfectly predicts tubular necrosis would be 1.0. As in the experi- ment described above in Figure 7-3, 7 -4 FigureKim-1 was the best-performing biomarker (AUC = 0.95) followed closely by clusterin (AUC = 0.93). Creatinine and BUN performed much better when all of the compounds were included in the analysis rather than just cisplatin but were definitely not as good as Kim-1 and clusterin. The area under the curve for creatinine was 0.83 and for BUN was 0.81. SOURCE: Vonderscher, 2007. 

72 EMERGING SAFETY SCIENCE 1 0.9 0.8 0.7 Area Under Curve: 0.6 Sensitivity Random = 0.5 0.5 Creatinine = 0.52 0.4 Urinary Protein = 0.86 0.3 Urinary Cystatin C = 0.91 0.2 Urinary b2- Microglobulin = 0.89 0.1 41 Diseased 0 291 Controls 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1-Specificity FIGURE 7-5  ROC (receiver operating characteristic) analysis to compare bio- Figure 7 -5 markers for glomerular alteration/damage in 10 studies with different nephrotox- icants. The area under the curve for a biomarker that perfectly predicts glomerular color alteration/damage would be 1.0. In this experiment, creatinine, which had an area under the curve of 0.52, was not a good marker for glomerular alteration. However, there were several other markers that were promising; the areas under the curve for urinary proteins, urinary β2-microglobulin, and urinary cystatin C were 0.86, 0.89, and 0.91, respectively. SOURCE: Vonderscher, 2007. score for Kim-1 dropped to 0.95 when all the different compounds and lesions were included was the inclusion of one compound that caused lesions in the tubular collecting ducts, where Kim-1 is not expressed and so cannot serve as an effective marker. In a similar analysis for glomerular alteration and damage (see Figure 7-5), creatinine once again performed little better than random (area under the ROC curve of 0.52). Thus the researchers concluded that creatinine is not a good marker for glomerular alteration, but that several markers are very promising for this sort of damage. For example, urinary proteins have an area under the ROC curve of 0.86. For urinary β2-microglobulin, the area under the curve was 0.89 and for urinary cystatin C was 0.91. 

QUALIFYING BIOMARKERS 73 SUMMARY In a very narrowly defined context, there appear to be some markers that could potentially be viewed as known valid biomarkers. As noted above, however, a panel of biomarkers will likely be required to charac- terize nephrotoxicity rather than a single ideal biomarker. A panel will be necessary in particular to specify different localizations in the kidney and to differentiate among toxicity types. While Novartis and the PSTC have not yet achieved this capability, their ultimate goal is to assemble a collec- tion of kidney toxicity markers that will be visible prior to histopathologi- cal changes and can serve as a panel covering most nephrotoxicity.