Integrated analysis of toxicity data of two pharmaceutical immunosuppressants and two environmental pollutants with immunomodulating properties to improve the understanding of side effects—A toxicopathologist׳s view

European Journal of Pharmacology 759 (2015) 343–355

Contents lists available at ScienceDirect

European Journal of Pharmacology journal homepage: www.elsevier.com/locate/ejphar

Integrated analysis of toxicity data of two pharmaceutical immunosuppressants and two environmental pollutants with immunomodulating properties to improve the understanding of side effects—A toxicopathologist's view C. Frieke Kuper a,n, Jack Vogels a, Jessica Kemmerling b, Ellen Fehlert c, Christine Rühl-Fehlert d, Hans-Werner Vohr b, Cyrille Krul a a

TNO, PO Box 360, 3700 AJ Zeist, The Netherlands Bayer Pharma AG, GDD-GED-TOX-IT-Immunotoxicology, Aprather Weg, 42096 Wuppertal, Germany Department of Medicine IV, Eberhard-Karls University, Otfried-Mueller Strasse 10, 72076 Tuebingen, Germany d Bayer Pharma AG, GDD-GED-TOX-P&CP Pathology, Aprather Weg, 42096 Wuppertal, Germany b c

art ic l e i nf o

a b s t r a c t

Article history: Received 27 January 2015 Received in revised form 3 February 2015 Accepted 12 March 2015 Available online 28 March 2015

Data in a toxicity test are evaluated generally per parameter. Information on the response per animal in addition to per parameter can improve the evaluation of the results. The results from the six studies in rats, described in the paper by Kemmerling, J., Fehlert, E., Rühl-Fehlert, C., Kuper, C.F., Stropp, G., Vogels, J., Krul, C., Vohr, H.-W., 2015. The transferability from rat subacute 4-week oral toxicity study to translational research exemplified by two pharmaceutical immunosuppressants and two environmental pollutants with immunomodulating properties (In this issue), have been subjected to principal component analysis (PCA) and principal component discriminant analysis (PC-DA). The two pharmaceuticals azathioprine (AZA) and cyclosporine A (CSA) and the two environmental pollutants hexachlorobenzene (HCB) and benzo(a)pyrene (BaP) all modulate the immune system, albeit that their mode of immunomodulation is quite diverse. PCA illustrated the similarities between the two independent studies with AZA (AZA1 and AZA2) and CSA (CSA1 and CSA2). The PC-DA on data of the AZA2 study did not increase substantially the information on dose levels. In general, the no-effect levels were lower upon single parameter analysis than indicated by the distances between the dose groups in the PCA. This was mostly due to the expert judgment in the single parameter evaluation, which took into account outstanding pathology in only one or two animals. The PCA plots did not reveal sex-related differences in sensitivity, but the key pathology for males and females differed. The observed variability in some of the control groups was largely a peripheral blood effect. Most importantly, PCA analysis identified several animals outside the 95% confidence limit indicating highresponders; also low-to-non-responders were identified. The key pathology enhanced the understanding of the response of the animals to the four model compounds. & 2015 Elsevier B.V. All rights reserved.

Keywords: Principal component analysis Discriminant analysis Outliers High responders Nonresponders

1. Introduction The traditional paradigm in toxicology is focusing on single toxic endpoints and statistically significant changes in incidences or means

Abbreviations: AHR, aryl hydrocarbon receptor; AZA, azathioprine; BaP, benzo(a) pyrene; CSA, cyclosporine A; DA, discriminant analysis; F, female animal identification when followed by number; HCB, hexachlorobenzene; PCA, principal component analysis; PC, principal component; PC1, first principal component; PC2, second principal component; PC-DA, discriminant analysis with principal component analysis; LD, low dose; MD, mid dose; HD, high dose; M, male animal identification followed by number. n Corresponding author. Tel.: þ 31 88 866 1736. E-mail address: [email protected] (C.F. Kuper). http://dx.doi.org/10.1016/j.ejphar.2015.03.045 0014-2999/& 2015 Elsevier B.V. All rights reserved.

between treated groups and controls. Criticisms on current practice are, amongst others, the appropriateness of stars as signs of (biological) significance and of the frequently used statistical tests like the ANOVA (analysis of variance) in bioassays (Hothorn, 2014). Reduction and visualization of the data is a tool to improve understanding the underlying structure of the data in a ‘blind’ (unsupervised) way, i.e. calculating the best discriminating components without foreknowledge about the groups. Multivariate analysis like the principal component analysis (PCA) is such a tool. A key question when analyzing data in toxicity studies is whether the information provided by the measured biological entities (parameters or variables) is related to the experimental conditions, or to experimental bias or artefacts. For the toxicologic pathologist, the additional challenge is the discrimination between background and

344

Table 1 Parameters included in the multivariate analysis of azathioprine (AZA1 and AZA2), cyclosporine A (CSA1 and CSA2), benzo(a)pyrene (BaP) and hexachlorobenzene (HCB). Parameter

Compound mg/kg bw rat AZA2 0, 5, 12.5, 25 (Crl:WI(WU)BR)

CSA1 0, 1, 5, 25 Thomae

No. of animals/group

6 Males

10 Males

Body weight Relative organ weight: Mesenteric lymph nodes Popliteal lymph nodes Spleen Thymus Hemoglobin Red blood cell count Bone marrow cellularity Spleen cellularity White blood cell count Eosinophils Neutrophils Lymphocytes Monocytes Histopathology: Bone marrow

X X X X X X X X X X X X – Decreased cellularity

BaP 0, 37.5, 75, 150 HsdCpb:Wu

HCB 0, 3, 30, 100 Thomae

10 Males and 10 Females 8 Males

5 males

X

X

X

X

10 Males and 10 Females X

X X X X X X X

X X X

X

X X X X X X

X X X

X X

X X

X X X X X

X X X X X

X X X X X

X X X X X X

X X X X X X

– Decreased cellularity paracortex – No. of germinal centers

Peyer's patches

– Decreased cellularity

– Decreased cellularity red pulp

– Congestion ¼Decreased cellularity red pulp

X X

– Decreased cellularity

Mediastinal, mesenteric and popliteal lymph nodes

Spleen

CSA2 0, 1.25, 5, 20 (Crl:WI(WU)BR)

– Increased adipocytes

– Hypercellularity – Increased myeloid: erythroid

– Decreased cellularity paracortex

– Prominent HEVs – Increased cellularity medulla – Increased cellularity paracortex

– Decreased cellularity

– Decreased cellularity interfollicular area – Mineralization

– Prominent HEVs

– Decreased cellularity

– Decreased cellularity

– Prominent lining sinuses

– Increased cellularity follicles – Increased cellularity marginal zone

C.F. Kuper et al. / European Journal of Pharmacology 759 (2015) 343–355

AZA1 0, 5, 10, 20 HsdRCC Han:Wist

Kidneys Kidneys – Mineralization – Mineralization – Basophilic – Regenerative tubules tubules ¼regenerative Liver tubules – Vacuolation

Skin – Hyperplasia endothelium – Leukostasis – Ulcerative dermatitis – Epidermal hyperplasia – Hyperkeratosis

345

induced morphologic changes in organs and tissues (McInnes and Scudamore, 2014; Shackelford et al., 2002). In contrast to the ideal normal morphology as depicted in Handbooks of histology and anatomy, the range of ‘normal appearance’ of organs and tissues can be quite wide. Therefore, it should be established whether the spectrum of organs and tissues of all control animals or of only a majority of the controls is to be considered normal. PCA not only investigates and visualizes patterns of responses, it also provides insights into sources of normal variation and interanimal variability (Watanabe et al., 2009). This increases the understanding of the response, and may provide clues to the mode of action of the substance under study. Discriminant analysis (DA) does more or less the same, the major difference with PCA being that DA calculates the best discriminating components (¼ discriminants) for groups that are defined by the user (supervised). The major drawback in shorterterm toxicity studies is the small number of animals per treatment or exposure group. It remains to be seen how valid the conclusions are based on the multivariate analyses (Hothorn, 2014). Papers on PCA and other types of multivariate analyses in in vivo toxicology studies are limited, to our knowledge (Festing et al., 1984; Hothorn, 2014; Ikarashi et al., 1994; Keil et al., 1999; Kropf et al., 1997; Leegwater and Kuper, 1984; Strauss et al., 2009). In other related areas, it has been used as well, for example in the evaluation of immune responses to viruses (Schountz et al., 2014) and antigen selection in maturing B cells (Kaplinsky et al., 2014). Exploration of the use of multivariate analysis in immunotoxicology demonstrated that single correlations were not effective to predict in general the impact of compounds on immune functioning like resistance to infections (Keil et al., 1999). This is not only because the immune system as a target is a complex and multifaceted system, but also because the system closely communicates with the endocrine and neural system (Elenkov et al., 2000). Therefore, in the course of 28 day studies primary target cells may be obscured by secondary involvement of other cell types (Kemmerling et al., In this issue). The results from the studies described in the paper The transferability from rat subacute 4-week oral toxicity study to translational research exemplified by two pharmaceutical immunosuppressants and two environmental pollutants with immunomodulating properties (Kemmerling et al., In this issue) have been subjected to multivariate analysis, namely principal component analysis (PCA) and principal component discriminant analysis (PC-DA). To limit complexity, the evaluation was performed on a core battery of parameters in main group animals. Thus, satellite animals and immune function tests were not included in the evaluation. In a two-dimensional distribution pattern of individual parameters/variables based on the loadings of the first and second principal components, correlations between the individual parameters were investigated and characterized into composite effects or descriptors. Individual animal responses were examined against the background of these newly formed, composite effects. Although the potential contribution of multivariate analysis to the evaluation and interpretation of toxicological studies is not defined yet, information on individual responses is badly needed in the light of personalized medicine, with respect to efficacy and safety.

– – – –

Decreased cortex:medulla Reduced cortex area Indistinct CMZ border Tingible body macrophages

– Decreased cellularity cortex ¼decreased cortex: medulla – Decreased cellularity medulla – Decreased epithelial-free areas

– Increased cortex: medulla – Tingible body macrophages, cortex

– Increased cortex: medulla

– Atrophy – Starry sky appearance ¼ Tingible body macrophages

Lungs – Hyperplasia endothelium – Alveolar histiocytosis – Leukostasis

C.F. Kuper et al. / European Journal of Pharmacology 759 (2015) 343–355

2. Materials and methods

Other organs

Thymus

2.1. Toxicology data from subacute rat studies The immunomodulating drugs azathioprine (AZA) and cyclosporine A (CSA), and chemicals hexachlorobenzene (HCB) and benzo(a)pyrene (BaP) were tested in a 28-day oral rat assay, according to the OECD (1995) test guideline 407, with added parameters to specifically address the immune system. In Table 1, the data included in the multivariate analysis are described. The studies with AZA and CSA have been performed

346

C.F. Kuper et al. / European Journal of Pharmacology 759 (2015) 343–355

at two institutes, Bayer (AZA1 and CSA1) and TNO (AZA2 and CSA2). The TNO and Bayer studies differed with respect to the Wistar rat strain used and the doses (Table 1). It should be noted that key liver toxicity data of HCB and BaP was accidentally excluded from the analysis, because the focus was on immunomodulation. However in the evaluation of the PCA results it became clear that it should have been analyzed as well.

2.2. Multivariate analysis 2.2.1. Principal component analysis (PCA) PCA is a well-established statistical method for the multivariate interpretation of complex data (Massart et al., 1997; Vandeginste et al., 1998). The linear combinations of the original parameters from the six studies lead to other, constructed, parameters with improved projection properties. The improved parameters are described as a linear combination of scores and loadings and are

X No. of Samples (n)

Data

T B

T =

scores

*

+ E

Loadings

m n

No of measurements (m) Fig. 1. Scheme of a mathematical description of the principal component analysis (PCA). PCA leads to improved parameters (linear combination of scores and loadings), which are called principal components or PCs. The PCA is described as X ¼TBT þ E, where X is the original nnm set of data, BT is a transpose Fnm matrix of variable coefficients (loadings), T is an nnF matrix of object scores and E is an nnm matrix containing the residuals not explained by the model using F PCs.

called principal components (PCs) (Fig. 1). The scores and loadings vectors give a concise and simplified description of the variance present in the data. The first PC explains the highest variance in the predicting data set, and monotonically decreasing variance is obtained for the remaining PCs. The number of PCs gives an indication about the model complexity. If the data are highly correlated, a few PCs will be sufficient to reproduce the original data. In the six studies with the four model compounds, two PCs were sufficient. Sometimes PCA may not be able to extract relevant information and may therefore provide meaningless principal components that do not describe experimental characteristics. The reason is that its linear transformation involves second order statistics (i.e. to obtain mutually nonorthogonal PCs) that might not be appropriate for biological data. A plot of score vectors (score plot) indicates clustering and trends present in the profiles. Scores of similar samples will tend to form clusters whereas more dissimilar samples will be found at greater mutual distances. Equally a plot of the loadings (loadings plot) can indicate an interpretation of the trends present. Plotting the scores and loadings together in one ‘biplot’ allows a direct graphical interpretation of the data where trends in the scores can directly be correlated to trends in the loadings and thus in the underlying ‘measured’ parameters (Greenacre, 2010). Key pathology was established by grouping highly correlating parameters with a loading o  2 or 42. 2.2.2. Principal component-discriminant analysis (PC-DA) Discriminant analysis (DA) is applied if the interest is focused on differences between groups of samples (Massart et al., 1997; Vandeginste et al., 1998). The technique is based on the

Fig. 2. Principal component analysis (PCA) of the studies with the 4 model compounds in male rats: 2 independent studies on azathioprine (AZA1 and AZA2), 2 on cyclosporine A (CSA1 and CSA2), 1 on hexachlorobenzene (HCB) and 1 on benzo(A)pyrene (BPA). The red circles depict the 95% confidence borders. Animals outside this border may show exaggerated responses (treated animals) or point to significant variation in the control group. X-axis (first principal component or PC1) and y-axis (second principal component or PC2), in brackets percentage of variance explained by the PC. Blue is control animals (1); green is low dose (2); red is mid dose (3); and turquoise is high dose (4). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

C.F. Kuper et al. / European Journal of Pharmacology 759 (2015) 343–355

AZA2

5 4

M26

3

PC2(10.60)

M74 M82

2 1 0 -1 -2 -3 -4 -10

-5

0

5

10

15

PC1(23.49)

AZA2 discriminant analysis 0.3 0.2 0.1

D2(28.34)

assumption that samples of the same group are more similar compared to samples of other groups. The goal of DA is to find and identify structures in the original data, which show large differences in the group means. This process involves a priori knowledge of which samples are similar. Therefore, DA is regarded to be a supervised analysis technique. This distinguishes it from principal component analysis (PCA), which does not require a priori knowledge about the samples. The first step in DA is to combine the original parameters into a set of mutually independent new parameters in such a way that the projection of the original samples in the space, spanned by a minimum number of these new variables, maximizes the difference between the group means. DA describes most efficiently the differences between groups of samples. However, the number of variables is often large compared to the number of samples. This may lead to degenerate solutions. The general rule of thumb is that the number of samples should be at least four times the number of variables. This rule will lead to problems in holistic experiments where the number of measurements is often large compared to the number of subjects (animals). However, there is a solution to this problem. Hoogerbrugge et al. (1983) developed a scheme in which the number of variables is reduced by PCA, firstly, followed by DA on the scores of the samples on the first PC axes. This technique is called principal componentdiscriminant analysis (PC-DA). Determining the exact number of PCs to include is difficult. The number should not be too small because including only the first few can result in a loss of the between-group information. The number should not be too large also, because it will exceed the number-of-samples-divided-by-four rule. Therefore, all PCs, which explained a significant amount of variance up to a maximum of the number of samples divided by four, were included.

347

0 -0.1 -0.2

3. Results 3.1. PCA plots of the 6 studies in male rats Table 1 presents an overview of the 6 studies with their investigated parameters. Fig. 2 presents the PCA male animal plots of the six studies, which are listed in Table 1. The males are plotted individually, and the color of the symbols marks the different treatment groups. The colored lines, which delineate the different treatment groups, have been drawn manually. 3.1.1. AZA1–AZA2 and CSA1 and CSA2 comparison The similarities in the AZA studies of the two labs were less than in the CSA studies (described in detail by Kemmerling et al. (In this issue)). This is in accordance with overall findings in the ring studies with the two compounds; The ICISIS Group Investigators (1998) ascribed the relatively low concordance with AZA to the lack in harmonization of the procedures. Based on the experience with AZA, harmonization was much stricter in the CSA studies. The PCA plots show with both AZA1 and AZA2 a separation between the controls and low-dose group with the mid- and high-dose groups (Fig. 2). However, the separation is less in the AZA2 study; the overlap between the mid- and high-dose groups is much bigger; the direction in which the groups are separated differs slightly (AZA1 showing an upwards curve and AZA2 a downwards curve); and the variability between the animals of the high-dose groups is much more prominent. In the AZA2 study three animals are outside the 95% confidence border and could be considered ‘outliers’ (see also Section 3.2.2). Exclusion of the outliers does not lead to a difference in separation of the groups (Fig. 3A). Supervised analysis (PC-DA) improves the separation marginally (Fig. 3B). The multivariate analysis plots illustrate the conclusions by Kemmerling et al. (In this issue) that there are inter-study and inter-animal differences in the response to AZA.

-0.3 -0.4 -0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

D1(56.22) Fig. 3. (A) PCA plot of AZA2 males from which the three high-dose outliers were removed. Later on the outliers were projected into the new PCA plot. The positions of the groups do not change much by leaving the outliers out. Dark blue are the outliers (0); green is controls (2); red is low dose (3); turquoise is mid dose (4); purple is high dose (5). (B) Processing of the data (supervised data; principal component discriminant analysis) can lead to increased separation of the groups. Also the individual responsiveness or non-responsiveness is more apparent. Blue is control (1); green is low dose (2); red is mid dose (3); turquoise is high dose (4). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Both the inter-study as well as the inter-animal differences may be due to differences in metabolizing enzymes (both labs used different Wistar rat strains), and be related only in part to lack of harmonization of procedures. The PCA plots of both the CSA1 and CSA2 study illustrated that the high-dose groups were distinctly separated from the other three groups including the controls (Fig. 2). Also the direction in which the high-dose group was separated from the other groups was identical. In the CSA2 study, the mid-dose group was slightly more separated from the controls and low-dose group than in the CSA1 study. Overall, the plots are in agreement with the conclusion that results with CSA in the studies performed at TNO and Bayer were strikingly similar (Kemmerling et al., In this issue).

3.1.2. PCA analysis compared to variable per variable (conventional) evaluation As stated above, the PCA plots of the AZA studies showed a separation between the controls and low-dose group with the mid-

348

C.F. Kuper et al. / European Journal of Pharmacology 759 (2015) 343–355

Table 2 Descriptors as provided by the principal component analysis of azathioprine (AZA1 and AZA2), cyclosporine A (CSA1 and CSA2), benzo(a)pyrene (BaP) and hexachlorobenzene (HCB). Male studies

Doses (mg/kg)

AZA1

5/10/20

AZA2

CSA1

CSA2

BaP

HCB

5/12.5/25

1/5/25

1.25/5/20

PC1/PC2

27.97/ 13.53

37.61/ 9.56

30.97/ 10.58

30.99/ 13.61

37.5/75/150 52.21/ 10.20

3/30/100

32.85/ 11.53

Variation explained by PC1 and PC2 [%]

PCA males: grouping of parameters/variables into descriptors

Characterization of the dose groups by the effect (mg/kg)

41.50

Thymus lymphocytes effect relative thymus and spleen weight; thymus C:M ratio; thymus cortex size; spleen cellularity; peripheral blood leukocytes and lymphocytes Toxicity effect body weight; dilated sinuses femur bone marrow; congestion spleen Prevaling lymphoid organ effect relative thymus weight; bone marrow cellularity; thymus cellularity cortex; thymus epithelial-free areas; spleen cellularity; mesenteric lymph node erythrophagocytosis; peripheral blood leukocytes and lymphocytes; hemoglobin; erythrocytes Toxicity and subordinate immune effect body weight; relative spleen weight, mediastinal lymph nodes no. of germinal centers; peripheral blood monocytes Toxicity and immune effect relative kidney weight; kidney regenerative tubules and mineralization; liver vacuolisation; thymus C:M ratio; Spleen PALS and follicles cellularity; lymph node cellularity; Peyer's patches cellularity Peripheral blood effect neutrophils; lymphocytes; erythrocytes

10 (partly) and 20

47.17

41.55

44.60

62.41

44.38

Female studies CSA1

HCB

HCB

Toxicity and prevaling immune effect body weight; relative kidney weight; kidney regenerative tubules and mineralization; thymus C:M ratio; spleen PALS cellularity; lymph node cellularity; Peyer's patches interfollicular area cellularity; peripheral blood leukocytes and lymphocytes Red blood cell and subordinate immune effect [hemoglobin; peripheral blood erythrocytes; relative spleen and lymph node weights; bone marrow cellularity; Peyer's patches mineralization] Stress-related immune effect body weight; relative thymus weight; bone marrow sternum and femur increased adipocytes; thymus atrophy; thymus tingible body macrophages; spleen prominent lining sinuses; peripheral blood total leukocytes, lymphocytes, erythrocytes; hemoglobin Peripheral blood effect neutrophils and monocytes Immune activation and endothelium relative spleen weight; spleen cellularity marginal zone; mesenteric lymph node cellularity medulla; mesenteric and popliteal lymph node and Peyer's patches prominent HEVs; lungs hyperplasia endothelium; lungs leukostasis; lungs alveolar histiocytosis Peripheral blood and immune effect peripheral blood leukocytes, all types; popliteal lymph node paracortex cellularity

10 (mostly) and 20 12.5 (mostly) and 25

12.5 (mostly) and 25

25

Explains variability in controls and 25 (divided into 2 subgroups) 20

5 (partly) and divides 20 into 2 subgroups 37.5 and 75 and 150

37.5 and 150 variability in controls 30 and 100

variability in 30 and 100

PCA females: grouping of parameters/variables into descriptors 1/5/25

3/30/100

3/30/100

26.53/ 10.53

44.93/ 10.57

34.24/ 14.01

37.06

55.50

48.34

Toxicity and prevaling immune effect kidney regenerative tubules and mineralization; thymus C:M ratio; thymus cortex and medulla cellularity; spleen cellularity follicles and PALS; spleen germinal centers; Peyers patches cellularity Peripheral blood and subordinate immune effect peripheral blood lymphocytes and erythrocytes; hemoglobin; terminal body weight; bone marrow cellularity; relative thymus and mesenteric lymph node weights Prevaling immune activation and endothelium effect relative thymus and spleen weight; bone marrow sternum cellularity; spleen total cell numbers; spleen cellularity marginal zone; mesenteric lymph node cellularity medulla; mesenteric and popliteal lymph node HEV; lungs hyperplasia endothelium; peripheral blood erythrocytes; hemoglobin Subordinate immune activation effect mesenteric and popliteal lymph node cellularity paracortex; lungs alveolar histiocytosis Skin pathology effect skin hyperplasia endothelium, ulcerative dermatitis, leukostasis, epidermal hyperplasia, hyperkeratosis Males and females combined Females: Immune-related inflammation effect relative popliteal lymph node weight; skin ulceration; skin epidermal hyperplasia; lungs endothelial hyperplasia Males: Toxicity and peripheral blood effect terminal body weight, peripheral blood erythrocytes, neutrophils, monocytes, total leukocytes and lymphocytes

25

5 (small part), divides 25 into 2 subgroups

30 (partly) and 100

30 and 100 (divided into 2 subgroups) 100 partly

All doses

All doses

C.F. Kuper et al. / European Journal of Pharmacology 759 (2015) 343–355

AZA1 male azathioprine

349

BaP male benzo(a)pyrene

AZA2 male azathioprine

5 2

3

3

1

2

2

PC2(9.56)

PC2(13.53)

1 0 -1

PC2(10.20)

4

4

1 0

-2 -3

-1

-4

-3

-2

-5

-4 -4

-3

-2

-3

-2

-1

0

1

2

3

-6 -3

4

-2

PC1(27.97)

-1

0

1

2

3

-3

4

1

PC1(30.97)

2

3

4

2

3

3

PC2(11.53)

PC2(13.61) 0

1

4

2 1 0 -1

2 1 0 -1

-2

-2

-3

-3

-4 -1

0

HCB male hexachlorobenzene

5

3

-2

-1

PC1(52.21)

CSA2 male cyclosporine

5 4 3 2 1 0 -1 -2 -3 -4 -5 -3

-2

PC1(37.61)

CSA1 male cyclosporine

PC2(10.58)

0 -1

-4 -3

-2

-1

0

1

2

3

4

PC1(30.99)

-3

-2.5

-2

-1.5

-1

-0.5

0

PC1(32.85)

Fig. 4. Descriptors of the animal responses against the four model compounds. The parameters/variables are depicted as lines. The direction and the length of the line determine the contribution of that particular parameter/variable to the animal plots.

and high-dose groups (Fig. 2). This is comparable with the single parameter evaluation (Kemmerling et al., In this issue; Schulte et al., 2002; The ICISIS Group Investigators, 1998) although it was questioned if the low-dose was indeed a no-effect level. Immune function tests in satellite group animals also questioned the low-dose as a no-effectlevel. It remains to be investigated if both new factors (first principal component PC1 and second principal component PC2) together sufficiently explained the variability (AZA1 41.50% and AZA2 47.17% of the variation) to fully understand the response (Table 2). The plots of the animals treated with CSA showed a separation between the high-dose group with the controls, low- and mid-dose groups, and the mid-dose group tended to be separate from controls and mid dose group. The single parameter evaluation demonstrated that the low-dose was the no-observed effect level, whereas the immune function tests in satellite group animals even had a no-effectlevel below the low-dose. PC1 and PC2 together accounted for about 42% of the variability (Table 2). The plots of the animals exposed to the low-, mid- and high-dose of BaP were clearly separate from the controls. The single parameter analysis could not convincingly establish the low-dose as a no-effectlevel and the immune function tests clearly showed that the no-effectlevel was below the low-dose. PC1 and PC2 together accounted for about 60% of the variability (Table 2). The plots of the animals exposed to the mid- and high-dose of HCB were separate from the control and low-dose groups. The single parameter evaluation pointed to a noeffect-level below the low-dose in the study. PC1 and PC2 together accounted for about 44% of the variability. In summary, the single parameter analysis was slightly more sensitive with respect to the dose–response, meaning that the established no-effect levels were lower than the distances between the plotted dose and control groups in the PCA suggested. This is not remarkable, because the single parameter evaluation uses expert judgment and can contribute a high value to pathology occurring in

only one or two animals, as long as that particular pathology is remarkable. 3.1.3. Grouping of parameters/variables: description of the response against the four model compounds The parameters/variables that correlated most with each other (loaded most on the X- and Y-axes, namely o  2 or 42) were identified (Fig. 4; Table 2). These groups of parameters (composite parameters or descriptors) were given effect names to help characterize the responses of the groups of animals upon treatment/exposure (presented in the text in italic). Within these effects some parameters contributed more than others, but this was not taken into account in the name of the effect. For instance, males treated with the high-dose of CSA (CSA1) were especially characterized by kidney and liver toxicity, and females treated with CSA (CSA1) more by thymus and spleen pathology, yet both were characterized by the Toxicity and immune effect (Table 2). One could also describe the effect in males as Toxicity and immune effect and in females as Immune and toxicity effect. Interestingly, in males the second effect did not include an immune parameter, hence the main effect was denoted Toxicity and immune effect. In the CSA2 study, both the main and the secondary effect included immune parameters, like the females in the CSA1 study, thus the main effect in both the male CSA2 and the female CSA1 was described as Toxicity and prevalent immune effect and the second effect as Toxicity and subordinate immune effect. 3.2. Animal responses 3.2.1. Sex differences PCA analysis of the two studies with males and females, namely one study with CSA (CSA1) and the one with HCB did not support the finding that there were distinct gender differences in responses, males

350

C.F. Kuper et al. / European Journal of Pharmacology 759 (2015) 343–355

being more sensitive to CSA and females more sensitive to HCB (Kemmerling et al., In this issue; Fig. 5A). Upon single parameter/variable evaluation, none of the CSA-treated animals responded to the low-dose, but at the mid-dose the incidence of increases in thymus cortex–medullary ratio (the most noticeable immunotoxicologic pathology effect) was higher in males than in females (10 out of 10 males and 4 out of 10 females) (Kemmerling et al., In this issue, Table 2). Of the 10 affected males, 2 had a moderate increase whereas all four females showed only a slight increase. In addition, 1 male showed a slightly decreased cellularity in the periarteriolar lymphoid sheets of the spleen, but none of the females had an affected spleen. All high-dose animals were affected, but males more severely than females. Male body weight was also affected, at the mid- and high-dose, and relative thymus weight was decreased at the high-dose. Females were practically unaffected in body and thymus weights. Both sexes were about equally affected in immune functions tests (performed in satellite groups of animals). Finally, the high-dose males were more severely affected by kidney and liver histopathology than the high-dose females. The PCA plots of the CSA-treated animals showed a clear separation between the high-dose and the other three groups of both males and females, but the overlap was large between the mid-dose and the controls and low-dose for both sexes (Figs. 2 and 5A). The parameters that determined the pattern of the plots of the males were more inclined to Toxicity (PC1: terminal body weight, kidney regenerative tubules and mineralization, and liver vacuolization; PC2: neutrophils), whereas the parameters for the females were more towards Immune effect (PC1: thymus cortex:medulla ration and tingible body macrophages; spleen germinal centers; possibly also PC2: bone marrow cellularity). Thus, the PCA plots did not reveal distinct differences in grouping between males and females, but the descriptors (composite parameters) differed. Females were affected more than males by HCB, based on single parameter evaluation. The PCA plots revealed that the direction in which the dose groups were plotted differed between males and females, the direction being linear in males and curved in females. The curved shape in the females was largely due to distinct skin pathology in part of the high-dose females. Females exhibited an Immune-related inflammation effect (relative popliteal lymph node weight; skin ulceration; skin epidermal hyperplasia lungs endothelial hyperplasia) and males a Toxicity and peripheral blood effect (terminal body weight, peripheral blood erythrocytes, neutrophils, monocytes, total leukocytes and lymphocytes), when the sexes were analyzed together (Fig. 5B). The higher sensitivity of females appeared related to a more distinct effect on the primary lymphoid organs. The variability between control, low- and mid-dose females was also low, thereby contributing to distinct separation between mid- and high-dose groups with controls and low-dose groups.

3.2.2. Biological outliers and statistical outliers: variability in controls and aberrant responders Toxicologists and toxicological pathologists may use the description of outliers as provided by Frame et al. (2014). They stated that outliers are extreme deviations in an individual finding from the group norm, as well as from historical (control) values. In assessing test article-related versus chance effects, ‘outlier’ also assumes the deviation from norm is not due to the test article. In statistics, an outlier is an observation point that is abnormally distant from other observations (Grubbs, 1969). This definition leaves room to decide what is considered normal and abnormal. Outliers may include the sample maximum or sample minimum, depending on whether they are extremely high or low. In large samples or small samples with scores of numerous parameters, some accidental outliers are to be expected. A frequent cause of outliers is a mixture of two distributions, which may be two distinct sub-populations.

Outlier points can indicate faulty data, erroneous procedures, or areas where a certain theory might not be valid. An outlier may be due to variability in the measurement or it may point to measurement error. Frame et al. (2014) recommend to remove outliers from evaluation and analysis, in case they are due to technical errors or occur secondary to disease states unrelated to test article administration and as such do not reflect a group effect of the test article. Statistical outliers may represent low-incidence occurrences of compoundrelated effects; therefore a Weight of Evidence approach must be taken into account when deciding to include or exclude an outlier. Limited variation was observed in the control groups of the HCB males and females, AZA2 males, CSA2 males and CSA1 females. A few control animals of the BaP males, CSA1 males and AZA1 males fell outside the 95% confidentiality interval (Fig. 2 for the animal plots outside the red circle representing the 95% confidentiality limit; Fig. 3 for the parameters). These were one BaP-exposed male (M3—characterized by remarkable peripheral blood total leukocyte number; neutrophils, eosinophils; monocytes; lymphocytes; basophils), two CSA1 males (M3—characterized by remarkable peripheral blood erythrocytes; lymphocytes; neutrophils; total leukocytes and M4—erythrocytes; lymphocytes; neutrophils; relative liver and kidney weight) and one AZA1 male (M5—characterized by remarkable peripheral blood leukocytes, lymphocytes, monocytes, basophils, neutrophils). Just within the 95% confidentiality range was the AZA1 male M3 (remarkable erythrocytes, hemoglobin, spleen congestion). In conclusion, the variability (presence of outliers) in the control groups of some studies depended largely on peripheral blood parameters/variables. All studies, with the exception of the males treated with AZA in the AZA1 study and the males treated with BaP, showed single animals outside the 95% confidentiality interval/limit. These so-called outliers are described below, with their distinctive parameters (single parameter/ variable evaluation) between brackets. In the AZA2 study the outlier animals were three high-dose males, mostly due to a marked response within the Primary lymphoid organ effect (Table 2) (M26—characterized by remarkable hemoglobin; erythrocytes; relative spleen and mesenteric lymph node weights; total number of blood leukocytes and lymphocytes, 74 and 82—both characterized by thymus and mesenteric lymph node cellularity, thymus EFA's; relative thymus weight; hemoglobin; erythrocytes; bone marrow cellularity; mesenteric erythrophagocytosis; total blood leukocytes and lymphocytes). The location of these three outliers in the animal plot figure, when compared to the other high-dose animals (also compared to all other animals in the study), was due especially to the effects on hemoglobin and erythrocytes; the severity of the thymus and bone marrow effects contributed to their position outside the 95% limit. The mid-dose animal M114 tended to stand out, with low numbers of blood monocytes and eosinophils, decreased cellularity bone marrow, decreased number of thymic epithelial free areas and low number of germinal centers in the lymph nodes. One of the high-dose males (M122) and three of the mid-dose males (M32, M36 and M66) were located within the groups of the controls and low-dose animals (M122). Indeed the high-dose male and two of the mid-dose males did not show an effect on any of the parameters/variables tested and were thus considered to be non-responders. The remaining mid-dose male (M66) demonstrated responses comparable to control with the exception of decreased cellularity in the red pulp of the spleen. Decreased cellularity in the splenic red was considered an AZA-induced effect, seen in 6 out of 10 high-dose males. In the CSA1 study there were two high-dose males, one mid-dose female and one high-dose female outside the 95% confidence limit (high-dose males M36 and M38), (mid dose female F65—characterized by hemoglobin; erythrocyte number; spleen cellularity; blood neutrophils and lymphocytes) (high-dose female F73—characterized by blood total leukocyte, neutrophils and lymphocyte numbers). Interestingly, the males scored consistent but not exceptional (except spleen and lymph node weight of M38) on the Toxicity and immune effect (see Table 2). The difference between both males depended on

C.F. Kuper et al. / European Journal of Pharmacology 759 (2015) 343–355

351

HCB male + female hexachlorobenzene 6

Females Females

4

Females

PC2(14.10)

2

Females

0

-2

Males Males

-4

-6 -10

Males

-8

-6

-4

-2

Males 0

2

4

6

8

10

PC1(34.24) Fig. 5. (A) PCA plots of the females, exposed to CSA or HCB. The descriptors are depicted at the right. Blue is control animals (1); green is low dose (2); red is mid dose (3); and turquoise is high dose (4). (B) PCA plots of HCB-exposed males and females, analyzed together. Blue males is control males (1); green is low dose males (2); red is mid dose males (3); turquoise is high dose males (4); purple is control females (5); beige is low dose females (6); gray is mid dose females (7) and blue females is high dose females (8). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

352

C.F. Kuper et al. / European Journal of Pharmacology 759 (2015) 343–355

4. Discussion

CSA2 males

6

PC2(13.61)

4

2

0

Potential ‘Outliers’

-2

-4

-6 -10

-8

-6

-4

-2

0

2

4

6

8

10

PC1(30.99) Fig. 6. Identification of potential outliers, exemplified by high-responders to the mid and the high dose of cyclosporine A (CSA). The dotted and red circles show each a different level of confidence. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

the hemoglobin and erythrocyte parameters/variables, thus related to the Peripheral blood effect. In the CSA2 study two high-dose males and one mid-dose male were outside the limit (high-dose M94 and M186—both characterized by remarkable body weight and distinct kidneys mineralization and regenerative tubules; M94 in addition by relative thymus weight; cellularity in the popliteal and mediastinal lymph nodes, Peyer's patches and spleen periarteriolar lymphoid sheets PALS; Fig. 6). M186 was characterized in addition by hemoglobin and erythrocytes; cellularity in bone marrow and mesenteric lymph node paracortex; mineralization in Peyer's patches (mid-dose M168—characterized by remarkable body weight; relative spleen weight; cellularity in the mesenteric lymph node paracortex). Both high-dose males scored high on the Toxicity and primary immune effect, and M186 scored also high on the Red blood cell and secondary immune effect. The mid-dose male scored on Toxicity and primary immune effect albeit without the toxicity effects. Finally, in the HCB study two high-dose males (M38 and M39) and two high-dose females (F72 and F77) were outside the limit when the sexes were analyzed separately. Evaluation of the individual parameters/variables showed that M38 stood out in the cellularity of the popliteal lymph node paracortex; M39 scored high in relative spleen weight and in spleen cellularity, number of blood neutrophils, and lung hyperplasia endothelium. Both animals scored overall high in lymphoid organs response to HCB, the Primary immune activation effect. The two males were widely apart in the animal plot, due to the Peripheral blood and immune effect (high number of neutrophils in M39 and the increased paracortex cellularity in M38). The two highdose females could not be examined in full (no body and organ weights, no blood parameters), but scored high on the Skin pathology effect and the Primary immune activation effect [notably high cellularity of the popliteal lymph node medulla and high myeloid:erythroid ratio in the bone marrow of femur and sternum (F77)]. When the sexes were analyzed together, one of the males (M38) no longer fell outside the limit. Instead the male tended to be located towards the female high- and mid-dose groups. This may be related to the relatively low scores on most body and organ weights and blood parameters. In summary, the PCA analysis identified several animals outside the 95% confidence limit. Most of these animals were indeed highresponders. In addition, low-to-non-responders could be identified. A single outlier animal scored especially high on peripheral blood parameters, which may show high variability also in controls (see above), and thus a relationship with the treatment/exposure is questionable.

Toxicology research aims to identify and explore mechanisms of responses to xenobiotics and to predict human responses, with regard to effect and dose, and preferentially also sensitive subpopulations (the individual response). In this study we are interested in the application of approaches to reveal insightful patterns of toxicology and toxicologic pathology induced by four model compounds while reducing the dimension in the data. The four model compounds were two immune-suppressing drugs and two environmental pollutants with immunomodulating properties. All four have been shown to affect the immune functioning in rodents and a potential to exert similar effects in man. However, the evidence of immune modulation of the environmental pollutants is less compared to the pharmaceutical compounds (Kemmerling et al., In this issue). The approaches, used to reduce the dimension in the data, are the principal component analysis (PCA) and the discriminant analysis, combined with the PCA (PC-DA). The PCA is an unsupervised analysis, meaning without previous insight into the data (in a way comparable to ‘blind scoring’ of tissue slides by a toxicologic pathologist). In contrast, DA is a supervised analysis. The DA was only used once, to explore if it added sufficiently to the evaluation. Because it did not add much, it was decided to restrict further evaluation of the data to PCA. Model predictions from multivariate analysis should ideally be tested in an independent experimental data set. We found good agreement between our model predictions and measured toxicity endpoints based on a comparison of the two independent studies with AZA and CSA. The analyzed data were restricted to the results in the main groups, the data of the satellite groups were not included. PCA assumes that all parameters have been measured in each animal, but this is unfeasible with parameters ranging from histopathologic endpoints to immune functioning. Keil et al. (1999) demonstrated that multivariate analysis could be performed with animals from the main and satellite groups together in case of dexamethasone but not with cyclosporine A, but consistent criteria for PCA use could not be given on the basis of only two compounds. Even with only the main groups, it remains a challenge to find the appropriate multivariate analysis for the small sample size design, the high dimension and the different scales (continuous, ordinary, binary) in complex layouts (Hothorn, 2014). PCA is particularly powerful if the toxicology and toxicologic pathology question is related to the highest variance, but can fail to be biologically and toxicologically meaningful when parameters do not follow a multivariate normal distribution or when the biological question may not be related to the highest variance in the data (Yao et al., 2012). PCA projects the data into a new space spanned by the principal components (PC). The PCs can successfully extract relevant information in the data and reveal experimental characteristics, as well as artefacts or bias. PCA is often used as a pre-processing step for subsequent analyses, so what have we learned from clustering of parameters? First of all it should be stressed, probably superfluously, that input is output. This applies especially to the selection of parameters included in the analysis (Table 1). The analysis of the AZA data were restricted mainly to effects on immune-related parameters, because other effects were not found in the rats (in man: liver, pancreas, intestines and flu-like symptoms including skin effects; Teml et al., 2007; Vermeire and Rutgeerts, 2005). The analysis of the CSA data included effects on kidneys, a major target in humans as well (Pallet and Legendre, 2010). The analysis of both environmental pollutants, BaP and HCB considered only parameters connected to the immune system. The compounds are known to influence besides the immune system a variety of cells and organ systems mainly by interaction with the ligand-activated transcription factor aryl hydrocarbon receptor (AHR) (Randi et al., 2008; Schraplau et al., 2015; Sparfel et al., 2010; Van Grevenynghe et al., 2005). The

C.F. Kuper et al. / European Journal of Pharmacology 759 (2015) 343–355

AHR can act variably on the immune system after binding with xenobiotics: on one side as a sensor with protective properties against deleterious changes such as inflammation but is on the other side in case of prolonged exposure associated with pathologic consequences like autoimmunity and tumor proliferation (Julliard et al., 2014). Rodent studies indicated that BaP causes its immunotoxic effects by upregulation of cytochrome P4501A1 (CYP1A1), which is regulated via the AHR, and P4501B1 (CYP1B1) (Carlson et al., 2004; Uno et al., 2006). For humans, the same CYP enzymes are responsible for this drug-metabolism (Martignoni et al., 2006). HCB is a weak agonist of the AHR and the predication that toxicity is mediated by AHR is restricted and inconclusive (Randi et al., 2008). The liver has been shown to be a main target organ of oral BAP in rats and mice (Uno et al., 2006). Likewise, for HCB the liver is a key target organ in rats, monkeys and humans (Giribaldi et al., 2011; Iatropoulos et al., 1976; U.S. Department of Health and Human Services, 2013). The selection of parameters in the present investigation will have influenced the pattern of responses and should be taken into account in the evaluation (key parameters or ‘effects’; see below). The value of the input also applies to inter-study comparisons, whereby one group of pathologists can record every small change whereas others restrict themselves to record only distinct changes thus leading to variations in threshold (Long and Hardisty, 2012). This type of difference was not encountered in the studies used here, in which strict diagnosis criteria were used. Secondly, random-bred rat strains are genetically heterogeneous and there is overwhelming evidence that responses to many substances are under genetic control, especially against immunotoxicants when the hematopoietic system is target (Festing et al., 2001; Palm et al., 2011). Specifically, toxicogenomic investigations showed for all compounds under investigation that major gene-regulated pathways were influenced in rodent tissues as by HCB in several organs of rats (Ezendam et al., 2004), by CSA and BAP in the mouse spleen (Baken et al., 2008) and by AZA in rat hepatocytes (Cho et al., 2014). The two studies with AZA and especially with CSA demonstrated marked similarities despite the use of different (sub) strains of rats. The genetic heterogeneity may explain the variability in peripheral blood parameters in some control groups, and in some test groups, as was evident in the second descriptor (PC2). Peripheral blood effects in rats may be a very sensitive marker for effects of substances or individual sensitivity. It has to be taken into account when investigating the immune system, that hematology and in particular white blood cell counts have shown considerable variation depending on age, sex, or genetic background in rats (Petterino and Argentino-Storino, 2006) and mice (Serfilippi et al., 2003). Likewise, in human populations from different countries variability of laboratory values was observed (MiriDashe et al., 2014; Zeh et al., 2011). In studies in human volunteers, unwanted variations can be reduced by setting of appropriate selection criteria (Sibille, 1990; Sibille et al., 1999). Depending on the disease and parameters studied, interindividual variation in controls can be larger than in patients (Wagener et al., 2013). Thirdly, the two grouped key parameters or ‘effects’ (Table 2) illustrate that the AZA response is described best as an immune effect. The AZA2 mid-dose male (M66) demonstrated responses comparable to control with the exception of decreased cellularity in the red pulp of the spleen. Decreased cellularity in the splenic red was considered an AZA-induced effect, seen in 6 out of 10 high-dose males. It remains to be determined whether this effect must be considered as a decisive part in the AZA-response; it is not included in the effects formulated on the basis of parameter/ variable loadings (Table 2). CSA and HCB were characterized by immune parameters as well as by toxicity to non-lymphoid organs. BaP is characterized by immune parameters in combination with changes in body weight, which may have influenced the immune parameters. In addition, the immune effects fit those related with glucocorticoid-dependent stress effect. The main

353

effect was therefore denoted Stress-related immune effect. In rats, BaP has been found to interfere with hormonal imprinting upon perinatal exposure, especially in relation to glucocorticoids (Csaba et al., 1991). It should be noted that for both environmental pollutants HCB and BaP the pattern of response may have been different when liver toxicity had been included (see above). The common practice in toxicology to rely almost solely on the mean response may be misleading. Outliers may be indicative of data points that belong to a different population than the rest of the sample set. The PCA plots of individual animals with 95% confidence limit are a way to identify ‘aberrant’ or ‘outlier’ responders. Outliers can be deleted as standard practice, but this practice is questionable. Although mathematical criteria can provide an objective and quantitative method for data rejection, these criteria may not be scientifically sound, especially in the case of small sets or when a normal distribution cannot be assumed. Rejection of outliers can be acceptable in the case of experimental error, but this should be verified for every case. In large sample sizes, ‘spontaneously’ occurring outliers are expected. Collection of historical data including spontaneous incidences of outliers within laboratory specific data is routinely done in toxicology (examples: Hayakawa et al., 2013; Lee et al., 2012; Okamura et al., 2011). In addition, the incidence of outliers in clinical chemistry or hematology and unusual histopathologic findings in controls is surveyed in larger databases. Laboratory-specific and inter-laboratory data collections aid in the interpretation of findings seen in dosed animals (Deschl et al., 2002; Elmore and Peddada, 2009; Hall and Everds, 2003; Kobayashi et al., 2010). For example, in normally distributed data and a large sample size, roughly 1 in 22 observations will differ by twice the standard deviation or more from the mean. The designated outliers should be examined carefully to establish that they really are over- or under-responding to the treatment or exposure, or if they show some unrelated, deviant pathology or experimental bias or artefact. It is also the other way around, animals may score exceptionally high on one particular treatment/exposure-related parameter, without being located outside the 95% confidence limit. Therefore, PCA cannot replace the single parameter evaluation, but is an additional tool to understand the response. The group response in PCA may change due to the presence of outliers, but this was not demonstrated here as elimination of outliers from the AZA2 study did not change the overall picture. Overall outliers or aberrant responders may represent low-incidence occurrences of compound-related effects; therefore a Weight of Evidence approach must be taken into account when deciding to include or exclude an outlier. Based on human data and mode of action (AZA, CSA and HCB) or mode of action alone (BaP) the highest variation in response was expected in the AZA and BaP groups. In contrast to this expectation, outstanding responders were observed with AZA2, CSA1 and CSA2, and with HCB, in the PCA analysis with a 95% confidence limit. This may point to differences between rat and man or to differences between rat strains, because the variation in AZA1 was low whereas the variation in AZA2 was high. Also, it may be a reason to reconsider the expectation or assumption and look again at both the experimental and human data, as PCA and other multivariate analyses are a tool to direct attention to unexpected or overlooked results and correlations.

5. Summary and conclusions In summary, application of PCA enabled the identification of highly correlated parameters and key pathology for further investigation, which will contribute to the mode of action of compounds. This is exemplified by BaP. It also revealed potential biomarkers, namely peripheral blood parameters for individual sensitivity of

354

C.F. Kuper et al. / European Journal of Pharmacology 759 (2015) 343–355

(control) rats. Finally high- and non-to-low-responders could be identified which can be of interest to the human situation. Individual patients showed different responses to drugs, which triggered investigations of pharmacogenetics as well as considerations of individualized immunosuppression in transplant patients to supplement therapeutic drug monitoring and to find an optimal strategy to minimize the risks of either rejection or toxicity (Abboudi and Macphee, 2012; Ekbal et al., 2008). Investigations in this direction are promising but challenging and have been considered for both calcineurin inhibitors (Elens et al., 2014) and thiopurines (Duley et al., 2012). Hence, a different view on preclinical animal studies might provide additional information regarding biomarker selection. It is acknowledged that there is room to improve unsupervised (‘blind’) and supervised multivariate analyses like the PCA and PC-DA for toxicology and toxicologic pathology, especially with respect to the relatively small sample size, the differently scaled parameters (continuous, ordinary, binary) and lack of normal distribution. Moreover it cannot replace evaluation by single parameters including expert judgment. Nevertheless, we believe that these analyses are useful to evaluate the data through visualization of the response per animals and to exhibit experimental characteristics (including bias and artefacts). Further investigations of preclinical studies are needed, including the correlation between morphology and (immune) functioning. It is concluded that PCA helps to identify key pathology by reducing the data and to identify the diversity in response to the substance under investigation.

References Abboudi, H., Macphee, I.A., 2012. Individualized immunosuppression in transplant patients: potential role of pharmacogenetics. Pharmgenomics Pers. Med. 5, 63–72. Baken, K.A., Pennings, J.L., Jonker, M.J., Schaap, M.M., de Vries, A., van Steeg, H., Breit, T.M., van Loveren, H., 2008. Overlapping gene expression profiles of model compounds provide opportunities for immunotoxicity screening. Toxicol. Appl. Pharmacol. 226, 46–59. Carlson, E.A., Li, Y., Zelikoff, J.T., 2004. Benzo[a]pyrene-induced immunotoxicity in Japanese medeka (Oryzias latipes): relationship between lymphoid CYP1A1 actifity and humoral immune suppression. Toxicol. Appl. Pharmacol. 201, 40–52. Cho, Y.E., Moon, P.G., Baek, M.C., 2014. An integrated proteomic and transcriptomic approach to understanding azathioprine-induced hepatotoxicity in rat primary hepatocytes. Electrophoresis 35, 911–922. Csaba, G., Inczefi-Gonda, A., Szeberenyi, Sz., 1991. Lasting impact of a single benzpyrene treatment in pre-natal and growing age on the thymic glucocorticoid receptors of rats. Gen. Pharmacol. 22, 815–818. Deschl, U., Kittel, B., Rittinghausen, S., Morawietz, G., Kohler, M., Mohr, U., Keenan, C., 2002. The value of historical control data-scientific advantages for pathologists, industry and agencies. Toxicol. Pathol. 30, 80–87. Duley, J.A., Somogyi, A.A., Martin, J.H., 2012. The future of thiopurine pharmacogenomics. Pharmacogenomics 13, 1549–1552. Ekbal, N.J., Holt, D.W., Macphee, I.A., 2008. Pharmacogenetics of immunosuppressive drugs: prospect of individual therapy for transplant patients. Pharmacogenomics 9, 585–596. Elenkov, I.J., Wilder, R.L., Chrousos, G.P., Vizi, E.S., 2000. The sympathetic nerve—an integrative interface between two supersystems: the brain and the immune system. Pharmacol. Rev. 52, 595–638. Elens, L., Bouamar, R., Shuker, N., Hesselink, D.A., van Gelder, T., van Schaik, R.H., 2014. Clinical implementation of pharmacogenetics in kidney transplantation: calcineurin inhibitors in the starting blocks. Br. J. Clin. Pharmacol. 77, 715–728. Elmore, S.A., Peddada, S., 2009. Points to consider on the statistical analysis of rodent cancer bioassay data when incorporating historical control data. Toxicol. Pathol. 37, 672–676. Ezendam, J., Staedtler, F., Pennings, J., Vandebriel, R.J., Pieters, R., Harleman, J.H., Vos, J.G., 2004. Toxicogenomics of subchronic hexachlorobenzene exposure in Brown Norway rats. Environ. Health Perspect. 112, 782–791. Festing, M.F.W., Hawkey, C.M., Hart, M.G., Turton, J.A., Gwynne, J., Hicks, R.M., 1984. Principal components analysis of haematological data from F344 rats with bladder cancer fed N-(ethyl)-all-trans-retinamide. Food Chem. Toxicol. 22, 559–572. Festing, M.W.F., Diamanti, P., Turton, J.A., 2001. Strain differences in haematological response to chloroamphenicol succinate in mice: implications for toxicological research. Food Chem. Toxicol. 39, 375–383. Frame, S.R., Mann, P.C., Caverly, J.M., 2014. Principles of pathology for toxicology studies. In: Hayes, A. Wallace, Kruger, Claire L. (Eds.), Hayes' Principles and Methods of Toxicology, Sixth edition CRC Press, Boca Raton, pp. 571–595.

Giribaldi, L., Chiappini, F., Pontillo, C., Randi, A.S., Kleiman de Pisarev, D.L., Alvarez, L., 2011. Hexachlorobenzene induces deregulation of cellular growth in rat liver. Toxicology 289, 19–27. Greenacre, M., 2010. Biplots in Practice. BBVA Foundation, Madrid, Spain, ISBN: 978-84-923846-8-6, Available for free download. Grubbs, F.E., 1969. Procedures for detecting outlying observations in samples. Technometrics 11 (1), 1–21. http://dx.doi.org/10.1080/00401706.1969.10490657. Hall, R.L., Everds, N.E., 2003. Factors affecting the interpretation of canine and nonhuman primate clinical pathology. Toxicol. Pathol. 31, S6–S10. Hayakawa, K., Mimura, Y., Tachibana, S., Furuya, M., Kodama, T., Aoki, T., Hosokawa, S., Fukui, M., Shibata, S., Yoshida, M., Masuyama, T., Narita, T., Kuwagata, M., Hisada, S., Maki, E., 2013. Study for collecting background data on Wistar Hannover [Crl:WI(Han)] rats in general toxicity studies—comparative data to Sprague Dawley rats. J. Toxicol. Sci. 38, 855–873. Hoogerbrugge, R., Willig, S.J., Kistemaker, P.G., 1983. Discriminant analysis by double stage principal component analysis. Anal. Chem. 55, 1710–1712. Hothorn, L.A., 2014. Statistical evaluation of toxicological bioassays—a review. Toxicol. Res. 3, 418–432. Iatropoulos, M.J., Hobson, W., Knauf, V., Adams, H.P., 1976. Morphological effects of hexachlorobenzene toxicity in female rhesus monkeys. Toxicol. Appl. Pharmacol. 37, 433–444. Ikarashi, I., Tauchi, K., Yukawa, M., 1994. Application of principal component analysis to the results of toxicological tests. Pharmacokinetics 48, 487–492. Julliard, W., Fechner, J.H., Mezrich, J.D., 2014. The aryl hydrocarbon receptor meets immunology: friend or foe? A little of both. Front. Immunol. 5, 458. Kaplinsky, J., Li, A., Sun, A., Coffre, M., Koralov, S.B., Arnaout, R., 2014. Antibody repertoire deep sequencing reveals antigen-independent selection in maturing B cells. Proc. Natl. Acad. Sci. USA 111, E2622–E2629. Keil, D., Luebke, R.W., Ensley, M., Gerard, P.D., Pruett, S.B., 1999. Evaluation of multivariate statistical methods for analysis and modeling of immunotoxicology data. Toxicol. Sci. 51, 245–258. Kemmerling, J., Fehlert, E., Rühl-Fehlert, C., Kuper, C.F., Stropp, G., Vogels, J., Krul, C., Vohr, H.-W., 2015. The transferability from rat subacute 4-week oral toxicity study to translational research exemplified by two pharmaceutical immunosuppressants and two environmental pollutants with immunomodulating properties (In this issue). Kobayashi, K., Sakuratani, Y., Abe, T., Nishikawa, S., Yamada, J., Hirose, A., Kamata, E., Hayashi, M., 2010. Relation between statistics and treatment-related changes obtained from toxicity studies in rats: if detected a significant difference in low or middle dose for quantitative values, this change is considered as incidental change? J. Toxicol. Sci. 35, 79–85. Kropf, S., Hothorn, L.A., Lauter, J., 1997. Multivariate many-to-one procedures with applications to preclinical trials. Drug Inf. J. 31, 433–447. Lee, J.M., Lee, M.A., Do, H.N., Song, Y.I., Bae, R.J., Lee, H.Y., Park, S.H., Kang, J.S., Kang, J.K., 2012. Historical control data from 13-week repeated toxicity studies in Crj: CD (SD) rats. Lab. Anim. Res. 28, 115–121. Leegwater, D.C., Kuper, C.F., 1984. Evaluation of histological changes in the kidneys of the alloxan diabetic rat by means of factor analysis. Food Chem. Toxicol. 22, 551–557. Long, G.G., Hardisty, J.F., 2012. Regulatory forum opinion piece: thresholds in toxicologic pathology. Toxicol. Pathol. 40, 1079–1081. Martignoni, M., Groothuis, G.M.M., De Kanter, R., 2006. Species differences between mouse, rat, dog, monkey and human CYP-mediated drug metabolism, inhibition and induction. Expert Opin. Drug Metab. Toxicol. 2, 875–894. Massart, D.L., Vandeginste, B.G.M., Buydens, L.M.C., De Jong, S., Lewi, P.J., SmeyersVerbeke, J., 1997. Handbook of Chemometrics and Qualimetrics: Part A. Elsevier, Amsterdam. McInnes, E.F., Scudamore, C.L., 2014. Review of approaches to the recording of background lesions in toxicologic pathology studies in rats. Toxicol. Lett. 229, 134–143. Miri-Dashe, T., Osawe, S., Tokdung, M., Daniel, N., Choji, R.P., Mamman, I., Deme, K., Damulak, D., Abimiku, A., 2014. Comprehensive reference ranges for hematology and clinical chemistry laboratory parameters derived from normal Nigerian adults. PLoS One 9, e93919. http://dx.doi.org/10.1371/journal.pone.0093919. OECD, 1995. Guidelines for the Testing of Chemicals/no. 407: Repeated Dose Oral Toxicity Test Method. Organization for Economic Cooperation and Development, Paris, France. Okamura, T., Suzuki, S., Ogawa, T., Kobayashi, J., Kusuoka, O., Hatayama, K., Mochizuki, M., Hoshiya, T., Okazaki, S., Tamura, K., 2011. Background data for general toxicology parameters in RccHan:WIST Rats at 8, 10, 19 and 32 weeks of age. J. Toxicol. Pathol. 24, 195–205. Pallet, N., Legendre, C., 2010. Deciphering calcineurin inhibitor nephrotoxicity: a pharmacological approach. Pharmacogenomics 11, 1491–1501. Palm, A., Havermark, A., Meyerson, B.J., Nylander, I., Roman, E., 2011. When is a Wistar a Wistar? Behavioral profiling of outbred Wistar rats from five different suppliers using the MCSF test. Appl. Anim. Behav. Sci. 135, 128–137. Petterino, C., Argentino-Storino, A., 2006. Clinical chemistry and hematology historical data in control Sprague–Dawley rats from pre-clinical toxicity studies. Exp. Toxicol. Pathol. 57, 213–219. Randi, A.S., Sanchez, M.S., Alvarez, L., Cardozo, J., Pontillo, C, Kleiman de Pisarev, D.L., 2008. Hexachlorobenzene triggers AhR translocation to the nucleus, c-Src activation and EGFR transactivation in rat liver. Toxicol. Lett. 177, 116–122. Schountz, T., Quackenbush, S., Rovnak, J., Haddock, E., Black, W.C.IV, Feldmann, H., Prescott, J., 2014. Differential lymphocyte and antibody responses in deer mice infected with Sin Nombre hantavirus or Andes hantavirus. J. Virol. 88, 8319–8331.

C.F. Kuper et al. / European Journal of Pharmacology 759 (2015) 343–355

Schraplau, A., Schewe, B., Neuschäfer-Rube, F., Ringel, S., Neuber, C., Kleuser, B., Püschel, G.P., 2015. Enhanced thyroid hormone breakdown in hepatocytes by mutual induction of the constitutive androstane receptor (CAR, NR1I3) and arylhydrocarbon receptor by benzo[a]pyrene and phenobarbital. Toxicology 328, 21–28. Schulte, A., Althoff, J., Ewe, S., Richter-Reichhelm, H.B., 2002. Two immunotoxicity ring studies according to OECD TG 407—comparison of data on cyclosporine A and hexachlorobenzene. Regul. Toxicol. Pharmacol. 36, 12–21. Serfilippi, L.M., Stackhouse-Pallman, D.R., Russell, B., Spainhour, C., 2003. Serum clinical chemistry and hematology reference values in outbred stocks of albino mice from three commonly used vendors and two inbred strains of albino mice. Contemp. Top. Lab. Anim. Sci. 42, 46–52. Shackelford, C., Long, G., Wolf, J., Okerberg, C., Herbert, R., 2002. Qualitative and quantitative analysis of nonneoplastic lesions in toxicology studies. Toxicol. Pathol. 30, 93–96. Sibille, M., 1990. Selection of healthy volunteers for phase I studies. Fundam. Clin. Pharmacol. 4 (Suppl. 2), S167–S176. Sibille, M., Deigat, N., Durieu, I., Guillaumont, M., Morel, D., Bienvenu, J., Massignon, D., Durand, D.V., 1999. Laboratory data in healthy volunteers: reference values, reference changes, screening and laboratory adverse event limits in Phase I clinical trials. Eur. J. Clin. Pharmacol. 55, 13–19. Sparfel, L., Pinel-Marie, M.L., Boize, M., Koscielny, S., Desmots, S., Pery, A., Fardel, O., 2010. Transcriptional signature of human macrophages exposed to the environmental contaminant benzo(a)pyrene. Toxicol. Sci. 114, 247–259. Strauss, V., Wiemer, J., Leibold, E., Kamp, H., Walk, T., Mellert, W., Looser, R., Prokoudine, A., Fabian, E., Herold, M., van Ravenzwaay, B., 2009. Influence of strain and sex on the metabolic profile of rats in repeated dose toxicological studies. Toxicol. Lett. 91, 88–95. Teml, A., Schaeffeler, E., Herrlinger, K.R., Klotz, U., Schwab, M., 2007. Thiopurine treatment in inflammatory bowel disease: clinical pharmacology and implication of pharmacogenetically guided dosing. Clin. Pharmacokinet. 46, 187–208. The ICICIS Group Investigators, 1998. Report of validation study of direct immunotoxicity in the rat. Toxicology 125, 183–201. U.S. Department of Health and Human Services, Agency for Toxic Substances and Disease Registry, 2013. Draft Toxicological Profile for Hexachlorobenzene.

355

〈http://www.atsdr.cdc.gov/ToxProfiles/tp.asp?id=627&tid=115〉 (last accessed 14.01.15.). Uno, S., Dalton, T.P., Dragin, N., Curran, C.P., Derkenne, S., Miller, M.L., Shertzer, H.G., Gonzalez, F.J., Nebert, D.W., 2006. Oral benzo[a]pyrene in Cyp1 knockout mouse lines: CYP1A1 important in detoxication, CYP1B1 metabolism required for immune damage independent of total-body burden and clearance rate. Mol. Pharmacol. 69, 1103–1114. Vandeginste, B.G.M., Massart, D.L., Buydens, L.M.C., De Jong, S., Lewi, P.J., SmeyersVerbeke, J., 1998. Handbook of Chemometrics and Qualimetrics: Part B. Elsevier, Amsterdam. Van Grevenynghe, J., Bernard, M., Langouet, S., Le Berre, C., Fest, T., Fardel, O., 2005. Human CD34-positive hematopoietic stem cells constitute targets for carcinogenic polycyclic aromatic hydrocarbons. J. Pharmacol. Exp. Ther. 314, 693–702. Vermeire, S., Rutgeerts, P., 2005. Current status of genetics research in inflammatory bowel disease. Genes Immun. 6, 637–645. Wagener, A.H., Zwinderman, A.H., Luiten, S., Fokkens, W.J., Bel, E.H., Sterk, P.J., Van Drunen, C.M., 2013. The impact of allergic rhinitis and asthma on human nasal and bronchial epithelial gene expression. PLoS One 8, 1–7. http://dx.doi.org/ 10.1371/journal.pone.0080257. Watanabe, K.H., Li, Z., Kroll, K.J., Villeneuve, D.L., Garcia-Reyero, N., Orlando, E.F., Sepulveda, M.S., Collette, T.W., Ekman, D.R., Ankley, G.T., Denslow, N.D., 2009. A computational model of the hypothalamic-pituitary-gonadal axis in male fathead minnows exposed to 17alpha-ethinylestradiol and 17beta-estradiol. Toxicol. Sci. 109, 180–192. Yao, F., Coquery, J., Le Cao, K.-A., 2012. Independent principal component analysis for biologically meaningful dimension reduction of large biological data sets. BMC Bioinform. 13, 1–15. http://dx.doi.org/10.1186/1471-2105-13-24. Zeh, C., Amornkul, P.N., Inzaule, S., Ondoa, P., Oyaro, B., Dufton, M.M., Vandenhoudt, H., Gichangi, A., Williamson, J., Thomas, T., DeCock, K.M., Hart, C., Nkengasong, J., Laserson, K., 2011. Population-based biochemistry, immunologic and haematological reference values for adolescents and young adults in a rural population in Western Kenya. PLoS One 6, 1–10. http://dx.doi.org/10.1371/ journal.pone.0021040.