Heat map of mRNA expression levels of 26 selected transporters and membrane markers under different culture conditions. B) Exemplary graphs of apical transporters organic anion transporter 4

(OAT4) and multidrug and toxin extrusion proteins 1 + 2 (MATE1+2) expression under different culture conditions. C) Exemplary graphs of basolateral transporters organic cation transporter 1 (OCT1) and organic anion transporters 2 + 3 (OAT2+3) expression. Mean ± SEM, n = 4 with technical triplicates (2D and 2.5D); n = 10 with technical duplicates (3D). One-way ANOVA with Bonferroni’s post-test. p ≤ 0.05 (*), p ≤ 0.01 (**), p ≤ 0.001 (***). D) Confocal microscopy of OCT substrate ASP⁺ (upper panel) and OAT substrate LY (lower panel) uptake for 24 h into 3D grown cells. Nuclei were stained with Hoechst (blue). Signal intensity was quantified in cells and lumen of tubules stained with Hoechst only (- ASP⁺/LY) or tubules exposed to ASP⁺ or LY (+ ASP⁺/LY). Mean ± SD, 15 regions of interest. 2D = grown on plastic, 2.5D = grown on transwell inserts, 3D = grown in matrigel sandwich, BF = bright field, RFU

= relative fluorescence unit.

105 Cisplatin causes delayed kidney injury in ~30% of patients after single injection (Ries and Klastersky, 1986). Assessing low-dose and in particular delayed nephrotoxicity in vitro remains difficult since many cells in culture need to be passaged after a few days and express only a subset of the transporters expressed in vivo. Using a culture method allowing for long-term cultivation of in vitro generated proximal tubular structures, we sought to overcome these problems. To validate our 3D proximal tubule model as nephrotoxicity model, we performed acute (72 h) and delayed (24 h exposure plus 10 days culture) cisplatin toxicity assays.

Using acute cisplatin treatment, we compared different methods to assess cytotoxicity and cell viability in our 3D proximal tubule model. Cell death, assessed via LDH leakage, was induced with a medium cisplatin concentration (50 µM) both in 2.5D and 3D cultured cells (Fig. 4 A, left). At the highest concentration of cisplatin (100 µM), ~45% of all cells were dead after 72 h as assessed via LDH leakage (Fig. 4 A, left). Using the resazurin assay which measures cell viability, we observed a dose-dependent decrease in viability in both cell culture conditions (Fig. 4 A, right), that however, was more strongly pronounced under 2.5D. Therefore, we proofed that both of the classical cytotoxicity endpoints, LDH leakage and resazurin reduction, can easily be applied to assess cisplatin-induced cytotoxicity in 3D cultured RPTEC/TERT1 cells with consistent outcome. Concomitantly, this provided evidence that 3D cultured RPTEC/TERT1 cells express functional OCT2 and CTR1 necessary for uptake of cisplatin.

Next, single dose treatment with various cisplatin concentrations was performed for 24 h with subsequent culture for 10 days. This setting resembles the clinic exposure scheme used for chemotherapeutic treatment of patients, of whom ~30% experience renal failure with a delayed onset (Ries and Klastersky, 1986). By monitoring glucose consumption and lactate secretion for an 11-day period, we were able to identify early metabolic responses in the highest cisplatin concentration (100 µM) (Fig. 4 C, D). Both parameters (glucose and lactate) peaked on day 3 (48 h after cisplatin exposure) and then started to drop significantly, thereby clearly indicating cell death. All other tested concentrations of cisplatin induced a delayed metabolic response in 3D cultured cells, starting on day 5 (Fig. 4 C, D). The transient increase in glycolytic activity during the first two days might represent a secondary stress response, since high concentrations of cisplatin are known to interfere with mitochondrial activity (Martins et al., 2008) in turn increasing glycolytic rates. Likewise, LDH leakage clearly indicates that cisplatin provokes a delayed and concentration-dependent cytotoxic response (Fig.

4 E). The highest cisplatin concentration (100 µM) showed highest cytotoxicity starting at day 3 and increased up to 80% on day 11. The medium cisplatin concentration (50 µM) reached maximum LDH leakage at day 11 with 67%. The lowest cisplatin concentration (10 µM) induced a slight but not significant increase in LDH leakage upon day 11.

In summary, we were able to induce and monitor cisplatin-induced cytotoxicity in our 3D model of the proximal tubule. Whereas this model seems to lack higher sensitivity in acute toxicity settings when compared to simpler culture conditions (Fig. 4 A, B), the assessment of delayed and concentration-dependent toxicity appears promising for future investigations of post-acute toxicity. However,

106

comparing data in simpler culture conditions is still missing. Quantification of glucose consumption and lactate production proofed to be a valid and reliable tool to assess cytotoxicity in complex matrices like matrigel cultures, whereas LDH leakage is less sensitive at low concentrations.

Fig. 4: Acute and delayed cisplatin-induced cytotoxicity in 3D tubular structures

A-B) Acute cisplatin treatment of RPTEC/TERT1 cells for 72 h cultured on transwell inserts (2.5D) or in a matrigel sandwich (3D). Toxicity was assessed by A) LDH leakage and B) resazurin reduction using four concentrations (0, 10, 50, 100 µM). Mean ± SD, n = 3 with technical triplicates. C-E) Delayed toxicity of 3D grown RPTEC/TERT1 cells exposed for 24 h with various concentrations of cisplatin with subsequent culture for 10 days. C) Glucose consumption, D) lactate secretion and E) LDH leakage was assessed over a period of 11 days. Mean ± SEM, n = 4 with technical triplicates.

107 The renal proximal tubule plays a major role in maintaining the body’s fluid, ion and glucose homeostasis.

Therefore, the underlying mechanisms of physiology and pathophysiology of proximal tubule cells are of key interest to researchers of many scientific areas. Since the proximal tubule is also a main target for drug-induced toxicity due to its role in drug transport and secretion, the detection of nephrotoxic lead compounds in nonclinical trials is of particular interest. Due to interspecies variability, today’s animal models have only limited predictivity for human nephrotoxicity (Tiong et al., 2014). The inability to predict drug-induced kidney injury early during pharmacological development is illustrated by the fact that nephrotoxicity only accounts for 2% and 5% of drug attrition in nonclinical and phase I studies (Redfern et al., 2010) but is responsible for 19% of drug attrition during large scale phase III studies (Tiong et al., 2014). Reliable nonclinical detection would allow for rapid cancellation of potentially nephrotoxic drug candidates resulting in huge money and time saving and reduction of patient harm. Alarming is the absolute lack of approved or validated in vitro models for prediction of nephrotoxicity while other organs systems are heavily researched in this regard (Tiong et al., 2014). Interestingly, the potential of an in vitro system to accurately predict adverse outcomes seems to be rather dependent on the differentiation status of the cell than on the selected end point (Tiong et al., 2014). However, a fully differentiated proximal tubule cell system is still lacking. Hence, it appears obvious that having a highly differentiated cultured proximal tubule cell in hand, would be of extreme advantage for research and development.

Here, we present a cell culture technique inducing tubular structure formation of the RPTEC/TERT1 cell line. This cell line not only highly resembles the morphology and functionality of in vivo proximal tubule cells, but expresses also a variety of drug transporters and can be cultured as differentiated monolayer for extended time periods (Aschauer et al., 2015; Wieser et al., 2008). Still, transporter expression levels are markedly lower when compared to in vivo data and expression and functional transport of organic anion transporters (OATs) is still controversial (this publication,(Aschauer et al., 2015)). We hypothesized that RPTEC/TERT1 cells have the capability to even better differentiate when cultured in a 3D matrix.

Indeed, we could show that solely providing the cells with a complex scaffold is sufficient to differentiate the RPTEC/TERT1 cell line more towards an in vivo-like proximal tubule without the need of additional growth factors. By proving glucose uptake, lactate secretion as well as lack of cell death during long-term culture of these tubular structures, we conclude that the 3D cultured RPTEC/TERT1 cells are highly stable and viable for extended culture periods up to 60 days. Using the sandwich culture method (culturing cells in between two matrigel layers), we provide a rather simple culture technique allowing e.g. for microscopic analyses since tubular structures form within one layer and morphology can be therefore easily monitored using bright-field microscopy.

So far, several cell types have been reported to form tubule-like structures when cultured in extracellular matrices like matrigel (Debnath et al., 2002; Hadley et al., 1990; Niemann et al., 1998). Indeed, there is data on 3D culture of primary rabbit (Han et al., 2004) as well as of primary human proximal tubule cells (DesRochers et al., 2013) and, very recently, the RPTEC/TERT1 cell line (Homan et al., 2016). In comparison to model systems using primary cells, our model has the advantage of unlimited and

well-108

defined cell source lacking interdonor variability. In contrast to our approach, Homan et al. used bioprinted 3D proximal tubules on a chip populated with RPTEC/TERT1 cells allowing also the formation of a differentiated and polar epithelium but lacking the ease of use and cost efficiency our sandwich culture of RPTEC/TERT1 cells provides. In strong contrast to the study of Homan et al., our system is based on the intrinsic capability of the RPTEC/TERT1 cells to form 3D, polarized structures without external trigger, e.g. seeding cells into a pre-existing tubular scaffold. The latter forces the cells to attach to the walls of the tube-shape scaffold resulting rather in an imitation of a 3D culture.

Our results provide strong evidence that the tubular structures grown by cultivation of RPTEC/TERT1 in a matrigel sandwich are morphologically and functionally similar to in vivo proximal tubules.

Macroscopically, the cells formed branched tubules characterized by a single layer of cells encircling a lumen. Using confocal and electron microcopy, we confirmed that the cells are highly polarized indicated by the basolateral localization of Na⁺/K⁺-ATPase, apical cell-cell connections via tight junctions and primary cilia located at the apical brush border. These results were supported by qPCR analyses showing highest mRNA expression of brush border enzymes like GGT1, LAP3 and ALPL in the 3D culture compared to simpler culture conditions. Our culture condition resulted in an improved differentiation status of the RPTEC/TERT1 cell line indicated by an overall increase in transporter expression. Of special interest is the strong increase in expression of OAT2 and OAT4 as well as the de novo expression of OAT3 since these transporters were barely or not detectable in RPTEC/TERT1 cultured in less complex conditions. This finding is a convincing example of how improved cell culture environment supports cellular differentiation status. All transporters (except for OCT2 and GLUT1) showed a higher expression levels in the 3D culture compared to the 2.5D culture on membrane inserts, the current culture system of choice for RPTEC/TERT1 cells (Aschauer et al., 2015). Since we provide strong evidence that the cells forming the tubule are correctly polarized, we hypothesize that the tested transporters are also correctly localized at their respective apical or basolateral membrane. Using fluorescent substrates for OCTs and OATs, we observed an uptake of both ASP⁺ and LY into the cells. Contrary to other studies (Freedman et al., 2015;

Han et al., 2004; Masereeuw et al., 1999) and our expectations, the substrates were not secreted into the tubular lumen. In contrast to the studies by others and the physiological situation in vivo, our system lacks continuous flow of a primary urine analog which is known to trigger widening of the inner and outer diameter of proximal tubules (Du et al., 2006; Raghavan and Weisz, 2016). Having this in mind, it appears likely that the absence of flow reduces both, the tubular diameter and the capacity to excrete substrates into the lumen of the tubules in our system.

Nevertheless, we provide first evidence that our 3D model can be used to assess acute and – most importantly – delayed cisplatin-induced cytotoxicity. LDH leakage and resazurin reduction can be easily assessed in matrigel-grown cells particularly in acute toxicity settings. Quantification of both glucose consumption and lactate production proofed to be useful tools for online monitoring of delayed metabolic changes resembling toxicity, since these secreted metabolites are sufficiently small to freely diffuse through the matrigel to be measured in the medium. LDH leakage appears to be slightly less sensitive when compared to glucose and lactate quantification, likely due to slower diffusion of LDH through matrigel. Whereas the 3D model appears to be less sensitive in acute toxicity settings with cisplatin when compared to 2.5D cultures, comparative studies of delayed toxicity are needed to

109 determine if the 3D culture of RPTEC/TERT1 cells provides significant advantages in nephrotoxicity assessment.

Taken together, the 3D culture model of RPTEC/TERT1 cells presented here shows morphological and functional similarity to human kidney proximal tubules in vivo. Culturing RPTEC/TERT1 cells in a matrigel sandwich induced tubule formation and further increased differentiation regarding transporter expression and polarity compared to classic 2D or 2.5D culture on transwell inserts indicating that this cell line is capable of improving its characteristics when provided with optimal culture conditions. The easy-to-use model described here represents an interesting and cost-efficient alternative method for addressing specific endpoints, e.g. discovering compounds interfering with tubule formation, differentiation and polarization as well as crude cytotoxicity.

Acknowledgements

We thank the Bioimaging Center (University of Konstanz) for providing the imaging equipment and support and the Electron Microscopy Center (University of Konstanz) for performing electron microscopy.

110

Protein (gene) name NCBI Ref. Seq. Forward primer Reverse primer OAT2 (SLC22A7) NM_006672 AGCCTACGTGAGTACCCTGG CACTCCAGCTCCAGTGGC OAT3 (SLC22A8) NM_004254 CACGAGCCCTCCAATCAGTA CTGGGTCTACAACAGCACCA OAT4 (SLC22A11) NM_018484 CCGCAGTAGATGACGAATGTT ATCCTGGTGGGCTCCTTTAT MRP1 (ABCC1) NM_004996 CTGGACTGATGACCCCATCG GCGATCCCTTGTGAAATGCC MRP2 (ABCC2) NM_000392 GGGATCTCTTCCACACTGGAT CATACAGGCCCTGAAGAGGA MRP3 (ABCC3) NM_003786 GTCCCATTCCGCTCCAAGAT TCAGGGTAGGGGTTAGGGTC MRP4 (ABCC4) NM_005845 TCTCCGTTTATGGCCAATTT CCGTGTACCAGGAGGTGAAG MRP5 (ABCC5) NM_005688 TGAGCTGAGAATGCATGGAG GAGAACCAGCACTTCTGGGA OCT1 (SLC22A1) NM_003057 CCCCTCATTTTGTTTGCGGT TTTCTCCCAAGGTTCTCGGC OCT2 (SLC22A2) NM_003058 TGCATATTTTCGGCTTCCTC ACCGGCTCACTAACATCTGG OCT3 (SLC22A3) NM_021977 AGGTGAATGCTCCAGTCAGG ACTCCACCATCGTCAGCG CTR1 (SLC31A1) NM_001859 GGTTGGGTGATGGTGAGAAG GCTAGTGGCTGGACTTGACC MATE1 (SLC47A1) NM_018242 TGATCAGGAACACCATCAGC GAGGCCACCCTTGAGGTC MATE2 (SLC47A2) NM_152908 TGCTTCCCAGTTCCTCTCAG GAAGATGTCATTGCCCTGGT OCTN1 (SLC22A4) NM_003059 CGTGACCGAGTGGAATCTGG AGCCATGGTTGCGAAGAGAA OCTN2 (SLC22A5) NM_003060 ACACCCACGAAGAACAAGGA ATGGCTGGGAGTTCAGTCAG PGP (ABCB1) NM_000927 ACAGAGGGGATGGTCAGTGT TCACGGCCATAGCGAATGTT BCRP (ABCG1) NM_004827 TGGTGTTTCCTTGTGACACTG TGAGCCTTTGGTTAAGACCG SGLT1 (SLC5A1) NM_000343 TGGCCACTTCCAATGTTACT GGGACTGTTGGAGGCTTCTT SGLT2 (SLC5A2) NM_003041 TTCACCAAGATCTCAGTGGACAT GAAGGTCTGTACCGTGTCCG GLUT1 (SLC2A1) NM_006516 GGCATTGATGACTCCAGTGTT ATGGAGCCCAGCAGCAA Cubilin (CUBN) NM_001081 GGACGGCCATTACTCACAAAAG TTTGTCCACCTCCTCAGTTCC AQP1 (AQP1) NM_198098 GGAGGGTCCCGATGATCT CTCTCAGGCATCACCTCCTC GGT1 (GGT1) NM_013421 CTCATAGCCTCGGATCTCCC ACAACAGCACCACACGAAAA ALPL (ALPL) NM_000478 GACCTCGTTGACACCTGGAA CTGGCTCGAAGAGACCCAAT LAP3 (LAP3) NM_015907 GCAGAAGCCTTGATGGAGAT CCCCCAGTCTTCTTGGAAAT βActin (ACTB) NM_001101 GTTGTCGACGACGAGCG GCACAGAGCCTCGCCTT HPRT1 (HPRT1) NM_000194 CACCCTTTCCAAATCCTCAG CTCCGTTATGGCGACCC RPL13A (RPL13A) NM_012423 GGTATGCTGCCCCACAAAACC CTGTCACTGCCTGGTACTTCCA

S1 Table: Primer pairs for quantitative real-time PCR. List of all used forward and reverse primers including NCBI Reference Sequence numbers

111 Fig. S1: mRNA expression levels of all tested transporters and membrane markers under

different culture conditions

Expression levels of A) anion transporters B) cation transporters C) glucose transporters D) protein and water transporters E) housekeeping genes. Mean ± SEM, n = 4 with technical triplicates (2D and 2.5D); n = 10 with technical duplicates (3D). One-way ANOVA with Bonferroni’s post-test. p ≤ 0.05 (*), p ≤ 0.01 (**), p ≤ 0.001 (***).

112

113 Drug-induced cytosolic and nuclear protein accumulations in rat kidney are thought to be a recurrent phenomenon arising in nonclinical safety studies (Dietrich et al., 2008; Gopinath and Mowat, 2014; Radi et al., 2013). However, the relevance for human risk assessment as well as the underlying mechanism remain to be determined. In addition, the protein accumulations – previously supposed to consist mainly of the peroxisomal protein D-amino acid oxidase (DAAO) (Dietrich et al., 2008) – needed to be better characterized and mechanism(s) inducing mislocalization and accumulation needed to be elucidated.

Subcellular compartmentalization into organelles is a key feature of eukaryotic cells allowing for separation of biological processes. Surprisingly, also the nucleus, hosting the genome and handling the genetic information, deals with various external proteins and is in constant exchange with the cytosol (Kirli et al., 2015).

Neurodegenerative diseases such as Parkinson’s, Alzheimer’s, Huntington’s disease and ALS are typically characterized by distinct inclusion bodies of aggregated proteins or fibrils in the cytosol, the nucleus or outside the cell (Ross and Poirier, 2005). The term inclusion body refers to the sequestration of aggregated proteins including active retrograde transport along microtubules (Kopito, 2000). Whereas cytosolic aggregates are transported on microtubules towards perinuclear regions (Kopito, 2000), the mechanism underlying nuclear inclusion body formation is poorly understood (Ross and Poirier, 2005).

Although the term inclusion body was also used in the context of drug-induced DAAO accumulations (Radi et al., 2013; Shimoyama et al., 2014), it remains to be determined whether DAAO accumulates in a native or an aggregated, non-functional state; concluding that the use of “inclusion body” is deceptive and should not be used in connection with DAAO accumulations. However, since cytosolic DAAO accumulations are not found in perinuclear regions but randomly distributed within the cytosol, active retrograde transport along microtubules seems unlikely. Furthermore, inclusion bodies observed in neurodegenerative diseases are significantly smaller in size compared to DAAO accumulations which fill the entire nucleus displacing the chromatin to the periphery of the nucleus. Moreover, inclusion bodies related to neurodegenerative diseases are restricted to several brain regions (Takalo et al., 2013).

Interestingly, the correlation of presence and toxicity of inclusion bodies such as in neurodegenerative diseases is questioned nowadays, assuming that the formation of inclusion bodies is rather protective than toxic in these cases (Ross and Poirier, 2005).

114

Intranuclear inclusions are also a recurrent pathological finding in human samples which can be used for diagnosis of various disease entities and tumor types. In this context, nuclear inclusions are defined as accumulation of substances in the nuclear matrix which are normally absent there and is commonly associated with displaced chromatin. Known nucleus-accumulating substances are viral particles (e.g.

herpes simplex), immunoglobulins (e.g. lymphoplasmacytic lymphoma), glycogen (nonspecific), biotin (e.g. pulmonary blastoma) or polyglutamine (Huntington’s disease), accumulating in various tissues. (Ip et al., 2010; Kasnic et al., 1982)

Since drug-induced DAAO accumulations are supposed to be kidney-specific, the occurrence and prevalence of nuclear protein accumulations in kidney cells is of key interest for this thesis. Indeed, several reports provide evidence that intranuclear inclusions are found exclusively in proximal tubule cells of several species after exposure to lead salts (Kilham et al., 1962; Shelton and Egle, 1982). Nuclear accumulation of lead is an early hallmark of lead intoxication and is thought to act as cellular sink for lead which enters the nucleus and is bound by several proteins and Ca²⁺ (Moore and Goyer, 1974). In addition, nuclear bismuth inclusions were occasionally observed after therapeutic treatment with bismuth both in the proximal tubule of humans and several animals including rats (Ghadially, 1988). Similar to DAAO accumulations, these cases of intranuclear inclusions induced by lead or bismuth showed large, spherical inclusions displacing chromatin, however they were found to consist mainly of the accumulating metal and only to a lesser extent of protein (Ghadially, 1988).

The question remains open as to how peroxisomal proteins obtain a nuclear localization. As highly oxidative organelles, the maintenance of proper protein import into peroxisomes can be assumed crucial for the cellular redox homeostasis. However, several peroxisomal matrix proteins have dually targeted isoforms executing the same or a similar function at another place (Ast et al., 2013). As an example, the 3-hydroxy-3-methylglutaryl coenzyme A lyase can be targeted to mitochondria and peroxisomes by possessing both a mitochondrial and a peroxisomal targeting signal, respectively (Ashmarina et al., 1996).

Whereas several peroxisomal proteins were described to localize additionally to the cytosol or mitochondria, there is only one known protein (glycerol 3-phosphate dehydrogenase, Gpd1p) in S. cerevisiae which can be targeted to the cytosol and – most interestingly for my topic – the nucleus (Jung et al., 2010). In this case, peroxisomal localization is favored upon phosphorylation of the PTS2 sequence of Gpd1p whereas nuclear localization was induced by osmotic stress or heat shock without the presence of a classic nuclear localization signal (Jung et al., 2010).

Interestingly, we found a putative NTS variant within the protein sequences of rDAAO and hDAAO.

However, neither mutational activation nor inactivation of the putative NTS resulted in altered DAAO localization concluding that the putative NTS sequences are not functional in DAAO. Consequently, we hypothesized that nuclear localization of DAAO must result from passive diffusion. Indeed, interference with peroxisomal transport via deletion of the PTS1 signal or PEX5 knockdown resulted in nuclear DAAO localization, thereby supporting the hypothesis. However, since our EYFP-hDAAO fusion protein accounts

115 for 135 kDa, this finding was highly contradictory to the often-cited size exclusion limit of ~40 kDa for passive diffusion into the nucleus in mammalian cells (Keminer and Peters, 1999; Knockenhauer and Schwartz, 2016; Ma et al., 2012; Paine and Feldherr, 1972). However, by employing variably sized mCherry- and/or EYFP-fusion proteins of DAAO and catalase, we showed that peroxisomal proteins until a size exclusion of approximately 135 kDa can indeed passively diffuse into mammalian cell nuclei in a size-dependent manner, thereby contradicting the often-cited 40 kDa diffusion limit.

Having shown how peroxisomal proteins can enter the nucleus, the question remained why DAAO and catalase actually are present in the nucleus. Three questions arose: 1) do peroxisomal proteins incidentally diffuse into the nucleus, 2) is there an underlying rationale for nuclear localization or 3) do they fulfill a function within the nucleus? Recently, Kirli et al. identified that a remarkably large number

Having shown how peroxisomal proteins can enter the nucleus, the question remained why DAAO and catalase actually are present in the nucleus. Three questions arose: 1) do peroxisomal proteins incidentally diffuse into the nucleus, 2) is there an underlying rationale for nuclear localization or 3) do they fulfill a function within the nucleus? Recently, Kirli et al. identified that a remarkably large number