Regression Modeling of the Antioxidant-to-Nephroprotective Relation Shows the Pivotal Role of Oxidative Stress in Cisplatin Nephrotoxicity

The clinical utility of the chemotherapeutic drug cisplatin is significantly limited by its nephrotoxicity, which is characterized by electrolytic disorders, glomerular filtration rate decline, and azotemia. These alterations are consequences of a primary tubulopathy causing injury to proximal and distal epithelial cells, and thus tubular dysfunction. Oxidative stress plays a role in cisplatin nephrotoxicity and cytotoxicity, but its relative contribution to overall toxicity remains unknown. We studied the relation between the degree of oxidative reduction (provided by antioxidant treatment) and the extent of nephrotoxicity amelioration (i.e., nephroprotection) by means of a regression analysis of studies in animal models. Our results indicate that a linear relation exists between these two parameters, and that this relation very nearly crosses the value of maximal nephroprotection at maximal antioxidant effect, suggesting that oxidative stress seems to be a pivotal and mandatory mechanism of cisplatin nephrotoxicity, and, hence, an interesting, rationale-based target for clinical use. Our model also serves to identify antioxidants with enhanced effectiveness by comparing their actual nephroprotective power with that predicted by their antioxidant effect. Among those, this study identified nanoceria, erythropoietin, and maltol as highly effective candidates affording more nephroprotection than expected from their antioxidant effect for prospective clinical development.


Introduction
Cisplatin is one of the most potent and widely used chemotherapeutic drugs for the treatment of a variety of solid cancers [1], but its dosage and clinical utility are limited by nephrotoxicity [2]. Nephrotoxicity occurs in 25-35% of adult [3] and 70% of pediatric [4] therapeutic courses. Direct effects on the renal vasculature are involved [2], but cisplatin nephrotoxicity mostly shows a tubular damage pattern of dysfunction and derangement, producing electrolytic disturbances (i.e., most typically hypomagnesemia and hypokalemia), acute tubular injury (ATI), and acute kidney injury (AKI), with elevated plasma creatinine (pCr) and urea (pUrea) levels [2,[5][6][7], which may occasionally progress to chronic fibrotic nephropathy [8,9]. As shown in Figure 1, tubular damage causes a reduction in glomerular filtration rate (GFR) by a number of mechanisms, including activation of the tubuloglomerular feedback (TGF) mechanism and renal vasoconstriction induced by inflammation and factors released by activated renal cells [2,10]. This pathophysiological pattern results from cisplatin accumulation in proximal (mainly the S3 segment) [11,12] and distal tubule cells [2,13], which causes diverse cellular alterations, chiefly including inhibition of membrane transporters [2,14], interference with metabolic pathways [15], and cell death [16,17]. Tubular cell death shows apoptotic and nonapoptotic phenotypes, depending on the level of exposure to cisplatin [16]. While lower concentrations induce apoptosis, higher concentrations cause a necrotic-like pheno-type [18,19]. Inside the cells, cisplatin becomes aquated and turns into a potent nucleophilic that binds to numerous targets, most prominently nucleic acids and many proteins [18,20]. Cisplatin cytotoxicity has been traditionally explained by formation of inter-and intrastrand adducts with nuclear DNA, which activates DNA repair mechanisms that, when overwhelmed, in turn, activate apoptosis. This cytotoxic mechanism is very effective in rapidly dividing cells, because nonrepaired DNA activates the p53-p21 cyclin-dependent kinase 2 (cdk2) pathway to make death/life decisions at cell division checkpoints [21].
Oxidative stress has been proposed as a prominent event and mediator of cisplatin cytotoxicity and nephrotoxicity [2,13,21], but its relative weight among other pathophysiological mechanisms, and its hierarchical and causality relation with them, are mostly unknown. In this article, we studied and modeled the relation between the degree of reduction in oxidative stress and the degree of protection of cisplatin nephrotoxicity bestowed by exogenous antioxidants in a number of studies with animal models. A key role of oxidative stress in cisplatin nephrotoxicity was inferred from the linear relation between the antioxidant and nephroprotective effects, with almost complete prevention of nephrotoxicity at maximal antioxidant effect.

Data Mining
The data used for this study were obtained from the literature search carried out in our previous meta-analysis [41], in which preclinical studies reporting molecules or products preventing cisplatin nephrotoxicity were identified. Among them, only those articles meeting the following criteria were used: (1) evaluating antioxidant nephroprotectants, (2) conducted on experimental animals, (3) providing number of individuals per experimental group, (4) using cisplatin as the nephrotoxic agent, (5) written in English, (6) fully accessible for authors (through journal subscriptions, request to authors, or open access), (7) using pUrea or blood urea nitrogen (BUN) level as the parameter to estimate nephroprotection, and (8) using malonyldialdehyde (MDA) to evaluate oxidative stress, as previously described [42]. The subsequent mathematical analysis was performed only with those studies reporting statistically significant nephroprotective and antioxidant effects (with respect to the group that received cisplatin and no nephroprotectant). Publication bias was evaluated with the asymmetry tests of Begg and Mazumdar [43], and Egger et al. [44].

Mathematical Modeling
With the objective of evaluating a potential relation between the antioxidant and the nephroprotective activity of the nephroprotectants included in the study, the following parameters were defined: MaxNP is the value of the nephrotoxicity biomarker (i.e., pUrea or pBUN) at the maximum toxicity time in the nephroprotectant+cisplatin group; BasNP is the value of the nephrotoxicity biomarker at basal time point in the nephroprotectant+cisplatin group; MaxNA is the value of the nephrotoxicity biomarker at the maximum toxicity time point in the cisplatin group; and BasNA is the value of the nephrotoxicity biomarker at the basal time point in the cisplatin group. Thereof, MaxNP-BasNP corresponds to the increment in the level of the nephrotoxicity biomarker in the nephroprotectant+cisplatin group; and MaxNA-BasNA corresponds to the increment in the level of the same nephrotoxicity biomarker in the aminoglycoside group. E nep > 0 denotes nephroprotection (i.e., reduced cisplatin nephrotoxicity due to the action of the nephroprotectant), with the higher the value of E nep , the higher the nephroprotective effect. E nep = 1 represents total nephroprotection. E nep = 0 means there is no effect exerted by the nephroprotectant.

•
Antioxidant index (E oxi ): The E oxi versus E nep relation was represented. We used an ordinary least squares (OLS) approach for building a linear regression model between E oxi as an independent variable, and E nep as a dependent variable. Nonlinear models were also taken into account, but they did not improve the performance achieved by their linear counterparts. As on the basal state of E oxi = 0, we should expect no nephroprotection effect (E nep = 0); we used this fact in the model assessment and supposed a zero-centered model that was tested with a proper model assessment using the Akaike information criterion (AIC). In particular, the final linear regression model was given by a weighted linear combination: The model was assessed by measuring the statistical significance of the coefficients; the variability of the relationship between the predictors and the target value was determined by the corresponding R 2 coefficient [45].
As the presence of outliers was considerably high, two additional robust techniques were considered, namely the Huber regression and the random sample consensus (RANSAC) algorithm [46,47]. The Huber regression is a robust technique that uses a Huber loss function instead of the standard least squares in order to penalize the error depending on their magnitude [47]. RANSAC is an iterative estimation algorithm which fits several iterative models on subsets of data, and then selects the subset with the least average error that, by assumption, is the subset with no outlier points [46]. The value of the slope coefficient corresponding to each of the three models was eventually compared as an evaluation metric about the influence of outliers in the coefficient estimation carried out by OLS. All three models were fitted using Python module Scikit-learn [48]; the rest of the processing was performed in Python programming language [49].

Results
The characteristics of the studies included in this work are provided in Table 1. The Begg-Mazumdar test applied to assess potential publication bias yielded a Kendall's tau of 0.74 (p < 0.001). Similarly, the Egger test provided a bias of 10.49 (95% CI = 9.56, 11.43; p < 0.001). Both tests showed the presence of asymmetry. However, in our study, this result was expected and is not necessarily reflective of publication bias. In fact, pursuant of our objective, only studies reporting a statistically significant nephroprotective effect were included, as stated in the Methods.
The OLS model was evaluated by checking the statistical significance of the coefficients with an alpha error threshold of 0.01. We obtained the following results: w = 0.938 (95% CI = (0.89, 0.987), p < 0.0001). We used the Akaike information criterion (AIC) [112] to assess the choice of including or excluding the bias term from the final model. We obtained an AIC intercept = −66.32 for the linear model with bias term, and AIC base = −61.90 for the model without bias term. Therefore, based on this result, we only kept the slope term in the resulting model, R 2 = 0.932, meaning that 93.2% of the variability of E nep could be explained by E oxi .
The potential influence of outliers in the final model was also assessed. In particular, Huber and RANSAC regressions were obtained, assuming the same model as for the OLS case (i.e., without intercept term). Both algorithms yielded similar slope values to that obtained by the OLS regression model: w Huber = 0.957 and w RANSAC = 0.965. We concluded that, under our assumptions, the outliers had no significant influence on the final model. The three models are depicted in Figure 2. Studies in which E oxi > 1 were removed from the models. In these studies, the antioxidant reduced oxidative stress beyond the basal level (i.e., the level of oxidative stress in the control group), which had a negative impact on nephroprotection. Specifically, E nep showed a negative slope beyond E oxi = 1 (data not shown). This is because normal (i.e., basal) production of reactive oxygen species (ROS) has been shown to have homeostatic signaling roles [113][114][115]. As a corollary, inhibition of basal ROS production may reasonably result in deleterious effects for cell and organ function [116]. Products located over the model provided more nephroprotection than expected from their antioxidant effect. Based on the RANSAC regression (the model with the most robust fit), these products were subclassified as showing 25, 50, 75, or 100% of additional nephroprotection ( Figure 3); they are identified and listed in Table 2.  Table 2. Letters (a through v) identify individual products, as listed in Table 2.

Discussion
The regression model best adjusting our experimental data shows a linear relationship between inhibition of oxidative stress and amelioration of cisplatin nephrotoxicity (Figure 2). This relation intercepts the nephroprotection axis (i.e., the y-axis) very near the E nep = 1 value at the maximal antioxidant point (i.e., E oxi = 1 in the x-axis). This indicates that a complete abrogation of oxidative stress apparently leads to a complete prevention of nephrotoxicity. Thus, oxidative stress might not only be a contributing, but a pivotal mechanism of cisplatin nephrotoxicity. Cisplatin nephrotoxicity is a tubulopathy, in which all pathophysiological and clinical manifestations derive from cytotoxic tubular injury as the primary event ( Figure 1) [2,13]. Consequently, oxidative stress must also be in the core of cisplatin cytotoxicity.
Mitochondria are the main intracellular site of cell life/death decision [117][118][119]. Mitochondria funnel and integrate stress signals arising from damaged subcellular structures and organelles, including themselves, and activate apoptotic and necrotic death programs that mostly pose no-return points for cell demise [120]. One of these signals is oxidative stress. Extramitochondrial sources of ROS exist (e.g., the cytosol and the endoplasmic reticulum) [121], but mitochondria are the main source of ROS production and overproduction [115]. Mitochondrial outer membrane permeabilization (MOMP) is a mandatory event for the release of proapoptotic factors (e.g., cytochrome c and AIF), apoptosome formation in the cytosol, and initiation of intrinsic apoptosis [119]. Intracellular death signals regulate MOMP by targeting the outer membrane through pro-and anti-apoptotic Bcl-2 family members, which directly modulate its permeability [118,122]. Inner membrane permeabilization (i.e., mitochondrial permeability transition, MPT) is also intimately related to cell death. MPT is bidirectionally linked to transmembrane mitochondrial potential (∆Ψ) dissipation, and causes intermembrane swelling, outer membrane disruption, and MOMP. MPT is mediated by a multiprotein complex, the permeability transition pore (PT pore or PTP). PTP is located at sites of inner-outer membrane connections (where Bcl-2 family members accumulate), is inhibited by anti-apoptotic Bcl-2 members, is critical for apoptosis, and participates in MOMP [123][124][125].
In isolated mitochondria [126], cisplatin interferes with the respiratory chain, produces oxidative stress [127] and rapid cytochrome c release [128], and causes calcium-dependent mitochondrial swelling and mitochondrial depolarization, as a consequence of PT pore opening [129]. In this scenario, oxidative stress may be the cause or the consequence of the other events. In fact, decoupling or inhibition of mitochondrial respiration induces both PT pore opening and oxidative stress [117,130,131]. PT pore opening (and, thus, MPT) is triggered by mitochondrial Ca 2+ and potentiated by oxidative stress [125,132,133], suggesting that alterations in respiration induce oxidative stress, and this, in turn, contributes to the opening of the PT pore. In agreement, antioxidants inhibit MPT [134]. However, vice versa is also possible: PT pore opening produces ∆Ψ dissipation, respiratory uncoupling, and oxidative stress [124,132,135]. As such, oxidative stress and mitochondrial dysfunction induce one another [136,137], and so a causality dilemma existed for cisplatin cytotoxicity [16]. Cisplatin also causes oxidative stress by directly damaging mitochondrial DNA (mtDNA) [128,[138][139][140], which impairs appropriate expression of mitochondrial enzymes forming the respiratory chain, and thus induces oxidative stress. Finally, cisplatin abates the antioxidant barrier by inhibiting superoxide dismutase (SOD), catalase, glutathione peroxidase, glutathione S-transferase [141][142][143], and glutathione reductase [144] in kidney tissues. Figure 4 summarizes the participation of oxidative stress in the tubular pathophysiological scenario. The results of the present study are more congruent with oxidative stress being mainly upstream of MPT and MOMP, because, after these mitochondrial events have occurred, the cell is irreversibly committed to dying [120]. Our study closely relates oxidative stress to the reduction in glomerular filtration (using pUrea as a proxy). GFR reduction is a pivotal alteration in cisplatin nephrotoxicity, derived mostly from tubular cytotoxicity (as shown in Figure 1) [2], and an internationally recognized hallmark of AKI, regardless of etiology [145]. However, tubular damage and GFR decline are not directly proportional. In fact, an undetermined degree of tubular damage may occur without affecting GFR [146,147], as undamaged nephrons may, to a certain extent, sustain (total) GFR by increasing their single-nephron GFR (SNGFR) [148]. This implies that additional injury mechanisms (unrelated to oxidative stress) might re-main under maximal antioxidant circumstances. Potential oxidative stress-independent mechanisms are known (see Figures 1 and 4, and [2,16,21]), but their weight in cisplatin toxicity is unknown. They would pose potential targets for pharmacological intervention in combination with antioxidants to optimize cisplatin nephrotoxicity prophylaxis. As previously reported [42], nephroprotectants whose effect lies above the model line are products showing greater protective effect than expected from their antioxidant effect. This suggests that additional protection mechanisms are involved, which makes them especially interesting candidates for clinical application. The most effective candidates include nanoceria, recombinant human erythropoietin, maltol, and the butanolic extract of Centaurea choulettiana Pomel (Table 2). On the contrary, those compounds lying below the model line are less effective than expected, implying that they also activate counteracting mechanisms, and are thus less interesting.
Along with its antioxidant effect, nanoceria (cerium oxide nanoparticles) also shows anti-inflammatory [101] and antiapoptotic properties [149]. Its anti-inflammatory effect has been shown to derive from the inhibition of inducible nitric oxide synthase (iNOS) expression [150] and of the NF-κB signaling pathway [151]. With regard to erythropoietin, multiple additional mechanisms have been invoked, including (i) the promotion of tubular cell regeneration, (ii) the reduction in vascular endothelial growth factor (VEGF), hemeoxygenase-1 (HO-1) and iNOS expression [89], (iii) the inactivation of macrophages via NF-κB [152], (iv) the inhibition of TGF-β1 expression [153], and (v) the reduction in polymorphonuclear cell infiltration [154]. Anti-inflammatory and antiapoptotic properties with involvement of the AMPK/PI3K/Akt pathway have also been attributed to maltol, an ingredient in the food industry [88]. Finally, traditional medicine has attributed antiinflammatory properties to Centaurea choulettiana [155]. However, oxidative stress is known to be involved in the development and perpetuation of inflammation [156,157], and in the activation of apoptosis [158]. Accordingly, their anti-inflammatory and antiapoptotic properties might be the consequence of their antioxidant capacity, and would thus not explain their additional properties, which need to be further explored. Because drug discovery from plant extracts is a complicated and long process, nanoceria, erythropoietin, and maltol hoard readier potential to become clinical applications, and should thus be further explored.

Conclusions
Our results have revealed that oxidative stress is not a contributing, but a central mediator of cisplatin nephrotoxicity in preclinical models. In agreement, a recent meta-analysis identified antioxidants as the most effective protectants of cisplatin nephrotoxicity in clinical studies [159]. Interestingly, several antioxidants have shown, in animal models, nephroprotective properties without interfering with the antitumor effect of cisplatin [160][161][162][163][164], a critical issue for clinical application. This might be attributed to cisplatin genotoxicity mostly impacting on proliferating cells, such as tumor cells. In perspective, this study provides a rationale for further clinical development of preventive strategies based on single or combined therapies containing antioxidants.

Data Availability Statement:
The data presented in this study are available in the article.

Conflicts of Interest:
The authors declare no conflict of interest.