Necessity of Bumped Kinase Inhibitor Gastrointestinal Exposure in Treating Cryptosporidium Infection

Summary Physiologically based pharmacokinetic models for bumped kinase inhibitors suggest that in vivo anticryptosporidial efficacy is associated with gastrointestinal concentrations of drug.

Cryptosporidium, a genus of protozoan parasites distributed worldwide, causes severe diarrhea disproportionately affecting those with poor nutrition and/or immunodeficiency. The Global Enteric Multicenter Study (GEMS) found Cryptosporidium to be the second leading cause of moderate to severe diarrheal disease and the leading pathogen associated with death in 6-18-monthold children at sites in both Africa and Asia [1][2][3]. Furthermore, children who survive cryptosporidiosis have associated growth and development stunting known to contribute to increased allcause mortality rates and developmental issues [4]. The sole drug for cryptosporidiosis, nitazoxanide, has only a 30% response rate in malnourished children and no efficacy in human immunodeficiency virus-infected individuals [5]. Improved therapeutics are the greatest hope in the near term for addressing death and disability from Cryptosporidium infection. Species and strain antigenic differences and a lack of cross-protection impedes the development of a broadly protective vaccine [6].
There are unique challenges in drug development for cryptosporidiosis in that Cryptosporidium is a protozoan parasite and is localized to the gastrointestinal (GI) tract. As a protozoan parasite, Cryptosporidium is not susceptible to antibiotics which target processes specific to bacteria. In addition, Cryptosporidium is more closely related to humans than to bacteria or viruses, and this increases the risk for potential drug toxicity. We have reduced host toxicity by developing bumped kinase inhibitors (BKIs) that selectively target calcium-dependent protein kinase 1 (CpCDPK1) compared with host kinases. A subset of the BKIs have shown inhibitor constant (K i ) values <1 nmol/L against CpCDPK1 and inhibit the growth of Cryptosporidium in vitro with EC 50 values <5 μmol/L against the parasite itself while demonstrating little toxicity to human cells or in preclinical animal models [7].
Although we and others have made progress in the development of selective inhibitors of Cryptosporidium growth, the pharmacokinetic (PK) parameters that predict efficacy are unknown for compounds addressing its distinct localization. PK/pharmacodynamic (PD) indices use PK parameters such as the maximal and average unbound drug concentration (C max , C avg ) and area under the drug concentration-time curve (AUC), along with in vitro-derived efficacy parameters, such as the 50% minimum inhibitory concentration or EC 50 , to predict the in vivo efficacy of drug candidates [8]. In the antimicrobial field, these indices have been well established and are used to guide the selection of dosing regimens for drug classes such as β-lactams [9,10] and fluoroquinolones [11][12][13]. For anti-Cryptosporidium drugs, not only are we unsure which PK/PD index should be used for predicting in vivo efficacy, it is unclear which tissue drug concentration will provide the information required for predicting the desired in vivo response.
Once inside the GI tract, Cryptosporidium attaches to and invades enterocytes, where it becomes enclosed in a parasitophorous vacuole that underlies the apical host cell plasma membrane facing the fecal stream. Unique membranous and cytoskeletal structures separate the parasite from host cell cytoplasm, and the level and direction of molecular exchange between them remains poorly understood [14][15][16][17]. Given the unique localization of Cryptosporidium, systemic exposure may not be required for BKI efficacy, and the in vivo efficacy of BKIs may correlate better with their GI exposure. Targeting this profile would be expected to minimize toxicity associated with systemic drug distribution. A similar strategy has been used for treating nonsystemic GI infections with rifaximin, a nonabsorbable rifamycin antibiotic [18]. For BKIs, a correlation between fecal concentrations of BKI and in vivo efficacy was reported elsewhere [7]. However, further work is required to clarify the correlation between BKI exposure in different parts of the GI tract and the observed in vivo efficacy. It is difficult to measure the concentration of a drug in solution along the GI tract.
In the current study, we applied computational modeling using Gastroplus software (Simulations Plus) to develop a physiologically based PK model to simulate systemic and GI BKI concentrations for 2 mouse models of cryptosporidiosis. To investigate how differences in compound pharmacokinetics may affect in vivo efficacy, we selected 8 BKIs that demonstrate similar potency against C. parvum in vitro but display divergent in vivo pharmacokinetics. Dose-normalized and simulated concentrations of BKIs predict the in vivo efficacy in a neonatal and adult mouse model of C. parvum infection. The information on the importance of GI drug exposure provided herein will be valuable for the entire research community currently engaged in developing therapeutics for cryptosporidiosis and other enteric diseases.

METHODS AND MATERIALS
Information regarding the C. parvum oocysts, generation of BKIs, CpCDPK1 enzyme inhibition, C. parvum growth inhibition by BKIs, plasma protein binding, and neonatal mouse model of C. parvum infection has been described in detail elsewhere [7]. Information on BKI solubility and permeability, Gastroplus modeling, the interferon (IFN) γ knockout (KO) mouse model, and measurement of BKI levels in the GI tract of mice is described in the Supplementary Material. All animal experiments were approved by the Institutional Animal Care and Use Committees at the participating universities (University of Washington, University of Georgia, and University of Arizona).
To predict the inhibition of C. parvum growth in the neonatal model of C. parvum infection, the dose-normalized C max was used to predict the inhibition of parasite growth along with a sigmoidal E max model as described in Equation 1: where A is the amount of RLUs measured the previous day; e is the mathematical constant Euler's number; k, the growth rate constant; and t, the time in days. The predicted RLU values were log-transformed and fit with a linear regression. The slope of the predicted RLU values was compared with that observed in the BKI-treated mice by analysis of covariance, using GraphPad Prism software, and if the slopes differed significantly (P ≤ .05) the prediction was considered inaccurate.

BKI Pharmacokinetics and Predicted Efficacy in a Neonatal Mouse Model of Cryptosporidium Infection
Although the 8 BKIs used in this study share the same pyrazolo[2,3-d]pyrimidine scaffold, differing chemical side chains generate variances in plasma PK after an oral dose. When a single oral dose of each BKI was administered to mice, 30-to 60-fold differences between the C max and AUC values were observed, respectively (Table 1). Given the lack of knowledge on the PK/PD relationship for BKI anti-Cryptosporidium efficacy, the single-dose BKI PK parameters and their respective EC 50 values do not provide the information necessary for prioritizing compounds to test in preclinical animal models. Therefore, the BKIs were tested in a neonatal model of C. parvum infection to determine whether any systemic PK parameters are associated with in vivo efficacy.
Although each of the BKIs generated a significant reduction in C. parvum infection when tested in a neonatal model of cryptosporidiosis across the set, the in vivo efficacy had a large range of 33%-70%. Surprisingly, BKI 1553 displayed the highest C max and largest AUC after a single oral dose, yet it had the least efficacy in clearing parasite infection, other than 1557 ( Table 1). The BKI 1369 had the most efficacy in clearing parasite infection, although its C max was the second lowest of the 8 BKIs, and its EC 50 value for in vitro parasite growth inhibition was nearly identical to that of the other BKIs tested.
The single-dose C max and AUC values were dose normalized and used to determine whether in vivo efficacy could have been predicted with BKI systemic PK values ( Figure 1A), with the assumption that the BKI have linear PK. No significant association was observed between the efficacy and the dose-normalized BKI unbound C max /EC 50 value, suggesting that the dose-normalized unbound C max is not a predictor of in vivo efficacy. A comparable result was obtained when the total dose-normalized C max /EC 50 was used to predict efficacy (data not shown). Similar to the C max data, the dose-normalized AUC values did not predict the in vivo efficacy ( Figure 1B). Finally, a sigmoidal E max model that incorporated the dose-normalized C max and the EC 50 of each BKI was evaluated for its ability to predict the in vivo efficacy of each compound [19]. However, neither the unbound nor total dose-normalized C max values predicted the in vivo efficacy ( Figure 1C).
To overcome the inability to predict in vivo efficacy with systemic levels of BKI, the in vivo pharmacokinetics and physiochemical properties of each BKI were incorporated into a physiologically based PK model to simulate the BKI GI concentrations. BKI concentrations were simulated using the dosing protocol of the neonatal study, and the GI concentrations reported refer to the segments of the Gastroplus advanced compartmental absorption and transit simulation model. The mouse advanced compartmental absorption and transit model was adjusted as described in Supplemental Table 1 to represent more appropriately the reported physiology of a neonatal mouse [20]. For each GI segment, simulated  Predicting in vivo efficacy with dose-normalized bumped kinase inhibitor (BKI) maximal concentration (C max ) and area under the curve (AUC). A, Observed reduction in a neonatal mouse model of Cryptosporidium parvum infection was plotted as a function of the dose-normalized C max /EC 50 ratio for each BKI where EC 50 is the BKI concentration that inhibits 50% of Cryptosporidium growth in vitro. There was not a significant association between the 2 values. B, Similar to the dose-normalized C max /EC 50 ratio, the dose-normalized AUC was not a predictor of in vivo efficacy. C, Finally, sigmoidal E max model was used to predict the in vivo efficacy, but neither the total nor the unbound dose-normalized C max predicted the in vivo efficacy.
concentrations of BKI in the lumen and the enterocytes were used to determine their respective C max and C avg values over the course of the neonatal efficacy study (Supplemental Table 2).
Representative GI concentration versus time profiles for the BKIs are shown in Figure 2 for BKIs 1294 and 1553. The C max for BKI 1553 is predicted to be greater than for BKI 1294 in plasma and a majority of the GI tract, whereas the C max for BKI 1553 in the cecum and ascending colon are predicted to be less than for BKI 1294. In addition, although BKI 1553 is predicted to have brief increases in concentration after each dose, BKI 1294 is predicted to remain in the cecum and ascending colon for an extended period. Therefore, the predicted C avg values for BKI 1294 in these GI sections are much higher than those for BKI 1553.
The simulated concentrations for BKIs 1294 and 1553 in the plasma and the GI tract were used to predict efficacy in the neonatal mouse model. As expected, both the plasma C max and the C avg were not predictors of BKI in vivo efficacy, and accounting for plasma protein binding did not improve the prediction ( Figure 3A and 3B).
The predicted BKI efficacy in each section of the GI tract was compared with the observed reduction in neonatal C. parvum infection (Supplemental Figure 2 and Figure 3). Similar to the efficacy predicted with systemic concentrations, BKI efficacy was not predicted using simulated concentrations in the duodenum, jejunum, or ileum (Supplemental Figure 2). However, when BKI concentrations in the enterocytes of the ascending colon were used to predict the in vivo efficacy, there was a significant (P < .05) positive association between the predicted and the observed efficacy ( Figure 3C). Furthermore, there was a nonsignificant trend between the predicted and observed efficacy when ascending colon BKI concentrations in the lumen were used for the predictions ( Figure 3D). In addition, when BKI concentrations in the lumen of the cecum were used to predict the in vivo efficacy, there was a significant (P < .05) positive association between the predicted and observed efficacy (Supplemental Figure 2D).

Adult Mouse Model of Cryptosporidium parvum Infection
The ability to predict in vivo efficacy with BKI ascending colon concentrations was further evaluated in an adult model of C. parvum infection. Female IFN-γ KO mice were infected with transgenic C. parvum strain UGA1 expressing nanoluciferase (Nluc), and the infection was allowed to proceed for 3-5 days [21]. Infected mice were treated with oral doses of BKI 1294 (60 mg/kg), BKI 1553 (10 mg/kg), or vehicle control Because the plasma half-life of BKI 1553 is longer than that of BKI 1294, the systemic accumulation is greater for BKI 1553 over the duration of the study. In the GI lumen, whereas BKI 1294 exposure is predicted to be maintained between doses in the cecum and the ascending colon, BKI 1553 levels are predicted to rapidly rise and fall in each section of the GI tract after each dose, producing long periods with little BKI 1553 exposure between dosing.
daily for 5 days, and oocyst shedding was measured by quantifying parasite-derived luminescence in the feces. To account for daily growth of parasites over the course of the study, the observed RLUs in the vehicle-treated mice were used to simulate the growth of the parasite over the course of the experiment ( Figure 4A). On each day of the study, the simulated parasite level was based on both the growth of the parasite and the predicted inhibition by BKI [22][23][24]. The BKIs 1553 and 1294 C max and C avg in the lumen and the enterocytes of the ascending colon were used to predict the daily in vivo efficacy of the BKI over the duration of the study. These parameters were calculated for each day of the study using the simulated BKI concentration versus time profile over the 5 days of treatment (Supplemental Table 3). The observed and predicted luminescence was log-transformed, and linear regression was used to fit the data for the 5 days after treatment was initiated. The predicted efficacy was accepted as accurate if the predictions approximated what was observed and if there was not a significant difference (P > .05) between the slopes of the predicted and observed change in luminescence. When the simulated ascending colon luminal C max and C avg for BKI 1553 and 1294 were used to predict the in vivo efficacy, there was a significant difference (P < .05) between the predicted and observed RLU slopes over the course of BKI treatment, demonstrating that these values did not accurately predict the observed efficacy ( Figure 4B and 4C). In contrast, the simulated enterocyte C avg values for BKI 1553 and 1294 accurately predicted the rate of luminescence decrease, because the slopes of the predicted and observed luminescence did not differ significantly (P > .05) ( Figure 4B and 4C). For BKI 1294, the simulated enterocyte C max accurately predicted the in vivo efficacy as well ( Figure 4C).
The utility of the model was further evaluated with a dose escalation study of BKI 1534. Control vehicle-treated mice  were again used to simulate parasite growth. Three days after infection, mice were treated once daily with a vehicle control or with BKI 1534 at doses of 6, 20, or 60 mg/kg ( Figure 5). The ascending colon enterocyte C max and C avg values were calculated from the simulated concentration versus time profiles and used to predict the in vivo efficacy. For the 6-mg/kg study, both the enterocyte C max and C avg accurately predicted the lack of BKI efficacy. However, though the simulated enterocyte C max accurately predicted the in vivo efficacy after doses of 20 or 60 mg/kg, the efficacy predicted with the enterocyte C avg did not match the observed inhibition of parasite growth ( Figure 5B and 5C).

In vivo BKI Exposure Along the GI Tract
To experimentally confirm that plasma concentrations of BKI are indeed not representative of their concentrations in the GI tract, BKI 1294 (60 mg/kg) and BKI 1553 (10 mg/kg) were each given as a single oral dose to 9 mice. The average plasma concentration of BKI 1553 was approximately 10 µmol/L by 0.5 hours, significantly (P < .05) higher than that for BKI 1294 at each time point ( Figure 6A and 6B), consistent with findings of the initial single-dose PK study. Although the systemic levels of BKI 1553 were higher than those for BKI 1294, GI concentrations of the latter were higher over the course of the study   Adult interferon γ knockout mice were infected with nanoluciferase-expressing C. parvum oocytes. Therefore, oocyst excretion can be quantified by measuring luminescence after lysis of the fecal content. A, In vivo growth rate of C. parvum was simulated by log-transforming the observed relative luminescence in the feces of 3 vehicle-treated mice and fitting the data with a linear regression. B, C, The resulting growth rate was incorporated into the in vivo efficacy studies with bumped kinase inhibitor (BKI) 1553 (B) and 1294 (C). Ascending colon lumen or enterocyte levels were used to predict the inhibition of C. parvum growth with 5 daily oral doses of BKI 1553 (10 mg/kg) (B) or BKI 1294 (60 mg/kg) (C). The slope of the predicted relative luminescence unit (RLU) values was compared with that observed in the mice treated with BKI by analysis of covariance, and if the slopes differed significantly (P ≤ .05), the prediction was considered inaccurate. In vivo efficacy was accurately predicted with BKI 1553 enterocyte values for the average and maximal concentrations (C avg ), but the lumen values overpredicted the in vivo efficacy (B). Similarly, the BKI 1294 efficacy was accurately predicted with the enterocyte C max and C avg , but the lumen BKI 1294 levels over predicted the in vivo efficacy (C). For each study, 3 mice were used in both the BKI-treated and the vehicle-treated groups. Open circles represent results generated from the pooled feces of the 3 mice.
( Figure 6C and 6D). Because BKI efficacy was associated with the ascending colon levels of BKI, the concentrations of BKI 1294 and 1553 in the ascending colon were compared using Student t test. At 4 hours, the average BKI 1294 concentration was significantly higher than that of BKI 1553 (46 vs 9 µmol/L, respectively; P < .01).

DISCUSSION
The pathogen localization within its host's organs and tissues is of utmost importance for the development of antimicrobial therapy. Simple models based on serum exposure alone are insufficient to predict cure, and knowledge of the specific site at which drug exposure is necessary is crucial to the development of novel therapeutics. As one of many examples, this is frequently discussed in the context of the blood-brain barrier and its impact on the treatment of different stages of African trypanosomiasis [25]. Information to build a reliable PK/PD index is not currently available for any class of anti-Cryptosporidium drugs; thus, we selected 8 BKIs with similar in vitro efficacy but divergent PK properties to identify PK characteristics associated with in vivo efficacy.
Plasma BKI levels do not predict in vivo efficacy, suggesting that systemic distribution of BKI is not required for anti-Cryptosporidium efficacy in mice. The lack of association between systemic exposure and in vivo efficacy is probably due to the GI localization of the parasite along the entire GI tract in both IFN-γ KO [26] and SCID/beige mice [27]. Dose-normalized PK values for the BKIs were used along with in vitro derived EC 50 values to predict their respective efficacy in a neonatal model of C. parvum infection, but no associations were identified between any of the PK/PD indices and in vivo efficacy. Previous work in our laboratory identified BKIs lacking systemic exposure that generated an in vivo reduction in C. parvum infection [7], and a similar lack of association between systemic exposure and efficacy has been reported for the treatment of bacterial diarrhea by oral bicozamycin [28] and aztreonam [29]. Owing to the poor efficacy predictions generated with systemic levels of BKI and a correlation previously reported between high fecal levels of BKI and in vivo efficacy [7], we chose to investigate the importance of GI drug exposure for in vivo BKI efficacy.
Simulated large intestinal BKI levels accurately predicted the efficacy of the BKIs. Based on the parasite localization throughout the GI tract, it is likely that exposure of BKI in both the small and large intestines is crucial for efficacy, because BKIs with low exposure in the large intestine fail to clear the infection. It is likely that permeable compounds are quickly absorbed in the small intestine (eg, BKI 1553) and less permeable compounds are carried further along the GI tract (eg, BKI 1294), allowing them to maintain anti-Cryptosporidium activity. Although these initial BKI observations provided encouraging results, we tested the model with an IFN-γ KO adult mouse model of C. parvum infection to confirm the ability to predict in vivo efficacy with ascending colon BKI levels.
In agreement with findings in the neonatal mouse model, simulated ascending colon BKI concentrations accurately predicted the in vivo response to BKI treatment in the IFN-γ KO adult mouse model of Cryptosporidium infection. Although the neonatal mouse model provides only a single end point for assessing BKI efficacy, IFN-γ KO mice infected with Nlucexpressing C. parvum allow for daily measurement of C. parvum infection. This enabled development of a dynamic BKI efficacy model based on previous PK/PD modeling efforts for antibacterials and antitumor drugs [12,[22][23][24]. This model can account for differences in proliferation rates between Cryptosporidium strains or even in batches of oocysts.
We confirmed that simulated levels of BKI in the enterocytes of the ascending colon accurately predicted the in vivo efficacy of 3 BKIs. However, the importance of the C max and/or the C avg in the efficacy prediction was not consistent across the data set, so it is unclear whether the pharmacodynamics of the BKIs are time or concentration dependent. The GI BKI values measured in vivo were in agreement with the predicted Gastroplus values. The average maximum BKI 1294 and 1553 concentrations measured in the ascending colon were slightly higher than the values predicted for the enterocytes. This elevation could be due to residual BKI remaining after the lumen was flushed with phosphate-buffered saline and BKI in cells other than enterocytes. Although the measured ascending colon concentrations were higher than those simulated, the BKI 1553 values were significantly less than those measured with BKI 1294, correlating to the efficacy underperformance of BKI 1553 compared with BKI 1294.
Taken together, the results presented here provide strong support for optimizing drug exposure along the entire GI tract when treating Cryptosporidium infection. Therefore, adjusting the formulation to achieve the desired GI exposure profile will be essential for developing anti-Cryptosporidium therapies. Formulation strategies for such treatments may be especially challenging, because it is unclear how disease and altered GI transit times will affect drug dissolution and absorption. Mice do not develop clinical diarrhea with C. parvum infection, but studies in infected calves that exhibit diarrheal symptoms of cryptosporidiosis suggest that BKI exposure is not significantly altered by infection [30]. Malnourished children are an important focus of the current target product profile for Cryptosporidium drugs [31,32], but it is unclear whether malnourishment will affect BKI pharmacokinetics, because previous studies have observed significant changes in antibiotic exposure in children with severe acute malnutrition [33][34][35][36][37][38][39][40].
In conclusion, to our knowledge this is the first study to investigate the importance of GI drug exposure for in vivo anti-Cryptosporidium efficacy. The results of this work have the potential to provide fundamental support for drug discovery programs in prioritizing lead drug candidates. In addition, these predictions can be tested in other GI infections and disorders in which large intestine epithelial cell delivery may be important.