On-target inhibition of Cryptosporidium parvum by nitazoxanide (NTZ) and paclitaxel (PTX) validated using a novel MDR1-transgenic host cell model and algorithms to quantify the effect on the parasite target

Cryptosporidium parvum is a globally distributed zoonotic protozoan parasite that causes moderate to severe, sometime deadly, watery diarrhea in humans and animals, for which fully effective treatments are yet unavailable. In studying the mechanism of action of drugs against intracellular pathogens, it is important to validate whether the observed anti-infective activity is attributed to the drug action on the pathogen or host target. For the epicellular parasite Cryptosporidium, we have previously developed a concept that the host cells with significantly increased drug tolerance by transient overexpression of the multidrug resistance protein-1 (MDR1) could be utilized to evaluate whether and how much the observed anti-cryptosporidial activity of an inhibitor was attributed to the inhibitor’s action on the parasite target. However, the transient transfection model was only applicable to evaluating native MDR1 substrates. Here we report an advanced model using stable MDR1-transgenic HCT-8 cells that allows rapid development of novel resistance to non-MDR1 substrates by multiple rounds of drug selection. Using the new model, we successfully validated that nitazoxanide, a non-MDR1 substrate and the only FDA-approved drug to treat human cryptosporidiosis, killed C. parvum by fully (100%) acting on the parasite target. We also confirmed that paclitaxel acted fully on the parasite target, while several other inhibitors including mitoxantrone, doxorubicin, vincristine and ivermectin acted partially on the parasite targets. Additionally, we developed mathematical models to quantify the proportional contribution of the on-parasite-target effect to the observed anti-cryptosporidial activity and to evaluate the relationships between several in vitro parameters, including antiparasitic efficacy (ECi), cytotoxicity (TCi), selectivity index (SI) and Hill slope (h). Owning to the promiscuity of the MDR1 efflux pump, the MDR1-transgenic host cell model could be applied to assess the on-parasite-target effects of newly identified hits/leads, either substrates or non-substrates of MDR1, against Cryptosporidium or other epicellular pathogens.


Introduction
Cryptosporidiosis is a globally distributed diarrheal disease of humans and animals. Among more than 40 Cryptosporidium species or genotypes, humans are mainly infected by C. parvum (zoonotic) and C. hominis (anthroponotic), while immunocompromised patients might also be infected by other species [1][2][3]. In people with weak or compromised immunity (e.g., infants, elderly and AIDS patients), cryptosporidial infection can be severe or deadly. Cryptosporidiosis is also a significant problem in farm animals and may cause death in neonatal calves and substantial weight loss in cattle [4,5]. On the other hand, only a single drug (i.e., nitazoxanide [NTZ]) is approved by the United States Food and Drug Administration (FDA) for treating human cryptosporidiosis. However, NTZ is not fully effective in immunocompetent patients and ineffective in immunocompromised individuals, and its mechanism of action remains undefined [6,7].
While the anti-cryptosporidial drug discovery has been impeded by some technical constraints (e.g., difficulties in manipulating the parasite in vitro and in vivo) and unique parasite biology (e.g., lack of conventional drug targets and epicellular parasitic lifestyle), an increasing effort in the past decade has resulted in the discovery of a number of leads showing excellent anti-cryptosporidial efficacy in vitro and in animal models [8][9][10][11][12][13]. Hits or leads might be identified by in vitro phenotypic screening or by target-based screening, followed by confirmation of efficacy in vitro and in vivo. For obligate intracellular parasites including Cryptosporidium, an efficacious drug may kill the parasite directly via acting on a parasite target or indirectly via acting on a host cell target, or both (i.e., the actions on both the parasite and host targets contributing to the killing of the parasite) (see illustration in S1 Fig). For simplicity, hereinafter we will use "on-target" effect to describe the action of an inhibitor "on the parasite target" and "off-target" effect to describe the action of an inhibitor "off the parasite target" (i.e., on the host target).
The validation and quantification on whether and how much a hit/lead truly inhibits the parasite by acting on the parasite target is technically challenging for Cryptosporidium and other obligate intracellular pathogens. There were actually few attempts to demonstrate on-

Transgenic HCT-8 cells overexpressing MDR1 allow relatively rapid development of drug resistance to both substrates and non-substrates of MDR1
We first generated a stable transgenic cell line by transfection of HCT-8 cells with a lentiviral vector carrying copepod GFP (copGFP) and human MDR1 genes driven by EF1α and CMV

PLOS NEGLECTED TROPICAL DISEASES
Validation and quantitation of on-target effect of anti-cryptosporidial inhibitors promoters, respectively (Fig 2A). Parental HCT-8 cells (wild-type) and those carrying blank vectors (negative control) or MDR1 gene were designated as HCT-8/WT, HCT-8/NC or HCT-8/MDR1 cells, or WT, NC or MDR1 cells for simplicity (Table 1). MDR1 cells continuously overexpressed MDR1 as demonstrated at both protein and mRNA levels (Fig 2B-2D). In comparison to NC cells, there were >1.5-fold increase of MDR1 protein and >9-fold increases of mRNA in MDR1 cells, respectively (Fig 2C and 2D). The fold increases were lower, but the . Panels E, F and G show that drug selections by PTX and NTZ had no or little effect on the expression of MDR1 at both mRNA and protein levels. Bars represent the standard errors of the means (SEMs; n = 3). Statistical significances were determined by Holm-Šídák multiple t-test between group pairs (**** = p <0.0001). https://doi.org/10.1371/journal.pntd.0011217.g002

PLOS NEGLECTED TROPICAL DISEASES
Validation and quantitation of on-target effect of anti-cryptosporidial inhibitors levels were more consistent over the time, than those in our previously reported transiently transfected cells (i.e., 2.12-to 3.37-fold for protein and >40-fold for mRNA) [15].
The transgenic cell lines were evaluated for their drug tolerance to nine compounds by MTS cytotoxicity assay, including the anti-cryptosporidial lead PTX and the only approved drug NTZ (see the description of compounds tested in this study in S1 Table) Fig 3B). The results agreed with the fact that PTX was a native substrate of MDR1, whereas NTZ was not [19][20][21]. The 1.63-fold increase of tolerance to PTX in the MDR1 cells was lower than the >2-fold increase in transiently transfected cells as previously reported [15]. MDR1 cells (vs. NC or WT cells) also exhibited increased tolerance to four of the other seven compounds, i.e., 1.54-to 1.76-fold increases to mitoxantrone (MTX), doxorubicin (DXR), vincristine (VCT) and ivermectin (IVM), but not to cyclosporin A (CSA), daunorubicin (DRC) and loperamide (LPM) (i.e., 0.95-to 1.03-fold changes) ( Table 2).
To test the hypothesis that MDR1-transgenic cells were more adaptable to drug selection for developing drug resistance to the "substrates of MDR1", we applied continuous drug pressures with stepwise increase of concentrations of PTX to MDR1 cells (vs. WT and NC cells; drug selection design in S2 Table). For clarity, a cell line after drug selection was named by adding abbreviation of the drug in parenthesis, e.g., WT(PTX), NC(PTX) or MDR1(PTX) ( Table 1). After selection with PTX, all three resulting cell lines [i.e., WT(PTX), NC(PTX) and MDR1(PTX)] increased tolerance to PTX. For comparison of cells before and after PTX selection, WT(PTX) and NC(PTX) cells showed smaller increases of PTX-resistance (i.e., 1.26-or 1.23-fold increase of TC 50 vs. WT or NC cells) ( Fig 3C; Fig 3D). It was notable that the resistance to DRC was successfully increased in MDR1(PTX) cells (2.46-fold vs. NC cells) that was unachieved in MDR1 cells (0.95-fold vs. NC cells).
We also tested the hypothesis that drug resistance to "non-substrates of MDR1" could be rapidly developed in MDR1-transgenic cells. NTZ was chosen here because it was the only  FDA-approved drug to treat human cryptosporidiosis, for which the mechanism of action still remained undefined. By applying continuous drug pressures of NTZ, all three resulting cell lines developed resistance to NTZ at varied levels (  Table 3), indicating that the developed resistance was specific for NTZ, rather than to multiple drugs.
Overexpression of MDR1 and selection with PTX or NTZ caused no apparent changes on the morphology and growth of host cells in vitro (Fig 4). Selection with either PTX or NTZ had no significant effects on MDR1 protein levels as shown by immunofluorescence assay (IFA) (Fig 4). Western blot analysis also showed that the ratios of MDR1 protein levels for the three pairs of cell lines [i.e., MDR1 vs. NC; MDR1(PTX) vs. NC(PTX) and MDR1(NTZ) vs. NC(NTZ)] were relatively consistent ( Fig 2E). Only the mRNA levels showed a relatively higher increase by the PTX-selection [i.e., MDR1(PTX) vs. NC(PTX)] (Fig 2F). How host cells after selection increased tolerance to PTX or NTZ with no significant increase of MDR1 protein level remains to be determined. One possible explanation currently under investigation is the mutations in endogenous and/or transgenic MDR1 gene as demonstrated by other investigators [22].
In short summary, stable overexpression of MDR1 in HTC-8 cells could increase tolerance of the cells to multiple MDR1 substrates (e.g., PTX), but at lower than 2-fold increase in general. The drug tolerance could be further increased by applying drug pressure. More importantly, stable overexpression of MDR1 allowed the development of drug tolerance of host cells

PLOS NEGLECTED TROPICAL DISEASES
Validation and quantitation of on-target effect of anti-cryptosporidial inhibitors to non-MDR1 substrates (e.g., NTZ) by applying drug pressure in a relatively short timeframe (e.g., in around three months to develop >2.0-fold increase of resistance to NTZ).

PTX and NTZ inhibited the growth of C. parvum by acting fully on the parasite targets, while DXR, IVM, MXT and VCT acted on both the parasite and host targets
The availability of host cells with >2 to 3-fold increase of drug resistance to NTZ, PTX and four other compounds made it possible to evaluate whether, and how much, the anti-cryptosporidial activities of these compounds were attributed to their actions on the parasite targets.
In theory, if a specified inhibitor inhibited the epicellular C. parvum in vitro by solely acting on the parasite target and its action on host cell target made no contribution to the antiparasitic activity, the increase of resistance to the inhibitor in the host cells would not affect the anti-cryptosporidial activity [15]. This could be achieved by comparing the anti-cryptosporidial efficacy (EC 50 values) with cytotoxicity (TC 50 values) of the inhibitor between MDR1 (PTX) and NC or between MDR1(NTZ) and NC cells. As a quality control, we first confirmed that overexpression of MDR1 and drug selection by either PTX or NTZ in the host cells had no effect on the infection and growth of C. parvum by a qRT-PCR-based 44-h infection assay, in which all nine cell lines showed virtually identical parasite loads (Fig 5A). There were no enrichment of MDR1 protein at the C. parvum infection sites in MDR1, MDR1(PTX) and MDR1(NTZ) cells (Fig 5B), indicating the overexpression of MDR1 would not alter the drug fluxes at the host cell-parasite interface or on the parasitophorous vacuole membrane (PVM) to complicate the drug action on the parasite. All compounds displayed no difference in their anti-cryptosporidial activities between WT and NC cells, confirming that transfection of cells with vector alone had no effect on the action of compounds Table 3

PLOS NEGLECTED TROPICAL DISEASES
Validation and quantitation of on-target effect of anti-cryptosporidial inhibitors to the parasite (Fig 6; Table 4). We then used these host cell lines to evaluate the on/off-target effects of PTX, NTZ and four other compounds by examining whether increased drug tolerance in host cells affected anti-cryptosporidial activities.

PLOS NEGLECTED TROPICAL DISEASES
Validation and quantitation of on-target effect of anti-cryptosporidial inhibitors effect on the killing of the parasite by PTX. This confirmed that PTX inhibited the parasite growth by solely acting on the parasite target (i.e., 100% on-target). The data agreed with previous observation using transient overexpression models [15]. For the other four compounds to which MDR1(PTX) cells also developed >2-fold increase of drug tolerance (i.e., DXR, IVM, MXT and VCT), their EC 50 values increased by 31.6% to 103.9% (vs. NC cells) (Fig 6B-6E; Table 4), meaning that the increase of drug tolerance affected the anti-cryptosporidial efficacies of the four compounds and their actions on host cells (i.e., off-target effect) also contributed to the killing of the parasite. Because the percent increases of EC 50 values were less than those of TC 50 values (e.g., for MXT, the percent change of EC 50 was 103.9% while that of TC 50 was 172.7%) ( Table 4), we might conclude that the off-target effects contributed partially to the killing of the parasite by the four inhibitors. In other words, both on-target and off-target effects contributed to the anti-cryptosporidial activities of the four inhibitors.
The development of NTZ-resistant cell line [i.e., MDR1(NTZ) cells] allowed us to evaluate the on-target effect of NTZ against C. parvum for the first time since its anti-cryptosporidial activity was discovered. In this assay, the increase of drug tolerance in host cells had no effect on the antiparasitic activity of NTZ based on the inhibitory curves and EC 50 values between WT, NC and MDR1(NTZ) cells (Fig 6F; Tables 4 and S3). This result confirmed that, like PTX, NTZ inhibited the C. parvum growth by fully acting on the parasite target (i.e., 100% on-target).

Full or partial on-parasite-target effects were further validated using the MDR1 inhibitor elacridar
If the increase of tolerance to a specified inhibitor in host cells was truly mediated by MDR1, specific inhibition of MDR1 would restore the sensitivity of the host cells to the inhibitor (as  . In all panels, there were no or little differences on the efficacy or cytotoxicity curves between WT and NC cells. Increased drug tolerance (i.e., reduced cytotoxicity) to paclitaxel (A) or nitazoxanide indicated by TC 50 ). Additionally, inhibition of MDR1 would not affect the anti-cryptosporidial efficacy (EC 50 ) if the drug that acted only on the parasite target, or partially affect the efficacy (EC 50 ) if the drug that also acted on host cell target. This notion was tested using elacridar, a third generation of MDR1 inhibitor [18,[23][24][25]. The concentration of elacridar at 300 nM was used based on previous studies that elacridar at this concentration could produce strong inhibition on the activity of MDR1 with no significant cytotoxicity to HCT-8 cells [15]. This study also confirmed that elacridar at 300 nM produced no or little effect on the growth of the six cell lines and on the growth of C. parvum cultured with NC, MDR1(PTX) and MDR1(NTZ) cells (Fig 7). We then examined the effect of elacridar on the cytotoxicity and anti-cryptosporidial efficacies of the six inhibitors in the range of TC 50 or EC 50 concentrations.
In NC, MDR1, NC(PTX) and MDR1(PTX) cells, elacridar dramatically reduced the tolerance of these cells to PTX, MXT, DXR, VCT and IVM (Fig 8A and 8B), indicating that: 1) the increased tolerance to the five inhibitors in MDR1 and MDR1(PTX) cells was MDR1-dependent; and 2) there was a basal level of MDR1 in NC and NC(PTX) cells (negative controls) that could be inhibited by elacridar (Fig 8A and 8B, columns in black). However, elacridar had no effect on the anti-cryptosporidial activity of PTX in both NC and MDR1(PTX) cells (Fig 8C), confirming that the killing of C. parvum by PTX was unrelated to the MDR1 activity. In other words, the action of PTX on the host cell target had no effect on anti-cryptosporidial activity by PTX, so that the killing of C. parvum was solely attributed to its action on the parasite target (100% on-target). On the other hand, elacridar reduced the anti-cryptosporidial activities of MXT, DXR, VCT and IVM in both NC and MDR1(PTX) cells, indicating that the killing of C. parvum by the four inhibitors was associated with MDR1 activity, so the actions of the four inhibitors on the host cell targets also contributed to the inhibition of the growth of C. parvum.
(F) in host cells had no effect on the anti-cryptosporidial efficacy, while increased drug tolerance (reduced cytotoxicity) to ivermectin (B), vincristine (C), doxorubicin (D) and mitoxantrone (E) in host cells reduced the anti-cryptosporidial efficacy. Bars represent the standard errors of the means (SEMs; n = 3).
https://doi.org/10.1371/journal.pntd.0011217.g006 Table 4. Anti-cryptosporidial efficacies (EC 50 ) of selected compounds in specified cell lines in comparison with corresponding drug resistance parameters and on/ off-target rates calculated based on the ratios of percent (Pct) changes between EC 50 and TC 50 values*.  In the case of NTZ, elacridar had little effect on the cytotoxicity of NTZ in NC(NTZ) cells (Fig 9A). This observation agreed with the fact that NTZ was not an MDR1 substrate, so that its cytotoxicity should not be affected by the inhibition of basal level MDR1. The tolerance to NTZ in MDR1(NTZ) cells was reverted by elacridar (Fig 9A), indicating that the NTZ-resistance developed in MDR1(NTZ) cells was related to overexpressed MDR1. In the efficacy assay, elacridar had no effect on the anti-cryptosporidial activity of NTZ in both NC and MDR1(NTZ) cells (Fig 9B), confirming the killing of C. parvum by NTZ was unrelated to the MDR1 activity, or in other words, the action of NTZ on the host cell target made no contribution to the killing of C. parvum.

Quantitative estimation of the relative contributions of on-target and offtarget effects to the observed anti-cryptosporidial activity
Estimation of on-target rate based on EC 50 and TC 50 values. We showed that stable MDR1-transgenic cells could increase drug tolerance to MDR1 substrates or non-substates in response to selection. These cell lines could serve as an in vitro model to assess whether an anti-cryptosporidial compounds killed C. parvum via acting fully (PTX and NTZ) or partially (MXT, DXR, VCT and IVM) on the parasite targets. We were also intrigued in quantifying the proportions of contributions of on-target effect (i.e., the action on the parasite target) to the antiparasitic activity. Because EC 50 and TC 50 were the most commonly used parameters for drug efficacy and cytotoxicity, we first attempted to develop a formula to calculate the on-target effect based on EC 50 and TC 50 values. The theory was that, the effect on the host target (i.e., off the parasite target) at 50% efficacy (denoted as E 50(off) ) was correlated to the ratio between the relative increase (or percent increase) of EC 50 and the relative increase of TC 50 , which could be calculated using the equation (see the equation derivations in the Methods section):

PLOS NEGLECTED TROPICAL DISEASES
while the on-target rate at 50% efficacy (denoted as E 50(on) ) could be calculated by: Using Eqs 1 and 2, we obtained theoretical on/off-target rates for the six compounds, in which the on-target rates for PTX and NTZ were 101.2% and 103.6%, respectively ( Table 4). The values were slightly higher than 100% (the theoretical maximum) due to the assaying errors. The other four compounds varied in their on/off-target rates, i.e., on-target rate 39.8% for MXT, 54.5% for DXR, 62.5% for VCT and 77.1% for IVM (Table 4). It was noticeable that the off-target effects of MXT and DXR contributed more than 50% or near 50% to the observed anti-cryptosporidial activity at 50% efficacy.
Estimation of on/off-target rates in the whole efficacy range. In theory, an inhibitor at different concentrations might act at varied levels on the parasite target and host cell target. In other words, an inhibitor's on/off-target effects might differ in their contributions to the antiparasitic activity at varied efficacy levels. We denote E i(on) or E i(off) as the on-target or off-target rate for a compound at EC i (the concentration of the compound inhibiting the parasite growth by i%; i = 0 to 100). The off-target rate E i(off) at the specified efficacy EC i could be estimated  https://doi.org/10.1371/journal.pntd.0011217.g009 using the following Eq 3 that was generalized from Eq 1:

Fig 8. Effect of MDR1-inhibition by elacridar (300 nM) on the cytotoxicity and anti-cryptosporidial activity of the five inhibitors using MDR1(PTX) cell model. (A) Effect of elacridar on the cytotoxicity of the five inhibitors on NC and MDR1 cells. (B) Effect of elacridar on the cytotoxicity of the five inhibitors on NC(PTX) and MDR1(PTX) cells. (C) Effect of elacridar on the anti-cryptosporidial activity of the five inhibitors against Cryptosporidium parvum cultured on NC and MDR1(PTX) cells. Cytotoxicity of inhibitors at specified concentrations on the host cells was evaluated by MTS cytotoxicity assay. Anti-cryptosporidial activity of inhibitors at specified concentrations was
while the on-target rate could be calculated by: where EC i and TC i (i = 0 to 100) values were calculated using a 4-parameter logistic (4PL) model (see Eq 9 and derivation in the Methods). Using Eq 3 and 4, we were able to plot the on/ off-target rates for the six inhibitors over the entire efficacy range (Fig 10). Based on the relative linearity of the 4PL model-derived sigmoidal curves (Fig 10), the on/off-target rates between EC 10 /TC 10 and EC 90 /TC 90 might be considered more reliable and biologically relevant

PLOS NEGLECTED TROPICAL DISEASES
Validation and quantitation of on-target effect of anti-cryptosporidial inhibitors (also see S5 Table for representative E on and E off values between EC 10 and EC 90 ). An open source Python code to plot the E on(i) and E off(i) curves from parameters EC 50 , TC 50 and Hill slope h was developed and deposited at the GitHub depository (https://github.com/alienn233/ PACOOTER). In the plots, both PTX and NTZ produced relatively parallel curves of E on and E off near 100% and 0%, respectively (Fig 10A and 10F). The E on and E off curves for IVM and VCT were also relatively parallel, showing higher contributions from the on-target effects (E 10(on) to E 90 (on) values = 81.6% to 73.4% for IVM, and 61.9% to 62.9% for VCT) (Fig 10B and 10C; S5 Table). There was a small surprise for DXR and MXT that showed lowest the E 50(on) values as described above, in which the E on and E off curves were non-parallel and intersected (Fig 10D  and 10E). The curves for DXR and MXT revealed that, at upper or lower effective concentrations, the on-target effect contributed more to the antiparasitic activity of DXR and MXT, but this trend was reversed at higher effective concentrations after concentrations reached to certain points (i.e., at E 62 and E 27 , respectively).

Relationships between in vitro selectivity index (SI), on-target ratio and cytotoxicity
It was noticed that an on-target inhibitor would have a larger selectivity index (SI; or SI 50 for accuracy as it was determined by the TC 50 /EC 50 ratio) [15]. Here we further observed a certain linear relationship between E 50(on) and SI in WT cells for the four partially on-target inhibitors (Fig 11A, green line). The SI values of the four inhibitors were all in single digits (i.e., SI = 0.21, 3.42, 4.1 and 4.91 for MXT, DXR, VCT and IVM, respectively) in comparison to those in double digits for NTZ and PTX (i.e., SI = 24.54 and 41.13, respectively). When nonlinear regressions were applied to all six compounds, the relationship between E 50(on) and SI roughly followed the 4PL model (Fig 11A, red line; h = 1.891; R 2 = 0.9882). The authenticity of the nonlinear relationship remained to be confirmed after more values were available, particularly those in the upper quartile of E 50(on) values. Apparently, the SI values for the four partially ontarget and the two fully on-target inhibitors were separated by the "10-fold selectivity window" that was commonly used as criterion at the hit stage of drug discovery [26].
We further examined the mathematical relationship between SI and cytotoxicity, aiming to explore whether SI in WT cells might serve as a hint for the quality of hits. The assumption was that, for a fully on-target inhibitor, the cytotoxicity would be null or minimal in the range of concentrations showing antiparasitic efficacy. Based the 4PL model, the following equation was derived (see the derivation of equations in the Methods section): where E i denoted the antiparasitic efficacy [i = 0 to 100(%)], T (Ei) denoted the cytotoxicity of the inhibitor at the concentration producing the efficacy E i , k represented selectivity index (SI) and h represented the Hill slope (note that the i values could only approach 0 or 100(%), but would never be equal to 0 or 100(%)). In this equation, h values in the efficacy and cytotoxicity curves were set to be the same based on the assumption that a specified inhibitor would possess the same or similar mode of action on the parasite and host cells. The notion was supported in part by the actual h values obtained in this study (S6 Table).
For the six compounds under investigation, both k and h were defined constants, thus allowing us to plot their relationship curves between the rates of calculated cytotoxicity and antiparasitic efficacy (Fig 11B). As expected, the cytotoxicity of all inhibitors rose along with the increase of efficacious concentrations but the trends were nonlinear and displayed as five concave and one convex rising curves. Curves were more skewed towards the two ends, e.g., at the 0-10% and 90-100% efficacy regions. Overall, the increase rates of the curves were negatively correlated to the SI and on-target rates, i.e., inhibitors with higher k and E 50(on) had a slower rate of increase of cytotoxicity, or vice versa. Apparently, the cytotoxicity values at EC 50 for the two on-target inhibitors NTZ and PTX were less than 5% (calculated values = 2.73% and 1.92%, respectively), while those for the four partially on-target inhibitors were much higher (i.e., between 14.40% and 20.56% for IVM, VCT and DXR, and up to 84.61% for MXT) (Fig 11B). Only MXT produced a convex curve due to the fact that its SI value was less than 1 (i.e., k = 0.21). In fact, SI was the determinant for the curve curvature (i.e., k = <1, 1, or >1 would produce convex, linear and concave curves, respectively).
Eq 5 also provided an opportunity to examine the relationship between an inhibitor's theoretical cytotoxicity and k (SI) and Hill slope (h) at specified efficacy (E i ). Based on the h values in this study (i.e., between 1.0 to 1.24), we plotted two sets of curves as examples to show the relationships between cytotoxicity, k (between 1 and 100) and h (between 0.8 and 2.0) at 50% and 90% efficacy concentrations (i.e., EC 50 and EC 90 ; the two commonly used parameters for drug efficacy) (Fig 11C and 11D). From the two plots, we observed that: 1) with fixed h and k values, an inhibitor's cytotoxicity was higher at higher efficacious concentrations (e.g., T (E90) > T (E50) ); 2) with fixed h and efficacy, all curves declined more sharply at lower k values, more apparently at k <10; and 3) with fixed k and efficacy, the cytotoxicity was negatively correlated with the h values (i.e., a lower toxicity at higher h value). Nonetheless, the plots gave us a new perspective to examine and compare the properties of inhibitors that might not be easily seen

PLOS NEGLECTED TROPICAL DISEASES
Validation and quantitation of on-target effect of anti-cryptosporidial inhibitors from the efficacy/cytotoxicity curves and commonly used parameters (e.g., EC 50 , TC 50 and SI values). The effect of Hill slope (h) on cytotoxicity was a novel observation, although it might be noticeable by careful comparison between the cytotoxicity and efficacy curves (Fig 6).

Discussion
The multidrug resistance protein 1 (MDR1; aka P-gp or ABCB1) is an ATP-dependent efflux pump with broad substrate specificity [16,17]. Both transient and stable overexpression of MDR1 in HCT-8 cells could increase the tolerance of cells to multiple compounds, but the two models have their own advantages and disadvantages. In the transient transfection model, host cells overexpressing MDR1 could be generated instantly to show >2-fold increases of drug tolerance to multiple compounds, but applicable only to MDR1 substrates [15]. In stable transfection model as demonstrated in this study, it might take months to first generate MDR1transgenic cells that only showed <2-fold increases of drug tolerance. However, drug tolerance in MDR1-transgenic cells could be quickly increased to much higher than 2-fold by applying continuous drug pressure. The notable advantage of the stable transfection model is the potential to generate novel resistance to non-MDR1 substrates in a reasonably short timeframe that is otherwise unachievable using transient transfection model. As exemplified in this study, NTZ-resistance was generated in MDR1-transgenic HCT-8 cells in three months (>2-fold in MDR1(NTZ) cell vs. WT or NC cells). This allows us to validate for the first time that NTZ, the only FDA-approved drug to treat human cryptosporidial infection, kills C. parvum by solely acting on the parasite target (100% on-target).
NTZ is a thiazolide compound with a relatively broad spectrum of activity against anaerobic bacteria and parasites by targeting pyruvate:ferredoxin/flavodoxin oxidoreductase (PFOR) involved in anaerobic metabolism. NTZ displays low micromolar inhibition constant (K i = 2 to 10 μM) on PFOR from the protozoan parasites Trichomonas vaginalis, Entamoeba histolytica and Giardia intestinalis and the bacterial pathogens Clostridium difficile, C. perfringens, Helicobacter pylori and Campylobacter jejuni [27]. However, the mode of action of NTZ against cryptosporidial infection is yet undefined. Cryptosporidium possesses a PFOR that is fused with an NADPH-cytochrome P450 reductase to form a unique bifunctional enzyme pyruvate:NADP + oxidoreductase (PNO) [28]. It should be noted that NTZ could also stimulate the host immune system [29][30][31]. This may partly explain why NTZ, which only displays lower micromolar in vitro anti-cryptosporidial efficacy (i.e., EC 50 from 1 to 5 μM as reported here and in other studies) [32,33], is only effective in immunocompetent patients. Nonetheless, the confirmation that NTZ is fully on-target justifies that the parasite PNO is worth to be investigated for developing more selective and effective anti-cryptosporidial drugs.
To our knowledge, this study represents the first attempt to develop algorithms for quantifying the proportional contributions of on-target and off-target effects to the overall anti-infective efficacy. Although the algorithms are developed under the experimental conditions used in this study, they are modifiable to suit other experimental settings. One noteworthy application of the algorithms is to evaluate and quantify the on/off-target effects in developing drugs targeting host cell targets localized inside the cells.
It is worth to clarify again that: 1) the MDR1-transfected cell models are applicable only to evaluating the on/off-target effects on epicellular pathogens (e.g., C. parvum) whose drug exposure would not be affected by MDR1 efflux. It is not applicable to pathogens residing in the host cytoplasm (e.g., Toxoplasma and Eimeria parasites) whose drug exposures would also be affected by MDR1 efflux; 2) The in vitro models and algorithms are used to evaluate whether an applicable inhibitor kills the parasite by acting fully or partially on the parasite target, rather than evaluating whether an inhibitor acts on a specific biochemical target; 3) The model can only be used to evaluate whether a compound acts on the parasite target, but unable to identify or confirm the target. It also cannot evaluate drug's effect on immune response.

Conclusions
We have developed an MDR1-transgenic cell-based model applicable to evaluating whether anti-cryptosporidial hits/leads kill the parasite by fully or partially targeting the parasite targets. The hits/leads can be either MDR1 substrates or non-MDR1 substrates. Using the model, we have validated that paclitaxel (PTX) and nitazoxanide (NTZ) kill C. parvum by fully acting on the parasite targets (100% on-target), while mitoxantrone (MTX), doxorubicin (DXR), vincristine (VCT) and ivermectin (IVM) kill the parasite by acting on both the parasite and host cell targets (partially on-target). We have also developed algorithms to quantify the percent contributions of on-and off-target effects to the observed anti-cryptosporidial efficacy in vitro, and to examine the relationships between anti-cryptosporidial efficacy (EC i ), cytotoxicity (TC i ), selectivity index (SI or k) and Hill slope (h).

In vitro culture of C. parvum and assays for anti-cryptosporidial efficacy and drug tolerance in host cell lines
A strain of C. parvum with subtype IIaA17G2R1 at the gp60 locus was propagated in the laboratory by infecting calves, from which oocysts were collected from feces and stored in PBS containing 200 U/mL penicillin and 0.2 mg/mL streptomycin at 4˚C until use. Prior to use, oocysts were purified by a sucrose/CsCl gradient centrifugation protocols, followed by a 5-min treatment of 10% house bleach in ice and extended washes with distilled water [34][35][36]. The viability of the oocysts was assessed by in vitro excystation in PBS containing 0.5% taurocholic acid sodium salt hydrate at 37˚C for 1 h, and only those with >85% excystation rates were used in experiments.
HCT-8 cells (National Collection of Authenticated Cell Cultures, Shanghai, China) was used as a parent wild-type (WT) cell line for generating MDR1-transgenic cells for assaying in vitro drug efficacy against C. parvum. Host cells were maintained in 25 cm 2 flasks containing RPMI-1640 medium, 10% fetal bovine serum (FBS) and 1.0% penicillin-streptomycin at 37˚C under 5% CO 2 atmosphere. Anti-cryptosporidial efficacy assay was performed using an established protocol [37,38]. Briefly, host cells including WT and its derived transgenic cell lines were seeded in 96-well plates overnight until~80% confluence and inoculated with C. parvum oocysts (2×10 4 per well). After 3 h incubation to allow excystation and invasion of C. parvum sporozoites, uninvaded parasites were removed by a medium exchange. Compounds at specified concentrations were added at this time point, and infected host cells were incubated for additional 41 h (total 44 h of infection). Cell lysates were prepared using an iScript qRT-PCR sample preparation reagent (50 μL/well) (Bio-Rad Labs, California, CA) [37,38]. Host cell drug tolerance to specified inhibitors was evaluated by 3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium (MTS) assay. WT and transgenic host cells were seeded in 96-well plates (10,000 cells/well) and cultured for 24 h, followed by the addition of specified compounds at serially diluted concentrations and continued culture for 41 h. After 3 washes with serum-free medium, MTS solution (Saint-Bio, Shanghai, China) was added into the plates (20 μL/well) and incubated at 37˚C for 2 h. Optical density at 490 nm (OD 490 ) was measured using a Synergy LX multi-mode reader (BioTek, Winooski, VT). Drug tolerance was indicated by half-maximal cytotoxic concentrations (TC 50 values) calculated by nonlinear regression using 4PL model. Selectivity index (SI) for each compound was determined by the ratio between TC 50 and EC 50 values (SI = TC 50 /EC 50 ) [39,40].

Development of stable transgenic cell lines and detection of MDR1 gene expression
A lentiviral expression vector system including pCDH-CMV-MDR1-EF1α-copGFP-T2Apuro lentivector, psPAX2 and pMD2G helper plasmids was used to generate MDR1-transgenic cell lines (Xiamen Anti-hela Biological Technology Co., Xiamen, China). Blank vector pCDH-CMV-MCS-EF1α-copGFP-T2A-puro was used as negative control (Fig 2A). Recombinant lentiviruses were prepared by co-transfection of 293T cells with the vectors/plasmids, followed by collection of lentiviruses in the supernatant and determination of the viral titers [41]. Parent HCT-8 cells (WT) were infected with the lentiviral preparations for 48 h, followed by selection with puromycin (4 μg/ml) for 7 days. The resulting transgenic cell lines were designated as HCT-8/MDR1 (or MDR1 for short) that overexpressed MDR1 and HCT-8/NC (or NC) that carrying negative control blank vector ( Table 1).
The morphology of WT, NC and MDR1 cells were examined by immunofluorescence assay (IFA), in which cells were cultured in 48-well plates containing glass coverslips coated with 0.1 mg/mL poly-L-lysin for 1 d. Cell monolayers were fixed in 4% paraformaldehyde for 10 min and permeabilized with 0.5% Triton X-100 in PBS for 5 min, followed by blocking with PBS buffer containing 3% BSA. MDR1 was detected by incubation with a rabbit monoclonal anti-MDR1 antibody (Cell Signaling Technology Co., Danvers, MA) (1:200 dilution) overnight at 4˚C and anti-rabbit antibody conjugated with Alexa Fluor 594. There were three washes with PBS for 5 min after each treatment step. The same IFA procedures was also used to detect whether there was any enrichment of MDR1 at the host cell-parasite interface in specified host cells infected with C. parvum for 24 h.
In qRT-PCR assay, total RNA was isolated from cells using Trizol RNA isolation kit (Takara, Shiga, Japan) and MDR1 and GAPDH transcripts were detected using a TransScript Green One-Step qRT-PCR SuperMix (TransGen Biotech, Beijing, China). The reactions (20 μL/reaction) contained 20 ng total RNA, 10 μL 2× SuperMix solution, 0.4 μL forward and reverse primers (10 μM each), 0.4 μL passive reference dye I, 0.4 μL TransScript One-Step RT/ RI Enzyme Mix and 5.4 μL RNase-free water, and were performed using a StepOnePlus realtime PCR system (Applied Biosystems, Waltham, MA). Primers for MDR1 and GAPDH were listed in S4 Table. Generation of cell lines with increased drug tolerance to MDR1 substrate paclitaxel (PTX) and non-substrate nitazoxanide (NTZ) Stable MDR1-transgenic cell line (MDR1 cells) was more resistant than WT and NC cell lines to five of the nine compounds tested in this study, but the increases were less than 2-fold (ranging from 1.54 to 1.76) ( Table 2), which were less ideal for evaluating on/off-target effects for these compounds and useless in evaluating other compounds. Since MDR1 was responsible for the development of multidrug resistance in cancer cells for a large number of therapeutics [42][43][44], we hypothesized that overexpression of MDR1 would make host cells more adaptable than WT and NC cells to the drug selection pressure for rapid increase of resistance to MDR1 substrates (e.g., PTX) and induction of resistance to non-substrates (e.g., NTZ). To test the hypothesis, WT, NC and MDR1 cells were subjected to selection by PTX and NTZ.
We employed a drug selection scheme similar to those reported by other investigators [45][46][47], in which cells were subjected to multiple rounds of drug selection with incrementally increased drug concentrations, each round containing 2-3 cycles of 2-d drug selection at 80% inhibition concentrations followed by 3-5 d of drug withdrawal to allow the growth of host cells to near confluence (see S2 Table for detailed drug selection design). More specifically, WT, NC and MDR1 cells were cultured in 6-well plates (2×10 5 cells/well) to confluence and incubated with PTX at 0.75 μM (WT and NC cells) or 1.5 μM (MDR1 cells) or NTZ at 3.0 μM (WT, NC and MDR1 cells) for 2 d (the drug concentrations were near their TC 80 values determined by 48-h cytotoxicity assay). Surviving cells were allowed to recover in drug-free medium for 3-5 d to near confluence (round 1). The selection/recovery cycle were repeated once (round 2). Cells were then subjected to a serial new rounds of selection/recovery cycles with incrementally increased drug concentrations until MDR1 cells could grow normally in the presence of 7.61 μM of PTX or 15.20 μM of NTZ (round 11). At this time point, WT and NC cells could grow normally in the presence of 1.70 μM PTX or 10.13 μM NTZ (S2 Table). Finally, all cells were cultured at the final selection concentrations of PTX or NTZ for �7 d, followed by culture in drug-free medium for 14 d. At this time point, cells were used for cytotoxicity and efficacy assays or cryopreserved in a liquid nitrogen tank. The resulting cell lines after PTX or NTZ selection were designated as WT(PTX), NC(PTX) and MDR1(PTX), or WT (NTZ), NC(NTZ) and MDR1(NTZ), respectively (Tables 1 and S2).

Mathematical models for quantitative estimation of relative contributions from the on-and off-target effects to the anti-cryptosporidial efficacy
Model based on EC 50 and TC 50 values. Let us denote E on and E off as the on-and off-target rates, and E obs as the observed as anti-parasitic efficacy, representing the proportions or precents of on/off-target effects contributing to the observed anti-cryptosporidial efficacy. The observed anti-parasitic efficacy (E obs ) is the sum of E on and E off that was set to 100%: Under the condition that the drug tolerance is significantly increased in the drug-resistant cell line (e.g., >2-fold increase between TC 50(MDR1) and TC 50(NC) ), where MDR1 represents MDR1-derived cell lines such as MDR1(PTX) and MDR1(NTZ) cells, the relative contributions of E on and E off to E obs can be indicated by whether, and how much, the anti-parasitic efficacy is also increased proportionally. More specifically, we may estimate the percent contribution of E off to E obs by calculating whether and how the relative increase of anti-parasitic efficacy (RI EC50 ) is proportionally correlated to the relative increase of drug tolerance between (RI TC50 ), or the ratio between RI EC50 and RI TC50 using the following equations: Eq 9 can be rearranged to obtain Eq 1 described in the Results section. Based on Eq 6, we also obtain Eq 2 described in the Results section.
Expansion of the model to the whole efficacy range from EC 0 to EC 100 . Dose-dependent efficacy and cytotoxicity kinetic curves generally follow a 4-parameter logistic (4PL) sigmoidal model [48]: where Y is the response (theoretically ranging from 0 to 1 probability values) and X is the drug concentration. E min and E max are the lower and upper plateaus of the curve (also termed Bottom and Top). The parameter h is the slope factor of the curve (Hill slope). The E 50 (= either EC 50 or TC 50 ) is the concentration to achieve the midway response between E min and E max . In a drug efficacy assay based on quantitation of relative parasite loads by qRT-PCR and a cytotoxicity test based on colorimetric or fluorescent assay, the response (Y) can be converted to the percent inhibition on the parasite or on host cell, in which E min is normalized to zero (i.e., the response to diluent in the negative controls). Eq 10 is then simplified to: Ideally, the parameter E max value is 1 (100%), by which E 50 (solved from the equation) is the inhibitor's concentration that truly achieves 50% inhibition, referred to as "absolute EC 50 or TC 50 " [48]. However, E max might not reach 100% in many assays, in which E 50 solved from Eq 10 is relative to the upper plateau, referred to as "relative EC 50 or TC 50 " (Note: this study reported relative EC 50 or TC 50 values).
Derivation of equations to visualize the relationship between selectivity index (SI) and cytotoxicity over a drug's efficacious concentrations. The principal here is to plot the cytotoxicity (inhibition rates on host cell growth; denoted by Y TC ) of a specified inhibitor against the concentrations of the inhibitor over the range showing anti-cryptosporidial efficacy (denoted by Y EC ) in WT cells. Based on the 4PL model (Eq 10), we have: where X ECi is the concentration of the inhibitor at anti-cryptosporidial efficacy EC. Y TC(Ei) is the cytotoxicity rate of the inhibitor at the concentration EC i (i = 0 to 100%). Since SI is defined by the ratio between TC 50 and EC 50 , we have: where k is SI for simplicity. The parameter k in Eq 13 can be introduced into Eq 12: The anti-cryptosporidial efficacy, denoted by Y ECi here for clarity, can be introduced into Eq 14 to replace X ECi based on the 4PL model again: which can be derived to: After placing Eq 16 into Eq 14 and some derivations, we obtain the following simplified equation to define Y TC(ECi) as the function of Y ECi , k and h, i.e., Eq 5 in the Results section.
In Eq 5, the h values in both efficacy and cytotoxicity curves are assumed to be the same after considering that a specified inhibitor would likely act on the same or similar targets in the parasite and the host cells. This assumption is also supported by the h values for the six inhibitors obtained in this study, in which the h values range from 1.0 to 1.24 and differ by 0.40% to 2.87% between efficacy and cytotoxicity curve pairs (S6 Table).

Data analysis and statistics
At least two independent experiments were conducted for each experiment condition. Each experiment contained minimal 2 or biological replicates for experimental groups or negative controls, respectively. In qRT-PCR assay used 2 or 3 technical replicates. In vitro efficacy and cytotoxicity data were analyzed using Prism (v9.0 or higher; GraphPad, San Diego, CA) using a 4-parameter logistic model. Statistical significances were evaluated by two-way analysis of variance (ANOVA) and Holm-Šídák multiple t-test between group pairs [40].