A computer-assisted systematic search for melatonin derivatives with high potential as antioxidants

A systematic rational search for newly designed melatonin derivatives, was performed using a computer-assisted protocol. A total of 116 derivatives were generated by adding functional groups (i.e., -OH, -NH2, -SH and -COOH) to the melatonin structure. A selection score (S ) was built to sample the search space, simultaneously considering ADME (absorption, distribution, metabolism, excretion) properties, toxicity and manufacturability (i.e., synthetic accessibility). The search characterized the whole set of designed melatonin derivatives and allowed the selection of a reduced subset of 20 melatonin derivatives that are expected to be the most promising, regarding drug-like behavior. For this subset, several reactivity indices were estimated, as well as their pKa values. According to the gathered data, 5 melatonin derivatives have been identified as the most likely candidates to act as chemical antioxidant (directly scavenging free radicals, by electron transfer and/or H transfer). All of them are predicted to be better for that purpose than melatonin itself or trolox (a water-soluble vitamin E analog). The findings from this work are expected to motivate further investigations on these molecules, using both theoretical and experimental approaches.


INTRODUCTION
Oxidative stress (OS) and chemical agents against it are currently the focus of numerous investigations.The interest in both, oxidative damage to biomolecules and its prevention by chemical species (usually referred to as antioxidants) is well justified.There is ample evidence that OS compromises human health leading to various diseases such as different kinds of cancers, cardiovascular and neurodegenerative disorders (1)(2)(3)(4)(5)(6).At the same time, there is also abundant data supporting the protective effects of varied chemicals against OS and associated diseases.Although naturally occurring molecules have been the most widely studied in this context, there is an increased interest in designing molecules capable of exerting such protection.

Scheme 1. Melatonin structure
A rational way to perform such a systematic search is starting by using computer-assisted protocols, which significantly reduce costs and expedite the process.Thus, that is the strategy followed here.The search mainly consisted on modifying the melatonin structure through the inclusion of different functional groups, specially chosen to potentiate versatile antioxidant activity, into all the available positions of the indole ring.Then using absorption, distribution, metabolism, excretion and toxicity (ADMET) properties a subset of the most promising candidates was chosen.For that reduced space, several reactivity indices were estimated using electronic structure calculations within the framework of the Density Functional Theory (DFT).They were used to select a few molecules, among the generated pool, which are novel molecules proposed for the first time as melatonin derivatives with high probabilities of being excellent multipurpose antioxidants.The presented results are expected to motivate further investigations on these molecules, using both theoretical and experimental approaches.

Physicochemical parameters
Several physicochemical parameters that are considered relevant for absorption, distribution, metabolism and excretion (ADME) properties were estimated for all the designed melatonin derivatives, using Molinspiration Property Calculation Service (www.munilnsiration.com)and DruLiTo software (http://www.niper.gov.in/pi_dev_tools/DruLiToWeb/DurLiTo_index.html).The estimated parameters are those necessary to investigate if the designed melatonin derivatives satisfy Lipinski's rule of five (88), Ghose's rules (89) and the Veber criteria (90).According to the Lipinski's rule, orally active drugs should have no more than 5 hydrogen bond donors (HB D ), no more than 10 (5x2) hydrogen bond acceptors (HB A ), a molecular weight (MW) under 500 (5x100) g/mol, and an octanol/water partition coefficient (logP) lower than 5. Compounds violating more than one of these rules may have difficulties with bioavailability.According to Ghose's rules, for preventing orally active drugs from having low permeation or absorption issues, they must have logP values ranging from -0.4 to 5.6, molar refractivity ( M R) from 40 to 130, MW from 160 to 480, and a number of non-hydrogen atoms ( X At) from 20 to 70.On the other hand, according to the Veber criteria, chemicals with 10 or fewer rotatable bonds (RB) and a polar surface area (PSA)  140 Å 2 (or 12 or fewer H-bond donors and acceptors) would have better chances of good oral bioavailability.
At this point it seems worthwhile to emphasize that all these criteria are empirical and intended to be general guidelines, not strict regulations.In fact, viable drugs must also fulfil other important requirements, including manufacturability and safety (91).Therefore, these features were also investigated here.

Toxicity
Two toxicity descriptors were used in this work, namely: -LD50: amount of the investigated chemical per body weight (mg/kg) leading to 50% death of rats, after oral ingestion.
-M: usually referred to as Ames mutagenicity.A chemical is positive if it increases (in a reproducible, dose-related manner) the number of revertant colonies in one or more strains of Salmonella.To calculate both indices, the Toxicity Estimation Software Tool (T.E.S.T.), version 4.1, was used.This software makes predictions based on quantitative structure activity relationships (QSAR), which are intended for screening untested compounds.The LD50 and M descriptors were computed with the consensus method, which makes predictions as the average of the toxicities predicted from several QSAR methodologies, considering the applicability domain of each of them (92).There is a general agreement that the consensus method usually provides higher accuracy and coverage than other methodologies.

Synthetic accessibility
The synthetic accessibility (SA) of the designed compounds was estimated with the SYLVIA-XT 1.4 program (Molecular Networks, Erlangen, Germany) (93).It uses several contributing criteria, including the similarity to commercially available compounds, the complexity of the molecular structure and of ring systems, and the number of stereo centers.These criteria are scaled and weighted to provide a value between 1 and 10.The larger the value, the more difficult to synthesize is the compound.The SYLVIA program has been validated for ranking virtual compounds during drug discovery processes (94).

Electronic calculations
In addition to the physicochemical parameters, toxicity and synthetic accessibility, there are specific reactivity indices that are expected to indicate antioxidant behavior.To estimate them, electronic calculations are necessary.Such calculations were all performed with Gaussian 09 package of programs (95).Local minima were identified by the absence of imaginary frequencies.Unrestricted calculations were used for open shell systems.Geometry optimizations and frequency calculations were carried out using the Density Functional Theory (DFT), in particular the M05-2X functional (96) in conjunction with 6-311+G(d,p) basis set and the solvation model density (SMD) (97) using water to mimic a polar environment.M05-2X is a global hybrid exchangecorrelation GGA functional designed for thermochemistry, kinetics and noncovalent interactions (96), it has also been recommended for calculating reaction energies involving free radicals (98).Furthermore, the M05-2X functional has been widely used for estimating the pKa values, the bonding dissociation energies and, in general, the free radical scavenging activity of several antioxidant molecules (99)(100)(101)(102)(103)(104)(105)(106)(107)(108)(109)(110)(111).SMD is considered a universal solvation model, due to its applicability to any charged or uncharged solute in any solvent or liquid medium for which a few key descriptors are known (97).In all cases, the absence of imaginary frequencies was identified to assure that structures found were local minima.

Reactivity indices
Several global reactivity indices were estimated to analyze the chemical behavior of the designed melatonin derivatives (Table 1).IE and EA values were calculated in the framework of the electron propagator theory (EPT) (112,113), because this approach usually produces values closer to those derived from experiments than other strategies.In particular, the partial third-order quasiparticle theory (P3) (114) was chosen because it has lower mean errors than other methods (115).
Measures the energy necessary for breaking donor(D)-H bonds.The lower the BDE, the higher the antioxidant activity, via H transfer.

*
The expressions for ,  and  correspond to the commonly used finite difference approximation.
Since many of other calculated reactivity indices are estimated from IE and EA values, for reliability purposes it is important that these two magnitudes are as accurate as possible.However, it is important to keep in mind that for the EPT approximations (including P3) to be valid, the values of the pole strength (PS) should be larger than 0.80-0.85(122,123).This has been confirmed to be the case for all the calculations performed in this work (Table S1).
In the BDE case, all sites that are likely to act as H donors were taken into account (Scheme 2).They correspond to those already present in the melatonin's framework (a to d), and also the new possibilities arising from incorporating functional groups (-OH, -NH2, -SH and -COOH) in sites R1 to R4.

Reference set
To put into perspective the data obtained for the newly designed melatonin derivatives, a reference set of molecules was used.It consists of 35 chemicals already clinically used as neuroprotectors.Their names and structures are shown in Table 2, while their properties are provided as Supporting Information (Tables S2 and S3).Neuroprotectors were used to construct this set because, as previously mentioned, neurodegenerative disorders are among the diseases that are attributed (at least partially) to oxidative stress.Thus, efficient antioxidants might also play a protective role in this context.However, it should be noted that the ways of action, identified so far, for the neuroprotectors in Table 2 do not necessary involves antioxidant protection.

pKa calculations
Acid constants, expressed as pKa, were calculated for the subset of the most promising melatonin derivatives.This is an important feature of molecules intended to be used as the medicinal drugs, since drug absorption across the gastrointestinal lining mainly takes place via passive diffusion (175).This is ruled by passing across lipid membranes; thus, neutral species are more likely to do so than charged ones.The proportion of neutral vs. charged species for molecules with acid-base bahaviour is ruled by the pKa-pH relationship.It is evident that for newly designed chemicals, pKa values have not been experimentally estimated.Fortunately, theoretical predictions can be used instead.Although this is usually a very challenging task, there is at least one approach that is easy to use and reliable enough.It is usually referred to as the fitted parameters approach (FPA).It involves using experimental pKa values of a set of small reference molecules to obtain two parameters (m and C0) from linear fittings: In this equation ΔGBA represents the difference in Gibbs energy between the conjugated base and the corresponding acid (Gcalc(A−) − Gcalc(HA)).The m and C0 parameters are currently available at numerous levels of theory, for phenols, amines, carboxylic acids and thiols (100, 101), i.e., the functional groups used here to construct melatonin derivatives.The values of m and C0 at the level of calculation used here are shown in Table 3.It has been demonstrated that for all of them, the pKa values calculated with the FPA approach deviate from experiments by less 0.5 pKa units, in terms of mean unsigned errors.Therefore, that is the approach used in this work to estimate the pKa values of the designed melatonin derivatives.

RESULTS
Computer-based design of drug-like molecules is a challenging task.There are at least three important issues that should be properly addressed: (i) building the candidate species; (ii) sampling the search space in an efficient way; and (iii) evaluating their potential for the intended purpose (176).The detailed criteria on each of them are provided next.

Building melatonin derivatives
The melatonin derivatives designed here were conceived to promote multifunctional antioxidant activity.Antioxidants can be classified as primary (Type I, or chain breaking), secondary (Type II, or preventive) and tertiary antioxidants (Type III, or repairing) (177,178).Type I are molecules that directly react with free radicals, thus they are usually referred to as free radical scavengers.Type II de-active free radicals otherwise, for example acting as • OHinactivating ligand (OIL) (179,180) through metal chelation.Type III restore oxidative damaged biomolecules to their original structure (mainly through H or electron transfer).Multifunctional antioxidants are molecules that exhibit more than one of these types of protection.
To take advantage of the desirable features of melatonin as much as possible, the new molecules were built by moderate structural modifications.To that purpose four functional groups (i.e., -OH, -NH2, -SH and -COOH) were used.They were chosen considering that, because of their chemical nature, they are expected to play, at least, one of the following roles: − They can influence the acid-base behavior, thus modulating the proportion of neutral species at specific pH values, which is important for drugs passing across lipid barriers via passive diffusion (175).− They may contribute to increase free radical scavenging activity (i.e., antioxidant activity, types I and III) via H donation, or electron donation.− They may contribute to increase metal chelating capability (i.e., antioxidant activity type II, OIL behavior).
Although the thiol group might be considered as an unwanted functionality in drug discovery because of its high reactivity (181), it has been included in this investigation.The reason is just that reactivity, since the compounds designed in this work mainly act as chemical (antioxidant) agents.In addition, the thiol functionality is expected to increase metal chelating abilities and has been identified as crucial for the free radical scavenging activity of widely recognized antioxidants such as glutathione.(182,183) Placing the above-mentioned functional groups in all the available positions of the indole ring (R1 to R4, Scheme 2), 116 melatonin derivatives were constructed (Table S4, Supporting Information).Sixteen of them with only one functional group (all possible species within the used substitution scheme), 96 of them with two functional groups (using any possible combination) and 4 with three functional groups.The latter were built from the most promising bi-functionalized species (details on this are provided in the next section).Melatonin itself was included in the set of molecules for comparison purposes.

Sampling the search space, using extended ADME properties
As previously mentioned, in addition to ADME (absorption, distribution, metabolism, excretion) properties, toxicity (T) and manufacturability (i.e., synthetic accessibility, SA) are also important features for viable medicinal drugs.Therefore, a selection score (S S ) including all these aspects has been constructed.It was used to characterize the whole set of designed melatonin derivatives (116 species) and choosing a reduced subset of 20 molecules that are expected to be the most promising, regarding drug-like behavior.
The selection score was constructed in such a way that the higher the value of S S the more likely that a melatonin derivative (dM) has a drug-like behavior:  In the particular case of S X A , only the upper limit of the Ghose's rules was used because many of the currently used drugs have S X A < 20.For the reference set used here 16 out of 35, i.e., 45%, including the parent molecule (melatonin).They are acetylcarnitine (S The results for S S are shown in Fig. 1, and the values of the individual properties used to calculate S S are reported in Table S5 (Supporting Information).In general, in this figure the molecules with higher S S values are expected to have lower toxicity, better synthetic accessibility, and better ADME properties.Based on this criterion, 20 melatonin derivatives (Scheme 3) were selected for the next stage of the investigation, i.e., to evaluate their potential as antioxidants based on reactivity indices.However, the selection score is rather general and might mask a particular case with high S S but a failure for a particular property.Therefore, in addition to S S , exclusion scores (S E ) were also used, inspired by previous proposals (184,185), to double-check if any molecule in the selected subset significantly deviates (in any of its properties) from the average value of the reference set (described in section 2.6).Four exclusion scores were used for that purpose.The first one, here referred to as S E,ADME2 is identical to that previously proposed (184,185) The third one, S E,ADMET , includes two other terms, related to toxicity; and the fourth (S E,ADMETSA ) also includes a term for synthetic accessibility: These exclusion scores (Fig. 2 and Table S6) measure deviations from the average values of already used oral drugs.S E,ADME2 values have been reported to be 1.5 and 1.2, when considering 152 and 1791 oral drugs, respectively (184,185).For the reference set used here (35 molecules), average S E,ADME2 = 1.3, with individual values ranging from 0.15 to 4.49.On the other hand, for the 20 melatonin derivatives with highest selection scores, average S E,ADME2 = 0.7 with individual values ranging from 0.4 to 1.1 (Table S6, Supporting Information).Therefore, according to S E,ADME2 they seem to be suitable as drug-like molecules.The S E,ADME2 score, however, only accounts for 2 of the analyzed properties and has rather similar values for all the analyzed molecules (Fig. 2).That is the reason the other 3 elimination scores were also implemented.They account for more properties and span the scales.Average S E,ADME8 = 4.1, with individual values ranging from 1.4 to 7.2; S E,ADMET = 6.5, with individual values ranging from 2.7 to 14.7; and S E,ADMETSA = 6.8, with individual values ranging from 3.1 to 14.8.It is important to consider, though, that high values of these scores may arise from either worse or better behavior (as oral drug-like species) than the average of the reference drugs.

Figure 2. Elimination score (S E ) for the most promising melatonin derivatives, according to S S
. Columns are divided to show the influence of the new contributions included in each score, with respect to the previous one.

Evaluating antioxidant likeliness, using reactivity indices
Reactivity indices were estimated for the subset of molecules chosen in the previous section.Since their involvement in acid-base equilibria may be relevant to the target behavior, their pKa values were estimated (Table 4) and the associated deprotonation routes elucidated (Figure S1, Supporting Information).The corresponding distribution diagrams are also provided as Supporting Information (Figure S2), while the molar fractions ( M f) of the different acid-base species, at physiological pH, are reported in Table 4.The reactivity indices for the acid-base species with non-negligible population ( M f > 0.1%), at pH=7.4, are reported in Table 5.

Elimination scores
Some discussions on the different contributions to the elimination scores seem worthwhile.Figure 2 clearly shows that synthetic accessibility and the two properties included in S E,ADME2 (logP and MW) both have rather small (and similar contributions) to the deviations from the reference molecules.On the contrary, the new properties included in S E,ADME8 and the toxicity indices have the largest contribution to S E,ADMETSA .
To analyze the individual contributions of the different properties to the S E,ADMETSA elimination score, a more detailed plot was constructed (Fig. 3).It was found that the largest deviations from the average value of the reference set of molecules arise from LD50, M, PSA, HB D and HB A .Regarding LD50, the derivatives deviating the most from the average (dM-104, dM-96, dM-72, dm-112, dM-96 and dM-100) are less toxic to rats than the reference average (LD50 = 960.8),with values 6960.5, 4733.5, 2892.9, 2861.8,2399.7 and 2303.2,respectively.Thus, these large deviations mean a more desirable behavior than that of the references and, consequently, these derivatives were not excluded from the subset selected as the most promising, based on ADMETSA properties.
A similar trend was found for the Ames mutagenicity, i.e., the compounds predicted as the least mutagenic are just those that deviate the most from the reference set (M = 0.41).They are dM-6, dM-7, dM-8, dM-64, dM-38 and dM-61, all with M  0.02.Thus, it is important not just to identify the designed compounds with the largest deviation from the reference set, but also what causes such deviations.Otherwise, good candidates might be eliminated for the wrong reasons.Based on what has been discussed in this section, none of the 20 melatonin derivatives identified as the most likely candidates, based on the selection score, was eliminated after further screening using the elimination scores.Accordingly, reactivity indices were estimated and analyzed for all of them.

Antioxidant-like behavior
The reactivity indices estimated in this work are expected to help predicting antioxidant behavior, via free radical scavenging activity, provided that such activity involves single electron transfer (SET) and/or formal hydrogen atom transfer (HAT) mechanisms.There is graphical tool, known as the full electron donator acceptor map (FEDAM) (186,187) that allows predicting, quickly and qualitatively, the direction of the electron flow in SET reactions (Fig. 4).It is based on the precept, that in SET reactions between two chemical species, that with the lower IE would be the electron donor, and that with the higher IE would be the electron acceptor.Thus, the FEDAM tool was used for the subset of melatonin derivatives selected in section 3.2 (Fig 5).Some reactive oxygen species (ROS) were also included in the map to facilitate the analyses, as well as the parent molecule for comparison purposes.The different acid-base species of the newly designed derivatives were explicitly included in Fig 5, since deprotonation is expected to play an important role on SET feasibility.In fact, this is clearly shown in the figure, where the acid-base species are located in a cluster-like way depending on their charge.According to this map, all of them except the protonated ones, are expected to donate one electron to ROS.Thus, the designed melatonin derivatives are predicted to behave as ROS scavengers, at least via SET.It is also interesting to note that some of them are also expected to be slightly more efficient for that purpose than melatonin itself.The trend obtained from the FEDAM is in line with that of the chemical potential (μ, Table 5).This is a logical result, since electrons are expected to flow from regions of high  to regions of low .Moreover  has a linear dependence with IE (Figure S3, Supporting Information).

Figure 5. FEDAM (Full Electron Donor Acceptor Map) for melatonin derivatives
On the contrary, electrophilicity () and electrodonating power ( − ) do not have linear dependences with IE (Figure S4, Supporting Information).In fact, for species with very low IE, the  and  − values increase.This behavior resembles to some extent that of the inverted region of the Marcus parabola, which indicates that for Gibbs (ΔG) energy of reaction much lower than minus the reorganization energy, reaction barriers increase as ΔG becomes more negative (188)(189)(190).Albeit this is a counterintuitive behavior, it suggests that species with very low IE are not expected to be very efficient as free radical scavengers acting as electron donors in SET reactions.That would be the case for the dianionic species of the investigated melatonin derivatives.However, such a behavior would need further confirmation.
The acid-base species that, based on  values, seem to be the most promising for deactivating free radicals via SET, acting as electron donors, are the mono-anions (Figure 6).They have the lowest values of  for each derivative; and as previously mentioned, in a chemical reaction involving two reactants that with the lower  is expected to act as the nucleophile (116,117).The mono-anionic species analyzed in Figure 5, all have lower  values than any of the investigated free radicals, thus they are expected to be efficient for scavenging free radicals via electron transfer.In addition, all the mono-anions of the subset of melatonin derivatives with better druglike behavior have similar electrophilicity.Thus, they are probably similarly efficient as free radical scavengers via SET.To tell them apart another criterion is necessary.Another graphical tool has been designed, simultaneously accounting for likeliness as electron donors (SET reactions) and H donors (formal HAT reactions).Here it is referred to as the electron and hydrogen donating ability map for antioxidants (eH-DAMA), and simultaneously includes electrodonating power ( − , accounting for SET feasibility) and bond dissociation energies (BDE, accounting for HAT feasibility).The BDE values for each species are provided as Supporting Information (Table S7). Figure 7 shows this map for melatonin derivatives, it also includes the parent molecule and trolox for comparison purposes and the H2O2/O2 •− pair as the potential oxidant target.The later has been chosen because it is usually harder to scavenge than other reactive oxygen species, and because it has been previously found that melatonin itself is not very efficient for chemically deactivating this radical.(191) The chemical species with lower  − are expected to be particularly efficient for scavenging free radicals acting as electron donors via SET, while the species with lower BDE are expected to be particularly efficient for scavenging free radicals acting as H donors via formal HAT.Therefore, the species located at the bottom and left side of the eH-DAMA are likely to act both ways, i.e., they are particularly interesting as antioxidants.This region has been highlighted in Figure 6, and shows that the species fulfilling both criteria are all mono-anions, including that of trolox.All the species in the target region are predicted to have similar electron-donor capability, but rather different H-donating power.Based on the data summarized in Fig. 7, derivatives dM-34, dM-115, dM-38, dM-114, dM-61 and dM-94 are predicted to be better hydroperoxyl scavengers than trolox, and also than the parent molecule.On the other hand, dM-64 should be better than melatonin for that purpose, but its antioxidant activity is not expected to surpass that of trolox.Since their most active species are expected to be the mono-anions, their molecular fractions are relevant in this context.For all these derivatives, mono-anions are the most abundant acid-base species at physiological pH (Table 4).However, to cross biological barriers it is also important that the molar fractions of the neutral species are not negligible.Most of above mentioned derivatives also fulfill this requirement.The only exception is dM-114.Therefore, the melatonin derivatives proposed as the most promising antioxidants are dM-34, dM-115, dM-38, dM-61 and dM-94 (in that order).Further, more detailed and quantitative, investigations on their antioxidant action are still needed and highly desirable, to confirm or refute the proposal from this work.

Other considerations
At this point it seems worthwhile to make some comments regarding the limitations of the present study, and the necessity of further investigations on the topic of this investigation using both theoretical and experimental approaches.Antioxidant protection is a complex process that involves different chemical and non-chemical routes, thus there are several aspects on the behavior of the designed compounds, in biological systems that need to be further explored.Some of them are: 1. Quantitative estimations of the kinetics involved in the free radical scavenging activity of the designed compounds.2. Identification of the primary products yielded from their reactions with free radicals.3. Investigations on the possible capability of the designed compounds to chelate redox metals and act as OH inactivated ligands.4. The possibility of pro-oxidant effects. 5. Antioxidant protection arising from melatonergic signaling, including the evaluation of the designed compounds as ligands to melatonin receptors.6. Enzymatic metabolism of the designed compounds, and characterization of the corresponding metabolites.7. Experimental assessments of toxicity of both the proposed compounds and their metabolites.8.The possibility that the melatonin derivatives identified here as the most promising antioxidants may undergo transnitrosation reactions, since they all contain a sulfhydryl group.As it is evident from these points, it is unfeasible to carry out all the necessary research on the designed compounds in a single investigation.Hopefully, the results from this work are promising enough to motivate further researches on these compounds, and help obtained a more complete picture regarding their possible use as antioxidant agents.

CONCLUSIONS
A systematic rational search for newly designed melatonin derivatives, performed using a computer-assisted protocol, is presented.A total of 116 derivatives were generated by adding functional groups (i.e., -OH, -NH2, -SH and -COOH) to the melatonin structure; 16 with only one functional group (all possible species within the used substitution scheme), 96 with two functional groups (using any possible combination) and 4 with three functional groups.
A selection score (S S ) was built to sample the search space, simultaneously considering ADME (absorption, distribution, metabolism, excretion) properties, toxicity and manufacturability (i.e., synthetic accessibility).It was used to characterize the whole set of designed melatonin derivatives and allowed the selection of a reduced subset of 20 melatonin derivatives that are expected to be the most promising, regarding drug-like behavior.
For this subset, several reactivity indices were estimated, as well as their pKa values.These indices account for electron and H donor capabilities; thus, they are expected to reflect free radical scavenging behavior through single electron transfer (SET) and formal hydrogen transfer (HAT) mechanisms.According to the gathered data, 5 melatonin derivatives have been identified as the most likely candidates to act as chemical antioxidant (by directly scavenging free radicals).They are dM-34, dM-115, dM-38, dM-61 and dM-94 (Scheme 3), in that order.All of them are predicted to be better for that purpose than melatonin itself and trolox.The findings from this work are expected to motivate further investigations on these molecules, using both theoretical and experimental approaches.

AUTHORSHIP
Miguel Reina, Romina Castañeda-Arriaga, Adriana Pérez-González and Eduardo Gabriel Guzmán-López: Contributed to acquisition of data, data analysis and interpretation, drafting of the manuscript and approval of the article.Dun Xian Tan and Russel J. Reiter: Contributed to the data analysis and interpretation, drafting of the manuscript, critical revision of the manuscript and approval of the article.Annia Galano: Contributed to the conception and design of the investigation, acquisition of data, data analysis and interpretation, drafting of the manuscript, critical revision of the manuscript and approval of the article.
to reports of their use as neuroprotectors.

Figure 1 .
Figure 1.Selection score (S S ) for the melatonin derivatives designed in this work.Vertical lines mark the arithmetic mean of the reference set (red) and the value for the parent molecule (melatonin, green).

Figure 3 .
Figure 3. Individual contributions to the elimination score (S E ), for the most promising melatonin derivatives.On the contrary, for the other indices (PSA, HB D and HB A ) larger S E values actually mean that the behavior of the investigated derivatives approaches the upper limits of the recommended range for the investigated parameters, although they still fulfil Lipinski's and Ghose's rules, as well as the Veber criteria.Regarding PSA the selected derivatives deviating the most from the reference set are dM-92, dM-96, dM-100, dM-104, dM-112 and dM-72.Their PSA values range from 128.7 to 117.5 (i.e., below the Veber's threshold, 140 Å 2 ).The largest deviations for HB D correspond to dM-72, dM-81, dM-10, dM-11, dM-92 and dM-96, with HB D = 5 or 4; and for HB A correspond to dM-92, dM-96, dM-100, dM-104, dM-112 and dM-72, with HB D = 8 or 7. Thus, they do not constitute violations of the Lipinski's rule.Based on what has been discussed in this section, none of the 20 melatonin derivatives identified as the most likely candidates, based on the selection score, was eliminated after further screening using the elimination scores.Accordingly, reactivity indices were estimated and analyzed for all of them.

Figure 6 .
Figure 6.Electrophilicity of the acid-base species of melatonin derivatives.

Figure 7 .
Figure 7.The electron and hydrogen donating ability map for antioxidants (eH-DAMA), including the acid-base species of melatonin derivatives, the parent molecule, trolox and the oxidant the H2O2/O2 •− pair