Mutagenicity of N-acyloxy-N-alkoxyamides – QSAR determination of factors controlling activity

Keywords: acyloxy alkoxyamides, ames mutagenicity, anomeric amides, bilinear QSAR, deamination, direct‐acting mutagenicity, DNA binding, groove binding, HERON reactions, intercalation, linear QSAR, mutagenic amides, nitrogen deletion, PAH, pyramidal amides, QSAR, quantitative structure‐activity relationship, skeletal editing, TA98, TA100​​​​​​.

N-Acyloxy-N-alkoxyamides (NAAs), 2, are members of the unusual class of anomeric amides 1, which we defined as amides bearing two electronegative heteroatoms bonded to the amide nitrogen (Chart 1).^[1] These have unusual properties. Firstly, through Bent’s Rule,^[2,3] the electron demand of the electronegative substituents in 1 imparts more p character to the nitrogen hybrid orbitals bonded to them, with a concomitant increase in s character in the nitrogen lone pair orbital. Hence, these amides possess distinctly pyramidal nitrogens.^[1,4–13] This, combined with the lower-energy lone pair, results in diminished amide resonance. Secondly, anomeric interactions between the nitrogen substituents results in unusual chemistry at the amide nitrogen.

Chart 1. 

In the case of N-acyloxy-N-alkoxyamides 2, the nitrogen is bonded to two oxygen atoms and, as a result, they are strongly pyramidal at nitrogen, evidence for which stems from X-ray structures^[7] as well as computational data.^[8,9,13] Several NAAs we have made are the most pyramidal amides known.^[7] The combined electronegativity results in a very significant reduction in resonance, estimated computationally to be ~50% that of the iconic standard amide N,N-dimethylacetamide, and manifests itself in longer amide bonds.^[7,8,13] These structural properties are borne out spectroscopically, most notably by infrared spectroscopy where the amide carbonyl stretch frequencies are considerably higher than found in primary, secondary and tertiary amides. Typically, these are in the range of 1718–1742 cm⁻¹ on average, some 50–54 cm⁻¹ higher than their precursor hydroxamic esters.^[8]

Although two anomeric interactions are possible in NAAs (Fig. 1), because the acyloxyl group would be more electron-demanding, the more favourable anomeric interaction at the amide nitrogen involves donation of the alkoxyl oxygen lone pair into the σ* orbital between nitrogen and the acyloxyl oxygen (n_OR–σ*_NOAcyl) for which there is structural and theoretical evidence.^[7,8,13] The destablisation of the bond to the acyloxyl group renders NAAs susceptible to both unimolecular and bimolecular reactions at nitrogen. Under conditions of acid catalysis, they react by the unusual A_Al1 mechanism, producing nitrenium ion intermediates.^[14,15] However, they undergo S_N2 reactions at nitrogen with a range of nucleophiles including amines (path i, Scheme 1),^[16–20] azides (path ii, Scheme 1),^[21] hydroxides (path iii, Scheme 1)^[22] and thiols (path iv, Scheme 1),^[23] each process producing new anomeric amide intermediates that react further. All these processes occur at ambient temperatures. At higher temperatures, NAAs undergo the novel HERON reaction, producing anhydrides and alkoxynitrenes in competition with homolysis (path vi and vii, Scheme 1).^[24]

Fig. 1.  Two possible anomeric interactions in anomeric amides: (a) n_OR–σ*_NOAcyl, and (b) n_OAcyl–σ*_NOR.

Scheme 1.  Reactions of N-acyloxy-N-alkoxyamides.

We originally investigated the chemical properties of NAAs as N-alkoxyl analogues of N-acetoxy-N-arylacetamides 4, one penultimate carcinogenic metabolite of aromatic amines 3 such as carcinogenic 2-acetylaminofluorene or 4-acetylaminobiphenyl (Scheme 2).^[25–29] These produce by solvolysis N-acetyl-N-arylnitrenium ions 7, the ultimate electrophilic carcinogenic metabolites, which react with DNA.^[30–33]

Scheme 2.  Parallels between aromatic amine metabolites N-acetoxy-N-arylacetamides 4 and N-acyloxy-N-alkoxyamides 5.

In foundation chemistry, we synthesised N-acetoxy-N-alkoxymides 5 for the first time from hydroxamic esters 6 by chlorination at nitrogen followed by reaction with silver acetate and found them to be mutagenic in the Ames test.^[14,34,35] N-acyl-N-alkoxynitrenium ions 8 and N-acetyl-N-arylnitrenium ions 7 are similarly resonance-stabilised,^[36,37] so by analogy with 4, we envisaged that 5 could undergo solvolysis to 8, which could react with nucleophilic centres in DNA (Scheme 2, dashed route). However, in vitro DNA damage studies demonstrated that NAAs bind and react intact with plasmid DNA primarily at G-N7 in the major groove and, to a lesser degree, at A-N3 in the minor groove (Scheme 2, solid arrow; path v, Scheme 1), rather than through the intermediacy of nitrenium ions 8.^[38,39] Furthermore, both the acyloxyl and an alkoxyl substituent at nitrogen are determinants of reactivity with DNA.^[38] Not surprisingly therefore, NAAs are direct-acting mutagens reverting the point mutation strain Salmonella typhimurium TA100 to wild type in the Ames reverse mutation assay without the need for metabolic activation.^[34,40] Reactions with N-methylaniline have been widely studied kinetically to determine factors affecting S_N2 reactivity, as this was deemed pivotal in their behaviour within the major groove of DNA.^[16,18–20]

In the full-plate incorporation assay used throughout our Ames investigations, point mutations result in countable colonies,^[41] and at the outset, we found that in nearly all cases the NAAs gave a linear dose response, the gradient of which afforded LogTA100, the log₁₀ of induced revertants at 1 μmol per plate.^[40] A comparison of activities at 1 μmol per plate enabled us to investigate factors that both promoted and inhibited their mutagenic activity. In all, we have determined comparative data for some 100 NAAs.

The Ames methodology involves some variability in testing medium and therefore background reversion rates. However, to counteract this, we tested every new compound in parallel with a single compound, N-butoxy-N-acetoxybenzamide (9, X = H). In this manner, all new data could be scaled to the activity of this standard. As a consequence, changes in LogTA100 could be sheeted home to alterations in structure rather than assay variability. This proved to be pivotal in determining structural influence on a range of cellular and intracellular processes. As the target giving rise to reverse mutation in bacterial TA100 must be cytosolic DNA, we have been able to use standardised Ames mutagenicity to deduce factors affecting the transport to, binding with and reactions with DNA from correlations between chemical structure and activity.

Fig. 2 illustrates the various intervening processes involved in the ultimate damage to DNA. Passage through the cell wall and associated equilibrium constant, K_L, is governed by lipophilicity and is a function of logP, log₁₀ of the octanol/water partition coefficient for the molecule. Reaction with cytosolic DNA must depend on binding in the major (G-N7) and minor (A-N3) grooves and be governed by binding constant K_M, which must also depend, in part, on logP. However, K_M must also be related to other factors such as steric bulk or the dimensions of the mutagen, or structural aspects that enhance binding to DNA such as intercalation. The final determinant of DNA damage is the rate constant for bimolecular reaction with the nucleophilic centres in DNA, k_M, which can be dictated by factors affecting the electrophilicity of the amide nitrogen and conformation for S_N2 attack at the amide nitrogen. In the course of our studies, we have encountered structural elements impacting on all of these processes.

Fig. 2.  Elemental processes in the reaction of N-acyloxy-N-alkoxyamides with DNA. Reproduced from Ref. [42] with permission from the Royal Society of Chemistry.

A review of the mutagenicity of NAAs covering early correlations appeared in 2008 as part of a wider study of the structure, reactivity and biological activity of NAAs and led to the first working quantitative structure–activity relationship (QSAR).^[8] This account serves to build on this and to demonstrate applications of higher-level QSARs to gain insights into transport, DNA binding and reactivity of NAAs with DNA.

As with most drug interactions, the hydrophobicity of NAAs is a critical determinant of activity.^[43–48] In this host–guest relationship, hydrophobicity is important for binding in the hydrophobic grooves of DNA. Hence, logP is an important molecular property. Throughout our studies on NAAs, we have computed this according to the Ghose–Crippen model,^[49] values for which are readily determined in most computational chemistry packages.

Chart 2. 

The importance of steric effects will be discussed at a later point, but initial mutagenicity data from series of mutagens bearing para-substituted phenyl groups on all three side chains (9–11, Chart 2) indicated that, whereas logP for para-tert-butylphenyl-substituted systems (X, Y and Z = Bu^t) was much larger than for para-methylphenyl-substituted systems (X, Y and Z = Me), activities as determined by LogTA100 were lower in each case. Similarly, para-phenyl groups (X, Y and Z = Ph) with much larger logP afforded only modest increases in LogTA100.^[50] In development of QSARs, Taft steric parameters for para-phenyl substituents were deemed to be representative of these steric effects and, where relevant, were incorporated as E_s¹, E_s² and E_s³ for para substituents on benzamide, benzyloxy and benzoyloxy phenyl rings respectively.^[43,51]

Reactivity at the amide nitrogen was seen to be a significant determinant of activity. As substituents on a benzamide or benzyloxyl sidechains are fairly remote from the reactive nitrogen, they were not expected to impact measurably on activity. In the case of a series of para-substituted N-acetoxy-N-butoxybenzamides 9, this was borne out by relative rates of S_N2 reactivity with N-methylaniline, which correlated very weakly with Hammett σ constants (reaction constant ρ = 0.13).^[18,19] The electronic effect of para-substituents on a benzoyloxyl group in N-alkoxy-N-benzoyloxyamides was very different. Bimolecular rate constants for the reaction of N-methylaniline with a series of N-benzyloxy-N-(para-substituted-benzoyloxy)benzamides 11 correlated strongly with Hammett σ constants (ρ = 1.69)^[18,19] in support of the modelled charge-separated S_N2 transition state, which invokes partial negative and positive charge on the leaving group and nucleophilic nitrogen respectively.^[8,16,17,19] In addition, we demonstrated that rates constants for S_N2 reaction of a range of N-alkanoyloxyl and benzoyloxyl amides with N-methylaniline, k³⁰³, correlate negatively with pK_a of the departing carboxylic acid group (Eqn 1).^[16] However, mutagenicity of a series of N-benzyloxy-N-(para-substituted-benzoyloxy)benzamides correlated negatively with Hammett σ constants (ρ = –0.57)^[22] or positively with the pK_a of the carboxylic acid (Eqn 2); the more reactive an NAA, the greater the probability that it will react with adventitious intracellular nucleophiles prior to binding to DNA, thereby lowering the effective concentration.

Several earlier QSARs based on limited data^[8,50] culminated in a QSAR based on logP, pK_a of the leaving group and Taft E_s¹, E_s² and E_s³, which performed well for 50 analogues (Eqn 3, Fig. 3, open squares).^[52] correlation coefficient, r, standard error, s, F-statistic and the leave-one-out cross validation index, Q², all point to reasonable predictive ability.

n = 50, r = 0.89, s = 0.16, F = 34, LOOCV Q² = 0.70.

Fig. 3.  Predicted vs experimental mutagenicities as LogTA100 for 50 members of the training set (open squares), 12a–e (filled triangles), 13a (filled diamond), 13b (open diamond) and 14, 15a–c, 16a (closed circles) in Salmonella typhimurium TA100 using the linear QSAR in Eqn 3.

At the time of publication of such relationships, it was deemed unusual for direct-acting mutagens to show a logP dependence because binding to activating enzymes in metabolism was not important.^[43,45,53] Where mutagens or other molecules require metabolic activation, the role of structural change and logP becomes less clear owing to its effect on binding to enzymes, which may be the determining factor. The modest dependence on logP is clear cut in the case of these direct-acting mutagens and relates to binding to DNA. These molecules are small and the bacterial strains developed by Ames are altered by the rfa mutation, rendering their cell walls more permeable to bulky molecules.^[41,54] Drug-like compounds are more than likely to have a logP < 5 and, similarly, direct-acting mutagens acting on Salmonella typhimurium would be expected to conform to this rule.^[55] However, in a series of NAAs bearing long aliphatic chains 12 (Chart 3), a significant deviation from the predicted LogTA100 was observed.^[52] Moreover, the deviation was logP-dependent (Fig. 3, solid triangles, Table 1). Whereas the activity of 12a was well predicted by the QSAR in Eqn 3, activities of 12b–e deviated progressively from their predicted values with increasing logP. Though unrelated to the benzamides, the deviation from the predicted activity for the naphthamide 13b bearing long aliphatic chains on the alkoxyl and acyloxyl group was also significant (Fig. 3, open diamond).

Table 1.  LogP_GC, pK_a, experimental and calculated LogTA100 for 12a–e and 13b.

Chart 3. 

The decrease in mutagenicity relative to their predicted activity can be ascribed to lipid entrapment through membrane localisation and a reduced K_L (Fig. 2) resulting in lower concentrations of mutagens available to interact with cytosolic DNA.^[43] With respect to the mutagenicity of N-acyloxy-N-alkoxybenzamides, there is clearly an optimum logP value in the range of logPs covered by mutagens 12a–e.

Several treatments are possible to generate QSARs catering for drugs covering a wide range of logP. To establish a QSAR that encompasses the activities of increasingly hydrophobic mutagens 12a–e, we chose Kubinyi’s bilinear model.^[52,56–58] In Eqn 4, β is a non-linear term calculated by an iterative procedure. In cases where logP is sufficiently small, βP becomes negligible and therefore the log(βP + 1) term decreases to zero, converting the model to the linear form (Eqn 3). We determined that a logP₀ of 6.4 is the optimum logP for activity and, by Kubinya’s method, can be derived from logP and log(βP + 1) coefficients, a and b, from the regression analysis using Eqn 5.^[52,56–58] Experimental and predicted LogTA100 from the bilinear model in Eqn 4 are given in Table 1 and are in good agreement when compared with the result from the linear model in Eqn 3. It will be reported in the section on ‘Naphthalene as an intercalator’ that a naphthalene substituent increases binding to DNA through intercalation. As illustrated in Fig. 3, the activity of 13a (closed diamond), given here for comparison,^[40] is significantly higher than predicted, whereas that of the lipophilic 13b is strongly suppressed (Fig. 3, open diamond). However, the activity of 13b is well reproduced by Eqn 4, indicating that the activity enhancement due to the naphthamide structure is largely negated owing to lipid entrapment. K_L in Fig. 2 is dominant and limiting in this case.

Logβ = −6.639, n = 55, r² = 0.75, adj. r² = 0.71, s = 0.18, F = 19.7; LOOCV Q² = 0.60.

Steric effects would be expected to impact on activity levels in several ways; they could influence the capacity of NAAs to bind in the hydrophobic grooves of DNA (reducing K_M) or, in the bound state, they could inhibit reaction with nucleotides (reducing k_M). Previous investigations found that mono tert-butylated mutagens (9, X = Bu^t, 10, Y = Bu^t, 11, Z = Bu^t) were well predicted by an earlier QSAR.^[59] Activity of the di-tert-butylated diaryl mutagen 14 (Chart 4) was also well predicted by the linear and bilinear QSARs in Eqns 3, 4 (Fig. 3, Table 2). For a series of di-tert-butylated triaryl mutagens 15a–c and 16a (Chart 4), we encountered significantly diminished activity with the linear QSAR in Eqn 3 (Fig. 3, filled circles, Table 2) and although dose responses were evident, with LogTA100 > 2, only 15c can be regarded as mutagenic. No dose response could be obtained with tri-tert-butylated substrates 15d and 16b, which must be regarded as non-mutagenic towards TA100. We originally ascribed this reduced activity to the dimensions of these triarylated systems, which from X-ray and computed structures are from end to end between 14 and 17 Å, at least of the order of the width of the major groove of DNA (12 Å).^[8,59] However, these substrates have logP values of 8.24, larger than the logP₀ for NAAs (logP₀ = 6.4). The bilinear QSAR in Eqn 4 corrects predicted values in relation to their high hydrophobicity and it is found that 15c is well predicted (Table 2); the deviation from expected activity from Eqn 3 is entirely due to the increase in hydrophobicity in this case. On the other hand, activities of mutagens 15a–b and 16a still deviate strongly from predicted values after taking their increased hydrophobicity into account, which reinforces the groove exclusion argument for these substrates. As a tert-butylbenzyloxy group is common to all triarylated mutagens with greatly reduced activity, it would appear to be a limiting feature, whereas tert-butyl groups on the benzoyloxyl group may be less critical. From models, it is also possible that the transition state complex for S_N2 reaction at G-N7 can only be achieved with one para tert-butyl group.^[8] Fig. 4a depicts the AM1-optimised transition state for the reaction of model N-formyloxy-N-methoxyformamide with G-N7. Modelling this into the major groove of a segment of DNA with 15e (Fig. 4b) indicates severe restrictions for tert-butyl groups on the benzoyloxy and benzamide rings, which would need to make adverse contact with the wall of the major groove.

Table 2.  LogP_GC, experimental and calculated LogTA100 for di-tert-butylated mutagens 14, 15a–c and 16a.

Fig. 4.  (a) AM1 optimised transition state for reaction of N-formyloxy-N-methoxyformamide with G-N7; (b) Depicted in yellow is the conformation of 15e in the major groove of DNA constrained to the transition state in (a) with side chains minimised by molecular mechanics (from Ref. [8]); black circle, tert-butyl group; white circles, para positions on the benzoyloxyl and benzamide side chains.

Chart 4. 

Branching adjacent to the amide carbonyl in NAAs has a profound effect on the mutagenic response. Whereas acetamide 17a is well predicted, propanamide 17b, 2-methylpropanamide 17c, 1-adamantanecarboxamide 17e and neohexamide 17f gave weak responses deviating significantly from the predicted activities from the linear QSAR in Eqn 3. Deviations from predicted values increased through the series CH₃ < Et < Prⁱ < Bu^t. The tert-butyl side chain in 17d resulted in no response at all. With adamantyl and neopentyl side chains (17e and 17f), experimental activities deviated strongly from predicted activities (Fig. 5, filled diamonds; Table 3). The results are broadly in line with the rates of S_N2 reactivity with N-methylaniline at 303 K in [D₄]methanol where amides with branched amide side chains in 17c, 17d and 17e were unreactive (Table 3). Branching α to the carbonyl is also known to impede S_N2 displacement of chlorine in α′-haloketones.^[60] Mutagenicity, which must involve reaction of DNA at G-N7, was observed, albeit weak in all but the 2,2-dimethylpropanamide substrate (17d). We have attributed this difference to a greater ease of S_N2 reaction in the hydrophobic groove of DNA; in solution, S_N2 transition states for reaction with amines are strongly charge-separated, resulting in large negative entropies of activation due to solvent reorganisation.^[8,16–18] However, in the hydrophobic grooves of DNA, a similar transition state would not be adversely affected by negative entropy of activation, making S_N2 reactivity in the grooves of DNA easier than in solution. It is clear that in this series, mutagenic activity is limited by k_M in Fig. 2

Table 3.  Rate constants for bimolecular reaction with N-methylaniline, logP_GC, pK_a, experimental and predicted LogTA100 for 17, 18 and 19 using linear QSAR in Eqn 3.

Branching adjacent to the alkoxyl oxygen and the presence of bulky benzyloxy groups in series 18 has almost no effect on mutagenic activity and the activity of all mutagens was well predicted by the linear QSAR in Eqn 3 (Fig. 5 – filled squares, Table 3). Their activities are controlled by their logP and pK_a values, which impact on K_M and k_M in Fig. 2. These are all small molecules, so groove binding would not be expected to be limiting and bulkiness at the alkoxyl group does not appear to impede S_N2 reactivity at G-N7. The rates of S_N2 reactions with N-methylaniline in [D4]methanol have also been measured for this series and rate constants at 303 K are somewhat smaller than their straight-chain n-butoxy analogue 19a (Table 3).^[16] 18c, with a tert-butoxy group, was unreactive but the steric interference is not as severe as is found with branching on the amide side chain, which we attribute to greater flexibility on the alkoxyl side chain. Clearly though, a reduction in the ease of S_N2 reactivity with branching does not manifest itself in the mutagenic response. Once again, the factors controlling reactivity in the hydrophobic grooves of DNA appear to be somewhat different from those governing S_N2 reactions with amines in solution.

A series of mutagens bearing bulky substituents adjacent to the acyloxyl carbonyl in 19 demonstrated a systematic suppression of the mutagenic activity relative to the predicted activities from Eqn 3 (Fig. 5, filled triangles; Table 3). All have small logP values and ought not to be subject to lipid entrapment or restricted access to the major groove. Although for a range of such compounds rates of reactivity with N-methylaniline correlated negatively with pK_a (Eqn 1),^[16] mutagenic activities were suppressed by an average of 0.59 LogTA100. While mutagen 19a with an acetoxyl leaving group (in training set) and to a degree 19b with a butanoyloxyl group are predicted well, the effect of branching on LogTA100 as in 19c–g is significant, with the interference of isopropyl (19c) and tert-butyl groups (19g) being the greatest. This reflects steric hindrance to reaction with DNA, possibly through inability to achieve a conformation suitable for reaction at G-N7 (lowering k_M in Fig. 2). It is possible that, unlike groups on the alkoxyl side chain, the acyloxyl group must make unavoidable contact with the wall of the major groove. A comparison of data for the benzoyloxy substrate 19j with the dimethylated analogues reinforces this. Although 19j was slightly overpredicted, the 2,6-dimethylphenyl side chain in 19i strongly inhibits activity (Table 3). The effect is reduced when the methyl groups are on the 3- and 5-positions in 19h. Clearly, in 19i the twisting that would be required to minimise steric interference with the carbonyl must have an influence (Fig. 6).

Fig. 5.  Predicted vs experimental mutagenicities as LogTA100 for training set (open squares), 17a–e (filled diamonds), 18a–g (filled squares), 19a–i (filled triangles) and 13a/20 (filled circles) in S. typhimurium TA100 using the linear QSAR in Eqn 3.

N-Acetoxy-N-butoxy-2-naphthamide mutagen 13a was part of an earlier study in which we demonstrated an increased degree of DNA damage relative to other NAAs with similar logP.^[38] Furthermore, 13a exhibited mutagenic activity nearly an order of magnitude higher than predicted by the QSAR in Eqn 3 (Fig. 3, solid diamond).^[40] This, together with a series of naphthalene-containing NAAs 20a–g (Chart 5), all showed a remarkably consistent increased activity relative to predicted LogTA100 levels based on the QSAR in Eqn 3, with most differences between experimental and predicted activities close to 1 LogTA100 (Fig. 5, filled circles; Table 4).^[42] Futhermore, the enhanced activity was evident irrespective of the regiochemistry, the locus of substitution on the naphthalene nucleus, or the length of aliphatic tethers where present. We attributed this enhancement to an increase in residence time on DNA through the capacity of the planar naphthalene to intercalate with the DNA bases.

Fig. 6.  Conformations in N-butoxy-N-(3,5-dimethylbenzoyloxy)- and N-butoxy-N-(2,6-dimethylbenzoyloxy)benzamides 19h and 19i.

Table 4.  Experimental and predicted LogTA100 for 13a and 20a–g in S. typhimurium TA100 when calculated using linear QSARs in Eqns 3, 6, and bilinear QSAR in Eqn 7.

Chart 5. 

Naphthalene has not generally been regarded as an intercalator, a property reserved for larger, fused polyaromatic systems with greater surface areas.^[62–64] The ring system should possess a minimum of three rings to give an optimum surface area of more than 28 Å.^[65] Various theories have been proffered regarding the capacity of planar aromatic systems to intercalate and the presence of positively charged side chains or acyl substitution was deemed important for naphthalene intercalation.^[66,67] The intercalation of naphthalene mono- and di-imides, established threading intercalators, is well known.^[68–70] Although in only two of the set studied, 13a and 20a, was naphthalene acyl-substituted, the remainder being bonded to methylene, it is hard to ascribe this increased activity to any process other than intercalation. These molecules have very modest computed logPs and no special features above those of similarly constituted benzamide systems in the training set. However, a modest, reversible ability to intercalate into DNA would increase residence time in the grooves of DNA, thereby increasing K_M in Fig. 2.

TA98 is a frameshift strain derived from Salmonella typhimurium that is used to detect intercalation. In effect a base deletion results in a reading error and a failure to proliferate in the plate media used in Ames assays. The space requirement for an intercalated polycyclic aromatic hydrocarbon(PAH) (3.4 Å) is similar to that required by a single base pair,^[63,71–76] and intercalation of a PAH between the bases of DNA reinstates the wild-type reading frame in S. typhimurium through an a + 1 frameshift.^[41,76,77] Furthermore, such restoration is particularly effective (10–100-fold increase) when the intercalative group is associated with an electrophilic centre, reaction through which can localise the intercalating group.^[76–79] Importantly, where the intercalator is dissociated in the reaction with DNA, the duality requirement is lost, resulting in a weak TA98 response. With their electrophilic amide nitrogens, NAAs appeared to be ideal candidates to effect such frameshifts. We subjected naphthalene-bearing mutagens 13a, 20a, 20d and 20g to TA98 tests and found linear dose responses and significant LogTA98 at 1 μmol per plate for 13a (490 revertants; average of six tests), 20d (287 revertants) and 20g (114 revertants), which was not evident in the control systems N-acetoxy-N-butoxybenzamide 19a and N-benzoyloxy-N-butoxyacetamide 17a, the phenyl analogues of 13a and 20a or naphthalene-bearing systems without electrophilic nitrogen such as 21a–b, both of which produced no dose response.^[42] 20a, where the intercalator dissociates from the mutagen in reaction with DNA, resulted, predictably, in a weak response (15 revertants) in TA98.

We consider this clear evidence for intercalation of naphthalene moieties.

To account for this activity enhancement, we regenerated a linear (Eqn 6) and a corresponding bilinear (Eqn 7) QSAR to enable prediction of this intercalative ability by the introduction of an indicator variable, I, which is given the value 1 if a naphthalene is present, otherwise 0.

n = 58, R² = 0.85, adj. R² = 0.83, s = 0.16, F = 48.2; LOOCV Q² = 0.85.

Logβ = −6.705, n = 63, R² = 0.81, adj. R² = 0.79, s = 0.18, F = 29.6; LOOCV Q² = 0.76.

From Eqn 5 and the coefficients in Eqn 7, a revised LogP₀ value of 6.44 can be calculated, slightly higher than 6.37 calculated from Eqn 3.

A comparison of QSAR results for 13a, 20a–g from Eqn 3 with those from revised linear and bilinear QSARs in Eqns 6, 7 using I = 1 is presented in Table 4. The differences between experimental and predicted mutagenicities are significantly diminished when compared with predictions from the linear QSAR in Eqn 3. Correlations between the predicted and experimental LogTA100, based on the new QSAR in Eqn 6, are illustrated in Fig. 7 (training set, filled squares).

As the correlations with logP and I in Eqn 6 are highly significant (P- values of 10⁻¹³ and 10⁻¹⁷ respectively), the respective coefficients of 0.26 and 0.83 indicate that the attachment of one naphthalene group increases activity equivalent to between 3 and 4 logP. Although the logP of 13a and 20a–g (logP  2.0–3.8) falls below the logP₀ for NAAs (6.44 from Eqn 7), they exhibit mutagenic activity that is equivalent to mutagens with an ‘effective’ logP ranging between logP = 5 and 7. This can be readily seen from a plot of linear (Eqn 8) and bilinear logP dependence (Eqn 9) over the range logP = 0–10 (Fig. 8), where the boxes reflect actual and effective logP for this series of naphthalene-bearing NAAs.^[42]

n = 55, R² = 0.677, s = 0.19, F = 98.5.

Logβ = −6.466, n = 55, R² = 0.677, s = 0.19, F = 35.55954, logP₀ = 6.37.

Fig. 7.  Predicted vs experimental mutagenicities as LogTA100 for training set (filled and empty squares), biphenyl side chains (filled diamonds) and fluorene side chains 22a–b, 23a–b (open triangles I = 1 and filled triangles I = 0) in S. typhimurium TA100 using the linear QSAR in Eqn 6.

This finding that a naphthalene enhances the activity of NAAs to the tune of between 3 and 4 logP units by its hitherto unknown intercalative ability is an important new discovery. Presumably, the DNA targeting function of naphthalene is not unique to NAAs. The Lipinski Limit is a hydrophobicity above which lipid entrapment diminishes the activity of a drug.^[55] A drug with logP close to the Lipinski Limit for that type of drug, but bearing a simple naphthalene substituent, can pass through the lipid membrane, yet exhibit binding to DNA equivalent to a molecule with a substantially higher hydrophobicity.

The role of naphthalene appears not to be unique. We have recently found that Eqn 6 (logP < 6.4) or Eqn 7 (logP > 6.4) can be used to differentiate between molecular substructure capable of intercalation into DNA and substructure that does not intercalate, based on the fit of measured LogTA100 with I = 1 or 0. If the fit to the QSAR is best with I = 0, the structure is a groove binder with mode of action similar to most of the mutagens we have studied. If the fit is best with I = 1, such molecules target DNA through a degree of intercalation akin to that of naphthalene-bearing mutagens.

Three substructures demonstrate this principle: NAAs bearing the biphenyl moiety, the fluorene ring system and the fluorenone group, which are related. In fluorenes, the biphenyl rings are bridged by a methylene, making fluorene groups largely planar but not cross-conjugated as is the case of fluorenones, where the rings are bridged by a carbonyl (Chart 6).

Chart 6. 

Measurement of TA100 activity for NAAs with potential intercalators incorporated into side chains was instructive. Firstly, biphenyl systems on all three side chains (Fig. 7, filled diamonds) have previously been shown to conform to linear QSARs and are part of the training set, which now incorporates eight naphthalene-bearing NAAs (Fig. 7, filled squares). The biphenyl group does not target DNA beyond its hydrophobic influence.

Fig. 8.  Linear (Eqn 8) and bilinear (Eqn 9) dependence on logP and the experimental activities of naphthalene-bearing compounds 13a, 20a–g at calculated (closed triangles) and effective logP range (open triangles).^[42]

The activity of the four fluorene-bearing mutagens (Chart 7) is greatly over-predicted with I = 1 (Fig. 7, open triangles, Table 5).^[80] Activities of 23a, 23b are well predicted and 22a is adequately predicted with I = 0. Activity of 22b would appear to be even lower than predicted with I = 0 (Fig. 7, filled triangles; Table 5). It would appear that all four NAAs bearing fluorene are not targeted to DNA through intercalation, as was the case with naphthalene. These structures possess an aliphatic bridge and no cross-conjugation, which may result in poor π–π stacking and a lack of polarisability. We ascribed the much diminished activity of 22b to the fact that the pendant group is directed into the bay region of the fluorene, resulting in more difficult access to nucleophilic DNA components. The same is true for 23b and both side chains would be twisted out of plane of the ring system. However, there is less flexibility in the pendant group in 22b.

Table 5.  Experimental and predicted LogTA100 22a–b, 22d–e and 23a–e in S. typhimurium TA100 when calculated using linear QSAR in Eqn 6 with I = 0 and 1.

Chart 7. 

22b, 23a and 23b gave negative results in the TA98 screen. Surprisingly, 22a gave a reasonably positive result (246 induced revertants at 1 μmol per plate) though it showed no enhancement in TA100 and in fact was underpredicted using the indicator variable I = 0. We have ascribed this to oxidation of the fluorene to the fluorenone 22c in the plate incubation or, alternatively, selective metabolism to 22c by bacterial S. typhimurium TA98.^[80]

4-Aminobiphenyl and 2-aminofluorene are well known carcinogens where the metabolic route involves cytochrome P450 oxidation of the nitrogen, ultimately leading to electrophilic nitrenium ion formation, as illustrated in Scheme 2.^[25–31] TA98 studies, and studies with other frameshift-sensitive strains, have implicated metabolites of these compounds as intercalators.^[76,78] Our results point to the fact that such intercalation is a function of the binding of the derived metabolites, the aryl nitrenium ions. The biphenyl or fluorene aromatic ring systems do not appear to promote intercalation.

Fluorenone derivative 22c could not be synthesised by existing protocols,^[80,81] but the remaining fluorenone-bearing compounds (Chart 8) yielded a clearcut result. Activities of two of the fluorenone substrates, 22d and 23c, are clearly well predicted when I = 1 and those of the other three, 22e and 23d–e, are greatly overpredicted (Fig. 9, Table 5). However, activities of these three are very well predicted with I = 0.^[80] Clearly, two of the NAAs bearing fluorenone, like naphthalene derivatives, benefit from intercalation, whereas the remainder do not.

Fig. 9.  Predicted vs experimental mutagenicities as LogTA100 for training set (filled and empty squares), fluorenone side chains 22d–e and 23c–e (open triangles I = 1, filled triangles I = 0) in S. typhimurium using the linear QSAR in Eqn 6.

Substitution at the 2- (in 22d) and the 1-position (23c) appears to facilitate intercalation. In both these, the tricylic conjugated system can most probably penetrate adequately between base pairs. Of the three systems that did not report enhanced activity, 22e and 23e have the pendant active centre directed towards the distal ring in the bay region. In such systems, the intercalation would have to be edge-on rather than end-on and, presumably, would lead to a poor intra-base penetration and π–π stacking ability.

Chart 8. 

23d appears not to intercalate even though it could do so in an end-on fashion. We have argued that the difference between 23d and intercalating 22d may lie in different electronic effects owing to the fact that in 23d, the fluorenone is bonded through the 2-position to the ester carbonyl whereas in 22d, it is bonded to the amide carbonyl although the argument is by no means a compelling one.^[80]

Fig. 10 shows AM1-optimised ground state geometries for 22c–e and 23c–e. It is clear, when comparing 22d with 22e, and also 23c with 23e that the accessible surface area of the tricyclic systems in 22e and 23e, where the pendant group is in the bay region, would be smaller than in 22d and 23c, which can bind in an end-on fashion.

Fig. 10.  AM1 minimum energy conformations of (a) 22c, (b) 22d, (c) 22e, (d) 23c, (e) 23d and (f) 23e.

TA98 studies confirmed that in keeping with its TA100 result, 22d is a strong intercalator (TA98 = 1458 revertants at 1 μmol per plate).^[80] In accord with the dissociation arguments above, 23c (a strong intercalator from TA100 studies) gave a weak response through disconnection on reaction, as did 22e and 23d–e, which were not predicted to intercalate by TA100. Though we were unable to synthesise 22c, it is possible that it was generated from fluorene 22a by oxidation during the TA98 procedure.

Application of our QSAR would appear to differentiate not only between substructure promoting intercalation or otherwise, but also between modes of intercalation. A study of the mutagenic activity of five other mutagens bearing a 1-naphthalene group (Chart 9) reinforced this concept. Although 24b could not be synthesised owing to steric impact of the 2-methyl group, LogTA100 data for five naphthalene-bearing mutagens 24a, 24c and 25a–c were acquired.^[81]

Chart 9. 

The activities of 24a and 25b and 25c were clearly overpredicted from Eqn 6 with I = 1 (Fig. 11, open triangles; Table 6), but were not in excess of the predicted value with I = 0 (Fig. 11, closed triangles; Table 6). Such naphthalenes would need to intercalate in an edge-on rather than an end-on orientation as the pendant groups would be twisted out of the plane of the ring system, thereby blocking penetration of the naphthalene ring in a substantive fashion.^[81]

Fig. 11.  Predicted vs experimental mutagenicities as LogTA100 for training set (filled and empty squares), 1-naphthalene side chains 24a, 24c and 25a–c (open triangles I = 1, filled triangles I = 0) in S. typhimurium using the linear QSAR in Eqn 6.

Table 6.  Experimental and predicted LogTA100 for 24a, 24c and 25a–c in S. typhimurium TA100 when calculated using linear QSAR in Eqn 6 using I = 1 and 0.

25a gave a reproducible ambivalent result while 24c surprisingly and reproducibly correlated with I = 1. Of the five mutagens bearing 1-substituted naphthalenes, only 24c was active in TA98 (3803 revertants at 1 μmol per plate), confirming the TA100 study.^[81] Although this was a seemingly incongruous result, there is some evidence from PAH intercalation that a methyl substituent can enhance intercalation, possibly by increasing polarisability of the ring system and enhancing electrostatic interactions and π–π stacking capability.^[82] It is deemed that the electronic influence of aliphatic substituents that enhance intercalation are more important than the steric effects that inhibit intercalation.

There is fairly recent evidence that derivatives of the ‘left half’ of Azinomycin B modified to remove alkylating activity, 26, and that have an acyl side chain naphthalene motif similar to that of 25 bind intercalatively to DNA.^[83] However, the unsubstituted naphthalene system in the same structure 27 was shown to be ineffective in DNA binding.^[83] The impacts of substituents on π-electron density distribution are likely to make a significant contribution to the intercalative potential of the naphthalene group. It is not a clearcut situation though; computational studies by Alcaro et al. have been unable to clearly define the role of the naphthalene moiety in Azinomycin,^[84,85] and Coleman et al.^[86,87] and Casely-Hayford^[88] suggest that there is no intercalation of the naphthalene moiety in 26 or 28. There is no doubt, though, that 24c intercalates and so too does 20g with a 1-naphthyl group on the alkoxyl side chain (Table 4).

Once again, the difference between end-on and edge-on intercalative ability can clearly be seen in a comparison between the AM1-optimised structures of 13a and 20a, both 2-naphthalene derivatives, and their 1-naphthalene analogues 24a and 25a (Fig. 12).

Fig. 12.  AM1 optimised structures of 1- and 2-substituted naphthalene-bearing mutagens (a) 13a, (b) 24a, (c) 20a and (d) 25a.^[81]

Pyrene and its derivatives have long been known to bind to double-stranded DNA by intercalation and the property has been studied in detail through spectroscopic studies or binding studies.^[89–95] Recently, kinetic studies have shown that the binding of a pyrene unit to DNA involves a two-step mechanism; the first step is the formation of a precursor complex, while the second step is the formation of the intercalated complex through penetration of the pyrene moiety between the base pairs of DNA.^[90] Recently, the intercalation of a pyrene unit has been exploited as a photophysical probe,^[93] in single-molecule atomic force microscopy (AFM) for double-stranded DNA mismatches,^[96] and as a surface modification for the immobilisation of double-stranded DNA (ds-DNA) on a solid support.^[91]

The capacity for pyrene to intercalate with ds-DNA with high binding constants spurred us to extend our QSAR work to NAAs bearing a pyrene substituent. On the premise that pyrene would localise the NAAs and the pendant electrophilic centre would need to access proximal guanines, 1-pyrene was attached directly and through a variable-length tether with a view to determining the impact of pyrene, or tethered pyrene, on mutagenic activity (Chart 10).

Chart 10. 

Table 7 gives experimental and predicted LogTA100 for pyrene-bearing mutagens, all of which have logP below 6, and the data are illustrated in Fig. 13. Four of the derivatives, 30a, 31, 32a and 32b, are better predicted with I = 1 (open triangles), although 30a, bearing a pyrene tethered through one methylene group on the amide side chain, is still underpredicted with I = 1. This was the most mutagenic of all the NAAs we have tested. Surprisingly, LogTA100 for 30b and 29 were better predicted with I = 0 although activities of both are somewhat overpredicted even with I = 0. 30a and 30b both bear the pyrene on the amide side chain but the propyl tether in 30b appears to render activity typical of a groove binder rather than an intercalator. All three substrates with pyrene on the acyloxyl group (31, 32a and 32b) are predicted to benefit from intercalation. 30a (9448 revertants at 1 μmol per plate), 31 (239 revertants at 1 μmol per plate) and 32a (385 revertants at 1 μmol per plate) gave a positive result in TA98, confirming their intercalative ability whereas 30b was negative in the TA98 test in accord with it reacting as a groove binder.

Table 7.  Experimental and predicted LogTA100 for 29, 30a–b, 31, 32a–b in S. typhimurium TA100 when calculated using the linear QSAR in Eqn 6 with I = 0 and 1.

Fig. 13.  Predicted vs experimental mutagenicities as LogTA100 for training set (filled and empty squares), pyrene-bearing N-acyloxy-N-alkoxyamides 29–32; (open triangles I = 1, filled triangles I = 0) in S. typhimurium using the linear QSAR in Eqn 6.

With pyrene-substituted mutagens, several scenarios are possible.

1. The pyrene could intercalate strongly, in which case the tether must be of suitable length to enable reaction at an N7 of an adjacent or distal guanine. Should this occur, the mutagen would register as benefiting from enhanced activity in TA100 and as a frameshift mutagen in TA98 through localisation of the pyrene. It would appear that 30a, 31 and 32a fall into this category. Two of these (30a and 31) bear a three-bond tether while 32a has a four-bond tether.

2. A mutagen could intercalate through pyrene but be unable to react at an adjacent or distal guanine. Intercalation is an equilibrium process, and such compounds could still damage DNA in a groove-bound situation and would register in TA100 with I = 0. However, they could also register as an intercalator in the TA98 test. 29 with a short, two-bond tether falls into this category. From Eqn 6, it correlates well with I = 0 but is moderately active in TA98 (281 revertants at 1 μmol per plate).

3. A mutagen could intercalate and facilitate a reaction of the tethered nitrogen at a proximal or distal G-N7 but if the tether becomes disconnected in the process, which, as described earlier, is the case where the pyrene is on the acyloxyl side chain, the pyrene may not localise sufficiently without the tether, and register weakly in TA98. 32b may fall into this category. The six-bond tether may inflict damage, probably at a distal G-N7, in the intercalated state and register as an intercalator in TA100 but disconnection through reaction at G-N7 results in poor pyrene retention; 32b is negative in TA98 (26 revertants at 1 μmol per plate)

4. A mutagen could register negatively for intercalation in both TA100 and TA98 if it reacts readily in the groove-bound state prior to intercalation but in the bonded state intercalation is difficult to achieve. 30b could be categorised as such. Its activity in TA100 does not benefit from intercalative enhancement and it is a very poor intercalator from TA98 (88 revertants at 1 μmol per plate).

Molecular modelling suggests that with a three-bond tether, an intercalated pyrene would be able to react at an adjacent guanine without severe disruption to the helix. Hence, 30a and 31 are of the most active forms.

Overall, though, throughout our investigations of intercalative ability, we have found that the result from the frameshift mutation strain TA98 was in agreement with the predictions from point mutation strain TA100 in 79% of the 24 compounds studied. In the majority of cases, those compounds showing enhanced TA100 mutagenicity were also mutagenic in TA98. In general, when I = 0, the mutagen was non-mutagenic in TA98, whereas with I = 1, the mutagen gave a positive result. Collectively, this adds credibility to the value of the indicator variable I being a reporter of intercalative ability of a PAH in the TA100 reverse mutation assay.^[81]

Early studies of substituent electronic effects for a series of N-acyloxy-N-benzyloxybenzamides 10 produced a result that at the time was little understood.^[40] Activity for N-acetoxy-N-(4-nitrobenzyloxy)benzamide 33 in TA100 was substantially greater than expected but much lower in the presence of S9 liver homogenate, which is used to reproduce enzymic metabolic activation in the Ames test.^[41] We subsequently investigated four other 4-nitrobenzyl-bearing NAAs, 34–37, as well as the 2-nitro- and 3-nitro- analogues of 33, 38 and 39 (Chart 11) (unpublished data, SA Glover, K Kavanagh, RR Schumacher).^[81]

Chart 11. 

The results in Table 8 can be summarised as follows:

1. Although the activity of 33 in the absence of S9 was extremely high, the activity in the presence of S9 appeared to be well predicted by Eqn 6.

2. To varying degrees, the activities of all 4-nitrobenzyl-substituted NAAs 34–37 in the absence of metabolic activation are greater than predicted by Eqn 6, pointing to a general model for their mutagenicity.

3. The activity of the 3-methylated derivative 37 is underpredicted by Eqn 6, but not to the same extent as 33.

4. The 2- and 3-nitro analogues 38 and 39 were suitably predicted by Eqn 6.

Table 8.  Experimental and predicted LogTA100 for 33–39 in S. typhimurium TA100 when calculated using Eqn 6.

We contend that the heightened activity of the 4-nitrobenzyl substrates 33–37 is most likely due to reduction of the nitro group to the hydroxylamine by nitro reductase present in bacterial TA100 (Scheme 3). Acetylase or sulfotransferase would lead to the ester or sulfate ester derivatives, which could react with DNA or solvolyse to an electrophilic phenylnitrenium ion, though these ions are predicted to have very short lifetimes in aqueous media.^[30–33] Debnath et al. invoked these pathways in correlating the activities of a wide range of nitro aromatics in TA100 without metabolic activation.^[46,97] Clearly, logP can relate to binding to nitroreductase or sulfotransferase as well as to DNA. In the case of 33 in the presence of S9, the cytochrome P450 oxidative enzymes most likely preserve the integrity of the nitro group. Hence, the LogTA100 in the presence of S9 is almost identical to the predicted value (Table 8). We demonstrated early on that S9 plays no role in the activity of NAAs and, in general, dose–response plots were extended linearly where toxicity was demonstrated at higher doses without S9.^[40]

Scheme 3.  Metabolic activation pathways of 4-nitrobenzyloxy systems in TA100.

The 2- and 3-nitrobenzyl configurations 38 and 39 are well behaved, with LogTA100 close to the predicted value based on Eqn 6. It is apparent that in these cases, nitro reductase would appear to be inactive, which may be due to poor host–guest interactions relative to para-substituted benzyl groups in 33–36.

The 3-methyl-4-nitro analogue 37 appears not to be as mutagenic as 33 even though computations show that the nitro group in such configurations deviates minimally from planarity with the ring. This parallels the mutagenic activities of 4-nitrobiphenyl and its 3-methyl analogue, both of which require reduction and activation according to Scheme 3.^[98] Boche showed that the reduced mutagenicity of the methylated derivative was more than likely due to interference with binding to nitroreductase in TA100 rather than disruption of coplanarity with the aromatic ring.^[98,99]

Results for the naphthamide 36 indicate that, while the naphthyl group is present, it does not enhance activity through intercalation; LogTA100 is overpredicted with I = 1. With I = 0, the activity is significantly underpredicted as is the case for the other systems bearing a 4-nitrobenzyl group.

Activation of the 4-nitro substituent would appear not to be a uniform process across 33–37, which among other things could reflect different binding capacities with reductive enzymes in TA100. However, the molecules have capacity to react through two modes, leading to moderately enhanced activities relative to predicted levels.

Prior to development of functional QSARs governing the mutagenic activity of NAAs, Clay synthesised two NAAs bearing two naphthalene substituents, for which we obtained mutagenicity data (unpublished data, AM Bonin, SF Clay, SA Glover).^[81] Although a single naphthalene has been shown to elevate activity through intercalation, the role of multiple naphthalene units in the same molecule was unknown. 40 and 41 (Chart 12) were synthesised with the expectation that the presence of two naphthalene units would further elevate activity relative to mutagens bearing one naphthalene. The logP values of 41 (logP  6.17) approached the logP₀ for NAAs (logP₀ 6.4) and application of both the linear and bilinear QSARs was appropriate.

Chart 12. 

From Table 9, it is apparent that neither mutagen exhibited activity in keeping with intercalative enhancement; differences between predicted and experimental LogTA100 values were small when activities were computed with I = 0 by both Eqns 6, 7. Both 40 and 41 behave as if they are groove binders and the activity is largely governed by their hydrophobicity. Although these are fairly large molecules, normal reaction with G-N7 in the major groove of DNA would appear to operate. Intercalation of either naphthalene must be prohibited by these configurations; intercalation of one planar group must inhibit reaction with a neighbouring guanine, presumably for steric reasons. Hence, reaction must occur in a groove-bound complex without intercalation.

Table 9.  Experimental and predicted LogTA100 for 40 and 41 in S. typhimurium TA100 when calculated using the linear QSAR in Eqn 6 and bilinear QSAR in Eqn 7 with I = 0 and 1.

It should be mentioned that dual intercalation is prohibited by the neighbouring exclusion principle, which states that, at most, intercalators can only bind every second possible base pair site and binding of additional intercalators adjacent to the first is prohibited.^[63,71] Owing to the changes in the DNA backbone conformation associated with the helix unwinding that is involved in the base pair separation essential for intercalation, there are significant limitations on the proximity of intercalation sites to one another. Bis-naphthyl NAAs have both intercalator units within relatively close proximity to one another; thus, even if one naphthalene unit were capable of intercalative binding, the other naphthalene would be prohibited by the interaction of the first.

A recent paper by Levin and coworkers in Nature described a method of editing nitrogen from secondary benzylic amines and related systems.^[100] The method is widely applicable to editing nitrogen from secondary amines and suitably constituted primary amines.^[101] The reaction employs a novel NAA, N-benzyloxy-N-pivaloyloxy-4-(trifluoromethyl)benzamide 42 (Scheme 4). Based on our precedent,^[102–104] 42 reacts with suitable secondary amines, 43, to produce an anomeric amide intermediate 44, which undergoes HERON rearrangement to give benzyl 4-(trifluoromethyl)benzoate 45 and a 1,1-diazene or amino nitrene 46. The reaction sequence was based on our discovery of both reactivity of NAAs with N-methylaniline,^{[8,16–19,50,59,105]} and our discovery of the HERON reaction.^{[13,102–105]} The 1,1-diazene 46 extrudes nitrogen, producing two alkyl radicals, one of which is conjugated, and solvent cage recombination yields alkane 47. Through this sequence, dibenzylamines generate bibenzyls in good yields, thereby editing nitrogen from the secondary amine.

Scheme 4.  Editing nitrogen from secondary amines using an N-acyloxy-N-alkoxyamide 42.

The merit of Levin’s reaction and the value of the reagent has been recognised by both Unsworth^[106] and Bräse.^[107] Importantly, however, Bräse cautioned that we had much earlier proved NAAs to be chemical mutagens, a fact seemingly overlooked by Levin and coworkers. The author apprised Nature of potential mutagenicity of 42 and a correction to the paper appeared addressing this fact and stating that 42 belongs to the class of NAAs ‘some of which have been found to mutate genetic material’. However, the aforementioned sections indicate that almost all NAAs are classified as mutagenic; of the 100 variants we have made, only five structures were found not to be mutagenic in the Ames test.

Recently, we expanded our QSAR based on all published mutagens to date. Incorporation of three fluorene-bearing (22a and 23a–b) and five fluorenone-bearing mutagens (22d–e and 23c–e) yields Eqn 10:

n = 67, R² = 0.84, adj. R² = 0.82 s = 0.17, F = 50.7; LOOCV Q² = 0.83.

The predictive power of this QSAR, which applies to NAAs with logP < 6.4,^[52] is demonstrated by the small standard error, high F-value and the LOOCV Q² index of 0.83. Eqn 10 is similar to the preceding Eqn 6.^[42,61]

Our QSAR allows us to predict mutagenic activity of congeners with some accuracy, and employing the QSAR in Eqn 10 indicates that 42 has a very high probability of being a direct-acting mutagen. Initial application of the QSAR to 42 predicts a very high level of mutagenic activity (LogTA100 = 3.14, Fig. 14). From our studies of steric effects on the acyloxyl side chain in substrates 19a–g, we reported that branching at the α-position and other bulky groups reduces activity. Based on Eqn 10, the average suppression was 0.55 LogTA100 (Fig. 14, filled triangles).^[61]

Fig. 14.  Predicted vs experimental activities of mutagenic N-acyloxy-N-alkoxyamides, at 1 μmol per plate; training set, filled and open squares (new data: fluorene in red, fluorenones in green); 18a–g, filled triangles; 42, filled circles.

Correcting the initial activity of 42 yields an activity of LogTA100 = 2.6 (Fig. 14) and 42 would generate between 270 and 580 induced revertants at 1 μmol per plate. The QSAR points to 42 being at least as mutagenic as our standard, N-acetoxy-N-butoxybenzamide (9, X = H), which we have reproduced many times in the Ames studies.

Levin used other NAAs, 48, 49a–c, 50 and 51 (Chart 13), in the development of reagent 42. 48 is a proven mutagen; in an early study, we reported its activity at 1 μmol per plate to be LogTA100 = 2.63.^[40] After steric correction for the pivaloyloxyl group, 49a–c and 50 are predicted to have very similar activity to that of 42 and 48 (Table 10) with activity of 51 slightly lower.

Table 10.  Relevant QSAR parameters and predicted LogTA100 of N-acyloxy-N-alkoxyamides 42, 49a–c, 50 and 51 using the QSAR in Eqn 10.

Chart 13. 

Of the wide range of some 100 NAAs for which we have mutagenicity data, which accounts for almost all known NAAs, only three, 15d, 16b and 17d, failed to generate a dose response in TA100 and can be considered to be non-mutagenic, while 15a and 15b generated very shallow dose responses and are weakly mutagenic according to the protocols (LogTA100 < 2).^[41,59] As outlined in ‘Steric effects on activity’, non-mutagenicity of 15d and 16b has been attributed to their inability to enter the major groove of DNA on account of their overall dimensions.^[52,59] The negative result for 17d in the Ames test was in accord with its resistance to S_N2 reaction with N-methylaniline.^[59] It is possible that 15d or 16b may be suitable candidates for nitrogen deletion from amines in spectral editing.

The QSAR we have developed would strongly suggest that 42 will be a direct-acting mutagen and must be handled very carefully by practitioners in the field. It is important that 42, or any analogues thereof, should not be declared to be safe reagents for wide use in skeletal editing, and for synthesis in large quantities for routine use, until it is proved not to produce a dose response in the Ames reverse mutation test. It is incumbent on the author to inform users of Levin’s skeletal editing method, those employing other NAAs for variations thereof or for other means, of the mutagenic properties of almost all known variants. Eqns 6, 7, 10 provide useful means of predicting a priori the likelihood and extent of mutagenic activity.

N-Acyloxy-N-alkoxyamides, as a class of anomeric amides, are mutagenic in the Ames reverse mutation assay with point mutation strain TA100, and select members bearing intercalating groups are active in frameshift strain TA98, in both cases without the need for metabolic activation. They have been shown to react with plasmid DNA, primarily at N7 of guanine in the major groove of DNA. As such, their mutagenic activity can be used to determine the factors controlling ease of access to bacterial DNA, binding modes with DNA and the ease of reaction with the guanosine nucleoside in the bound state. The Ames test requires minimal quantities of these compounds, which are easily synthesised in small amounts with a high degree of structural diversity. Almost all are heavy oils, which can be safely procured, purified, characterised and tested without exposure.

By measuring and comparing the mutagenicity at 1 μmol per plate with TA100 in a standardised manner, we have been able to establish how their activity is related to their structural features. For comparisons with our data, any new compound should be tested with the standard (9, X = H), which serves both as a cross-check of the test system as well as a means of relating activities. Quantitative structure–activity relationships have been established enabling identification of trends and processes of importance to small molecule–DNA interactions. We have developed QSARS catering for hydrophobic properties of the molecules, chemical reactivity at the electrophilic nitrogen, steric effects on all three side chains and intercalative ability of planar polycyclic aromatics. Based on the mutagenic activity of many structural and regio-isomers, the QSARs are accurate to the point of identifying impacts of small structural changes on ease of groove binding or of intercalation. They can also be used to focus on bacterial metabolic processes and mechanisms of DNA damage and to identify substructure potentially deleterious to DNA binding and reaction with DNA, as well as to predict, in advance, the mutagenic properties of new congeners.

All this is made possible on account of their anomeric substitution at nitrogen with both a donor alkoxyl and an acceptor acyloxyl group. The electronegativity of oxygen atoms at nitrogen results in greatly reduced amide resonance and nitrogens that are quite atypical of those in more common primary, N-alkyl secondary and N,N-dialkyl tertiary amides. Their pyramidality, combined with the anomeric interaction between the nitrogen substituents, both properties of which we have developed a thorough understanding, are entirely responsible for their reaction, directly, with nucleophilic sites in DNA.

Supplementary material is available online.

The data used in this account are supplied as Electronic Supplementary Material.

The author declares no conflict of interest.

This work did not receive any specific funding.

The skeletal editing agent 42 has recently been shown to be a direct-acting mutagen in the Ames II assay.^[108] Qualitative results from the Ames II method, which detects point mutations in a mix of S. typhimurium strains, are generally in good agreement with those from the full plate Ames methodology using TA100.^[109] Contrary to Levin’s recently published view,^[108] quantitative Ames II results for 42 with S. typhimurium TA7001-7006 cannot be compared meaningfully to predicted plate counts from the standard Ames test with S. typhimurium TA100.