Chemoselectivity for Alkene Cleavage by Palladium-Catalyzed Intramolecular Diazo Group Transfer from Azide to Alkene

Abstract

Alkenes can be cleaved by means of the (3+2) cycloaddition and subsequent cycloreversion of 1,3-dipoles, classically ozone (O₃), but the azide (R–N₃) variant is rare. Chemoselectivity for these azide to alkene diazo group transfers (DGT) is typically disfavored, thus limiting their synthetic utility. Herein, this work discloses a palladium-catalyzed intramolecular azide to alkene DGT, which grants chemoselectivity over competing aziridination. The data support a catalytic cycloreversion mechanism distinct from other known metal-catalyzed azide/alkene reactions: nitrenoid/metalloradical and (3+2) cycloadditions. Kinetics experiments reveal an unusual mechanistic profile in which the catalyst is not operative during the rate-controlling step, rather, it is active during the product-determining step. Catalytic DGT was used to synthesize N-heterocyclic quinazolinones, a medicinally relevant structural core. We also report on the competing aziridination and subsequent ring expansion to another N-heterocyclic core structure of interest, benzodiazepinones.

Nitrogen is found extensively in organic molecules: natural products, pharmaceuticals, and materials. Thus, the construction of C–N bonds is a continual core goal of synthetic organic chemists. Organic azide (R–N₃) reactions are a common strategy for incorporating nitrogen into target molecules of various sizes and complexities. This broad utility is due to the immense diversity of azide reactivity: 1,3 dipole, nitrene, nucleophile, electrophile, and radical.^[1] Although this grants organic azides synthetic versatility, it also necessitates the development of methods to control selectivity. In the context of thermolytic azide/alkene reactions, the chemoselectivity challenge is two-fold (Scheme 1a, top). There is an initial divergence of mechanistic paths: nitrene formation versus (3+2) cycloaddition. Nitrenes often react with alkenes to form aziridines, but are so energetic they tend to indiscriminately aminate at C–H bonds as well.^[2] The (3+2) cycloadditions form triazolines, some of which are isolable, but most rearrange to form imines and aziridines with the expulsion of N₂.^[3] However, Huisgen et al. observed that triazolines can also rearrange through N–N bond cleavage,^[4] and Fusco et al. discovered that both N–N and C–C bond cleavage can occur.^[5] Regitz et al. categorized these mechanisms as azide to alkene DGT, as imines and diazos (R–N₂) result.^[6] This is analogous to alkene ozonolysis in which a carbonyl and carbonyl oxide (R–O₂) form. However, cleavage of the alkene C–C bond subsequent to azide (3+2) cycloaddition is usually disfavored, in contrast to the ozonolysis of C–C bonds. Chemoselectivity for azide to alkene DGT with C–C bond cleavage has thus far been based solely on substituent effects. DGT may be favored over other pathways with sufficiently electron-poor azide diazo donors (R=SO₂R³ or POR³₂) and electron-rich alkene diazo acceptors (R¹=N or O).^[6a,7] Catalyst-based chemoselectivity for DGT is desirable, as it could allow for reactivity comparable to ozonolysis, but with the incorporation of nitrogen into the products, namely imines and potentially synthetically versatile diazos.

Scheme 1.

Chemoselectivity challenge of azide/alkene reactions, metal catalysis solutions and this work.

Significant strides have been made in controlling the chemoselectivity of azide/alkene reactivity with metal catalysts (Scheme 1a, bottom). It is well established that catalysts (M=Co, Fe, Ru, Rh) can selectively cleave azides to nitrenoids/metalloradicals and impose control for aziridination (Scheme 1a, left).^[8] Although a vast amount of research has been aimed at metal-catalyzed azide/alkyne (3+2) cycloadditions (M=Cu, Ru, Ag),^[9] there are few reports of metal-catalyzed azide/alkene (3+2) cycloaddition reactions. Rare earth metals, Ce and Sm, are thought to catalyze cycloadditions with subsequent decomposition of the triazolines to imines (Scheme 1, center).^[10] Pd^II-catalysts can also affect azide/alkene reactivity; however, it is unclear if there is a consistent mechanism involved. Migita et al. reported the PdCl₂(PhCN)₂-catalyzed allylic ether isomerization with subsequent azide cycloaddition and rearrangement to imines.^[11] Recently, Ramasastry et al. reported Pd(OAc)₂ altered the chemoselectivity of a thermal azide/alkene ring-closing reaction to form the pyridine ring of β-carbolines.^[12] However, all of these transformations involve only the expulsion of N₂. Herein, we report a Pd^II-catalyzed cycloreversion to achieve azide to alkene DGT with C–C bond cleavage (Scheme 1, right).

We centered our studies of catalytic azide to alkene DGT on the reactions of carbamoyl azides 1 with Pd^II-complex PEPPSI^™-IPr. (Scheme 1b). The kinetics of catalytic DGT are fascinating in that the reaction is overall zeroth order in Pd, indicating that the catalyst effects the product-determining step and not the rate-controlling step. Based on our evidence and literature precedence we propose the plausible key mechanistic step to be a Pd^II-promoted β-carbon elimination/cycloreversion from an intermediate cycloadduct. Through this, we chemoselectively synthesized quinazolinones 2. Under noncatalyzed thermolytic conditions we synthesized benzodiazepinones 3 through the ring expansion of azirinoquinazolinones 4. Our synthetic focus was in accordance with the noteworthy strategy of employing azide/alkene cycloadditions/rearrangements for the synthesis of other N-heterocyclic targets.^[12,13] Both quinazolinones and benzodiazepinones are core structures of many bioactive molecules of interest and new methods for their construction are continually sought.^[14]

Our studies commenced with heating a solution of 1a in [D₈]toluene to 130°C in an NMR tube. (Table 1, see section (a) in Scheme, and entry 1). We were initially not hypothesizing any of the DGT product 2a to form as this was neither a sulfonyl/phosphoryl azide nor an enamine/enol ether reaction. Nor were we expecting enamine product 3a as this would appear to indicate an unlikely vinylic C–H nitrene amination. A thermolytic Curtius rearrangement of an acyl azide should be a concerted process that does not involve an intermediate nitrene.^[15] However, after 14 h, 1a was fully consumed and, surprisingly, a mixture of primarily 2a (29%) and 3a (48%) was observed with quantitative ¹H NMR (qNMR) along with a very minor quantity of imine 5a (Table 1, entry 1). Formation of 2a could be rationalized as the carbamoyl azide is still an electron-poor diazo group donor and the alkene is ortho to the secondary aniline and therefore could still serve as a relatively electron-rich diazo acceptor. Rationale for the formation of 3a remained elusive.

Table 1.

Thermolysis reaction and catalyst optimization.

Entry	Catalyst	2a[a]	3a	4a	5a
1	none[b]	29%	48%	<1%	1%
2	none	31%	<1%	46%	1%
3	PdCl2(PhCN)2	19%	27%	1%	10%
4	Pd(OAc)2	16%	27%	6%	9%
5	PdCl2	27%	20%	17%	4%
6	PdCl2(PPh3)2	54%	2%	10%	4%
7	PEPPSI™-IPr	68%	1%	5%	5%
8	PEPPSI™-IPr[c]	59%	2%	6%	4%
9	PEPPSI™-IPr[d]	62%	1%	6%	3%

^[a]Product yields;

^[b]time=14 h;

^[c]5 mol%;

^[d]20 mol%.

Next, we reacted 1a for just 2 h and in this case 1a was already 94% consumed, and a mixture of 2a (31%) and aziridine 4a (46%) was obtained (Table 1, entry 2). Because an approximately equimolar quantity of 4a formed prior to 3a, we proposed that 3a results from the ring expansion of 4a. Indeed, Woerner et al. reported a single example of an analogous thermal aziridine ring expansion.^[16] Subsequent attempts to isolate 4a with silica gel column chromatography were unsuccessful. Instead, we isolated urea 6a, a plausible hydrolysis product (see the Supporting Information).^[17] Thus, our structural assignment of 4a as the azirinoquinazolinone was predicated on both the observed ring expansion to 3a, as well as the isolation of 6a. Also, the overall mass balance of identifiable products was non-ideal; thus, we hypothesized a Curtius rearrangement was diverting a minor portion of 1a to unidentified side products (see the Supporting Information).^[18]

With the relevant products of the thermolysis reaction identified we began testing Pd^II catalysts for their effect on chemoselectivity (Table 1, section (b) in Scheme). PdCl₂(PhCN)₂ (Migita’s conditions)^[11] altered the product distribution significantly, decreasing 2a and promoting formation of 3a from 4a or producing 3a through an alternative mechanism (Table 1, entry 3). Interestingly, the yield of 5a also increased considerably. Pd(OAc)₂, (Ramasastry’s conditions)^[12] produced a similar product distribution to PdCl₂(PhCN)₂ (Table 1, entry 4). PdCl₂ improved the mass balance compared to the first two Pd^II catalysts (Table 1, entry 5). However, no significant chemoselectivity was achieved with the initial entries. PdCl₂(PPh₃)₂ was selective for DGT, producing 2a in 54% yield, favoring it over the aziridination/ring expansion pathway 4.5:1 (2a:(3a+4a)) (Table 1, entry 6). PEPPSI^™-IPr improved the yield and chemoselectivity for 2a (68%, 11.3:1) (Table 1, entry 7). Altering the loading of PEPPSI^™-IPr to 5% and 20% each decreased the yield and selectivity slightly (Table 1, entries 8 and 9).

We performed reaction progress kinetics with qNMR to assess the general kinetic behavior of these unique catalytic and thermolytic reactions (Figure 1). The PEPPSI^™-IPr-catalyzed and the thermolysis reactions were monitored until maximum yields of 2a were observed. This occurred between 120 and 150 min for both reactions, at ≈95–97% conversion of 1a, as we observed a minor degradation of 2a at further time points. The plot illustrates the stark chemoselectivity achieved with PEPPSI^™-IPr, but interestingly we did not observe a change in the rate of 1a consumption. Rather, the rate of 2a formation drastically increased while the rate of 4a formation decreased.

Figure 1.

Reaction progress kinetics.

We expanded our kinetic analysis by performing a series of initial rate measurements. The initial rates of 1a conversion for the thermolysis reaction (ν₁a) and the PEPPSI^™-IPr reaction (ν_1a(Pd)) were approximately equal, as observed with reaction progress kinetics (Table 2, entries 1 and 2). Halving [PEPPSI^™-IPr] produced an initial rate within error of ν_1a and ν_1a(Pd), supporting an overall zeroth order in Pd (Table 2, entry 3). The presence of catalyst only affected the initial rates of formation of 2a (ν_2a) and 4a (ν_4a). Halving [1a] resulted in an approximately two-fold reduction in ν_1a(Pd), supporting an expected first order reaction in azide (Table 2, entry 4). We synthesized azide 7 with no alkene as a proxy for measuring the rate of the hypothesized Curtius rearrangement that may compete with productive (3+2) cycloaddition. The initial rate of conversion of 7 (ν₇) under thermolytic conditions was measured (Table 2, entry 5). Remarkably, the sum of ν₇ with the rates of product formation ν_2a and ν_4a was within error of ν_1a (−2.64e⁻⁶ ± 1.16e⁻⁷ MS⁻¹ ≈ −2.61e⁻⁶ ± 1.46e⁻⁷) This accounts for the full mass balance. When 7 was reacted in the presence of PEPPSI^™-IPr the rate of conversion increased slightly (Table 2, entry 6). This may indicate that some nonproductive interaction of Pd and azide could occur prior to cycloaddition.

Table 2.

Initial rate kinetics experiments.

Entry	[1a]	[Pd]	ν1a [MS−1]	ν2a [MS−1]
1[a]	[0.01]	–	−2.61e−6 ± 1.46e−7	8.14e−7 ± 5.46e−8
2[b]	[0.01]	[0.001]	−2.51e−6 ± 7.90e−8	1.74e−6 ± 2.65e−7
3	[0.01]	[0.0005]	−2.59e−6 ± 5.77e−8	1.79e−6 ± 2.38e−7
4	[0.005]	[0.001]	−1.31e−6 ± 6.26e−8	9.63e−7 ± 5.52e−8
Entry	[7]	[Pd]	ν7 [MS−1]	–
5	[0.01]	–	−5.42e−7 ± 3.24e−8	–
6	[0.01]	[0.001]	−6.89e−7 ± 3.21e−8	–

^[a]ν_4a=1.28e⁻⁶ ± 9.70e⁻⁸;

^[b]ν_4a(Pd) below detection threshold for Pd-catalyzed reactions.

Next, we broadened our studies to other azide 1 substrates (Scheme 2). We first targeted derivatives substituted para to the alkene with groups of varying electronic character. Substrates 1b–1f were subjected to both sets of reaction conditions. Yields were measured with qNMR at two timepoints (2 and 2.5 h) to gauge if a significant change in rate or decomposition occurred. The duration of the thermolysis reactions was then extended (overnight) to observe the ring expansion of 4 to 3. Strikingly, the chemoselectivity for the thermolysis of 1 was greatly influenced by the electronic character of the ring. More electron-rich substrates 1b and 1c had decreased preference for DGT and more electron poor substrates 1d–1f had increased preference for DGT products. 1f, the most electron-poor substrate tested, favored DGT over aziridination 1.5:1. We established a Hammett linear free energy relationship (LFER) that illustrates this electronic bias for chemoselectivity (Figure 2). In addition, σ_para correlated very well with the observed chemoselectivity (R²=0.99) and ρ=0.68. Despite the increased electronic bias for aziridination with more electron-rich substrates, PEPPSI^™-IPr still catalyzed the formation of 2b to a yield of 65% and chemoselectivity 6.4:1. 2c–2f yields were also good (67–71%) and good chemoselectivity was achieved. For the reaction of 1f, 2f was the sole observed product.

Scheme 2.

Substrate scope: [a] yields assessed at both 2 h and 2.5 h, and maximum yield is reported; [b] see the Supporting Information for specific substrate reaction time; [c] chemoselectivity ratios calculated with maximum yields from any timepoint; [d] 2 h; [e] 2.5 h; [f] same yield at both timepoints; [g] 9% 4e remained; [h] 14% likely 5h obtained; [i] 12% 4i remained; [j] timepoints taken at 2 h and 20 h due to slower reaction, 2j not isolated but downfield ¹H NMR peak at δ=8.54 indicative of quinazolinone product; [k] ratio given is 2j:5j, 5j being the H-shift product equivalent to 1a.

Figure 2.

Chemoselectivity Hammett LFER.

We synthesized 1g with an electron deficient p-NO₂ benzyl-group as well as 1h with N–Me group to test if the N–Bn protecting group has a role in catalyst activity. Both substrates behaved similarly to 1a. PEPPSI^™-IPr catalyzed the formation of 2g and 2h, and ring expansion products 3g and 3h were obtained in the uncatalyzed reaction. Next, we synthesized derivatives with alterations to the alkene; butenyl azide 1i and styrenyl azide 1j. Azide 1i reacted in an analogous manner to 1a, however the catalyzed reaction gave slightly reduced yield of 2i (46%). Under thermolytic conditions 1j selected significantly for the H-shift product 5j instead of 2j and 4j. 2j was speculated to have formed in only a minor quantity based on the ¹H NMR spectrum (see the Supporting Information). However, PEPPSI^™-IPr was still slightly active and increased the yield and chemoselectivity for DGT product 2j.

We propose the following mechanisms to account for our kinetic data and synthetic results. Under thermolysis conditions rate-controlling (3+2) cycloaddition (A) leads to intermediate triazoline I and competing Curtius rearrangement (B) leads to intermediate aminoisocyanate II (Scheme 3a). I is favored 3.9:1 over II based on our initial rates measurements of 1a. Based on reported triazoline decomposition mechanisms, I ring opens (C) to the diazonium III, at which point the chemoselectivity divergence of interest occurs.^[6a] C–C bond cleaving DGT (D) leads to quinazolinone 2a with elimination of CH₂N₂, which may further decompose under the reaction conditions.^[19] However, III favors 3-exo-tet aziridination (E) with expulsion of N₂ to form 4a, which undergoes relatively slow rearrangement to benzodiazepinone 3a through a plausible 1,2-hydride shift (F).^[16] Since the product ratios for our Hammett LFER analysis were calculated as (2:(3+4)) and ρ=0.68, it indicates that negative charge builds as 2 forms (Scheme 3b). Therefore, the electronic influence on chemoselectivity arises from the rings’ ability to stabilize or destabilize the negative charge that builds at the benzylic position, which alters the barrier of DGT. On the other hand, 3-exo-tet aziridination should be less impacted by changes to the ring electronics (Scheme 3c). Thus, the σ_para LFER observed is likely a reflection of mostly the ΔΔG^‡ for DGT.

Scheme 3.

Plausible PEPPSI^™-IPr catalysis and thermolysis mechanisms.

In the presence of PEPPSI^™-IPr, we propose an interaction between either intermediate I or III and Pd alters the ΔΔG^‡ between the DGT path and aziridination. Because the reaction is zeroth order overall in Pd, kinetics data is unable to lend evidence as to how this process works. Presumably, rate-controlling cycloaddition produces minute concentrations of I, and then I or III reacts with Pd in a bimolecular process. If Pd irreversibly traps I or III then the DDG^‡ between the two paths may result either from the barrier of aziridination increasing, the barrier of DGT decreasing, or a combination of both. If the Curtin–Hammett principle is operative, and binding between an intermediate and Pd is reversible, then the barrier for DGT from the Pd-bound intermediate may be lower relative to the noncatalyzed DGT (Scheme 3d). We propose this to be most plausible and triazoline I exchanges reversibly with 3-chloropyridine binding to Pd at the once terminal nitrogen of the azide to form I·Pd(G).^[20] From this species a β-carbon elimination may initiate a cycloreversion process to cleave the C–C bond and eliminate 2a from the complex (H) (Scheme 3e).^[21] This forms IV with CH₂N₂ bound to the Pd, which may either exchange with the pyridine to reform PEPPSI^™-IPr (I) or the cycle continues by exchange of CH₂N₂ with I (J).^[22]

In summary, we have presented a Pd^II-catalyzed intramolecular DGT from azides to alkenes. Catalyst-based chemoselectivity is achieved after the rate-controlling step, plausibly through interaction with the intermediate triazoline. This represents a new mode of metal-catalyzed chemoselective control in azide/alkene reactions. We applied the DGT reaction to synthesize medicinally relevant N-heterocyclic quinazolinones and demonstrated the competing aziridination may be a useful method to synthesize benzodiazepinones.