An organophotocatalytic late-stage N–CH3 oxidation of trialkylamines to N-formamides with O2 in continuous flow

We report an organophotocatalytic, N–CH3-selective oxidation of trialkylamines in continuous flow. Based on the 9,10-dicyanoanthracene (DCA) core, a new catalyst (DCAS) was designed with solubilizing groups for flow processing. This allowed O2 to be harnessed as a sustainable oxidant for late-stage photocatalytic N–CH3 oxidations of complex natural products and active pharmaceutical ingredients bearing functional groups not tolerated by previous methods. The organophotocatalytic gas–liquid flow process affords cleaner reactions than in batch mode, in short residence times of 13.5 min and productivities of up to 0.65 g per day. Spectroscopic and computational mechanistic studies showed that catalyst derivatization not only enhanced solubility of the new catalyst compared to poorly-soluble DCA, but profoundly diverted the photocatalytic mechanism from singlet electron transfer (SET) reductive quenching with amines toward energy transfer (EnT) with O2.


Introduction
A quintessential theme in medicinal chemistry is probing structure activity relationships. While strategies such as de novo and diversity-oriented synthesis are powerful tools to achieve this task, late-stage functionalization (LSF) has gained traction over the past decade as it offers a quicker route to access libraries of complex bioactive molecules from a dened core structure. 1,2 Among the myriad of methods that are applied in LSF, C-H functionalization is undeniably an attractive and potent addition to a synthetic chemist's arsenal, given the ubiquity of C-H bonds in molecules. 1-3 This umbrella term stretches over traditional transition metal catalysis to alkali and base-metal catalysis to organocatalysis and photocatalyticallyenabled transformations. Recent examples demonstrate the value of C-H functionalization of simple and complex amides through ionic 4 or radical 5 mechanisms. Trialkylamines are especially important targets since they are well represented in the alkaloids, a family of potent bioactive molecules that has shaped the natural sciences. 6,7 N-CH 3 groups are attractive loci for C-H functionalization in pharmaceutical research, since incremental structural variations carry substantial pharmacological effects (Fig. 1), for example in bioactivities of opiates. 8 However, C-H bonds a to N are relatively inert. Access to derivatives was historically carried out stepwise, leveraging the nucleophilicity of the N atom, usually requiring initial demethylations of trialkylamine N-CH 3 groups to free N-H secondary amines for subsequent transformations. 9 That is until the renaissance of single electron transfer (SET) redox methods, partly driven by photoredox catalysis, which have revolutionized practices in organic synthesis. 10 This allowed direct C-H functionalizations a to N, of benzylic amines with nucleophiles 11 and a few examples of aliphatic amines with electrophiles. 12 A direct and highly N-CH 3 -selective LSF of trialkylamines was achieved using stoichiometric quantities of an SET-generated hydrogen atom transfer agent (DABCOc + ). 13 Powerful catalytic processes, 31 we envisaged that continuous ow would allow us to efficiently and safely handle O 2 as a simple, abundant, sustainable terminal oxidant. The rapid uptake of continuous ow reactors in the synthesis of ne materials and pharmaceuticals is worth noting, as is their innovative marriage with visible light irradiation which drastically enhanced the efficiency, sustainability and safety of photochemical processes. 32 Previous organophotocatalytic activations of trialkylamines using O 2 were reported, but those instead targeted (i) N-demethylations of opiates and tropanoids, 33 (ii) endocyclic C-H cyanations a to N 34 or (iii) oxidations of benzylic amines. 35 We were particularly drawn to 9,10-dicyanoanthracene (DCA) as used by Santamaria and co-workers (Fig. 2B). Using DCA as a potent photooxidizing catalyst (E 1/2 [ 1 DCA*/DCAc À ] ¼ +1.99 V vs. SCE), 30 air (O 2 ) as terminal oxidant, and an LiClO 4 additive, they reported variable amounts of N-formyl products (2) in batch, however N-demethylation (nor-amines (3)) competed or dominated reactions. 36,37 Herein, we report a late-stage organophotocatalytic oxidation of N-CH 3 groups that selectively delivers N-formyl compounds (2). Our method leverages mild conditions and continuous ow processing to handle O 2 safely as a terminal oxidant (Fig. 2C). Key to the aforementioned achievements was the design of a novel dicyanoanthracene catalyst that not only enhanced solubility for ow processing, but switched the excited state mechanism from single electron transfer with amines toward energy transfer with O 2 .

Results and discussion
Photocatalyst and process design At the onset, our attempts to use DCA using modied reaction conditions of Santamaria in batch (and in ow) were severely obstructed by its poor solubility in MeCN (i.e., turbidity and sedimentation were observed). The suspended, undissolved photocatalyst was detrimental to photochemistry due to hindering light penetration of the reaction. Furthermore, in continuous ow this oen led to ow channel blockages and longer reaction times (for details, see ESI † le). Thus, design of a catalyst with enhanced solubility was required (Fig. 3).
Intuitively, introduction of polar substituents improves solubility of compounds in polar aprotic solvents. Nitro-and sulfonic acid-groups are good choices for polyaromatic compounds as the synthetic process to access them is straightforward. Glöcklhofer and co-workers reported the synthesis of a dinitro derivative of DCA with improved solubility. 38 On the other hand, sulfonic acids carry the advantage of further derivatization via their sulfonyl chlorides. Inspired by intermediates reported in the synthesis of a water-soluble DCA analogue, 39 we began our catalyst synthesis (Fig. 3). Anthraquinone-2,6-disulfonic acid 4, commercially supplied or easily synthesized from cheap anthraquinone, 40 was reduced by activated Zn in aq. (NH 4 ) 2 CO 3 to afford anthracene-2,6disulfonic acid 5 in good (65%) yield aer acidic workup and recrystallisation from aq. KCl. Electrophilic bromination of the central ring of 5 gave 6 in high (80%) yield. At this stage, our synthesis deviated from the literature cyanation which digested the crude product (containing CuCN) in conc. HNO 3 and liberated toxic HCN gas. However, both Rosenmund von-Braun and Pd-catalysed cyanations failed to cyanate 6 due to its poor solubility in organic solvents. Tohnai and co-workers had reported that the derivatization of anthraquinone disulfonic acids (ADS) as their organic ammonium salts (i.e., n-heptyl and npentyl) prevented p-stacking interactions of ADS as observed by crystallography. 41,42 Instead of ammonium salts which would hinder characterization and reaction workup, we achieved this covalently with sulfonamides. Therefore, 6 was derivatized to increase its solubility in polar aprotic organic solvents and to increase prospects for successful cyanation. Chlorination of 6 with POCl 3 and subsequent trapping of 7 with secondary amines of various chain lengths gave 9,10-dibromoanthracene-2,6-disulfonamides (DBAS) 8a, 8b and 8c in 87, 90 and 74% yields, respectively. Pleasingly, Rosenmund von-Braun cyanations under microwave-assisted (15 min) or thermal (see ESI †) heating afforded 9,10dicyanoanthracene-2,6-disulfonamides 9a, 9b (DCAS) and 9c as 'brilliant yellow' solids in 66%, 89% and 26% yields, respectively. We note that our entire synthesis to 9 is carried out on gram scale, with straightforward purication via recrystallisation instead of chromatography. Photocatalyst 9b (henceforth coined 'DCAS') was progressed to evaluation in reactions since it: (i) displayed the highest solubility in MeCN (1.900 AE 0.100 mg mL À1 vs. 0.340 AE 0.006 mg mL À1 for DCA) consistent with its calculated physical property values (which showed that it was the least lipophilic and had the highest topological polar surface area, see ESI †), 43 and (ii) was obtained in the highest overall yield (42% over 5 steps).

Studies using a homogeneous liquid ow photoreactor
Next, DCAS was tested under some initial photocatalytic ow conditions (Table 1) in a commercial tubular coil continuous ow photoreactor (Vapourtec Ltd R-series/UV-150). Using 1a (12 mM) as our substrate and 5 mol% of DCAS at rt, a maximum yield of 25% for 2a (with 4 : 1 of 2a : 3a selectivity) was obtained under recycling conditions (90 min) no matter whether dry air, O 2 , or (1 : 1) N 2 /O 2 were used (entry 2). The absence of catalyst (entry 2) or O 2 led to no reaction. We found out that in the absence of LiClO 4 , single pass conditions gave a similar yield (25%) and with much improved selectivity for 2a (entry 4, 3a was not detected). When the temperature was increased to 40 C, the yield improved to 40% (entry 5). Under similar conditions but employing DCA as catalyst afforded 2a in 15% yield, conrming superiority of DCAS under ow conditions. A batch reaction mimicking Santamaria and co-workers' condition (entry 8) afforded a complex reaction mixture (see ESI †). Our previously reported batch anaerobic conditions for SET oxidation of Nalkyl tetrahydroisoquinolines with [Ru(bpy) 3 ] 2+ photocatalysis 11b in batch (entry 7) gave no reaction, and when the more potent photooxidizing SET catalyst [Ru(bpz) 3 ] 2+ was used only traces of 3a were observed. As such and due to cost of the catalysts, we did not examine these any further in ow.
When under N 2 protection (entry 3), a purple coloration in the post-reactor owing reaction mixture was observed (see ESI †) which hinted at formation of DCASc À . We note that the related parent structure DCAc À is well-known to be purple in color. 44 When the purple post-reactor reaction mixture was collected and exposed to air, immediate discoloration back to yellow was observed. From these observations, we had initially assumed  a reductive SET quenching of 1 DCAS* by the amine, as originally proposed by Santamaria and co-workers (Fig. 2B). 36,37 However, this was later refuted (see the Mechanistic studies for details). Based on this mechanistic assumption, we reasoned that formation of 2a reached its upper limit due to limiting oxygen solubility at ambient conditions in the tubular reactor, preventing catalyst turnover. The solubility of O 2 in an O 2 -saturated solution of MeCN is 8.1 mM, 45 and considering the theoretical requirement of 2 equiv. O 2 to remove 2 electrons from the trialkylamine, mass transfer limits full conversion of a reaction mixture containing 12.0 mM trialkylamine (later in the revised mechanism, we nd that [O 2 ] is still a limiting factor for the reaction yield).

Studies using a gas-liquid ow photoreactor
Considering the abovementioned observations, we opted for a photoreactor designed for biphasic gas-liquid reactions. A commercial microuidic continuous ow photoreactor (Corning Lab Photoreactor©) designed for excellent mixing via turbulent slug ow allowed us to safely operate up to 60 C and 8 bar backpressures. The hazard of the ammable reaction mixture was safely contained by the thermal isolation of the ow path and the small volume of reaction mixture (2.7 mL) at any given time. A summary of reaction condition optimization is shown in Table 2 (see ESI † for full optimization). Transferring conditions from the previous tubular reactor (Table 1, entry 5), 2a was afforded in 22% yield ( Table 2, entry 1), as expected since the decreased yield exactly consists with (is proportional to) the decreased residence time (R T ). However, the yield almost doubled when 395 nm LEDs were used (entry 2), which accorded with a higher extinction coefficient of DCAS's UV-vis band at ca. 395 nm compared to its 420 nm band (vide infra). At 24 mM 1a and double the residence time, the yield increased to 44% (entry 7). At 48 mM of 1a the yield decreased to 24% (entry 8), presumably again due to the limiting [O 2 ]. At T ¼ 60 C and 24 mM 1a, the yield of 2a marginally improved to 46% (entry 9). The inherent back pressure on the ow by the microuidic module was sufficient to ensure precise, reproducible, low ow rates (down to 0.1 mL min À1 ) up to 60 C. To our delight, tropine 1b afforded 2b in 60% under reaction conditions at T ¼ 40 C and R T ¼ 27 min (entry 10) despite its free 2 alcohol typically prone to oxidation under similar oxidative conditions. [23][24][25][26][27][28]46 Decreasing catalyst loading decreased the yield (entries 11 and 12). Like the case of substrate 1a, a marginal increase of yield to 61% occurred at 60 C (entry 13). At this stage, we explored the effect of a back pressure (8 bar) to evaluate higher O 2 solubility (entries 14-16). At lower backpressures, the ow was heterogeneous slug ow but at 7-8 bar, homogenous ow was observed indicating full solubilization of O 2 and higher dissolved [O 2 ]. At 7-8 bar, doubling the concentration to 48 mM or using a residence time as short as R T ¼ 6.8 min negatively impacted the yield of 2b (entries 14 and 15), but we found that yield (61%) was preserved at R T ¼ 13.5 min (entry 16 vs. 13). This doubled productivity of 2b to 0.65 g per day which was the upper limit of the gas-liquid organophotocatalytic ow reaction in this system. Next, we tested the scope of the reaction (Table 3). Since isolations of polar formamides were oentimes challenging due to the N-formyl group being a weak chromophore, the following discussion deems 1 H NMR yields more representative of reaction efficiency. Compounds 2c (59%) and 2d (67%) were obtained from natural products tropane and (free alcohol-bearing) atropine. Even scopolamine, which has a free alcohol, an ester, and an epoxide, afforded 2e in 62% yield with no norscopolamine detected, albeit requiring 2 passes through the reactor (total R T ¼ 27 min). This contrasts with Santamaria and co-workers' conditions using DCA and without LiClO 4 , which afforded a 1 : 1 mixture of 2e : nor-scopolamine. 36,37 Compared to 2b, the yield of 2f was lower (34%) presumably due to the presence of the Si protecting group known to stabilize radicals and quench excited photosensitizers via different pathways. 47 Benzoyl-containing compound 2g was afforded in good (73%) yield. Electron-poor (-CF 3 ) and electron-rich (-OMe) substituents on the benzoyl group were tolerated equally, affording 2h (56%) and 2i (58%) respectively. We note both 2g and 2i are natural products; novel tropanoid compound 2g was recently isolated from Pellacalyx saccardianus and our method corroborated its proposed structure. 48 Compound 2g (confoline) was isolated from Convolvulus subhirsutus and our method accessed it from convolvamine in a single step (in the literature, semi- synthesis of 2i was achieved by formylation of norconvolvamine, hence a demethylation step from convolvamine culminates in a two-step process). 17 Compounds 2j to 2o were obtained from piperazines as common API fragments (such as those present in sildenal and danooxaxin). 49 Despite having 3 possible sites for functionalization (one exocyclic N-CH 3 and two endocyclic N-CH 2 -R sites), selective oxidation at the N-CH 3 (exo-: endo-¼ 5.7 : 1 for 2j, 3.4 : 1 for 2k, 3.7 : 1 for 2l, and 6.5 : 1 for 2p, see ESI †) was apparent, affording N-formyl compounds in respectable yields. Despite the modest yields of products 2l (55%), 2m (21%), and 2n (30%) (as well as 2h), we were surprised by the tolerance of halogen-bearing substrates under the reaction conditions. Especially, given the aforementioned putative presence of DCASc À via reductive quenching of 1 DCAS* by trialkylamines (well known for DCA's case) 50,51 and given that photoexcited radical anions are known to reductively cleave aryl halides and other strong bonds. 50-53 C-F bonds and N-Ts groups are also prone to reductive cleavage under reductive photocatalysis 54 or by photoexcited super electron donors. 55 A simple piperidine 2p (39%) was also tolerated. Our success with 2b, 2d, 2e, and 2p whose precursors bore free alcohol groups encouraged us to explore more complex molecules. Gratifyingly, conditions were successfully applied to macrolide antibiotics with dense functionalities (free alcohols, an oxime ether, and a ketone). Erythromycin, clarithromycin and roxithromycin afforded 2r, 2s, and 2t in 61%, 44%, and 24% yields, respectively. However, benzylic amines and trialkylamines containing benzylic alcohols or free carboxylic acids such as 1u, 1v and 1w, were unsuccessful. Benzaldehyde formation (C-N cleavage, possibly via endocyclic iminium ion formation and then hydrolysis) and intractable complex reaction mixtures were observed for these substrates.

Mechanistic studies
Cyclic voltammetry (CV) revealed DCAS (E 1/2 [DCAS/DCASc À ] ¼ À0.59 V vs. SCE) is substantially easier to reduce than DCA (E 1/ 2 [DCA/DCAc À ] ¼ À0.98 V vs. SCE), due to the electronwithdrawing sulfonamide groups at the 2,6-positions (Fig. 4,  le). UV-vis absorption and emission spectra were measured for DCA and DCAS (Fig. 4, right) and their comparison revealed that the 2,6-sulfonamides hardly affect the absorptive or emissive proles of the dicyanoanthracene core. In both cases, overlap of the longest wavelength absorption band (l max ¼ 422 nm) and shortest wavelength emission band (l max ¼ 435 nm) allows to approximate E 0-0 for the singlet excited state (z2.90 eV). Taking this value together with measured redox potentials, the photocatalyst excited state oxidation potentials were approximated by Table 3 Scope of organophotocatalytic flow N-CH 3 to N-formyl oxidation a R T ¼ 27 min, O 2 (ambient pressure). b 2 passes. c 12 h recycling. d 12 mM. e 6 mM. Yields in parenthesis determined by 1 H NMR of the reaction mixture using 1,3,5-TMB as internal standard. c.r.m. ¼ complex reaction mixture. a derivative of the Rehm-Weller equation. 56 1 DCAS* (E 1/ 2 [ 1 DCAS*/DCASc À ] ¼ +2.31 V vs. SCE) is a notably more potent photooxidant than 1 DCA* (E 1/2 [ 1 DCA*/DCAc À ] ¼ +1.93 V vs. SCE). Our initial hypothesis thus continued to align with the SET mechanism proposed by Santamaria (Fig. 5). 36,37 In this premise, 1 DCAS* was assumed to behave like 1 DCA* which underwent reductive quenching by trialkylamine 1, and oxidation of DCASc À by O 2 regenerated DCAS. Deprotonation of radical cation 1 0 and radical combination of the a-amino radical and superoxide would ultimately afford N-formyl product 2. We note SET reactions were also proposed as the main pathways for trialkylamine activations by thiazine and uorescein organophotocatalysts, either via oxidation to N-oxides or via Ndemethylations. 57 To test this initial hypothesis, two control batch experiments with stoichiometric (2.0 equiv.) DCA and DCAS were conducted under strict N 2 protection in PhCN solvent to promote solubility. Both afforded clean conversion of 1a to a 1 : 1 mixture of 1a : 3a (Fig. 6A), although DCA's reaction required >3.5Â reaction time due to poorer solubility. Upon irradiation, the reaction mixtures changed from a pale yellow color to dark purple (Fig. 6B). Removal the light and exposing to air, the colors of reaction solutions quickly reverted to yellow (consistent with aforementioned observations of the ow reaction under N 2 ). The UV-vis spectra of DCAc À is well studied in the literature, 51,53 and it is known to be purple in color. 44 We conrmed the presence of DCASc À spectroscopically by matching the spectra of a sample of DCAS treated by cathodic electrolysis to that treated photochemically in the presence of a trialkylamine reductive quencher (see ESI † for details). Both gave a new, broad absorption spectrum at the visible-green region (l max z 544 nm, Fig. 6C), thus an apparent purple color.
A Time-Dependent Density Functional Theory (TD-DFT) calculation of the UV-vis transitions of DCASc À consisted with this green absorption peak (l max ¼ 547 nm). The detection of these radical anions together with N-demethylation reaction conrmed that the SET oxidation of 1 to 1 0 by the organophotocatalysts was possible, at least under anaerobic conditions. Considering the preceding discussion supportive of an SET mechanism in Fig. 5 and catalysts' redox potentials, one would expect 1 DCAS* to undergo more rapid uorescence quenching than 1 DCA* by trialkylamines. Very surprisingly, the opposite was clearly true (Fig. 7). The Stern-Volmer rate constant for quenching of 1 DCAS* by 1a (k q ¼ 1.44 Â 10 9 M À1 s À1 ) was two orders of magnitude smaller than that of 1 DCA* (k q ¼ 1.69 Â 10 11 M À1 s À1 ). 58,59 Presumably, either (i) the 2methoxyethyl groups of DCAS affects the kinetics of bimolecular quenching by sterically obstructing the approach of trialkylamine, or (ii) aggregation of DCA accelerates reductive quenching by trialkylamines 60 where DCAS exhibits a different kind of aggregation in solution. 61 To probe the mechanistic role of the structural changes on the catalyst, and to rationalize the unexpected trend between the order of redox potentials of 1 DCA* and 1 DCAS* vs. their uorescence quenching rates, DFT and TD-DFT calculations were performed. The activation energies (DG ‡ SET ) for the   photoinduced single electron transfer (SET) from 1b to photoexcited dicyanoanthracenes were determined using Marcus theory (Table 4). Aside from free energy (DG SET ), another key parameter of Marcus theory is the reorganization energy (l) which accounts for the properties of the solvent, the size of, and the distance between reacting species. The calculated vertical excitation energies (DG Ex z 3.3 eV) were close to E 0-0 value obtained from optical spectroscopy (vide supra). As expected from the experimentally-determined photocatalyst redox potentials, SET of 1b with 1 DCAS* is 1.3Â more exergonic (DG SET ¼ À49.9 kcal mol À1 ) than with 1 DCA* (DG SET ¼ À37.3 kcal mol À1 ). However, the kinetic barrier is notably (3Â) higher for 1 DCAS* (DG ‡ SET ¼ 13.7 kcal mol À1 ) than 1 DCA* (DG ‡ SET ¼ 4.4 kcal mol À1 ). This agreed with the relatively slower uorescence quenching of 1 DCAS* by trialkylamines. However, this is juxtaposed with the greater synthetic efficiency of the reaction catalysed by DCAS compared to DCA. Taken together, these results show that although SET between excited cyanoanthracenes and trialkylamines can occur under anaerobic conditions, an alternative mechanism must operate for DCAS under aerobic conditions in order for it to deliver higher synthetic efficiencies.
Elsewhere, 1 DCA* is also known as an efficient singlet oxygen sensitizer (k q ¼ 4.3 Â 10 9 M À1 s À1 ) via a photosensitized energy transfer (E n T) mechanism. 62 The high reported quantum yield (reaching 2.0) supports the generation of 2Â 1 O 2 molecules per 1Â 1 DCA*. 62 This quenching rate constant of 1 DCA* by E n T is more than double that of the reductive SET quenching of 1 DCAS* by trialkylamines. Thus, as [O 2 ] increases and approaches that of the trialkylamine ([O 2 ] z [trialkylamine]), singlet oxygen generation dominates in the case of 1 DCAS*. This consists with the increase in yields observed at higher back pressures, temperatures and thus higher dissolved [O 2 ]. The k q for quenching of 1 DCAS* by O 2 was slightly higher (4.89 Â 10 9 M À1 s À1 Fig. 7, le) than that reported for 1 DCA*. 62a Taken together with the uorescence quenching rates (k q ) with trialkylamines, this points to a photochemical mechanistic switchover: 1 DCA* is quenched faster (39Â) by 1a than O 2 , while 1 DCAS* is quenched faster (3Â) by O 2 than 1a. Thus, under aerobic reaction conditions, 1 DCA* favors an SET mechanism while 1 DCAS* favors an E n T mechanism.
We then studied the behaviour of the excited cyanoanthracenes under the aerobic reaction conditions (i.e., catalyst, trialkylamine and O 2 (from air) are all present). This was done by comparing the relative intensity change of light (420 nm) transmitted through the coil of the tubular ow reactor. The principle is as follows: the faster the excited state catalyst can relax to the ground state, the greater the steady-state population of ground state photocatalyst is, leading to more absorption of light (therefore less transmission). The aerated, owing reaction mixture of DCA (5 mol%) + 1a (12 mM), under the conditions of Table 1, entry 6, gave minimal light absorption (Fig. 8A). As discussed earlier, the reductive quenching of 1 DCA* by 1a is even faster than quenching by O 2 and does not directly afford DCA but affords DCAc À whose absorption 51,53 is shied far into the visible green region and thus is not detected by the probe. Regeneration of the ground-state catalyst relies on the oxidation of DCAc À by O 2 , which is comparatively slow. In the absence of 1a, the aerated solution of DCA (Fig. 8B) gave strong light absorption (decrease of transmitted light intensity to roughly half). O 2 no longer competes with 1a and is now the exclusive a Geometry optimization, molecular radius (r) and free energies calculated using DFT (ground state) or TD-DFT (excited state) at CAM-B3LYP (or uB97X-D in parentheses)/6-31++g(2d,p), CPCM(acetonitrile) level of theory (see ESI). b From Stern-Volmer analyses (Fig. 7), in M À1 s À1 . c Vertical excitation energy. d Photoinduced-SET free energy. e Photoinduced-SET activation energy. All free energy units in kcal mol À1 . PC ¼ photocatalyst. For further details, see ESI. quencher, regenerating and sustaining a large steady-state concentration of absorbing DCA via rapid E n T quenching of 1 DCA*. In contrast, the reaction mixture (Table 1, entry 4) of DCAS (5 mol%) in MeCN gave notable light absorption even when 1a (12 mM) was present (Fig. 8C), since quenching of 1 DCAS* by O 2 now outcompetes reductive quenching by 1a, ensuring a larger steady-state concentration of absorbing DCAS. This agrees with the aforementioned differences in quenching rate constants. Finally, the light absorption of an aerated solution of DCAS in the absence of 1a (Fig. 8D) was greater in the absence of competing 1a, and was more pronounced than in the case of DCA (Fig. 8B). This reects the enhanced uorescence quenching of the former with O 2 (for light transmission measurements under N 2 or with 380 nm, see ESI †). The lifetimes of 1 DCA* and 1 DCAS* as measured by Timecorrelated Single Photon Counting (TSCPC) in MeCN under Ar were similar, at 14.5 and 13.8 ns, respectively ( Table 5).
The lifetime of 1 DCA* was 1.8 ns lower in presence of air, while the lifetime of 1 DCAS* was 4.7 ns lower, conrming the slight enhancement of quenching by O 2 (and consistent with the Stern-Volmer k q s of 1 DCAS* and 1 DCA*, vide supra). Further experiments supported the photosensitized E n T quenching of 1 DCAS* as the dominant mechanism, rather than photoinduced SET to afford O 2 c À (Fig. 9A and B). Firstly, when a-terpinene was employed as the substrate, ascaridole was formed in 65% yield as quantied by 1 H NMR. Endoperoxide formation is a hallmark reporter for 1 O 2 through its Diels-Alder [4 + 2]cycloaddition with dienes (Fig. 9A, le), 63 thus conrming 1 DCAS* is capable of 1 O 2 generation. Secondly, the presence of DABCO as an additive inhibited conversion in 1b's reaction (Fig. 9B, right). Despite DABCO's low oxidation potential (E p ox ¼ +0.66 V), 13 this inhibition was not due to its competitive SET reductive quenching of 1 DCAS*, since the quenching rate constant (k q ¼ 7.28 Â 10 8 M À1 s À1 ) conrmed it was even less efficient as a quencher of 1 DCAS* than O 2 or 1a/1b (Fig. 7). Rather, DABCO is a well-known physical quencher of 1 O 2 . 64,65 This was conrmed by a linear correlation (R 2 ¼ 0.997) between the reciprocal relative rate and [DABCO], an experiment designed by Lapi and co-workers. Finally, as proof of the direct xation of oxygen atoms from O 2 gas into trialkylamines, 18 O-2b was detected by HRMS when isotopically-enriched oxygen ( 18 O 2 ) gas was employed in the batch reaction of 1b (Fig. 9B).
In summary, increased efficiency of DCAS over DCA in the reaction is not only attributed to the former's enhanced solubility. The sulfonamide substituents at the 2,6-positions of the dicyanoanthracene markedly decrease the reductive quenching of 1 DCAS* by trialkylamines, compared to that of 1 DCA*. This observation may be explained by a change in the aggregation state of the organophotocatalyst, 61 where the ordered p-stacking of DCA aggregates creates a large effective volume for collisions with amines, while DCAS behaves differently. The distance of psandwich planes for DCA ¼ 3.37Å and the usual range for 2 interacting planes is 3.3 to 3.8Å. 66 In the X-ray diffraction (XRD) structure of DCAS (Fig. 9C), the distance between p-planes of anthracene ¼ 13.60Å and considering that 2r ¼ distance between molecules, this value agrees with 2Â the calculated spherical radii of DCAS in MeCN (Table 4). From this, we tentatively propose that the bulky, freely-rotating sulfonamide substituents sterically inhibit bimolecular (or unimolecular) 59 quenching events with trialkylamines. The smaller O 2 molecules outcompete larger trialkylamines to reach the cyanoanthracene core, diverting the mechanism to 1 O 2 sensitization. This consists with the need for constrained trialkylamine substrates with protruding N-CH 3 groups herein, and may rationalize DABCO's inefficiency as a reductive quencher on steric grounds. 13 A similar "steric-bulk" strategy was recently employed using tert-butyl substituents to prevent an unproductive EDA complexation in a catalytic reaction. 67 In light of all the above, we propose the following mechanism (Fig. 10). Photoexcitation of DCAS affords 1 DCAS* which undergoes E n T with 3 O 2 . The generated 1 O 2 interacts with the trialkylamine via a well-studied exciplex, 62,64,68,69 which can  68 is endergonic, consistent with our DFT calculations of an endergonic free energy (DG ¼ 6.0 kcal mol À1 ). Thus, we deemed SET within the exciplex as the minor pathway. Conversely, HAT within the exciplex was slightly exergonic (DG ¼ À0.1 kcal mol À1 ) suggesting this is the major pathway. Combination of 1 00 with proximally-generated peroxyl radical affords 10 (which could also be accessed by SET oxidation of a-amino radical 1 00 by 1 O 2 followed by combination of 11 with O 2 c À and subsequent protonation is also possible). Finally, liberation of H 2 O from 10 affords 2 and DCAS is regenerated by the reported triplet-triplet annihilation of 3 DCAS* with a second molecule of 3 O 2 . 62 In a recent study by Rovis, Schoenebeck and co-workers on the photocatalytic functionalizations of cyclic trialkylamines, 14e they proposed that a reversible and fast HAT is responsible for their endocyclic selectivity. Our computational studies point to a rapid, irreversible HAT in the 1 O 2 -trialkylamine exciplex, thus steric factors must govern the selectivity (i.e. at the less sterically demanding N-CH 3 position). In the case of less-constrained trialkylamines (1u, 1v), the 1 O 2 -bound exciplex can react promiscuously in HAT with endocyclic/non-N-CH 3 positions (e.g. benzylic groups, free alcohols) leading overall to degradation.

Conclusions
Herein, we report DCAS as a new organophotocatalyst for latestage N-CH 3 to N-formyl oxidations of complex trialkylaminecontaining natural products and pharmaceuticals, using molecular oxygen and continuous ow. Redox sensitive functionalities were tolerated, allowing the LSF post-modication of alkaloids and macrolide antibiotics to their N-formyl derivatives in good yields with excellent chemo-and regioselectivities, all in a continuous manner. The safe handling of O 2 under increased back pressures and temperatures via gas-liquid continuous ow in turn promoted mass transfer of O 2 to the reaction, increasing yields, shortening reaction (residence) times to several minutes and unleashing synthetically useful productivities (0.65 g per day). Mechanistic insights demonstrate how seemingly minor structural variations in an organophotocatalyst can not only increase solubility, but profoundly divert the excited state mechanism from photoinduced SET to E n T, followed by a downstream HAT mechanism. Precious metal photocatalysts of Ru-and Ir-based polypyridyl complexes are well known to participate in both E n T and SET, where structural tuning of ligands can affect switching between the divergent pathways. To our knowledge, such a concept has rarely been exploited in organophotocatalysis on the same core, privileged organophotocatalyst structures are typically developed either for SET or E n T pathways. Switching the mechanism offers opportunities to control selectivity, as indicated by the tolerance of reductively-labile groups herein. With the generation of 1 O 2 revealed, our study showcases one of few successful applications of 1 O 2 as a reagent in complex natural product synthesis. 31c,70 Further investigations on the selectivity of 1 O 2 's reactions with trialkylamines and the nature of interactions between DCAS, O 2 and trialkylamine quenchers are ongoing. 71

Data availability
Respectfully, all experimental and computational data is adequately available and retreivable from the ESI le. †

Conflicts of interest
There are no conicts to declare.