Benzoates as photosensitization catalysts and auxiliaries in efficient, practical, light-powered direct C(sp3)–H fluorinations

Of the methods for direct fluorination of unactivated C(sp3)–H bonds, photosensitization of SelectFluor is a promising approach. Although many substrates can be activated with photosensitizing catalysts, issues remain that hamper fluorination of complex molecules. Alcohol- or amine-containing functional groups are not tolerated, fluorination regioselectivity follows factors endogenous to the substrate and cannot be influenced by the catalyst, and reactions are highly air-sensitive. We report that benzoyl groups serve as highly efficient photosensitizers which, in combination with SelectFluor, enable visible light-powered direct fluorination of unactivated C(sp3)–H bonds. Compared to previous photosensitizer architectures, the benzoyls have versatility to function both (i) as a photosensitizing catalyst for simple substrate fluorinations and (ii) as photosensitizing auxiliaries for complex molecule fluorinations that are easily installed and removed without compromising yield. Our auxiliary approach (i) substantially decreases the reaction's induction period, (ii) enables C(sp3)–H fluorination of many substrates that fail under catalytic conditions, (iii) increases kinetic reproducibility, and (iv) promotes reactions to higher yields, in shorter times, on multigram scales, and even under air. Observations and mechanistic studies suggest an intimate ‘assembly’ of auxiliary and SelectFluor prior/after photoexcitation. The auxiliary allows other EnT photochemistry under air. Examples show how auxiliary placement proximally directs regioselectivity, where previous methods are substrate-directed.


Introduction
Although numerous methods have been developed for uorination reactions of C-H bonds, such as the uorination of carbanions, 1 arylpalladium complexes, 2 and the addition of electrophilic uorine to alkenes, 3 few methods permit the direct uorination of unactivated C(sp 3 )-H bonds. 4 Such a transformation is particularly advantageous for the late-stage functionalization (LSF) of complex molecules. Especially, since F atoms as H atom bioisosteres impart vastly differentand usually favorableproperties (physical, ADME, lipophilicity) in a pharmaceutical/agrochemical context. 5 Of these, methods that involve photosensitization of SelectFluor (SF) are particularly attractive for their mild conditions and use of light as a sustainable, uncontaminating energy source to drive reactivity. 6 Here, SF functions as both a uorine source and an activator of C(sp 3 )-H bonds; since quinuclidinium-type radical cations are powerful and selective hydrogen atom transfer (HAT) agents. 7 Pioneering studies by Chen 8 and Tan 9 demonstrated photocatalytic, direct uorinations of unactivated C(sp 3 )-H bonds by employing acetophenone or anthraquinone (AQN) as photosensitization catalysts (PSCats, Fig. 1A). Fluorination regioselectivity was governed by thermodynamic factors endogenous to the substrate; the most electron-rich C(sp 3 )-H bond that forms a stable 2°radical intermediate is uorinated. Lectka used benzil as an alternative PSCat for the late-stage photocatalytic uorination of complex molecules, 10 including steroid derivatives, and found that ketones/enones innate to the substrate could direct regioselective C(sp 3 )-H uorination in a proximal or distal fashion (Fig. 1A). Further examples of photosensitized C(sp 3 )-H uorination have successfully engaged benzylic positions of simple substrates 10a or complex peptides, 10c as well as ketal substrates. 10e In most cases, PSCats belonged to the family of aryl ketones as a privileged architecture for photosensitization, whose success has been attributed to the matching of triplet energies with SF. 9a While the utilized aryl ketone catalysts alone do not absorb visible light, their assembly with SF affords a charge transfer complex with a tailing absorption >400 nm. 9b However, in all cases uorinations of molecules containing 1°/2°alcohols or amines were unsuccessful (only 3°alcohols were tolerated), likely due to competing nucleophilic destruction of SF when the photochemical reaction is kinetically slow. 11 This severely hampers applications to complex bioactive molecules such as pharmaceuticals. Recently, Egami and Hamashima 12 found molecules with N-alkylphthalimides underwent directed photochemical C(sp 3 )-H uorination of their alkyl chains (Fig. 1B). However, both an N-alkyl tetrachlorophthalimide and an N-alkyl benzamide derivative were unsuccessful, leading authors to suggest a key role for the phthalimide's triplet energy. The phthalimide could be considered as a 'photosensitization auxiliary' (PSAux) that can be cleaved to reveal a uorinated 1°amine, though this was not realized and would typically require toxic, explosive hydrazine.
We envisioned a PSAux that is easily incorporated and removed could increase the generality and rapidity of photochemical C(sp 3 )-H uorinations. The PSAux may form an intimate assembly with SF for inner-sphere photochemistry, increasing reaction efficiency. Moreover, an appropriately designed PSAux (e.g. containing a C]O group 10a ) may even provide an exogenous user handle to direct regioselectivity for the rst time, where all previous reports follow factors/ functionality endogenous to the substrate. Herein, we report the discovery of benzoyl groups as novel, versatile photosensitizer architectures. Methyl 4-uorobenzoate serves as a photosensitizing catalyst for uorination of simple substrates, while the 4-uorobenzoyl group serves as a PSAux for the uorination of complex molecules containing alcohols and amines and can be easily installed and removed without compromising product yields. The PSAux strategy (i) markedly increases the efficiency, rapidity, reliability and practicality of C(sp 3 )-H uorination reactions; (ii) enables reactions of substrates that cannot be engaged with photocatalysis (alcohols and amines, where previous catalytic methods were generally only applicable to 3°a lcohols); (iii) allows reactions to succeed under air atmosphere and (iv) offers promise to direct uorination regioselectivity (Fig. 1C).

Discovery
When examining the C(sp 3 )-H uorination of amyl benzoate 1a to 2a as reported by Tan and co-workers 9a ( Table 1, entries 1 and  2) with AQN and derivatives, we found that no catalyst was necessary; the reaction proceeded in its absence at 365 and 400 nm (entries 3 and 4). The PSCat-free reaction did not proceed at 451 nm or in the absence of light (entries 5 and 6).
Assuming that 1a was acting as a photosensitizer for its own self-uorination, we reasoned that arylbenzoate esters may function as exogenous PSCats in the C(sp 3 )-H uorination of 1adamantanol 1b to 2b and 3b. However, 1a was an ineffective PSCat (entry 7). Assuming that self-uorination of 1a may deactivate its catalysis of the uorination of 1b, methyl benzoate (4a) was attempted but was similarly unsuccessful (entry 8).
While other para-substituted benzoate esters (4, where X = Cl, Br or CF 3 , entry 9) were also ineffective (also see ESI †), methyl 4-uorobenzoate (MFB) was a surprisingly effective PSCat (entry 10), delivering a 46% yield of 2b(+3b) compared to an ineffective control reaction without PSCat (entry 11). Employing 10 mol% of MFB provided 2b+3b in an improved 71% yield (entry 13). Doubling the overall concentration of 1b to 0.31 M and decreasing catalyst loading at this higher overall concentration also improved the yield (entry 15). For both the low and high concentration conditions, a higher power LED further increased the yield of 2b+3b in a shorter reaction time (entries 14 and 17, termed 'conditions A' and 'conditions B', respectively), although 6 h was here insufficient for full conversion (entry 16). Air, protic additives and using an equimolar ratio of SelectFluor (vs. substrate) were all detrimental to the reaction (for full investigations, see ESI †).
Despite a reasonably broad scope of applications of this catalytic method, we noticed that apart from 3°alcohols (e.g. 1b), free alcohols (5a, 5b ′ , 5c-5e, 5g) and free amines (6a-6d) all failed under photocatalytic C(sp 3 )-H uorinations (Table  3). We note that in the earlier reports of Chen, 8 Tan, 9a and Lectka, 10a-e free 1°or 2°alcohols were never reported, only hindered 3°alcohols were tolerated. Neither were free 1°/2°/3°a mine moieties; 1°and 2°amines always required protection. Even protected 1°amide 7 did not uorinate in this case. One explanation could be that if the desired reactivity is too slow, nucleophilic destruction of SelectFluor (known to occur with 1°and 3°amines) 13 may emerge as a competitive thermal reaction, arising lower yields or no reactivity. The steric hindrance of 3°alcoholsthe only class of free alcohols that worked in the previous reportsmay explain their tolerance. However, since nucleophilic 3°amines did proceed (1s, 1z) albeit in low (33%, 18%) yields, there may be other factors responsible for the failure of 1°and 2°alcohols/amines/ amides. SF's known interactions with a wide range of functional groups in ground state chemistrysuch as Lewis acid interactions with C-O bonds and uorinations of nucleophilic (S,N) atoms 14are likely detrimental to its use in photocatalytic C(sp 3 )-H uorinations. Another explanation is that hydrogen bonding networks around alcohol/water-solvated SF prohibit its non-covalent assembly with the aromatic PSCat that is needed for successful photochemistry (vide infra). 9b Interestingly, although 5b has no unprotected alcohol and contained a privileged benzylic C(sp 3 )-H bond, it also reacted poorly (8% yield).

Synthetic scope of PSAux method
At this juncture, we considered that a key advantage of our 4-uorobenzoate photosensitizer compared to previouslyemployed aryl ketone PSCats was the ability to reversibly attach it to alcohol and amine substrates and then detach it aer photochemical uorination (both under simple conditions): a 'photosensitization auxiliary' (PSAux) approach.
Not only would this enable C(sp 3 )-H uorinations that were catalytically inefficient or impossible (such as complex molecules), it would accelerate reaction kinetics and improve reproducibility (vide infra). Moreover, it would improve practicality by decreasing atmospheric sensitivity and simplify puri-cation of otherwise highly polar unprotected substrates. Thus, a variety of alcohols 8 and amines 10 were loaded with the 4-uorobenzoate PSAux in near-quantitative yields. To our delight, auxiliary-loaded alcohols 8a-i all underwent C(sp 3 )-H uorination (Table 3), generally in good to excellent (57-90%) yields. Interestingly, where benzoate ester 5b provided only 8% yield, its 4-uorobenzoate derivative 8b was uorinated to afford 9b in 75% yield, highlighting the benet of the PSAux in improving efficiency (even when unprotected alcohols/amines are absent). Moreover, while the catalytic uorination affording 2k was completely ineffective under air, the yield of 9b (74%) was unaffected by setting up under air. Although the reaction of 1a without any catalyst gave a 44% yield of 2a (Table 1, entry 4), however, its PSAux derivative, amyl 4-uorobenzoate, led to a 78% yield of 9j again conrming the benet of the F atom in the PSAux. Though lithocholic acid derivative 8k only afforded a 23% yield of 9k, the C(sp 3 )-H uorination of steroid derivatives was only possible previously with in-built enone/ ketone/ketal protecting/directing groups. 10 Gratifyingly, the C(sp 3 )-H uorination of amines was also enabled via their auxiliary-loaded amide forms 10a-10e, affording 11a-11e in moderate to very good (32-72%) yields. The profound increase in reactivity from having the uorine in the 4-position of the benzoate group was apparent here. Substrate 10a was successfully engaged, while the previously attempted photochemical uorination of N-butylbenzamide (Fig. 1B) 12 gave no reaction. Although phthalimides 12 offer the potential to serve as a PSAux, the 4-uorobenzoyl PSAux is both more effective and generally applicable since (i) phthalimides can only function in this way for 1°amines and ii) benzoates undergo straightforward deprotection (vide infra). To our surprise, the auxiliary loaded amide derived from Fingolimod 10f did not require any subsequent protection of its two free 1°a lcohols. In summary, incorporation of a PSAux accelerates the rate of C(sp 3 )-H uorination to a point that it (i) outcompetes the catalytic method, (ii) outcompetes degradation of SF, (iii) allows energy transfer (E n T) photosensitization reactions to succeed under air. This renders the PSAux an attractive strategy for C(sp 3 )-H uorination of complex, bioactive molecules like active pharmaceutical ingredients, exemplied by the successful uorination of phenylcyclohexylglycolic acid (8i), a derivative of lithocholic acid (8k), multiple sclerosis active pharmaceutical intermediate (API) Fingolimod (10f) and antidiabetic API vildagliptin (10g).
Noticing that antipsychotic API Haloperiodol 12 contains a 4-uoroacetophenone moiety, we reasoned its photochemical C(sp 3 )-H uorination may occur without any catalyst or auxiliary. Indeed, its reaction successfully gave 13 in 43% yield ( Fig. 2A). Haloperidol and Fingolimod feature in the top 200 pharmaceuticals by retail sales in 2020 15 and their late-stage uorination opens avenues to new chemical space and pharmacological activities. Finally, we investigated the impact of PSAux location on C(sp 3 )-H uorination selectivity by the late- Table 2 Scope of photocatalytic C(sp 3 )-H fluorination method a a For substrates with more than one regioisomer, the major isomer is depicted. NMR yields were determined by 19 F NMR with triuorotoluene as internal standard (IS). Unless explicitly dened, yields correspond to a single regioisomer. Isolated yields in parenthesis. b Yield corresponds to the combined mixture of regioisomers. c 2.0 eq. of SF used. stage uorination of Dextromethorphan 1z (Fig. 2B). The photocatalytic method with MFB was inefficient, affording an unsatisfactory (18%) yield of 2z. Following attachment of the PSAux to 1z by tandem N-or O-demethylation/PSAux loadingaffording 10h and 8l in high yields over 2 steps (85 and 87%, respectively)their photochemical C(sp 3 )-H uorinations afforded 11h and 9l in satisfactory yields (42 and 48%, respectively). Interestingly, 11h was afforded as two uorinated regioisomers on the 'B' ring (possibly assisted by a developing uorine gauche effect) 16 while 9l was a single regioisomer on the 'C' ring. Therefore, our PSAux auxiliary strategy offers future promise as an attachable handle to direct late-stage C(sp 3 )-H uorination selectivity which (i) is not possible by previous catalytic methods and (ii) allows uorination to be exogenously directed by the chemist, rather than endogenously directed by the molecule via its inherent functionality. 9a, 10,12 Of key importance and justifying the use of a PSAux strategy, the 4-uorobenzoate auxiliaries can be cleaved by known methods 17 to give alcohols or amines in near-quantitative (93-98%) yields (Fig. 3).

Mechanistic studies of PSCat method
Luminescence spectra of the reaction components were measured at their representative excesses (conditions B, 10× diluted to ensure SF would fully dissolve). While MFB showed no change at all aer repeated excitations at l = 375 nm (Fig. 4A), SF underwent a slow photodecomposition over 6-18 min (Fig. 4B; for photodecomposition rates, see ESI †). Upon Table 3 Scope of photosensitization auxiliary C(sp 3 )-H fluorination method. (A) Unreactive, unprotected substrates. (B) Scope of photosensitization auxiliary method a a For substrates with more than one regioisomer, the major isomer is depicted. NMR yields were determined by 19 F NMR with triuorotoluene as internal standard (IS). Unless explicitly dened, yields correspond to a single regioisomer. Isolated yields in parenthesis. b Yield corresponds to a mixture of regioisomers. c Reaction completed in 10 h. d Yield corresponds to a single regioisomer. e Isolated yields of gram scale reactions. f Reaction completed in 5 h.
mixing MFB (12.5 mol% w.r.t. SF) with SF (8 eq. w.r.t MFB), the peak at l = 434 nm decreased in intensity by 42% (Fig. 4C). Its shape did not resemble SF (no peak at l = 467 nm) and even aer 5 min of repeated excitations the intensity barely decreased. Lu, Soo, Tan and co-workers also observed that the emission peak shape of AQN differed in the presence of SF. 9b Mixing MFB with larger excesses of SF (Fig. 4C-E) led to buildup of a peak at l = 467 nm that more closely resembled SF. Interestingly, the presence of MFB led to faster photodecomposition of the SF peak compared to SF alone (Fig. 4, E vs. B). The measured lifetime of SF also decreased in the presence of MFB (see ESI †). These data corroborate an energy transfer between MFB and SF, in line with that proposed for previous arylketone catalysts. 9,10 This suggests 3 MFB* is the excited state involved, and its T 1 energy is well matched with that of SF (see Table 4, vide infra). A radical trapping experiment with 1.5 eq. of TEMPO was performed. As clear evidence of the alkyl chain radical intermediate, a product was detected by LC-MS which matched the TEMPO-bound 16 and no uorinated product (2k) was detected (Fig. 5). Therefore, the mechanism of the MFB-catalyzed reaction mirrors that previously proposed and elegantly investigated in the literature for AQN (Fig. 6). 9b, 12 Upon photoexcitation and intersystem crossing, 3 MFB* undergoes E n T with SF. An exciplex is formed involving substrate molecule (1k) which facilitates the N-F bond cleavage and HAT process. 9b Intermolecular HAT occurs between the radical dication of SF (18) and the most hydridic C(sp 3 )-H bond of the substrate (1k) that affords the more stable 2°alkyl radical 17 and SF-H. Once 17 is generated, uorine atom transfer (FAT) with SF affords 2k. This may propagate a chain reaction, but a previously reported quantum  Table 3B were used, with 1.5 eq. substrate and 1.0 eq. SF. Isolated yields given unless stated otherwise. b Yield determined by 19 F NMR with trifluorotoluene as IS. c Overall yields over two steps.  Table 3 for regioselectivity.  yield of ca. 0.12 seems to oppose this (or suggests an inefficient chain propagation). 9b

Mechanistic studies of PSAux method
The majority of efforts were directed at understanding (i) the pronounced benet of the PSAux method over the catalytic method, and (ii) the promoting role of the 4-uorine atom in the PSAux. The rst observation was that compounds 1c, 1d and 1e all gave no uorination in the absence of MFB (Table 2), despite electronically resembling amyl benzoate (1a) and MFB. The lack of reactivity of 1e in the absence of PSCat is consistent with the previously reported lack of reactivity of n-butylbenzamide (Fig. 1B). 12 The triplet energies (T 1 ) for directly related compounds are summarized in Table 4 To probe further, we examined substrate 5b for which the catalytic method was ineffective, giving product 21 in only 8% yield ( Table 5, entry 1). A control experiment without catalyst (entry 3) revealed this trace reactivity derived from photochemical self-uorination, akin to amyl benzoate 1a but in very low efficiency. When a stoichiometric amount (150 mol%) of MFB was employed (entry 4), activity of the PSCat method was observed, affording product 21 in 47% yield. That the reaction of 8b without catalyst gave almost double (75%) the yield of 9b (entry 7) conrmed the PSAux provides a benecial effect beyond simply higher 'effective' catalyst loading. It supports the notion that a preassembly co-locating substrate and SF in close proximity sets up a more efficient inner-sphere photochemical process. This is further supported by the fact that the synthetic reactions work equally well under the PSAux method without degassing (entry 7 vs. 8), but not under the PSCat method (entry 4 vs. 6) which requires strict freeze/pump/thaw cycling. NFSI as a different uorinating source gave results that were (i) far inferior to SF (entries 8 and 9) and (ii) similar with both PSCat and PSAux methods (see ESI †), suggesting a different, unrelated mechanism operates here. UV-vis spectra of the reaction mixtures of 5b and 8b revealed, in both cases, an absorption band at ca. 400 nm (Fig. 7, see ESI for details †). This absorption was notably higher than any individual reaction component and was accessible at the synthetic experiments' LED wavelength, suggesting a charge-transfer complex 20 forms between the PSAux and SF. However, the peak shape and absorption intensity was no different in the case of 5b's and 8b's reaction mixtures. Given the drastic difference in the efficiency of 5b's and 8b's reactions, the promoting role of the F atom in the PSAux's clearly does not involve chromophoric changes.
a NMR yield, based on 19 F NMR and triuorotoluene as IS. b Under air. c Instead of SF, NFSI was used as a uorine source. Given the inability to determine the origin of the promoting effect of the PSAux's F atom by optical spectroscopy, we turned to examine reaction kinetics. The kinetics of 8b's reaction were examined by a photoirradiation probe allowing on-line LED irradiation within the NMR spectrometer (see ESI for details †). 21 Consumption of all starting materials and formation of all products could be tracked by time-resolved 1 H{ 19 F} NMR. Samples were compared at the synthetic reaction concentration (0.31 M) to be as representative as possible. SF was not fully dissolved under these conditions, meaning unagitated reactions in NMR tubes could not reach yields/conversion rates as high as the synthetic reactions. Nonetheless, this approach was deemed sufficient to observe relative effects. Interestingly, an induction phase was apparent where scarcely any product forms. When repeating the reaction with different amounts of substrate, a clear trend emerged: the induction phase shortened as the loading of 8b increased. This occurred even in spite of the decreasing (measured) solubility of SF as the overall mixture became more non-polar with increasing [8b]. While the standard reaction with 1.5 eq. of substrate 8b had an induction phase of 9.2 h, when 3 eq. 8b are used it only took 1.9 h until product formation began with a prole typical of a rst-order reaction. In summary, by doubling the amount of substrate 8b, the induction phase can be shortened by 79% (Fig. 8A). The rates of product formation within the linear build-up and the overall yield were, however, independent of substrate loading (Fig. 8B). This may be due to (i) the lack of stirring (an inevitable drawback of in situ NMR kinetics), meaning only traces of undissolved SF are available for the ongoing reaction, and/or (ii) lower light intensity of the in situ irradiation setup (see Table 1, entry 15 vs. 17).
We hypothesized that a de-aggregation of SF may rationalize the induction period. Therefore, Diffusion ordered spectroscopy (DOSY) was performed. The volume of 8b hardly changed as a function of its concentration (431.4Å for 0.62 M and 410.1Å for 0.31 M; Table 6, entries 1 and 2) and was unchanged in the presence of SF (entries 4-6). In contrast, the volume of SF decreased by 13% from 952.2Å (entry 3) to 825.3Å (entry 7) in the presence of 1.5 eq (8b) and by 12% to 836.9Å with 3 eq. 8b (entry 5) compared to pure SF with the same concentration (952.2Å, entry 3). This conrmed the role of 8b as a deaggregation agent. According to the estimated (spherical) molecular volume of SF (see ESI †), such a change suggested a change in aggregation from 3 to 2 SF molecules. We then examined the size of the SF aggregate in the presence of 5b. The volume of 5b alone (413.9Å) was very similar to that of 8b, consistent with the fact that the uorine atom is a known bioisostere for H. 6 However, 5b only de-aggregated SF by 4% (entry 9)a de-aggregating effect 73% smaller than 8b. Therefore, the F atom in the PSAux clearly plays a role in de-aggregating SF. This could be explained by a halogen-p interaction as suggested by DFT calculations (see ESI †). However, we note that (i) aer this initial de-aggregation with 1.5 eq. 8b, the volume of components remain unchanged during the reaction and (ii) doubling the concentration of 8b does not further de-aggregate SF. Thus, while de-aggregation of SF can explain 8b's success vs. 5b's failure, the substrate concentration benet on induction period shortening (see Fig. 8) cannot be explained by deaggregation. Further studies are ongoing in this direction.
Taking together the synthetic benets of the PSAux strategy (handle to direct regioselectivity, remarkable tolerance to O 2 , ability to engage molecules with unprotected functionality), our mechanistic studies suggest that the PSAux plays two key roles. Firstly, the benzoate PSAux engages SF in an intimate chargetransfer preassembly (Fig. 9) that absorbs 400 nm (see Fig. 7). The second key role of the PSAux is to de-aggregate SF (Fig. 9), for which the 4-uoro atom is especially effective. Moreover, the measured quantum yield of the PSAux reaction for 8b (F < 0.01) suggested against a radical chain mechanism. The downstream mechanism then resembles that of the PSCat method (Fig. 6) and that proposed in the literature, 9,12 however we propose subsequent steps all occur within the exciplex/solvent cage. E n T from 3 MFB* to SF leads to N-F bond cleavage. The benzoate then directs the dication radical to the locally most hydridic C(sp 3 )-H bond(s) for HAT that forms a stable (2°, 3°) radical to undergo uorination. We propose that the benecial role of deaggregation is not related to the success of charge-transfer complex formation (see Fig. 7 and the poor reactivity of 5b vs. 8b), but rather it increases the efficiency of the E n T step following photoexcitation.
We cannot rule out that the C]O of the 3 PSAux* engages in photochemical HAT 22 with the C(sp 3 )-H bond, but this seems inconsistent with (i) the catalytic lack of reactivities of other para-substituted or unsubstituted benzoyl groups (4b, 5b) in PSCat or PSAux uorinations and (ii) the clear promotion of SF's photodecomposition as seen in the emission spectra (Fig. 4). The ability to conduct photochemical C(sp 3 )-H uorination reactions in shorter overall reaction times is highly attractive to practitioners, and useful for achieving high yields since SF undergoes a competing (photo)decomposition process.

Application of benzoyl auxiliary in other E n T photochemistry
While the main purpose of our study targeted applications to (and understanding of) C(sp 3 )-H uorinations, we found that the 4-uorobenzoyl photosensitizer could successfully deliver E n T [2 + 2]-photocycloadditions in air. The 4-uorobenzamide derivative of cinnamyl amine (22) afforded cyclobutane 23 (31%) in exclusive head-to-head regioselectivity, but as a mixture of 'all-trans' and 'all-cis' diastereomers (5 : 1). In contrast, 4-uorobenzoylated cinnamyl alcohol 24 underwent [2 + 2]-photocycloaddition to give exclusively the head-to-head, 'all-trans' diastereomer of 25 (87%). The previously reported direct UV excitation of cinnamyl alcohol (10 mM in MeCN, 450 W Hg bulb) affords a 2 : 1 mixture of cis-/trans-isomers and only traces of the cyclobutane (<5%) 23a Previously, high yield and selectivity for the cyclobutane was only possible by an iterative (2-step) covalent tethering of two cinnamyl alcohols by di-iso-propyldichlorosilane (compared to which, 4-uorobenzoyl chloride is ∼2.5× cheaper) under UV-light and strict degassing. 23 Herein, this is achieved non-covalently, presumably via p-stacking of the photoactive 4-uorobenzoyl motifs under milder visible-light conditions. Where contemporary photocatalytic [2 + 2]-cycloadditions require degassing due to redox or E n T quenching of the photocatalyst by O 2 , 24 the 4-uorobenzoyl photosensitization auxiliary provides an alternative strategy than can be run open to air (Fig. 10).

Conclusions
In conclusion, we report the discovery of a highly efficient visible-light harvesting photosensitizer architecture for the   direct uorination of unactivated C(sp 3 )-H bonds. Our new photosensitizing (4-uorobenzoyl) group was successfully used both as photosensitizing catalyst and photosensitizing auxiliary. Both catalytic and auxiliary methods were successfully applied to the late-stage functionalization of complex molecules. While a reasonably broad substrate scope of uorinated products was accessible with catalytic method, the photosensitization auxiliary strategy was found to be far superior in scope and practicality. Substrates that for which the catalytic method was inefficient or ineffective were successfully engaged in higher yields and shorter reaction times, furnishing a general platform for photosensitized uorination that even tolerates free alcohols/amine functionality and works under air. Mechanistic studies reveal the auxiliary's promoting effects derive from two angles, (i) harnessing SF in an intimate charge transfer assembly with SF and (ii) in de-aggregating SF. The 4-uorobenzoyl auxiliary allows other E n T photochemistries to occur in air, where its non-covalent p-stacking interactions alter mechanism and selectivity. Assemblies between photocatalyst and substrate such as EDA complexes, 20 hydrogen bonding, 22c,d non-covalent interactions, 25 and ordering of solvent 26 attract ever more attention to uncover the next generation of photocatalytic transformations and provide new frontiers in selectivity and efficiency. This study highlights the emerging importance of how changes in aggregation states of photocatalysts can profoundly inuence photochemical reaction mechanisms. 27 Further studies into the structural nature of the charge transfer assembly and the origins of the induction period in photochemical C(sp 3 )-H uorination reactions are ongoing.

Data availability
Experimental and computational data are located in the ESI. †

Author contributions
S. Y. discovered and optimized the photocatalytic and photochemical reaction conditions, synthesized most substrates, synthesized and puried all products, performed synthetic experimental mechanistic studies, measured UV-visible spectra and grew crystals for XRD. W. J. S. conducted all studies on NMR kinetics and DOSY experiments and discovered the induction period and different aggregation states of SF. W. J. S. assisted with 2D NMR structural assignments of complex uorinated products.

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