The complexes were prepared by first synthesizing the corresponding ligands 1,4-di(picolyl)-1,4,7-triazacyclononane L ^N3 and 1-thia-4,7-di(picolyl)-diazacyclononane L ^N2S . We followed published procedures with slight adjustments (see also the Supporting Information). Briefly, L ^N3 was prepared beginning with the tosylation of diethylenetriamine and glycol (Friscourt et al., 2012). The next step was a Cs₂CO₃ catalyzed macrocyclization using the tosylated compounds to form tri-tosyl-triazacyclononane (ts₃-tacn) (Cao et al., 2010). Partial detosylation of the formed macrocycle using HBr in acetic acid resulted in the formation of ts-tacn (Cao et al., 2010), which was transformed into dipicolyl-tosyl-tacn using two equivalents of 2-picolylchloride hydrochloride in a base-catalyzed S_N2 reaction (Stavila et al., 2008). To obtain the desired ligand L ^N3 the remaining tosyl protecting group was cleaved using conc. H₂SO₄ (Luk’yanenko et al., 1990; Stavila et al., 2008). L ^N2S was prepared in a similar fashion: First N,N’-bis-tosyl-bis(2-aminoethyl)sulfide was synthesized via sodium ethoxide driven S_N2 reaction of cysteamine hydrochloride and 2-chloroethylamine hydrochloride (Wilson, 2007), followed by tosylation as described earlier for L ^N3 (Friscourt et al., 2012). The macrocyclization was started with LiOH (2.5% in water) in toluene and using NaBu₄Br as phase-transfer catalyst (Wilson, 2007). To obtain L ^N2S a detosylation and substitution with 2-picolylchloride hydrochloride was performed (Chak et al., 1994; Wilson, 2007). Both proligands were used immediately to obtain the complexes, due to sensitivity towards oxidation. Hence it was not possible to store the proligands for longer time periods.

For both compounds, L ^N3 and L ^N2S , the analytical data nonetheless agreed with those published earlier. Metal complexation could be achieved by reaction of the corresponding ligand with an equimolar amount of simple metal salt, e.g., metal triflates M(OTf)₂ (M = Fe, Co, Ni). Work-up of the reaction mixture was done under air atmosphere leading to oxidation of the iron and cobalt compounds. Purification of the crude product was obtained via washing with CH₂Cl₂ to remove excess of ligand and washing with toluene (Fe(OTf)₂ and Co(OTf)₂) or THF (Ni(OTf)₂) to remove unreacted metal salt. The desired complexes were obtained as solids, which were redissolved in a small amount of MeCN and precipitated using Et₂O. Since L ^N3 was synthesized as hydrochloride salt (stemming from column chromatographic purification using dichloromethane), Fe-L ^N3 and Co-L ^N3 were obtained as complexes with one chloride ligand and two uncoordinated triflate counter anions, which could be verified by mass spectrometry and X-ray diffractometry. Ni-L ^N3 was isolated as a chloride free complex by precipitating AgCl using Ag(OTf). The complexes M-L ^N2S were prepared in the same fashion giving yields in the range of 44–50%. All complexation reactions should be done soon after preparation of the proligands, due to sensitivity against oxidation of the compounds L ^N3 and L ^N2S .

The successful complex formation was supported by high resolution MS revealing characteristic peaks at m/z 516.0974 (Fe-L ^N3 ), 519.0975 (Co-L ^N3 ), 518.1018 (Ni-L ^N3 ), as well as 533.0584 (Fe-L ^N2S ), 536.0568 (Co-L ^N2S ) and 535.0590 (Ni-L ^N2S ) for the corresponding [M + OTf]⁺ fragments (Supplementary Figures S13–S18). To verify the oxidation state of +3 for both the Fe and the Co complexes, different approaches have been made. Since Co-L ^N3 and Co-L ^N2S represent diamagnetic compounds, i.e., they are in the low-spin d⁶ form, NMR spectroscopy could be applied. Thereby, Co-L ^N3 gives a nicely resolved ¹H NMR spectrum (Supplementary Figure S11) where all signals can be assigned. The ligand conformation leads to an asymmetry in the complex in contrast to the symmetric nature of the ligand itself (see also crystal structure in Figure 1). Thus, the pyridyl substituents give rise to eight individual signals in the aromatic region, ranging from 9.5 to 7.2 ppm. The methylene groups arise at 5.1, 4.6, and 4.5 ppm and the macrocyclic protons from 4.2 to 2.0 ppm. In case of Co-L ^N2S the ¹H NMR spectrum shows broad signals in the region from 2 to 9 ppm. It seems that the S-donor atom in the ligand macrocycle is less strongly bound leading to a higher mobility in the system and hence stronger dynamics.

On the other hand, Fe-L ^N3 and Fe-L ^N2S are paramagnetic compounds, which give rise to non-interpretable ¹H NMR spectra. The magnetic properties indicate a +3 oxidation state in octahedral environment with a strong ligand field, as for +2 oxidation state a d⁶ low-spin state, and thus a diamagnetic compound, would be expected. The +3 oxidation state was confirmed by using EPR spectroscopy for both compounds giving a Fe^III typical spectrum with signals having a g value of around 4.0 and 2.0 (Supplementary Figure S23) (Srinivasan and Gralla, 2002).

Electrochemical and spectroscopic properties were determined via cyclic voltammetry (CV) and UV/vis measurements. The UV/vis studies in DMF solution (Table 1; Supplementary Figures S19–S21) reveal a strong ligand-based absorption peaking at 268 nm for all compounds, which can be assigned to be situated at the aromatic picolyl substituents. These maxima are outside the visible spectrum and do not have an impact on the color of the complexes. For the two Fe complexes the dark brown color derives from a broad absorption feature ranging from 600 to 350 nm with a shoulder at around 432 nm (Fe-L ^N3 ) or 502 nm (Fe-L ^N2S ). The Ni complex Ni-L ^N3 only has an additional absorption maximum at 308 nm and hence is colorless. As the Ni-L ^N2S absorbs in a broader area from 500 to 300 nm it is isolated as light brown solid. The most striking difference is revealed for the Co complexes where d-d transitions are observed at 362 nm (ε = 320 dm³/mol*cm) for Co-L ^N3 and 490 nm (ε = 390 dm³/mol*cm) for Co-L ^N2S leading to a light red/pink color.

In the CV measurements of the complexes Fe-L ^N3 and Fe-L ^N2S a reversible redox event is observed at a half-wave potential of −0.09 and 0.20 V (referenced to Fc/Fc⁺), respectively, representing the Fe^II/Fe^III redox couple (Table 1; Supplementary Figure S22). In addition, there are irreversible reduction events at −2.50 V and −2.91 V for Fe-L ^N3 and −2.29 V and −2.77 V for Fe-L ^N2S , which can tentatively be assigned to the Fe^II/Fe^I reduction and the formal Fe^I/Fe⁰ reduction. However, there is certainly the possibility that the true electronic distribution is different and ligand-based reduction cannot be neglected without further experiments.

Interestingly, Fe-L ^N2S is thus easier to reduce (by about 200 mV) than the sulfur free analogue. This result can be attributed to the electron donating property of the sulfur atom, and a similar trend is also observed for the related cobalt complexes. It is important to note, though, that for Fe-L ^N2S an additional reduction process occurs at −0.34 V, which we tentatively relate to the sulfur donor atom. A similar behavior is observed for Co-L ^N2S that shows and additional reduction process at −0.71 V, which is absent for Co-L ^N3 .

In case of Co-L ^N3 a reversible redox event at −0.59 V is indicative for the Co^II/Co^III redox couple. This event is accompanied by two irreversible reduction processes at −2.17 V and −2.59 V resulting in the formation of Co^I and finally the Co⁰ state – most likely being a Co^I(L ^{N3 −} ) species. In accordance with the observations for Co-L ^N3 , Co-L ^N2S shows one reversible reduction at −0.35 V and three irreversible reduction events at more negative potentials at −1.93 V, −2.43 V and −2.88 V that are shifted to higher voltages when compared to Co-L^N3 . An assignment of these redox processes is not straightforward and currently not possible. The Ni complexes Ni-L ^N3 and Ni-L ^N2S do not show any redox processes in the investigated solvent window.

X-ray quality crystals of Co-L ^N3 and Fe-L ^N3 were obtained by slow vapor diffusion of Et₂O to a high concentrated MeCN solution. Both complexes are isostructural (Figure 1; Supplementary Figure S24; Supplementary Tables S1, S2) with all five nitrogen atoms acting as donor atoms. The octahedral coordination sphere is completed by coordination of one chloride ion. The crystal structure shows a hydrogen atom at the secondary amine which confirms the +3 oxidation state (two triflate counter ions were also found per cation).

In case of Co-L ^N3 shorter Co-N bond lengths (1.94–1.97 Å) are observed, which is in line with Co³⁺ being smaller than Fe³⁺. The Co-Cl bond distance, on the other hand, is almost identical to the Fe-Cl bond with ∼2.23 Å (Table 2; Supplementary Table S2). For Co-L ^N3 an isostructural complex [CoL^N3 (OH₂)][CIO₄]₃•H₂O was published earlier (McLachlan et al., 1995) having similar bond angles and distances, but a shorter Co-O bond since H₂O is a smaller ligand than the chloride ion. Furthermore, there are a few crystal structures of similar Fe compounds with the L ^N3 ligand described in the literature. Two prominent examples are the complexes [FeL ^N3 (Cl)](PF₆) and [FeL ^N3 (NCS)](PF₆), which were obtained via reaction under inert conditions and, hence, possess an oxidation state for Fe of +2 (Spiccia et al., 1998). Both complexes nonetheless show a similar ligand geometry, with Fe in an octahedral coordination environment surrounded by five nitrogen and one (pseudo-)halide donor atom.

In case of the L ^N2S complexes we were not able to obtain crystals so far. Nonetheless, crystal structures have been reported for similar Fe^II and Ni^II complexes (Wasielewski and Mattes, 1993; Zhang et al., 1998).

The photocatalytic activity of the prepared complexes was determined using Ir(dFppy)₃ and triethylamine (TEA) as sacrificial electron donor. In earlier studies we successfully applied the following conditions (Giereth et al., 2021): 5 × 10⁻⁵ M catalyst concentration in 5 ml DMF with 5 × 10⁻⁵ M Ir as photosensitizer (PS) and 5 Vol% TEA as sacrificial electron donor (SR). The catalysis mixture is placed in front of a 200 W Hg-lamp equipped with a 400 nm low-wavelength cut-off filter to exclude UV light. Ir was selected because of its ability to still absorb solar photons in the region of 400–460 nm (Becker et al., 2020) and its potentially well suited (excited state) redox properties (Table 1) (Koike and Akita, 2014; Lee and Han, 2020; Giereth et al., 2021). For this work we just focused on the gaseous reduction products formed and did not investigate the formation of liquid products. Further information is given in the supplementary data.

It can be observed, that not only the ligand but also the metal centre do have an impact on both the formed products and the catalytic activity (Figure 2; Table 3). While Fe-L ^N3 shows a slightly higher TON for H₂ than for CO formation after 24 h reaction time, Co-L ^N3 shows no H₂ formation and the highest TON_co for the here investigated complexes. This result might be due to the fact that Fe tends to form hydride intermediates (Drosou et al., 2020). When a sulfur atom is added to the ligand framework, considerable H₂ formation is observed. It is known, that sulfur donor atoms can take part in hydrogen evolution reaction by forming a sulfur-hydride species (Darmon et al., 2014; Drosou et al., 2020). On a side note, the illumination of the reaction solution without any catalysts results in no CO or H₂ release, indicating that the photosensitizer is stable and its (photo) decomposition products are not responsible for the formation of the gaseous products.

Further evidence for molecular catalysis could be obtained by adding mercury to a catalysis solution with Co-L ^N3 . Similar TONs were obtained indicating that Co colloids did not form. In case of Co-L ^N2S the ratio of CO:H₂ changes to about 2:1 (i.e., 66 % selectivity for CO formation) in comparison to no H₂ formation for Co-L ^N3 . In case of Fe-L ^N2S the CO:H₂ ratio stays very similar to the one of Co-L ^N2S , but the activity drops by about one third, and about half in comparison to Fe-L ^N3 . For the Ni complexes the electrochemical data predicted no (photo)catalytic activity, because they do not show a reduction process in the solvent window investigated. This could be confirmed for Ni-L ^N3 , whereas Ni-L ^N2S shows a moderate TON_H2, which again is probably induced by the sulfur atom in the ligand system.

In order to investigate if the product ratio or the activity can be adjusted by adding an additional proton source, water was added to a Co-L ^N2S catalytic mixture in different concentrations. Thereby a shift to higher TON_H2 is observable, while TON_CO is decreasing (Supplementary Table S4). When adding 1 Vol.% water a ratio of CO:H₂ 18:24 was observed, while the addition of 5 Vol.% water led to a ratio of CO:H₂ 5:54. This result strikingly demonstrates that the product selectivity can be manipulated to a large amount by the addition of a proton source.

To investigate the influence of the type of PS, catalytic experiments were performed using Cu. This PS was chosen due to its similar absorption and redox features as compared to Ir (Table 1) (Giereth et al., 2021). Unexpectedly, the combination of Cu and Co-L ^N3 shows only a TON_CO of 13 after 24 h of illumination, i.e., only a seventh part compared to the reaction using Ir. The much lower overall TON_CO can be attributed to the rather fast photodecomposition of Cu (Giereth et al., 2021). In the first 2 hours of illumination, Cu actually shows a better performance than Ir (Supplementary Figure S25; TON_CO of 9 vs. 3). However, there is almost no further increase in CO evolution for Cu indicating that most of the photosensitizer is decomposed already after a few hours of reaction time. The initial better performance of Cu vs. Ir is tentatively assigned to the more suitable excited state oxidation potential of Cu (Table 1) and, hence, more efficient oxidative quenching of the excited state by the catalyst (see next section).

Luminescence quenching experiments were performed to obtain some insight into the reaction mechanism. The excited state of the (Ir) photosensitizer can be quenched oxidatively by the catalyst Co-L^N3 (Supplementary Figure S26; Table 1). Specifically, Stern-Volmer luminescence quenching experiments in acetonitrile yield a quenching rate constant k _q = 1.1·10⁹ M⁻¹s⁻¹ (for Ir), which is roughly a factor of seven (Montalti et al., 2006) below the diffusion limit. Under the catalytically relevant reaction conditions, 5·10⁻⁵ M Co-L^N3 was present, and given a rate constant of 1.1·10⁹ M⁻¹ s⁻¹ for oxidative Ir luminescence quenching, this leads to a pseudo first-order rate constant of 5.5·10⁴ s⁻¹ for photoinduced electron transfer from Ir to Co-L^N3 . Given an inherent excited-state decay of 5·10⁵ s⁻¹ (lifetime of Ir of about 2 μs) (Giereth et al., 2021) this implies that 11 in 100 photo-excitations (5.5·10⁴ s⁻¹/5·10⁵ s⁻¹ = 0.11) will lead to electron transfer from Ir to Co-L^N3 under the catalytically relevant conditions with 5·10^-5 M Co-L^N3 .

Similar bimolecular rate constants were determined by Chan and co-workers (Chan et al., 2015), who investigated the photocatalytic CO₂ reduction in acetonitrile with [Co^II(tpa)Cl]Cl (tpa = tris(2-pyridylmethyl)amine) and using Ir(ppy)₃ (ppy = 2-phenylpyridine anion) as photocatalyst. Stern–Volmer analysis revealed quenching rate constants for the photosensitizer of 5.29 × 10⁹ M⁻¹s⁻¹ for [Co^II(tpa)Cl]Cl and 4.14 × 10⁴ M⁻¹s⁻¹ for TEA, respectively. It was further noted that reductive quenching of excited [Ir(ppy)₃] is not feasible from thermodynamic consideration (E_1/2 [(Ir(ppy)₃)^*/(Ir(ppy)₃)⁻] = +0.31 V vs. SCE) and, hence, oxidative quenching of the PS is the more favourable reaction path. These findings are also in line with previous studies by some of us (Giereth et al., 2021), in which we used Ir in the presence of a large excess of TEA and a dirhenium catalyst. In the photocatalytic experiments, Ir (E_1/2[(Ir(dFppy)₃)*/(Ir(dFppy)₃)⁻] = +0.36 V vs. Fc/Fc⁺) (Teegardin et al., 2016) seemingly had no influence on the catalytic performance, because the dirhenium compound was itself an efficient photocatalyst and because of very inefficient reductive quenching of Ir by TEA. It is interesting to note that smaller quantities of a stronger reductant (such as BIH, 1,3-dimethyl-2-phenylbenzimidazoline) gave significantly greater turnover numbers and led to very fast CO₂ transformation.

Though the fluorinated Ir complex is a somewhat stronger photo-oxidant than Ir(ppy)₃ (+0.36 V vs SCE compared to +0.31 V vs SCE, see above) one may assume that the rate constant for bimolecular quenching of photoexcited Ir by TEA is on a similar order of magnitude, roughly 4.14·10⁴ M⁻¹s⁻¹ (see above). Under the catalytically relevant reaction conditions in which 0.36 M TEA is present, this leads to a pseudo-first order rate constant of 1.5·10⁴ s⁻¹. Co-L ^N3 is present at 5·10^-5 M concentration under the catalytically relevant conditions, and given a rate constant of 1.1·10⁹ M⁻¹s⁻¹ for bimolecular electron transfer from photoexcited Ir to Co-L ^N3 (see above), one obtains a pseudo-first order rate constant of 5.5·10⁴ s⁻¹. Thus, under the assumption that reductive excited-state quenching of Ir by TEA is not more rapid than reductive quenching of [Ir(ppy)₃] despite its 0.05 V higher oxidative power (see above), oxidative quenching of Ir by Co-L ^N3 is a factor of 3.6 (= 5.5·10⁴ s⁻¹/1.5·10⁴ s⁻¹) faster than reductive quenching by TEA.

Therefore we speculate that the reaction mechanism is as illustrated in Scheme 2. Following the absorption of light by the photosensitizer, the excited PS* transfers an electron to the catalyst. Subsequently, PS⁺ is reduced by TEA to reform PS, which then restarts the cycle (Pellegrin and Odobel, 2017). The reduced catalyst cat⁻ is probably only able to interact with protons, derived from TEA decomposition, to form a hydride intermediate. Consecutive electron and proton transfer results in hydrogen formation (Dempsey et al., 2009; Darmon et al., 2014; Wiedner and Bullock, 2016). In order to be able to reduce CO₂ to CO, the catalyst very likely needs to be present in the double reduced form cat²⁻. The double reduced species could form in a dark reaction by further reduction of cat⁻ by the radical cation TEA^•+, which is a potent reductant (Shimoda et al., 2018). Alternatively, cat²⁻ could result from a disproportionation reaction (indicated in Scheme 2) as has been proposed for other 3d transition metal compounds (Takeda et al., 2017; Dalle et al., 2019). Such disproportionation reactions have likewise been demonstrated to play a key role in photoinduced charge accumulation (Skaisgirski et al., 2017). In another alternative, the attack of two single reduced catalyst species on CO₂ could be a possible reaction path, as has been suggested for rhenium complexes (Morris et al., 2009), though this requires the formation of a ternary encounter complex in solution. Further investigations to elucidate the reaction mechanism by characterization and isolation of intermediates are currently performed.

After purification Fe-L ^N3 was obtained as brownish red solid (38 mg, 54%).

ESI-MS: m/z: calc. for [M + OTf ^- ]⁺, 516.0974; found, 516.0974.

UV/vis: λ 268, 312, 432 nm.

After purification Co-L ^N3 was obtained as pink solid (49 mg, 38%).

¹H NMR (400 MHz, CD₃CN): δ 9.24 (d, J = 5.8 Hz, 1H), 8.33 (td, J = 1.5, 7.8 Hz, 1H), 8.13 (td, J = 1.5, 7.8 Hz, 1H_pico), 7.86 (m, 2H_pico), 7.72 (d, J = 7.9 Hz, 1H_pico), 7.47 (t, J = 7.4, 1H_pico), 7.21 (d, J = 5.9 Hz, 1H_pico), 6.67 (s, NH), 5.09 (d, J = 16.5 Hz, 1H_alkyl), 4.64 (d, J = 16.5 Hz, 1H_alkyl), 4.46 (m, 2H_alkyl), 4.10 (m, 2H_macro), 3.79 (m, 3H_macro), 3.45 (m, 1H_macro), 3.31 (dd, J = 6.6, 15.2 Hz, 1H_macro), 3.23 (dd, 5.8, 12.5 Hz, 1H), 3.01 (dd, 5.3, 13.7 Hz, 1H_macro), 2.77 (td, 5.9, 13.7 Hz, 1H_macro), 2.54 (dd, 5.9, 13.7 Hz, 1H_macro), 2.01(m, 1H_macro).¹³C NMR (400 MHz, CD₃CN): δ 164.83 (C_quart), 162.48 (C_quart), 153.65 (C_pico), 149.45 (C_pico), 141.83 (C_Pico), 140.83 (C_Pico), 126.69 (C_Pico), 126.61 (C_Pico), 125.51 (C_Pico), 129.94 (C_Pico), 69.43 (C_alkyl), 67.27 (C_alkyl), 63.92 (C_macro), 62.37 (C_macro), 61.53 (C_macro), 60.16 (C_macro), 54.23 (C_macro), 52.93 (C_macro). ¹⁹F NMR (300 MHz, CD₃CN): δ -76.95 (F_triflat).

ESI-MS: m/z: calc. for [M + OTf ^- ]⁺, 519.0956; found, 519.0975.

UV/Vis: 268, 362, 507 nm.

General procedure for the synthesis of Ni-L ^N3 , Fe-L ^N2S , Co-L ^N2S and Ni-L ^N2S . The macrocycles L ^N3 and L ^N2S were dissolved in dry MeCN (5 ml) and M(OTf)₂ (1 eq.) was added. The suspension was stirred for 3 days at room temperature and the solvent was removed subsequently. The resulting oil was washed with CH₂Cl₂ and toluene [Fe(OTf)₂ and Co(OTf)₂] or THF [Ni(OTf)₂], dissolved in MeCN and the remaining chloride ions were precipitated using an excess of Ag(OTf). The filtrate was evaporated and afterwards the residue was taken up in a small amount of MeCN and precipitated using Et₂O.

After purification Ni-L ^N3 was obtained as beige solid (39 mg, 30%).

ESI-MS: m/z: calc. for [M + OTf ^- ]⁺, 518.0984; found, 518.1018.

UV/Vis: 268, 308 nm.

After purification Fe-L ^N2S was obtained as dark brown solid (34 mg, 46%).

ESI-MS: m/z: calc. for [M + OTf ^- ]⁺, 533.0586; found, 533.0584.

UV/vis: 268, 490, 502 nm.

After purification Co-L ^N2S was obtained as red solid (31 mg, 50%).

ESI-MS: m/z: calc. for [M + OTf ^- ]⁺, 536.0571; found, 536.0568.

UV/vis: 268, 490 nm.

After purification Ni-L ^N2S was obtained as light brown solid (28 mg, 44%).

ESI-MS: m/z: calc. for [M + OTf ^- ]⁺, 535.0590; found, 535.0590.

UV/vis: 268 nm.