New chlorophylls designed by theoretical spectroscopy and machine learning

Abstract The main pigment for oxygenic photosynthesis, chlorophyll (Chl) a, is structurally related to several other Chl variants, naturally occurring mostly with mono-oxidized substitutions. These include Chl b, Chl d and Chl f, divinyl chlorophyll (DVChl) a and b, and 8¹-hydroxy-Chl and (the artificial) 3-acetyl-Chl a. In this contribution, we computationally explore an expanded set of over 250,000 Chl variants, looking for potentially interesting targets for synthetic biology. We focus on optical properties, employing a machine learning (ML) approach and subsequently verifying the corresponding predictions using time-dependent density functional theory (TD-DFT) and multireference DFT (DFT/MRCI). We find that (i) Chl f is the best monosubstituted red-shifted Chl, as no other Chl in our set exceeds Chl f in terms of both red shift and absorption intensity, (ii) Chl b is not the best Chl to harvest photons from the green region of the optical spectrum, as several other Chls with the same or better green absorbance were identified (most notably DVChl b) and (iii) the T₁ energy of Chls can be slightly adapted. The latter would enable experiments to check if it is beneficial to have the T₁ transition energy located between the two lowest O₂ singlet state transitions, as it is found for Chl a; this might be a prerequisite for stable, efficient oxygen generation. Our ML approach thus provides a thorough overview on an extensive subset of potential Chl modifications which could be used for tuning oxygenic photosynthesis, if suitable synthesis pathways can be found.