Discovery and molecular basis of subtype-selective cyclophilin inhibitors

Although cyclophilins are attractive targets for probing biology and therapeutic intervention, no subtype-selective cyclophilin inhibitors have been described. We discovered novel cyclophilin inhibitors from the in vitro selection of a DNA-templated library of 256,000 drug-like macrocycles for cyclophilin D (CypD) affinity. Iterated macrocycle engineering guided by ten X-ray co-crystal structures yielded potent and selective inhibitors (half maximal inhibitory concentration (IC50) = 10 nM) that bind the active site of CypD and also make novel interactions with non-conserved residues in the S2 pocket, an adjacent exo-site. The resulting macrocycles inhibit CypD activity with 21- to >10,000-fold selectivity over other cyclophilins and inhibit mitochondrial permeability transition pore opening in isolated mitochondria. We further exploited S2 pocket interactions to develop the first cyclophilin E (CypE)-selective inhibitor, which forms a reversible covalent bond with a CypE S2 pocket lysine, and exhibits 30- to >4,000-fold selectivity over other cyclophilins. These findings reveal a strategy to generate isoform-selective small-molecule cyclophilin modulators, advancing their suitability as targets for biological investigation and therapeutic development. DNA-templated compound library screening and structure-guided hit optimization resulted in the identification of selective macrocyclic inhibitors of cyclophilin isoforms CypD and CypE.

for biological probes and for potential therapeutics to minimize unwanted side-effects from off-target cyclophilin perturbation. To our knowledge, no subtype-selective cyclophilin inhibitors have been described for any cyclophilin isoforms 1,3 . Efforts to develop selective cyclophilin inhibitors are stymied by the high sequence identity (61-86%) and highly conserved structures of human PPIase domains 1 .
A key structural feature shared among all cyclophilin isoforms is a shallow and highly conserved PPIase active site 1,30 (Extended Data Fig. 1a,b). Classic cyclophilin inhibitors such as cyclosporine A (CsA) bind to this active site 1,30 and thus equipotently inhibit most cyclophilins (Extended Data Fig. 1a,c and Supplementary Fig. 1). Adjacent to the active site is the S2 pocket, a distinct exo-site that forms part of a long substrate-binding groove for peptides (Extended Data Fig. 1a). S2 pocket residues are much more diverse than active site residues among cyclophilins, providing an opportunity for isoform-selective binding 1 (Extended Data Fig. 1d). In particular, three gatekeeper residues (positions 123, 124, and 145 in CypD) and one far S2 pocket residue (position 118 in CypD) are highly diverse among cyclophilins (Extended Data Fig. 1e). While S2 pocket binding has been recognized as a strategy to achieve isoform-selective inhibition 1 , current small-molecule cyclophilin inhibitors when profiled show poor isoform selectivity and are not known to exploit interactions with non-conserved S2 exo-site residues 3,[30][31][32][33][34][35][36][37] .
In this study we report the development of novel cyclophilin inhibitors from a DNA-templated macrocycle library and iterated structural biology and small-molecule engineering, yielding potent and isoform-selective CypD and CypE inhibitors. These findings establish a strategy for the development of subtype-selective = 0.18 μM), retain CypD potency, but no longer potently inhibit NKTR, CypH, or CypG, with some selectivity over PPWD1 (Fig. 2a,b and Supplementary  Fig. 5a,b). The S2 pockets in these cyclophilins are more sterically occluded or rigid than in the other seven tested cyclophilins, likely clashing with large hydrophobic moieties in B1 and B2 (Fig. 2d). To gain further insight into the basis of emerging CypD subtype selectivity, we solved co-crystal structures of B1, B2, and a smaller derivative B3 bound to CypD ( Fig. 2c and Supplementary Fig. 5c; PDB IDs 7TGU, 7TGV, 7TH1). These structures revealed deeper S2 pocket binding by the benzophenone and biphenyl groups of B1 and B2, respectively, while retaining the original active site interactions of the parent compound A26. We also observed migration of the R124 side-chain contingent on the size of the S2-binding moiety. As the ligand grew in size from B3 → B2 → B1, the R124 side-chain gradually moved from tucked in the pocket to fully solvent-exposed ( Fig. 2c and Supplementary Fig. 5c).
These observations suggest the need to displace some residues within the S2 pocket during ligand binding to achieve partial CypD subtype selectivity. While the polarity and hydrophilicity of R124 in CypD minimizes the energetic cost of doing so, CypG, CypH, NKTR, and PPWD1 contain large hydrophobic groups or inflexible residues buried in the S2 pocket, and moving their side-chains out of the pocket would likely incur a large energetic penalty (Fig. 2d). This observation was supported by additional macrocycles for which the largest, most hydrophobic derivatives (such as B5) are more selective for CypD than smaller or more hydrophilic analogs (B4, B6, B7) (Supplementary Fig. 6a-d).
To further improve selectivity for CypD, we diversified the biphenyl ring of B2, which was positioned in our co-crystal structures near K118, S123 and R124 within CypD's S2 pocket. We reasoned that establishing favorable interactions with these residues might further improve subtype selectivity. While derivatives that contain various alkyl, alkoxy, phenols, or heterocycles at varying positions on the distal ring of the biphenyl (B8-B14) provide no additional benefit, these biphenyl derivatives maintain the same selectivity profile as B2 ( Supplementary Fig. 7a-g). We also observed that para-carbonyl moieties, such as B15 ( IC CypD 50 = 0.038 μM), show promising potency with a preference of 2-to 20-fold for CypD over CypC and Cyp40 (B15-B19; Supplementary Fig. 8a-e). Replacement of the acetyl group of B15 with a carboxylate resulted in a tenfold decrease in CypD potency (B20, IC = 0.0030 μM; Fig. 3a,b and Supplementary Fig. 9f) results in potency and selectivity profiles that are similar to B23. The selectivity and potency of these carboxylates are attenuated upon migrating the position of the carboxylate (B26, B27; Supplementary Fig. 10a,b), or masking the negative charge of the carboxylate through an ester or nitro group (B28-B30) ( Supplementary Fig. 10c-e). Replacing the carboxylate with an amine results in a 500-to 1,000-fold reduction in potency for CypD, and a changed selectivity profile that favors PPWD1 and CypC (B31, B32) ( Supplementary Fig. 11a, b). Nitriles B33 and B34 also exhibit a similar linker-length-dependent potency profile, with B34 showing similar potency to B23, yet with reduced selectivity ( Supplementary Fig. 12a,b). Collectively, these results suggest that carboxylate-containing biphenyl-substituted inhibitors gain CypD potency though a novel interaction that is both charge selective and hydrogen-bond dependent. Moreover, B23, B24, and B25 establishes that iterative diversification deeper into the S2 pocket with precisely tailored chemical properties can result in enhanced selectivity for CypD.
Structural and mutagenesis studies improve CypD selectivity. We solved a co-crystal structure of B23 or B25 bound to CypD ( Fig. 3c; PDB IDs 7TH7 and 7THC). While retaining the same interactions with CypD as B2, we also observed a novel hydrogen bond with the backbone of S119 and a salt bridge with the side-chain of K118, which is usually oriented away from the S2 pocket (Extended Data Fig. 4a). The latter interaction explains why cyclophilins that replace the K118 residue with a glutamate (Cyp40, PPWD1), or a hydrophobic residue (CypC, PPIL1) are much less potently inhibited by B23   Fig. 1 | Selection of a DNA-templated library of 256,000 macrocycles for CypD affinity reveals novel cyclophilin inhibitors. a, Generalized trends in inhibition potency of CypD prolyl-isomerase activity from cyclophilin inhibition profiles of library hits and A-series macrocycles. b, JOMBt, showing weak and promiscuous cyclophilin inhibition of prolyl-isomerase activity. c, Compound A26 showing improved CypD potency but promiscuous inhibition. d, X-ray co-crystal structure of compound A26 (cyan) bound to CypD (PDB ID 7TGT, 1.06 Å resolution), shown as a space-filling model. A26 has a dual-binding mode involving the active site (red) and S2 pocket (green) of CypD. e, Active site binding interactions with A26. The phenyl group provides the primary hydrophobic interactions with F102, M103, A143, F155, L164, and H168. Black dashes show predicted hydrogen bonding interactions with R97, Q105, G114, N144, and W163 and the backbone of the A26. f, S2 pocket binding pose of the furan of A26, exhibiting a shallow interaction that does not engage non-conserved residues K118 (orange), S123 (magenta), and R124 (magenta) on the far side of the pocket. IC 50 values reflect mean ± s.e.m. of three technical replicates. Data points and error bars reflect mean ± s.d. of individual assays at one dose.
= 0.003 μM), consistent with the role of a salt bridge with K118 in enhancing CypD potency and selectivity ( Supplementary Fig. 13b,c).
These observations refined our model for B23 binding. Given the proximal placement of the carboxylate of B23 to S123  and R124, these two residues may also contribute to the partial selectivity of B23 for CypD. We therefore generated S123E, R124A, and R124K CypD mutants, which retain PPI activity ( Supplementary Fig. 3c), and assayed inhibition by B23 and B25. B23 and B25 exhibit reduced potency against CypD S123E as compared to wild-type CypD ( Supplementary Fig. 14b,c).  CypD S123E shares the exact same S2 pocket residues with CypB ( Supplementary Fig. 14f), and for both proteins we observe a decrease in IC 50  = 0.008 μM). We reasoned that S123 can better tolerate the carboxylate on B23 in comparison to a negatively charged Glu at this position in CypB. This hypothesis was further supported by the conformational difference in the S123 backbone loop between co-crystal structures of B2 versus B23 bound to CypD (Extended Data Fig. 4a,b). Migration of the S123 loop in the B23-CypD structure compared to B2-CypD suggests that this conformational shift must be tolerated for any cyclophilin bound to B23 (Extended Data Fig. 4a,b). Indeed, other cyclophilins with Glu at the 123 position (CypA, CypB, CypC, Cyp40) show weaker inhibition by B23 compared to CypD. Surprisingly, neither B23 nor B25 show any appreciable difference in potency for CypD R124A or R124K compared to wild-type CypD, suggesting that R124 plays a minimal role in B23 or B25 binding ( Supplementary  Fig. 15b,c). Collectively, these observations suggest that the S123 gatekeeper plays a role in the selectivity profile of B23 and B25.
Using these co-crystal structures and mutational analyses, we sought to achieve improved CypD selectivity over cyclophilins that share K118 (CypA, CypB, CypE) by installing carboxylate groups predicted to be positioned closer to the non-conserved S123 in CypD. Several of the resulting inhibitors retain similar potencies to B23, but none offer improvements in selectivity (B35-B39; Supplementary  Fig. 16a-e). Many deleterious modifications were observed, including functionalizing the ethylene linker of B23, or altering the position of the carboxylate on the biphenyl group (B40-B50; Supplementary  Fig. 17a-k). These modifications presumably change the location and binding mode of the carboxylate with respect to K118 and S119, thereby decreasing potency with no selectivity benefit. We reasoned that to gain further CypD selectivity, we must maintain the carboxylate's interactions with K118 and S119 and add groups that can be presented to the S123 position. As the analogous S123 position is a more sterically bulky lysine in CypE, and a negatively charged glutamate in CypA and CypB, presenting a carboxylate group closer to this region could contribute additional CypD selectivity over these cyclophilins (Extended Data Fig. 1d,e). We subsequently made a disubstituted nitrile/carboxylate biphenyl derivative (B51), similar to B34, and two dicarboxylate moieties (B52, B53) to test this hypothesis.
While the nitrile of B51 only hampers the selectivity compared to the B23 carboxylate ( Supplementary Fig. 18a), B52 and B53, which contain malonate or glutarate moieties, respectively, retain CypD potency (B52 IC = 0.057 μM) compared to B23, while further improving CypD selectivity (Fig. 4a,b and Supplementary Fig. 18b,c). B52 and B53 show ~15-to 30-fold selectivity for inhibiting CypD over the most closely related cyclophilins, CypE and CypB, 60-to 900-fold selectivity over CypA and PPIL1, and >900-fold selectivity against the remaining six cyclophilins (Extended Data Fig. 5). The 100-fold selectivity of B52 and B53 for CypD over CypA is especially noteworthy since CypA is the most abundantly expressed cyclophilin in human cells, and one of the most abundantly expressed intracellular proteins (Supplementary Table 1). We then solved co-crystal structures of B52 and B53 bound to CypD ( Fig. 4c; PDB IDs 7THD, 7THF). We observed a similar binding mode as B23, maintaining a salt bridge with K118, but now presenting a second carboxylate in the proximity of the S123 and R124 residues, as designed. For B52, we also observed a novel hydrogen bond with the backbone atoms between S123 and R124. We therefore attributed the selectivity of B52 and B53 for CypD over CypA, CypB, and CypE to the presentation of the second carboxylate near S123 (Fig. 4d).
To test this basis of selectivity, we assayed B52 and B53 against CypD gatekeeper mutants S123E, R124A, and R124K (Supplementary Figs 14d,e and 15d,e), along with K118A and K118E mutants ( Supplementary Fig. 13d,e). We observed similar trends as seen with monocarboxylates B23 and B25 compared with wild-type CypD-decreased potency for CypD S123E (B52 = 4 μM) mutants upon removing the possibility of an inhibitor-protein salt bridge. These results suggest that B52 and B53 achieve substantial CypD selectivity on the basis of favorable contacts with its K118 and S123 residues, and on the absence of such interactions with homologous positions in other cyclophilins.
To further establish the dependence of CypD inhibitor potency on S2 pocket residues, we sought to rescue the attenuated potency of B52 and B53 for both CypB and CypA by installing gatekeeper mutations to match the residues of CypD in this pocket (CypB  Supplementary Fig. 20d,e). These trends were also conserved to a lesser degree for monocarboxylates B23 and B25 (Supplementary Figs 19b,c and 20b,c). Inhibition trends were maintained with fluorescein-labeled A26 (A26-Fl), B52 (B52-Fl), and B53 (B53-Fl), using a fluorescence polarization assay to measure binding against 11 prolyl-isomerase-active and five catalytically impaired or inactive cyclophilins (Extended Data Fig. 7a-c). While A26-Fl promiscuously binds the 11 prolyl-isomerase-active cyclophilins, B52-Fl and B53-Fl exhibit selective CypD binding, corroborating prolyl-isomerase inhibition (Figs 1c and 4b). We also observed weak binding to the five impaired or non-PPIase active cyclophilins by A26-Fl, B52-Fl, and B53-Fl, suggesting that their macrocycle scaffold cannot target these cyclophilins. Collectively, these results confirm that interactions between inhibitors and S2 pocket residues are strong determinants of CypD potency and selectivity.
We also reasoned that we could modify the S2 pocket of CypD to accommodate the amine containing compound B32, as it shows an alternate selectivity profile compared to most of our inhibitors, but has poor overall potency (Extended Data Fig. 6c,d and Supplementary Fig. 11b). We observed that replacing cationic residues within the S2 pocket of CypD (K118E, R124A) improves B32 potency by 5-to 6-fold compared to wild-type CypD (IC = 0.06 μM), a 100-fold improvement compared to wild-type CypD (Extended Data Fig. 6c). B32 exhibits good selectivity for CypD K118E/R124A over wild-type cyclophilins, with at least 13-fold preference over the nearest cyclophilin, PPWD1 ( Supplementary Fig. 11b) and >100-fold selectivity over CypA. Overall, B32 and CypD K118E/  K118 B52 S119 S123 R124 K118 B53 S119 S123 R124 Active site binding is identical to A26 (Fig. 1e). Yellow dashes indicate predicted hydrogen bonds. d, List of residues on the far side of the S2 pocket of cyclophilins that are proximal to the ligand carboxylates. Both compounds retain CypD potency similar to that of mono-carboxylate B23, while enhancing selectivity over CypA, CypB, Cype, and PPIL1. The malonic and glutaric acids of B52 and B53, respectively, position the carboxylate in a similar pose as B23 (Fig. 3c), while presenting a second carboxylate to the S123 residue. B52 forms a predicted hydrogen bond with the peptide backbone of S123-R124. R124 is pushed out of the S2 pocket, consistent with other macrocycles containing large S2-binding groups such as B1. B52 and B53 achieve selectivity over CypA and CypB through charge repulsion with a glutamate at the analogous 123 position, while creating a steric clash with PPIL1 and the lysine of Cype at this same position. IC 50 values reflect mean ± s.d. of four independent replicates (each comprising three technical replicates). Graphs show a representative single independent replicate (independent replicate 1 is shown, containing three technical replicates) with data points and error bars reflecting mean ± s.d. of individual assays at one dose. Further independent replicates are shown in Supplementary Fig. 18b,c.
R124A provide another cyclophilin-ligand pair with selective inhibition dependent on S2 pocket identity.

CypD inhibitors are active in mitochondria.
The lack of reliable functional CypD assays in intact cells makes it difficult to directly measure CypD inhibition in intact cell culture. Consistent with multiple previous reports, our efforts to use a calcein-Co 2+ assay and mitochondrial membrane potential dyes in cells to characterize our compounds with CsA as a control failed to reliably assess mPTP opening in a CypD-inhibition-dependent manner and were also prone to confounding signals [39][40][41][42][43][44] . For these reasons, CypD inhibitors have mostly been characterized biochemically in vitro, or in isolated mitochondria 13,16,[45][46][47] .
We tested the ability of our CypD-selective macrocycles to inhibit CypD in active mitochondria isolated from mouse liver. First, we verified that modifying groups on the 'tail' position of B52 and B53 (for example, B52-A, B53-A) do not affect potency or selectivity, and that their enantiomers do not inhibit any cyclophilin (*B52-A, *B53-A) ( Supplementary Fig. 21a-d). Since CypD is exclusively found in mitochondria, we appended Cy5, an established mitochondrial localization group 48 , to all four macrocycles at the 'tail' position (B52-Cy5, B53-Cy5, *B52-Cy5, *B53-Cy5). We observed minimal change in potency or selectivity for CypD in vitro upon addition of the Cy5 group ( Supplementary Fig. 22a-d). We then tested compounds B52-Cy5 and B53-Cy5 in isolated mouse liver mitochondria, measuring their calcium retention before a mPTP opening event. In comparison to the dimethysulfoxide (DMSO) control, we observed substantially increased calcium retention capacity before mPTP opening when pretreated with CsA, B52-Cy5, or B53-Cy5 (Fig. 5a-d and Supplementary Fig. 23a,b). Importantly, inactive enantiomers *B52-Cy5 and *B53-Cy5 do not inhibit CypD in vitro (Fig. 5a-d and Supplementary Fig. 23a,b) and do not influence calcium retention capacity before mPTP opening, strongly supporting CypD-dependent inhibition of mPTP opening by B52-Cy5 and B53-Cy5.
As expected, while B52-Cy5 and B53-Cy5 engage CypD in isolated mitochondria, we did not observe efficient plasma membrane permeability of B52-Cy5, B53-Cy5, or their inactive enantiomers by fluorescence microscopy (Extended Data Fig. 8a-i and Supplementary Fig. 24b-f), presumably because of the presence of dicarboxylate groups. To improve cell permeability, we used a pro-drug strategy and prepared both sets of active and inactive enantiomers as ethyl esters (B52-Et-Cy5, B53-Et-Cy5, *B52-Et-Cy5, *B53-Et-Cy5) ( Fig. 5e and Supplementary Fig. 24a). We observed strong mammalian cell permeability and mitochondrial localization for all four ester-containing compounds (Fig. 5f, Extended Data Fig. 8a-i and Supplementary Fig. 24b-f). These ester derivatives did not potently inhibit the prolyl-isomerase activity of CypD, consistent with the importance of the dicarboxylate groups ( Supplementary Fig. 25). All four pro-drug compounds were hydrolyzed to their active dicarboxylate forms in vitro with human carboxylesterase (CES1), and B52-Et-Cy5, *B52-Et-Cy5, and B53-Et-Cy5 were readily converted to the corresponding di-acids intracellularly by endogenous esterases in a variety of human (A549, HeLa, HEK293T, HepG2) and mouse (mouse embryonic fibroblasts (MEFs)) cell lines (Extended Data Fig. 9a-h).
These findings, coupled with robust CypD-dependent inhibition of mPTP opening in isolated mitochondria, suggest that pro-drug versions of Cy5-conjugated B52 and B53 can access mitochondria and release active CypD-selective inhibitors in mammalian cells. These observations make esterified B52 and B53 attractive candidates for potent and selective probes of CypD activity in biological systems.

Development of CypE-selective inhibitors.
Since CypD-selective inhibitors B52 and B53 were designed from a slightly promiscuous inhibitor B2, we next explored the possibility of selectively modulating a different cyclophilin by using covalent inhibition. While most covalent inhibitors target catalytic residues, the non-conserved residues in cyclophilin S2 pockets are both solvent-exposed and non-catalytic. Many cyclophilins contain non-catalytic lysines in their S2 pocket (Extended Data Fig. 1d,e). Although non-catalytic lysines are typically protonated at physiological pH, aryl boronic acid carbonyls have been shown to modify lysine and N-terminal groups covalently and reversibly through the formation of iminoboronates. Previous studies have incorporated aryl boronic acid carbonyl warheads on inhibitors to covalently modify non-catalytic lysines 49 . We hypothesized that installing this reactive group on the biphenyl of inhibitor B2 might allow a reversible covalent interaction with one of these cyclophilins, potentially providing a new subtype-selective inhibitor. We synthesized ketone (C1A, C2A) and aldehyde boronic acids (C3A, C4A) that could place this warhead close to cyclophilin S2 pocket lysine residues ( Supplementary Fig. 26a-d).
We screened these four compounds in a fluorescence polarization competition assay with A26-Fl against seven lysine-containing cyclophilins ( Supplementary Fig. 26a-d), identifying C3A as a potent inhibitor of A26-Fl binding to CypE (K i of 0.072 µM) (Fig. 6a,b and Supplementary Fig. 26c). Additionally, C3A showed ≥10-fold selectivity for CypE over other tested cyclophilins. Shifting the aldehyde group to the meta position of the ring (C4A) attenuated CypE potency and selectivity, suggesting improper placement of C4A's aryl boronic acid carbonyl towards the lysines of CypE ( Supplementary Fig. 26d). Replacing the aldehyde with a ketone (C1A, C2A) also decreased potency for CypE compared to C3A ( Supplementary Fig. 26a,b). Compounds C5A and C6A containing either the aldehyde or the boronic acid alone, respectively, showed reduced potency by 16-fold and 100-fold, respectively, relative to C3A ( Supplementary Fig. 27a-c). Since the potency of C3A is contingent on both the boronic acid and the aldehyde, these results suggest C3A is acting covalently through iminoboronate formation with CypE, consistent with the lysine-rich S2 pocket of CypE (Extended Data Fig. 1d,e).
Mass spectrometry did not detect a clear lysine-modified covalent adduct between C3A and CypE (Extended Data Fig. 10a). We speculated that reducing the iminoboronate intermediate with sodium cyanoborohydride (NaCNBH 3 ) might trap the covalent adduct. Indeed, co-incubating C3A in fivefold excess for 1 h with CypE followed by treatment with 25 mM NaCNBH 3 resulted in observation of a +806 Da adduct, corresponding to the reductive amination product minus water (Fig. 6d), but not for boronic acid C6A (Extended Data Fig. 10b). The loss of water has been reported for other aryl boronic acid carbonyl inhibitors treated with NaCNBH 3 49 . These findings suggest that C3A functions in a reversible covalent manner. We also observed +779 Da covalent adduct formation under these conditions with aldehyde-only C5A, suggesting that carbonyl groups at this position can form imines with S2 pocket lysines of CypE (Extended Data Fig. 10b). Nonetheless, the 75-fold higher CypE inhibition potency of C3A over C5A suggests that this reversible interaction is substantially stronger for C3A. Little or no covalent modification for 12 other cyclophilins was observed when treated with C3A and NaCNBH 3 , establishing the selectivity of this covalent interaction ( Supplementary Fig. 29a,b).

Discussion
We tailored promiscuous initial ligands from a DNA-templated library that bind cyclophilin active sites through iterated cycles of structural studies, mutagenesis, and small-molecule engineering to generate what are, to our knowledge, the first reported potent and subtype-selective inhibitors of CypD and of CypE. Cyclophilin subtype selectivity emerged through carefully engineered engagement with poorly conserved residues in the S2 pocket, an exo-site adjacent to each cyclophilin's active site. Our results also suggest the feasibility of discovering subtype-selective cyclophilin ligands through parallel selections using a panel of subtypes to identify putative selective binders through their selective enrichment.
Structure-activity relationships and mutational analyses refined our model for CypD selectivity, which we applied to develop other selective cyclophilin inhibitors. CsA, an active-site-binding cyclophilin inhibitor, shows minimal selectivity for CypD over other cyclophilins and mutants (Extended Data Figs 1a-c and 5, Supplementary Table 2, and Supplementary Figs  13a, 14a, 15a, 19a, 20a and 31a). By contrast, our inhibitors that bind both the active site and S2 pocket of CypD offer improved selectivity (Extended Data Fig. 5 and Supplementary Table 2). The inclusion of large hydrophobic groups (B1, B2) into the S2 pocket eliminated off-target inhibition of CypG, CypH, and NKTR owing to the presence of inflexible or hydrophobic residues within their S2 pockets (Fig. 2a- Selectivity for CypD was further improved by installing carboxylates (B23, B25) that can hydrogen bond with S119 and K118, achieving selectivity over non-lysine-containing PPIL1, Cyp40, CypC, and PPWD1 ( Fig. 3a- Table 2). To confer selectivity over K118-containing CypA, CypB, and CypE, we installed a second carboxylate (B52, B53) to interact with the gatekeeper residues on each cyclophilin, which show high potency and selectivity for CypD (Fig. 4a-d). CypD S123 can sterically and electronically tolerate dicarboxylates, while Glu at the analogous position in CypA and CypB presumably repels the dicarboxylate, and the more bulky lysine of CypE may sterically clash with the dicarboxylate (Fig. 4d, Extended Data Fig. 5, and Supplementary Table 2). Complementing these trends is the overall migration of S2 pocket residues in CypD to accommodate ligand binding (Extended Data Fig. 4b and Supplementary Fig. 32). Conformational flexibility of K118, S123, and R124 is dependent on the size and functionality of the binding ligand, as growing the S2-binding moiety of the ligand, while installing hydrogen-bond interactions, induces this residue migration (Supplementary Video 1).
Dicarboxylate groups on the inhibitors developed in this study thus provide a chemical environment that CypD, but not other cyclophilins, can accommodate. This model was further supported with molecular footprinting analysis of JOMBt compared to B52 (Supplementary Fig. 33). JOMBt is reliant on weak binding energy per residue on highly conserved cyclophilin active site residues, while B52 owes its potent CypD binding to unique S2 pocket residues that are poorly conserved between other family members. The conformational flexibility of K118, S123 and R124 is consistent with previous reports 1 that suggest the S2 pocket of each cyclophilin may confer endogenous substrate-binding selectivity, whereby prolyl isomerization of a peptide substrate at the active site is only accessible if the S2 pocket can tolerate upstream amino acids. Indeed, the S2 pockets of each cyclophilin vary in their tolerance of different groups in our inhibitors. This observation is consistent with the S2 exo-site serving as a substrate-binding pocket separate from the active site that confers substrate selectivity. We further established that S2 pocket interactions are the basis of observed subtype selectivity through rational mutagenesis of the S2 pocket of CypD to accommodate a positively charged amine-containing ligand B32 (Extended Data Fig. 6c,d and Supplementary Table 2).
We further extended these insights to achieve selective CypE inhibition. C3A selectively targets CypE though reversible covalent bonding with its S2 pocket K217 residue (Fig. 6a-e). To our knowledge, C3A represents the first CypE-selective inhibitor and the second type of subtype-selective cyclophilin inhibitor after B52 and B53 (Supplementary Table 2). The aryl aldehyde boronic acid forms a reversible covalent bond with the K217 residue of CypE, while the presence of the boronic acid minimizes binding to the other cyclophilins. CypE is one of many cyclophilin proteins reported as a structural component of the spliceosome, but the assessment of its mechanistic role has been hampered by the lack of a cyclophilin-selective inhibitor 26 . The use of CypE-selective inhibitors such as C3A may serve to illuminate the role of CypE in the spliceosome without confounding inhibition of the seven other spliceosome-associated cyclophilins 26 .
The CypD-selective inhibitors developed in this work increase calcium retention capacity and delay mPTP opening in isolated mitochondria (Fig. 5a-d and Supplementary Fig. 23a,b). Ester pro-drug derivatives of these CypD-selective inhibitors also exhibit strong plasma membrane permeability, mitochondrial localization, and the ability to release active inhibitor in mammalian cells ( 8a-i and 9a-h). These compounds may serve as tools to explore the mechanism of action of the mPTP or as probes of the role of CypD in disease models with mPTP-oxidative stress as a phenotype, such as neurodegenerative disorders, IRI, liver diseases, and mitochondrial disorders.
More broadly, this study establishes that subtype-selective cyclophilin inhibition can be achieved through the design of ligands that interact with unique residues in cyclophilin exo-site S2 pockets (Extended Data Fig. 1d,e and Supplementary Table 2). Applying this strategy to develop additional selective cyclophilin inhibitors may unlock the biological or therapeutic potential of this family of targets, which have been relatively unexplored compared to others such as kinases, phosphatases, and proteases.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41589-022-01116-1.

Methods
General. Fmoc-protected amino acids and peptide coupling reagents were purchased from Chem-Impex International. Boronic acids or aryl halides were purchased from Combi-Blocks and Enamine, verified >95% pure by the manufacturer's standards. All other chemical reagents (verified >95% pure by manufacturer), PPIL4 (full length), and PPIL6(C-Myc/DDK), were purchased from Millipore-Sigma. SDCCAG-10(GST) was purchased from Abnova. Human recombinant CES1 and CES2 were purchased from R&D Systems. All purchased proteins were verified >85% pure by the manufacturer using SDS-PAGE. Nuclear magnetic resonance (NMR) spectra were gathered using a Bruker Ascend 400 MHz NMR. Quantification of DNA was completed using a NanoDrop One Microvolume UV-Vis Spectrophotometers (ThermoFisher Scientific). Preparative high-performance liquid chromatography (HPLC) reverse phase purification was conducted with an Agilent 6100 Quadrupole liquid chromatography-mass spectrometry (LC-MS) system using a Kinetex 5 µm C18 100 Å, AXIA Packed LC Column 150 ×30.0 mm (Phenomenex). Silica gel column chromatography was conducted with a Biotage SP1 Flash Chromatography system. Recombinantly expressed proteins were purified using an AKTA pure fast protein liquid chromatography (FPLC). Quantitative PCR analysis was conducted using a CFX96 Touch Deep Well Real-Time PCR System (Bio-Rad). 1  plus GlutaMAX (ThermoFisher Scientific) for HEK293T/HeLa/MEFs, minimum essential medium (MEM; Corning) for HepG2, and F-12K (ATCC) for A549, each supplemented with 10% (v/v) fetal bovine serum (FBS; Gibco, qualified). All cell types were incubated, maintained, and cultured at 37 °C with 5% CO 2 . Cell lines were authenticated by their respective suppliers and tested negative for mycoplasma.

HPLC purity analyses of key compounds.
Analytical analyses for compound purity was performed using an Agilent 6100 Quadrupole LC-MS system with a Kinetex 5 µm C18 100 Å LC Column 150 ×2.1 mm (Phenomenex). Five microliters of 0.5-1 mM compound in DMSO-d 6 was injected and run for 3 min at 10% water/ acetonitrile with 0.1% trifluoroacetic acid to elute off DMSO-d 6 . From 3 min to 15 min, gradient was increased to 100% acetonitrile, and held for 2 more min. Compound peaks were identified using the mass spectrometry trace. Purity analyses was quantified by percentage area of the compound peak at 214 nm absorbance relative to all identified peaks within the 3-17 min analysis window (to exclude the DMSO peak at 1.5 min).
Site-directed mutagenesis of CypD. Mutant CypD constructs were generated with primers for one-piece uracil-specific excision reactions (USER) containing a mutant overhang by PCR amplification of starting plasmids (Supplementary Table 3). PCR product was purified on microcentrifuge membrane columns (MinElute, Qiagen) and quantified by Nanodrop. Fragment (0.2 pmol, 7.5 µL) was combined in a 10 µL reaction mixture containing 0.75 µL (15 U) DpnI (NEB), 0.75 µL USER mix (endonuclease VIII and uracil-DNA glycosylase, NEB), 1 µL 10× CutSmart Buffer (NEB). Reactions were incubated at 37 °C for 30 min followed by heating to 80 °C for 3 min and slow cooling to 12 °C at 0.1 °C s −1 . Constructs were directly transformed into One Shot Mach1 T1 Phage-Resistant Chemically Competent E. coli (Invitrogen) by heat shock for 30 s and streaked on 100 µg mL −1 carbenicillin containing LB agar. Selected colonies with correct construct were verified by Sanger sequencing and cultured overnight in 2× YT supplemented with 100 µg mL −1 carbenicillin. Plasmid was extracted by microcentrifuge membrane columns (QIAprep Spin Miniprep Kit, Qiagen) as per manufacturer's instructions and quantified by Nanodrop.
Site-directed mutagenesis of CypA, CypB, CypD (only for K175I mutant), and CypE. The K175I mutation for reduction of surface entropy 50 was introduced into the PPIF (CypD) construct and the gatekeeper residue mutations were introduced into the PPIA (CypA), PPIB (CypB), and PPIE (CypE) constructs via Quikchange site-directed mutagenesis (Agilent Technologies). Primers for the respective mutations are included in Supplementary Table 4. The template plasmid (~100 ng, 1 µL) was combined in a 25 µL mixture with forward and reverse primers (125 ng, 1.25 µL each), Q5 High-Fidelity DNA Polymerase (NEB) (1 µL, 2000 U mL −1 ), deoxynucleoside triphosphate (dNTP) mix (10 mM, 1 µL), Q5 Reaction Buffer (NEB) (5×, 5 µL), and deionized H 2 O (14 µL). Reactions were incubated at 98 °C for 120 s, followed by 30 cycles of 98 °C (10 s, melting), 55-60 °C (30 s, annealing), and 72 °C (5 min, elongation), followed by a final elongation cycle of 7 min. PCR products were transformed into E. coli DH5α competent cells with a heat shock at 42 °C for 45 s and streaked onto Luria Broth agar plates containing 100 µg mL −1 ampicillin (2BT) or 50 µg mL −1 kanamycin (pET28α LIC). Single colonies were isolated for inoculation and plasmid extraction via microcentrifuge membrane columns (QIAprep Spin Miniprep Kit, Qiagen), and mutations were verified by Sanger sequencing. Table 5) by transforming into One Shot BL21(DE3) chemically competent E. coli (Invitrogen) by heat shock at 42 °C for 30 s. Cells were streaked onto agar plates containing 100 µg mL −1 carbenicillin and incubated at 37 °C for 16 h. Individual colonies were grown up in a 2-L culture of LB medium supplemented with 100 µg mL −1 carbenicillin at 37 °C until optical density reached 0.8. The culture was then cooled to 16 °C for 1 h and protein production was induced by adding 2 mL of 1 M isopropyl β-d-1-thiogalactopyranoside and left to incubate for 16 h. Cells were pelleted at 4,000g for 5 min at 4 °C and resuspended in 50 mL cold Tris-HCl pH 8.0, 50 mM NaCl, 5% glycerol (NiA Low Salt). Two Pierce protease inhibitor tablets (ThermoFisher Scientific) was added to the suspension and cells were subsequently lysed using an Avestin Emulsiflex-C3 homogenizer at 17,000-20,000 p.s.i. Lysed cells were pelleted, and supernatant was purified by FPLC affinity chromatography using a Histrap HP 5 mL (GE) and a gradient of 0-100% NiA low salt/NiB low salt (NiB low salt: NiA low salt + 500 mM imidazole). Protein eluted off around 60-70%, confirmed by SDS-PAGE analysis of fractions. Isolated fractions were additionally purified by FPLC cation exchange, using a HiTrap SP 5 ml column (GE) and a gradient of 0-100% SA buffer/SB buffer (SA buffer: Tris-HCl pH 7.0, 1 mM dithiothreitol (DTT), 5% glycerol; SB buffer: SA buffer + 1 M NaCl). Fractions corresponding to >90% pure CypD came off around 40%, confirmed by SDS-PAGE. Combined fractions were dialyzed in 20 mM Tris-HCl pH 8.0, 50 mM NaCl, 1 mM DTT, 5% glycerol, using a Slid-A-Lyzer molecular weight cutoff = 3,000 dialysis cassette according to the manufacturer's instructions, overnight at 4 °C. Protein purity was >90% on the basis of SDS-PAGE gel electrophoresis and Coomassie staining. Pooled fractions were concentrated, flash frozen in liquid nitrogen, and stored at −80 °C. Protein was quantified using Pierce BCA Protein Assay Kit.  Table 5) were transformed into E. coli BL21(DE3) competent cells with heat shock at 42°C for 45 s. Cells were streaked onto Luria Broth agar plates containing 100 µg mL −1 ampicillin (2BT) or 50 µg mL −1 kanamycin (pET28α LIC) and incubated overnight at 37 °C. Single colonies were inoculated into 2× YT medium supplemented with 1% glucose, 1 mM Mg 2+ , and 100 µg mL −1 ampicillin (2BT) or 50 µg mL −1 kanamycin (pET28α LIC) and shaken at 37 °C until reaching an OD 600 = ~0.6-0.8. The cultures were then cooled to 16 °C for 1 h, and protein production was induced with 1 mM isopropyl β-d-1-thiogalactopyranoside overnight for 16-18 h. Cells were pelleted at 4,000g for 10 min at 4 °C before being homogenized in an Avestin Emulsiflex-C3 High Pressure Homogenizer at 17,000-20,000 p.s.i three times and resuspended in buffer containing 20 mM Tris pH 8.0, 50 mM NaCl, and 5% glycerol. Lysates were centrifuged for 1 h at 17,000 r.p.m. at 4 °C in a Sorvall SLC6000 Fixed-Angle Rotor. Protein was purified from the supernatant via nickel affinity chromatography followed by cation exchange chromatography and size exclusion chromatography. First, the recombinant His 6 -tagged proteins were purified with Ni(II)-affinity chromatography (HisTrap FF, GE Healthcare). The supernatant was run over the column, which was subsequently washed twice with buffer to remove non-specific binding. Two-milliliter fractions were eluted over 12 column volumes by increasing the imidazole concentration to 500 mM. For PPIL2 and PPIL3, the His 6 -tag was cleaved with TEV (2BT) or thrombin (pET28α LIC) overnight in pH 8.0 dialysis buffer containing 20 mM Tris base, 100 mM NaCl, 5 mM BME, 5% glycerol with a 3 kDa molecular weight cutoff filter. Proteins both with or without His-tag were diluted to reduce NaCl concentration in pH 8.0 buffer containing 20 mM Tris, 1 mM DTT, and 5% glycerol and loaded onto a cation exchange column (HiTrap SP, GE Healthcare). The column was washed with six column volumes of buffer to remove non-specific binding. Two-milliliter fractions were eluted over 12 column volumes with increasing salt gradient up to 1 M NaCl. Pooled fractions containing protein were concentrated and loaded onto a size exclusion column (HiLoad 16/600 Superdex 200 prep grade, GE Healthcare). Protein was eluted in pH 7.3 buffer containing 50 mM NaH 2 PO 4 , 100 mM NaCl, and 2 mM EDTA. Protein purity was >90% on the basis of SDS-PAGE gel electrophoresis and Coomassie staining. Pooled fractions were concentrated, flash frozen in liquid nitrogen, and stored at −80 °C. Protein concentration was quantified using Pierce BCA Protein Assay Kit.

Recombinant expression and purification of
Recombinant expression and purification of CypA E81S/K82R, CypB E121S, CypE (131-301) K212A, K217A, and K218A mutants. CypA, CypB, and CypE mutants were purified in a similar manner to wild-type with modification as indicated (see Supplementary Table 5 for expression plasmids). Following initial nickel affinity chromatography and overnight dialysis with thrombin-mediated His 6 -tag cleavage, the remaining protein was again run over a Ni(II)-affinity chromatography column (HisTrap FF, GE Healthcare) and flow-through and wash were collected to remove any unbound protein and His 6 tags. Protein purity in flow-through and wash was >95% on the basis of SDS-PAGE gel electrophoresis and Coomassie staining. Pooled flow-through and wash was concentrated, flash frozen in liquid nitrogen, and stored at −80 °C. Protein concentration was quantified using Pierce BCA Protein Assay Kit.
In vitro selection of a 256,000-member DNA-templated library with human His 6 -CypD. The selection protocol utilized a 256,000-member library and recombinant His 6 -CypD 45-207 adapted from previous work 38 . Resuspended magnetic Ni-NTA beads (DynabeadsHis-Tag Isolation and Pulldown, Invitrogen) (25 μL) were immobilized on a MagJET Separation Rack (ThermoFisher) with the supernatant subsequently removed. Beads were washed twice with 300 μL of 50 mM sodium phosphate pH 8.0, 300 mM NaCl, 0.01% Tween-20, 2 mM TCEP (Tris(2-carboxyethyl)phosphine) (PBST). Forty micrograms of CypD protein was loaded onto solid support in 300 μL PBST, incubated on rotary for 1 h at 4 °C, and washed twice with 200 μL of 50 mM Tris-HCl pH 8.0, 150 mM NaCl, 0.05% Tween-20, 2 mM TCEP (TBST). Immobilized protein was suspended in 100 μL blocking buffer (TBST + 0.1 mg mL −1 bovine serum albumin + 0.6 mg mL −1 yeast total RNA) for 25 min at 4 °C. After supernatant was removed, 20 pmol second-generation 256,000-member DTS library was suspended with bead immobilized protein in 50 µL blocking buffer and incubated for 1 h at 4 °C. Supernatant was removed (FT fraction) and beads were washed three times with TBST, retaining the supernatants (W1, W2, W3). CypD was eluted off Ni-NTA beads by incubating with 50 µL PBST supplemented with 300 mM imidazole (Elution Buffer) for 5 min. The elution was PCR amplified and barcoded with Illumina Primers as previously reported 38 . PCR amplicons were purified by PAGE, extracted, and quantified with KAPA quantitative PCR analysis and QuBit (Invitrogen). High-throughput DNA sequencing was performed on an Illumina MiSeq according to the manufacturer's instructions for sample preparation. Reads generated were analyzed using the Python scripts (shared in Supplementary Note 1 of Supplementary Information) to quantify library barcodes and calculate change in percentage abundance for each library member compared to the pre-selection library.
Chymotrypsin coupled prolyl-isomerase assay. Adapting a previously described protocol 51 , with an Agilent Bravo Velocity 11 with a 384ST head, 90 µL assay buffer (25 mM HEPES pH 7.3, 100 mM NaCl, 0.01% Triton X-100) containing 5.28 nM cyclophilin was added into a flat, clear-bottom, black 384-well plate pre-chilled in a Corning CoolBox at 2-3 °C. Five microliters of compound pre-dissolved in 5% DMSO/assay buffer was then added to appropriate wells and incubated at 2-3 °C for 5 min. Five microliters of 0.5 mM α-chymotrypsin from bovine pancreas (Type II, lyophilized powder, Millipore-Sigma) in 20% 1 mM HCl/Assay Buffer was then added to each well and incubated at 2-3 °C for 5 min. The plate was then quickly transferred to a FLIPR Tetra High-Throughput Cellular Screening System (Molecular Devices). Using the internal liquid handling system of the FLIPR, 1 µL 0.5 mM Suc-AAPF-AMC (Chymotrypsin Substrate II, Millipore-Sigma) dissolved in 0.55 M LiCl/2,2,2-trifluoroethanol was added to each well, which was then mixed for 5 s, followed by immediate fluorescence measurements every 1 s for 330 s using a 360-380 nm excitation LED module and a 400-460 nm emission filter. Final concentrations of the plate include 5 nM cyclophilin, 0.25% DMSO, 25 µM α-chymotrypsin, 5 µM Suc-AAPF-AMC. Raw fluorescence data was analyzed by non-linear regression analysis with Prism v.9 by fitting one-phase association curves to each well. Analyses were halted after the calculated maximum fluorescent value owing to observed fluorescent bleaching during the sustained fluorescent plateau. Rate constants calculated for each well (s −1 ) were then normalized to substrate only (no prolyl-isomerase) and enzyme + substrate only controls. IC 50 values were calculated as the value at which 50% inhibition was achieved on the non-linear regression fitted curve of each compound.
Anisotropy binding assay. Adapted from previous work 52 , titrated cyclophilin in assay buffer (25 mM HEPES pH 7.3, 100 mM NaCl, 0.01% Triton X-100) was incubated with 0.5 nM fluorescein-labeled macrocycle for 6 h at room temperature in a flat-bottom black untreated 96-well plate (Corning). Final assay volume was 100 µL with 0.25% DMSO. Fluorescence anisotropy was then measured using a Tecan Spark plate reader using 492 nm/523 nm excitation/emission settings. Raw fluorescence polarization data was analyzed by non-linear regression analysis with Prism v.9.3.1 by fitting to a one site-total binding equation, providing a dissociation constant (K D ) for each cyclophilin-compound pair.
Competition anisotropy binding assay with A26-Fl. Adapted from previous work 52 , cyclophilin in assay buffer (25 mM HEPES pH 8.0, 100 mM NaCl, 0.01% Triton X-100) at a pre-determined concentration from Supplementary Table 6 was incubated with 0.5 nM A26-Fl for 10 min at room temperature in a flat-bottom black untreated 96-well plate (Corning). Competitor macrocycle was then added to each well and incubated for 24 h. Final assay volume was 100 µL with 0.25% DMSO. Fluorescence anisotropy was then measured using a Tecan Spark plate reader using 492 nm/523 nm excitation/emission settings. Raw fluorescence polarization data was normalized to protein + fluorescent probe (0%) and buffer + fluorescent probe (100%) and K i values were calculated using one site-competitive binding equation with Prism v.9.3.1, importing K D values from A26-Fl (Extended Data Fig. 7).
Surface plasmon resonance analysis of CypD ligands. Adapting a previously used protocol 53

Assessment of covalent modification of CypE and other cyclophilin proteins.
In PCR strips, CypE or other cyclophilin in assay buffer (25 mM HEPES pH 8.0, 100 mM NaCl, 0.01% Triton X-100) was incubated with compound for 1 h at room temperature. Final concentrations were 20 µM CypE, 100 µM compound, 0.5% DMSO, at a final volume of 20 µL. For analysis of lysine-iminoboronate modification, samples were directly submitted to Harvard's Center for Mass Spectroscopy for LC-MS analysis. For amine-lysine modification, samples were treated with 5 μL of 125 mM sodium cyanoborohydride dissolved in 25 mM ammonium bicarbonate solution (pH 8.0) and incubated for 4 h at room temperature. For all other wild-type cyclophilins, protein was treated for 4 h with C3A, followed by 16 h treatment with NaCNBH 3 . The final concentration of NaCNBH 3 was 25 mM. Samples were then submitted for LC-MS as described above. Crystal structure refinement. Diffraction data for JOMBt, A26, B1, B2, and B3 was indexed, integrated, and scaled with autoPROC, and diffraction data for B21, B23, B25, B52, and B53 was indexed, integrated, and scaled with FastDP. Both programs rely on additional functionality within XDS and CCP4. Phases were assigned via molecular replacement in Phaser 54 with the apo structure of CypD K175I (PDB ID 2BIT) for CypD-JOMBt and subsequently with our previously solved structures of CypD K175I complexed with macrocyclic inhibitors as the search model. All refinements to the model were made in PHENIX 55 . Model building was performed in Coot 56 with ligands and waters fit into the initial |F o | − |F c |. Macrocycle restraints were generated using eLBOW (JOMBt, A26, B1, B2, B3) and the ProDrg server 57 (B21, B23, B25, B52, B53). The coordinates of the holo-structures of CypD have been deposited in the PDB. Additional crystallographic and data collection statistics are listed in Supplementary Table 9. Ligand electron densities are displayed in Supplementary Fig. 34.

Co-crystallization of
Molecular footprinting analysis. Molecular footprints, defined as per-residue decomposition of the Van der Waals, electrostatic, and hydrogen bonding energies between the ligand and the receptor, were generated with the crystal structures in the DOCK6.9 molecular modeling software as described by Balius et al. 58 . In brief, two crystal structures were structurally superimposed in UCSF Chimera on the basis of lowest pairwise root-mean square deviation using the Needleman-Wunsch alignment algorithm. Both the reference ligand and B52 were saved in relation to the CypD-B52 protein structure and were parameterized using the GAFF force field and the Gasteiger charging method. The ligands were then rigidly docked into the receptor using DOCK6.9, and the pairwise interaction energies for the top fifty contributing residues were visualized using matplotlib in Python.
Ligand morph movie. The video morphing between the crystal structures of CypD K175I in its apo state (PDB: 2BIT) through JOMBt, A26, B2, B23, and B52 was made in the molecular visualization software suite UCSF Chimera 59 . The six structures were structurally superimposed on the basis of lowest pairwise root-mean square deviation using the Needleman-Wunsch alignment algorithm and morphed with corkscrew interpolation at an interpolation rate of 150 steps between each conformation. Each crystallized ligand pose was saved as a separate Mol2 file in relation to the final trajectory structure. The trajectory was visualized with ligands using a per-frame script and recorded forwards and backwards.
Isolation of mouse liver mitochondria. All procedures used in animal studies were approved by the Institutional Animal Care and Use Committee at Massachusetts General Hospital. Female C57BL/6J mice (Jackson Labs) age 10-12 weeks were anesthetized by isoflurane. The liver from one mouse was rinsed in ice-cold PBS, minced in ice-cold isolation buffer containing 0.28 M sucrose, 10 mM Hepes-KOH pH 7.2, 0.2 mM EDTA and 1% (w/v) bovine serum albumin, gently homogenized with four strokes of a tight-fitting Teflon pestle at 1,000 r.p.m., and then centrifuged for 10 min at 600g at 4 °C. The supernatant was recovered and centrifuged for 10 min at 8,000g at 4 °C. The loose outer buffy coat was rinsed off and the pellet resuspended gently in isolation buffer and the spins were repeated. The remaining buffer coat layer was rinsed off and the pellet resuspended in buffer containing 137 mM KCl, 10 mM Hepes-KOH pH 7.2, 2.5 mM MgCl 2 for a final concentration between 40-80 mg mL −1 as assessed by Bradford. The suspension was kept on ice and all assays performed within 4-6 h following isolation. Quality control was done with every preparation by adding 250 mcg mitochondria to 500 µL assay buffer containing 137 mM KCl, 10 mM Hepes-KOH pH 7.2, 2.5 mM MgCl 2 , 5 mM each of glutamate and malate, and 3 mM KH 2 PO 4 . Sequential 150 µM ADP pulses were then delivered. Only preparations with respiratory control ratio >6 (as assessed by the ratio of ADP-coupled state 3 respiration to state 4 respiration) were used for further experiments. Measurements were made in a custom-built spectrometer.
Mitochondrial calcium retention capacity assays. Two hundred and fifty micrograms of mouse liver mitochondria isolated as above were added to 500 µL assay buffer containing 125 mM KCl, 20 mM Hepes-KOH pH 7.2, 1 mM MgCl 2 , 5 mM each of glutamate and malate, and 3 mM KH 2 PO 4 . 0.5 µM Calcium Green 5N (Molecular Probes) was included to monitor extramitochondrial free Ca 2+ . Fluorescence was continuously monitored in the custom-built fluorimeter described above, with excitation 470 nm and emission 520-560 nm. Sequential 60 µM CaCl 2 pulses were delivered until Ca 2+ uptake ceased and an abrupt release of previously taken up Ca 2+ was observed, consistent with mPTP opening. The calcium retention capacity ratio was obtained by normalizing the number of Ca 2+ pulses that could be taken up in a given condition by that taken up under DMSO control conditions, as previously described 60 . Student's t-test was used to determine statistical significance between predefined comparisons (DMSO versus CsA, inactive versus active enantiomers of B52-Cy5, and inactive versus active enantiomers of B53-Cy5).
Fluorescence microscopy on Hela cells. Cells were seeded into 96-well, black, clear-bottom, TC-treated plates, in DMEM supplemented with 10% FBS 24 h prior at a density where wells were ~80% confluent at the time of the experiment. Medium was removed and cells were then stained with 50 µL mixture containing Cy5-labeled compound (6 µM) in Kreb's ringer solution HEPES buffered pH 7.2 (KRB), for 1 h at 37 °C. Then, 50 µL Hoechst 33342 (8 µM) and Mitotracker Green FM (0.1 µM) in KRB was added directly to wells containing cells and incubated for 15 min at 37 °C. Medium was removed and cells were washed twice with 100 µL KRB. Cells were then kept adherent to the wells in 100 µL KRB for imaging.
Fluorescence microscopy was conducted on an Opera Phenix Plus High-Content Screening System, using Alexa 488, Alexa 647, and Hoescht 33342 channels. Images were analyzed using Harmony v.4.9.

In vitro esterase activity on ester pro-drug Cy5 derivatives of CypD inhibitors.
In PCR tubes, ester compound (6 µM) and either buffer only, CES1 (0.25 µM), or CES2 (0.25 µM) was diluted in 100 mM Tris-HCl buffer, pH 7.4, with a final DMSO concentration of 1%. Samples were maintained on a PCR block at 37 °C for 8 h. After incubation, samples were diluted with 20 µL acetonitrile and analyzed by LC-MS at Harvard's Center for Mass Spectroscopy. Di-ester, mono-ester, and di-acid abundances were quantified by total ion count of the primary isotope and the three compounds were summed and each one expressed as a fraction of the total sum. Cellular esterase activity on ester pro-drug Cy5 derivatives of CypD inhibitors. Cells were seeded in 96-well, clear, flat-bottom, TC-treated plates in DMEM supplemented with 10% FBS for HEK293T, HeLa, and MEFs, F-12K supplemented with 10% FBS for A549, or MEM supplemented with 10% FBS for HepG2. Cells were seeded 24 h prior at a density where wells were ~70% confluent at the time of the experiment. Medium was removed and cells were incubated with 100 µL mixture containing each Cy5-conjugated compound (6 µM) in their respective medium at 1% DMSO for 48 h (36 h for HepG2) at 37 °C at 5% CO 2 . Supernatant was removed, and cells were washed twice with 100 µL PBS pH 7.4 (Gibco). Cells were then lysed for 15 min with 30 µL 10 mM Tris-HCl pH 8.0 buffer containing 0.05% SDS. Lysates were diluted with 100 µL acetonitrile and filtered. Lysates were analyzed by LC-MS at Harvard's Center for Mass Spectroscopy. Di-ester, mono-ester, and di-acid abundances were quantified by total ion count of the primary isotope, and the three compounds were summed and each one expressed as a fraction of the total sum.  Table 1). Parts per million values were reported from calculated whole organism (integrated) abundance. Fig. 8 | Quantification of mitochondrial localization in HeLa cells by fluorescence microscopy of Cy5-conjugated compounds. HeLa cells were treated with Cy5-conjugated compounds and analyzed for a, total identifiable Cy5 spots; b, Cy5 spots per well; c, mean fluorescence intensity of identified Cy5 spots; d, sum of fluorescence intensity of identified Cy5 spots; e, mean Cy5 fluorescence intensity per cell; f, sum of Cy5 fluorescence in all measured cells; g, percent of Cy5 spots that overlap >70% with Mitotracker Green co-stain; and h, fluorescence intensity of Cy5 spots that overlap >70% with Mitotracker Green. i, values of data shown in a-h. Values and error bars reflect mean ± s.d. of three technical replicates.