Our studies, and those described by Kowalinski and coworkers in the accompanying article 
, provide the first molecular insights into the mechanism of inhibition of the essential influenza enzyme PA endonuclease, and we have confirmed that it represents an ideal target for drug discovery. Previous mutagenesis studies have shown a direct correlation between PAN
endonuclease activities and RdRp transcription activities, suggesting that the isolated PAN
domain contains the same structure in the context of the intact RdRp 
. Our biochemical studies show that inhibitors of RdRp transcription also inhibit PAN
endonuclease activity, and this validates the use of the isolated PAN
endonuclease domain for drug development.
Our structural studies provide the framework to develop novel inhibitors of the influenza virus PA endonuclease. However, two-metal active sites are ubiquitous in enzymes that process nucleic acids, and it may be challenging to develop drugs that specifically target PAN endonuclease. We therefore analyzed the PAN active site for conserved and unique features for drug discovery by aligning ~13,000 PA amino acid sequences to identify the consensus sequence for PAN of influenza types A, B, and C (). Thirty residues are highly conserved and 17 are more than 99.9% identical. Unsurprisingly, most are in the active site pocket and include the metal-binding residues His41, Glu80, Asp108, and Glu119 and the catalytic residue Lys134 (). The central scaffolds of our characterized inhibitors interact with these residues and are likely to be resistant to mutation but are unlikely to be useful for specificity.
Conserved residues and ordered water molecules in the PAN active site cleft.
Our studies have shown that interactions with residues further away from the two-metal center substantially increase potency. The same conclusion has been drawn by Kowalinski and coworkers who specifically identified four pockets that can be exploited for inhibitor optimization 
. maps out how compounds 1
engage these pockets, and it can be seen that none of the compounds bind pockets 1 and 2, which only appear to become available upon side-chain rotation and inhibitor binding 
. However, our structures reveal two additional pockets 5 and 6. Compounds 2
occupy pocket 3 and interact with Tyr24, which is a highly conserved aromatic residue. The biological role of Tyr24 is revealed in the studies of Kowalinski and coworkers which show that it forms a crucial stacking interaction with the base of the mononucleotide 
. The new pocket 5 is revealed by the binding of the benzylpiperidine group of compound 3
; it comprises conserved residues Arg84, Trp88, Phe105, and Leu106, and is an excellent target for further exploration (, ). The same is true for the new pocket 6 that engages the acetamide group of compound 5
and comprises highly conserved residues Thr123, Tyr130, Lys134 and Lys137 (, ). Mutation of Arg84, Tyr130, or Lys137 to Ala reduces but does not eliminate endonuclease activity, suggesting that inhibitor resistance could develop, although possibly at a cost to virus fitness 
. Similarly, the interactions between molecule B of compound 5
and pocket 4 residues Lys34 and Arg124 are unlikely to be useful for drug development because these residues are not well conserved. However, π-stacking interactions have been shown to be very productive in terms of increasing potency 
, and Tyr24, His41, F105, Tyr130, and F150 offer potential opportunities. These data reveal the potential for the use of growing and linking strategies to design potent inhibitors.
The entropic contribution to binding can be substantial when ordered water molecules are displaced 
, and the PAN
active site offers opportunities in this regard. PAN
contains a large, deep active site (over 3000 Å3
) with several ordered water molecules, 17 of which are found in at least three of the four PAN
molecules in the asymmetric unit (). A large network of water molecules near Val122 becomes displaced by molecule B of compound 5
, and a network of four water molecules between Mn2 and Arg84 is displaced by the benzylpiperidine group of compound 3
, and both can be targeted for inhibitor optimization. Ordered water molecules can also be mimicked by oxygen atoms introduced during inhibitor optimization (see for example 
). Our studies provide an example of this. One water molecule (H2
) that interacts with Mn1, Glu119, and Lys134 becomes displaced by an oxygen atom from compounds 1
also forms a hydrogen bond with water molecule H2
, which in turn forms hydrogen bonds with Val122 (backbone amide), Tyr130, and another water molecule. Modification of inhibitors that displace H2
but preserve its hydrogen bonds should significantly improve inhibitor binding via gains in both entropy and enthalpy.
Another important consideration in the design of optimal inhibitors is the location and coordination sphere of each Mn2+
ion in the PAN
active site. Detailed structural analyses on the Bacillus halodurans
RNase H revealed that the distance between the metal ions changes at different stages of phosphodiester hydrolysis 
. Consistent with this is the observation that the metals are approximately 2.9 Å apart in PANΔLoop
–Apo and move to 3.8–4.0 Å apart when an inhibitor is bound. This mobility seems to occur in Mn2 because Mn1 is in a similar location in both the unbound and inhibitor-bound structures. Our data suggest that the inhibitor-bound form of PAN
represents the enzyme-substrate complex stage in which the metals are separated by about 4.0 Å 
. Thus, computational modeling or docking of inhibitors may best be suited with the inhibitor-bound form of PAN
Furthermore, metal coordination appears to play an important role in compound binding. Specifically, the compound oxygen atoms that coordinate Mn1 in all the complexes described here and in the accompanying article 
are separated by two atoms (), and this allows them to ideally contribute to the octahedral geometry completed by the Mn1-coordinating oxygen atoms from H41, D108, E119, and I120.
Finally, our studies support the potential for developing antiviral inhibitors that target the endonuclease activity of other negative strand and cap-snatching segmented RNA viruses, specifically the Orthomyxoviridae
, and Arenaviridae
families. Recent crystal structures of the endonuclease domains from La Crosse orthobunyavirus L protein and lymphocytic choriomeningitis virus L protein reveal clear structural homology to the influenza A virus PAN
endonuclease domain with dependence on manganese ions for activity 
). However, low sequence homology and structural variation between virus family endonucleases suggest opportunities for developing virus family-specific inhibitors.