In situ click chemistry generation of cyclooxygenase-2 inhibitors

Atul Bhardwaj, Jatinder Kaur, Melinda Wuest, Frank Wuest•View original

OverviewBalancedalloy voice

Imagine a lock that doesn't just accept a key, but builds it. That's the idea here. Cyclooxygenase-2, or COX-2, doesn't passively wait for a drug. It helps assemble one inside its own active site. And that's a big deal because this family of enzymes has been a therapeutic headache for decades. Cyclooxygenase-1, or COX-1, is the steady, housekeeping isozyme, making prostanoids that protect your gut lining and keep your cardiovascular system balanced. COX-2, by contrast, is the emergency responder, induced during inflammation and implicated in cancer and neurodegenerative disease. Block them both, as traditional nonsteroidal anti-inflammatory drugs do, and you get relief, plus ulcers and gastrointestinal trouble. Try to be selective for COX-2, as drug makers did with agents like rofecoxib, and you can shift the balance between prostacyclin and thromboxane toward clotting. That's why some COX-2-selective drugs were withdrawn. The isoforms look so similar, with over sixty percent sequence homology and almost superimposable structures, that designing a clean COX-2 hit without collateral damage has been hard. There is, however, a structural chink in that armor. COX-2's active site is bigger — about a quarter larger — and it carries a secondary binding pocket that COX-1 lacks. That extra space arises from a subtle swap: an isoleucine in COX-1 becomes a valine in COX-2 at two key positions, opening the room just enough for bulky motifs to dock. The secondary pocket is defined by residues like histidine 90, arginine 513, and valine 523. Alongside it runs a hydrophobic channel shaped by tryptophan 387, tyrosine 385, phenylalanine 518, phenylalanine 381, and leucine 352. Medicinal chemists have aimed at that pocket for years, threading sulfonyl groups into it to pick up selectivity. But Bhardwaj, Kaur, Wuest, and Knaus pushed the logic one step further: let the pocket not just recognize the right scaffold, but let it assemble it. They turned to kinetic target-guided synthesis, a flavor of in situ click chemistry, and made COX-2 the reaction vessel. Picture two reactive fragments, an azide and an alkyne, each small enough to enter the active site. If they sit in the right orientation, they "click" — they cycloadd to form a triazole ring right there in the protein. The trick is to bias one fragment so it finds the secondary pocket and holds tight. Here, that anchor was a five-azido-pyrazole bearing a methanesulfonyl group, which is SO2CH3, the same sulfonyl handle that celecoxib uses to engage COX-2. The team built a small set of these azides, varied in size and substitution, and paired them with a panel of aryl alkynes. Then they let COX-2 sift the pool. The libraries are worth a quick tour because the details matter. Four azides carried the day: compounds labeled five, fourteen, twenty-seven, and thirty-one. Two of them, five and fourteen, looked like cousins; both had the sulfonyl pharmacophore, but fourteen was slightly trimmer, with a molecular volume of about three hundred twenty-six cubic angstroms compared with three hundred thirty-three for five. One, twenty-seven, was bulkier. And thirty-one intentionally broke the rules by dropping the sulfonyl group entirely. Against that, they brought eleven alkynes with different para substituents. Under standardized, copper-catalyzed azide-alkyne cycloaddition conditions, they incubated tiny aliquots of each azide and alkyne with recombinant human COX-2 in buffered solution at room temperature, then followed the chemistry by mass spectrometry over a full day. Most pairs didn't react. That's the point. Library one, which included the larger azide five with various alkynes, yielded nothing. But when fourteen took the field, two products popped up as if the enzyme had tapped them on the shoulder. Starting at around six hours and continuing to build into the twelve to fifteen-hour window, the instrument began to see two triazoles we'll call eighteen and twenty-one. Both were clean one, four-regioisomers — the topologically preferred outcome of this click chemistry — and both matched the expected mass and retention profiles. Swap in the azide that lacked the sulfonyl group, the one labeled thirty-one, and the reaction died, even after thirty-six hours. Make the azide too big and it didn't work either. Mix everything together — four azides and eleven alkynes — and COX-2 still chose the same two products, just at about half the signal you see in the simpler pairings because of competition. Run the entire experiment with COX-1 instead, or without any protein, or spike in bovine serum albumin, or flood the system with celecoxib at one hundred micromolar so the pocket is preoccupied, and nothing forms. That's a chemist's way of saying the enzyme isn't a bystander. It's the matchmaker. Docking studies help explain how the matchmaking works. COX-2's binding cavity has room — roughly three hundred ninety-four cubic angstroms compared with about three hundred sixteen for COX-1 — which makes it physically plausible to host two reactive fragments at once. When the team computationally placed the azide building block fourteen into COX-2, its sulfonyl oxygens reached out like grappling hooks into the secondary pocket, forming hydrogen bonds with arginine 513 and histidine 90. The modeled distances were tight, on the order of two angstroms. That anchoring let the azide group sit near the right alkynes. With a para-fluoro alkyne, the reactive carbon nestled within about one point eight angstroms of the azide nitrogen; with a para-amine alkyne, it was roughly two point nine angstroms. Other alkynes drifted elsewhere in the channel, farther from the azide and less likely to click. In COX-1, by contrast, the models couldn't get the two partners into a workable orientation at all. The smaller site and missing pocket broke the geometry the reaction needed. Once the products formed, they fit like a glove. When compounds eighteen and twenty-one were docked back into COX-2, both made stable, low-energy complexes — in the ballpark of minus sixteen kilocalories per mole — with the sulfonyl group still holding court in the secondary pocket. For eighteen, the sulfonyl oxygens again reached arginine 513 and histidine 90, and the pyrazole nitrogens caught a hydrogen bond to alanine 527. For twenty-one, the sulfonyl leaned into tyrosine 355 at a little over two angstroms. If you know the old COX-2 inhibitor SC-558, the picture will feel familiar: a four-methanesulfonyl-phenyl motif tucked into the accessory pocket and the other aryl rings skating down the hydrophobic channel. Same dance, new partner. But now the protein helped choreograph it. That would all be clever if the products were weak. They weren't. Across the whole set of triazoles they generated, COX-2 inhibition ranged from a few hundredths of a micromolar up to a couple dozen micromolars, while COX-1 inhibition sat an order of magnitude or more worse, often above one hundred micromolar. The headliners were the in situ hits. Compound eighteen checked in with a COX-2 half-maximal inhibitory concentration of zero point zero nine micromolar and no measurable effect on COX-1 up to the one hundred micromolar ceiling, a selectivity index of roughly eleven hundred. Compound twenty-one was even cleaner at zero point zero five micromolar on COX-2 and again beyond the one hundred micromolar mark for COX-1, pushing the selectivity index to about two thousand. For context, celecoxib — the poster child for COX-2 targeting — landed at about zero point zero seven micromolar on COX-2 with a selectivity index around fourteen hundred. Same ballpark potency, but the enzyme-assembled molecules were even more isoform-selective. Cellular data told the same story without the pipettes. In COX-2-expressing HCA-7 cells, which make prostaglandins through the inducible pathway, the new compounds kept their edge. The cell-based COX-2 half-max values sat at roughly zero point zero six micromolar for eighteen and zero point zero eight for twenty-one, while celecoxib clocked in around zero point zero nine. That's not just proof-of-concept chemistry. That's pharmacology sneaking into the room. The thermodynamics add texture to the picture. Isothermal titration calorimetry — that's the technique where you drip one binding partner into another and watch the heat of interaction — measured how the key azide fragment, compound fourteen, engages COX-2. The stoichiometry was essentially one-to-one, about one point zero two. The binding free energy came out to around minus thirty-six kilojoules per mole, with an enthalpic contribution of about minus thirty-four point five and a small favorable entropy term near one point seven. In plain language, the interaction is driven by the making of specific, exothermic contacts — hydrogen bonds, van der Waals snugness, maybe a touch of ion-dipole charm — and it doesn't cost the system much in lost motion to do it. That's exactly what you'd expect from a fragment that slides into a defined pocket and latches on with a sulfonyl moiety. But the payoff isn't on a microcalorimeter. It's in an inflamed paw. In a standard rat model of acute inflammation — you inject carrageenan to make the hind paw swell, then see what reduces the edema — both in situ hits were remarkably effective when given orally. Across four ascending doses, the median effective dose at three hours was about zero point four four milligrams per kilogram for eighteen and a striking zero point one two for twenty-one. Stretch the clock to five hours and those numbers eased to about zero point nine nine and zero point three four. For orientation, celecoxib in the same setup needed roughly ten point eight milligrams per kilogram at the three-hour mark. That's more than a twenty-five fold difference for the best of the new compounds. The sample sizes were small — four or five rats per condition — but the gap was large enough to feel confident about the direction of effect. Underneath the pharmacology, you can see a theme running from structure to function. COX-2's slightly roomier active site, and that extra pocket with residues like histidine 90 and arginine 513, let two small fragments sit close enough, long enough, and in the right orientation to click into a potent inhibitor. COX-1 can't host the same geometry, so it doesn't catalyze its own demise and doesn't get inhibited as well by the resulting products. The enzyme quite literally chooses the chemistry, and in doing so, dictates the architecture of the inhibitor that will later block it. Bhardwaj, Kaur, Wuest, and Knaus showed that in the cleanest possible way: the only in situ products in COX-2 were eighteen and twenty-one, the same two no matter whether you mixed one azide-alkyne pair or the full soup, and nothing formed under any of the controls. The docking lines up with that story, residue by residue, down to hydrogen bond distances. And the activity data — test tube, cell, whole animal — make the structural bet feel worthwhile. A couple of nuances are worth pausing on. First, not all sulfonyl-bearing azides worked. The largest one, twenty-seven, could make a contact to arginine 120 in the models, but it just didn't line up to foster the reaction, and empirically it failed to generate products. Second, size and placement were everything. The smaller sulfonyl azide, fourteen, placed the reactive azide right where the best alkynes — a para-fluoro and a para-amine — could approach within two or three angstroms. Shift a substituent, and those distances stretched, and the chemistry stalled. That kind of steric and electrostatic choreography is exactly what you'd hope an enzyme pocket would impose. Here it became a filter. What does this mean beyond COX-2? It suggests a very practical route for targets with a clear anchoring pharmacophore and a pocket big enough to seat two fragments. You let the protein do the structure-activity relationship, or SAR, by templating only the pairs that fit best, then you harvest the products and test them. It's efficient, it's selective by construction, and when the biology cooperates, as it did here, you end up with leads that are not only potent on the bench but carry weight in vivo. There are limits. This approach leans on a secondary pocket that COX-1 doesn't share, so it may generalize best to targets with that kind of extra real estate. And the chemistry still needed a copper catalyst and a well-behaved protein. But the core lesson holds. Sometimes the shortest path between a binding pocket and a good inhibitor is a straight line — and sometimes it's a circle you draw right inside the protein. In this case, COX-2 traced that circle twice, and the resulting rings, eighteen and twenty-one, beat the standard bearer on selectivity and matched or exceeded it where it counts. That's not just clever chemistry. That's a new way to listen to what the target is trying to tell you.

Imagine a lock that doesn't just accept a key, but builds it. That's the idea here. Cyclooxygenase-2, or COX-2, doesn't passively wait for a drug.

It helps assemble one inside its own active site. And that's a big deal because this family of enzymes has been a therapeutic headache for decades. Cyclooxygenase-1, or COX-1, is the steady, housekeeping isozyme, making prostanoids that protect your gut lining and keep your cardiovascular system balanced.

COX-2, by contrast, is the emergency responder, induced during inflammation and implicated in cancer and neurodegenerative disease. Block them both, as traditional nonsteroidal anti-inflammatory drugs do, and you get relief, plus ulcers and gastrointestinal trouble. Try to be selective for COX-2, as drug makers did with agents like rofecoxib, and you can shift the balance between prostacyclin and thromboxane toward clotting.

That's why some COX-2-selective drugs were withdrawn. The isoforms look so similar, with over sixty percent sequence homology and almost superimposable structures, that designing a clean COX-2 hit without collateral damage has been hard.

There is, however, a structural chink in that armor. COX-2's active site is bigger — about a quarter larger — and it carries a secondary binding pocket that COX-1 lacks. That extra space arises from a subtle swap: an isoleucine in COX-1 becomes a valine in COX-2 at two key positions, opening the room just enough for bulky motifs to dock.

The secondary pocket is defined by residues like histidine 90, arginine 513, and valine 523. Alongside it runs a hydrophobic channel shaped by tryptophan 387, tyrosine 385, phenylalanine 518, phenylalanine 381, and leucine 352. Medicinal chemists have aimed at that pocket for years, threading sulfonyl groups into it to pick up selectivity.

But Bhardwaj, Kaur, Wuest, and Knaus pushed the logic one step further: let the pocket not just recognize the right scaffold, but let it assemble it.

They turned to kinetic target-guided synthesis, a flavor of in situ click chemistry, and made COX-2 the reaction vessel. Picture two reactive fragments, an azide and an alkyne, each small enough to enter the active site. If they sit in the right orientation, they "click" — they cycloadd to form a triazole ring right there in the protein.

The trick is to bias one fragment so it finds the secondary pocket and holds tight. Here, that anchor was a five-azido-pyrazole bearing a methanesulfonyl group, which is SO2CH3, the same sulfonyl handle that celecoxib uses to engage COX-2. The team built a small set of these azides, varied in size and substitution, and paired them with a panel of aryl alkynes. Then they let COX-2 sift the pool.

The libraries are worth a quick tour because the details matter. Four azides carried the day: compounds labeled five, fourteen, twenty-seven, and thirty-one. Two of them, five and fourteen, looked like cousins; both had the sulfonyl pharmacophore, but fourteen was slightly trimmer, with a molecular volume of about three hundred twenty-six cubic angstroms compared with three hundred thirty-three for five.

One, twenty-seven, was bulkier. And thirty-one intentionally broke the rules by dropping the sulfonyl group entirely. Against that, they brought eleven alkynes with different para substituents.

Under standardized, copper-catalyzed azide-alkyne cycloaddition conditions, they incubated tiny aliquots of each azide and alkyne with recombinant human COX-2 in buffered solution at room temperature, then followed the chemistry by mass spectrometry over a full day.

Most pairs didn't react. That's the point. Library one, which included the larger azide five with various alkynes, yielded nothing.

But when fourteen took the field, two products popped up as if the enzyme had tapped them on the shoulder. Starting at around six hours and continuing to build into the twelve to fifteen-hour window, the instrument began to see two triazoles we'll call eighteen and twenty-one. Both were clean one, four-regioisomers — the topologically preferred outcome of this click chemistry — and both matched the expected mass and retention profiles.

Swap in the azide that lacked the sulfonyl group, the one labeled thirty-one, and the reaction died, even after thirty-six hours. Make the azide too big and it didn't work either. Mix everything together — four azides and eleven alkynes — and COX-2 still chose the same two products, just at about half the signal you see in the simpler pairings because of competition.

Run the entire experiment with COX-1 instead, or without any protein, or spike in bovine serum albumin, or flood the system with celecoxib at one hundred micromolar so the pocket is preoccupied, and nothing forms. That's a chemist's way of saying the enzyme isn't a bystander. It's the matchmaker.

Docking studies help explain how the matchmaking works. COX-2's binding cavity has room — roughly three hundred ninety-four cubic angstroms compared with about three hundred sixteen for COX-1 — which makes it physically plausible to host two reactive fragments at once. When the team computationally placed the azide building block fourteen into COX-2, its sulfonyl oxygens reached out like grappling hooks into the secondary pocket, forming hydrogen bonds with arginine 513 and histidine 90.

The modeled distances were tight, on the order of two angstroms. That anchoring let the azide group sit near the right alkynes. With a para-fluoro alkyne, the reactive carbon nestled within about one point eight angstroms of the azide nitrogen; with a para-amine alkyne, it was roughly two point nine angstroms.

Other alkynes drifted elsewhere in the channel, farther from the azide and less likely to click. In COX-1, by contrast, the models couldn't get the two partners into a workable orientation at all. The smaller site and missing pocket broke the geometry the reaction needed.

Once the products formed, they fit like a glove. When compounds eighteen and twenty-one were docked back into COX-2, both made stable, low-energy complexes — in the ballpark of minus sixteen kilocalories per mole — with the sulfonyl group still holding court in the secondary pocket. For eighteen, the sulfonyl oxygens again reached arginine 513 and histidine 90, and the pyrazole nitrogens caught a hydrogen bond to alanine 527.

For twenty-one, the sulfonyl leaned into tyrosine 355 at a little over two angstroms. If you know the old COX-2 inhibitor SC-558, the picture will feel familiar: a four-methanesulfonyl-phenyl motif tucked into the accessory pocket and the other aryl rings skating down the hydrophobic channel. Same dance, new partner. But now the protein helped choreograph it.

That would all be clever if the products were weak. They weren't. Across the whole set of triazoles they generated, COX-2 inhibition ranged from a few hundredths of a micromolar up to a couple dozen micromolars, while COX-1 inhibition sat an order of magnitude or more worse, often above one hundred micromolar.

The headliners were the in situ hits. Compound eighteen checked in with a COX-2 half-maximal inhibitory concentration of zero point zero nine micromolar and no measurable effect on COX-1 up to the one hundred micromolar ceiling, a selectivity index of roughly eleven hundred. Compound twenty-one was even cleaner at zero point zero five micromolar on COX-2 and again beyond the one hundred micromolar mark for COX-1, pushing the selectivity index to about two thousand.

For context, celecoxib — the poster child for COX-2 targeting — landed at about zero point zero seven micromolar on COX-2 with a selectivity index around fourteen hundred. Same ballpark potency, but the enzyme-assembled molecules were even more isoform-selective.

Cellular data told the same story without the pipettes. In COX-2-expressing HCA-7 cells, which make prostaglandins through the inducible pathway, the new compounds kept their edge. The cell-based COX-2 half-max values sat at roughly zero point zero six micromolar for eighteen and zero point zero eight for twenty-one, while celecoxib clocked in around zero point zero nine.

That's not just proof-of-concept chemistry. That's pharmacology sneaking into the room.

The thermodynamics add texture to the picture. Isothermal titration calorimetry — that's the technique where you drip one binding partner into another and watch the heat of interaction — measured how the key azide fragment, compound fourteen, engages COX-2. The stoichiometry was essentially one-to-one, about one point zero two.

The binding free energy came out to around minus thirty-six kilojoules per mole, with an enthalpic contribution of about minus thirty-four point five and a small favorable entropy term near one point seven. In plain language, the interaction is driven by the making of specific, exothermic contacts — hydrogen bonds, van der Waals snugness, maybe a touch of ion-dipole charm — and it doesn't cost the system much in lost motion to do it. That's exactly what you'd expect from a fragment that slides into a defined pocket and latches on with a sulfonyl moiety.

But the payoff isn't on a microcalorimeter. It's in an inflamed paw. In a standard rat model of acute inflammation — you inject carrageenan to make the hind paw swell, then see what reduces the edema — both in situ hits were remarkably effective when given orally.

Across four ascending doses, the median effective dose at three hours was about zero point four four milligrams per kilogram for eighteen and a striking zero point one two for twenty-one. Stretch the clock to five hours and those numbers eased to about zero point nine nine and zero point three four. For orientation, celecoxib in the same setup needed roughly ten point eight milligrams per kilogram at the three-hour mark.

That's more than a twenty-five fold difference for the best of the new compounds. The sample sizes were small — four or five rats per condition — but the gap was large enough to feel confident about the direction of effect.

Underneath the pharmacology, you can see a theme running from structure to function. COX-2's slightly roomier active site, and that extra pocket with residues like histidine 90 and arginine 513, let two small fragments sit close enough, long enough, and in the right orientation to click into a potent inhibitor. COX-1 can't host the same geometry, so it doesn't catalyze its own demise and doesn't get inhibited as well by the resulting products.

The enzyme quite literally chooses the chemistry, and in doing so, dictates the architecture of the inhibitor that will later block it. Bhardwaj, Kaur, Wuest, and Knaus showed that in the cleanest possible way: the only in situ products in COX-2 were eighteen and twenty-one, the same two no matter whether you mixed one azide-alkyne pair or the full soup, and nothing formed under any of the controls. The docking lines up with that story, residue by residue, down to hydrogen bond distances.

And the activity data — test tube, cell, whole animal — make the structural bet feel worthwhile.

A couple of nuances are worth pausing on. First, not all sulfonyl-bearing azides worked. The largest one, twenty-seven, could make a contact to arginine 120 in the models, but it just didn't line up to foster the reaction, and empirically it failed to generate products.

Second, size and placement were everything. The smaller sulfonyl azide, fourteen, placed the reactive azide right where the best alkynes — a para-fluoro and a para-amine — could approach within two or three angstroms. Shift a substituent, and those distances stretched, and the chemistry stalled.

That kind of steric and electrostatic choreography is exactly what you'd hope an enzyme pocket would impose. Here it became a filter.

What does this mean beyond COX-2? It suggests a very practical route for targets with a clear anchoring pharmacophore and a pocket big enough to seat two fragments. You let the protein do the structure-activity relationship, or SAR, by templating only the pairs that fit best, then you harvest the products and test them.

It's efficient, it's selective by construction, and when the biology cooperates, as it did here, you end up with leads that are not only potent on the bench but carry weight in vivo.

There are limits. This approach leans on a secondary pocket that COX-1 doesn't share, so it may generalize best to targets with that kind of extra real estate. And the chemistry still needed a copper catalyst and a well-behaved protein.

But the core lesson holds. Sometimes the shortest path between a binding pocket and a good inhibitor is a straight line — and sometimes it's a circle you draw right inside the protein. In this case, COX-2 traced that circle twice, and the resulting rings, eighteen and twenty-one, beat the standard bearer on selectivity and matched or exceeded it where it counts.

That's not just clever chemistry. That's a new way to listen to what the target is trying to tell you.