Bergman Cyclization: The Molecule That Grows Teeth

Folding an Enediyne Shut

The parent substrate is (Z)-hex-3-ene-1,5-diyne, numbered C1≡C2–C3=C4–C5≡C6. The cis double bond in the middle is a hinge: it holds both alkyne arms on the same side, pointing their far ends — C1 and C6 — at each other across empty space. In the floppy acyclic parent that gap is about 4.1 Å, far too wide to react below its pyrolysis temperature; a strained ten-membered ring squeezes it to about 3.16 Å.

Heat does the rest. Thermal motion bends the alkynes away from linearity, which costs energy but brings C1 and C6 together. At the transition state their separation is roughly 2.0 Å and the in-plane p orbitals already overlap. Past that point the molecule commits: the C1···C6 gap closes from 3.16 Å to 1.39 Å, essentially a benzene bond (benzene C–C = 1.397 Å), and the six carbons flatten into an aromatic ring. The bond-forming passage over the transition state takes about 100 femtoseconds — one vibrational period. Everything slow about this reaction is the waiting, not the doing.

Count the electrons. The new ring needs an aromatic sextet plus the new σ bond, and two are left over. They end up in two in-plane sp2-like orbitals on C2 and C5 — the internal alkyne carbons that flanked the original double bond, now sitting para to each other. The product is 1,4-didehydrobenzene, or p-benzyne. Those two radical lobes are about 2.7 Å apart across the ring, far too distant to overlap directly, so they communicate through the σ framework instead. That through-bond coupling puts the singlet below the triplet by only 3.8 kcal/mol (Wenthold and Squires, 1998) — a formally spin-paired diradical that still behaves like two hungry aryl radicals, because the coupling is far too feeble to make a bond.

The Energy Ledger: 28 Up, 8.5 Uphill, 20 Down Per Hydrogen

Robert G. Bergman reported the reaction in 1972 from gas-phase pyrolysis near 200 °C; later gas-kinetic work (Roth, Hopf and Horn, 1994) pinned the numbers at ΔH‡ = 28.2 kcal/mol (118 kJ/mol), with p-benzyne sitting about 8.5 kcal/mol uphill of the enediyne.

Worked example. Put 28 kcal/mol into the Eyring equation, k = (k_BT/h)·exp(−ΔG‡/RT), and assume ΔS‡ ≈ 0 for a first pass. At 473 K (200 °C), k_BT/h = 9.9 × 10¹² s⁻¹ and RT = 0.940 kcal/mol, so exp(−28/0.940) = 1.2 × 10⁻¹³ and k ≈ 1.1 s⁻¹ — a half-life of about 0.6 seconds. At 298 K, RT = 0.592 kcal/mol, exp(−28/0.592) = 2.9 × 10⁻²¹, and k ≈ 1.8 × 10⁻⁸ s⁻¹ — a half-life of roughly a year. A 175-degree change buys a factor of about 6 × 10⁷ in rate. Setting ΔS‡ = 0 flatters both figures, though: the real transition state is pre-bent and more ordered, with ΔS‡ near −10 cal/(mol·K), a temperature-independent brake of exp(ΔS‡/R) ≈ 10⁻² that makes the true half-life a minute or two at 200 °C and a couple of centuries at 25 °C. Either way the parent enediyne is bench-stable at 25 °C and a runaway at 200 °C.

Aromatisation is usually assumed to pay for everything, but the product sits 8.5 kcal/mol above the starting material: breaking two π bonds to make one σ bond and two radicals costs more than the sextet returns. Retro-Bergman ring opening therefore has a barrier of only 28.2 − 8.5 ≈ 19.7 kcal/mol, giving the diradical a lifetime on the order of 100 microseconds at 200 °C before it unzips again.

What makes the reaction irreversible is the next step. An aryl C–H bond costs about 113 kcal/mol to break; a deoxyribose C–H bond costs about 92 kcal/mol, so each hydrogen the diradical steals is roughly 20 kcal/mol downhill. Two abstractions release about 38 kcal/mol in all, dragging the sequence from +8.5 to roughly −30 kcal/mol overall. The two are not identical, and the weaker one comes first: capping one lobe of p-benzyne to give a phenyl radical forms a C–H bond worth about 109 kcal/mol (17 downhill), and only capping that phenyl radical to give benzene is worth the full 113 (21 downhill). That ~4 kcal/mol difference is the same diradical stabilisation that surfaces as the 3.8 kcal/mol singlet–triplet gap.

Nicolaou's Distance Criterion — and What It Really Measures

If the barrier is the cost of bending the alkynes together, a ring that has already done part of the bending should react colder. In 1988 K. C. Nicolaou and co-workers made a series of cyclic enediynes and found where the switch flips: ten-membered-ring enediynes whose reacting carbons sit 3.2–3.31 Å apart cyclize at 25 °C; at 3.35 Å and beyond they are shelf-stable. Four hundredths of an ångström separates a reagent from a reaction.

Worked example. What barrier does a molecule need to fire at body temperature? For a half-life of one minute at 310 K, k = 0.693/60 = 1.2 × 10⁻² s⁻¹. With k_BT/h = 6.5 × 10¹² s⁻¹, exp(−ΔG‡/RT) = 1.8 × 10⁻¹⁵, so ΔG‡ = 34.0 × RT = 34.0 × 0.616 ≈ 21 kcal/mol. Geometry has to find about 7 kcal/mol relative to the parent's 28 — a factor of exp(7/0.616) ≈ 10⁵ in rate. That is the whole budget the ring strain must supply.

The criterion is a good predictor inside the ten-membered series and a poor one outside it. Schreiner and others showed in the late 1990s that what really sets the barrier is the difference in strain energy between the ground state and the transition state, not the ground-state gap. A ring already strained at the transition-state geometry reacts fast even at a nominally safe distance; a floppy acyclic enediyne can sit at 3.2 Å in one conformer and stay inert because reaching the transition state costs it entropy. Treat 3.2–3.31 Å as a design heuristic for one scaffold class, never as a law.

Calicheamicin: A Safety Catch, a Trigger and a Staggered Cut

Calicheamicin γ1-I was isolated at Lederle Laboratories from Micromonospora echinospora, cultured from chalky caliche soil collected in Texas. Its enediyne is locked into a bicyclic cage at a ground-state gap of 3.35 Å — just outside the reactive window. It is a loaded spring with the safety on.

The catch is a methyl trisulfide. Inside a cell, glutathione (or another thiol) attacks the central sulfur and cleaves the trisulfide, unmasking a thiolate on the drug itself. That thiolate then adds intramolecularly to a bridgehead enone in a Michael fashion. The consequence is purely geometric: the bridgehead carbon rehybridises sp2 → sp3, its bond angles pull from 120° toward 109.5°, the cage puckers, and the enediyne gap contracts from 3.35 Å to about 3.16 Å. That is past the tight edge of Nicolaou's reactive range, not merely inside it, and the Bergman cyclization runs at 37 °C, in seconds.

Meanwhile the drug's aryltetrasaccharide arm — four unusual sugars, an N–O glycosidic linkage, an iodinated aromatic ring — has been holding the assembly in the DNA minor groove, selectively at 5′-TCCT-3′. The diradical is born with both lobes already pointed at sugar C–H bonds, and abstracts H• from C5′ of one strand and from C4′ three bases away on the complementary strand. Oxygen then converts those carbon radicals into strand breaks: the C4′ radical gives 3′-phosphoglycolate ends and a base propenal, the C5′ radical a 5′-aldehyde.

The result is a staggered double-strand break — not two independent nicks, which a cell repairs routinely, but a two-strand cut with a three-base overhang that end-joining machinery handles badly. Picomolar concentrations kill; calicheamicin is roughly 1000× more potent than doxorubicin. That potency is oxygen-dependent, which is why hypoxic tumour cores resist it.

How Enediynes Are Made, Measured and Specified

Making them. The standard route to the enediyne core is a double Sonogashira coupling of cis-1,2-dichloroethylene with two terminal alkynes under palladium/copper catalysis — the cis alkene is bought in, not built, so the critical geometry is never in doubt. The hard step is the macrocyclisation that installs the strain, which is the same step that makes the product want to destroy itself.

Measuring the barrier. Kinetics are run in neat 1,4-cyclohexadiene, whose allylic C–H bond (~77 kcal/mol) is a fast, clean hydrogen source; the product is benzene, followed by ¹H NMR or UV-Vis. Trap concentration matters: because retro-Bergman ring opening has a barrier of only ~19.7 kcal/mol, a thin trap lets the diradical revert and the observed rate turns donor-dependent — a classic way to publish a wrong rate constant. Differential scanning calorimetry gives the complementary number: exotherm onset temperature is the practical way to rank a series of solid enediynes.

Proving the mechanism. Bergman's 1972 evidence was a labelling experiment: deuteriums on the terminal alkyne carbons scrambled into the olefinic positions on heating, which only a symmetric intermediate allows. Trapping in carbon tetrachloride gave 1,4-dichlorobenzene, fixing the radicals as para. The 3.8 kcal/mol gap came from negative-ion photoelectron spectroscopy of the p-benzyne radical anion.

Specifying a drug. DNA damage is assayed on supercoiled plasmid: Form I relaxing to Form II (nicked) counts single-strand breaks, and Form III (linear) counts double-strand breaks — the ratio is the enediyne's signature. Cleavage sites are mapped on sequencing gels with 5′-³²P-labelled oligonucleotides. Calicheamicin conjugates are then specified by drug-to-antibody ratio and handled under nanogram-per-cubic-metre occupational exposure limits.

History, and the Drug That Was Approved Twice

Bergman published the cyclization in 1972 as a curiosity of gas-phase hydrocarbon chemistry with no obvious application. Fifteen years later, groups working independently isolated bacterial natural products — calicheamicin (1987), esperamicin (1987), dynemicin A (1989) — whose structures contained exactly that motif, caged and triggered. Neocarzinostatin, known since 1965 as a chromoprotein antitumour agent, turned out to hide a related nine-membered enediyne. A physical-organic footnote had been a weapon for a billion years. The synthetic race that followed produced total syntheses of calicheamicin γ1-I by Nicolaou in 1992 and by Danishefsky in 1994 — landmark work precisely because the targets were built to self-destruct.

Calicheamicin is far too toxic to give free, so it is delivered as an antibody–drug conjugate. Mylotarg (gemtuzumab ozogamicin, anti-CD33, for acute myeloid leukaemia) received accelerated FDA approval in 2000, the first ADC ever licensed. It was voluntarily withdrawn in 2010 after the confirmatory SWOG S0106 trial showed no survival benefit and more fatal induction toxicity. Re-analysis blamed the dose and schedule rather than the chemistry, and a lower, fractionated regimen was re-approved in 2017. Besponsa (inotuzumab ozogamicin, anti-CD22, for B-cell acute lymphoblastic leukaemia) followed the same year. Both carry boxed warnings for hepatic veno-occlusive disease — the signature of a warhead that occasionally goes off in the liver instead of the tumour.

Failure Modes and the Look-Alikes

It is not downhill aromatization. The commonest misconception is that forming a benzene ring drives the reaction. It does not: p-benzyne is 8.5 kcal/mol uphill, and without a hydrogen donor the diradical mostly reverts. Aromatisation only makes the barrier climbable; hydrogen abstraction makes the outcome permanent.

It is a singlet, not a triplet. The singlet ground state lies 3.8 kcal/mol below the triplet, so p-benzyne is EPR-silent under ordinary conditions and hunting for an EPR signal is not a valid test for it.

It is not o-benzyne. ortho-Benzyne (1,2-didehydrobenzene) has a genuine, if weak, in-plane π bond and a singlet–triplet gap of 37.5 kcal/mol — an order of magnitude larger. It behaves as an electrophilic strained alkyne and a superb Diels–Alder dienophile, and does not abstract hydrogen from sugars; meta-benzyne sits between at 21.0 kcal/mol.

It is not Myers–Saito. Neocarzinostatin's activated chromophore is an enyne–cumulene, not an enediyne, and it cycloaromatises by the C2–C7 Myers–Saito route to α,3-didehydrotoluene — one radical benzylic and π, the other aryl and σ, a chemically different species from p-benzyne with a more single-strand-dominant damage pattern. Bulky substrates divert instead to the Schmittel (C2–C6) fulvene biradical.

How it fails in practice. Circulating thiols trigger the warhead off-target, linker hydrolysis releases free calicheamicin into the bloodstream, hypoxia starves the oxygen-dependent scission step, and an enediyne designed a tenth of an ångström too tight cyclizes in the flask and is never isolated.

Bergman cyclization against the cycloaromatizations and arynes it is most often confused with

Reaction	Substrate	New bond and product	Singlet–triplet gap
Bergman (C1–C6)	(Z)-enediyne	C1–C6 σ bond → p-benzyne (1,4-didehydrobenzene)	3.8 kcal/mol, singlet ground state
Myers–Saito (C2–C7)	Enyne–allene (cumulene)	C2–C7 bond → α,3-didehydrotoluene, a σ,π-biradical	~3 kcal/mol, singlet
Schmittel (C2–C6)	Enyne–allene with bulky alkyne terminus	C2–C6 bond → fulvene σ,π-biradical	Small, singlet
C1–C5 (Schreiner–Pascal) closure	(Z)-enediyne	C1–C5 bond → fulvene diradical, a minor competing channel	Small, singlet
o-Benzyne (aryne) generation	1,2-disubstituted arene, not an enediyne	Loss of two substituents → strained in-plane π bond	37.5 kcal/mol, singlet
m-Benzyne	1,3-didehydrobenzene precursors	1,3-didehydrobenzene, ring pinched by strong through-space coupling	21.0 kcal/mol, singlet