Predictable hydrogen-bond supramolecular synthons for rational API co-crystal engineering — from CSD mining to CMC-ready candidate
Co-crystal design begins not with crystallization, but with molecular self-recognition theory. Before any experiment is run, the API structure is dissected into its hydrogen-bonding functional groups, and Etter's rules are used to predict, on paper, which supramolecular synthon is most likely to form in the solid state — turning co-crystal screening from trial-and-error into a rational, hypothesis-driven exercise.
Margaret Etter's empirical rules (Acc. Chem. Res. 1990) remain the conceptual foundation of co-crystal design:
1. All good proton donors and acceptors are used in hydrogen bonding. A molecule crystallizing from the melt or solution will saturate every strong donor (O–H, N–H) and strong acceptor (C=O, pyridine-N, P=O) it possesses, rather than leaving them unsatisfied.
2. Six-membered-ring intramolecular hydrogen bonds form in preference to intermolecular hydrogen bonds. Intramolecular motifs are checked first — if an API already satisfies a donor/acceptor pair intramolecularly, it is unavailable for co-crystal formation.
3. The best proton donor and the best proton acceptor remaining after intramolecular hydrogen-bond formation will preferentially form intermolecular hydrogen bonds to one another. This is the rule that makes co-crystal design tractable: rank the API's donors and acceptors by strength, then look for a coformer bearing the complementary partner.
Donor strength (descending): –COOH > –CONH2 (amide N–H) > –OH (phenol) > –OH (aliphatic) > aromatic C–H (weak). Acceptor strength (descending): carboxylate O⁻ > pyridine-type N > amide C=O > ketone/ester C=O > ether O > aromatic π-system (weak).
A carboxylic acid API with no available intramolecular partner is the single most reliable co-crystal building block in medicinal chemistry — it almost always forms a heterosynthon with a pyridine-N or primary-amide coformer, because the acid is simultaneously the strongest donor and (as carboxylate) a strong acceptor.
Etter also introduced graph-set notation, a formal descriptor used throughout co-crystal literature to unambiguously classify a hydrogen-bonded motif:
Gₐᵈ(n) — where G is the graph-set descriptor (C=chain, R=ring, D=discrete/dimer, S=intramolecular), a=number of acceptors, d=number of donors, n=total atoms in the repeat unit.
Common examples relevant to co-crystals: • R2,2(8) — the classic carboxylic acid homodimer (two COOH groups, two O–H···O=C bonds, 8-membered ring) — this is what must be "broken" by a coformer to force heterosynthon formation. • R2,2(8) heterosynthon — carboxylic acid⋯amide, same ring topology but between two different functional groups on API and coformer. • C(4) or C(6) — chain motifs, common for coformers like urea that bridge multiple API molecules into extended 1-D hydrogen-bonded chains rather than discrete dimers.
Because API homodimers (acid⋯acid, amide⋯amide) are thermodynamically robust, successful heterosynthon formation with a coformer requires that the heterosynthon be at least comparably stable — this is why coformer selection favors functional groups (pyridine, imide) known from CSD statistics to out-compete the API's self-association.
With the API's donor/acceptor inventory in hand, the Cambridge Structural Database (CSD) is queried using ConQuest/Mercury to determine, empirically, how often a candidate heterosynthon actually occurs across the >1.28 million deposited organic and metal-organic crystal structures. This converts Etter's qualitative rules into quantitative occurrence probabilities that rank coformers before a single gram is weighed out.
CSD synthon frequency analysis is performed with the Cambridge Crystallographic Data Centre's ConQuest search engine and Mercury visualization software:
1. Define a substructure query: e.g., a carboxylic acid fragment (–C(=O)OH) within 3.5 Å of a pyridine-type nitrogen on a different molecule (intermolecular contact filter). 2. Search restricted to organic, non-polymeric, R-factor <7.5% structures to ensure data quality. 3. Output: N structures containing both fragments in the crystal; of those, how many show the specific heterosynthon geometry (D···A distance 2.6–3.2 Å, D–H···A angle >140°). 4. Frequency = (structures showing target synthon) / (structures containing both fragment types) — this is the empirical probability the synthon will form given both groups are present in the same crystal.
Key published synthon frequencies (Aakeröy, Desiraju literature): • Carboxylic acid⋯pyridine: ~85–90% — one of the most reliable heterosynthons known, favored because pyridine-N is a much stronger acceptor than the amide/acid it displaces. • Carboxylic acid⋯primary amide (R2,2(8) heterodimer): ~55% — competes with both the acid homodimer and the amide homodimer; less predictable. • Carboxylic acid⋯carboxamide (catemer, C(4)): secondary motif seen when the heterodimer is sterically blocked. • Phenol⋯pyridine: ~70%, weaker than acid⋯pyridine but useful for APIs lacking a carboxylic acid. • Sulfonamide N–H⋯carbonyl O: common with saccharin- and sulfathiazole-type coformers.
For a co-crystal to be viable as a drug product, the coformer must itself be safe for human administration — this restricts the practical library to GRAS (Generally Recognized As Safe, FDA), approved excipients, or other API compounds ("drug-drug co-crystals"). A typical first-pass screening library (20–50 compounds) is built around a short list of workhorse coformers with well-characterized CSD synthon behavior:
• Nicotinamide (vitamin B3 amide): pyridine-N acceptor + primary amide donor/acceptor pair; extremely versatile, used in the marketed co-crystal Depakote-analog research and countless screening studies. • Saccharin: imide N–H donor + two sulfonyl acceptors; well known from the co-crystal that improved the dissolution of several BCS Class II APIs. • Citric acid: tricarboxylic acid, three independent COOH donor/acceptor pairs; useful when the API is a strong base (pyridine, amine) needing an acidic partner. • Urea: small, symmetric, bifunctional amide; forms chain (catemer) motifs bridging multiple API molecules, useful for lattice densification. • 4-Aminobenzoic acid (PABA): aromatic amine + carboxylic acid, dual-functionality building block. • Malonic, succinic, fumaric, tartaric, glutaric acids: the GRAS dicarboxylic acid series, systematically varied chain length used to tune melting point and hygroscopicity of the resulting co-crystal.
Selection strategy: cross-reference CSD-predicted synthon frequency (Stage 2 output) against a coformer's GRAS/IID (FDA Inactive Ingredient Database) status, aqueous solubility (a highly soluble coformer often improves co-crystal dissolution) and melting point (a high-melting coformer generally raises co-crystal lattice stability).
A rule of thumb from Desiraju's crystal engineering framework: rank coformers first by predicted CSD synthon probability, then break ties using GRAS status and aqueous solubility — this "probability-then-safety" ordering typically halves the number of physical experiments needed to find a hit versus unranked random screening.
Computational synthon ranking narrows a library of dozens of coformers to a short list of high-probability candidates, but only physical experiments confirm that a new crystalline phase actually nucleates. Liquid-assisted grinding (LAG) and slow solvent evaporation are the two workhorse low-material-consumption techniques used in early co-crystal screening, each requiring only 20–50 mg of API per experiment.
LAG (also called solvent-drop grinding) is the standard rapid-screening method because it is fast, solvent-minimal, and scalable directly to manufacturing-relevant methods (twin-screw extrusion):
1. Weigh API and coformer in stoichiometric ratio (typically 1:1 molar; 2:1 or 1:2 tested if the API/coformer has multiple donor or acceptor sites). 2. Add a catalytic quantity of solvent: η (eta) value = µL solvent per mg solid, typically η=0.25–0.5. Too little solvent (η<0.1) gives incomplete conversion; too much (η>1) approaches slurry conversion and can favor the thermodynamically most stable polymorph rather than exploring the full landscape. 3. Mill in a stainless-steel or agate jar with 1–2 grinding balls, 25–30 Hz, 20–60 minutes (Retsch MM400 mixer mill or equivalent). 4. The added solvent acts as a molecular lubricant/catalyst that mobilizes molecules at grain-boundary contact points, dramatically accelerating solid-state proton transfer and lattice reorganization compared to dry (neat) grinding. 5. Solvent choice matters: a solvent in which both API and coformer are sparingly soluble tends to favor co-crystal nucleation over simple recrystallization of the more soluble component.
LAG has a distinct mechanistic advantage over solution crystallization: because so little solvent is present, differential solubility of the API and coformer cannot cause selective dissolution/recrystallization of only one component — both are forced to interact at the solid interface.
Slow solvent evaporation from a stoichiometric solution provides both a screening method and, if repeated across a concentration series, the coformer/API ternary phase diagram needed to define a manufacturable crystallization process:
1. Dissolve API and coformer together at 1:1 (or other) molar ratio in a minimal volume of solvent (methanol, ethanol, acetone, or acetonitrile are common first choices), often warming gently to achieve full dissolution. 2. Filter (0.2 µm PTFE) to remove seed particles/dust, then allow slow evaporation at controlled temperature (typically 4–25°C) over 1–14 days. 3. Isolate any crystals that form and screen by PXRD before pursuing further characterization. 4. Ternary phase diagram (API–coformer–solvent, constructed via a solubility grid of multiple concentration ratios): identifies the co-crystal "existence region" — the composition window in which the co-crystal alone is thermodynamically stable versus regions where a physical mixture of the two starting materials, or a single-component polymorph, forms instead.
Other screening techniques used in parallel: hot-stage microscopy (observe eutectic melting/recrystallization directly under a polarized-light microscope as two solid particles are heated together — a classical Kofler contact method predates modern PXRD screening and remains a fast qualitative first pass), and slurry conversion (excess solid stirred in a solvent where the co-crystal is the least soluble phase — the thermodynamically most stable form is isolated after equilibration, typically 24–72 h at controlled temperature).
A new solid recovered from LAG or solvent evaporation is not automatically a co-crystal — it could be an amorphous blend, a simple eutectic physical mixture, or unreacted starting material. Three complementary, low-material-consumption analytical techniques are used together to unambiguously confirm that a genuine new hydrogen-bonded crystalline phase, built on the predicted supramolecular synthon, has formed.
PXRD is the primary and most decisive technique: every crystalline solid has a unique diffraction pattern determined by its unit cell, and a co-crystal — being a genuinely new crystal structure, not a mixture — produces a pattern with peak positions that match neither the pure API nor the pure coformer.
Protocol: 1. Gently grind the isolated solid to a fine powder (avoid excessive grinding, which can induce amorphization). 2. Load into a zero-background silicon holder; scan 2θ = 3–40° at 0.02° step size, Cu-Kα radiation (λ=1.5406 Å), typically 15–30 minutes per pattern on a bench-top diffractometer (e.g., PANalytical Empyrean, Bruker D8). 3. Overlay the new pattern against reference patterns for pure API, pure coformer, and (if available) a simple 1:1 physical mixture prepared by gentle blending (not grinding) — the physical mixture pattern is simply the linear superposition of the two component patterns. 4. Positive co-crystal indicators: appearance of new peaks not present in either component pattern, and/or disappearance of characteristic API or coformer peaks, indicating those crystal forms no longer exist in the sample. 5. Indexing (using software such as DASH, TOPAS, or Materials Studio Reflex) can assign a new unit cell to confirm single-phase purity and, combined with Rietveld refinement, provide a full crystal structure without single-crystal data if a good-quality powder pattern is obtained.
A physical mixture of API and coformer, even ground together, will show a PXRD pattern that is simply the arithmetic sum of the two pure-component patterns — peak positions unchanged, only relative intensities shifted by composition. Any shift in peak position, not just intensity, is the diagnostic signature of a true new crystalline phase.
FTIR (Fourier-transform infrared spectroscopy) probes the vibrational signature of the specific functional groups predicted to be involved in the new heterosynthon:
• Carboxylic acid C=O stretch: free/dimerized COOH ~1710 cm⁻¹ shifts to ~1680–1700 cm⁻¹ when hydrogen-bonded to a pyridine-N acceptor (heterosynthon formation weakens the C=O bond by delocalizing electron density into the new H-bond). • Broad O–H/N–H stretch (2500–3300 cm⁻¹) sharpens or shifts position/intensity as the donor group moves from a homodimer to a heterosynthon environment. • Primary amide N–H₂ asymmetric/symmetric stretches (~3350, 3170 cm⁻¹) and amide C=O (~1650–1680 cm⁻¹, amide I band) shift measurably when the amide participates in a new hydrogen bond to an acid rather than to itself. • ATR-FTIR (attenuated total reflectance) is typically used for speed and minimal sample prep (a few mg pressed directly onto the diamond crystal, no KBr pellet required).
Differential scanning calorimetry (DSC) provides thermal-behavior confirmation:
• Pure API and pure coformer each show their own characteristic melting endotherm at their literature melting points. • A simple physical mixture typically shows a small eutectic endotherm at a lower temperature (from limited mutual solubility in the melt) followed by the two separate component melts. • A true co-crystal shows a single, sharp melting endotherm at a temperature distinct from (usually intermediate to, but not always) both starting materials, with no residual endotherms from either pure component — because no unreacted starting material remains. • Sample preparation: 2–5 mg crimped in an aluminum pan, heated 10°C/min from 25°C to above the higher of the two component melting points, under 50 mL/min dry N2 purge (TA Instruments Q2000 or Mettler-Toledo DSC are common platforms). • Hot-stage microscopy under polarized light run in parallel visually confirms a single, congruent melt rather than two sequential melting events.
Confirming that a novel hydrogen-bonded crystalline phase exists is necessary but not sufficient — of the several co-crystal hits typically surviving analytical validation, only one is advanced into formulation development. Final candidate selection weighs each hit against Chemistry, Manufacturing and Controls (CMC) criteria: melting point and thermal stability, intrinsic dissolution rate, hygroscopicity, and physical stability under accelerated ICH storage conditions.
A co-crystal is only useful as a drug substance if it survives the full manufacturing and shelf-life journey. The CMC evaluation panel typically scores each surviving hit on:
1. Melting point and thermal stability: a co-crystal melting point too close to processing temperatures (granulation drying, hot-melt extrusion, or even accelerated storage) risks partial conversion during manufacture; ideal candidates melt comfortably above processing temperatures (commonly targeting Tm >150°C for oral solid dosage intermediates) with no decomposition detected by TGA below the melting endotherm.
2. Intrinsic dissolution rate (IDR): measured with a rotating- or stationary-disc apparatus (USP <1087>), compacted co-crystal pellet exposed to dissolution medium at fixed surface area and stirring rate — IDR (mg/min/cm²) is compared directly to the free-form API. Co-crystals are frequently pursued specifically to improve the IDR of BCS Class II (low solubility, high permeability) APIs; a ≥2× IDR improvement is a common go/no-go threshold.
3. Hygroscopicity: dynamic vapor sorption (DVS) across 0–90% relative humidity identifies the critical relative humidity (RH0) above which the co-crystal absorbs significant moisture or undergoes phase conversion (often back toward hydrate forms of the free API). A robust candidate should remain a stable anhydrous co-crystal up to at least 60% RH — well above typical packaging/storage humidity excursions.
4. Solid-state physical stability: samples stored per ICH Q1A(R2) accelerated conditions (40°C/75% RH, 6 months) and long-term conditions (25°C/60% RH, 12–24 months) are periodically pulled and re-analyzed by PXRD to confirm no reversion to the individual starting crystal forms and no new polymorph appearance.
5. Mechanical/manufacturability properties: powder flow, compactibility (tabletability profile via a compaction simulator), and particle morphology are assessed since co-crystals frequently show altered mechanical behavior (often improved plasticity) relative to the parent API crystal form.
Several well-documented industry examples illustrate how synthon-driven design translates into approved or late-stage products:
• A carbamazepine–saccharin co-crystal (widely studied academic model system, O'Nolan/Zaworotko and others) demonstrates the amide⋯imide heterosynthon and served as an early proof-of-concept that co-crystal engineering could rationally improve dissolution behavior of an anticonvulsant API without covalent modification.
• Entresto (sacubitril/valsartan, Novartis) is technically a molecular complex/co-crystal-like supramolecular assembly of two active pharmaceutical ingredients combined via calcium-mediated and hydrogen-bonded interactions with water and solvent molecules in the lattice — illustrating "drug-drug co-crystal" logic at commercial scale, where the "coformer" is itself an approved API rather than a GRAS excipient.
• Caffeine-based co-crystal families (with oxalic, malonic, glutaric, and other GRAS diacids) are heavily represented in the academic literature as systematic model systems for studying how coformer alkyl-chain length tunes co-crystal melting point and hygroscopicity — informative for selecting a homologous acid series when tuning a real API co-crystal's CMC profile.
The consistent lesson across these programs: the coformer that gives the cleanest, highest-frequency CSD-predicted synthon is not always the coformer that survives CMC review — a slightly lower-probability synthon partner with better GRAS status, lower hygroscopicity, and a more processable melting point is frequently the one advanced into formulation development.
Industry rule of thumb: budget for 20–50 coformers computationally ranked, 8–15 taken into physical LAG/evaporation screening, 2–5 confirmed as genuine new crystalline phases by PXRD/FTIR/DSC, and typically only 1 advanced past full CMC review into formulation — synthon prediction shrinks the search space, but does not replace experimental developability testing.