💎 Polymorph High-Throughput Crystallization Screen

Solvent Selection & 384-Well Crystallization Matrix Design

Before any liquid is dispensed, a polymorph screen lives or dies on the design of its experimental matrix. The goal is to sample the widest practically achievable region of crystallization phase space — solvent polarity, hydrogen-bond donor/acceptor character, antisolvent ratio, and thermal history — so that no accessible polymorphic form of the API is missed during CMC preformulation.

• 8: Solvent panel size (spanning Hansen δ-space)
• 8×4×3×4: Matrix dimensions (solvent × antisolvent × temp × reps)
• 384: Total wells (four 96-well plates)
• 50–200 µL: Dispense volume (per well, robotic)

Hansen solubility parameters and solvent panel diversity

Polymorph screens are only as good as the chemical diversity of the conditions tested. The Hansen solubility parameter (HSP) framework decomposes solvent character into three additive components: δD (dispersion forces), δP (dipolar interactions), and δH (hydrogen bonding), each in MPa^0.5. Plotting candidate solvents in this 3-D space and selecting 8 that maximize pairwise Euclidean distance ensures the panel is not accidentally redundant (e.g., choosing both acetone and MEK, which cluster tightly).

A representative 8-solvent panel for a small-molecule API: • Ethanol (δD 15.8, δP 8.8, δH 19.4) — protic, moderate polarity, common recrystallization solvent • Acetonitrile (δD 15.3, δP 18.0, δH 6.1) — aprotic, high dipolarity • Ethyl acetate (δD 15.8, δP 5.3, δH 7.2) — moderate polarity ester • Acetone (δD 15.5, δP 10.4, δH 7.0) — fast-evaporating ketone • Isopropanol (δD 15.8, δP 6.1, δH 16.4) — bulkier protic alcohol • Tetrahydrofuran (δD 16.8, δP 5.7, δH 8.0) — aprotic ether, good API solubility • Dimethyl sulfoxide (δD 18.4, δP 16.4, δH 10.2) — high-boiling, strong H-bond acceptor • N-methyl-2-pyrrolidone (δD 18.0, δP 12.3, δH 7.2) — high-boiling amide solvent

Antisolvents are chosen from the opposite end of the polarity spectrum: heptane and methyl tert-butyl ether (MTBE) are the two most common antisolvents for organic APIs, added at 0%, 20%, 40%, and 60% v/v to trigger supersaturation-driven nucleation that would not occur in the pure solvent alone.

Building the 384-well experimental matrix

The full factorial design crosses:

• 8 solvents (rows A–H equivalent, cycling across 4 plates) • 4 antisolvent ratios: 0%, 20%, 40%, 60% v/v heptane or MTBE • 3 temperature/cooling profiles: rapid quench (4 °C direct), slow cooling (0.1 °C/min from 50 °C to 4 °C), and room-temperature evaporation • 4 replicates per condition, to distinguish stochastic nucleation events from reproducible hits

8 × 4 × 3 × 4 = 384 wells — exactly filling four standard 96-well glass or PP microplates. API stock solutions are prepared at a fixed supersaturation ratio (typically 1.2–2.0× the solvent's equilibrium solubility at 50 °C, determined by a rapid turbidimetric solubility pre-screen) so that all 384 wells start from a comparable thermodynamic driving force.

Well-to-well dispense volumes: 50 µL of API stock + 100–150 µL solvent/antisolvent mixture, totaling 150–200 µL per well — small enough that a full 384-well plate consumes under 200 mg of API, a critical constraint during early-phase development when drug substance supply is limited to grams, not kilograms.

ICH Q6A (Specifications for New Drug Substances) explicitly calls out polymorphism as a critical quality attribute requiring characterization when the API exists in multiple crystal forms with materially different properties. A well-designed HTS matrix is the primary tool for satisfying this expectation before a single form is locked into a regulatory filing.

Automated Parallel Crystallization — Cooling, Evaporation, and Antisolvent Methods

Once the matrix is designed, execution must be as reproducible as the plan demands. Robotic liquid handlers such as the Freeslate (formerly Symyx) Core Module or the Formulatrix Rock Imager-integrated dispense stations execute all 384 wells with microliter precision, and programmable temperature-controlled plate hotels drive each well through an identical thermal or evaporative history.

• 0.1–1.0 °C/min: Cooling ramp range (programmable profiles)
• ±2 µL: Dispense precision (robotic liquid handler)
• 96-well glass/PP: Plate format (×4 plates = 384 wells)
• 6: Imaging timepoints (0, 4, 24, 72, 168, 336 h)

Robotic dispensing and plate sealing

A Freeslate/Symyx-class robotic platform (or equivalent Chemspeed SWING/ISYNTH system) executes the full 384-well fill sequence unattended:

1. API stock solution (pre-filtered 0.2 µm to remove seed particles) is aliquoted into each well via a positive-displacement syringe module, accurate to ±2 µL 2. Solvent and antisolvent are added per the matrix design, with antisolvent addition rate controlled at 10 µL/s to standardize local supersaturation spikes across wells 3. Plates are sealed with a heat-sealed aluminum/polypropylene laminate film (gas- and vapor-tight) to prevent uncontrolled evaporation in the "closed" cooling-crystallization wells, while evaporation wells use a gas-permeable membrane seal instead 4. Sealed plates are loaded into a Formulatrix Rock Imager storage hotel or an in-house temperature-controlled carousel for the programmed thermal or evaporative protocol

Crystallization methods run in parallel across the plate

Four fundamentally distinct crystallization methods are represented across the matrix so the screen is not blind to kinetically-favored forms that only appear under one nucleation pathway:

• Cooling crystallization: API dissolved near saturation at 50 °C, then cooled at a programmed rate of 0.1–1.0 °C/min to 4 °C. Slow cooling (0.1 °C/min) favors the thermodynamically most stable form; rapid quench (effectively >5 °C/min) can kinetically trap metastable forms. • Evaporative crystallization: wells left open to a controlled-humidity chamber; solvent evaporates over 3–14 days, progressively concentrating the solution until spontaneous nucleation. Excellent for capturing solvates, since the last solvent molecules to leave are often incorporated into the lattice. • Antisolvent (drown-out) crystallization: heptane or MTBE added stepwise to the API/solvent solution, driving the mixture rapidly past the metastable zone width. High supersaturation ratios from fast antisolvent addition tend to favor faster-nucleating, sometimes amorphous or metastable, forms. • Slurry (thermodynamic) conversion: excess solid API stirred in each solvent for 48–72 h at controlled temperature, allowing Ostwald-ripening-driven conversion toward the most stable form present — used as a cross-check against the kinetic hits from the other three methods.

Seeding-free wells (the majority of the matrix) probe genuine nucleation behavior; a parallel seeded sub-set (10% of wells, seeded with previously known Form I or Form II crystals) is included to test whether seeding suppresses discovery of novel forms or accelerates conversion to known ones.

Cooling rate is one of the single most consequential experimental variables in a polymorph screen. Dropping from 1.0 °C/min to 0.1 °C/min can shift the population of a 384-well plate from >80% amorphous/oil to >60% well-formed crystalline hits, because slower cooling keeps supersaturation inside the metastable zone width where ordered nucleation dominates over uncontrolled precipitation.

In-Situ Imaging, Automated Hit Classification, and Spectroscopic Triage

With 384 sealed wells running unattended thermal or evaporative programs, the screen needs eyes that never blink. In-situ polarized light imaging captures every well at fixed timepoints, and an automated image classifier triages the resulting tens of thousands of images into actionable categories long before a human scientist looks at a single picture.

• 2,304: Images captured (384 wells × 6 timepoints)
• 3: Classifier categories (clear / amorphous / crystalline)
• 57 / 384: Crystalline hits found (14.8% hit rate)
• ~3 min/well: Raman/PXRD triage time (per hit, automated stage)

Polarized light microscopy and birefringence detection

A Formulatrix Rock Imager (or Crystal16/Crystalline parallel reactor platform with an integrated turbidity/imaging probe) captures brightfield and cross-polarized images of every well at each timepoint without breaking the plate seal:

• Brightfield imaging: detects gross precipitate, oiling-out, or clear solution • Cross-polarized light imaging (crossed polarizers at 90°): ordered crystalline lattices rotate the plane of polarized light and appear bright ("birefringent") against a dark background; amorphous solids and true solutions remain dark • A birefringence intensity threshold (mean pixel brightness under crossed polarizers, background-subtracted) provides a fast, quantitative first-pass signal: wells above threshold are flagged as candidate crystalline hits

Six imaging timepoints (0, 4, 24, 72, 168, 336 h) capture both fast-nucleating and slow-nucleating conditions — some polymorphs only appear after a week of standing, particularly hydrates that require ambient humidity equilibration.

CNN-based image classification and spectroscopic confirmation

Raw images are automatically triaged by a convolutional neural network classifier (trained on thousands of historical well images from prior screens) into three categories:

• Clear solution — no visible solid, birefringence signal at background level • Amorphous precipitate — solid present but no birefringence (isotropic); often an oiled-out or glassy phase, not a true crystal • Crystalline hit — clear birefringence signal, distinct particle/needle/plate morphology visible in brightfield

On a representative 384-well matrix: roughly 41% of wells remain clear solution (insufficient supersaturation), 44% show amorphous precipitate (161 wells), and 57 wells (14.8%) are classified as genuine crystalline hits — the wells that proceed to spectroscopic confirmation.

Every crystalline hit is triaged by one of two rapid, non-destructive techniques directly at the well:

• Raman microspectroscopy: a 785 nm or 532 nm laser probes lattice vibrational modes through the sealed plate in under a minute per well; distinct polymorphs of the same molecule often show shifted or split peaks in the fingerprint region (400–1800 cm⁻¹) even when bulk composition is identical • Micro-PXRD (transmission mode, ~50 µm beam, synchrotron or micro-focus lab source): collects a low-resolution diffraction pattern (2θ 3–35°) directly from the small crystal mass in a well, sufficient to distinguish a handful of major peak positions between forms without extracting the sample

A 14.8% hit rate is typical for a well-designed HTS polymorph matrix; screens that return >40% "hits" are usually flagging amorphous or oiled-out precipitate as false positives, while screens returning <5% usually indicate a supersaturation ratio set too low in the API stock preparation.

Polymorph Clustering — PXRD, DSC, and Ranking Distinct Crystal Forms

Fifty-seven crystalline hits do not mean fifty-seven polymorphs — most are duplicates of the same underlying crystal form obtained under different conditions. Full-pattern PXRD comparison and DSC thermal analysis, combined with hierarchical clustering, collapse the raw hit list down to the true number of distinct polymorphic families present in the screen.

• 2θ 3–40°: PXRD scan range (Cu Kα, λ=1.5406 Å)
• 10 °C/min: DSC heating rate (N₂ purge, 25→250 °C)
• 4: Distinct forms clustered (Form I, Form II, hydrate, solvate)
• 187 °C: Form II melting point (ΔHfus = 112 J/g)

Full-pattern PXRD comparison and hierarchical clustering

Each of the 57 hits is scaled up to a few milligrams (via slurry or slow evaporative regrowth of the identified condition) and run on a benchtop or synchrotron powder X-ray diffractometer over 2θ = 3–40° using Cu Kα radiation (λ = 1.5406 Å):

1. Baseline-subtract and normalize each diffractogram 2. Extract peak positions and relative intensities for the 10 most intense reflections per pattern 3. Compute a pairwise dissimilarity matrix between all 57 patterns using a weighted Euclidean distance over matched peak positions (allowing ±0.1° 2θ tolerance for sample-height/instrument variation) 4. Apply hierarchical (agglomerative) clustering with average linkage; cut the dendrogram at a dissimilarity threshold empirically set to separate known reference patterns of Form I and Form II by more than one cluster level

On the representative 57-hit dataset, this clustering resolves 4 distinct families: • Form I (22 hits) — the previously known, thermodynamically stable anhydrous form, appearing across ethanol, acetonitrile, and ethyl acetate conditions at low antisolvent ratios • Form II (19 hits) — a previously known metastable anhydrous form, favored by rapid antisolvent addition (40–60% heptane) and fast cooling • A monohydrate (11 hits) — appearing predominantly in evaporative wells run from aqueous-miscible solvents (ethanol, isopropanol), consistent with water incorporation during slow solvent loss • A novel THF solvate (5 hits) — a previously uncharacterized channel solvate found only in tetrahydrofuran wells, flagged for full single-crystal structure determination

DSC thermal analysis and thermodynamic stability ranking

Differential scanning calorimetry (DSC), run at 10 °C/min from 25 °C to 250 °C under nitrogen purge, provides the second independent line of evidence and quantifies each form's thermal behavior:

• Form I: single sharp melting endotherm at 201 °C, ΔHfus = 128 J/g — no solid-solid transitions observed on heating, consistent with a stable, well-ordered lattice • Form II: melting endotherm at 187 °C, ΔHfus = 112 J/g, frequently preceded by a small exotherm around 165 °C corresponding to solid-state conversion of a fraction of the sample to Form I before it melts — a classic signature of a monotropically related metastable form • Monohydrate: broad endotherm at 90–110 °C (dehydration, mass loss confirmed by TGA) followed by recrystallization exotherm and a final melt matching Form I, indicating the hydrate desolvates to the anhydrous stable form on heating • THF solvate: desolvation endotherm at 78 °C, followed by melting behavior distinct from both Form I and Form II, supporting its classification as a genuinely novel form rather than a mixture

Because Form II converts to Form I on heating (rather than the reverse), the pair is classified as monotropic: Form I is more stable at all temperatures below the melting point, and no solid-solid transition temperature exists where their relative stability inverts. This ranking, combined with solubility measurements (Form I solubility in water: 0.42 mg/mL at 25 °C vs. Form II: 0.61 mg/mL, a 1.45× advantage for the metastable form), sets up the central trade-off resolved in scale-up selection: thermodynamic stability (Form I) versus higher apparent solubility and potentially better bioavailability (Form II).

A form that is more soluble is not automatically the better development candidate. Form II's 45% solubility advantage over Form I must be weighed against its tendency to convert to Form I under stress (humidity, compaction, elevated storage temperature) — an uncontrolled polymorphic transition during shelf-life is a major regulatory and manufacturing risk that PXRD/DSC form-ranking is specifically designed to catch before it reaches Phase III.

Gram-Scale Confirmation and Developability-Driven Polymorph Selection

A polymorph identified in a 200 µL microplate well is not yet a manufacturable drug substance. The final stage reproduces the best-ranked hit at a scale that can be handled, filtered, dried, and characterized by conventional bulk techniques — and folds in the non-crystallographic factors (stability, bioavailability, patent landscape) that ultimately decide which form advances into formulation development.

• 5–20 g: Scale-up batch size (jacketed lab reactor)
• >99%: Phase purity required (single form by PXRD)
• Form II: Selected form (higher solubility, controlled process)
• Form I only: FTO patent check (claimed by originator, expired 2031)

Reproducing the hit at gram scale

The winning microplate condition — for example, ethyl acetate with 40% heptane antisolvent added at controlled rate, cooled at 0.3 °C/min from 45 °C to 5 °C — must be translated into a jacketed lab reactor (typically 100 mL–1 L) capable of producing 5–20 g of material:

1. Solubility curve confirmation: re-measure API solubility in the selected solvent system across the full temperature range at gram scale, since well-plate solubility can be skewed by container surface effects at microliter volumes 2. Controlled antisolvent addition: replace the robotic microliter dispense with a syringe pump delivering antisolvent at a rate calibrated to reproduce the same local supersaturation ratio measured in the well (addition time scaled proportionally to batch volume) 3. Seeding: gram-scale batches are deliberately seeded (0.1–1% w/w of the target form) near the metastable zone width boundary to control nucleation onset and avoid the stochastic, sometimes multi-form outcomes seen in unseeded microwells 4. Isolation: vacuum filtration through a Büchner funnel, wash with a small volume of cold antisolvent to remove mother liquor without dissolving the cake, then dry under vacuum at a temperature at least 20 °C below the form's desolvation/melting onset

Confirming phase purity and choosing the developable form

The scaled-up cake is characterized by the same PXRD and DSC methods used in Stage 4, now with sufficient material for confirmatory techniques not feasible on microwell quantities:

• Full PXRD pattern match (Rietveld or simple peak-overlay) against the reference pattern for the target form, confirming >99% phase purity with no detectable co-existing form • Thermogravimetric analysis (TGA) to confirm absence of unexpected solvate/hydrate mass loss • Particle size distribution (laser diffraction) and morphology (SEM) to characterize the habit relevant to downstream filtration, drying, and milling behavior • Dynamic vapor sorption (DVS) to test hygroscopicity and hydrate-formation risk under accelerated humidity (40–80% RH) • Competitive slurry stability: the candidate form is slurried together with the other identified forms in a saturated solution for 1–2 weeks at 25 °C and 40 °C — if it does not convert, it is confirmed as the thermodynamically dominant (or kinetically persistent) choice under those storage-relevant conditions

Final selection weighs multiple factors beyond crystallography alone: • Thermodynamic stability — favors Form I, lower risk of uncontrolled conversion during shelf life • Solubility/bioavailability — favors Form II, 45% higher aqueous solubility translating to improved dissolution rate for a BCS Class II compound • Process robustness — favors whichever form has a wider, more reproducible metastable zone width and is less sensitive to minor process deviations • Freedom-to-operate — a patent landscape search shows the originator's existing composition-of-matter and polymorph patents claim Form I specifically (expiring 2031), while Form II remains unclaimed, an important strategic consideration for a follow-on or generic development program

In this representative program, Form II is selected as the development candidate despite being thermodynamically metastable: its solubility advantage materially improves projected oral bioavailability, its conversion to Form I is shown to be slow and controllable under the intended manufacturing and packaging conditions (confirmed by 6-month accelerated stability data), and it sits outside the originator's claimed polymorph patent estate — illustrating that polymorph selection in CMC development is a multi-factor engineering decision, not simply "pick the most stable form."