A freeze-drying cycle has no temperature ceiling of its own. The ceiling belongs to the formulation, and it is fixed before the vials go into the dryer — by which solutes are present, at what ratio, and whether each one ends up amorphous or crystalline. Get that wrong and shelf-temperature tuning will not recover it: either the cake sinks, or it looks immaculate and the peptide inside it is not what you froze. This is the formulation side of lyophilisation, separate from the question of why a peptide ships dry in the first place.
Three stages, three different control problems
Freezing, primary drying and secondary drying are not three settings on one operation. Tang and Pikal's process-design review separates them explicitly: control of ice nucleation and crystallisation during freezing; selection of a target product temperature for primary drying, referenced to the collapse temperature and to the thermal transition denoted Tg′; and a separate set of rules for optimising secondary drying [1].
What freezing controls is geometry. Ice crystals become the pores through which water vapour later escapes, so the nucleation temperature propagates into mass transfer for the whole cycle. Konstantinidis and colleagues manipulated chamber pressure to nucleate every vial simultaneously. In 5% w/w mannitol, uncontrolled nucleation occurred randomly between −8.0 °C and −15.9 °C as vials cooled to −40 °C, whereas controlled nucleation was achieved between −2.3 °C and −3.7 °C. Nucleating warmer enlarged the effective pore radius from 13 to 27 µm, and cycles generated from the resulting resistance parameters for that 5% mannitol system cut primary drying time by 41% when tested experimentally [2]. Warmer nucleation gives fewer, larger crystals; larger crystals give wider channels.
The same geometry carries a cost. Ice surface is interface, and interface is where unfolding starts. Sarciaux and colleagues freeze-dried bovine IgG with no lyoprotective solute and detected no insoluble aggregate after freezing alone, but always detectable insoluble aggregate after freeze-drying — consistent with denaturation at the ice/freeze-concentrate interface that is reversible on freeze-thaw and becomes irreversible after freeze-drying and reconstitution. Rapid cooling by quenching in liquid nitrogen produced more and larger aggregates than slow cooling on the shelf, matching the higher specific surface area of the fast-cooled solids, and annealing the rapidly cooled solutions gave significantly less aggregation with a corresponding fall in specific surface area [3]. The aggregation figures belong to that antibody; the geometric argument — fast cooling buys small crystals, small crystals buy surface area, surface area is a liability — is general to freezing.
Tg′ and the collapse temperature are two different numbers
As ice forms, everything else is excluded into a shrinking unfrozen phase until that phase is too viscous to concentrate further. The glass transition of that maximally freeze-concentrated solute phase is Tg′, and it is a property of the whole solute mixture rather than of the peptide. Kasraian and colleagues measured Tg′ at −32.3 °C for a 7% w/v sucrose solution, and adding glycine drove it down linearly, reaching −51.5 °C at a sucrose/glycine ratio of 2:5 [4]. Every excipient you add moves this number, and the small-molecule ones — amino acids, salts, polyols — generally move it downward.
Collapse is not a synonym for Tg′. In freeze-drying microscopy of L-arginine–citric acid mixtures, Yamaki and colleagues observed collapse onset at temperatures roughly 10 °C above the corresponding Tg′ values, and in actual freeze-drying at low chamber pressure, physical collapse appeared at shelf temperatures close to or slightly above that collapse temperature. Raising chamber pressure slowed sublimation and moved collapse to lower shelf temperatures, which the authors read as confirming the significance of sublimation-related heat loss on the interface temperature during primary drying [5]. Shelf temperature is not product temperature. The roughly 10 °C gap belongs to those arginine–citrate systems; what generalises is the ordering, not the interval.
Whether you must stay below Tg′ is genuinely contested. Depaz and colleagues ran primary drying above Tg′ for four proteins from three different classes and obtained pharmaceutically acceptable cakes at the higher protein concentrations, provided the product stayed below the collapse temperature; using an IgG4 monoclonal antibody as the example, minor extents of collapse appeared as protein concentration decreased. They observed no other impacts on product quality or on physical or chemical stability across the drying conditions tested [6]. Read that as a reason to measure the collapse temperature rather than assume it equals Tg′ — not as general permission. The result covers those formulations and those attributes.
Cryoprotection and lyoprotection are two jobs
The freezing stress and the dehydration stress are separable, which is exactly what the bovine IgG result shows: damage a thaw undoes is not the damage the dryer leaves behind [3]. Excipients are therefore asked to do two different things, and the two dominant explanations of how sugars do the second one have not converged.
The water-replacement account says the sugar hydrogen-bonds to the solute where water used to. Carpenter and Crowe's infrared work is the classic evidence. Freeze-drying trehalose, lactose or myo-inositol with lysozyme produced large frequency shifts, decreased absorbance and loss of band splitting in the carbohydrate fingerprint region (900–1500 cm−1), mimicking the effects of water on hydrated trehalose, while dehydration-induced shifts in the protein's amide I and amide II bands were partially and fully reversed respectively when lysozyme was dried with trehalose or lactose. myo-Inositol was less effective at shifting the amide bands. Across carbohydrate concentrations, the ability to shift amide II back toward its hydrated position coincided — at least in the extreme cases — with the capacity of that same level of carbohydrate to preserve rabbit skeletal muscle phosphofructokinase activity through freeze-drying [7].
The vitrification account says the sugar glass simply immobilises everything in it. Cicerone and Soles tested that with neutron scattering on a series of trehalose glasses diluted with glycerol, and found that stabilisation of horseradish peroxidase and yeast alcohol dehydrogenase correlated with suppression of short-length-scale, high-frequency dynamics rather than with the bulk glass transition: in that series, the glass conferring maximum room-temperature stability did not have the highest Tg [8]. If you were planning to rank candidate formulations by Tg alone, that is the result to read first.
Chang and colleagues put the two accounts against each other using an IgG1 antibody and recombinant human serum albumin. Sorbitol only slightly protected against aggregation and had no effect on chemical degradation, whereas chemical degradation and aggregation rates both fell as the sucrose-to-protein weight ratio rose from 0 to 2:1. In a separate measurement on the same system, the stretched relaxation time τβ plotted against sucrose fraction passed through a maximum at a 1:1 sucrose-to-protein weight ratio, so molecular mobility did not correlate with stability at high sucrose content. Storage stability correlated best with retention of native structure as measured by FTIR, though the authors state plainly that the evidence for a specific stabiliser–protein interaction is not compelling [9]. Two ratios, two different measurements — they are not interchangeable.
A crystallising excipient stops protecting
Protection requires the sugar to share a phase with the peptide. Izutsu and colleagues showed what happens when it does not: the crystalline morphology of mannitol and inositol in freeze-dried cakes depended on the pre-freezing solute concentration and on the freeze-drying method used, and while the additives in their amorphous state stabilised Aspergillus oryzae β-galactosidase in a concentration-dependent way, additive crystallisation during freeze-drying decreased those effects. Heat treatment before freeze-drying likewise caused crystallisation and diminished the stabilising effect [10]. A crystal lattice excludes the very solute it was meant to protect.
Worse, the crystallisation can happen after you have checked for it. Pyne and colleagues followed mannitol by in situ X-ray powder diffraction with a vacuum pump attached to the low-temperature diffractometer stage, simulating the process in the sample chamber. Rapidly cooled 10% w/w mannitol stayed amorphous, but annealing or primary drying — both below Tg′ — crystallised mannitol hydrate. With trehalose present at 5% w/w or more, annealing below Tg′ produced nucleation but not crystallisation, yet mannitol hydrate still crystallised during primary drying [11]. A frozen-state thermogram showing no crystallisation event does not establish that the dried cake is amorphous.
None of this makes crystalline excipients useless. It makes their job structural rather than protective. In the sucrose/glycine system, mixtures at 3:4 and 2:5 sucrose-to-glycine had glass transitions of −48 °C and −51.5 °C and nonetheless freeze-dried readily at a product temperature above Tg′ on a fast cycle at +20 °C shelf temperature and 100 mTorr chamber pressure, with no sign of collapse; collapse was avoided wherever the DSC thermogram showed a transition from recrystallisation of the amorphous glycine, and was severe for mixtures dried above Tg′ without it [4]. The bulking agent holds the cake up. It is not what keeps the peptide folded, and if it withdraws mass from the amorphous phase you have traded protection for appearance.
A peptide-specific test of the same choice: Schaal and colleagues compared a trehalose-based PEGylated peptide formulation against reformulations replacing trehalose with mannitol or with a 75:25 sucrose–mannitol blend, processed under four spin-freezing conditions differing in cooling and crystallisation rate and stored up to 13 weeks at 2–8 °C or 50 °C. Varying the freezing parameters produced no consistent or statistically significant effect on peptide or monomer levels; formulation composition emerged as the dominant stability determinant. Trehalose-based samples held up under both storage conditions, mannitol-based samples showed moderate degradation at the elevated temperature, and the sucrose–mannitol blend showed pronounced instability at 50 °C with cake collapse, browning, and interference in peptide quantification, which the authors attribute to likely sucrose hydrolysis and formation of aromatic degradation products [12]. Note the second-order consequence: the excipient's own degradation products broke the assay used to measure the peptide.
The sugar can be the reactant
Reducing sugars carry a reactive carbonyl. Li and colleagues lyophilised recombinant human relaxin with several sugar and polyol excipients and saw rapid covalent modification in the presence of glucose. LC/MS and tryptic mapping identified Maillard adducts of glucose with amino groups on lysine and arginine side chains, plus a significant amount of serine cleavage from the C-terminal of the B-chain. That cleavage occurred to a greater extent in the solid state than in solution — drying does not switch chemistry off, it selects which chemistry runs. Mannitol and trehalose produced stable lyophilised formulations of the same molecule [13].
Buffers move during freezing
Freeze concentration drives buffer components toward their solubility limits, and where one member of a conjugate pair crystallises out and the other does not, the effective acidity of the remaining phase shifts. Lay-Fortenbery and colleagues measured apparent acidity in six trehalose-containing lyophilised buffer systems using two independent probes: an isotopically labelled carboxylic acid read by solid-state NMR, and bromocresol purple read by UV/Vis diffuse reflectance. Sodium citrate showed little change in apparent acidity relative to the pre-lyophilisation solution, whereas potassium phosphate, sodium phosphate and histidine all showed greater change. The two methods trended similarly but disagreed on the numerical equivalent pH values [14]. Treat the absolute numbers as method-dependent and the ranking as the usable output.
Counterion identity also does work that is not pH. In the bovine IgG system, sodium phosphate consistently gave more turbid reconstituted solids than potassium phosphate at the same concentration, and the authors state explicitly that this effect was not attributable to a pH shift during freezing. Added sodium chloride or potassium chloride contributed markedly to insoluble aggregate formation, and both made the reconstituted solid more turbid than either phosphate buffer at similar ionic strength, with sodium chloride giving substantially higher aggregate levels than potassium chloride — and at a given cooling rate the specific surface area of the dried solids was about a factor of two higher for the sodium chloride formulation [3].
Residual moisture is a window, not a floor
Water plasticises the glass, so residual moisture and Tg are one variable read two ways. Breen and colleagues lyophilised a humanised monoclonal antibody in a sucrose/histidine/polysorbate 20 formulation to residual moistures from 1% to 8% and measured Tg by modulated DSC from 80 °C at 1% moisture down to 25 °C at 8%. High-moisture cakes aggregated faster than drier samples when stored above their Tg values. But intermediate-moisture vials were more stable to aggregation than dry vials, and the authors' conclusion separates the two failure modes: high moisture decreased chemical stability regardless of whether the protein was glassy or rubbery, while physical stability was not compromised, and may even be enhanced, by increasing residual moisture if storage is below Tg [15]. Driving secondary drying as far as it will go is a decision with a cost, not a default. Sarciaux's group saw the asymmetry from the other side — turbidity in the reconstituted bovine IgG was influenced by the extent of secondary drying, which they read as reason to minimise secondary drying for that system [3].
What a good-looking cake licenses you to conclude
Very little about the molecule. Schersch and colleagues induced collapse two independent ways — by varying the ratio of crystalline mannitol to amorphous sucrose, and by running an aggressive collapse cycle against a conventional protocol on identical formulations — and quantified the extent of collapse as the decrease in specific surface area. Collapsed cakes had residual moisture levels comparable to non-collapsed lyophilisates, reconstitution times were not increased, and protein stability was not relevantly different, across an IgG1 antibody, a second protein and lactate dehydrogenase [16]. Sarciaux's group put the blunter version on record: there was no correlation between the pharmaceutical acceptability of the freeze-dried cake and insoluble aggregate levels in the reconstituted product [3].
Cake appearance reports on one thing — whether the amorphous matrix stayed rigid enough to hold its own pore walls during drying. That is worth knowing, and a collapsed cake remains evidence that the cycle exceeded a critical temperature. It is not an assay. Monomer content, covalent modification, secondary structure and residual moisture each require their own measurement, and cake appearance predicts none of them reliably.
Assembling the formulation
The order of decisions follows the physics. Choose the amorphous stabiliser first and fix its ratio to the peptide, because that ratio is what the stabilisation data are indexed to [9]. Prefer a non-reducing sugar unless you have specific evidence for the alternative [13]. Establish whether any bulking agent will crystallise, and establish it in the dried cake rather than only in the frozen solution [10, 11]. Select the buffer for its behaviour on freezing, not only its aqueous pKa [14]. Then measure Tg′ by DSC and the collapse temperature by freeze-drying microscopy on the actual formulation, since every excipient shifts both [4, 5], and set the target product temperature from those measurements rather than from a generic rule [1]. Finally, treat residual moisture as a target window with an upper and a lower bound [15].