Extractables and Leachables in Peptide Container-Closure Systems

A container is a solid solution of its own processing chemistry. An elastomeric closure carries antioxidants, vulcanisation accelerators, fillers and cure residues; a polyolefin body carries stabilisers and mould-release agents; and both carry whatever those compounds became during vulcanisation, moulding and sterilisation. The fill is a sink at zero concentration for all of them. Migration is the default state; the study asks how much arrives and what it does on arrival.

Two words describe two different measurements, and the distinction carries the whole assessment. Extractables are what a material releases under exaggerated laboratory conditions — an aggressive solvent, an elevated temperature, a contact time chosen to push the system toward equilibrium. Leachables are what actually appears in the fill under its own storage conditions, in its own vehicle, over its own storage period. The first is a property of a material and a protocol; the second, of a finished unit.

A migration study on a model container-closure system shows how loosely the two are coupled. Jenke and colleagues assembled a system from a linear low-density polyethylene bottle, a polypropylene cap with an elastomeric liner, an adhesive label and a foil overpouch, filled it with three simulating solvents — an aqueous salt/acid mixture at pH 2.5, an aqueous buffer at pH 9.5, and 1:1 v/v isopropanol/water — and stored the units upright or inverted for up to six months at 40 °C ^[1]. The profile that came back could be correlated qualitatively with the composition of the materials of construction, but the chemical nature of both the extracting medium and the individual extractable markedly affected which substances appeared and at what level. The same study found that direct contact between the solution and a material is not a prerequisite: substances associated with a label applied to the outside of the bottle accumulated as leachables inside it ^[1].

The converse case is equally instructive. Across five lyophilised and nineteen liquid protein products held in a range of rubber-stoppered vial and syringe systems, the organic and elemental leachable profiles sat below the limit of detection at every time point examined ^[2]. A long extractables list is a hazard inventory, not a measured burden. What a study owes is a demonstration that those entries either do not migrate or migrate below the threshold you set before you started.

What the extraction solvent decides

A controlled extraction study is not a survey of everything the material contains. Li and Sobańtka's systematic analysis works through the chain that governs what a solvent recovers: the degree of solvent–material interaction, the swelling of the material, and the diffusion rate of its constituents, with the extent of swelling in particular setting the molecular-weight range that can escape ^[3]. The number of compounds in an extract, their concentrations and their molecular weights are therefore all solvent-dependent. There is no absolute extractables profile; there is a profile relative to a chosen solvent set, and the set is part of the result.

How small can that set be? The ELSIE materials working group ran a comprehensive protocol — several extraction solvents, several extraction techniques, several analytical techniques — against polyethylene and polyvinyl chloride, specifically to test whether the protocol could be reduced to fewer techniques and fewer solvents while still supporting material selection ^[4]. For PVC at least, such reductions proved feasible. Two secondary findings from that work travel further than the primary one: extractable elemental impurity levels in both materials were low, and orthogonal analytical detection techniques were confirmed as necessary to understand an extraction profile at all ^[4].

Exaggeration has a limit that follows from the swelling mechanism. Push the solvent hard enough to open the polymer matrix well beyond anything the aqueous fill will do, and the extract reports on a material state the storage condition never reaches ^[3]. That is not an argument against exaggeration, which exists to surface candidates early, but it is why an extractables hit is a hypothesis about a leachable rather than a measurement of one.

Where the reporting threshold comes from

Chromatographic screening returns peaks, many of which will never be identified, and the reporting threshold decides which of them get chased. Its numerator is toxicological. Jenke and Carlson collated toxicological information for over 540 extractables identified in laboratory testing of polymeric materials used in pharmaceutical applications, taking endpoints such as no-observed-effect and no-observed-adverse-effect levels for those compounds or structurally similar surrogates, and running them through a systematic scheme with four uncertainty factors, each assigned a value of 10, covering the quality and relevance of the data among other differences ^[5]. A threshold assembled that way is a statistical construct, not a physical constant.

The analytical half is where the arithmetic usually breaks. Quantitation in extractables and leachables screening is normally done by comparing an analyte's response to that of an internal standard added at a known amount — simple, and resting on the largely false premise that all extractables, leachables and internal standard candidates share a response factor ^[6]. A poorly responding compound will sit below the AET while its true concentration is above it. The correction is an uncertainty factor that lowers the AET, its size set by the magnitude of response factor variation.

The analytical evaluation threshold establishes which chromatographic peaks produced during organic screening require toxicological risk assessment, and how large the uncertainty factor should be has been answered empirically. Working from a database of relative response factors for nearly 1200 extractables and leachables measured by GC/MS, LC/MS with atmospheric pressure chemical ionisation, and LC/MS with electrospray ionisation, Jenke and colleagues found that an uncertainty factor of 3 correctly flags nearly 92 % of the database compounds as above the adjusted threshold ^[7]. Ninety-five per cent coverage is reached with a factor of 4 applied to the GC/MS and LC/MS combination; coverage rises to 97 % when a factor of 4 is applied to GC/MS individually and a factor of 10 to the LC/MS methods individually. The same data suggest that running both APCI and ESI in LC/MS screening, rather than either alone, gives the greatest coverage of extractables and leachables ^[7].

What the uncertainty factor is not is a safety margin on the compound. It is a correction for not knowing the compound's response factor, and it disappears once you do know it. The recommendation that follows is to base quantitation on compound-specific relative response factors collected in a database, and specifically not to use the uncertainty factor in the opposite direction to inflate reported concentrations ^[6].

Method coverage across three techniques

Assigning a method by volatility is the reflex, and it is under challenge. Li assembled 776 extractables and leachables from public sources with their molecular weights and boiling points, computed hydrophobicity as log P_o/w using an Abraham solvation parameter model, and mapped the resulting three-dimensional space ^[8]. The space corresponding to a molecular weight of 1000 Da runs to boiling points of 857 °C and log P_o/w between −2.3 and 19. A window that wide is a severe demand on reversed-phase retention and resolution, and the study concludes that the current concept of building analytical strategies around volatility needs to be challenged ^[8].

Neither GC-MS nor LC-MS sees the inorganic axis; that is a separate list needing a separate technique. ICP-MS work under ICH Q3D(R1) treats utilities, manufacturing equipment, container-closure systems and excipients as contributing sources that each have to be assessed and qualified; in one such programme, the elements of toxicological concern in multiple protein products came in well below the control threshold the guideline recommends, 30 % of the limit ^[9]. Elemental impurities have their own guide; the point here is that an organic screen returning nothing says nothing about metals.

The usual suspects, and what they turn into

The elastomer inventory is old and well characterised. A thin-layer chromatographic survey of 150 filled preparations found one or more of 2-mercaptobenzothiazole (MBT), 2-mercaptobenzothiazole disulphide (MBTS) and 2-mercaptobenzimidazole (MBI) in small-volume presentations, with one set of units also containing the hindered phenol 2,2′-methylene-bis(4-methyl-6-α-methylcyclohexylphenol) ^[10]. In syringe-cartridges, which held both the lowest and highest values, concentrations ran 8.3–13.8 µg/mL for MBT, 2.9–9.3 µg/mL for MBTS and 2.8–11.1 µg/mL for MBI. No contamination was found in the large-volume flasks of 250–1000 mL — surface-to-volume ratio doing exactly what you would predict. Two further observations travel further than the compound names: results varied between batches of the same preparation, indicating that rubber formulations or vulcanisation conditions differed, and 2-(2-hydroxyethylthio)benzothiazole was absent, which the authors read as evidence the rubbers had not been ethylene-oxide sterilised ^[10]. A sterilisation-derived reaction product is a process marker as much as a hazard.

Identification is where these studies stall. Three major extractables from rubber closures used with an injectable powder were isolated by semi-preparative HPLC combined with silica-gel chromatography and then identified using 1D NMR (¹H, ¹³C, DEPT135) and 2D NMR (COSY, HSQC, HMBC) alongside HPLC, GC-MS, ESI-MS/MS and HRMS, as N-(2-(2,2,4,4-tetramethylcyclohexyl)allyl)benzo[d]thiazol-2-amine, 2,6-di-tert-butyl-4-methylphenol and sulfur ^[11]. The first was a new compound, isolated and elucidated for the first time in that work. Two things follow: what came out of the closure was a substituted benzothiazol-2-amine rather than a parent additive, and mass spectrometry gave a formula while the structure needed isolation and NMR.

Antioxidants behave the same way. Two leachable impurities in an ophthalmic product were traced to Irganox 1010, a common antioxidant in polyolefin materials for container-closure systems, and identified by orthogonal LC-MS and GC-MS as 7,9-di-tert-butyl-1-oxaspiro[4.5]deca-6,9-diene-2,8-dione and 3-[3,5-bis(tert-butyl)-1-hydroxy-4-oxocyclohexa-2,5-dienyl]propanoic acid, both of which leached into solution during storage ^[12]. Neither is Irganox 1010. A target list built from the additive names on a supplier declaration will miss both.

Some leachables do not survive to be measured because they react. Thiuram disulfides are frequently used as accelerators in rubber stoppers for injectables and sterilised powders, and they exchange rapidly with thiols to give mixed disulfides. Corredor and colleagues formed and identified the mixed disulfides of tetramethylthiuram disulfide and tetrabutylthiuram disulfide with captopril, a thiol-containing small molecule, by reversed-phase HPLC and MS at m/z 337.1 and 420.9 respectively; formation was pH-dependent, with the captopril–TMTD adduct reaching a maximum of 31.3 % in 4.1 h at pH 6.0 and still detectable after 20 h at pH 4.0 and 2.2 ^[13]. The analyte there is a small molecule, so the kinetics do not transfer to a cysteine-containing peptide; the reaction type does. The chromatographic consequence is general: a leachable being consumed reports low as itself, and its adduct appears where you were not looking.

What this means for a peptide

Trace metal is catalytic, not stoichiometric

A microgram of iron does not consume a microgram of peptide once; it turns over. Metal-catalysed oxidation of human parathyroid hormone (1–34), a 34-residue peptide, by ferrous EDTA is site-specific, localised primarily to Met8 and His9, converting them to methionine sulfoxide and 2-oxo-histidine ^[14]. Beyond that, the same system produces hydrolytic cleavage of the backbone between Met8 and His9, a reaction requiring both Fe(II) and oxygen and preventable by diethylenetriaminepentaacetic acid and phosphate buffer; conditions leading to the cleavage also transform Met8 into homocysteine, which the authors read as indicating a common intermediate ^[14]. Metal-driven damage on a peptide is therefore not a diffuse oxidative haze that a total-oxidation assay will catch. It concentrates at a metal-binding site, and it includes backbone scission, which shows up as a new peak rather than as a mass shift of +16.

Adding a chelator is not automatically the fix. In a dithiothreitol/iron/oxygen system that oxidises Met to Met sulfoxide in the model peptides GGGMGGG and GHGMGGG, polyaminocarboxylate chelators did not suppress oxidation but significantly changed the product pattern: the yield of Met sulfoxide dropped substantially while oxidation products from the N-terminus and from His emerged, and the reaction became rather non-selective overall ^[15]. The kinetics were accelerated by nitrilotriacetate, ethylenediaminediacetate and EDTA, and slowed by EGTA and DTPA; the chelators themselves degraded, and scavengers of hydrogen peroxide or hydroxyl radicals gave only partial inhibition. The authors rationalise this as chelators extracting iron from potential peptide binding sites, impairing site-specific oxidation, while promoting reactive oxygen species different from those formed at the peptide's own metal-binding sites ^[15]. If your stability assay is tuned to methionine sulfoxide, a chelator that relocates the damage will read as an improvement.

Adsorption runs the other way across the same interface

Leaching moves material off the container into solution; adsorption moves peptide out of solution onto the container. The interface is the same one, and a recovery deficit does not say which happened. Livesey and Nicholls radiolabelled a panel of peptides and proteins with ¹²⁵I and measured competitive adsorption against human serum albumin in polystyrene or polypropylene tubes, defining stickiness from the albumin concentration required to halve adsorption ^[16]. Stickiness varied significantly between polypeptides (p < 0.00001), from 0.99 ± 0.07 for angiotensin II to 5.30 ± 0.07 for tyr⁰-urocortin II on their log scale, against 1.62 ± 0.09 for albumin and 1.92 ± 0.05 for γ globulin; no significant difference was found between polystyrene and polypropylene. They conclude that some peptides are sticky enough to risk adsorptive loss during sampling and analysis, and that peptides so sticky as to remain uncharacterised may exist ^[16]. Two consequences: sequence rather than the plastic's name dominates the loss, and a mass balance assuming that peptide missing from solution has degraded is unsafe. The adsorption guide covers mitigation.

Silicone is an interface before it is a leachable

Silicone oil applied to a syringe barrel migrates into the fill, and the peptide-relevant question is what the resulting interface does. Working with liraglutide, a lipidated 31-residue peptide, Hamada and colleagues found minimal structural change or higher-molecular-weight aggregate formation under mild agitation or limited headspace, while enhanced headspace and agitation together promoted fibrillation as followed by size-exclusion chromatography, circular dichroism and thioflavin T fluorescence ^[17]. Stimulated Raman scattering imaging resolved peptide adsorption on the silicone oil surface at submicrometre resolution, reported there for the first time, and ¹H NMR showed line broadening and signal loss consistent with aggregation or surface adsorption but no chemical shift changes, which the authors take as supporting the absence of strong, specific interactions. Their data suggest that the dual air–water and silicone-oil–water interfacial stress, rather than either alone, plays a significant role ^[17]. A silicone measurement on its own therefore does not predict the outcome, and removing silicone does not remove interfacial stress while headspace and agitation remain.

Measuring the silicone load is its own problem. Silicone oil in empty syringes or in formulated products can be extracted with organic solvents and quantified by HPLC with a universal detector, but direct extraction from a formulated protein product proved very challenging until the product was pretreated by pepsin digestion, which enabled analysis at low ppm levels ^[18]. Note the matrix: that digestion step was developed against a proteinaceous formulation, and a peptide formulation is a different matrix in which recovery has to be re-established rather than assumed.

Designing a study whose conclusions hold

Keep the two questions apart on paper. The material-level question is answered by a controlled extraction whose solvent set is declared as part of the result ^[3], with orthogonal detection ^[4]. The product-level question is answered on real units under real storage, and the answer can legitimately be that nothing is detectable ^[2]. Fix the threshold and the uncertainty factor before the screen runs, recording which factor went with which method, since the coverage differs sharply between them ^[7], and replace the factor with measured relative response factors as they arrive ^[6]. Run ICP-MS as a separate exercise on a separate list ^[9], and keep enough sample in reserve that a peak above threshold can be isolated and taken to NMR ^[11].

For a peptide, the endpoints that decide the question are on the peptide rather than in the leachate. Site-specific oxidation and backbone cleavage need a method that resolves a new peak, not only a mass shift ^[14]; a chelator's effect has to be judged on the full product distribution rather than on one marker ^[15]; recovery has to be measured against the surface before it is attributed to degradation ^[16]; and interfacial state has to be assessed with headspace and agitation held as variables in their own right ^[17].