Peptide APIs:Separation & Preparative Purification

SummaryTen core conclusions

1Starting materials and sources of impurities

The difficulty in preparative peptide purification lies not in the target molecule but in how closely the impurities in the crude resemble it structurally. Understanding which step generates each impurity determines which property to look for a difference in.

1.1 Three main source routes

1.2 How small the difference between impurity and main peak is

Most impurities from a synthetic route differ from the target peptide by only one residue, one acyl group or one chiral centre. For a deletion peptide, the change in hydrophobicity from one missing non-polar residue may correspond to only a few percentage points of organic phase in reversed phase; a racemate has exactly the same molecular formula, and its reversed-phase retention difference is often near the limit of detection. This is why preparative peptide work generally uses gradients of the order of 0.1 %B/min — resolution has to be bought from gradient slope and column efficiency.

Asn, asparagine

The deamidation precursor, with a neutral amide side chain

Asp, aspartic acid

The deamidation product, carrying one more negative charge — better resolved by IEX

L-Ala

Target configuration

D-Ala

The racemic impurity, with a molecular formula identical to the L form

Fig. 1　Four residue-level examples. The middle pair is deamidation before and after: a neutral amide becomes a carboxyl, changing reversed-phase retention little but altering the net charge, so ion exchange resolves it. The right pair is a pair of enantiomers: neither reversed phase nor ion exchange resolves them, so a chiral column or pre-column derivatization is needed. The structures show only the backbone and key functional groups; stereochemistry follows the original literature.

2The role of each substrate: silica, polymer and soft gel

Silica is not the only substrate for preparative chromatography. Choosing it is not a default but a consequence of the pressure, packing scale and resolution required in preparative peptide work.

The limits of a silica substrate are equally clear: a conventional bonded phase run above pH 8 for long periods hydrolyses the siloxane skeleton; where alkaline conditions are required, use a base-stable product or move to a polymer substrate. See Chapter 18.

3Six adjustable parameters of surface modification

In process terms, "modified silica" resolves into six parameters that can be specified independently. These six form the coordinate system for every selection discussion that follows.

3.1 What the surface groups actually look like

Surface silanols

The unmodified substrate, the source of hydrogen bonding and ionic interaction

C18 · ODS

The strongest hydrophobic retention; the mainstay for conventional synthetic peptides

C8 · octyl

Moderate hydrophobicity; common for lipidated peptides and insulin-type molecules

C4 · butyl

Weak hydrophobicity; for strongly hydrophobic samples and denatured proteins

Alkyl phenyl (C4Ph)

Hydrophobicity plus π–π interaction, giving selectivity for aromatic residues

Pentafluorophenylpropyl (PFP)

Dipole and configurational selectivity, for epimers and diastereomers

Aminopropyl (APS)

HILIC and normal phase, for hydrophilic peptides and carbohydrates

Trimethylsilyl endcapping

Occupies residual silanols, suppressing tailing of basic residues

Fig. 2　Eight surface groups. The grafting anchor is drawn throughout as (HO)₃Si–, the convention used across this site: in the real material that silicon is linked to the silica surface network by one to three Si–O–Si bonds, and no free trihydroxysilane exists. All structures are rendered by RDKit from SMILES and have passed molecular formula and ring count assertions. Product parameters are at the Daisogel product pageandChromatographic Silica Fundamentals。

4Deriving the separation handle

The "separation handle" is the physicochemical difference between the target peptide and its main impurities that chromatography can amplify. Fixing the separation handle amounts to fixing which class of surface modification to use.

5The reversed-phase backbone: system distribution across 59 preparative records

The statistics below come from mobile phase records for 59 preparative steps across 21 products. The basis is "this component appears in the mobile phase of this step"; one step may contain several components, so the percentages sum to more than 100%. This section gives only the system distribution and parameter ranges, not the complete gradient for each product.

Trifluoroacetic acid (TFA)

Used in 21 of 59 steps. Pairs with the peptide's basic residues to form a neutral ion pair, improving peak shape; residues form the trifluoroacetate salt and must be exchanged in a later step

Heptanesulfonic acid

An alkylsulfonic ion-pairing agent. Provides an extra hydrophobic anchor for hydrophilic short peptides, pulling the retention window back from near the dead volume into a usable range

Fig. 3　Two classes of ion-pairing agent. TFA is a volatile acid: strongly effective, but salt formation must be considered in work-up; alkylsulfonic agents are non-volatile, with a marked retention effect but a corresponding demand on downstream desalting.

6Orthogonal axes: where IEX, HILIC, SEC and normal phase fill the gaps

Reversed phase deals with most differences in hydrophobicity; the remaining impurities need a different separation handle. Which orthogonal axis to choose is determined by the type of impurity left.

The seven classes below follow the order "structural features determine the separation difficulty, and the difficulty determines the modified silica". Each gives the structural difficulty, the recommended combination of surface parameters, and the system and parameter ranges obtained from the preparative records. The Daisogel grades and specifications cited are the stated values from theproduct pageson this site.

7A · Linear lipidated metabolic peptides

Representative products　Semaglutide, tirzepatide, retatrutide, cagrilintide, liraglutide, mazdutide

7.1 Structural difficulty

What these have in common is a C16–C20 fatty diacid side chain on the backbone, usually attached through a γ-Glu and AEEA linker. The lipid chain has two consequences: the molecule is far more hydrophobic than an ordinary peptide of the same length and is over-retained on C18; and it tends to self-associate in aqueous solution, forming aggregates at the column head and within the bed, showing as peak broadening, reduced loading and unstable recovery. The impurities closest in properties to the target are mono- and di-acylation by-products, des-AA deletion peptides and lipid chain positional isomers, all three differing very little from the target in polarity.

Octadecanedioic acid

The acyl part of the lipidated side chain. One end forms an amide to the linker, the other retains a free carboxyl

γ-Glu–AEEA linker fragment

Connects the lipid chain to the lysine side chain of the backbone, with two ethylene glycol units providing flexibility and a hydrophilic balance

Fig. 4　The two parts of the lipidated side chain. The hydrophobic end and the hydrophilic linker together determine how these molecules behave in reversed phase: the hydrophobic end provides strong retention while the linker reduces the tendency to aggregate.

7.2 Modified silica selection

7.3 Process notes

8B · Disulfide and cyclic disulfide peptides

Representative products　Linaclotide, octreotide, bivalirudin, etelcalcetide, somatostatin, terlipressin, carbetocin

8.1 Structural difficulty

Peptides with one to three disulfide bonds generate scrambled disulfide isomers during oxidative folding: identical in sequence, differing only in disulfide topology, with the same molecular formula and almost identical hydrophobicity. These impurities are both hard to resolve and liable to keep forming during chromatography — neutral to alkaline conditions and transition metals both promote on-column disulfide exchange. The process constraint is therefore not only resolution but "do not create new impurities".

Cystine · disulfide bridge

Two cysteine residues joined by S–S. With several cysteine pairs on one chain the pairing is not unique, and the wrong pairing gives a scrambled isomer

Substrate surface

Transition metals (Al, Fe, Ti, Zr) residual in the silica substrate can catalyse disulfide exchange; ultra-high-purity substrates keep the total below 10 ppm

Fig. 5　The two concerns with disulfide peptides: the disulfide bridge itself, and the metals in the substrate that may catalyse its exchange.

8.2 Modified silica selection

8.3 Process notes

9C · Macrocyclic and synthetic macrocyclic peptides

Representative products　Oral PCSK9 macrocyclic peptides, complement inhibitor macrocycles, synthetic macrocycles containing D-amino acids

9.1 Structural difficulty

A macrocyclic backbone restricts conformational freedom, giving two characteristic impurity types: conformational isomers, which interconvert slowly on the chromatographic timescale and therefore appear as several peaks; and cyclic dimers and epimers. Oral products carry a further constraint: the limits on deletion peptides and potential genotoxic impurities are usually stricter, and the method must be MS-compatible for on-line confirmation.

9.2 Modified silica selection

9.3 Process notes

• System　The reversed-phase aqueous component is low-concentration formic acid or TFA. The recorded racemate separation used a low-pH ammonium salt aqueous phase on a pentafluorophenyl column, run isocratically at low organic — isocratic rather than gradient, which indicates the separation relies on selectivity rather than eluting strength. Record statistics
• Column temperature　Configurational selectivity is temperature sensitive, and a small change in temperature can widen or narrow the resolution of epimers; this is the main optimization variable for this class.
• Orthogonality　Reversed phase resolves configuration and normal phase resolves conformation; the two correlate weakly, which suits them to cross-checking release purity.

10D · Peptide radioconjugates

Representative products　¹⁷⁷Lu-labelled somatostatin receptor and PSMA-targeting peptide conjugates

10.1 Structural difficulty

These molecules comprise a targeting peptide, a linker and a chelator (of the DOTA class), and purification faces three constraints at once: the chelator is highly sensitive to trace metals, and any metal from the substrate, reagents or containers occupies coordination sites and lowers specific activity; the radionuclide half-life is limited, so everything after labelling must be completed within hours; and purification of the cold precursor and purification after labelling are two different kinds of step.

10.2 Modified silica selection

10.3 Process notes

• Labelling buffer　A sodium acetate system at weakly acidic pH, with an antioxidant to suppress radiolysis. Concentration and temperature are set by the radionuclide and chelator.Industry consensus
• Time budget　The whole sequence from labelling to release must fit within the window allowed by the radionuclide half-life, so chromatographic steps should be as short and as few as possible.
• Metal removal　Metal leaching from glassware, corrosion products from stainless steel tubing and the metal background of mobile phase reagents must each be assessed separately.

11E · Peptide–protein and PEG conjugates

Representative products　Antibody–peptide conjugates, PEGylated complement inhibitor peptides, peptide–Fc fusions

11.1 Structural difficulty

The finished product is a biomacromolecule, and the quality attributes shift from "peptide purity" to conjugation degree distribution, PEG polydispersity, aggregates, host protein and endotoxin. The peptide chemistry exists only before conjugation: the peptide fragment itself still goes through reversed-phase preparative work, while purification after conjugation moves into the protein process paradigm.

11.2 Modified silica selection

12F · Fermentation-derived peptides and lipopeptides

Representative products　Micafungin, telavancin and daptomycin-class cyclic lipopeptides

12.1 Structural difficulty

Fermentation-derived material has two characteristics: the lipid chain homologue profile varies between batches, with homologues differing by only one or two carbons, so reversed-phase retention differences are small but numerous; and the broth itself contains pigments, endotoxin and a large impurity load, so the burden on the crude is far higher than for a synthetic peptide. Some products also have characteristic chemical instability, requiring limits on temperature, light and pH.

12.2 Modified silica selection

12.3 Process notes

• System　The aqueous phases recorded for this class are mainly low-concentration ammonium salts or organic acids, with single steps running to 90 min or more and main-segment slopes of the order of 0.14 %B/min; there are also two-step processes that switch both the acid and the strong eluent at once. Record statistics
• Stability limits　Products prone to isomerization need defined temperature limits, protection from light and pH control, with the limits set by that product's stability studies.To be measured
• Batch variation　Variation in the homologue profile of fermentation material requires margin in the process, usually absorbed by multidimensional chromatography and dynamic adjustment of the collection window.

13G · Short peptides, D-amino acid and hydrophilically modified peptides

Representative products　Thymopentin, thymalfasin, difelikefalin, gonadorelin, carbetocin, synthetic peptides containing several D-amino acids

13.1 Structural difficulty

Three separate problems: short and hydrophilic peptides are barely retained on conventional C18 and co-elute with salt; products containing D-amino acids have diastereomers with the same molecular formula and similar hydrophobicity; and short peptides are hard to desalt, because their retention window overlaps that of the salt.

13.2 Modified silica selection

13.3 Process notes

• System　Two approaches were recorded for this class: pushing the pH to the alkaline side to enhance retention differences between short peptides, with main-segment slopes as low as the order of 0.08 %B/min; and keeping the same acid system across both steps while moving the pH from acidic to near neutral, achieving orthogonality through the change in net charge. Record statistics
• Column temperature　Diastereomers and hydrophilic peptides are often separated at a lower column temperature, to amplify the selectivity difference.Industry consensus
• Ion-pair residues　Non-volatile ion-pairing agents create a downstream desalting burden; confirm the removal scheme before choosing one.

14Master matrix: seven structural classes × eight dimensions

The grades in the table are starting points matched on parameters, not the only answer. Actual selection should take account of crude purity, the pressure rating of the available equipment, the target loading and the work-up scheme, and be confirmed at bench scale. Full specifications are at the Daisogel product page, and documentation at 。

Read across the seven classes and the selection rules converge to four: pore size scales with molecular size, chain length is chosen by the separation difficulty rather than by hydrophobicity alone, endcapping and bonding density set the balance between peak shape and loading, and metal content and pH tolerance are prerequisites that decide whether a medium can be used at all.

15The pore size ladder: from 60 Å to 2000 Å

The purpose of pore size is to let the target move freely in and out of the pores. When the molecule's hydrodynamic radius approaches the pore size, intraparticle diffusion is impeded, showing as peak broadening and reduced recovery; when the pore is far larger than the molecule, surface area falls geometrically and loading falls with it. Bigger is therefore not better; the pore should just cover the molecule.

The surface areas are the ranges stated on the Daisogel product pages; different grades at the same pore size differ.Manufacturer data

16Alkyl chain length and carbon load

Chain length determines the type and strength of hydrophobic retention, and carbon load determines the depth of retention and the loading at a given chain length. The two must be read together: a low carbon load C18 does not necessarily retain more strongly than a high carbon load C8.

17Endcapping, bonding density and residual silanols

The bonding reaction cannot cover every silanol on the silica surface, and what remains are the "residual silanols". Endcapping uses a small silylating reagent after bonding to occupy those sites. Whether to endcap is a directional choice in peptide purification.

Endcapping: trimethylsilyl occupying the site

Residual silanols are occupied by a small silane and ionic interaction is suppressed

Non-endcapped C4

Butyl provides weak hydrophobicity while the neighbouring unoccupied silanols provide hydrogen bonding, the two combining into a mixed mode

Fig. 6　The difference between endcapped and non-endcapped is what remains on the surface besides the alkyl chain. The process criterion is whether the separation needs that hydrogen bonding interaction.

18Metal content and pH tolerance

18.1 Metal content

The Daisogel "P" grade denotes ultra-high-purity silica, with total Al, Fe, Ti and Zr below 10 ppm; the BIO and HSA series, although not marked P, also use an ultra-high-purity silica substrate. The manufacturer analyses metal impurities by XRF and ICP and provides the data to customers.Manufacturer data

18.2 pH tolerance and column regeneration

A conventional bonded silica normally runs at pH 2–8. The Daisogel BIO series product page states a pH stability range of 1.5–11, regenerable by flushing with aqueous NaOH; the alkali tolerance data given for the PK series is that the bare silica withstands more than 170 column volumes of dilute NaOH flushing before the particles dissolve.Manufacturer data

19Gradient, column temperature and loading

19.1 Gradient form: three segments

The 59 preparative steps collected are highly consistent in gradient form, which resolves into three segments:

The above are summary statistics from the preparative records of 21 products; complete methods for individual products are not published. Gradient design for a new product can start from these ranges, but the actual slope must be set by the resolution obtained at bench scale. Record statistics

19.2 Column temperature

Column temperature is a two-way variable in preparative peptide work: raising it improves mass transfer, lowers viscosity and column pressure, and improves peak shape; lowering it suppresses on-column chemistry (disulfide exchange, isomerization, epoxide opening) and may amplify configurational selectivity. The direction is set by the on-column stability of the target peptide, not by mass transfer efficiency alone.

The exact temperature must be established from the product's on-column stability studies; no general figure is given here.To be measured

19.3 Loading

Loading depends on crude purity, target purity, the acceptable yield loss and the surface area of the medium; it is not an inherent constant of the medium. The usable loading on the same column can differ several fold between crude at 50% and at 75% purity. In practice the inflection point is found from a bench-scale loading curve, then scaled linearly by column volume, with margin left for industrial batches. To be measured

20From the analytical column to the industrial DAC column

A substantial part of the resolution in preparative peptide work comes from the quality of the bed rather than from the medium. The same lot of medium can give very different resolution in a uniform bed and in a non-uniform one, and this becomes more pronounced at large column diameters.

21Structural class → Daisogel grade mapping

This chapter maps the parameter conclusions above back onto the Daisogel specifications available on this site. The parameters listed are the stated values from the product pages, and grades are read under the coding rule "SP-pore size-particle size-bonded phase-grade". Manufacturer data

21.1 The Daisogel coding rule at a glance

"P" denotes ultra-high-purity silica with total Al, Fe, Ti and Zr below 10 ppm; the BIO and HSA series, although not marked P, also use ultra-high-purity silica. The full rule and all specifications are at the Daisogel product page。

21.2 Grades by structural class

21.3 Supporting product lines

22One-page reference and data basis

22.1 One-page reference: from symptom to action

22.2 Data basis

22.3 Scope

• The parameter statistics in this text are based on 21 products, mostly synthetic peptides with some fermentation peptides, and cannot represent every peptide product; the slope and duration ranges are a starting point for design, not a specification.
• The seven structural classes are divided by separation difficulty, and one product may fall into two classes at once (both a disulfide peptide and containing D-amino acids, for example); in that case the separation handle for the first step should be set by the impurity present at the highest level or with the strictest limit.
• The grades listed are starting points matched on parameters and do not constitute a process recommendation for any particular product; actual selection must take account of the properties of the crude, the equipment available and the regulatory requirements, and be confirmed at bench scale.
• The pentafluorophenyl phase is not in the current Daisogel range; what is said about it here comes from published literature and one instance in the preparative records, and it can go through a bonded phase customization assessment if required.

AppendixPrincipal source categories