Week 8 — Nov 2: Separation Methods — Theory Through Proof of Performance
Everything separations, in one session: the theory behind every separation (retention, selectivity, efficiency, resolution, van Deemter), the workflow that turns an analytical target profile into a validated LC method, forced degradation and the stability-indicating method, system suitability as the running proof a method still works, the wet-chemistry workhorses (Karl Fischer, titrimetry, ion chromatography), and how the same separation logic carries into biologics and advanced therapies.
A one-page overview graphic for this week is still to be produced.
(Lecture 7. The midterm is behind us.) Most real pharmaceutical questions aren’t answered by a measurement that needs no separation first — the sample is a mixture, and the answer depends on pulling it apart before anything gets measured. Chromatography is how that pulling-apart happens, and it sits behind the large majority of assay, impurity, and identity methods in a small-molecule QC lab. This session covers the whole arc in one sitting: the theory, the method-development workflow, proving a method keeps working over a product’s shelf life, the wet-chemistry techniques that round out everything a separation doesn’t cover, and the same logic doing heavier work on biologics and advanced therapies.
The one idea
A separation is a controlled competition: every component partitions back and forth between a stationary and a mobile phase, and small reproducible differences in how long each stays stuck are amplified, over a column, into baseline resolution. A validated method proves that competition can resolve what needs resolving; system suitability proves it is resolving it, now, before any sample result is trusted — and the physics doesn’t change with molecular weight: a 150,000-dalton antibody is still just partitioning between two phases, it’s just that “purity” now needs a panel of separations to answer instead of one.
What we cover this week
| # | Section | The one idea |
|---|
| 1 | Chromatographic Theory | Retention factor, selectivity, efficiency, and resolution: the four numbers that describe a peak pair, and the van Deemter equation that explains why peaks are as wide as they are — plus the modes (reversed-phase, HILIC, ion exchange, size exclusion, chiral, and TLC) and detectors that make up an LC method. |
| 2 | LC Method Development | The workflow from analytical target profile to a validated, robust method under ICH Q14, and why the search is mostly a search for selectivity. |
| 3 | Stability-Indicating Methods | Forced degradation, resolving every degradant with margin, specificity by DAD peak purity, mass balance, and system suitability as the running proof — plus the worked case in the specificity trap. |
| 4 | The Wet-Chemistry Workhorses | Karl Fischer titration, other titrimetry, and ion chromatography — the tests that round out almost every small-molecule specification without an optical or mass detector. |
Applied case — separations in biologics and advanced therapies
The same separation science extends directly to larger, more heterogeneous molecules, where it usually has to work harder because “the molecule” is really a population of related variants:
| Molecule | Method | What it separates | What can go wrong |
|---|
| Monoclonal antibody — charge variants | Imaged capillary isoelectric focusing (icIEF), cation-exchange chromatography (CEX) | Species differing by charge — deamidation, C-terminal lysine, sialylation, glycation | A shift in the charge-variant profile after a process change is often the first sign something upstream moved — see Week 10’s worked case, where mass spec runs it down |
| Monoclonal antibody — size, purity | CE-SDS (reduced and non-reduced) | Fragments and clips versus intact assembly | Reduced CE-SDS can hide a disulfide-linked aggregate that non-reduced CE-SDS would catch — the two runs answer different questions |
| Oligonucleotides (ASO, siRNA) | Ion-exchange and reversed-phase HPLC | Sequence-related impurities: n−1, n+1, depurination | These are the bridge case between small and large molecule — solid-phase-synthesized like a small molecule, but resolved and characterised like a biologic |
These are the same retention, selectivity, and resolution concepts from earlier this session — capillary electrophoresis separates by electrophoretic mobility in a buffer-filled capillary rather than partitioning on a packed column, but the goal (baseline-resolve closely related species) and the failure modes (poor resolution, migration-time drift, sample-matrix effects) are the same conversation in a different geometry.
Where the analyst sits
The four-number table in Chromatographic Theory looks like a formula you plug numbers into. The judgment is in choosing which lever to pull — a hard separation is almost always a selectivity problem wearing an efficiency-sized bill — and development software will optimise a separation against whatever critical pair you give it, which makes choosing the right critical pair the actual skill. Once a method exists, “the chromatogram looks fine” is not a sentence a reviewer accepts — you will be asked to point to the specific system-suitability numbers that prove it. And system suitability itself has a blind spot: it can only watch the critical pair someone already identified, on a small molecule or, at larger scale, on a biologic’s charge-variant or size panel. A method that has never been challenged with a known-defective batch or lot hasn’t earned its trust yet, no matter how many it has passed. That is the STEAM “A”. The refrain: science → evidence → reduced uncertainty → control → regulatory confidence → patient trust.
On the job
- Development software will optimise a separation against whatever critical pair you give it — recognising which pair actually matters, often from forced-degradation data you don’t have yet, is the judgment call nobody automates.
- Your first method-development task is more likely to be executing someone else’s scouting plan than designing one — know how to read a design-of-experiments robustness study before you’re asked to build one.
- Karl Fischer titration is one of the most-run tests in a QC lab and one of the easiest to get subtly wrong (reagent titer drift, sample introduction technique) — a common early competency check.
- “The model still fits” and “the method is still valid” are not the same claim — system suitability is what actually stands between a running method and a wrong result.
- A charge-variant or CE-SDS shift after a manufacturing change lands on an analyst’s desk as a triage problem first: is this attribute one that would actually affect safety or efficacy, or a chemically explained, clinically silent difference?
For discussion
- You can double N by doubling column length (and run time), or improve α from 1.05 to 1.10 by changing the mobile-phase pH. Which gains more resolution, and why is that the general rule?
- Q14 frames method development as “designed against validation targets from the start.” What would a method developed the old way — separation first, validation after — be likely to get wrong?
- A method passes robustness testing at every DoE point you tested, but fails in a receiving lab during transfer. What does that tell you about the DoE design, and what would you change?
- System suitability passed on a run that we later learned gave a wrong result. Was the test inadequate, or is this an inherent limit? What would you add?
- Karl Fischer and loss-on-drying give different numbers for the same sample. Which is right, and what does the difference tell you?
- Reduced and non-reduced CE-SDS give different purity numbers for the same mAb lot. Which is “right,” and what does the difference tell you about the sample?
Source note. Each section carries its own source note for the literature and compendial chapters it covers. The applied biologics/ATMP separations case connects to ICH Q6B and the charge-variant and CE-SDS methods discussed in the biopharmaceutical analytical literature. (Instructor: this session now absorbs what were two separate lecture weeks — confirm the pacing works in a single 3-hour slot.)
1 - Chromatographic Theory — Retention, Selectivity, Efficiency, Resolution
The four numbers that describe a peak pair — retention factor, selectivity, efficiency, resolution — and why resolution scales the way it does; the van Deemter equation and what it says about particle size and UHPLC; and the modes and detectors that make up an LC method.
A one-page overview graphic for this section is still to be produced.
This week named the one idea behind every separation: a controlled competition between a stationary and a mobile phase. This page puts numbers on that competition — the four quantities that describe whether two peaks come apart, and the physics that decides how wide each peak is to begin with.
The one idea
Once retention is in a reasonable range, chasing more theoretical plates has square-root returns; a small gain in selectivity moves resolution a lot. Method development is mostly a search for selectivity.
The four numbers that describe a peak pair
| Quantity | Symbol | Controlled by | What it does |
|---|
| Retention factor | k | Mobile-phase strength, stationary phase | Retention relative to an unretained marker; aim for k ≈ 2–10 |
| Selectivity | α | Stationary-phase chemistry, mobile-phase pH and modifier, temperature | The ratio of two components’ retention — the strongest lever for a hard separation |
| Efficiency | N (plates) | Particle size, column length, flow, viscosity | How narrow the peaks are |
| Resolution | Rs | All of the above | The actual separation; Rs ≥ 1.5 is baseline. Roughly Rs ∝ √N · (α−1)/α · k/(1+k) |
Practical reading: you can double N by doubling the column length (and the run time), or you can improve α from 1.05 to 1.10 with a change in mobile-phase pH — and the second move usually buys far more resolution, for less cost, than the first.
Van Deemter — why peaks are as wide as they are
Plate height H = A + B/u + C·u vs. linear velocity u: eddy diffusion (A, reduced by smaller particles), longitudinal diffusion (B/u, rarely limiting in modern LC), and mass-transfer resistance (C·u, flattened by sub-2-µm and core–shell particles). That last term is the whole case for UHPLC: the same resolution in a fraction of the time, at the cost of back-pressure and tighter demands on system dispersion.
Modes and detectors
Modes: reversed-phase (the default, separates on hydrophobicity), HILIC (very polar analytes), ion exchange (charge), size exclusion (hydrodynamic size), chiral (stereochemistry). Detectors: UV/diode array (the workhorse; DAD gives peak purity), fluorescence, refractive index, ELSD/CAD, and mass spectrometry (covered in its own week). TLC is not obsolete — cheap, parallel, and still compendial for many identity tests.
Where the analyst sits
The four-number table looks like a formula you plug numbers into. The judgment is in choosing which lever to pull: a hard separation is almost always a selectivity problem wearing an efficiency-sized bill, and reaching for a longer column before trying a different pH or stationary phase is the single most common wasted afternoon in method development.
For discussion
- You can double N by doubling column length (and run time), or improve α from 1.05 to 1.10 by changing the mobile-phase pH. Which gains more resolution, and why is that the general rule?
- UHPLC trades back-pressure and tighter system-dispersion requirements for the mass-transfer gains of smaller particles. What would make you decide a method is not a good UHPLC candidate?
- A reversed-phase separation of two very polar, poorly retained analytes keeps failing to reach k ≈ 2. What would you try before concluding reversed-phase is the wrong mode?
Source note. Chromatographic theory follows Snyder, Kirkland & Dolan, Practical HPLC Method Development, and Harris. Compendial basis: USP ⟨621⟩, ⟨1058⟩; Ph. Eur. 2.2.46.
2 - LC Method Development
The LC method-development workflow: defining the analytical target profile, scouting for selectivity, optimising, robustness (DoE) and the method operable design region, validation, and transfer — with each step tied back to ICH Q14 and Q2(R2).
A one-page overview graphic for this section is still to be produced.
Chromatographic theory named the levers — retention, selectivity, efficiency — that decide whether two peaks resolve. Method development is the disciplined process of pulling those levers on a real sample, in an order that doesn’t waste weeks chasing the wrong one.
The one idea
Nobody hands a method developer a finished separation to optimise. They hand you an analytical target profile — what has to be quantified, at what level, with what accuracy — and the column, mobile phase, and gradient are all still open questions.
The method-development workflow
| Step | What happens | Ties to |
|---|
| Define the ATP | What must be quantified, at what level, with what accuracy/precision — before a column is chosen | Q14 |
| Scout | Orthogonal phases, pH, modifier, temperature — a coarse grid for selectivity | Chromatographic Theory |
| Optimise | Gradient slope, temperature, pH; resolve the critical pair with margin | |
| Robustness (DoE) | Vary the factors that drift in a real lab; map the method operable design region (MODR) | Q14, Q9 |
| Validate | Specificity, linearity, range, accuracy, precision, LOD/LOQ | Q2 |
| Transfer & monitor | Same answer in every receiving lab; monitor and revise over its life | Week 1 · QC |
The order matters. Scouting for selectivity before optimising efficiency is the direct application of the square-root-returns lesson: a coarse screen across phases, pH, and modifier finds a workable α far faster than iterating on gradient shape ever will.
The analytical target profile, and why it comes first
An ATP states the requirement, not the method: the analyte(s), the matrix, the concentration range, and the accuracy and precision the result must deliver — deliberately silent on column chemistry or gradient. Under ICH Q14, this is the analytical-QbD starting point: the method is designed against the ATP and its later validation targets, rather than developed first and validated as an afterthought. Skipping this step is the most common reason a method later fails robustness testing — it was optimised against “does it separate,” not against “does it meet the requirement the specification actually needs.”
Robustness and the MODR
A method that resolves the critical pair once, under one set of conditions, has not been shown to be robust. Design of experiments (DoE) deliberately varies the factors that drift in a real lab — pH, column temperature, flow rate, mobile-phase composition, column lot — to map the method operable design region (MODR): the multidimensional space of conditions where the method is proven to keep working. Inside the MODR, a small drift is expected performance, not a deviation; outside it, the method needs to be requalified.
Where the analyst sits
Development software will optimise a separation against whatever critical pair you give it. Choosing the right critical pair — the two components most likely to co-elute, not just the two that happen to be hardest to resolve today — is analytical judgment, and it’s usually informed by what forced degradation turns up, not by the software’s own optimisation run.
For discussion
- An ATP specifies accuracy and precision but says nothing about run time. Who decides how much run time is acceptable, and on what basis?
- A method passes robustness testing at every DoE point you tested, but fails in a receiving lab during transfer. What does that tell you about the DoE design, and what would you change?
- Q14 frames method development as “designed against validation targets from the start.” What would a method developed the old way — separation first, validation after — be likely to get wrong?
Source note. Method-development workflow follows Snyder, Kirkland & Dolan, Practical HPLC Method Development. Regulatory basis: ICH Q14 (analytical procedure development) and ICH Q2(R2) (validation).
3 - LC Stability-Indicating Methods
Building and proving a stability-indicating method: forced degradation, resolving every degradant with margin, specificity by DAD peak purity, mass-balance as a check on what you might be missing, and system suitability as the running proof that a validated method is still working — plus a worked case in the specificity trap.
A one-page overview graphic for this section is still to be produced.
A method developed against today’s known impurities is not automatically ready to watch a product over its shelf life. A stability-indicating method has to resolve the API from degradants that don’t exist yet at release — and to keep proving, run after run, that it still can.
The one idea
A validated method proves the separation can work; system suitability proves it is working, now, before any sample result is trusted.
The stability-indicating method
- Force degradation (acid, base, oxidation, heat, humidity, light) to generate the degradants the method must see.
- Resolve every degradant from the API and from each other, with margin.
- Prove specificity — DAD peak purity on the API; confirm with an orthogonal method or LC–MS.
- Check mass balance — assay loss should equal the sum of degradation products; a gap means a degradant you are not seeing.
- Lock system suitability around the real critical pair.
This is the analytical machinery behind ICH Q1: a stability program is only as good as the method’s ability to actually see what’s changing.
System suitability — the running proof
A validated method proves the separation can work; system suitability proves it is working, now, before any sample result is trusted: resolution of the critical pair, tailing factor, plate count, retention reproducibility, replicate-injection %RSD, and S/N at the reporting threshold. Fail it and no data from that run is usable — regardless of how good the method looked in validation.
Worked case — the specificity trap
A stability-indicating assay reports 99.1% — in spec, batch released. Two years later a longer-gradient, different-selectivity method finds a degradation product had been co-eluting under the API peak the whole time; the true assay was 96.8% and a specified impurity was over its limit. Nothing looked wrong — system suitability was built around the known critical pair, and this degradant wasn’t in it. What would have caught it: DAD peak purity, an orthogonal method in development, and mass balance — the assay loss didn’t match the sum of impurities. A separation’s most dangerous failure mode is the impurity it was never designed to resolve.
Where the analyst sits
“The chromatogram looks fine” is not a sentence a reviewer accepts — you will be asked to point to the specific system-suitability numbers that prove it. And system suitability itself has a blind spot: it can only watch the critical pair someone already identified. The specificity trap above is what happens when that identification was wrong, or went stale as the process changed.
For discussion
- System suitability passed on a run that we later learned gave a wrong result. Was the test inadequate, or is this an inherent limit? What would you add?
- Mass balance “should” close to 100%, but real methods often report 97–102% even when nothing is wrong. How would you decide whether a mass-balance gap is real or just measurement uncertainty?
- Forced degradation is normally done once, early in development. What would make you decide a method needs to be re-challenged with fresh forced-degradation samples later in its life?
Source note. Stability-indicating method development follows Snyder, Kirkland & Dolan, Practical HPLC Method Development, and ties directly to ICH Q1 and ICH Q3. Compendial basis: USP ⟨621⟩, ⟨1225⟩.
4 - The Wet-Chemistry Workhorses — Karl Fischer, Titrimetry, Ion Chromatography
The tests that round out a small-molecule specification without an optical or mass detector: Karl Fischer titration for water content, loss on drying and residue on ignition, potentiometric titration for acid/base and counterion content, and ion chromatography — a genuine chromatographic separation for the small ions LC–UV can’t see.
A one-page overview graphic for this section is still to be produced.
Not everything on a small-molecule specification is an LC method. These tests are on almost every one of them, and — Karl Fischer especially — among the most-run assays in a QC lab.
The one idea
These four tests answer questions an LC–UV method structurally can’t: how much water, how much total volatile content, how much inorganic residue, and how much of a small ion that never gives a useful chromophore.
The wet-chemistry workhorses
| Test | Method | Measures |
|---|
| Water content | Karl Fischer (volumetric / coulometric) | Water, specifically — not total volatiles (that’s LOD) |
| Loss on drying (LOD) | Gravimetric | Total volatiles |
| Residue on ignition / sulfated ash | Gravimetric | Inorganic residue |
| Assay / content of a salt or counterion | Potentiometric titration; ion chromatography | Acid/base content; specific counterions and small ions |
Why ion chromatography belongs here, and belongs to chromatography
Ion chromatography separates ions on a column by ion-exchange retention, exactly the competition between stationary and mobile phase that opened this session — it just detects by conductivity rather than UV, and it resolves species (chloride, sulfate, small organic acids, counterions) that give an LC–UV method little or nothing to see. It sits in this “wet chemistry” group by purpose — it’s finishing the same job as a titration — while sitting in chromatography by mechanism.
Karl Fischer — the fine print
Volumetric KF suits higher water levels; coulometric KF reaches much lower levels (ppm) without a burette. Either way, the result is only as good as the titer (KF reagent’s water-equivalence factor), which drifts and must be checked routinely, and the sample-introduction technique — a sample that isn’t fully dissolved or that absorbs atmospheric moisture during handling will bias the result before the titration even starts.
Where the analyst sits
Karl Fischer titration is one of the most-run tests in a QC lab and one of the easiest to get subtly wrong — reagent titer drift and sample-introduction technique are common early competency checks, not edge cases. And a discrepancy between Karl Fischer and loss-on-drying is not a contradiction to explain away: they are measuring different things, and the gap between them is itself informative about what else is volatilising.
For discussion
- Karl Fischer and loss-on-drying give different numbers for the same sample. Which is right, and what does the difference tell you?
- A coulometric Karl Fischer result drifts upward over a week of use with no change in sample type. What would you check first?
- Why is ion chromatography, and not LC–UV, the default method for a chloride or sulfate counterion assay?
Source note. Compendial basis: USP ⟨921⟩ (water determination), ⟨281⟩ (residue on ignition), ⟨731⟩ (loss on drying); Ph. Eur. 2.5.12 (water: semi-micro determination), 2.2.47 (ion chromatography).