Size-Exclusion Chromatography

In biopharmaceutical quality control, SEC (specifically SE-HPLC) is the workhorse assay for quantifying monoclonal antibody aggregation — high-molecular-weight species that elute before the monomer peak are a regulatory red flag because aggregates can be immunogenic. Every lot of a therapeutic antibody is released against a specification such as "≥ 95% monomer by SEC," making this technique one of the most frequently run analytical measurements in the entire biotech industry. In polymer manufacturing, GPC/SEC with triple detection (refractive index for concentration, light scattering for absolute $M_w$, viscometry for $[\eta]$ and branching index $g'$) characterizes everything from commodity polyethylene pellets to specialty polymers for drug-delivery hydrogels, verifying that a batch meets its molar-mass-distribution and dispersity specification before it enters a product. In structural biology, analytical SEC (often coupled in-line to multi-angle light scattering, SEC-MALS) confirms the oligomeric state of a purified protein complex before crystallography or cryo-EM, distinguishing a genuine dimer from a monomer-plus-aggregate mixture that would otherwise confound structure determination. Preparative SEC is also the final polishing step in most recombinant protein purification workflows, exchanging buffer and removing residual aggregates or small-molecule contaminants in the same run. In environmental and materials science, SEC characterizes humic substances, lignin fragments, and natural organic matter — heterogeneous, poorly defined macromolecular mixtures for which molar-mass distribution, rather than any single molecular formula, is the only meaningful descriptor, echoing exactly the polymer-dispersity problem this technique was built to solve.

:::quiz[Why is universal calibration built on the product $[\eta]M$ rather than $M$ alone?]

• Because $[\eta]M$ is proportional to hydrodynamic volume, the physical quantity SEC actually separates by ✓
• Because intrinsic viscosity is easier to measure than molar mass
• Because $[\eta]M$ always equals the Stokes radius exactly
• Because polystyrene standards only report $[\eta]M$, never $M$

Ask the tutor: How would the calibration curve and elution order change if I used a semi-rigid gel with a much narrower pore-size distribution than a standard silica GPC column?

A size-exclusion column is packed with porous particles — cross-linked dextran, agarose, or polyacrylamide gels for aqueous work; rigid silica or styrene–divinylbenzene beads for organic-phase GPC. The total volume of the packed column, $V_t$, divides into three parts: the volume of the solid matrix itself, $V_m$; the volume between particles, called the interstitial or void volume $V_0$; and the volume inside the pores, $V_i$ (sometimes written $V_p$). A molecule flowing through the column samples $V_0$ regardless of its size — every solute, however large, must travel through the interstitial channels. Whether it also samples $V_i$ depends on whether it can diffuse into the pores. Define the size-exclusion distribution coefficient $K_{SEC}$ as the fraction of the internal pore volume that is thermodynamically accessible to a given solute:

A molecule much larger than the largest pores never enters them: $K_{SEC} = 0$ and it elutes at $V_R = V_0$. A molecule much smaller than the smallest pores explores the entire pore volume just as readily as the interstitial volume: $K_{SEC} = 1$ and it elutes at $V_R = V_0 + V_i$, the total permeation volume $V_t'$. Molecules of intermediate size elute somewhere between these two limits, with larger molecules (which fit into fewer of the pores) eluting earlier than smaller ones. This is the single most counter-intuitive feature of SEC for students trained on reversed-phase HPLC: retention order is inverted relative to size — big molecules first, small molecules last — and there is no analog of a retention factor $k'$ that grows without bound, because $K_{SEC}$ is capped between 0 and 1. Because entry into a pore costs a polymer coil (or a globular protein, treated as an equivalent hydrodynamic sphere) conformational entropy — it must contract or reorient to fit — the governing thermodynamics is entropic, not enthalpic. Ideal SEC assumes $\Delta H_{transfer} \approx 0$; any residual adsorption to the packing (hydrophobic patches on silica-based GPC gels, ion-exchange sites on underivatized silanols) shows up as anomalous retention beyond $V_t'$ or as peak tailing, and is the chief systematic error to guard against.

The relationship between $K_{SEC}$ and the molecule's size depends on the pore-size distribution of the particular packing, which is why calibration must be done empirically rather than computed from first principles for most practical work. Two idealized geometric models are commonly invoked: the cylindrical pore model, where a rigid sphere of radius $r$ excludes itself from a cylindrical pore of radius $R_p$ according to

for $r \le R_p$ and zero otherwise, and empirical log-linear calibration, which simply asserts that over the useful separation range of a given column,

with $A$ and $B$ column-and-chemistry-specific constants determined by regression against standards. The log-linear form is what is actually used in the vast majority of routine GPC/SEC work, precisely because real pore-size distributions are broad and irregular enough that the cylindrical-pore formula is only a qualitative guide.

The central retention relationship in SEC:

where:

• $V_R$ = retention (elution) volume of the solute, mL
• $V_0$ = interstitial (void) volume, measured with a fully excluded marker (e.g., blue dextran, DNA, or the column-manufacturer's exclusion-limit standard), mL
• $K_{SEC}$ = size-exclusion distribution coefficient, dimensionless, $0 \le K_{SEC} \le 1$
• $V_i$ = total accessible internal pore volume, measured as $V_t' - V_0$ using a small fully-permeating marker (acetone, deuterium oxide, ¹³⁷Cs⁺), mL

Reading: this equation says elution volume is a linear interpolation between the void volume (size-independent baseline) and the pore volume, weighted by how much of the pore network the molecule can enter. It replaces the retention-factor equation $V_R = V_0(1+k)$ used in partition/adsorption modes; there $k$ can be arbitrarily large, but here $K_{SEC}V_i$ saturates at $V_i$ because it is a volume fraction, not an equilibrium ratio of amounts. The empirical calibration equation used to convert measured $V_R$ into molar mass:

where:

• $M$ = molar mass of a standard eluting at volume $V_R$, g/mol
• $A$ = intercept, the extrapolated $\log M$ at $V_R \to V_0$ (nominal exclusion limit)
• $B$ = slope magnitude (mL⁻¹), the column's selectivity — steeper $B$ means better resolving power per unit volume but a narrower usable mass range

Reading: because pore volume accessibility falls off roughly log-linearly with size over a limited range, a plot of $\log M$ against $V_R$ is close to a straight line between the exclusion and permeation limits; outside that range the curve bends sharply (all large solutes pile up at $V_0$, all small ones at $V_t'$), which is the resolution-vs-range tradeoff intrinsic to every SEC column. The universal calibration relationship, which lets one column calibration serve for any polymer chemistry:

where:

• $[\eta]$ = intrinsic viscosity of the polymer at the column temperature, dL/g
• $[\eta]M$ = the hydrodynamic volume parameter (proportional to the actual pervaded volume of the coil in solution), the quantity SEC "really" separates by
• $A', B'$ = universal-calibration constants for that column/solvent/temperature combination, obtained from a single set of standards (almost always narrow polystyrene) and applicable to any polymer for which $[\eta]$ can be measured or estimated

Reading: SEC does not literally measure molar mass — it measures hydrodynamic size. Two different polymers of the same $M$ can have very different coil sizes (a branched polymer is more compact than a linear one of equal $M$), so they elute at different $V_R$. But if you plot $\log([\eta]M)$ instead of $\log M$, all polymers collapse onto the same master curve for a given column, because $[\eta]M$ is proportional to hydrodynamic volume for any coil shape (Flory–Fox relation, $[\eta] = \Phi \langle r^2\rangle^{3/2}/M$).

:::derivation[Universal calibration: why $[\eta]M$, not $M$, is the true separating variable] Goal: derive the universal calibration relation from the physical requirement that SEC separates by hydrodynamic volume, and show it collapses different polymer chemistries onto one curve.

Step 1 — SEC retention depends on hydrodynamic volume, not chemical identity. The pore-exclusion argument of the previous derivation depended on the solute only through its size $r_s$ (equivalently, its pervaded hydrodynamic volume $V_h \propto r_s^3$). Nothing in the geometric argument referenced the polymer's chemical structure. Therefore, for any two polymers A and B that happen to have the same hydrodynamic volume at a given column/solvent/temperature, they must show the same $K_{SEC}$ and elute at the same $V_R$ — regardless of their molar masses or chemical composition.

Step 2 — Relating hydrodynamic volume to intrinsic viscosity. The Einstein–Simha viscosity relation for a dilute suspension of equivalent hydrodynamic spheres of pervaded volume $V_h$ per mole gives

(the numerical prefactor 2.5 is the Einstein coefficient for spheres; for real coils it is absorbed into a shape-dependent Flory constant $\Phi$, giving the more general Flory–Fox form $[\eta] = \Phi \langle r^2\rangle^{3/2}/M$). Either way,

so the product $[\eta]M$ is directly proportional to the hydrodynamic volume, with a universal proportionality constant independent of polymer chemistry.

Step 3 — Substituting into the calibration law. Because Step 1 established that $V_R$ is a function of $V_h$ alone (through $K_{SEC}$), and Step 2 shows $V_h \propto [\eta]M$, any calibration curve built as $\log(\text{quantity proportional to } V_h)$ vs. $V_R$ must be chemistry-independent:

Concretely, if polystyrene standards give $A'_{PS}, B'_{PS}$ by plotting $\log([\eta]_{PS}M_{PS})$ vs. $V_R$, then for an unknown polymer X eluting at the same $V_R$,

so

Step 4 — Practical use with Mark–Houwink constants. Since $[\eta] = K M^{\alpha}$ (Mark–Houwink equation, $K,\alpha$ tabulated per polymer/solvent/temperature), Step 3's relation becomes

so that

This is the operational formula used whenever GPC is calibrated with polystyrene standards but the true analyte is, say, poly(methyl methacrylate) or a branched polyolefin.

Step 5 — Limiting-case check. If X is polystyrene ($K_X=K_{PS}, \alpha_X=\alpha_{PS}$), the formula collapses to $M_X = M_{PS}$ identically — the trivial, self-consistent limit — confirming the algebra introduces no spurious chemistry-dependence when none should exist.

Example 1 (foundational) — Stokes radius of a protein from a calibration curve. A size-exclusion column is calibrated with globular protein standards of known Stokes radius $r_s$. Regression of $\log_{10}r_s$ (nm) against elution volume over the linear range gives

valid for $8.0\text{ mL} \le V_R \le 16.5\text{ mL}$ (the exclusion limit is at $V_R=8.0$ mL, total permeation at $V_R=16.5$ mL). An unknown protein elutes at $V_R = 11.35$ mL. Find its Stokes radius and comment on whether the result is trustworthy. Step 1 — check the value falls within the calibrated range. $8.0 \le 11.35 \le 16.5$ mL — yes, comfortably interior, roughly 45% of the way from exclusion to permeation. Good. Step 2 — plug into the regression line.

Step 3 — invert the logarithm.

So the unknown protein has a Stokes radius of $r_s = 9.9$ nm (two sig figs, matching the precision of the calibration slope/intercept as given). For comparison, bovine serum albumin monomer has $r_s \approx 3.5$ nm; a value near 9.9 nm signals either a much larger native oligomer or, if BSA was expected, that the sample is substantially aggregated (consistent, for instance, with the presence of BSA trimers/tetramers or nonspecific aggregates). This is exactly the kind of check a quality-control chemist performs before releasing a biologic: does the observed size match the expected oligomeric state? Limiting-case sanity check: if $V_R \to 8.0$ mL (exclusion limit), $\log_{10}r_s \to 1.20-0.144=1.056$, giving $r_s\to 11.4$ nm, close to the largest resolvable size on this column — consistent with the column's stated exclusion limit. If $V_R\to16.5$ mL, $r_s \to 10^{1.20-0.297}=10^{0.903}=8.0$ nm — note that even the "fully permeating" limit here still corresponds to a finite, nonzero size on the log scale, which is a reminder that this particular regression line is only the locally linear approximation valid inside the stated window; extrapolating it outside $[8.0,16.5]$ mL would not correctly predict $r_s\to 0$, exactly as the geometric derivation warned.

Example 2 (applied) — Molar-mass averages from a GPC chromatogram. A GPC trace of a polyethylene sample (RI detector, response $\propto$ mass concentration) is digitized into six retention-volume slices of equal width. Using the column's calibration $\log_{10}M = 10.20 - 0.310\,V_R$ ($V_R$ in mL), the slice midpoints, detector heights $h_i$ (arbitrary units, proportional to mass in that slice), and derived $M_i$ are:

Compute $M_n$, $M_w$, and the dispersity $Đ$. Step 1 — number-average molar mass. Since $h_i \propto c_i \propto n_i M_i$ (RI response is proportional to mass concentration in each slice, and $n_i = h_i/M_i$ is proportional to the number of moles in that slice),

Compute $\sum h_i = 2.1+6.8+9.4+7.2+3.0+0.9 = 29.4$. Compute each $h_i/M_i$ (units of $10^{-6}$ mol-equivalent per arbitrary unit; only ratios matter):

Sum: $2.90\times10^{-6}+1.915\times10^{-5}+5.402\times10^{-5}+8.461\times10^{-5}+7.194\times10^{-5}+4.412\times10^{-5} = 2.7674\times10^{-4}$.

Step 2 — weight-average molar mass.

Compute each $h_iM_i$ (in units of $10^{5}$ g/mol$\times$arb.):

Sum $= 1.5204+2.4140+1.6356+0.61272+0.1251+0.01836$ (all $\times10^6$) $=6.3262\times10^{6}$.

Step 3 — dispersity.

This dispersity ($Đ\approx 2.0$) is typical of a conventional (Ziegler–Natta or free-radical) polyethylene with a "most probable" (Flory) molar-mass distribution, for which theory predicts $Đ \to 2$ exactly in the high-$M$ limit — a useful internal consistency check on the whole calculation. Limiting-case check: if the distribution were perfectly monodisperse (all mass in one slice), $M_n=M_w=M_i$ and $Đ=1$ exactly, which the formulas above trivially reproduce by inspection (single term, ratio of a number to itself).

Example 3 (exam/competition level) — Universal calibration correction for a branched polymer. A GPC column is calibrated with narrow polystyrene (PS) standards in THF at 35 °C, Mark–Houwink constants $K_{PS} = 1.14\times10^{-2}$ mL/g, $\alpha_{PS} = 0.716$ (typical literature values for PS/THF at this temperature). A branched polybutadiene (PBD) sample elutes at the same retention volume as a PS standard of $M_{PS} = 2.50\times10^{5}$ g/mol. The Mark–Houwink constants for linear PBD in THF at 35 °C are $K_{PBD}=3.07\times10^{-2}$ mL/g, $\alpha_{PBD}=0.750$, but this particular sample is known to be long-chain branched, with a branching correction factor $g' = 0.72$ relating its intrinsic viscosity to that of a linear PBD of the same molar mass:

Find the true molar mass of the branched PBD sample. Step 1 — hydrodynamic volume equivalence at the shared retention volume. Universal calibration requires

Step 2 — express both intrinsic viscosities via Mark–Houwink. For polystyrene,

Compute the exponent term: $\log_{10}(2.50\times10^5) = 5.3979$; multiply by 0.716: $3.8649$; so $(2.50\times10^5)^{0.716}=10^{3.8649}=7.327\times10^{3}$.

Step 3 — hydrodynamic volume product.

Step 4 — express branched PBD's intrinsic viscosity in terms of its unknown $M_{PBD}$.

Step 5 — set the hydrodynamic volume products equal and solve for $M_{PBD}$.

Take the 1/1.750 power: $\log_{10}(9.4475\times10^8) = 8.9753$; divide by 1.750: $5.1287$; so

So the true molar mass of the branched PBD is $M_{PBD}\approx 1.35\times10^{5}$ g/mol, whereas naive use of the polystyrene calibration alone (reading $M=2.50\times10^5$ g/mol directly off the calibration curve without any correction) would have overestimated the true mass by a factor of about 1.86. This is the expected direction of error: branching compacts the coil ($g' <1$ shrinks $[\eta]$ at fixed $M$), so a branched polymer must have a larger true $M$ than the naive PS-equivalent reading only if $g'$ effects were ignored in the wrong direction — here, because both branching and the different Mark–Houwink chemistry act, the correct algebra (not intuition alone) must be trusted, which is exactly why the derivation above carries every factor through explicitly rather than estimating by inspection. Limiting-case check: if the polymer were unbranched linear PBD ($g'=1$) with PS-identical Mark–Houwink constants ($K_{PBD}=K_{PS}$, $\alpha_{PBD}=\alpha_{PS}$), Step 5 reduces immediately to $M_{PBD}=M_{PS}=2.50\times10^5$ g/mol, the trivial and necessary check that universal calibration introduces no distortion when there is genuinely no chemistry difference.

Modern SEC traces its lineage to two largely independent developments that later merged. In 1955, the British biochemist Grant Henry Lathe and Colin R. J. Ruthven, working at the Postgraduate Medical School of London, first showed that starch and later dextran gels could separate proteins by molecular size rather than by charge, laying the empirical groundwork for gel filtration. The technique was transformed into a practical, reproducible tool by the Swedish biochemist Jerker Porath and his colleague Per Flodin, who in 1959 introduced cross-linked dextran gels (commercialized as Sephadex) with controlled, reproducible pore sizes — the first packing material engineered specifically for size-based protein separation rather than adapted from an unrelated application. On the synthetic-polymer side, the technique was independently invented (and named "gel permeation chromatography") by John C. Moore at the Dow Chemical Company in 1964, who cross-linked polystyrene beads to a controlled pore structure suitable for organic solvents and demonstrated, crucially, that molar-mass distributions of industrial polymers — previously accessible only through laborious fractional precipitation — could be read off a single chromatogram in under an hour. Moore's insight that pore size, not chemical functionality, should be engineered into the packing is the conceptual core that the entire field still rests on.