Integral and Peripheral Membrane Proteins

A bacteriorhodopsin molecule sits in the purple membrane of Halobacterium salinarum. Seven α-helices pack into a compact bundle. It absorbs a photon and flips a retinal chromophore from 13-cis to all-trans. This happens in under a picosecond. That single conformational twitch is relayed through the helix bundle. This opens a proton-conduction pathway. It pumps H⁺ against a gradient. Ultimately, this synthesizes ATP in the dark-adapted desert sun. None of this chemistry is possible without a protein that has solved a problem. No water-soluble enzyme ever faces this problem: how to fold, fully functional, inside a slab of hydrocarbon roughly 30 Å thick. Its polar catalytic machinery stays buried at its core. Its greasy, featureless exterior faces the lipid bilayer. Membrane proteins constitute an estimated 20–30% of all open reading frames in sequenced genomes. They are the targets of over half of all approved pharmaceuticals. Yet they remain the most underrepresented class in the Protein Data Bank relative to their abundance. This is a direct consequence of the same amphipathic paradox that makes them functionally indispensable. Every membrane protein must reconcile two contradictory chemical environments. One is a bilayer interior of ε ≈ 2–4, comparable to olive oil. The other is an aqueous exterior of ε ≈ 80. Between them lies a headgroup interfacial zone of steep, non-uniform polarity. The thermodynamics of partitioning a polypeptide across this discontinuity is the unifying chemical problem. This article addresses that problem, rather than any single "membrane protein fold." This article classifies the strategies life uses to solve that problem. Integral membrane proteins embed permanently. They are built from continuous stretches of hydrophobic sequence. These are organized either as bundles of α-helices or as closed β-barrels. These two folds satisfy the hydrogen-bonding requirements of the peptide backbone. They do so by entirely different geometric tricks. Peripheral membrane proteins never cross the bilayer at all. Instead, they dock reversibly onto its surface. They use amphipathic helices, electrostatic contacts with anionic headgroups, or specific lipid-binding domains. A third strategy is covalent lipid anchoring. This lets an otherwise soluble protein moonlight as a membrane protein. It does so by splicing on a fatty acyl chain, a prenyl group, or a glycosylphosphatidylinositol (GPI) anchor post-translationally. By the end you will be able to read a hydropathy plot. You will predict transmembrane topology from it. You will explain why β-barrels are confined to outer membranes. You will also explain why α-helices dominate everywhere else. You will compute the free-energy cost of inserting a charged residue into the bilayer core. This uses the Wimley–White hydrophobicity scale. You will place lipid-anchoring chemistry in its correct biosynthetic and functional context. This includes myristoylation, palmitoylation, prenylation, and GPI attachment. This foundation underlies Membrane Transporters and Active Transport, Ion Channels and Selective Permeability, and Passive Diffusion and Membrane Permeability. All of these depend on the architectural principles developed here.

Key takeaways

Integral membrane proteins span the bilayer using α-helical bundles (most membranes) or β-barrels (bacterial/mitochondrial/chloroplast outer membranes only).
The α-helix satisfies all backbone H-bonds internally, letting a ~20-residue hydrophobic helix cross ~30 Å of hydrocarbon with no polar cost.
β-barrels satisfy backbone H-bonding via a closed, hydrogen-bonded antiparallel sheet — geometrically demanding but chemically robust to the lack of a chaperoned insertion machinery inside the periplasm.
Peripheral proteins bind reversibly via amphipathic helices, electrostatic/anionic-lipid contacts, or curvature-sensing wedge insertion — no bilayer-spanning segment required.
Lipid anchors (myristoyl, palmitoyl, prenyl, GPI) covalently tether otherwise soluble proteins to one leaflet, often reversibly and regulated by acylation/deacylation cycles.
Insertion free energies are quantitatively predicted by hydrophobicity scales (Wimley–White, Hessa–von Heijne biological scale) built on the translocon's own amino-acid discrimination.
Key idea: Membrane protein architecture is dictated by how a polypeptide backbone can fully hydrogen-bond in a low-dielectric medium — every fold, anchor, and topology rule follows from that single constraint.

Where this lives today

The distinction between integral and peripheral membrane association underlies rational drug design at every level: G-protein-coupled receptors (GPCRs), the largest class of drug targets, are polytopic seven-TM-helix integral membrane proteins whose helix-packing rearrangements upon agonist binding are now routinely captured by cryo-EM, and structure-based drug design explicitly exploits predicted TM helix boundaries to define binding-pocket depth and lipid-facing surfaces that must remain untouched by candidate small molecules. Bacterial outer-membrane β-barrel porins and TonB-dependent transporters are targets for a new generation of antibiotics designed to exploit or block the BAM (β-barrel assembly machinery) complex, since Gram-negative pathogens' intrinsic antibiotic resistance often traces directly to restricted porin permeability. Ras and other small GTPases, whose oncogenic signaling depends entirely on prenylation (farnesylation or geranylgeranylation) for plasma-membrane localization, are the basis of an entire drug class — farnesyltransferase inhibitors — aimed at blocking the lipid anchor rather than the enzyme's catalytic site. GPI-anchor biosynthesis is itself a validated antiparasitic target, since Trypanosoma and Plasmodium species depend heavily on GPI-anchored surface antigens for immune evasion. Finally, engineered lipid-anchored fusion proteins (myristoylation or palmitoylation tags fused to fluorescent reporters) are a routine cell-biology tool for visualizing membrane microdomains and lipid-raft dynamics, directly exploiting the same acylation chemistry covered here. All of these applications sit downstream of the same architectural logic developed in Membrane Fluidity and Phase Behavior and feed forward into the transport and channel mechanisms of Membrane Transporters and Active Transport and Ion Channels and Selective Permeability.

Ask the tutor: How does the Sec61 translocon actually recognize a hydrophobic segment and decide, in real time, whether to release it laterally into the bilayer or keep threading the chain through the aqueous channel?

The physical picture

The defining chemical fact about the bilayer interior is its dielectric constant, ε ≈ 2–4, indistinguishable from a simple alkane. In water (ε ≈ 80) a peptide bond's carbonyl and amide dipoles can be solvated by reorienting neighboring water molecules; in the bilayer core there is no such luxury. Any unsatisfied hydrogen-bond donor or acceptor buried in the hydrocarbon region pays a severe electrostatic penalty — on the order of 3–5 kcal/mol per buried polar group — because the local medium simply cannot stabilize the resulting dipole. This is the same physics that makes oil and water immiscible, scaled up to the level of a folded polypeptide backbone. Two — and only two — secondary-structure solutions satisfy this constraint for a polymer that must cross the full bilayer thickness. The α-helix intramolecularly hydrogen-bonds every backbone C=O to the N–H four residues downstream (i → i+4), so a single continuous helix has no exposed backbone polarity along its hydrophobic length; only its very ends need solvation, and those ends are arranged to poke out into the headgroup/water interface. The β-barrel instead uses intermolecular (inter-strand) hydrogen bonds within a closed, continuous sheet — every strand pairs with its neighbors on both sides, and because the sheet closes on itself (barrel, not open sheet) there is no free edge exposing unpaired backbone groups to lipid. Both solutions are, at bottom, hydrogen-bond bookkeeping problems solved by different topologies. A membrane protein's sequence encodes this geometry directly. A single transmembrane (TM) α-helix needs roughly 20 residues to span a 30 Å hydrophobic core at 1.5 Å rise per residue, and those 20 residues must be predominantly hydrophobic (Leu, Ile, Val, Phe, Ala) with the flanking few residues enriched in Trp and Tyr — aromatic residues that preferentially partition into the headgroup interfacial belt, anchoring helix ends at the correct depth ("aromatic girdle" or "belt"). A β-strand, by contrast, only needs about 9–11 residues to cross the membrane because the extended (β) backbone has a rise of ~3.4 Å per residue pair, and because barrel strands alternate hydrophobic (facing lipid) and hydrophilic (facing the aqueous pore) side chains in a strict odd/even pattern — a signature immediately visible as an alternating pattern in the primary sequence.

Figure

Two folds, one constraint: α-helix versus β-barrel backbone H-bonding in a low-ε bilayer

Between the two aqueous faces, a membrane protein's topology — how many times it crosses, which loops face which side — is set at the moment of insertion by a translocon complex (the Sec61 channel in eukaryotes, SecYEG in bacteria) that reads hydrophobicity directly out of the nascent chain and partitions TM segments laterally into the bilayer. The positive-inside rule (von Heijne's rule) says that cytoplasmic loops are statistically enriched in positively charged Arg/Lys residues relative to extracytoplasmic loops; the translocon and the inside-negative membrane potential together orient each new helix so that basic residues stay on the cytosolic side, and this single statistical bias is powerful enough that topology of a new TM protein can often be predicted correctly from charge distribution alone. Peripheral proteins never present the translocon with a fully hydrophobic stretch: instead they display, on one face of an otherwise soluble domain, a strip of nonpolar and cationic residues that inserts shallowly into just the headgroup/acyl-chain interfacial region (the top ~5–10 Å of one leaflet) without crossing to the other side. This shallow insertion mode, combined with electrostatic recognition of anionic lipids such as phosphatidylserine (PS) or PIP₂, gives reversible, regulatable membrane binding — essential for signaling proteins that must be recruited to the membrane only on demand (e.g., upon a Ca²⁺ transient or a kinase-generated phosphoinositide).

Figure

Modes of membrane association: transmembrane, monotopic, amphipathic-helix, and lipid-anchored

The core ideas

The energetics of membrane insertion can be quantified with a two-state partitioning model. For a stretch of sequence moving from water into the bilayer, the transfer free energy is the sum of residue-level contributions read off an experimentally derived hydrophobicity scale:

\Delta G_{\text{transfer}} = \sum_{i=1}^{n} \Delta g_i(\text{position}) + \Delta G_{\text{helix}}

where:

$\Delta g_i$ is the position-dependent transfer free energy of residue $i$ (kcal/mol), tabulated from bilayer-depth-resolved scales (Wimley–White interfacial scale, or the Hessa–von Heijne "biological" scale measured directly with the Sec61 translocon)
position-dependence matters because the same residue costs far less at the interface than buried in the hydrocarbon core
$\Delta G_{\text{helix}}$ is a roughly constant favorable term (~ –4 to –5 kcal/mol) from backbone H-bond formation upon helix folding, which is "free" once the helix forms because it replaces water H-bonds lost on dehydration

A segment is predicted to insert as a stable, autonomous TM helix when the net free energy is favorable overall:

\Delta G_{\text{transfer}} < 0

Reading: a candidate TM segment does not need every residue to be hydrophobic — it needs the sequence-summed cost of desolvating its total surface, weighted by depth, to be outweighed by the hydrophobic driving force and the H-bond payoff of helix formation. This is why a single charged residue (Asp, Glu, Lys, Arg) can sometimes be tolerated mid-helix (as in many ion channels and transporters, where it plays a catalytic role) provided the rest of the segment is hydrophobic enough to make the sum negative — though the cost is never trivial and such residues are almost always compensated by salt bridges or by facing a water-filled pore rather than the lipid. The second governing relationship links the number of TM crossings to a purely geometric requirement — the positive-inside rule as a statistical/electrostatic bias rather than a strict law:

\Delta G_{\text{topology}} = \Delta G_{\text{translocon}} + z_{\text{loop}} F \psi_m

where:

$\Delta G_{\text{translocon}}$ is the intrinsic partitioning free energy of the loop/segment as read by the translocon machinery
$z_{\text{loop}}$ is the net charge of the loop connecting two TM helices
$F$ is the Faraday constant and $\psi_m$ the transmembrane electrical potential (cytosol negative relative to the exterior, typically –60 to –90 mV in the plasma membrane, though the effect operates cotranslationally at the ER/plasma membrane during insertion)

Reading: a positively charged loop is electrostatically stabilized by staying on the cytosolic (negative) side, biasing topology so that the translocon inserts subsequent helices with the correct orientation — this is why engineered charge reversals (adding Arg/Lys to a normally extracellular loop) can experimentally flip a protein's topology in the membrane, a classic proof of the rule's causal (not merely correlative) role.

Derivation AFree-energy accounting for a transmembrane helix — the Wimley–White partition

Goal. Show quantitatively why ~20 hydrophobic residues suffice to cross the bilayer spontaneously, and why one buried charge nearly abolishes that stability.

Step 1 — Define the reference states. Take the unfolded polypeptide in water as the initial state and the folded α-helix embedded across the bilayer as the final state. The overall transfer free energy decomposes into (i) dehydration of the peptide backbone and side chains, (ii) backbone hydrogen-bond formation upon helicity, and (iii) van der Waals packing against lipid acyl chains. Because backbone H-bonding is essentially complete in a well-formed transmembrane helix, term (ii) is treated as a fixed, favorable constant, leaving side-chain hydrophobicity as the discriminating variable.

Step 2 — Assign per-residue transfer free energies. The Wimley–White interfacial scale (measured using host–guest peptides partitioning into POPC vesicles) gives representative values (kcal/mol, favorable = negative):

\Delta g_{\text{Leu}} \approx -0.6,\quad \Delta g_{\text{Ile}} \approx -0.5,\quad \Delta g_{\text{Ala}} \approx +0.2,\quad \Delta g_{\text{Asp}^-} \approx +3.6,\quad \Delta g_{\text{Lys}^+} \approx +2.7

These reflect the cost of moving each side chain from water to the bilayer interface; the fully hydrophobic-core-only version of the scale (relevant to residues buried deep in the acyl-chain region rather than near the interface) pushes charged residues even higher, often above +5 kcal/mol, because there is no residual water or headgroup dipole available to help solvate the charge.

Step 3 — Sum over a representative 20-residue hydrophobic stretch. Take a segment such as poly-Leu/Ile/Val/Phe of composition 8 Leu, 6 Ile, 4 Val, 2 Phe (a realistic TM helix composition), each contributing roughly –0.5 kcal/mol on average:

\Delta G_{\text{side chains}} \approx 20 \times (-0.5\ \text{kcal/mol}) = -10\ \text{kcal/mol}

Add the helix backbone H-bond term, empirically about –4 kcal/mol net favorable for a fully formed 20-residue helix once water reorganization is accounted for:

\Delta G_{\text{transfer}} \approx -10 + (-4) = -14\ \text{kcal/mol}

This strongly negative value explains why hydrophobic 20-mers insert spontaneously and essentially irreversibly once partitioned — consistent with the very slow off-rates (months to years) measured for reconstituted TM helices in vesicles.

Step 4 — Insert a single buried charge and recompute. Replace one central Leu with Asp (a genuinely disruptive substitution, since it lands deep in the hydrocarbon core, not near the interface where partial charge stabilization is possible). Using a buried-core estimate of $\Delta g_{\text{Asp}^-} \approx +5.0$ kcal/mol (uncompensated, ionized side chain in ε ≈ 3):

\Delta G_{\text{transfer}} \approx (19 \times -0.5) + (-4) + (+5.0) = -9.5 - 4 + 5.0 = -8.5\ \text{kcal/mol}

Still net favorable in this toy sum — but this dramatically understates the real cost, because burying a fully ionized carboxylate in ε ≈ 3 typically costs far more than the linear hydrophobicity-scale increment suggests; Born-model electrostatics (transferring a unit charge from ε = 80 to ε = 3) predicts an additional penalty of order 10 kcal/mol on its own. In real membrane proteins, such "buried" charges are essentially never left unsatisfied: they are neutralized by an intramolecular salt bridge, face a water-filled pore rather than bulk lipid (as in the selectivity filter of ion channels), or are protonated/deprotonated to become neutral before insertion. The qualitative lesson survives the crude arithmetic: single buried charges are tolerated only when compensated by additional structural chemistry, never "for free."

Step 5 — Limiting case. As helix length $n \to 10$ residues (below the ~15–17 needed to span the hydrophobic core plus interfacial belts), the favorable hydrophobic sum shrinks proportionally while the fixed geometric requirement to bridge ~30 Å is not met — the segment cannot form a stable, fully membrane-spanning helix at all, and instead partitions as a surface-associated or reentrant loop. This matches the empirically observed minimum TM helix length of roughly 15–20 residues in solved structures.

Go deeper · Derivation B · 2 minStrand number and pore diameter in a β-barrel — geometric closure condition

Derivation BStrand number and pore diameter in a β-barrel — geometric closure condition

Goal. Derive the relationship between the number of β-strands $n$ in a barrel and its pore diameter, and show why outer-membrane β-barrels cluster at even strand numbers (8, 10, 12, 14, 16, 18, 22) rather than arbitrary values.

Step 1 — Model the barrel as a cylinder of tilted strands. Each antiparallel β-strand runs nearly parallel to the membrane normal but tilted at an angle $\alpha$ (typically 30–45°) relative to the barrel axis, to satisfy inter-strand hydrogen-bonding geometry around the curved wall. The shear number $S$ — the number of residues one must travel around the barrel circumference to return to the same strand, offset by one rung of the H-bonding ladder — is related to the number of strands $n$ by:

S = n \tan(\alpha) \cdot \frac{d_{\parallel}}{d_{\perp}}

For real barrels $S$ is close to $n$ itself for small tilt, and empirically $S \approx n$ to $n + 2$ .

Step 2 — Relate shear number to barrel radius. Structural analysis of solved barrels (OmpA, OmpF, FhuA, etc.) gives an approximately linear relationship between the mean barrel radius $R$ and the shear number:

R \approx \frac{S \cdot d_{\text{strand}}}{2\pi \cos(\alpha)}

where $d_{\text{strand}} \approx 4.4$ Å is the mean inter-strand spacing (Cα–Cα distance across adjacent antiparallel strands, close to the standard β-sheet spacing) and $\alpha$ the shear/tilt angle.

Step 3 — Evaluate for OmpF-type porins (n = 16). Using $S \approx 16$ , $d_{\text{strand}} = 4.4$ Å, and a typical tilt $\alpha \approx 41°$ (cos 41° ≈ 0.755):

R \approx \frac{16 \times 4.4}{2\pi \times 0.755} = \frac{70.4}{4.74} \approx 14.8\ \text{Å}

giving a barrel diameter of order 30 Å — in good agreement with the crystallographically observed OmpF pore diameter (the trimeric channel's individual barrel diameter is close to this estimate, with the physiologically restrictive constriction created by an internal loop rather than by the barrel wall itself).

Step 4 — Why even numbers dominate. Because strands are antiparallel and each strand's two flanking H-bond partners must themselves be antiparallel to it, closing the barrel with a continuous, unbroken hydrogen-bonding ladder around the full circumference requires an even number of strands; an odd-strand barrel would leave one "seam" where two adjacent strands run parallel (a geometrically strained, poorly hydrogen-bonded junction), which is strongly disfavored energetically. This is why every solved outer-membrane β-barrel (from 8-stranded OmpA to 26-stranded FadL-family transporters) has an even strand count.

Step 5 — Limiting case. As $n$ decreases toward the minimum observed value ( $n = 8$ , as in OmpA's membrane domain), the pore essentially closes to a diameter comparable to a single water molecule's hydration shell (a few Å), consistent with OmpA's barrel functioning structurally rather than as an open channel; conversely, as $n$ grows toward 22–26 strands (as in the TonB-dependent transporters like FhuA or BtuB), the barrel lumen becomes large enough to require a folded "plug" or "cork" domain occluding the pore from within, since an unplugged 22-stranded barrel would otherwise be a nonselective, dangerously large hole in the outer membrane.

Worked examples

Example 1 — Foundational: predicting a TM helix from a hydropathy plot. Consider a 25-residue stretch drawn from a real single-pass membrane protein (a simplified glycophorin-A-like segment): `ITLIIFGVMAGVIGTILLISYGIRR`. Apply the Kyte–Doolittle hydropathy scale (values per residue, window-averaged over 19 residues as is conventional) to the central 19-residue window `LIIFGVMAGVIGTILLISY`. Kyte–Doolittle values: Leu +3.8, Ile +4.5, Phe +2.8, Gly –0.4, Val +4.2, Met +1.9, Ala +1.8, Thr –0.7, Tyr –1.3. Summing the 19 residues (L,I,I,F,G,V,M,A,G,V,I,G,T,I,L,L,I,S,Y):

\sum = 3.8+4.5+4.5+2.8-0.4+4.2+1.9+1.8-0.4+4.2+4.5-0.4-0.7+4.5+3.8+3.8+4.5-0.8-1.3 = 40.0

\text{mean hydropathy} = \frac{40.0}{19} \approx 2.1

A window mean above the standard threshold of +1.6 (Kyte–Doolittle's empirical cutoff for reliable TM prediction) flags this as a strongly predicted transmembrane segment. Note the two terminal Arg residues (`RR`) are excluded from the hydrophobic window deliberately — they flank the helix and, per the positive-inside rule, signal that this C-terminal end faces the cytosol, matching the known topology of single-pass type I membrane proteins where a cleavable N-terminal signal sequence directs the bulk of the protein to the exoplasmic side and a C-terminal cytoplasmic tail carries the basic residues.

Example 2 — Applied: distinguishing peripheral from integral association biochemically. A membrane fraction is treated in parallel with (a) 1 M NaCl (high ionic strength), (b) 0.1 M Na₂CO₃ at pH 11 (alkaline carbonate extraction, which converts sealed vesicles into open sheets), and (c) 1% Triton X-100 (nonionic detergent). After each treatment, the membrane is re-pelleted by ultracentrifugation and pellet versus supernatant fractions are analyzed by SDS-PAGE. A protein found in the supernatant after NaCl treatment alone is a peripheral protein held by electrostatic contacts (its ionic interactions are screened by the high salt). A protein retained in the pellet after NaCl but released into the supernatant by carbonate treatment is a peripheral protein bound more tightly — via an amphipathic helix or hydrophobic wedge insertion into one leaflet — but still not spanning the bilayer, since carbonate at pH 11 disrupts the bilayer's sealed topology and mechanically strips any protein not physically threaded through it. A protein that remains in the pellet through all three treatments except Triton X-100, and is solubilized only by detergent (which itself dissolves the lipid bilayer into mixed micelles), is a bona fide integral membrane protein. This three-step operational assay (salt → carbonate → detergent) is one of the standard classical methods for classifying an unknown protein's membrane association mode before any structural information is available.

Example 3 — Competition-level: computing the topology change from an engineered charge reversal. A synthetic two-TM-helix construct is designed with the following native loop composition: an extracytoplasmic loop bearing net charge $z_1 = -1$ (2 Asp, 1 Arg → net –1) and a cytoplasmic loop bearing net charge $z_2 = +3$ (4 Lys, 1 Asp → net +3). The construct inserts with the expected topology (N-terminus out, per the positive-inside rule strongly favoring the +3 loop staying in). Now mutate the cytoplasmic loop to reverse its charge to $z_2' = -3$ (mutate the 4 Lys to Asp) while leaving the exoplasmic loop at $z_1 = -1$ . Estimate the free-energy consequence for topology using the electrostatic term from the governing relationship, with $\psi_m = -70$ mV (physiological, cytosol negative) and $F = 96{,}485$ C/mol: For the wild-type construct, the favorable (stabilizing) electrostatic contribution from having a +3 loop on the negative (cytosolic) side is:

\Delta G_{\text{elec, WT}} = z_2 F \psi_m = (+3)(96{,}485\ \text{C/mol})(-0.070\ \text{V}) = -20{,}262\ \text{J/mol} \approx -4.84\ \text{kcal/mol}

(favorable, because a positive charge is stabilized on the negative side of the membrane — consistent with the sign convention $z F \psi_m$ giving negative, favorable energy for positive $z$ on negative $\psi_m$ ). For the reversed-charge mutant, that same loop (now net –3) sitting on the cytosolic (negative) side would instead be destabilized:

\Delta G_{\text{elec, mutant, same topology}} = (-3)(96{,}485)(-0.070) = +20{,}262\ \text{J/mol} \approx +4.84\ \text{kcal/mol}

a swing of roughly 9.7 kcal/mol between keeping the same topology (now unfavorable by +4.84 kcal/mol relative to wild type) versus flipping topology so that the mutant –3 loop moves to the extracytoplasmic (positive-relative) side instead, restoring a favorable arrangement. Because 9.7 kcal/mol vastly exceeds thermal energy ( $RT \approx 0.6$ kcal/mol at 37 °C), this charge reversal is predicted — and experimentally confirmed in classic von Heijne-lab constructs — to flip the protein's topology by 180°, with both TM helices now inserting in the opposite orientation so that the mutated (now negative) loop ends up on the extracytoplasmic side where the membrane potential no longer penalizes it. This calculation illustrates, quantitatively, why the positive-inside rule is not merely descriptive statistics but reflects a real, calculable electrostatic driving force acting during cotranslational insertion.

Limiting cases, scope & common traps

Situation	What happens	Why
Hydrophobic stretch < ~15 residues	Fails to span bilayer as stable TM helix; becomes reentrant loop or surface helix	Insufficient length to bridge ~30 Å hydrocarbon core plus interfacial belts
Single buried charge, uncompensated	Strongly destabilizing (often >5–10 kcal/mol penalty)	Low-ε medium cannot solvate an ionized side chain (Born penalty)
Odd number of β-strands	Essentially never observed in native barrels	Leaves one parallel-strand seam that cannot fully H-bond antiparallel-style
Very large barrel (n ≥ 22)	Requires internal plug/cork domain	Otherwise pore is too large to maintain outer-membrane permeability barrier
High ionic strength wash	Releases electrostatically bound peripheral proteins only	Screens Debye length of ionic interactions, no effect on inserted hydrophobic segments
Palmitoylation site mutated (Cys→Ala)	Protein loses regulated membrane cycling	Removes the labile thioester that acyl-protein thioesterases can hydrolyze
GPI-anchor removed enzymatically (PI-PLC)	Protein released quantitatively from outer leaflet	GPI is the sole membrane-anchoring element; no polypeptide crosses the bilayer

Common trapMisconception: "hydrophobic = integral membrane protein"

A high average hydrophobicity score is necessary but not sufficient evidence of an integral, bilayer-spanning fold. Many peripheral proteins carry substantial hydrophobic character in an amphipathic helix or a hydrophobic loop that inserts only into the outer leaflet's acyl-chain region without ever reaching the opposing headgroups — annexins and many Bin/Amphiphysin/Rvs (BAR) domain proteins are classic examples. Conversely, some genuinely transmembrane segments (particularly in polytopic transporters) contain a buried charged residue essential for catalysis, and a naive hydropathy algorithm run without windowing or biological-scale correction will sometimes miss them as "not hydrophobic enough." Topology and membrane association must ultimately be confirmed by biochemical fractionation (salt/carbonate/detergent extraction), protease-accessibility mapping, or structural data — not by sequence inspection alone.

Common trapMisconception: "β-barrels can occur anywhere a protein needs to cross a membrane"

β-barrel folds are, with essentially no exception, restricted to the outer membranes of Gram-negative bacteria, mitochondria, and chloroplasts — organelles and organisms with a double-membrane, endosymbiotic ancestry. This is because β-barrel assembly requires a dedicated insertion machine (the BAM complex in bacteria, the SAM/TOB complex in mitochondria) that works from the periplasmic/intermembrane space side, chaperoning the unfolded polypeptide across an aqueous compartment and catalyzing lateral strand closure — a mechanism entirely different from and incompatible with the co-translational Sec61-based insertion used for the plasma membrane and ER. No plasma-membrane or ER-resident β-barrel proteins are known.

The chemist behind it

Gunnar von Heijne (b. 1951) is the biophysical chemist most responsible for turning membrane protein topology from a descriptive curiosity into a quantitative, testable science. Working first at Stockholm University and later at the SciLifeLab/Stockholm University joint center, von Heijne systematically analyzed the statistical distribution of charged residues flanking transmembrane segments across hundreds of then-known membrane protein sequences, discovering in 1986 the strong asymmetry now called the positive-inside rule. He went on to build, with the pioneering biochemist IngMarie Nilsson and collaborators including Stephen White, a series of in vitro translation/translocation assays using engineered constructs inserted directly into microsomal membranes via the Sec61 translocon — allowing him to measure, residue by residue, the actual free-energy cost the translocon "reads" for each amino acid at each position within a candidate transmembrane segment. This effort culminated in the Hessa–von Heijne "biological hydrophobicity scale" (2005, 2007), a set of experimentally calibrated transfer free energies that remain the gold-standard input for modern membrane protein topology predictors, complementing the physicochemical Wimley–White scale built from simple peptide-vesicle partitioning.

Origins · 1 minGunnar von Heijne and how this was discovered

Gunnar von Heijne

b. 1951 · Stockholm University; membrane protein topology and translocon biochemistry

Trained originally in theoretical physics before turning to computational and experimental membrane biochemistry, von Heijne bridged bioinformatics and wet-lab translocon assays. His statistical rule (1986) and later quantitative insertion-efficiency scale (with Tara Hessa, 2005–2007) together transformed topology prediction from sequence-pattern guesswork into a physically grounded, falsifiable free-energy calculation — directly enabling the worked charge-reversal analysis above.

History

- 1972 — Singer and Nicolson propose the fluid mosaic model, first framing membrane proteins as embedded rather than merely surface-adsorbed.

- 1985 — Deisenhofer, Huber, and Michel solve the photosynthetic reaction center structure, the first atomic-resolution integral membrane protein structure (Nobel Prize in Chemistry, 1988).

- 1986 — von Heijne formulates the positive-inside rule from statistical analysis of membrane protein sequence topology.

- 1991 — Cowan and coworkers solve the porin (OmpF/PhoE) β-barrel structures, establishing the β-barrel as a distinct membrane-spanning fold.

- 2005–2007 — Hessa, White, and von Heijne measure the Sec61-translocon-based "biological hydrophobicity scale," directly quantifying per-residue insertion free energies.

What is the difference between an integral and a peripheral membrane protein, precisely?

An integral membrane protein contains at least one segment (an α-helix or set of β-strands) that is permanently embedded within, and typically spans, the hydrophobic core of the bilayer; it cannot be removed without disrupting the bilayer itself (e.g., with detergent). A peripheral membrane protein associates with the membrane surface reversibly, through electrostatic interactions with lipid headgroups, shallow insertion of an amphipathic helix or hydrophobic loop into just the outer region of one leaflet, or through a specific lipid-binding domain (such as a PH or C2 domain) — but no part of the protein crosses the bilayer. Operationally, the two classes are distinguished biochemically: high salt releases many peripheral proteins, alkaline carbonate extraction releases the more tightly surface-bound ones, and only detergent solubilization releases genuinely integral proteins.

Why does the "positive-inside rule" work, mechanistically?

The rule reflects an electrostatic bias acting during cotranslational membrane insertion: the cytosolic face of most membranes is electrically negative relative to the exterior (the resting transmembrane potential), so a positively charged loop is energetically stabilized by remaining on the cytosolic side while the translocon (Sec61/SecYEG) is threading the nascent chain. In addition, the translocon itself and cytosolic chaperones/machinery may kinetically favor retention of basic residues on the cytosolic side independent of the potential. The rule is strong enough, statistically, that experimentally reversing loop charge (mutating basic residues to acidic ones or vice versa) can flip a protein's topology by 180°, as shown quantitatively using the $zF\psi_m$ electrostatic term.

Can a single protein be both integral and lipid-anchored?

Yes — these are not mutually exclusive categories. Many polytopic integral membrane proteins carry additional palmitoylation sites on cytosolic-facing cysteines near a transmembrane helix, which can stabilize local membrane curvature, promote lipid-raft partitioning, or regulate trafficking without contributing to the primary bilayer-spanning architecture. Conversely, a protein can be purely peripheral apart from a single covalent lipid anchor (e.g., a prenylated Ras protein), in which case the "integral" character comes entirely from the lipid tail rather than from any transmembrane segment of the polypeptide itself — this is sometimes called a "lipid-anchored" or "membrane-anchored" protein to distinguish it from true integral membrane proteins.

What is a GPI anchor and how is it attached?

A glycosylphosphatidylinositol (GPI) anchor is a glycolipid built from a phosphatidylinositol lipid tail linked through a glucosamine-containing glycan core to a phosphoethanolamine bridge, which in turn is amide-linked to the C-terminal carboxyl of a mature protein. Attachment occurs post-translationally in the ER lumen: a C-terminal GPI-attachment signal sequence is proteolytically cleaved by a transamidase complex, which simultaneously forms the new amide bond to the pre-assembled GPI anchor. The result is a protein tethered entirely by a lipid tail in the outer (extracellular/luminal) leaflet, with no polypeptide crossing the bilayer at all — cleavable specifically by phosphatidylinositol-specific phospholipase C (PI-PLC), a diagnostic test for GPI anchoring.

Why can prenylation but not myristoylation alone often achieve stable, long-term membrane attachment?

Myristoylation (N-terminal Gly-linked C14 saturated acyl chain, added co-translationally by N-myristoyltransferase) provides only weak, reversible membrane affinity on its own because a single C14 chain's partition coefficient into bilayers is modest; myristoylated proteins typically need a second signal — a polybasic patch or an additional palmitoylation — for stable localization (the "myristoyl-plus" mechanism, as in Src-family kinases and MARCKS). Prenylation (farnesyl C15 or geranylgeranyl C20 groups thioether-linked to a C-terminal CAAX-box cysteine) provides a longer, more hydrophobic, and more rigid isoprenoid tail, often combined with a second prenyl or palmitoyl group (as in Ras proteins), giving substantially tighter and more stable membrane binding sufficient for constitutive plasma-membrane localization.

How do hydropathy plots actually predict transmembrane segments, and what are their limitations?

A hydropathy plot slides a window (typically 19–21 residues) along the sequence, averaging a per-residue hydrophobicity score (Kyte–Doolittle, Wimley–White, or the Hessa–von Heijne biological scale) at each position. Windows whose average exceeds an empirical threshold (~+1.6 for Kyte–Doolittle) are flagged as candidate transmembrane helices. Limitations include: the algorithm cannot distinguish a genuine TM helix from a strongly hydrophobic surface loop or signal sequence without additional context; it systematically underweights charged residues that are functionally essential but locally compensated (e.g., ion-channel selectivity-filter residues); and it cannot predict β-barrel strands at all, since individual strands are too short (9–11 residues) and only alternate hydrophobic/hydrophilic residues rather than being uniformly hydrophobic — barrel prediction requires dedicated pattern-recognition or machine-learning methods trained on known barrel structures.

Why don't peripheral membrane proteins need to satisfy the same backbone hydrogen-bonding constraint as integral proteins?

Because peripheral proteins never leave the aqueous/interfacial environment — their bilayer-facing surface sits at most in the headgroup region and the very outer edge of the acyl-chain region, both of which retain substantial water and polar headgroup contacts (phosphate, choline, glycerol oxygens) capable of solvating exposed backbone or side-chain polarity. Only the deep hydrocarbon core (ε ≈ 2–4, essentially anhydrous) imposes the severe hydrogen-bond-satisfaction requirement that forces integral proteins into α-helix or β-barrel folds; peripheral proteins simply never enter that region, so ordinary globular folds (with occasional amphipathic helices or shallow hydrophobic loops) suffice.