From sequence to structure: a receptor in 3D

In this practical you will follow one drug target – Toll-like receptor 4 (TLR4) – from its entry in a protein database to its three-dimensional structure, compare an experimental structure with an AlphaFold prediction, and finish with a labelled figure of the kind you need for your coursework.

About 90 minutesNo software to installWorks in Chrome, Edge, Firefox or Safari

Everything happens on this page

Instructions are on the left. The workbench on the right has three tools: the 3D viewer (replacing ICM-Browser), a UniProt explorer and a PDB explorer. They use live data from the real databases.

buttons do a step for you. Try each step yourself first and use them if you get stuck.
Activity steps tick themselves when the page sees you do them. You can also tick them yourself.
Your answers are saved in this browser. Use My answers (top right) to download them at any time.

Your name (optional – it goes on your downloaded answers):

Why TLR4?

TLR4 is the receptor that lets our innate immune system detect lipopolysaccharide (LPS, endotoxin) from the outer membrane of Gram-negative bacteria. Over-activation of TLR4 drives the inflammation of sepsis, so it has been a target for drugs such as the LPS antagonist eritoran and the small molecule resatorvid (TAK-242). It is a good example of the four questions structure can answer (from the lecture):

How does shape affect function?

The curved, horseshoe-shaped TLR4 ectodomain – Chapter 3.

How does a protein interact with partners?

TLR4 needs its co-receptor MD-2, and forms a dimer – Chapters 2 & 3.

How do mutations affect function?

The common Asp299Gly / Thr399Ile variants – Chapters 1 & 3.

How do ligands and drugs interact?

LPS in MD-2's pocket, and the antagonist eritoran – Chapter 3.

Warm-up: your first structure

Activity 0 · Load, rotate, identify

At the top of the 3D viewer, type 3FXI in the PDB ID box and press Load.
Left-drag to rotate, scroll (or pinch) to zoom, right-drag (or Ctrl + drag) to move. Press ⟲ (top right of the viewer) to reset the view.
Hover over the molecule: a label tells you what each part is. Click an atom to select its residue (green markers appear), then press Esc to clear.

To…	Mouse / trackpad
Rotate	Left-drag (one-finger drag on a touch screen)
Zoom	Scroll wheel, pinch, or Shift + left-drag
Move	Right-drag, or Ctrl/⌘ + left-drag
Select	Click (Shift-click adds, Ctrl/⌘-click toggles) · Esc clears
Centre on a residue	Double-click it

Plan for today

Chapter	What you will do	Time
	Find human TLR4 and read its entry: function, location, variants, drugs, domains	25 min
	Find experimental structures and understand what is in them	10 min
	Explore TLR4–MD-2–LPS: styles, colours, selections, the LPS binding site	30 min
	Judge a predicted model with pLDDT and PAE and compare it with experiment	15 min
	Colours, key, labels and schematics – then save a high-resolution image	10 min
	Repeat the workflow for the receptor in your coursework	after class

Sections marked Extension are optional and are not counted in your progress bar.

UniProt: what do we know about a protein?

≈ 25 minWorkbench: UniProt tab

UniProt (uniprot.org) is a worldwide consortium that collects and annotates protein sequences and their functions. Each protein has an entry – like its home page – gathering information from experiments, the literature, analysis software and dozens of other databases.

Two halves of UniProtKB

Far too much sequence data arrives for every entry to be checked by hand, so UniProtKB has two parts: Swiss-Prot – reviewed entries curated by experts (★) – and TrEMBL – unreviewed entries annotated automatically. Every entry has a stable accession (e.g. O00206) and a readable entry name (e.g. TLR4_HUMAN).

1.1 Find human TLR4

Activity 1.1 · Search UniProtKB

In the , search for toll-like receptor 4.
Like a Google search, this returns everything. Look down the list: other Toll-like receptors (TLR9, TLR5…) and TLR4 from other animals (cat, horse, cow…) appear around the human protein – some of them above it. Always double-check the protein name and the organism!
Tick Reviewed (Swiss-Prot) only and Human only to narrow the results.
Click the accession of human Toll-like receptor 4 to open its entry.

What is the accession of human TLR4?

O00206 (entry name TLR4_HUMAN). Note that it starts with the letter O, not a zero.

Why is it important to check the organism and the Reviewed status before you use an entry?

Many proteins share a name across species (TLR4 in cat, horse, cow…) and related proteins appear in the same search (TLR9, TLR5…). Using the wrong entry gives the wrong sequence, residue numbers, variants and structures. Reviewed (Swiss-Prot) entries have been checked by curators against the literature; unreviewed (TrEMBL) entries are annotated automatically and may contain unverified predictions.

1.2 Function

The entry is organised into sections, listed down the left of the entry. Function summarises what the protein does, with references. Gene Ontology (GO) terms describe roles in words that computers can search; the code after each term shows the evidence (e.g. IDA = direct assay, IEA = automatic). Pathway databases such as Reactome place the protein in signalling pathways – e.g. R-HSA-166016, Toll Like Receptor 4 (TLR4) Cascade.

Activity 1.2 · Read the Function section

Open Function and read the summary.
Scan the GO terms and the Reactome pathways. Open R-HSA-166016 if you would like to see the pathway diagram (it opens Reactome in a new tab).

Which bacterial molecule does TLR4 respond to, and which protein does it cooperate with at the plasma membrane?

Lipopolysaccharide (LPS) from Gram-negative bacteria, together with LY96 – better known as MD-2. (CD14 is also part of the LPS receptor complex.)

Name two adaptor proteins through which TLR4 signals inside the cell. What is the end result of the signal?

MYD88 and TIRAP (also called MAL), with TRAF6 acting downstream. At the plasma membrane this leads to activation of NF-κB, secretion of pro-inflammatory cytokines and an inflammatory response. After TLR4 is taken into the cell by endocytosis, a MyD88-independent pathway uses TRAM (TICAM2) and TRIF (TICAM1), leading via TBK1 and IRF3 to production of type I interferon.

1.3 Names & taxonomy

Different research groups often give the same protein different names, and the protein name may differ from its gene name. This section sorts that out. The official human gene symbol is set by the HUGO Gene Nomenclature Committee (HGNC).

Activity 1.3

Open Names & Taxonomy.

Immunologists often use CD (“cluster of differentiation”) names. What is TLR4's CD name?

CD284. (Its alternative name is hToll; the gene is TLR4, HGNC:11850.)

1.4 Where is it? Subcellular location and topology

Activity 1.4 · From topology to 3D

Open Subcellular Location. Compare the UniProt annotation with the Gene Ontology cellular component terms – they do not always agree.
Look at the Topology bar: extracellular (blue), transmembrane (orange), cytoplasmic (peach).
Click the orange transmembrane segment. The page loads the AlphaFold model of TLR4 into the 3D viewer and highlights the helix that crosses the membrane. Come back to the UniProt tab afterwards.

Which residues form the transmembrane helix?

Residues 632–652. The extracellular domain is 24–631 and the cytoplasmic domain 653–839.

TLR4 is a “single-pass type I membrane protein”. Where is its N-terminus?

Outside the cell (extracellular) In the cytoplasm Buried in the membrane

1.5 Disease, variants and drugs

Pharmacology link

This section links a protein to medicine: natural variants (differences between people), mutagenesis experiments (residues changed on purpose to test their role), associated diseases, and – under Chemistry – drugs and compounds from DrugBank, ChEMBL and the IUPHAR/BPS Guide to Pharmacology.

Activity 1.5

Open Disease & Variants. Read the Polymorphism note.
In the natural-variants table, click the row for position 299 (or 399) to find that residue in 3D.
Scroll to Chemistry and look at the compounds listed.

Which two amino-acid changes make up the TLR4*B allele, and what is their reported effect?

Asp299Gly (D299G) and Thr399Ile (T399I). The allele is associated with a reduced (blunted) response to LPS, e.g. to inhaled LPS.

Mutating proline 714 abolishes binding to which adaptor protein?

MYD88 – P714H abolishes MyD88 binding and the LPS response. The equivalent mutation in mouse TLR4 (Pro712His) is why C3H/HeJ mice do not respond to LPS; it helped identify TLR4 as the LPS receptor (Poltorak et al., 1998).

Name one compound listed under Chemistry that was developed to block TLR4. What happened to it?

Eritoran (E5564) – a synthetic analogue of lipid A that competes with LPS for MD-2. It reached a phase III trial in severe sepsis (ACCESS) but did not reduce 28-day mortality (Opal et al., 2013). Resatorvid (TAK-242) instead binds Cys747 inside the cell, in the TIR domain.

1.6 PTM / processing

After translation, proteins are processed and modified. Proteins with an extracellular domain usually lose a signal peptide, and are often glycosylated and stabilised by disulfide bonds.

Activity 1.6

Open PTM / Processing. Click a green glycosylation pin to see where the sugar is attached in 3D.

How many N-linked glycosylation sites are annotated?

10 (Asn-35, 173, 205, 282, 309, 497, 526, 575, 624 and 630). The PTM note explains that Asn-526 and Asn-575 are needed for TLR4 to reach the cell surface.

Residues 1–23 are a signal peptide. What happens to them, and why might an experimental structure start at residue 24 or later?

The signal peptide directs TLR4 into the ER for the secretory pathway and is then cleaved off, so it is not part of the mature receptor. Structures are made from the mature protein (often a shorter fragment), so they start after the signal peptide – 3FXI starts at residue 27. UniProt keeps numbering from the precursor's Met-1, and most structures use the same numbers, so residues can be compared directly. But not always: 1CDW numbers TBP 4 lower than UniProt (its Ser155 is UniProt's Ser159). The DBREF line at the top of a PDB file shows how the two numbering schemes line up.

1.7 Expression and interactions

Activity 1.7

Open Expression . Large proteomics studies are summarised in the Human Protein Atlas (link).
Open Interaction . Interactions can be shown by experiments on the human protein (a PubMed reference) or inferred from a similar protein in another species (“by similarity”).

In which tissues and cells is TLR4 most highly expressed? Why does this make sense for its function?

Highly expressed in placenta, spleen and peripheral blood leukocytes; detected in monocytes, macrophages, dendritic cells and some T cells. These are innate immune cells that must detect invading bacteria quickly.

DERP2 is listed as an interaction partner. What is it, and why might it bind TLR4?

Der p 2, a major allergen of the house-dust mite (Dermatophagoides pteronyssinus). It resembles MD-2 in structure and can stand in for it, helping to activate TLR4 – which helps explain why it is such a potent allergen in asthma (Trompette et al., 2009).

1.8 Family & domains, and the sequence

Programs that recognise sequence patterns identify domains and repeats. The bar diagram shows where they lie along the protein – a format you can reproduce yourself in PowerPoint. UniProt numbering is the reference system: other numbering schemes (e.g. in PDB files) can be related to it.

Activity 1.8

Open Family & Domains.
Open Sequence & Isoforms. Use Go to residue to check which amino acid is at position 299.

How many leucine-rich repeats (LRRs) are annotated?

18 LRRs (residues 55–565), followed by an LRR C-terminal cap (LRRCT, 579–629). The repeats stack to form the curved ectodomain you will see in 3D.

Which domain is in the cytoplasmic part of TLR4?

The TIR (Toll/interleukin-1 receptor) domain, residues 672–815. TIR domains bind the TIR domains of adaptors such as MYD88 and TIRAP.

How many amino acids long is the canonical sequence?

839 amino acids (95,680 Da). The entry describes 3 isoforms made by alternative splicing; the canonical one is shown.

1.9 Which parts have structures?

The Structure section lists every PDB entry that contains this protein – with method, resolution, chains and the residues covered – and links to the AlphaFold model.

Activity 1.9

Open Structure. Study the coverage map.

Which part of TLR4 has no experimental structure? What could you use instead?

Every X-ray and cryo-EM structure covers only the extracellular domain (about 23–631); two NMR entries (5NAM, 5NAO) contain short transmembrane fragments. The cytoplasmic TIR domain (672–815) and C-terminal tail have no experimental structure – the AlphaFold model covers the whole sequence (1–839). Chapter 4 looks at how far to trust it.

Extension: practise on TBP Extension

The original version of this tutorial used the transcription factor TATA-box binding protein (TBP). Use it to practise.

Activity 1.10 · TATA-box binding protein

Search UniProt for TBP and open the human entry.
Open Family & Domains . One program has found two repeats in the sequence (165–241 and 255–332). Another has marked regions of compositional bias: stretches made mostly of one or a few amino acids.
In Sequence & Isoforms, type QQQQ in the find box to highlight the long run of Qs – a striking example of biased composition that you can see by eye.

TBP contains two similar repeated regions. What evolutionary event probably produced them?

A gene duplication: part of an ancestral gene was copied and the two copies fused, so the two halves of TBP's core have similar sequences. In 3D the two repeats fold into the two similar halves of the saddle-shaped core that sits astride the DNA (see 1CDW in the PDB chapter).

What is the accession of human TBP?

P20226 (TBP_HUMAN).

Which amino acid does the long run of Qs represent?

Glutamine (Gln, Q). The canonical sequence has 38 in a row (residues 58–95). Expansion of this poly-Q tract to about 47–63 repeats causes spinocerebellar ataxia 17 (SCA17) – see Disease & Variants.

In the Structure section, which chains of PDB entry 1C9B are TBP?

Chains B, F, J, N and R (residues 158–337). 1C9B contains five copies of a TBP–TFIIB–DNA complex. By contrast 1CDW has a single TBP (chain A) bound to DNA.

The Protein Data Bank: experimental structures

≈ 10 minWorkbench: PDB tab

The wwPDB is the worldwide archive of experimentally determined structures of proteins, nucleic acids and their complexes. It has three sites with different tools – RCSB PDB (USA), PDBe (UK) and PDBj (Japan) – but they share the same data. Each entry has a four-character ID such as 3FXI (upper or lower case).

Things to check for every entry

Method – X-ray crystallography, cryo-electron microscopy (cryo-EM) or NMR.
Resolution (X-ray and cryo-EM) – a smaller number means finer detail. At ~1.5 Å individual atoms are clear; at ~3 Å you can trace the chain and most side chains; beyond ~4 Å mainly the overall shape.
What was actually studied – structures are experiments, so they often contain only a fragment of the protein, engineered mutations or fusions, and molecules added to help crystallisation.

2.1 Look up 3FXI

Activity 2.1

In the , search for 3FXI.
Read the method, resolution and paper, then the Macromolecules and Small molecules tables.

How was 3FXI determined, and at what resolution?

X-ray diffraction, 3.1 Å Cryo-EM, 2.24 Å Solution NMR X-ray diffraction, 1.7 Å

How many copies of TLR4 and of MD-2 (gene name LY96) does 3FXI contain? What does this tell you about the activated receptor?

Two of each: TLR4 = chains A, B; MD-2 (lymphocyte antigen 96) = chains C, D. LPS binding brings two TLR4–MD-2 units together into an “m”-shaped 2:2:2 complex. In the cell this dimerisation brings the two cytoplasmic TIR domains together, creating a platform for the adaptors MyD88 and TIRAP.

LPS appears in pieces: fatty acids (FTT, MYR, DAO), phosphate (PO4) and a branched sugar chain. Why might one ligand be split like this in a PDB entry?

Every chemical building block in the PDB is described by an entry in its Chemical Component Dictionary. A large, branched molecule like LPS (lipid A + core oligosaccharide) is therefore modelled as several components – glucosamines, phosphates, fatty-acyl chains, Kdo and heptose sugars. Parts that are disordered (e.g. the outer sugars) are not seen at all.

2.2 It's just text!

A structure file is a text file listing every atom with its x, y, z coordinates. The older .pdb format is read by most programs; the newer .cif (mmCIF) format is needed for very large structures.

Activity 2.2

In the 3FXI entry, press Look inside the .pdb file.
In one ATOM line, find the atom name, residue name, chain, residue number, the x/y/z coordinates and the B-factor.

What is the difference between ATOM and HETATM lines?

ATOM lines are standard amino acids or nucleotides in a polymer chain. HETATM (“hetero atom”) lines are everything else: ligands, ions, water, sugars and modified residues.

Downloading files for other software

Use the PDB format / mmCIF buttons in the entry to save a file (e.g. for ICM-Browser, PyMOL or ChimeraX). If you download from a website and the file ends up as 3fxi.pdb.txt, your computer has treated it as text: rename it to remove .txt (you may need to turn on “show file extensions” to see it).

2.3 Search the PDB

Activity 2.3 · Three ways to search

Search by protein name: Toll-like receptor 4. Look at the organisms in the results.
Tick Human only. How many are left?
Now search with the UniProt accession O00206.
Sort by Newest first. Which method was used for the newest human TLR4 structures?

Why is searching with the UniProt accession more reliable than searching with a name or free text?

Names are ambiguous: a name search also finds other species (e.g. mouse TLR4) and related proteins, and free text matches any mention in titles or papers. The accession identifies exactly one protein sequence, and every PDB chain is mapped to UniProt (by the SIFTS project), so the search returns exactly the entries containing human TLR4 – 15 at the time of writing.

What method was used for the newest human TLR4 structures, and what did they study?

Cryo-EM: 8WO1 (2.24 Å) and 8WTA (2.9 Å), released in 2024, and 9J03 (2.7 Å, 2025) show human TLR4–MD-2 bound to synthetic lipid A mimetics. Cryo-EM now reaches resolutions that used to need crystals, and is well suited to large membrane-protein complexes. (Newer entries may have appeared since this was written.)

Extension: TBP bound to DNA Extension

Activity 2.4 · Refining a search

Search by protein name for TATA-box-binding protein, then tick Human only and Contains DNA.
Open the coverage map for TBP (UniProt P20226 → Structure). Careful: the bars show which part of the protein was in the sample, not which part could be seen. Many cryo-EM entries used full-length TBP, so their bars span 1–339, but only the core is visible in their maps – 5IY6, for example, models residues 153–337.

Most X-ray structures of TBP contain only residues ~159–339. Why do you think researchers crystallised just this part?

Residues ~159–339 form the conserved C-terminal core that binds the TATA box. The N-terminal region (with the poly-Q tract) is flexible and disordered, which stops proteins crystallising, so it is usually cut off. Cutting long proteins into stable pieces is a very common strategy – so a complete picture may need several entries (or a prediction).

Exploring TLR4–MD-2–LPS in 3D

≈ 30 minWorkbench: 3D viewer

A molecular graphics program turns the coordinates in a structure file into a picture you can explore. Here you will do in the browser everything the ICM-Browser worksheet did: change styles and colours, select parts of the molecule, look at a binding site, measure, label and save views.

3.1 The layout

Part of the viewer	What it does	ICM-Browser equivalent
Load bar (top)	Load a PDB entry, an AlphaFold model or your own file	PDB search box · File › Open
Tabs: Display, Colour, Select, Labels & key, Measure, Views, Save image	Change how the selection looks, and make pictures	display / labels tabs
Selection bar (green when something is selected)	Tells you what the buttons will change	“no selection” box
Workspace (left)	Chains and ligands, with eye icons to show/hide	Workspace panel
Sequence bar	Sequence of one chain – drag along it to select	“+” next to a chain
Command line (bottom)	Every button is written here as a command; you can type commands too	Terminal

Activity 3.1 · What is in 3FXI?

Load 3FXI if it is not already loaded.
Look at the Workspace: chains A–D, LPS and LPS*, N-glycans, magnesium ions and water (hidden). Hover over each part of the molecule to check which is which.
Press ▸ next to LPS to list its pieces: glucosamines (PA1, GCS), phosphates (PO4), fatty-acyl chains (FTT, DAO, MYR) and core sugars (KDO, GMH). Click one, e.g. FTT A1003, to select it and see what that part of LPS looks like.

Chain letters of LPS: in the file, the fatty-acyl chains and phosphates of LPS are labelled chain A and its sugars chain E (B and H for LPS*), even though LPS sits in MD-2 (chain C). The Workspace gathers the pieces for you as LPS and LPS*, but commands and interaction labels use the file's letters – e.g. A:1010 is one of LPS's phosphates.

Which chains are TLR4 and which are MD-2?

A and B are TLR4; C and D are MD-2 A and C are TLR4; B and D are MD-2 A is TLR4, B is MD-2, and C and D are LPS

3.2 Sequence and structure

Activity 3.2

The sequence bar shows chain A. Drag along the letters from residue 27 to about 100: those residues are marked with green dots in 3D. (Press ▸ next to another chain in the Workspace to show its sequence.)
Scroll to the end of the sequence. Chain A stops at Cys627, although the protein used in the experiment went on to Lys631: those last residues were too disordered to be seen, so they were not modelled. (Gaps inside a chain are shown as a red ⋯ – you will see some in 4KXF in Activity 3.9.)
Press Esc (or Clear selection) to clear the selection.

Chain A starts at residue 27, not residue 1. Why?

The protein made for crystallisation was only the ectodomain, without the signal peptide (1–23) – structures often contain only part of a protein. The residue numbers match UniProt, which makes comparison easy. Residues 628–631 were in the protein but disordered, so they could not be modelled.

3.3 One unit or two?

3FXI contains two TLR4–MD-2–LPS units. Following the paper, the second copy is marked with an asterisk: TLR4* (chain B), MD-2* (chain D) and LPS*.

Activity 3.3 · Show only one unit

Use the eye icons in the Workspace to hide chains B and D and LPS*. (Their sugars and ion are hidden with them.)
Click TLR4 (chain A) in the Workspace to select it, then press ⌖ (centre on selection) at the top right of the viewer.
Bring the second unit back with the eye icons (or Display › Show all).

Why is it important that LPS brings two TLR4–MD-2 units together?

Dimerisation is the activation step: it brings the two cytoplasmic TIR domains close together (they are not in the crystal), creating a platform that recruits the TIR-containing adaptors (TIRAP/MyD88, TRAM/TRIF) and starts signalling to NF-κB. Each LPS directly bridges the two units.

3.4 Display styles

Styles apply to the current selection – or to everything if nothing is selected. Click a style once to add it, and again to remove it. The lecture's three basic styles are cartoon (the fold), sticks (bonds) and spheres (atoms at their real size).

Activity 3.4 · Try each style

Select MD-2 (chain C) in the Workspace, then in the press Surface and then See-through.
Select LPS and press Spheres to see how LPS fills MD-2's pocket.
Try Lines, Ball & stick and Tube on part of TLR4. Use No style to remove all styles from the selection.

You want to show exactly which atoms of LPS make hydrogen bonds with TLR4. Which style is best?

Sticks Cartoon Spheres Surface

3.5 Colour schemes

Activity 3.5 · Colour with a purpose

Select TLR4 (chain A) and choose Rainbow N→C in the : the chain runs blue (N-terminus) → red (C-terminus).
Now choose Secondary structure.
Select MD-2 (chain C) (still a see-through surface) and choose Hydrophobicity. Show LPS as sticks inside it.

Which secondary structure dominates TLR4, and where is it?

β-strands. Each leucine-rich repeat contributes one short strand, and together they form a continuous parallel β-sheet lining the concave (inner) face of the curved “horseshoe”. The convex outer face is mostly loops with a few short helices. MD-2 binds on the concave side.

What kind of residues line MD-2's pocket, and why does that suit LPS?

Mainly hydrophobic residues (orange: Ile, Leu, Phe, Val…). They pack against the fatty-acyl chains of lipid A, which are hydrophobic. The polar phosphates and sugars stay at the mouth of the pocket, near polar and positively charged residues.

3.6 The LPS binding site

Pharmacology link

How a ligand binds – which residues it touches, and whether through hydrogen bonds, salt bridges or hydrophobic contacts – is the starting point for designing drugs that mimic or block it.

Activity 3.6 · Binding-site analysis

Click LPS in the Workspace to select it. In the press Binding site. Residues within 4.5 Å appear as sticks, and dashed lines show hydrogen bonds and salt bridges.
Hover over a dashed line to read which atoms it joins. Use the list of residue buttons to jump between them.
Select three residues that contact the phosphates or sugars (click them) and press Label residues in the .

Which part of LPS is buried inside MD-2, and which parts contact TLR4?

Five of the six fatty-acyl (lipid) chains of lipid A are buried in MD-2's hydrophobic pocket. The sixth chain sticks out of MD-2 and packs against conserved phenylalanines of the second TLR4 (TLR4*), and the phosphates and core sugars form ionic and hydrogen-bond contacts with positively charged and polar residues of TLR4 and MD-2 (Park et al., 2009). This is how LPS bridges the two units.

List three residues that make polar contacts (hydrogen bonds or salt bridges) with LPS, with their chain.

For example: TLR4 (chain A) Arg264, Asp294, Tyr296, Lys341; MD-2 (chain C) Ser118, Ser120, Lys122. Across the dimer interface, TLR4* (chain B) Gln436 hydrogen-bonds the exposed acyl chain of LPS. Exact lists depend on the distance cut-off used.

3.7 Box selection and measuring

Activity 3.7

In the , press Box select and drag a rectangle around part of the molecule (like ICM's manual selection). Colour the selection or change its style, then press Esc.
In the , press Measure distance and click two atoms – e.g. the NE nitrogen of Arg264 and oxygen O3 of the phosphate next to it (A:1010), or the NZ nitrogen at the tip of Lys341 and an oxygen of the Kdo sugar it touches.

What distance did you measure, and between which atoms? Is it close enough for a hydrogen bond or salt bridge?

For example Arg264 NE to phosphate O3 is about 3.3 Å. Hydrogen bonds and salt bridges are typically 2.6–3.5 Å between the heavy (non-hydrogen) atoms, so this is a genuine interaction. Note that X-ray structures usually do not show hydrogen atoms.

3.8 Labels and saved views

3D labels stay attached to their residue when you rotate; 2D text stays fixed on the screen (like a title). Saved views work like ICM's slides: they remember the orientation, styles, colours and labels.

Activity 3.8

In the , add a 2D text title. Rotate the molecule and watch the 3D labels move while the title stays put. Drag labels to move them; double-click to edit.
In the , save a view, rotate to a new orientation, and save another. Click the thumbnails to switch between them.

3.9 Many copies: 4KXF

Some entries load with many copies of the same protein because of how the protein packed in the crystal.

Activity 3.9 · Remove extra copies

Load 4KXF (the immune sensor NLRC4). Hide 3FXI with its eye icon if both are shown.
Leave just one copy visible: select one chain and press Display › Show only, or use the eye icons.
When finished, remove 4KXF (Workspace › ⋯ › Remove) and show 3FXI again.

4KXF contains eight copies of NLRC4. Is NLRC4 an octamer? How would you check?

No. The asymmetric unit of the crystal just happened to contain eight molecules (crystal packing). The biological assembly defined by the authors is a monomer – check the “Biological assemblies” section of the entry in the PDB tab. Contacts between copies in a crystal are not necessarily biological.

Extension: an agonist and an antagonist Extension

2Z65 is MD-2 bound to eritoran, the lipid A analogue that went to phase III trials. To help it crystallise, the TLR4 in this entry is a hybrid: its N-terminal part fused to a hagfish protein (VLR).

Activity 3.10 · LPS vs eritoran

Load 2Z65 as well as 3FXI.
Superpose the MD-2 of 2Z65 onto the MD-2 of 3FXI: The RMSD is shown in the message.
Show just the two MD-2s with their ligands:

Compare eritoran (magenta) with lipid A. How many lipid chains does each have, and why might eritoran block TLR4 rather than activate it?

Eritoran has four lipid chains, all buried in MD-2's pocket. E. coli lipid A has six: the extra chains push the phosphorylated glucosamine backbone ~5 Å out towards the solvent, so one lipid chain and the phosphates can contact the second TLR4 and drive dimerisation. Eritoran fills the pocket (competing with LPS) but cannot make this bridge, so no dimer forms and there is no signal – it is an antagonist.

Extension: where are the TLR4*B variants? Extension

Activity 3.11

Find Asp299 and Thr399 in 3FXI chain A: LPS* is shown too, because part of it touches chain A across the dimer interface. Measure from each residue to the nearest atom of LPS and of LPS*.

Are Asp299 and Thr399 close to LPS? What might this suggest about how they reduce the LPS response?

No. Asp299 is about 14 Å from LPS, and Thr399 about 15 Å from LPS* (the other unit's LPS), so neither touches a ligand directly. The crystal structure of the variant receptor (4G8A; Ohto et al., 2012) still forms the normal “m”-shaped complex with LPS, with only local changes near residue 299 that alter the surface of TLR4. Their effect is therefore subtle, not a blocked binding site.

AlphaFold: predicted structures

≈ 15 minWorkbench: 3D viewer

AlphaFold2 (Google DeepMind) predicts a protein's 3D structure from its sequence. It is a neural network trained on structures from the PDB and on evolutionarily related sequences from UniProt – so its success rests on fifty years of experimental work. The AlphaFold Protein Structure Database (alphafold.ebi.ac.uk, run with EMBL-EBI) holds predictions for over 214 million proteins.

Two confidence scores

pLDDT (0–100) – confidence in each residue's local structure. Dark blue > 90 very high · light blue 70–90 confident · yellow 50–70 low · orange < 50 very low (often disordered).
PAE (predicted aligned error, in Å) – confidence in the relative position of two residues. Low PAE between two domains means their arrangement is reliable; high PAE means it is not.

A prediction is a model, not an experiment: AlphaFold DB models are single chains, without partners, ligands, ions or sugars.

4.1 The TLR4 model

Activity 4.1

Load the AlphaFold model of TLR4: type O00206 in the AlphaFold box of the viewer and press Load model.
The model is coloured by pLDDT, and the AlphaFold confidence panel (top right) summarises it. Open pLDDT along the sequence in the panel.

Need the model file for other software (ICM-Browser, PyMOL, ChimeraX…)? Use the .pdb file button in the , or download it from the model's page on AlphaFold DB. UniProt's Structure section links to the same page.

What is the mean pLDDT of the model?

About 89 (89.2). Around 71% of residues are “very high” confidence.

Which regions have very low confidence (orange)? Suggest a reason for each.

The N-terminus (≈1–25) – the signal peptide, which is cleaved off and has no fixed structure; and the C-terminal tail (≈821–839), which is flexible. The linker between the membrane helix and the TIR domain (≈655–672) is also low. Very low pLDDT usually indicates flexible or intrinsically disordered regions rather than a “wrong” structure.

4.2 A complete receptor – almost

Activity 4.2

Rotate the model: you can see the LRR ectodomain, the single transmembrane helix and the cytoplasmic TIR domain – everything in one chain.
In the UniProt entry, open Family & Domains and click the TIR domain to highlight it on the model.
Resatorvid (TAK-242) binds Cys747. Find it: Hover over it to read its pLDDT.

The TIR domain has high confidence, but look at where it sits relative to the ectodomain. Could it be arranged like that in a cell?

No. In a cell the ectodomain is outside and the TIR domain is in the cytoplasm, on opposite sides of the membrane, with the helix crossing the ~4 nm lipid bilayer. AlphaFold has no membrane, so the relative orientation of the domains is essentially arbitrary – the PAE plot (next) confirms that it is not predicted with any confidence.

4.3 The PAE plot

Activity 4.3

In the confidence panel, open Predicted aligned error (PAE) plot. Dark green = low expected error; white = high.
Two dark squares lie along the diagonal – the ectodomain and the TIR domain. Drag a box over the pale area between them (x ≈ 670–815, y ≈ 30–620). The residues are selected and the mean PAE is shown.

What does the PAE tell you about how the ectodomain and TIR domain are arranged in this model?

The expected error between them is about 27 Å – close to the maximum (~32 Å) – so AlphaFold cannot tell how the two domains sit relative to each other. Within each domain the PAE is low (~5 Å), so each fold is reliable on its own. Interpret domains separately, not their arrangement.

4.4 Prediction versus experiment

Activity 4.4

Make sure 3FXI is loaded too.
Superpose the model on the crystal structure – press Superpose onto… in the confidence panel, or
Colour 3FXI chain A grey and hide chains B and D, to compare.

What RMSD did you get (in Å)?

About 0.7 Å over ~590 Cα atoms – an excellent match (differences below ~1–2 Å are similar to the differences between two crystal structures of the same protein). The prediction of the ectodomain is essentially as good as the experiment. One caveat: 3FXI (released in 2009) was among the structures AlphaFold2 was trained on, so this is not a fully independent test.

What does the AlphaFold model not show that 3FXI does? Why does this matter when thinking about drugs?

No MD-2 partner, no LPS (or any ligand), no second TLR4 (the dimer), no glycans, ions or water. The LPS-binding pocket is in MD-2, so a drug-design project using only the TLR4 model would miss the binding site. Ligands and partners can also change a protein's shape. Complexes need experiments or newer tools that model complexes and ligands (e.g. AlphaFold3).

A residue in a model has pLDDT 45. Which statement is best?

Its side-chain position is accurate to about 1 Å This region is probably flexible or disordered – don't interpret its position The residue is buried in the protein core

Extension: TBP Extension

Activity 4.5

Load the AlphaFold model of TBP: Compare its confidence with the TBP coverage map from Chapter 2.

Which part of TBP is predicted confidently, and which is not? How does this match the experimental structures?

The C-terminal core (≈160–339), which binds DNA, has very high confidence (mean pLDDT ≈ 96) – the same part seen in the X-ray structures. Most of the N-terminal region (≈1–57 and 96–159) has low or very low pLDDT: it is disordered, which is why it is cut off before crystallisation.

The poly-Q tract (58–95) is the surprise: AlphaFold builds it as one long helix with “confident” pLDDT (≈ 80). But the PAE between this helix and the core is ≈ 28 Å, so its position relative to the core is unknown, and no experimental structure shows the helix. Treat confident-looking helices in repetitive sequence with caution.

Make a labelled figure

≈ 10 minWorkbench: 3D viewer

Your coursework asks for a 3D view of your receptor, labelled, with a clear colour code and key. Labels help a reader understand the picture without having to read the legend. Here you will recreate the classic picture of the TLR4–MD-2–LPS dimer.

What makes a good molecular figure

One message. Decide what the picture should show (the overall fold? the binding site?) and hide everything else – extra copies, water, clutter.
Few, meaningful colours – e.g. one colour per molecule – and a key that explains every colour.
Labels on the important parts only, with leader lines where needed.
Orientation that shows the feature clearly (save it as a view).
White background and high resolution for reports (2× or more).
A figure legend that gives the PDB ID and reference, what each colour means, and the software used.

Activity 5.1 · Build the figure

Start afresh with 3FXI.
Give each molecule its own colour: select a chain in the Workspace and click a colour in the . Show LPS and LPS* as sticks.
Rotate until you can see the “m” shape clearly – a side view with the ends that lead into the membrane at the bottom – then save it in Views.
In , press Colour key. Double-click the entries to rename them (e.g. “TLR4”, “TLR4*”). Drag the key to a corner.
Add 3D labels: select each molecule and press 3D text… (or Label residues) – TLR4, MD-2, LPS and the starred copies.
Add a 2D title.
In , download a PNG at 2× (or copy it straight into Word or PowerPoint).
In Views, press Save session so you can carry on later (open it again with Open session).

Schematics

A schematic can show what the structure cannot. 3FXI contains only the extracellular part of the receptor; you can indicate the rest with simple shapes.

Activity 5.2 · Add a schematic optional

With the complex in the side view (), use Box to draw the plasma membrane below the ectodomains, and Ellipse to draw the two TIR domains inside the cell. Label them with 2D text (“plasma membrane”, “TIR domains – not in 3FXI”) and add Arrows if useful.

Prefer PowerPoint? Export the PNG and add labels and shapes there with the text and drawing tools – that is how the original tutorial did it. Either way is fine.

Writing the figure legend

Example legend

Figure 1. Structure of the human TLR4–MD-2–LPS complex. Crystal structure (PDB 3FXI; Park et al., 2009) viewed from the side, with the membrane-proximal ends at the bottom. TLR4 (dark blue) and TLR4* (light blue) are shown as cartoons with their co-receptors MD-2 (green) and MD-2* (light green); LPS molecules are shown as red sticks. LPS binding in MD-2's pocket bridges the two units into an “m”-shaped dimer. Image made with NGL Viewer (Rose et al., 2018).

Write a legend for the figure you made.

Reference

Viewer commands

Everything in the viewer can also be typed in the Command line (bottom of the viewer). Several commands can be joined with ;. Parts of a structure can be named as: a chain (A), a residue or range (A:126, A:264-300), a group from the Workspace (LPS, glycans), sele (the current selection), several of these joined with or, or NGL selection language. Put a structure name first to choose one of several (3FXI A).

Command	What it does
`load 3fxi` · `load af O00206`	Load a PDB entry / AlphaFold model
`show cartoon A` · `hide all water` · `only sticks LPS`	Styles: cartoon, tube, sticks, ballstick, spheres, surface, glass, lines
`undisplay B or D` · `display B` · `isolate A or C` · `showall`	Hide / show parts (like the eye icons)
`color rainbow A` · `color orange C` · `color #1f77b4 LPS`	Schemes: chain, molecule, rainbow, ss, hydrophobicity, restype, element, cpk, plddt, bfactor, default – or any colour
`select C:126` · `select near LPS 4.5` · `select none`	Selections
`site LPS 4.5` · `contacts LPS` · `contacts off`	Binding site and interaction lines
`distance A:264.NE A:1010.O3` · `distance clear`	Measure between two atoms (chain:residue.atom)
`label sele` · `label3d "MD-2" C` · `label2d "Title" 0.05 0.05` · `labels clear`	Labels
`key on` · `key auto` · `key add red "LPS"` · `key title "Key"`	Colour key
`superpose AF-O00206-F1:A onto 3FXI:A`	Superpose chains (reports RMSD)
`center LPS` · `reset` · `spin on` · `background black` · `opacity 0.3`	View settings
`view save "Pocket"` · `view 1` · `camera` · `orient …`	Saved views; print/restore an exact orientation
`export png 2` · `remove 4KXF` · `clear` · `help`	Other

Coming from ICM-Browser?

ICM-Browser	Here
PDB search box / File › Open	PDB ID box / Open file (or drag a file onto the viewer)
Workspace panel, check boxes	Workspace, eye icons
“+” next to a chain shows its sequence	▸ next to a chain shows it in the sequence bar
Select, then display tab (wire, xstick, CPK, ribbon, skin)	Select, then Display tab (Lines, Sticks, Spheres, Cartoon, Surface, See-through)
Colour palette	Colour tab: schemes and single colours
Manual selection (drag a box)	Select tab › Box select
Labels tab: 2D / 3D labels, HQ	Labels & key tab; images are always rendered at high quality on export
Camera button → slides	Views tab › Save view
Shutter button (copy image)	Save image tab › Copy image
Terminal	Command line

Glossary

Accession	Stable identifier of a UniProt entry, e.g. O00206
Swiss-Prot / TrEMBL	Reviewed (expert-curated) / unreviewed (automatically annotated) parts of UniProtKB
Isoform	Protein variant produced from the same gene, e.g. by alternative splicing; one is chosen as “canonical”
PTM	Post-translational modification, e.g. glycosylation, phosphorylation, disulfide bonds
Resolution	Level of detail in an X-ray or cryo-EM structure, in Å – smaller is better
B-factor	How much an atom's position is smeared out (mobility/uncertainty) in an experimental structure
Asymmetric unit	What was deposited: the smallest part of the crystal from which the whole unit cell can be generated by symmetry – may contain several copies
Biological assembly	The form believed to exist in the cell (e.g. a monomer or a dimer)
pLDDT	AlphaFold's per-residue confidence (0–100)
PAE	Predicted aligned error: AlphaFold's confidence in the relative position of two residues (Å)
RMSD	Root-mean-square deviation between matched atoms after superposition (Å)
Hydrogen bond / salt bridge	Polar interactions, typically 2.6–3.5 Å between heavy atoms; a salt bridge is between oppositely charged groups

For teaching staff

Model answers can be hidden for a class: set showModelAnswers: false in assets/js/config.js. Add ?answers to the page address to see them anyway.
The tutorial structures (3FXI, 2Z65, 4G8A, 4KXF, 1VTO, 1CDW, 1C9B) and the TLR4 AlphaFold model are stored in data/ so that chain names and numbers always match the text. All other structures and all UniProt/PDB searches are live.
Any ▶ button can run a list of viewer commands: <button class="do" data-cmd="load 1vto; color ss">. See the README for how to edit the content.

Credits and data

Molecular graphics: NGL Viewer – Rose AS et al. (2018) Bioinformatics 34:3755–3758; Rose AS & Hildebrand PW (2015) Nucleic Acids Res 43:W576–W579 (MIT licence).
Structures from the Protein Data Bank (RCSB PDB, rcsb.org; data CC0). 3FXI: Park BS et al. (2009) Nature 458:1191–1195. 2Z65: Kim HM et al. (2007) Cell 130:906–917. 4G8A: Ohto U et al. (2012) J Biol Chem 287:40611–40617. 4KXF: Hu Z et al. (2013) Science 341:172–175. 1CDW: Nikolov DB et al. (1996) PNAS 93:4862–4867. 1C9B: Tsai FT & Sigler PB (2000) EMBO J 19:25–36. 1VTO: Kim JL & Burley SK (1994) Nat Struct Biol 1:638–653.
Protein information from UniProt (uniprot.org; CC BY 4.0) – The UniProt Consortium, Nucleic Acids Res (2025).
Predicted structures from the AlphaFold Protein Structure Database (CC BY 4.0) – Jumper J et al. (2021) Nature 596:583–589; Varadi M et al. (2024) Nucleic Acids Res 52:D368–D375.
Other references: Poltorak A et al. (1998) Science 282:2085–2088; Trompette A et al. (2009) Nature 457:585–588; Opal SM et al. (2013) JAMA 309:1154–1162; Takashima K et al. (2009) Br J Pharmacol 157:1250–1262.
Adapted from the module's UniProt, PDB, ICM-Browser and “adding labels and schematics” tutorials and worksheets.