Ligand–RNA interactions — where the annotations come from

These annotations are drawn from several methods; there is no single unified RNA–ligand interaction standard. Each type is tagged by its source:

DSSR base-pairing and motif annotations are RNA-internal and reported separately.

Downloads — what's in the bundle

The Download pair (ZIP) button packages everything needed to reproduce this comparison offline; the card also offers each file individually:

The bundle follows the selected Apo state / Apo copy: switching the Apo state/Apo copy dropdown re-bundles that apo state's superposed apo + record. The morph trajectory and 2D structures aren't zipped — grab them from the individual file links on the card (or the API).

Per-residue displacement (apo → holo)

Each bar is one nucleotide, and its height is how far that residue's C1′ atom moved between the apo and the holo structure after they are superposed — a structure-derived, per-residue apo→holo shift in ångströms. Blue bars are binding-site residues (any atom within ≤ 8 Å of the ligand); grey bars are the rest of the chain. Click a bar to fly the 3D viewer to that residue.

When a pocket spans more than one RNA chain there is one bar chart per chain:

How the superposition is built. Three separate steps — which chain, which algorithm, and what the result is applied to:

1. Pick the reference chain. The host chain = lig_auth_chain, the chain the PDB entry deposits the ligand on.
2. Solve one rigid transform from that chain's paired C1′ atoms. The pairing is sequence-anchored — an apo–holo sequence alignment decides which residue matches which, rather than letting geometry guess (which can lock onto the wrong register in a pseudo-symmetric helix). A Kabsch fit then gives the rotation and translation that minimise the C1′ RMSD of those pairs. A US-align fit is computed as an alternative and the lower-RMSD one is kept; because Kabsch is the analytic minimiser of exactly that quantity, it wins on every served pair.
3. Apply it to the whole apo structure. Every other chain is moved by the same transform and is never re-fitted — that is why a secondary chain's bars measure how well it lands, not how well it could fit on its own.

Base-pair rewiring — formed / broken / unchanged on binding

For each pocket chain we compare the RNA–RNA base pairs in the apo and holo structures - structure-derived from x3dna-dssr - matched one-to-one through the apo–holo sequence alignment (a geometric mapping). The arc diagram draws the holo pairs above the nucleotide axis (holo numbering) and the apo pairs below. (A nucleobase ligand such as SAM or adenine also base-pairs to the RNA, but that is ligand recognition - shown in the Ligand–RNA interactions panel — not secondary-structure rewiring, so it is excluded here.)

Dashed arcs are non-canonical (Leontis–Westhof) pairs — anything other than a cis Watson–Crick / wobble (cWW); these matter most for structured RNA pockets. This is a descriptive comparison of two experimental endpoints, not a kinetic pathway or binding mechanism.

Partner swaps (the ⇄ marker). Some nucleotides keep a base pair but with a different partner in the holo than in the apo. Such a nucleotide carries a formed arc above and a broken arc below that meet at the same node — a register shift (a slide to an adjacent partner) or a longer-range re-pairing. These often accompany a reorganisation of the pocket, e.g. the junction rearrangement of an adenine riboswitch. Each swap node is drawn with an accent dot on the axis (hover for the old and new partner) and flagged with a ⇄ badge in the Changes table.

Two-strand (inter-strand) view. Base pairs whose two ends lie on different chains - the ribosomal A-site decoding duplexes and the inter-strand rungs of multi-chain pockets — can't sit on one axis. Pick “chains X ⇆ Y · inter-strand” in the chain selector (the default when a pocket has no single-chain pairs) to draw them as a ladder. Chain X sits on a top axis and chain Y on a bottom axis drawn 3′→5′, so an antiparallel duplex reads as vertical rungs. Each cross-strand pair is a rung coloured formed / broken / unchanged, and each chain's own intra-chain pairs arch outward. This shows, e.g., whether an aminoglycoside binding the A-site actually rewires the decoding duplex.

Reading the axis. The nucleotide axis is numbered by the holo residue (auth) numbers. Hover any nucleotide to see its base and number and the apo residue it aligns to (matched · ≠ a differing base · or no aligned residue). In the zoomed-in (non-fit) view a two-row sequence ribbon (holo H / apo A) shows the bases side by side - coloured by match / differing base / gap, with binding-site residues underlined; the two-strand view puts one ribbon beside each strand. Both come from this pair's own apo–holo alignment, so they agree with the arcs exactly.

Sequence differences (the ≠ marker). Apo and holo are matched at ≥90% sequence identity, not necessarily 100% — so some pairs come from slightly different constructs (often different organisms or engineered variants). A formed / broken change whose endpoints sit at an alignment column where the apo and holo nucleotides differ could reflect that sequence difference rather than the ligand: we draw it faded and mark it ≠. The caption reports how many of the changes fall at identical-sequence positions (identical nucleotide on both sides, so the rewiring is genuinely conformational) versus differing positions. Across the dataset ~89% of changes are at identical-sequence positions.

Base-pair rewiring — inter-strand (two-strand) pairs

This pair's base pairs run between two separate strands rather than within a single chain — typically the short ribosomal A-site fragments (~9–23 nt), which are self-complementary duplexes. x3dna-dssr resolves these inter-strand pairs, and they are drawn in the two-strand ladder view: pick “chains X ⇆ Y · inter-strand” in the chain selector (it is the default when a pocket has no single-chain pairs).

Each chain becomes one axis (the lower one drawn 3′→5′) and every cross-strand pair a rung, coloured formed / broken / unchanged like the single-chain arcs.

RNA type — how it's assigned

The Type is the curated functional class of this pocket's host RNA chain, taken from the Nucleic Acid Knowledgebase (NAKB) - an expert-curated annotation of every nucleic-acid structure in the PDB, not a keyword match on the deposited title. Each pair is annotated by resolving the host chain to its mmCIF entity and reading NAKB's per-entity functional classification (e.g. function › riboswitch › purineswitch); the label shown is the most specific NAKB term.

When NAKB assigns no functional class (bare model duplexes / quadruplexes) the type falls back to NAKB's structural descriptor (e.g. A-form double helix, Quadruplex). For the few structures NAKB does not annotate at all, it falls back to a keyword match on the deposited _struct.title. NAKB integrates external RNA resources (RNAcentral, Rfam, BGSU).

Local vs Global identity – two ways to compare the apo and holo sequences

Every pair carries two apo ⇆ holo sequence-identity numbers. They answer different questions, so they can differ by several points on the same pair.

A worked example – 1EHT. The holo and apo declared sequences differ at a single 3′ position (holo …GGCAGCGUC, apo …GGCAGCGCC). BLAST trims that overhang out of the aligned region and reports pident 100% over the 31 columns it aligns; the global alignment keeps all 33 columns and scores 32/33 = 97%. Neither is wrong – they are measuring different windows.

Global identity is computed from the declared (SEQRES) sequences, not the modeled coordinates, so unmodelled (disordered) residues do not penalise it: a structure that resolves only part of its construct still scores 100% against an identical construct. Where a pocket spans more than one RNA chain, each chain carries its own pair of numbers, shown on that chain’s own alignment header below. The single value published for the pair – in the table, the filters and the downloads – is the ligand’s own (auth) chain’s. That is the chain the PDB entry assigns the ligand to (what RCSB prints as [auth B], and the third token of the pocket identifier). It was the minimum over the chains until 2026-07-28, which let one loosely-related interface strand set the published value for a pair whose chain under study matched far better. It is a deposited fact rather than a geometric tie-break, so anyone can re-derive it from the PDB entry. The two per-chain sets are independent – they need not cover the same chains, so read them separately rather than pairing them up chain by chain; where the upstream BLAST run never keyed a chain, that chain simply shows no local identity.

Both identities are available for an alternative apo state as well: each alternative's own BLAST pident is looked up in the same upstream run. Local identity is therefore blank only for the few alternatives that run never scored – and for query-page uploads, which are not part of it. Global identity is computed here, so it is always available. Both are sequence comparisons only. The alignment box below is the visual counterpart of Global identity: it aligns the same declared (SEQRES) sequences, marking each position as modeled, declared-but-unmodeled, or a true indel – see its own ? – so its column tally agrees with the Global identity figure. On a pocket spanning several RNA chains there is one box per chain, and each box reports its own chain: the numbers on a chain’s header are that chain’s, never an aggregate over the pocket. The value published for the pair is the ligand’s own (auth) chain’s, so it is the number beside that chain’s box – not the lowest of them. A chain that scores worse is still shown, on its own header, where it belongs. (If you narrow the binding-site cutoff far enough to drop the ligand’s chain out of view, no box claims to carry the published value: a distance control must not re-home a sequence metric.)

Reading the apo vs holo alignment — modeled, unmodeled & true indels

The box below aligns the two declared (SEQRES) sequences — the full deposited constructs, with modified nucleotides resolved to their parent base — by Needleman–Wunsch in your browser. That is the same basis as the Global identity printed on this chain’s own header just above it, so the two agree: if the box shows no mismatch, that chain’s global identity is 100%. (The Local identity beside it is the upstream BLAST pident, measured only inside the BLAST-aligned region, so terminal overhangs are excluded — see the ? on that same header for the difference.) Every position falls into one of the three classes below. Nucleotide-analog ligands that x3dna-dssr folds into the strand (e.g. a bound SAM or nucleotide) are excluded from this polymer alignment and its residue count, but still appear in the 2D structure view.

Hover a modeled residue for its original PDB number. Because the box is built on SEQRES coordinates, disorder at different positions in apo and holo can no longer be mistaken for a mismatch – what you see here is the same comparison the Global identity reports.

This applies to every apo shown for a pocket: the pocket's default apo, an alternative apo state and an apo copy selected from the dropdowns are all indexed on their own declared sequence. All three therefore read on the same basis, and the three symbols above always mean what they say. (A structure you upload on the Query page has no deposited declared sequence, so that panel aligns the modeled residues only and says so in its caption.) On a pocket spanning more than one RNA chain there is one box per chain, and each box is scored on its own chain. The Global identity printed on a chain’s header is that chain’s own, and the box below it tallies exactly those columns. The single value published for the pair – in the table, the filters and the downloads – is the ligand’s own (auth) chain’s (what RCSB prints as [auth B], and the third token of the pocket identifier). The published figure is therefore the one beside that chain’s box. It is not a minimum, a mean or any other aggregate over the chains; see the identity ? for how the two metrics are measured.

Reading this alignment — modeled residues only

This panel has no declared (SEQRES) sequence to index on — it happens for structures uploaded on the Query page, which are not deposited entries. The box therefore aligns the residues that are resolved in the 3D model instead of the full deposited constructs. That is a different basis from the Global identity printed on the header above it, so the two numbers are not comparable: cite the identity figure, not this column tally.

Every pair served from the database itself uses the declared-sequence box instead – see its ? for the three-class reading (modeled / declared-but-unmodeled / true indel).

Secondary structure & base pairs — what x3dna-dssr extracts

We run x3dna-dssr on each 3D structure (both apo and holo). From the atomic coordinates — derived from the experimentally determined model, not predicted from sequence — it reports:

Leontis–Westhof code: each base has three edges - Watson–Crick, Hoogsteen, Sugar; a pair is named by cis/trans glycosidic orientation + the two edges (12 families). e.g. cWW = the standard A–U/G–C/G–U pairs; tHS = a common non-canonical pair. Read from 3D, so non-canonical pairs the dot-bracket can't show still appear in the table.

“All chains (joined)” view — the inter-chain connector lines

When a binding pocket spans more than one RNA chain (e.g. an aminoglycoside in a two-strand A-site duplex), the All chains (joined) option draws every pocket chain as one horizontal strand (5′→3′) and connects the base pairs between chains with lines. Two base-paired chains form an antiparallel ladder (partner strand reversed so paired bases line up in columns); three or more chains are stacked and joined by straight connectors.