CRISPR-Cas immunity and anti-CRISPR counter-defence

SCOPING BIOLOGY_DOMAIN

Species: BPZF4

Genes: AcrF8

CRISPR-Cas immunity and anti-CRISPR counter-defence

Two module documents, curated as a matched pair: the host pathway
(crispr_cas_adaptive_immunity) and
the phage counter-defence that targets it
(anti_crispr_suppression). Keeping them
separate is the substantive modelling decision — each is reusable alone, and the
pairing makes visible that anti-CRISPR mechanisms are individuated by which step
of the host pathway they attack, not by homology, because they have none.

Why two modules rather than one

Suppression could have been an optional regulatory part of the host module. It is
not, for three reasons:

  1. Different organisms. The host module is bacterial and archaeal
    (NCBITaxon:2, NCBITaxon:2157); the suppression module is viral
    (NCBITaxon:10239), acting in GO:0030430 host cell cytoplasm. A single
    module would have to assert both taxon scopes at once.
  2. Different GO branches. The host pathway sits under GO:0099048 CRISPR-cas
    system → GO:0099046 clearance of foreign intracellular nucleic acids.
    Suppression sits under GO:0098672 → GO:0052031 symbiont-mediated
    perturbation of host defense response. These are not parent and child; they
    are assertions about two different agents.
  3. Independent reuse. The host module is the right target for curating any
    cas locus; the suppression module is the right target for any phage genome.
    Folding one into the other would make each half unusable on its own.

The cost is that connections are document-scoped, so no edge can be drawn from
an Acr node to the host node it inhibits. Each suppression part names its host
target in prose instead, and the correspondence is tabulated below.

The attack surface, step by step

Every characterised anti-CRISPR class maps onto exactly one step of the host
pathway. That is the organising claim of the pair, and it is falsifiable: a
mechanism that attacked no step of the host module would mean the host module is
missing a part.

Host step (host module) Suppression part Mechanism Exemplar
Adaptation (Cas1–Cas2 integration) none curated — —
crRNA biogenesis enzymatic subversion guide-RNA cleavage AcrVA1 (UniProtKB:A0A5H1ZR47)
Interference — target recognition recognition blockade Csy-complex and crRNA binding AcrIF8 (UniProtKB:H9C181)
Interference — target recognition recognition blockade DNA mimicry at the PAM site AcrIIA4 (UniProtKB:A0A247D711)
Interference — Cas3 recruitment nuclease neutralisation Cas3 sequestration (GO:0140311) AcrF3-family (UniProtKB:L7P7R7)
Interference — Cas9 cleavage nuclease neutralisation HNH-domain occlusion AcrIIC1 (UniProtKB:A0A2D0TCG3)
Interference — cOA signalling enzymatic subversion cOA ring nuclease AcrIII-1 (UniProtKB:Q8QL27)

Two asymmetries in that table are worth reading as biology rather than as gaps in
the curation.

Nothing attacks adaptation. Every characterised Acr acts on biogenesis or
interference — the stages that happen during the current infection. Blocking
adaptation would spare the phage's descendants but not the phage, which is the
wrong incentive; killing the effector is what saves the infecting genome. This is
a prediction, not just an absence: an Acr that blocked Cas1–Cas2 integration
should be expected in temperate phages and prophages, where the element persists
in a lineage whose future immunity it has an interest in, rather than in lytic
ones.

Cas3 sequestration has no class 2 counterpart. It cannot have one. In class 1
systems recognition and destruction live in different proteins, so there is a
recruitment step to intercept; in class 2 they live in the same protein, and the
equivalent attack has to be domain occlusion (AcrIIC1) instead. The mechanism
repertoire is constrained by the host architecture, which is an argument for
modelling the effector classes as variants of one slot rather than as separate
modules — the variants are what make the constraint visible.

Stoichiometric versus catalytic inhibition

The module separates blockers from enzymes, and the distinction does real work.
AcrIF8, AcrIIA4, AcrF3 and AcrIIC1 each occupy their target and must therefore be
present in molar excess of it. AcrVA1 and AcrIII-1 are enzymes: they destroy the
crRNA guide and the cOA messenger respectively, and one molecule can disarm many
targets. This predicts different expression requirements — and bears directly on
the Aca autoregulatory loop, since a catalytic Acr can tolerate much earlier
repression than a stoichiometric one.

The Aca feedback loop

aca genes are co-transcribed with acr genes from a shared promoter, and the
Aca repressor then shuts that promoter down: the inhibitor's own expression
produces the switch that turns it off. The module records this as a closed loop
(recognition_blockade → aca_autoregulation → recognition_blockade), which the
renderer's feedback check confirms. Pectobacterium phage ZF40 supplies both
halves from one locus — acrIF8 (UniProtKB:H9C181, reviewed under
genes/BPZF4/AcrF8/) and aca2 (UniProtKB:H9C180) — which is why it is the
exemplar for the regulatory part.

Ontology gaps

Six knowledge gaps are recorded across the two modules. Three are ontology gaps
that between them make the defining properties of this entire domain
unrepresentable in GO:

Where no term exists, these modules assert no id. That is a deliberate choice: an
omitted id says "not established", whereas a plausible-looking wrong id says
something false in a field other tooling will believe.

The remaining gaps are curation-shaped. Most characterised Acr proteins have no
reviewed UniProt entry, and two of the exemplars used here have published
structures bound to their targets yet are named only "Phage protein" (L7P7R7)
and "Anti-CRISPR protein" (A0A2D0TCG3). InterPro has caught up faster, with
mechanism-bearing family names such as IPR049085 "Cas3 inhibitor AcrF3-like" —
which is why both modules ground families in InterPro and leave the member labels
as fetched.

Status and next steps

Both modules validate and render with no leaf-grounding gaps. All 35 UniProt
groundings now have gene reviews — the 33 added in this batch plus the two that
existed. Nine are DRAFT rather than COMPLETE, deliberately: the Cascade and
class 2 sets each turn on a judgement call that wants a second opinion rather
than on anything mechanical.

Five ontology terms are drafted across the reviews, each argued as a missing
member of an existing series rather than a one-off: cyclic oligoadenylate
synthase activity and cyclic oligoadenylate binding (type III), guide
RNA-directed target recognition and guide RNA-dependent nucleic acid
endonuclease activity (kept as a complementary pair, since a catalytically dead
but guide-loaded effector retains the first and loses the second), and nucleic
acid mimicry-based competitive inhibitor activity (converged on independently by
three sequence-unrelated anti-CRISPR families). A CRISPR RNA-guided surveillance
complex CC term is requested too, on the finding that GO's entire CRISPR
vocabulary is biological-process terms.

Reviewing the exemplars corrected the modules in eight places, which is the
argument for doing it: Cas2 does have a molecular function, Cas3's asserted
function was the wrong one of its two activities, Cas12a's GOA term imports
DNase I chemistry it does not have, AcrF2's blocked surface is established and
makes it the strongest cross-class mimicry case, AcrVA1's catalyst is not
established enough to be a family membership criterion, GO:0140721 covers
AcrIIC1, both complexes can be grounded at GO:1990904, and the AcrIII-1
accession was resolvable after all.

The remaining work is curation of the drafted terms with GO, and the open
questions each review records.

Neither module has a -deep-research-*.md report, and neither do the gene
reviews: no provider API keys are configured in this environment, so findings
went to GENE-notes.md files with verbatim [PMID:x "..."] provenance instead
of a fabricated provider file. The QC panel reports the absence.