======================================================================
NF1-Loss Pan-Cancer Dependency Atlas
Phase 4-5 Evaluation: Drug Targets, PRISM, and Cross-Atlas Comparison
Analyst: analyst | Division: cancer | Date: 2026-03-24
======================================================================

EXECUTIVE SUMMARY
------------------
Phase 4 drug target mapping is built on statistically non-significant
CRISPR data (all per-type targets FDR>0.9; only RIT1 pan-cancer at
FDR=0.071 — and RIT1 is undruggable). PRISM analysis yields no
mechanistically relevant NF1-selective compounds (top hits are NMDA
antagonist, antimalarial, anti-inflammatory). MPNST candidates are
mostly literature-derived, not DepMap-discovered. SHP2 is the only
candidate with BOTH CRISPR evidence (Phase 2) and clinical-stage
drugs. Phase 5 cross-atlas comparison finds 259 "NF1-specific"
dependencies, but the list is contaminated by artifacts.

======================================================================
1. DRUGGABLE DEPENDENCY TARGETS
======================================================================

43 targets at d<-0.3 across all contexts. CRITICAL: zero have FDR<0.1
except RIT1 pan-cancer (FDR=0.071, which has no direct inhibitors).

Credibility-stratified assessment:

TIER 1 — CRISPR-supported + druggable:
  SHP2/PTPN11: d=-0.389 pan-cancer (FDR=0.556), d=-0.808 Lung (FDR=1.0)
    Drugs: TNO155, RMC-4630, GDC-1971 (all Phase 1/2)
    Rationale: ONLY target with Phase 2 validation (FDR=0.017 in
    MAPK baseline). Consistent signal across Lung, Bowel, CNS.
    SHP2 is upstream of RAS — SHP2i+MEKi combos are in clinical trials.
    STRONGEST drug target candidate from this atlas.

  RAF1/C-RAF: d=-0.368 pan-cancer (FDR=0.696), d=-0.602 PNS (FDR=0.97)
    Drugs: LY3009120, Belvarafenib (pan-RAF, Phase 1/2)
    Rationale: Phase 2 validated (FDR=0.052). BRAF-sparing pan-RAF
    inhibitors are in development. RAF1 dependency is mechanistically
    distinct from BRAF dependency in this context.

TIER 2 — CRISPR-suggestive, not FDR-significant:
  CDK6: d=-0.125 pan-cancer (FDR=0.918), d=-0.638 Bowel (FDR=0.90)
    Drugs: Palbociclib, Ribociclib (FDA-approved)
    Rationale: Appears in 5/7 cancer types with d<0, but pan-cancer
    effect is small. Literature supports CDK4/6 in NF1-loss (in vivo
    validated). Disconnect between weak CRISPR signal and published
    efficacy may reflect cell-line vs in-vivo differences.

  CHEK1: d=-0.241 pan-cancer (FDR=0.84), d=-0.729 Bowel (FDR=0.90)
    Drugs: Prexasertib, SRA737 (Phase 1/2)
    TP53 CONFOUND: Cannot attribute to NF1 without TP53 stratification.

  ATR: d=-0.025 pan-cancer (FDR=0.98), d=-0.875 Bowel (FDR=0.91)
    Drugs: Ceralasertib, Berzosertib (Phase 2)
    TP53 CONFOUND: Same issue as CHEK1. Bowel signal is n=6.

  WEE1: d=-0.051 pan-cancer (FDR=0.98), d=-0.691 Breast (FDR=0.96)
    Drugs: Adavosertib, ZN-c3 (Phase 1/2)
    TP53 CONFOUND: Same issue.

TIER 3 — Undruggable but biologically informative:
  RIT1: d=-0.646 pan-cancer (FDR=0.071)
  SHOC2: d=-0.485 pan-cancer (FDR=0.245)
  SPRED1: d=-0.446 pan-cancer (FDR=0.289)
  SPRED2: d=-0.409 pan-cancer (FDR=0.245)
  GRB2: d=-0.418 pan-cancer (FDR=0.522)
  No clinical-stage inhibitors exist for any of these targets.
  SHOC2 inhibitors are in preclinical development (SHP099 analog space).

TIER 4 — No CRISPR support from this atlas:
  EZH2: d=+0.115 pan-cancer (positive — NF1-lost LESS dependent)
  BRD4: d=+0.107 pan-cancer (positive)
  mTOR: d=+0.022 pan-cancer (null)
  PARP1: d=+0.127 pan-cancer (positive)
  These are NOT supported as NF1-specific dependencies by DepMap CRISPR
  data, despite literature interest. Including them in recommendations
  without this caveat would be misleading.

======================================================================
2. PRISM NF1-SELECTIVE COMPOUNDS
======================================================================

1,516 compounds identified. Assessment: NOISE-DOMINATED.

Top 5 hits are biologically irrelevant:
  L-689560 (d=-0.435, FDR=0.64): NMDA receptor antagonist
  bamaquimast (d=-0.409, FDR=0.64): anti-inflammatory (H3R)
  KAF-156 (d=-0.383, FDR=0.93): antimalarial (PfATP4)
  CCG-1423 (d=-0.381, FDR=0.75): Rho/SRF inhibitor
  pyrethrins (d=-0.372, FDR=0.93): insecticide

None of these have plausible NF1/RAS pathway mechanisms. All FDR>0.6.

Key mechanistic drug classes MISSING from PRISM 24Q2:
  - MEK inhibitors (selumetinib, trametinib): NOT FOUND
  - mTOR inhibitors (everolimus): NOT FOUND
  - CDK4/6 inhibitors (palbociclib, ribociclib): NOT FOUND
  - SHP2 inhibitors (TNO155, RMC-4630): NOT FOUND

Only RAF inhibitors were found: encorafenib (d=-0.211, FDR=0.93),
RAF709, belvarafenib, lifirafenib — none NF1-selective.

Verdict: PRISM analysis is uninformative for this atlas because the
relevant drug classes are absent from the dataset. The analysis cannot
validate or invalidate the CRISPR-derived drug targets.

======================================================================
3. LITERATURE CROSS-REFERENCE (PMID 41036607)
======================================================================

15 published NF1-SL compounds matched. 4 with in vivo + selumetinib
synergy validation: everolimus, temsirolimus, AZD8055 (mTORi) and
palbociclib/ribociclib (CDK4/6i).

DISCONNECT: Our DepMap CRISPR data does NOT support mTOR or CDK4/6
as NF1-specific dependencies (both non-significant at pan-cancer level).
This could reflect:
  a) Cell-line screening limitations (monolayer, short-term viability
     vs in-vivo tumor microenvironment effects)
  b) Synthetic lethal interactions requiring drug-level perturbation
     (partial inhibition) rather than complete genetic knockout
  c) Context-specific effects present in vivo but not in vitro

KAT6A/B + Menin inhibitors (WM-1119, PF-07248144) for NF1-loss as
ESR1 resistance mechanism: interesting but orthogonal to the
dependency atlas approach. Not validated by CRISPR data here.

======================================================================
4. MPNST THERAPEUTIC CANDIDATES
======================================================================

6 recommended: MEK, mTOR, CDK4/6, SHP2, BRD4, EZH2.

Evidence assessment per candidate:

  MEK: AGAINST from this atlas. MAP2K1 d=+0.265 (NF1-lost LESS
    dependent). Clinical ORR<10% in MPNST confirms this.
    Recommendation: MEKi only as backbone for combo, not monotherapy.

  mTOR: NO CRISPR SUPPORT. MTOR d=+0.022. Include only based on
    published combo (selumetinib+sirolimus) rationale, not DepMap data.

  CDK4/6: WEAK CRISPR signal. CDK6 d=-0.424 in PNS (FDR=0.98).
    Literature-supported (CDKN2A loss frequent in MPNST). Rational
    but not a DepMap discovery.

  SHP2: SUPPORTED by Phase 2 CRISPR data. Only candidate with direct
    DepMap evidence. SHP2+MEK combo in PNS has biological rationale.

  BRD4: NO CRISPR SUPPORT (d=+0.107). Literature only.

  EZH2: PARADOXICAL. PRC2 loss is common in MPNST (SUZ12/EED
    deletions), which would make EZH2 inhibition irrelevant in those
    lines. Only MPNST lines with intact PRC2 could benefit. Not a
    general MPNST target. No CRISPR support (d=+0.408 in PNS).

Phase 5 MPNST-specific top dependencies (SH2D6, FKBP6, TGM6):
  - Very large effects (d=-2.19, -2.16, -1.99) but ALL undruggable.
  - All FDR>0.9 (except TGM6 FDR=0.13).
  - SH2D6/FKBP6 are poorly characterized genes with no established
    cancer biology. Likely noise from n=8 vs n=27 comparison.
  - These cannot be recommended as therapeutic targets.

======================================================================
5. PHASE 5 CROSS-ATLAS: NF1 vs KRAS COMPARISON
======================================================================

Classification: 1 shared, 259 NF1-specific, 16 KRAS-specific.

Quality assessment of "NF1-specific" list: POOR.
  Top hits are artifacts (KRTAP21-1, SPRR1A, USP9Y, EIF1AY — all
  flagged in Phase 3 as lineage/sex confounds).
  RIT1 ranks #5 — first biologically credible NF1-specific hit.

The shared dependency count (1 gene: ACADM) is suspiciously low.
This likely reflects the stringent thresholds used rather than true
biological independence. ACADM (medium-chain acyl-CoA dehydrogenase)
is a metabolic enzyme with no established RAS pathway role — its
shared status may be coincidental.

Per-tumor-type recommendations contain large effect sizes (d<-2) with
FDR>0.5 — these are noise at n=5-8 and should not be reported as
validated targets.

Key valid finding: NF1-loss and KRAS-mutation DO create distinct
dependency profiles at the pathway level (Phase 2 showed NF1 depends
on GRB2/SHP2/SHOC2 while KRAS typically depends on RAF1/MEK). The
gene-level classification has too many artifacts to be informative
without additional filtering.
