======================================================================
NF1-Loss Pan-Cancer Dependency Atlas
Phase 3 Critical Assessment: Genome-Wide Differential Dependency Screen
Analyst: analyst | Division: cancer | Date: 2026-03-24
======================================================================

EXECUTIVE SUMMARY
------------------
The genome-wide screen is severely underpowered at the per-cancer-type
level and modestly underpowered pan-cancer. Only 1 hit meets the
FDR<0.1, |d|>0.5 threshold (RIT1, which is biologically plausible).
Zero per-cancer-type hits survive FDR correction. The top hit list
contains a mix of credible biology (RIT1, SHOC2, SPRED1/2) and clear
artifacts (KRTAP21-1, SPRR1A, USP9Y). TP53 confounding remains
uncontrolled. The study does NOT meet the success criterion of >=5
NF1-loss-specific dependencies.

======================================================================
1. STATISTICAL ADEQUACY
======================================================================

Total tests: 160,947 (9 contexts x ~18,000 genes)
FDR method: Benjamini-Hochberg (appropriate)

Power assessment:
  Per-cancer-type (n=5-18 lost):
  - n=5-6: <15% power for d=0.8 at genome-wide FDR<0.1
  - n=10: ~20% power for d=0.8
  - n=18 (CNS): ~35% power for d=0.8
  Verdict: Per-type analyses are FUNDAMENTALLY UNDERPOWERED for
  genome-wide discovery. No amount of effect size can overcome the
  multiple testing penalty at these sample sizes. All per-type hits
  with FDR>0.5 are expected under the null.

  Pan-cancer RAS-excluded (n=67 lost):
  - ~70% power for d=0.8 at genome-wide FDR<0.1
  - ~30% power for d=0.5
  Verdict: Adequately powered for large effects only. The single hit
  (RIT1 d=-0.646) is consistent with borderline detection power.

Multiple testing burden: With 160,947 tests and BH-FDR, a raw p-value
of ~4e-06 is needed for FDR<0.1. This is appropriate but means only
very strong, consistent signals can be detected.

======================================================================
2. PAN-CANCER HIT ASSESSMENT (RAS-EXCLUDED)
======================================================================

Hit #1: RIT1 (d=-0.646, FDR=0.071, composite=0.74)
  Biological plausibility: HIGH
  - RIT1 is a RAS-family GTPase (RRAS subfamily) that activates
    RAF/MEK/ERK and PI3K/AKT pathways.
  - NF1 is a GAP for RAS but has minimal activity on RIT1 (different
    GTPase domain specificity). Thus, NF1-loss does NOT directly
    hyperactivate RIT1 — rather, NF1-lost cells may become more
    dependent on RIT1 as a parallel pathway input to maintain
    sufficient MAPK flux when upstream RAS regulation is disrupted.
  - RIT1 mutations are oncogenic in lung adenocarcinoma (PMID:25049390),
    confirming it as a driver in the RAS pathway.
  - RIT1 is per-type strongest in Breast (d=-1.115) and Lung (d=-0.446),
    absent in PNS (d=+0.188). Not MPNST-relevant.
  Confound risk: LOW. Not a TP53-associated dependency.
  Druggability: LOW. No direct RIT1 inhibitors exist. SOS1 inhibitors
    (BI-3406) may partially address RIT1 activation but specificity is
    unclear.
  Verdict: CREDIBLE but SOLE high-confidence hit. Insufficient alone.

Sub-FDR hits warranting attention (FDR 0.2-0.3):

  SHOC2 (d=-0.485, FDR=0.245):
  - Already validated in Phase 2. Consistent signal.
  - Mechanistically linked to RAF1 activation pathway.
  - Novel SHOC2 inhibitors in preclinical development.
  Assessment: CREDIBLE, consistent with Phase 2 baseline.

  SPRED1 (d=-0.446, FDR=0.289) and SPRED2 (d=-0.409, FDR=0.245):
  - SPRED proteins are RAS/MAPK negative regulators.
  - SPRED1 physically interacts with NF1 to enhance RAS-GAP activity.
  - Biological interpretation: NF1-lost cells retain partial MAPK
    pathway regulation through SPRED. Additional loss of SPRED pushes
    RAS/MAPK hyperactivation beyond a tolerable threshold, consistent
    with "RAS toxicity" models of synthetic lethality.
  - SPRED1 shows strongest signal in CNS (d=-1.069) and Uterus
    (d=-1.163) — large effects, per-type underpowered.
  Assessment: BIOLOGICALLY COHERENT. Top-tier for follow-up despite
  marginal FDR. SPRED-NF1 functional interaction is well-established
  (PMID:28135245).

  RAD51 (d=-0.402, FDR=0.296):
  - DNA repair (homologous recombination). NF1 has reported roles in
    DNA damage response (PMID:24573287).
  - BUT: 71.5% TP53 co-mutation confound. TP53-null cells show
    increased RAD51 dependency independent of NF1.
  Assessment: INCONCLUSIVE. Cannot attribute to NF1 without TP53
  stratification.

  EIF2B5 (d=-0.455, FDR=0.245):
  - Translation initiation factor. No known connection to NF1/RAS.
  Assessment: UNCLEAR biological rationale. May be noise.

  SRRM5 (d=-0.567, FDR=0.211):
  - Splicing regulator. No established NF1 connection.
  Assessment: UNCLEAR. Possibly novel but lacks mechanistic support.

ARTIFACT FLAGS (top list contamination):

  SPRR1A (d=-0.728, FDR=0.522):
  - Small proline-rich protein expressed in skin keratinocytes.
  - NF1 loss is enriched in melanoma/cutaneous tumors.
  - LINEAGE CONFOUND: likely reflects NF1-loss enrichment in skin
    lineages where SPRR1A is constitutively essential, not a true
    NF1-specific dependency.
  Assessment: ARTIFACT. Exclude from findings.

  KRTAP21-1 (d=-0.756, FDR=0.562):
  - Keratin-associated protein. Same lineage confound as SPRR1A.
  Assessment: ARTIFACT. Exclude.

  USP9Y (d=-0.666, FDR=0.522):
  - Y chromosome gene. SEX IMBALANCE confound (NF1-lost group may
    have different M:F ratio than intact).
  Assessment: ARTIFACT. Exclude.

  OR10T2 (d=-0.509, FDR=0.403):
  - Olfactory receptor. No biological rationale for cancer dependency.
  - Likely CNV artifact or mapping error.
  Assessment: ARTIFACT. Exclude.

======================================================================
3. PER-TUMOR-TYPE SIGNALS
======================================================================

Zero per-tumor-type hits reach FDR<0.1.

Notable large effects (all FDR>0.9, treat as EXPLORATORY ONLY):

  Bowel (n=6/37):
  - ATR:    d=-0.875 — Large, but FDR=0.91. n=6 is inadequate.
  - EZH2:   d=-0.778 — FDR=0.90. TP53 confound (6/6 TP53-mut).
  - CHEK1:  d=-0.729 — FDR=0.90. TP53 confound + small n.
  - CDK6:   d=-0.638 — FDR=0.90.
  - PTPN11: d=-0.633 — FDR=0.91. Consistent with pan-cancer signal.

  Lung (n=10/90):
  - PTPN11: d=-0.808 — Large but FDR=1.0. Per-type FDR is extremely
    conservative here due to ~18,000 tests with n=10.
  - CDK6:   d=-0.507 — FDR=1.0.

  Breast (n=5/34):
  - RIT1:   d=-1.115 — Largest RIT1 effect, but n=5, FDR=0.95.

  CNS/Brain (n=18/51):
  - SPRED1: d=-1.069 — Strongest SPRED1 signal. FDR=0.64.
  - PTPN11: d=-0.585 — Consistent with pan-cancer. FDR=0.98.
  - SPRED2: d=-0.833 — Strong. FDR=0.98.

  PNS (n=8/27):
  - CHEK1:  d=-0.706 — FDR=0.97. TP53 confound (7/8 TP53-mut).
  - RAD51:  d=-0.649 — FDR=0.97. Same TP53 concern.
  - RAF1:   d=-0.602 — FDR=0.97. Consistent with Phase 2.

Interpretation: Per-type signals are NOISE at the statistical level
(all FDR>0.6). However, some show consistency with pan-cancer trends
(PTPN11 in Lung/CNS/Bowel; SPRED1/2 in CNS; RAF1 in PNS). These
directional consistencies increase confidence in pan-cancer findings
but do NOT independently validate them.

======================================================================
4. HYPOTHESIS GENE SET ASSESSMENT
======================================================================

All hypothesis gene sets show FDR>0.5 at pan-cancer level. None are
statistically significant NF1-loss dependencies.

The heatmap shows highly heterogeneous patterns across cancer types,
with effects often pointing in opposite directions for the same gene
in different lineages. This heterogeneity explains why pan-cancer
pooling dilutes per-type signals.

mTOR/PI3K pathway: No signal. RPTOR is the "best" at d=-0.169.
  This argues against mTOR cross-talk as a major NF1-loss dependency,
  despite clinical interest in MEK+mTOR combinations.

Cell cycle (CDK4/6): CDK6 d=-0.125 (FDR=0.92) is weak. CDK4 positive.
  Per-type CDK6 signals in Bowel (d=-0.638) and Lung (d=-0.507) are
  underpowered. Does not support CDK4/6 as NF1-specific dependency.

Epigenetic (PRC2, BRD4): No signal pan-cancer. EZH2 d=+0.115.
  Per-type EZH2 signal in Bowel (d=-0.778) is isolated. Not NF1-specific.

DDR (ATR, CHEK1, WEE1): CHEK1 shows the most consistent signal
  (d=-0.241 pan-cancer, d=-0.729 Bowel, d=-0.706 PNS) but all FDR>0.8.
  CRITICAL: 71.5% TP53 co-mutation makes ALL DDR signals suspect.
  TP53-null cells are known to be synthetically lethal with ATR/CHEK1
  (PMID:22137880). Cannot attribute to NF1 without TP53 adjustment.

RAS feedback: SPRED1/2 are the notable signals (discussed above).
  DUSP4/6 show positive d (NF1-lost LESS dependent) — inconsistent
  with simple feedback model. RASA2 positive (d=+0.241).

======================================================================
5. TP53 CONFOUNDING IMPACT ON PHASE 3
======================================================================

The 71.5% TP53 co-mutation rate is expected to inflate signals for:
  - DDR genes (ATR, CHEK1, WEE1): DIRECT confound. Known TP53-SL genes.
  - RAD51: Homologous recombination dependency is TP53-context-dependent.
  - CDK6: Cell cycle regulation is TP53/RB pathway-dependent.
  - EZH2: PRC2 dependency can be TP53-driven in some contexts.

Genes UNLIKELY to be TP53-confounded:
  - RIT1: No established TP53-RIT1 synthetic lethal interaction.
  - SHOC2: RAS scaffold, not TP53-related.
  - SPRED1/2: RAS negative regulators, not TP53-related.
  - GRB2, PTPN11: Upstream MAPK, not TP53-related.

Recommendation: The TP53-stratified analysis (developer task #1335)
is ESSENTIAL before DDR/cell-cycle hits can be reported. MAPK pathway
hits (RIT1, SHOC2, SPRED1/2, GRB2, PTPN11) are more credibly
NF1-specific.

======================================================================
6. OVERALL PHASE 3 VERDICT
======================================================================

LARGELY NEGATIVE but with credible biology in the sub-threshold hits.

Credible NF1-loss dependencies (RAS/MAPK axis):
  1. RIT1 (d=-0.646, FDR=0.071) — ONLY hit meeting success criteria
  2. SHOC2 (d=-0.485, FDR=0.245) — consistent with Phase 2
  3. SPRED1 (d=-0.446, FDR=0.289) — strong biological rationale
  4. SPRED2 (d=-0.409, FDR=0.245) — same pathway, independent signal
  5. GRB2/PTPN11 — from Phase 2, confirmed by genome-wide trends

The emerging picture is that NF1-loss creates dependency on the
RAS/MAPK regulatory network (RIT1, SHOC2, SPRED1/2, GRB2, SHP2)
rather than on downstream effectors (MEK, ERK) or cross-pathway
targets (mTOR, CDK, PRC2). This is biologically coherent but
statistically under-powered for formal validation.

Success criteria status: 1/5 hits at FDR<0.1 (needs >=5). NOT MET.
However, the 4 additional RAS-axis hits at FDR 0.2-0.3 form a
mechanistically coherent gene set. If treated as a pathway rather
than individual genes, significance may improve via GSEA-style
analysis.
