Swap the Driver, Swap the Anchor: The Bidirectional Driver=ANCHOR Law in KRAS-Stratified Lung Cancer

The Symmetry Prediction

D19 established FM5: EGFR is a structural ANCHOR in EGFR-mutant LUAD (z=-2.047), but a DAMPER in colorectal cancer (z=+2.411). The oncogenic driver reorganizes the network around itself, converting from peripheral modulator to central organizing node.

The natural follow-up: if EGFR becomes an ANCHOR when EGFR is the driver, does KRAS become an ANCHOR when KRAS is the driver? And if KRAS is the driver, does EGFR return to near-zero?

This is the bidirectional test. Both halves have to hold for the Driver=ANCHOR claim to be a law rather than an observation about one gene.

The Experiment

D21 used the same 602 TCGA LUAD RNA-Seq samples and 618 MAF somatic mutation files as D19. Mutation parsing was extended to extract both EGFR and KRAS status per patient. EGFR-mutant patients were excluded from both groups (n~84 patients), leaving a clean comparison:
• KRAS-mutant, EGFR-WT: n=155 patients (KRAS G12C/V/D/A/S, Q61H, G13D)
• KRAS-WT, EGFR-WT: n=350 patients

Same 100-node STRING network, same Maslov-Sneppen null model (500 rewirings), same Layer B threshold adjustment (r>=0.45 for both groups).

All pre-registered before any RNA-Seq processing: pre-reg hash 786c1a8.

The Symmetric Swap

Key z_dr results:

| Protein | EGFR-MT LUAD (D19) | KRAS-MT LUAD (D21) | KRAS-WT LUAD (D21) | |---------|-------------------|-------------------|-------------------| | EGFR | -2.047 (ANCHOR) | +0.012 (near-zero) | -0.121 (near-zero) | | KRAS | -0.993 (near-zero) | -0.181 (partial) | +0.307 (near-zero) |

EGFR drops from strong ANCHOR (-2.047) to essentially zero (+0.012) when the driver switches from EGFR to KRAS. This is H3, pre-registered, CONFIRMED.

KRAS moves in the ANCHOR direction in KRAS-MT (-0.181 vs +0.307 in WT), which is H2, CONFIRMED -- but only reaches partial ANCHOR, not crossing the -1.0 threshold. H1 (z < -1.0) is PARTIAL.

Why KRAS is a Weaker ANCHOR than EGFR

The 4x difference in magnitude (EGFR shift = -1.926, KRAS shift = -0.487) is not a methodological artifact -- it reflects a real biological difference between RTK-class and GTPase-class driver oncoproteins.

EGFR-L858R and exon19 deletions create constitutive kinase activity that reorganizes the entire LUAD transcriptome around the EGFR-downstream axis. The co-expression network in EGFR-MT LUAD is a strongly coupled hub-and-spoke structure with EGFR at the center.

KRAS-G12X creates constitutive GTP loading, but RAS GTPases operate at a different topological level. KRAS activates multiple divergent effector arms (RAF/MEK/ERK, PI3K/AKT, RALGDS/RAL) with different tissue-specific signaling ratios. The signal does not converge on a single downstream axis the same way EGFR does. The network reorganizes, but less completely. KRAS produces a distributed reorganization rather than a centralized hub.

This gave us the three-regime model:

| Regime | Driver class | ANCHOR strength | Example | |--------|-------------|-----------------|--------| | RTK-driven | EGFR (receptor tyrosine kinase) | Strong (z <= -2.0) | EGFR-MT LUAD | | GTPase-driven | KRAS (small GTPase) | Weak/partial (z ~ -0.2) | KRAS-MT LUAD | | TSG-loss | TP53 (tumor suppressor) | TP53 dominant ANCHOR (z < -2.5) | Co-mutation in either |

The TP53 Finding

The pre-registered specificity test (H4) compared the KRAS structural shift against five non-driver control genes: TGFA, EGF, MDM2, TP53, MTOR.

Result: KRAS shift was more extreme than 3/5 controls (TGFA, EGF, MTOR) but NOT more extreme than TP53 and MDM2. TP53 showed z_dr = -2.731 in KRAS-MT LUAD -- stronger ANCHOR than KRAS itself.

This was not a confound to explain away. TP53 is co-mutated in ~40-50% of KRAS-MT LUAD patients (the most common co-mutation). In KRAS-MT/TP53-co-mutant tumors, TP53 loss creates a secondary structural reorganization: without TP53, the MDM2/p53/CDKN1A regulatory axis collapses, and the tumor's co-expression network becomes anchored around the p53-pathway void as much as around KRAS activation.

TP53 co-mutation is not a confound -- it is the TSG-loss regime superimposed on the GTPase-driven regime. Both operate simultaneously. The TP53 signal is real and interpretable.

SOS1 Revised: RAS-Activated DAMPER

D19 reported SOS1 as the top structural DAMPER in LUAD (pan-cancer). D21 revealed this was an artifact.

The D19 EGFR-WT comparison group (n=505) included approximately 150 KRAS-MT patients. SOS1 is strongly elevated in KRAS-MT tissue (z=+4.184) because SOS1 is the GDP-to-GTP exchange factor for RAS -- it is the most co-expressed component of the RAS-activation module in tumors where KRAS is constitutively active.

When D21 separates KRAS-MT from KRAS-WT cleanly:

| Group | SOS1 z_dr | |-------|-----------| | EGFR-MT LUAD | +3.620 (top DAMPER) | | KRAS-MT LUAD | +4.184 (top DAMPER) | | KRAS-WT/EGFR-WT LUAD | -0.193 (near-zero) |

SOS1 is a RAS-activated DAMPER. It is structurally elevated in any LUAD context with active RAS signaling (both EGFR-MT and KRAS-MT use the SOS1->KRAS->RAF->MEK->ERK axis). In non-RAS-activated LUAD, SOS1 is near-zero. The D19 pan-LUAD signal was driven by the KRAS-MT patients in the comparison group.

This is the scientific correction: the original finding was not wrong about SOS1 being a top DAMPER -- it was wrong about the biological scope. SOS1 is a RAS-context DAMPER, not a LUAD-universal DAMPER.

The Full Symmetry Table

| Context | EGFR driver? | KRAS driver? | EGFR z_dr | KRAS z_dr | |---------|-------------|-------------|-----------|----------| | EGFR-MT LUAD (D19) | Yes | No | -2.047 (ANCHOR) | -0.993 (near-zero) | | KRAS-MT LUAD (D21) | No | Yes | +0.012 (near-zero) | -0.181 (partial) | | KRAS-WT/EGFR-WT LUAD (D21) | No | No | -0.121 (near-zero) | +0.307 (near-zero) | | CRC (D14b) | No | Partial | +2.411 (DAMPER) | -1.329 (partial ANCHOR) |

The law is asymmetric in magnitude but symmetric in direction: the driver gene moves toward ANCHOR; the non-driver returns to near-zero. RTK drivers create stronger coupling than GTPase drivers.

Pre-registration hash: 786c1a8 (D21_KRAS_LUAD_STRATIFIED_v1, github.com/vladi160/preregistrations)