Checkpoint Inhibitors: Drug Mechanism Design Pattern

In progress

Checkpoint Inhibitors: Drug Mechanism Design Pattern

Status: Phase 1 Complete -- Module + 5 Pilot Entries

1. Landscape: Which Diseases Are Treated by Checkpoint Inhibitors?

Diseases with checkpoint inhibitor treatments currently in dismech

Disease Drugs Referenced Has target_mechanisms? Has Immune Evasion Pathophys Node?
MSI-High Colorectal Cancer pembrolizumab, nivolumab, ipilimumab, dostarlimab Yes Yes ("PD-L1 Upregulation and Immune Evasion")
Clear Cell Renal Cell Carcinoma nivolumab, pembrolizumab, ipilimumab Yes Yes ("Immune Evasion via PD-L1 and Immunosuppressive Microenvironment")
Hepatocellular Carcinoma atezolizumab, durvalumab, tremelimumab Yes Yes ("Immune Evasion via PD-L1 and Immunosuppressive Microenvironment")
BRAF V600 Mutant Melanoma pembrolizumab, nivolumab, ipilimumab Yes Yes ("Immune Evasion via PD-L1 Upregulation")
NRAS Mutant Melanoma checkpoint inhibitors (narrative) No Partial
KIT Mutant Melanoma checkpoint inhibitors (narrative) No Partial
Cutaneous Squamous Cell Carcinoma cemiplimab No Partial
Non-Small Cell Lung Cancer checkpoint inhibitors (narrative) No Partial
Small Cell Lung Cancer checkpoint inhibitors (narrative) No Partial
KRAS G12C Mutant NSCLC checkpoint inhibitors (narrative) No Partial
Cervical Cancer pembrolizumab No Partial
Triple-Negative Breast Cancer atezolizumab/pembrolizumab (narrative) No Yes (immunomodulatory subtype)
Nasopharyngeal Carcinoma PD-1 inhibitors Yes Yes ("Immune Evasion")
HPV-Positive Head and Neck Cancer checkpoint inhibitors (narrative) No Yes ("Immune Evasion")
Merkel Cell Carcinoma avelumab (narrative) No Partial
MSI-High Endometrial Cancer checkpoint inhibitors (narrative) No Partial
FGFR-Altered Cholangiocarcinoma durvalumab No No
IDH-Mutant Cholangiocarcinoma durvalumab No No
Malignant Mesothelioma checkpoint inhibitors (narrative) No Partial
Basal Cell Carcinoma cemiplimab (narrative) No Partial
EBV-Associated Gastric Cancer checkpoint inhibitors (narrative) No Partial
Osteosarcoma checkpoint inhibitors (narrative) No Yes ("Tumor Immune Microenvironment Remodeling")
Uveal Melanoma checkpoint inhibitors (noted as poor response) No Noted as "immunologically cold"

Diseases where checkpoint molecules appear in pathophysiology (not treatment)

Disease Context
Hepatitis B T cell exhaustion with PD-1, CTLA-4, TIM-3 upregulation
Addison's Disease CTLA4 as autoimmune susceptibility gene
Diabetes Mellitus Checkpoint inhibitor exposure as disease trigger
Type 1 Diabetes Immune checkpoint pathway involvement
Autoimmune diseases (multiple) CTLA-4/PD-1 polymorphisms in autoimmune susceptibility

2. Current Representation Gaps

Treatment side

Pathophysiology side

The disconnect

The treatment and pathophysiology sections describe the same biological process from opposite directions but are not structurally linked: - Pathophysiology says: "Tumor upregulates PD-L1 → suppresses T cells → immune evasion" - Treatment says: "Anti-PD-1 blocks this interaction → restores T cell function"

But there's no target_mechanisms edge connecting the treatment to the pathophysiology node.

3. Proposal: Drug Mechanism Classes as Design Patterns

Concept

Just as kb/modules/fibrotic_response.yaml defines a conserved pathological process that multiple diseases conforms_to, we should create mechanism-of-action modules that capture conserved drug response patterns. These would serve as design patterns for how treatments connect to pathophysiology.

Proposed Module: immune_checkpoint_blockade

kb/modules/immune_checkpoint_blockade.yaml

This module would define the conserved pattern:

Pathophysiology side (the disease process being targeted):
  1. Neoantigen Generation → high TMB produces immunogenic peptides
  2. Anti-Tumor Immune Response → CD8+ T cells recognize and infiltrate tumor
  3. Adaptive Immune Resistance → tumor upregulates PD-L1/PD-L2 in response to IFN-gamma
  4. T Cell Exhaustion and Immune Escape → chronic checkpoint engagement → dysfunctional T cells → tumor evades destruction

Treatment side (the therapeutic intervention):
  Anti-PD-1/PD-L1 therapy:
    target_mechanisms:
      - target: "Adaptive Immune Resistance"
        treatment_effect: INHIBITS
      - target: "T Cell Exhaustion"
        treatment_effect: INHIBITS
    → downstream effect: restores "Anti-Tumor Immune Response"

  Anti-CTLA-4 therapy:
    target_mechanisms:
      - target: "T Cell Exhaustion" (priming phase)
        treatment_effect: INHIBITS
    → downstream effect: expands T cell repertoire

How diseases would conform

# In kb/disorders/MSI_High_Colorectal_Cancer.yaml
pathophysiology:
- name: Neoantigen-Driven Immune Response
  conforms_to: "immune_checkpoint_blockade#Anti-Tumor Immune Response"
  # organ-specific: high TMB from dMMR
  ...
- name: PD-L1 Upregulation and Immune Evasion
  conforms_to: "immune_checkpoint_blockade#Adaptive Immune Resistance"
  ...

treatments:
- name: Pembrolizumab
  ...
  target_mechanisms:
  - target: PD-L1 Upregulation and Immune Evasion
    treatment_effect: INHIBITS
    description: Anti-PD-1 blocks PD-1/PD-L1 interaction, restoring T cell cytotoxicity

Why this matters

  1. Consistency: Every cancer with checkpoint inhibitor treatment would model immune evasion the same way, with organ-specific substitutions (just like fibrotic response uses organ-specific fibroblasts).

  2. Predictive power: If a cancer has a pathophysiology node conforming to the immune evasion pattern, it suggests checkpoint inhibitor sensitivity. Conversely, cancers like Uveal Melanoma that lack the pattern ("immunologically cold") explain treatment resistance.

  3. Completeness checking: If a disease has a checkpoint inhibitor treatment but no immune evasion pathophysiology node (e.g., FGFR-Altered Cholangiocarcinoma), that's a curation gap.

  4. Bidirectional linking: target_mechanisms formally connects treatment → pathophysiology, closing the current structural gap.

Additional mechanism-of-action modules to consider

Module Diseases Pattern
immune_checkpoint_blockade ~23 cancers Neoantigen → immune infiltration → adaptive resistance → checkpoint blockade restores immunity
kinase_inhibition ALK NSCLC, BRAF Melanoma, FGFR cancers, CML Constitutive kinase activation → oncogenic signaling → TKI blocks kinase → pathway shutdown
antibody_dependent_cellular_cytotoxicity HER2+ cancers, lymphomas Surface antigen overexpression → antibody binding → Fc-mediated immune recruitment
angiogenesis_inhibition RCC, HCC, CRC VEGF overproduction → neovascularization → anti-VEGF starves tumor
differentiation_therapy APL (ATRA/ATO) Differentiation block → pharmacologic override → terminal differentiation
synthetic_lethality BRCA-mutant cancers (PARP inhibitors) DNA repair deficiency → PARP inhibition → unrepaired damage → cell death
cdk_inhibition HR+ breast cancer, liposarcoma Cyclin D-CDK4/6 overactivation → Rb phosphorylation → CDK4/6i arrests G1/S → senescence

4. Implementation Plan

Phase 1 (DONE)

Phase 2 (TODO)

5. Key Questions

  1. Schema evolution: Should conforms_to on Treatment (not just pathophysiology) point to a mechanism module? Currently only pathophysiology nodes conform to modules.
  2. Granularity: Should anti-PD-1 and anti-CTLA-4 be separate sub-patterns within the module, or one unified checkpoint blockade pattern?
  3. Resistance modeling: How to represent acquired resistance (e.g., loss of B2M, JAK1/2 mutations) that breaks the checkpoint blockade pattern?
  4. Adverse events as pattern: Checkpoint inhibitor-induced autoimmunity (irAEs) is itself a conserved pattern -- model as a separate module or within the same one?
  5. Biomarker integration: PD-L1 expression, TMB, and MSI status predict checkpoint response. Should the module include biomarker nodes?

6. Observations on Autoimmune Mirror

Checkpoint inhibitors work by removing immune brakes. This is the mirror image of autoimmune diseases where checkpoint pathways are insufficient. The same module could potentially be used bidirectionally: - Cancer: Checkpoint engagement = disease mechanism → blockade = treatment - Autoimmune: Checkpoint insufficiency = disease mechanism → checkpoint agonism = (theoretical) treatment - irAEs: Checkpoint blockade (cancer treatment) → autoimmune pathology (adverse event)

This triangulation supports the "design pattern" framing -- the same biological pattern manifests differently depending on which direction it's perturbed.