checkpoint-biomarkers

Checkpoint Biomarkers

Predictive biomarkers for immune checkpoint blockade: what PD-L1 IHC scores are and why expression data cannot produce them, tumour mutational burden, microsatellite instability, and the expression signatures (IFN-gamma, TIS, TIDE) that are computable from RNA. Written to keep the assay-derived biomarkers and the expression-derived ones apart, because conflating them is the common failure.

When to Use This Skill

Activate when the user requests:

• PD-L1 status, CPS, TPS, or IC/TC scoring
• TMB-high classification for checkpoint inhibitor eligibility
• MSI-high or dMMR determination
• IFN-gamma, Tumour Inflammation Signature, or TIDE scoring
• Predicting or stratifying response to anti-PD-1, anti-PD-L1, or anti-CTLA-4
• Combining several checkpoint biomarkers into one call

Inputs

Data Type	Form	Produces
Stained slide	PD-L1 IHC, a named antibody clone	CPS, TPS, IC/TC — only from here
Somatic variants	VCF or MAF plus panel definition	TMB
Reads or variants	BAM, or MSI marker loci	MSI status
Bulk expression	genes x samples, TPM or normalized counts	IFN-gamma, TIS, TIDE
Clinical	response, PFS, OS	evaluation, never input to the score

---

The Division That Matters

Two families of checkpoint biomarker, and they are not interchangeable.

ASSAY-DERIVED, from a slide or a sequencer
  PD-L1 CPS / TPS      IHC. Requires counting stained cells.
  TMB                  mutations per megabase. Panel-dependent.
  MSI / dMMR           marker instability or MMR protein loss.

EXPRESSION-DERIVED, computable from RNA
  IFN-gamma signature  mean of a small gene set
  TIS / GEP            18-gene inflammation signature
  TIDE                 dysfunction and exclusion modelling

Most confusion in this area comes from treating a member of the first group
as though it belonged to the second.

CPS and TPS Cannot Be Computed From Expression

This is the correction worth stating first, because the request arrives
often and sounds reasonable.

  TPS = (PD-L1-stained viable TUMOUR cells / total viable tumour cells) x 100
  CPS = (PD-L1-stained tumour cells + lymphocytes + macrophages
         / total viable tumour cells) x 100

Both are counts of individual stained cells on an IHC slide, and CPS
requires telling a stained tumour cell apart from a stained lymphocyte or
macrophage. That is a morphological judgement at cellular resolution.

Bulk RNA gives you one CD274 value per sample. It has no cells in it, so it
cannot enumerate them and cannot assign them to a compartment. CD274 mRNA
does correlate with PD-L1 IHC, and reasonably well, but a correlation is not
a score: you cannot recover CPS from it, and a "CPS estimated from
expression" is a number with no regulatory or clinical meaning.

What you CAN honestly do with expression:
  - report CD274 (PD-L1) expression as its own continuous variable
  - test its association with outcome
  - state that it is a correlate of, not a substitute for, IHC

If the analysis requires CPS or TPS, it requires a stained slide scored by a
pathologist. Say so rather than approximating.