Biomarker-driven patient stratification test — HER2, PD-L1 and ctDNA co-diagnostics for targeted therapy
Before any treatment decision, oncologists need to know the molecular identity of a patient's cancer. Two complementary sampling strategies answer different questions: tissue biopsy reveals the spatial architecture and protein expression of the primary tumor; liquid biopsy from blood captures circulating tumor DNA that represents the whole tumor — including metastases that are invisible to biopsy.
Formalin-fixed paraffin-embedded (FFPE) processing is the century-old backbone of clinical pathology. But its chemistry matters for molecular testing:
Formalin fixation chemistry: • Formaldehyde (HCHO) forms methylene bridges between amino groups in proteins • These crosslinks preserve tissue architecture by preventing autolysis • Side effect: DNA crosslinks with proteins, making it difficult to extract high-quality nucleic acids • RNA is particularly vulnerable: mRNA degrades within hours; IHC and ISH work but RNA-seq is challenging from FFPE
FFPE blocks for IHC: • 4 µm sections cut from paraffin block on microtome • Sections mounted on positively charged slides (prevent detachment) • Dewaxing: heat 60°C, then xylene to dissolve paraffin • Rehydration: ethanol gradient → water • Antigen retrieval: heat-induced epitope retrieval (HIER) — boiling in citrate buffer pH 6 breaks formaldehyde crosslinks and restores protein epitope conformation
Contemporary challenges: • Intratumor heterogeneity: a single biopsy core samples < 0.01% of a large tumor volume • Different regions of the same tumor may have different molecular profiles • This is why liquid biopsy is complementary: ctDNA represents average of all tumor clones, including distant metastases • Baseline biopsy at diagnosis, re-biopsy at progression: acquired resistance mechanisms differ from primary tumor
Immunohistochemistry is a molecular microscopy technique that uses antibodies as precision targeting agents to dye specific proteins visible under a light microscope. By conjugating antibodies to enzymes that produce colored precipitates, pathologists can directly see which cells express a biomarker, how intensely, and in what cellular compartment — information invisible to molecular assays that homogenize tissue.
The IHC signal amplification chain transforms a single protein binding event into a visible colored deposit:
Direct IHC (historical, rare): • Primary antibody directly conjugated to enzyme (HRP or AP) • Simple, fast; poor sensitivity — only 1 signal molecule per target
Indirect IHC (standard): 1. Primary antibody (unconjugated, high-affinity anti-HER2) binds HER2 protein in tissue 2. Secondary antibody (anti-IgG, conjugated to HRP polymer) binds primary Polymer-HRP systems (EnVision, OptiView): 40–100 HRP molecules per secondary 3. DAB chromogen added: HRP catalyzes H2O2 oxidation of diaminobenzidine (DAB) into insoluble brown polymer Precipitates exactly where HRP is located = where the target protein is 4. Hematoxylin counterstain: binds nucleic acids (DNA), staining nuclei blue-purple Result: brown HER2 protein on blue/purple nuclear background
Critical controls every IHC run: • Positive control: cell line known to express target at 3+ • Negative control: same antibody concentration, but replaced with non-immune IgG (isotype control) If negative control shows brown staining = non-specific background • External quality assurance: College of American Pathologists (CAP) and NordiQC send proficiency testing samples to >2500 labs globally
HER2 IHC scoring per ASCO/CAP 2018 guidelines: • 0: no staining or < 10% cells with faint/incomplete membrane • 1+: > 10% cells with faint/incomplete basolateral or lateral membrane • 2+: > 10% cells with weak-to-moderate, complete or incomplete membrane (equivocal — requires FISH confirmation) • 3+: > 10% cells with strong, complete membrane staining — diagnostic for HER2 positivity
New 2023 revision — HER2-low category: • IHC 1+ or IHC 2+/FISH-negative now classified as "HER2-low" • Trastuzumab deruxtecan (T-DXd) is effective in HER2-low patients — this reclassification doubled the eligible population from 15% to 50–55% of breast cancer patients
A whole-slide image at 40× magnification contains 4 billion pixels and perhaps 500,000 individual cancer cells. Human pathologists reading 50 slides per day for a decade acquire unparalleled pattern recognition — but they are slow, expensive, and show inter-observer variability on borderline cases. AI algorithms trained on millions of annotated cells can score every cell on a slide in 3 seconds, with better reproducibility than humans on IHC assays.
Digital pathology AI pipelines involve multiple models working in sequence:
1. Tissue detection model: • Identifies tissue regions vs. background (air/glass) • Low-resolution inference (4×) to create tissue mask • Model: typically simple U-Net or even OpenCV thresholding
2. Tumor detection model: • Classifies patches (256×256 px) as tumor, stroma, immune infiltrate, necrosis • Attention-MIL (multiple instance learning): trained on slide-level labels (cancer or not) without pixel-level annotation • Foundation models (PathFoundation, UNI, CONCH): 100M–1B parameter ViT trained on 50–500M image patches
3. Cell segmentation and classification: • Instance segmentation: Mask R-CNN or StarDist identifies individual cell boundaries • Classifies each cell: carcinoma, lymphocyte, fibroblast, endothelial • For HER2: quantifies membrane staining intensity per carcinoma cell using RGB channel analysis
4. HER2 H-Score computation: • For each tumor cell: extract 50 pixel average along detected cell membrane boundary • Map brown-purple ratio in DAB Hematoxylin color space (via Macenko stain deconvolution matrix) • Classify intensity: 0, 1+, 2+, or 3+ per cell • H-Score = Σ(%3+ cells × 3) + Σ(%2+ cells × 2) + Σ(%1+ cells × 1), range 0–300
FDA-cleared AI CDx algorithms: • PathAI/Lilly HER2 companion diagnostic: clearance 2023 • Paige EGFR/ALK for lung: clearance 2022 • Proscia Concentriq: FDA-cleared primary diagnosis viewer
Regulatory pathway: FDA De Novo or 510(k) pathway for software as a medical device (SaMD); requires validation on prospective locked dataset with locked algorithm
Cancer cells shed fragments of their DNA into the bloodstream as they die. This circulating tumor DNA (ctDNA) represents the entire tumor genome — including all subclones, all metastatic lesions, and acquired resistance mutations that emerged after treatment. With digital PCR and next-generation sequencing, we can detect one mutant DNA molecule among 10,000 wild-type molecules, using only 2 milliliters of plasma.
Cell-free DNA (cfDNA) biology: • All nucleated cells shed DNA into plasma 24 hours a day via apoptosis and necrosis • Apoptotic cfDNA: 167 bp fragments corresponding to mono-nucleosome unit (146 bp DNA wrapped around histone octamer + 21 bp linker) • Cancer cells suffer higher turnover rates and shed more DNA: ctDNA may be 0.01% (early stage I) to 50% (late metastatic) of total cfDNA • ctDNA half-life in plasma: ~1–2 hours (cleared by kidneys and liver); real-time pharmacodynamic marker
cfDNA isolation: • Streck BCT (blood collection tube): formalin preserves cell integrity, preventing genomic DNA contamination • Benzonase-free protocol: spin 1600×g 10 min then 3000×g 10 min to remove cells and platelets • Silica-membrane spin column or size-exclusion bead-based extraction • Yield: 2–30 ng cfDNA per 4 mL plasma
Digital droplet PCR (ddPCR) for known mutations: • Sample partitioned into 20,000 nano-droplets; each contains 0 or 1 template molecules • Droplets amplified to endpoint (40 PCR cycles) • FAM-labeled probe: wild-type → high fluorescence; VIC-labeled probe: mutant → intermediate fluorescence • Poisson statistics: ratio of mutant:total droplets = mutant allele fraction • Dynamic range: 0.01%–100% VAF • Applications: KRAS G12C monitoring during osimertinib, ESR1 D538G in aromatase inhibitor-resistant ER+ breast cancer
Next-generation sequencing panel (ctDNA-NGS): • Hybrid capture (IDT xGen) enriches target regions • 500–1000× sequencing depth — deep enough to detect 0.1% VAF variants • Unique Molecular Identifiers (UMI): barcodes ligated to each cfDNA fragment before PCR PCR duplicates sharing identical UMI removed; sequencing errors (1 in 1000 bases) distinguished from true mutations (appear in all UMI families) • Foundation Medicine F1 CDx: FDA-approved pan-cancer companion diagnostic covering 324 genes • FoundationOne Liquid CDx: plasma-based, 324-gene, FDA-approved for multiple tumor types and companion drugs
Modern precision oncology does not pick a treatment first and then find patients who respond. It measures first — comprehensively, at the DNA, RNA, and protein level — and then assigns each patient to the therapy that targets their tumor's specific vulnerability. The result: dramatically higher response rates, less toxicity, and a transformation of cancer from a tissue-of-origin diagnosis to a molecular one.
Multi-biomarker integration algorithm:
Step 1: Tumor mutational burden (TMB) • Calculate mutations per megabase from NGS panel • TMB-High (≥10 mut/Mb): FDA-approved pembrolizumab regardless of tumor type — first pan-tumor biomarker approval (2020) • TMB-Low: proceed to other biomarkers
Step 2: Microsatellite instability (MSI) / mismatch repair (MMR) • MSI-High or dMMR: FDA-approved pembrolizumab (any solid tumor, 2017 — first tissue-agnostic approval in oncology history) • Detected by fragment length analysis or IHC for MLH1, MSH2, MSH6, PMS2
Step 3: HER2 status • IHC 3+ or FISH amplification ratio ≥2.0: trastuzumab + pertuzumab + chemotherapy • IHC 1+ or IHC 2+/FISH-negative (HER2-low): trastuzumab deruxtecan (T-DXd) — 2022 approval
Step 4: PD-L1 (programmed death-ligand 1) expression • 22C3 pharmDx (Dako): approved CDx for pembrolizumab in lung,gastric,cervical,esophageal cancers • TPS (tumor proportion score): percentage of tumor cells with membrane staining • CPS (combined positive score): includes immune cells; used for gastric/cervical
Step 5: Actionable mutations • EGFR exon 19 del, exon 21 L858R: osimertinib (FLAURA), erlotinib • ALK/ROS1 fusion: alectinib (ALEX), lorlatinib • KRAS G12C: sotorasib (CodeBreaK100), adagrasib (KRYSTAL-1) • BRCA1/2 germline/somatic: olaparib, niraparib, rucaparib (PARP inhibitors) • NTRK1/2/3 fusion: larotrectinib, entrectinib (pan-tumor) • RET fusion: selpercatinib, pralsetinib • MET exon 14 skip: capmatinib, tepotinib
Failed precision oncology (lessons): • EGFR-mutant NSCLC responds to gefitinib → initially tried in all NSCLC without biomarker selection → failed in unselected trials → biomarker-positive subgroup analysis rescued it → PDX-based co-clinical trial design now standard
The 2023 CROWN trial showed lorlatinib (ALK inhibitor) produced 60% 5-year progression-free survival in ALK-positive NSCLC patients — better than any chemotherapy ever achieved in lung cancer. The companion diagnostic ALK IHC D5F3 CDx test was developed concurrently with the drug, enabling a 1:1 patient-drug match that makes the 60% survival rate possible. This is precision oncology working as intended.
FDA requires companion diagnostic approval in parallel with the drug when the drug's labeling claims efficacy only in biomarker-defined population:
Co-development timeline: • Phase 1–2: exploratory biomarker analysis identifies potential CDx candidate • Phase 2b: CDx assay locked and analytically validated (ISO 13485 quality system) • Phase 3: prospective collection of paired tissue/blood specimens; CDx used to allocate treatment • Submission: drug NDA/BLA + CDx PMA submitted simultaneously • FDA review: drug and CDx reviewed by different CDER/CDRH divisions but coordinated
CDx analytical validation requirements: • Precision: repeatability (same lab same day) + reproducibility (different labs) • Accuracy: concordance with reference method (clinical truth derived from clinical outcomes) • Sensitivity/specificity vs. reference standard (RNA-ISH, NGS, or adjudicated clinical dataset) • Lot-to-lot variability: 3 independent lots tested • Cut-point validation: pre-specified in Statistical Analysis Plan before any efficacy analysis
Global regulatory differences: • FDA: CDx required for on-label use of companion drug; off-label use possible with LDT • EMA: IVD Regulation (IVDR) 2022 — stricter classification, requires Notified Body certification • Japan PMDA: parallel CDx review program similar to FDA • China NMPA: domestic CDx manufacturing preference through preferential review
Laboratory developed tests (LDT): • Developed and validated internally by a single CLIA-certified lab • No FDA pre-market review historically — "enforcement discretion" • FDA LDT Final Rule (2024): phased oversight, high-risk LDTs require 510(k)/PMA review