The promise of immunotherapy PD-1/PD-L1 checkpoint inhibitors have transformed the treatment of non-small cell lung cancer (NSCLC), with some patients achieving durable long-term remissions. Testing tumor tissue for PD-L1 expression became the standard approach to identify who might benefit, yet response rates remain unpredictable and many high-expressors still fail therapy.
The core problem PD-L1 is not a static, reliable protein target. Its expression is regulated by post-translational modifications, shaped by glycosylation, altered by prior therapy, and scored inconsistently across different immunohistochemistry (IHC) platforms. This review systematically explores why current PD-L1 testing misses important biology and what alternative strategies are being developed.
Scope of this review Ferrari and colleagues examine molecular mechanisms governing PD-L1 stability and detection, discuss how treatment-induced changes confuse biomarker interpretation, and survey multi-omics approaches that may replace or supplement single-protein IHC in clinical decision-making.
Transcriptional drivers PD-L1 expression in tumor cells is activated by interferon-gamma (IFN-gamma) through the JAK/STAT1 pathway, which upregulates IRF1 binding to the CD274 promoter. Oncogenic pathways including EGFR/RAS/MAPK, PI3K/AKT, and MYC amplification can also drive constitutive PD-L1 expression independent of immune signaling.
Post-translational modifications Once synthesized, PD-L1 protein undergoes extensive glycosylation at four N-linked sites (N35, N192, N200, N219). This glycosylation is not decorative - it physically shields PD-L1 from ubiquitin-mediated degradation and dramatically increases its half-life on the cell surface, making glycosylated PD-L1 the dominant functional form.
Protein stability and turnover Multiple E3 ubiquitin ligases including STUB1, SPOP, and beta-TrCP target non-glycosylated PD-L1 for proteasomal degradation. Conversely, deubiquitinases such as USP7 and OTUB1 can stabilize PD-L1. This dynamic turnover means that measured protein levels reflect a momentary equilibrium rather than a fixed tumor property.
Antibody epitope interference The four clinically approved PD-L1 IHC assays (22C3, 28-8, SP142, SP263) use antibodies that bind different epitopes on the PD-L1 protein. When glycan chains are attached at N192 and N200 - the most heavily modified sites - they physically block antibody binding, causing assays to underestimate true protein levels.
Platform discordance in practice Head-to-head comparisons of the four approved assays on matched tumor sections routinely show 20-30% discordance in TPS (tumor proportion score) classification, particularly around the clinically critical 1% and 50% cutoffs. This means the same tumor can appear PD-L1 positive or negative depending on which reagent kit was used.
Deglycosylation as a fix Treating tissue sections with PNGase F to remove N-glycans before staining significantly improves antibody binding uniformity across platforms and may reduce inter-assay discordance. While not yet standard practice, deglycosylation pretreatment is under active investigation as a way to normalize IHC results.
Treatment-induced upregulation Platinum-based chemotherapy, taxanes, and ionizing radiation all increase PD-L1 expression in surviving tumor cells through stress-response pathways and DNA damage signaling. This means a biopsy taken after neoadjuvant treatment may score substantially higher than the pre-treatment specimen that actually guided the original therapy decision.
Heterogeneity across biopsy sites PD-L1 expression varies significantly between the primary tumor and metastatic sites, and even between different cores from the same tumor. Studies show that spatial heterogeneity alone can cause discordant PD-L1 classification in 15-25% of cases where multiple sites are tested.
Dynamic instability Longitudinal studies following patients through treatment show PD-L1 scores changing by one or more clinical cutoff categories in a substantial fraction of patients. This temporal instability means that a PD-L1 score from a prior biopsy may poorly predict the tumor's current immune landscape, raising questions about the optimal timing of biomarker testing.
Tumor mutational burden (TMB) TMB measures the total number of somatic mutations per megabase of tumor DNA, reflecting the potential neoantigen load that can activate T cells. Unlike PD-L1, TMB is not affected by IHC artifacts. FDA has approved pembrolizumab for TMB-high tumors across cancer types, but the optimal cutoff and the interaction between TMB and PD-L1 scoring remain under investigation.
Transcriptomic immune signatures RNA-based approaches such as the 18-gene T-cell inflamed gene expression profile (GEP) and tumor inflammation signature (TIS) quantify multiple immune pathway components simultaneously, potentially capturing immune exclusion, dysfunction, and suppression that protein-level PD-L1 misses. These transcriptome-based tools show better correlation with actual immunotherapy response in some cohorts.
Combining biomarkers Multi-analyte algorithms that jointly consider PD-L1 IHC, TMB, gene expression signatures, and potentially liquid biopsy markers may substantially outperform any single test. The authors argue that the field needs to move beyond PD-L1 alone toward integrated biomarker panels, analogous to how HER2 testing in breast cancer now incorporates both protein expression and gene amplification.
Circulating PD-L1 Soluble PD-L1 shed from tumor cells and immune cells can be detected in plasma and may reflect tumor burden and immune activation. Elevated soluble PD-L1 levels have been associated with worse outcomes in some studies, though the biology of this circulating form is not fully understood and its clinical utility remains investigational.
Circulating tumor DNA (ctDNA) ctDNA-based TMB assessment from liquid biopsies can be performed without invasive tissue sampling and may be repeated longitudinally to track clonal evolution. While ctDNA TMB correlates imperfectly with tissue TMB due to shed variability and clonal heterogeneity, it offers a non-invasive window into tumor genomics that can capture spatial and temporal heterogeneity missed by single biopsies.
Circulating immune cells Flow cytometry of peripheral blood can measure T-cell exhaustion markers, regulatory T-cell abundance, and NK cell activity. These systemic immune metrics complement tumor-based assays and may help distinguish patients who lack immune response capacity from those whose tumors are simply PD-L1 low due to assay artifacts.
Standardizing glycoprotein detection The most direct technical fix is developing glyco-specific antibodies that intentionally target glycosylated PD-L1 or developing standardized deglycosylation pretreatment protocols for routine IHC. Either approach would reduce the platform-dependent variability that currently undermines clinical decision-making.
Beyond PD-L1/PD-1 The broader B7-H family of immune checkpoint ligands - including B7-H3, B7-H4, and VISTA - are expressed in NSCLC and largely orthogonal to PD-L1. Simultaneous multiplex profiling of the full immune checkpoint landscape may better define which patients need PD-1 blockade, combination checkpoint blockade, or entirely different strategies.
Artificial intelligence integration Deep learning analysis of histopathology slides can extract spatial immune features such as T-cell proximity to tumor cells, immune exclusion patterns, and tertiary lymphoid structure presence that correlate with immunotherapy response beyond PD-L1 score alone. The authors envision AI-augmented pathology as an integral part of next-generation biomarker panels.