Abstract
Immune checkpoint blockade, a form of immunotherapy, mobilizes a patient’s own immune system against cancer cells by releasing some of the natural brakes on T cells. Although our understanding of this process is evolving, it is thought that a patient response to immunotherapy requires tumor presentation of neoantigens to T cells, and patients whose tumors present a wider array of neoantigens are more likely to derive benefit from immune checkpoint blockade. Manczinger et al. recently reported findings that would appear contrarian to this notion, suggesting that patients with HLA alleles which bind more diverse peptides (higher promiscuity) are less likely to respond to immunotherapy. To estimate HLA promiscuity, they looked at the HLA-peptide binding repertoires for class I alleles contained in the IEDB and obtained consistent results when performing robustness checks. Here we show that the proposed HLA promiscuity values can vary significantly across source data types and individual experiments.
Introduction
In cancer immunotherapy, Human Leukocyte Antigen (HLA) molecules play a critical role by presenting peptide fragments (neoantigens) on the cell surface for T cell recognition. Traditionally, a broader repertoire of presented neoantigens was considered beneficial for patient response. However, a 2021 study by Manczinger et al. proposed a counter-intuitive finding: "promiscuous" HLA alleles—those capable of binding a highly diverse array of peptides—are associated with poorer responses to immune checkpoint blockade. This study re-evaluates the stability of the "promiscuity" metric ($Pr$), investigating whether these findings reflect biological reality or are artifacts of inconsistent experimental data sources.
Methods and Data
The authors analyzed the Promiscuity ($Pr$) metric, defined as the inverse of the average Kullback–Leibler (KL) divergence across a peptide's positions ($Pr = 1/KL$). A low KL divergence indicates that the amino acids at specific positions in HLA-bound peptides closely resemble the background distribution of the human proteome, signifying high promiscuity.
To test the robustness of this metric, $Pr$ values were recalculated using various subsets of data from the Immune Epitope Database (IEDB). The analysis specifically compared:
- HLA Binding Assays: Laboratory tests measuring the potential of a peptide to bind to an HLA molecule.
- Mass Spectrometry (MS) Data: Identification of peptides naturally presented on the surfaces of real cell lines (immunopeptidomes).
The study incorporated large-scale MS datasets from multiple sources (Abelin et al., Sarkizova et al., Faridi et al., etc.) and utilized the human proteome as the background distribution. Original R code was ported to Python to ensure exact reproduction of the initial study's results before performing comparative analyses.
Results
Lack of Correlation Between Data Types
A primary finding is the near-total lack of correlation between promiscuity values derived from binding assays and those derived from mass spectrometry. As shown in Figure 1A, the correlation between these two data types was non-existent ($\rho = 0.02, p = 0.92, N = 34$). This suggests that the original study's conclusions may be compromised by mixing fundamentally different types of experimental evidence.
Experimental and Study-Specific Variability
Even within a single data type (MS), $Pr$ values for identical HLA alleles varied significantly across different laboratories and cell lines. For instance, the allele A01:01* exhibited vastly different promiscuity levels depending on whether the source data originated from B cells or K562 cells (Figure 1H). This variability makes it difficult to assign a stable "raw" promiscuity score to any given allele.
Data Composition and Database Drift
The analysis revealed that alleles identified as "highly promiscuous" in the original study were often those with the highest proportion of binding assay data relative to MS data (Figure 3A). Furthermore, as the IEDB grows, the calculated promiscuity of alleles shifts over time. Comparing 2021 data to the 2019 data used by Manczinger et al. showed significant changes in $Pr$ values, indicating that the metric is highly sensitive to the specific snapshot of the database used.
Discussion
The findings demonstrate that the proposed HLA promiscuity metric is highly unstable and sensitive to experimental noise and data source composition. The "anchor position" problem—where specific positions in a peptide drive the KL divergence—is further complicated by the fact that cellular processing machinery is not random, making the whole-proteome background potentially inappropriate for MS data.
While the concept of HLA promiscuity remains potentially interesting for clinical research, current metrics are too inconsistent for use in prognostic models. Future work should focus on standardizing data integration across experimental sources and developing context-specific background distributions (e.g., using RNA-seq data) to better account for cellular processing biases. Rigorous validation is essential before HLA promiscuity can be reliably used to predict immunotherapy outcomes.
Summary updated