Patient-Specific Platelet Phenotypes in Patients with Immune Thrombocytopenia

Despite recent advances exploring the pathogenic mechanisms underlying immune thrombocytopenia (ITP), a challenging minority of patients remain refractory to multiple lines of treatment and experience repeated bleeding episodes. The unpredictable nature of bleeding in patients with low platelets indicates the presence of platelet functional defects. However, patients often present with severe thrombocytopenia, making platelet functional assessment difficult. Research-based evaluation of platelet parameters may aid in the diagnosis and clinical management of ITP and other conditions associated with low platelets.

We collected blood samples from 104 clinically annotated cases with thrombocytopenia to assess platelet properties and compared them with healthy donors. Most patients were diagnosed with ITP (82/104), while the remaining 22 patients had a platelet count < 100 x 10^9/L due to causes other than ITP and were included as a control thrombocytopenic group. Flow cytometry was employed to quantify a repertoire of platelet surface molecules, including stable surface proteins (αIIb, α2, ADAM10, and CD9), receptors susceptible to metalloproteolytic shedding (glycoprotein (GP) VI and GPIbα), and markers of α-granule release detectable on platelet surfaces or exported via microvesicles (P-selectin and Trem-like transcript-1 (TLT-1)). Levels of soluble receptor fragments, citrullinated histone-DNA (Cit3H-DNA) complexes, and thrombopoietin (TPO) in plasma or serum were quantified by enzyme-linked immunosorbent assay (ELISA), while whole blood clotting was evaluated using rotational thromboelastometry (ROTEM).

Elevated levels of GPVI (p=0.0047), platelet-bound immunoglobulin (p=0.0051), Cit3H-DNA complexes (p=0.0009), and TPO (p<0.0001) were observed in ITP patients experiencing symptoms of bleeding and bruising. Newly diagnosed ITP patients exhibited increased levels of platelet surface αIIb (p=0.0311) and soluble TLT-1 (p=0.0479). Patients undergoing second-line treatment showed raised mean platelet volume (p=0.0012) and soluble GPVI (p=0.0013). Using ROTEM analysis of amplitude at 10 minutes adjusted for platelet count, the contribution of platelets to clot size (Platelet A10) demonstrated a significant correlation (r=0.7000, p<0.0001) with platelet count in symptomatic patients.

We then employed multivariate analyses to assess all platelet and plasma measurements and functional outcomes in unison. Dimension reduction techniques like Pairwise Controlled Manifold Approximation Projection (PaCMAP), showed ITP cluster distinct from healthy donors and the control group. This clustering was independent of platelet count. Next step was to train machine learning (ML) models, so the dataset was randomly divided into a training and a test dataset, and platelet parameters (excluding platelet count) were used to construct six different ML models. After comparing the performance of the algorithm, the random forest model showed high discriminating capacity for ITP against the control group.

Our findings reveal that using research-based tools and approaches, unique platelet functional defects could be quantified in ITP. While no single measure could detect ITP platelet dysfunction in isolation, the evaluation of multiplexed platelet data could distinguish ITP from other types of thrombocytopenia independent of platelet count. By using machine learning training of platelet features, we have developed a novel statistical model for ITP diagnosis.

No relevant conflicts of interest to declare.