Page 181 - 2019_06-Haematologica-web

High-throughput elucidation of thrombus formation
for microspot M3, did the subjects’ age, but not sex con- tribute to component 2. Together, these results underscore the earlier conclusion that parameters of platelet aggrega- tion across microspots covariate.
Combined PCA of microspot parameters and hemato- logic parameters (Figure 4B) resulted in a similar configu- ration of component 1 (97.5%) as above, with in particular high contributions of microspots M1-6 for P2-5. Platelet count and platelet crit appeared in component 1 for microspots M1-2, while these variables contributed to component 2 (1.1%) for M3-6.
Combined PCA of microspot parameters and glycopro- tein expression levels (Figure 4C) revealed a marked con- tribution of most CD variables (although not of CD36) to the first component (95.5%) regarding M1, M2 and M6, suggesting that these expression levels were relevant to thrombus formation on these microspots. Component 2 (0.8%) contained most of the glycoprotein expression lev- els, particularly on M3 and M5 spots.
The combined PCA of microspot parameters with ago- nist-induced platelet secretion and integrin activation (Figure 4D) revealed a component 1 (89.0%), again show- ing large contributions for M1-5 on parameters P1-5 (platelet adhesion, aggregation and thrombus morpholo- gy). For microspot M6, component 1 also included the platelet activation markers (A2-4, i.e., ADP, CRP-XL and TRAP). Interestingly, component 2 (1.9%) revealed large contributions of the platelet activation parameters on all other microspots M1-5. These results suggested that microspot M6 (VWF-BP + fibrinogen) is the most sensitive to variation in common agonist-induced platelet activa- tion traits, regardless of the agonist. A final PCA, combin- ing glycoprotein expression markers with platelet activa- tion markers, revealed strong correlations for the variables linked to integrin αIIbb3 (Figure 4E).
Together, these results suggested that a considerable part of the variability in thrombus outcomes between the 94 subjects was linked, in a surface-dependent way, to hematologic traits (platelet count and crit), expression pat- terns of glycoproteins and to agonist-induced secretion and integrin αIIbb3 activation. Another part seemed to be linked, across surfaces, to platelet aggregation and throm- bus morphology.
Prediction models of platelet traits contributing to variation in thrombus formation
Based on the PCA, predictive models were built to iden- tify the quantitative platelet traits that correlate with thrombus formation parameters. In a first set of partial least squares (PLS) models, we aimed to find the covari- ance between the individual thrombus parameters and the other platelet parameters. By predicting one variable of the thrombus microspot - parameters matrix at a time, 14 (of 48) parameters had a reduction in the mean square error of prediction, meaning that they could be predicted more accurately than just using the mean (in cross valida- tion). Eleven of the 14 models appeared to be robust (sin- gle component, or improvement with an additional com- ponent). Most of these models captured a limited amount of the variation, although, for parameter P1, three models (M4P1, M5P1, M2P7) predicted >5% (6-11%) of the total variation (Figure 5A). This analysis explained a small part of the variance, although with a focus on platelet adhesion (P1) across the surfaces. Given the limited biological insight of this effort, additional modeling was performed.
Since thrombus formation on three out of the six microspots (M1, M2, M4) was GPVI-dependent, addition- al prediction models were made to relate the values of GPVI-induced integrin αIIbb3 activation (A3-Int) and secre- tion (A3-Sec) to the thrombus parameters. All models were checked by leaving-one-out cross-validations. To predict GPVI-induced integrin activation, an orthogonal PLS model was built with two components. This resulted in a b matrix for each of the microspots and parameters (Figure 5Bi). For the collagen microspots (M1-2), parame- ters of platelet adhesion and activation (P1, P6-7) gave a positive weight to the prediction, as well as the majority of parameters of M6 (VWF-BP + fibrinogen). Interestingly, for the strongest GPVI-dependent surfaces (M1, M4), parameters determining platelet aggregation under flow (P2-5, P8) weighed negatively on the prediction. Concerning GPVI-induced secretion, the most suitable PLS model had a single principal component. The resulting b matrix was similar to that predicting GPVI-induced inte- grin activation, albeit with more negative predictive weights for some M5-6 parameters (Figure 5Bii). Taken together, this analysis of the collagen surfaces indicates a positive relationship between GPVI-induced platelet inte- grin activation/secretion (assessed by flow cytometry) and platelet adhesion and activation in a thrombus.
Prediction of genetic variants of platelet proteins associated with variation in thrombus formation
Genome-wide association studies have identified sever- al hundreds of genetic variants associated with quantita- tive platelets traits.16 Epigenetic mapping has revealed that many of these variants are regulatory and lie within super- enhancer regions that are implicated in megakaryocyte differentiation and platelet production.20 For three variants associated with platelet traits, we analyzed whether these were also associated with thrombus formation parame- ters. Univariate linear regression analysis for the whole M- P matrix was used to identify associations with the fol- lowing single nucleotide variants: rs1613662 (GP6), rs3557 (FCER1G) and rs2363877 (VWF-CD9) (Figure 5C).
The single nucleotide variant rs1613662 is a non-syn- onymous variant in the GP6 gene, while rs3557 is located in the 3’ untranslated region of FCER1G (Fcγ receptor chain, a co-receptor of GPVI) in a megakaryocytic super- enhancer region. Subjects carrying the major allele of either variant (rs1613662, AA; and rs3557, TT) have high- er levels of platelet GPVI, and higher CRP-XL-induced platelet activation.20 For the present data set, regression analysis indicated inter-allelic differences, in the same direction, in thrombus formation parameters at the colla- gen surfaces M1-2 for P1 (platelet adhesion) with rs1613662 (Figure 5Ci, ii). In addition, for rs3557 we iden- tified associations at M1-2 and M4, i.e., the other GPVI- dependent surface, regarding P8 (integrin activation). Unexpectedly, alleles of both variants were also associated with parameters at the CLEC-2-dependent microspot M5 (VWF-BP + rhodocytin).
The variant rs2363877 is located in a megakaryocyte- specific super-enhancer site that interacts with the gene promoters of VWF and CD9. This single nucleotide vari- ant is linked to opposite changes in expression levels of platelet-stored VWF and the surface levels of the tetraspanin CD9.20 Here an allelic association was identi- fied at microspot M5 for secretion (P7) (Figure 5Ciii). Trends were also seen for platelet adhesion (P1) at the
haematologica | 2019; 104(6)
1263