Novel gene variants in patients with platelet-based bleeding using combined exome sequencing and RNAseq murine expression data

Abstract

Background

The UK Genotyping and Phenotyping of Platelets (GAPP) study has recruited and analysed 129 patients with suspected heritable bleeding. Previously 55 individuals had a definitive genetic diagnosis based on whole exome sequencing (WES) and platelet morphological and functional testing. A significant challenge in this field is defining filtering criteria to identify the most likely candidate mutations for diagnosis and further study.

Objective

Identify candidate gene mutations for the remaining 74 patients with platelet-based bleeding with unknown genetic cause, forming the basis of future re-recruitment and further functional testing and assessment.

Methods

Using python-based dataframe indexing, we first identify and filter all novel and rare variants using a panel of 116 genes known to cause bleeding across the full cohort of WES data. This identified new variants not previously reported in this cohort. We then index the remaining patients, with rare or novel variants in known bleeding genes against a murine RNAseq dataset which models proplatelet forming megakaryocytes.

Results

Filtering against known genes identified candidate variants in 59 individuals, including novel variants in several known genes. In the remaining cohort of ‘unknown’ patients, indexing against differentially expressed genes revealed candidate gene variants in several novel unreported genes, focussing on 14 patients with a severe clinical presentation.

Conclusions

We identified candidate mutations in a cohort of patients with no previous genetic diagnosis. This work involves innovative coupling of RNA sequencing and WES to identify candidate variants forming the basis of future study in a significant number of undiagnosed patients.

Introduction

The UK Genotyping and Phenotyping of Platelets (GAPP) ¹ study previously reported a cohort of 55 patients with Whole Exome Sequencing (WES) recruited from haemophilia centres nationwide with a significant clinical history of bleeding.² In this group of patients, candidate variants and a genetic diagnosis of platelet based bleeding was provided in 40 patients, yielding a detection rate of 72.72% candidate variants in this initial cohort using the ‘manual’ filtering method of Johnson et al.³ (Figure 1A).

Figure 1. Distribution of known and unknown GAPP patients and the identification of new variants in bleeding genes.

(A) In our previous report of the GAPP cohort in Johnson et al., a total of 55 exomes were reported with a total of 40 patients identified as having a clear genetic candidate for their inherited platelet based bleeding. (B) As the cohort expanded to include 129 exomes, only 55 patients have an established genetic cause of bleeding. To date, the cohort with unresolved causes remains at 74 patients. (C) By cross indexing rare and novel variants in GAPP exomes with known bleeding related genes we identified new candidates genes in a further 59 patients, reducing the size of the ‘unknown’ cohort to 14.

More recently, the GAPP study has expanded patients with available WES data to include 129 individuals with evidence of clinical bleeding. Of these, 55 individuals have had a definitive genetic diagnosis to identify novel disease causing genes (e.g. SLFN14)⁴ or new pathogenic variants in known genes (e.g. GP1BA/B, WAS) yielding an overall detection rate of 42.63% (Figure 1B).^{5, 6} Through an extensive panel of platelet function testing⁷ and morphological assessment,⁸ these genetic findings have been correlated to the physiological and clinical effects of these rare genetic variants. However, the genetic causes of bleeding in a significant number of GAPP-recruited patients remain unknown, despite an established clinical and family history. Here, we assess these individuals through a series of iterative bioinformatic filtering approaches to identify novel and ‘known’ candidate disease causing variants.

Methods

Patient recruitment and testing

Patients were consented and recruited to the GAPP study from multiple collaborating Haemophilia Centres across the UK and Ireland as previously described¹ and approved by the UK National Research Ethics Service by the Research Ethics Committee of West Midlands (06/MRE07/36). Peripheral blood was collected from patients and Platelet phenotyping using lumi-aggregometry or flow cytometry was performed on Platelet rich plasma as previously described.^7,8

Our study cohort consisted of 129 patients with a strong history of bleeding and suspected of having a PFD of unknown cause as previously described.^1,2 The ISTH Bleeding Assessment Tool showed a mean BAT score of 9.825 overall (range 2 - 23). Platelet counts for the patient group ranged from 43 - 428×10⁹/L and the average platelet count was 232 x10/L. The Mean Platelet Volume (MPV) in patients tested ranged between 8.3fL and 15.1fL.

RNA Sequencing

RNA sequencing data was provided by KRM and generated as reported in Machlus et al.⁹ Briefly 2 independent isolations of mouse megakaryocytes (at the round MK, proplatelet, and preplatelet/releasate stages) were performed (C57b1/6, 1 male and 1 female pool, 4-8 mice). Sequencing and analysis were performed as described by Rowley et al.,¹⁰ using the Useq analysis package (applying DESeq’s negative binomial test).^{11, 12} For analysis, a p-value of < 0.05 was used and a False Discovery Rate (FDR) of 5%, to yield a total of 7,094 (3,235 up-regulated and 3859 down-regulated) and 8,845 (4,125 up-regulated and 4,720 down-regulated) differentially expressed genes in round vs. proplatelet forming MKs and proplatelet MK vs. preplatelet releasate respectively. PCA plots of the samples are included as supplementary figure 3.

Analysis approaches

Whole exome sequencing (WES) analysis was performed in patient genomic DNA as previously reported.² To improve filtering of candidate genetic variants, further bioinformatic analysis was performed using python in the Pycharm IDE. Exomes were first uploaded as pandas dataframes, before filtration to identify rare (as in previous work this is classified as below a frequency of 0.0001 in EVS and the 1000 genomes) and novel non-synonymous variants as described by Johnson et al. These were cross-referenced against a known list of genes that have been previously implicated in inherited causes of bleeding. Relevant gene panels used were Inherited Bleeding Disorder, High Evidence ‘Green’, Medium Evidence ‘Amber’ and Low Evidence ‘Red’, gene lists, containing 116 genes (Supp. Table S1) derived from Panel app (https://panelapp.genomicsengland.co.uk/) and were applied in the project and deemed suitable for research purposes. However, of this gene panel only 88 genes from the Genomics England website (R90) are considered as suitable for clinical use at this present time.

The resulting variants were collated and graphed using PRISM Graphpad 8. DESeq2 output data was collated as pandas dataframes depending on the samples analysed, and visualised using matplotlib and seaborn. Unsupervised hierarchical clustering of RNAseq data was performed to stratify groups of variants according to their level of expression (up-regulated/downregulated) over the course of thrombopoiesis. The annotated scripts are available in GitHub for public use (https://github.com/aokhan/gapp-python-processing) and further details are available on request.

Results and Discussion

Of the 129 patients recruited in this study and assessed by the GAPP study, 55 patients have been reported in previous work and are now classified as having ‘known’ causes of bleeding (Figure 1B). The ‘unknown’ cohort is comprised of 74 patients with extensive clinical bleeding histories, platelet function testing, and WES analysis (Figure 1C). Therefore, to identify new candidate gene variants we developed a python-based indexing platform which first filtered the 129 patient exomes for rare and novel non-synonymous and indel variants, before cross referencing the resulting dataframe with a list of 116 genes already established as clinically causative of bleeding (detailed in methods and Supplementary Figure 1). We reasoned that an automated approach which uses an up to date list of clinically relevant platelet-based bleeding genes would better identify candidate mutations in the full GAPP cohort. When we compare the detection rate using our initial methodology in the cohort of 55 exomes/patients reported by Jonson et al. to our python workflow, we identify the original 40 candidate genes, as well as an additional 8 candidates through cross-indexing with the list of clinical bleeding genes. We observe an increase from 72.3% (40 out of 55 exomes) to 87.3% (48 out of 55). Expanding this analysis to the full cohort of 129 exomes, rare and novel genetic variants in a further 59 individuals, previously part of the ‘unexplained platelet disorder with bleeding’ cohort were identified (Figure 1C). These include new and unreported variants in established genes (NBEAL2, TUBB1, RUNX1, DIAPH1, VWF),¹³ as well as candidates in lesser-known genes more recently associated with bleeding (PTPRJ, MECOM, TPM4).^14–16

Our automated indexing of rare and novel variants has effectively highlighted a number of new candidates in ‘known gene’ categories (which we have labelled as ‘newly identified’), but there still remained a cohort of 14 patients with no rare variants in any known disease-causing genes. Given that this remaining cohort of patients is a challenging one to dissect, we developed a filtering approach based on an expression data resource generated from maturing and proplatelet forming murine megakaryocytes (MKs) published by Machlus et al.⁹

RNAseq was performed on populations of murine mature round MKs, proplatelet forming MKs, and released proplatelets to generate a palette of gene expression data representative of thrombopoiesis using an established in vitro model (Figure 2A).⁹ The data was analysed and processed by DESeq2 and filtered to omit low read counts before plotting the resulting 13,062 genes which were significantly differentially expressed across the maturation of MKs into the process of platelet production (Figure 2B, 2C). These genes were plotted as log2 Fold change in expression in the round MK vs proplatelet forming MK population (y-axis, Figure 2C) and the proplatelet forming MK vs released proplatelets (x-axis, Figure 2C). Therefore, these data represent an array of genes, which are up- and down-regulated during and throughout thrombopoiesis.

Figure 2. Cross indexing RNAseq data of maturing and proplatelet producing MKs yields novel candidates in ‘unknown’ cohort of patients.

(A) An RNASeq dataset derived from mature, round MKs, proplatelet forming MKs, and pre/platelet releasate was obtained (n =2). (B) A heatmap of differential expression and (C) a plot of fold change across round MK: proplatelet MKs and proplatelet MKs: released proplatelets showed that this data set effectively models gene expression during thrombopoiesis. (D) Filtering this data against known disease-causing genes and GAPP variants identified up-regulated genes and candidate variants which include a number of key known genes. These data can be further clustered to segment populations of differentially up-regulated and down-regulated genes. (E) Cross referencing rare and novel non-synonymous variants from the GAPP cohort of 14 unknown patients allowed for the segmentation of candidate novel variants which were significantly up-regulated.

We next filtered this expression data set against the array of known disease-causing genes to validate this approach, and found that the 116 candidates in this list are evident as a population of genes which are differentially expressed in this sample, with base mean read sizes visualised by point size and colour (Figure 2D). Of these, 25 genes are downregulated during thrombopoiesis, while the remaining 91 are upregulated. An annotated crop of this plot in Figure 2D shows established genes well represented by the data (GATA1, THPO, TUBB1, MYH9, RUNX1). We then cross-referenced our rare GAPP-study patient variants with genes represented in these plots to generate a list of 7034 candidate patient variants which were present in this thrombopoiesis data. We then performed an unsupervised hierarchical clustering of these genes to stratify groups of variants according to their level of expression (up-regulated/downregulated) over the course of thrombopoiesis. Finally, we restricted this analysis to our remaining cohort of 14 patients with unidentified genetic causes of bleeding (Supplementary Figure 2), and upon filtration and clustering, identified 360 novel variants in genes differentially upregulated during thrombopoiesis. These genes have not been previously associated with bleeding in the literature, and therefore represent an opportunity to identify novel players in thrombopoiesis that may be the cause of bleeding in this patient cohort.

Amongst these novel candidates, we found a novel variant in DNAI1¹⁷ in a patient with a documented platelet secretion defect (Table 1). Recent work from our group investigating the mechanisms by which the polymodification of TUBB1 mediates platelet production reported DNAI1 as a motor protein which is carefully spatially regulated by the post-translational modification of TUBB1.¹⁸ Mutations in DNAI1 have previously been documented to cause primary ciliary dyskinesia in a very small number of cases with no evidence of a platelet disorder documented.¹⁹ Similarly, in a patient with a severe thrombocytopenia (platelet count of 60 × 10⁹/L) and severe bleeding (BAT score 13) we found a frameshift insertion in MTMR11 (encoding human myotubularin 11), a protein which remains unstudied but is abundant in megakaryocytes in human bone marrow sections.²⁰ This variant has not been reported in a patient or population previously (gnomAD accessed August 2020) and with a significant predicted pathogenicity score of 37 (CADD). In another unrelated individual with a severe macrothrombocytopenia (platelet count of 43 × 10⁹/L, size of 15 fL), a stop/gain mutation in an alpha-granule protein, MEGF11,²¹ was found which was again absent in population databases (gnomAD accessed August 2020) (Table 1, Supplementary Figure 2).

Table 1.

Summary of novel candidates in GAPP patients with unexplained platelet related bleeding. Individuals with established platelet defects (secretion and thrombocytopenia) and significant BAT scores are reported with newly uncovered candidate mutations in genes expressed in megakaryocytes, but previously unstudied. Frequencies reported according to the latest data on gnomAD (https://gnomad.broadinstitute.org/).

Sample ID	Platelet Defect	Platelet count (x109/L)	Platelet Size (fL)5	BAT Score	Candidate gene	Mutation	Variant	Consequence	Frequency (Gnomad)
1010	Secretion	N/A	N/A	8	DNAI1	Missense	c.738G>C	p.E246D	Novel
					TRIM17	Missense	c.1289C>T	p.S430F	Novel
2521	Thrombocytopenia	53	11.7	5	COL7A1	Frameshift	c.7786delG	p.G2596fs	Novel
					MAP7	Missense	c.1822G>C	p.E608Q	Novel
1859	Thrombocytopenia	9.3	9.5	10	RAB36	Splicing	c.426-4A>G	p.?	Novel
					MYH7B	Splicing	c.4316-2T>C	p.?	Novel
1255	Thrombocytopenia	60	8.6	13	MTMR11	Frameshift ins.	c.8_9insG	p.W3fs	Novel
1248	Thrombocytopenia	43	14	5	MEGF11	Stop/gain	c.438C>A	p.C146X	Novel

In summary, one of the largest challenges in the study of inherited bleeding lies in interrogating a significant cohort of patients who, despite extensive study, have yet to yield a conclusive genetic cause of disease. This is not only true of the GAPP cohort, but of other studies focusing on such rare heritable disease.^{13, 22} Therefore, we have created an iterative filtration process to highlight new candidate variants in a population of patients which, to date, had no identified cause of bleeding. This approach is open source, automated, and easily adaptable to apply filtration to any list of genes of interest.

We use a murine MK model of thrombopoiesis as a model RNAseq data set against which we filter our ‘unknown’ cohort of patient genomic data. To our knowledge, this is the first such attempt to interrogate a clinical cohort of patients with bleeding using a combined WES and RNAseq bioinformatics approach. While murine models of thrombopoiesis are well established and extensively used in the study of platelet-based bleeding due to the high degree of homology between the species, restricting our analysis to this data set is a likely limitation of the study. Future work will focus on applying this approach in parallel to human models of thrombopoiesis, including CD34 and iPSC derived MKs.

By applying our methods, we generated a large pool of candidates which will become the focus of future studies as we attempt to dissect a well-studied but poorly understood cohort of patients. Further study is needed to definitively prove the involvement of candidate genes and as such, future work will focus on the re-recruitment of patients for mechanistic investigation.

Essentials.

• ◦Identifying genetic variants in platelet disorders is challenging due to its heterogenous nature
• ◦We combine exome sequencing, RNAseq, and python-based bioinformatics to identify novel gene variants
• ◦We find novel genes in patient data which are upregulated in a murine RNAseq data set modelling thrombopoiesis
• ◦This innovative combined bioinformatic approach provides novel data for future research in the field