Functional analysis of germline RUNX1 variants identified in individuals with suspected familial platelet disorder

Key Points

• •Functional studies support reclassification of novel germ line RUNX1 variants, informing Variant Curation Expert Panel on variant curation.
• •Transactivation assays alone are insufficient for RUNX1 variant classification, highlighting the need for complementary functional assays.

Visual Abstract

Abstract

Deleterious germline RUNX1 variants lead to familial platelet disorder with myeloid malignancies (FPDMM), an autosomal dominant disorder characterized by thrombocytopenia, qualitative platelet defects, and an increased risk of hematologic malignancies. Interpretation of variants of uncertain significance (VUSs) remains a major barrier to precision medicine for affected individuals and families, particularly because many RUNX1 variants are private. To address this challenge, we developed a comprehensive pipeline of well-established in vitro functional assays, including transactivation reporter assays, electrophoretic mobility shift assays, western blot with cell fractionation, and immunofluorescence imaging to assess subcellular localization. We evaluated the functional impact of 26 RUNX1 VUSs identified in individuals referred to the National Institutes of Health RUNX1 Natural History Study. Among these, 5 established benign and 6 pathogenic variants were included as controls. Functional data were integrated with population allele frequencies, in silico predictions, and clinical phenotype and segregation data, enabling reclassification of 17 variants. Our findings also reveal key limitations of the current Myeloid Malignancy Variant Curation Expert Panel guidelines for variant interpretation, including overly stringent thresholds that exclude functionally impaired variants from classification. In addition, we demonstrate that currently approved assays inadequately capture the pathogenic impact of C-terminal variants. These results underscore the importance of comprehensive functional data to support RUNX1 variant classification and inform clinical interpretation. Moreover, this work facilitates investigation of FPDMM pathogenesis and supports refinement of RUNX1-specific variant curation guidelines. We further advocate for development of additional functional assays, especially for variants affecting the RUNX1 C-terminus.

Introduction

The RUNX transcription factors, such as RUNX1, RUNX2, and RUNX3, play important roles in cell proliferation, differentiation, and lineage specification. All RUNX proteins contain a highly conserved N-terminal Runt homology domain (RHD), which is responsible for DNA binding and interaction with its heterodimeric partner CBFβ.1, 2, 3, 4 The less-conserved C-terminus contains regulatory domains commonly known as the transactivation domain (TAD) and inhibitory domain (ID), which are important for the transcriptional regulation function of heterodimeric CBF complexes.5, 6, 7, 8 As a transcription factor, RUNX1 resides mainly in the nucleus and contains both a nuclear localization signal (NLS) and a nuclear matrix targeting signal (Figure 1).^5,9

Figure 1.

Germline RUNX1 variants investigated in this study. Schematic representation of RUNX1c protein isoform (NP_001745.2). RUNX1 variants are colored according to their MM-VCEP classification, with benign/likely benign variants indicated in purple, pathogenic/likely pathogenic in pink, and VUS in teal. RUNX1 variants are shaped according to their type, with missense variants depicted as circles, nonsense as triangles, and frameshift as squares. Amino acid residue numbers are indicated in black. Dashed lines indicate exon boundaries. RHD (AA 77-204)^5,10; NLS (AA 194-210)⁵; TAD (AA 318-398)⁵; NMTS (AA 351-381)⁹; ID (AA 399-438)⁵; and VWRPY (AA 476-480). Variants marked with “#” were reclassified during this study. NMTS, nuclear matrix targeting signal.

RUNX1 and CBFβ are master regulators of hematopoiesis.^11,12 As the heterodimeric non-DNA binding partner of RUNX factors, CBFβ can increase RUNX1’s DNA binding affinity and stability of the CBFβ–RUNX1–DNA complex.¹³ Somatic variants and chromosomal abnormalities affecting RUNX1 are commonly observed in hematologic malignancies such as myelodysplastic syndromes and acute myeloid leukemia.¹⁴ Heterozygous deleterious germline variants in RUNX1 cause familial platelet disorder with associated myeloid malignancies (FPDMM; also known as FPD, RUNX1-FPD, and FPD/acute myeloid leukemia),¹⁵ an autosomal dominant disorder characterized by thrombocytopenia, qualitative platelet dysfunction, and a 30% to 50% lifetime risk of hematologic malignancy.15, 16, 17 In 2019, we launched the first longitudinal natural history study of patients with germline RUNX1 variants to better understand FPDMM,^18,19 with the goal of guiding improved clinical care.

As precision medicine advances and genomic sequencing becomes increasingly integrated into clinical care, the identification of variants of uncertain significance (VUSs) still remains a major clinical challenge. This is particularly true for RUNX1 given that most variants are private and segregation data are often limited, complicating variant interpretation.^20,21 For affected individuals and their families, unresolved VUSs limit diagnostic certainty, risk assessment, and genetic counseling.

The Clinical Genome Resource (ClinGen) Myeloid Malignancy Variant Curation Expert Panel (MM-VCEP) has developed RUNX1-specific variant curation rules based on the guidelines of the American College of Medical Genetics and Genomics and the Association for Molecular Pathology.^22,23 These rules incorporate data on variant type, population frequency, clinical data, and functional evidence to support rigorous and consistent classification of variants. Providing definitive variant pathogenicity is essential for clinical management, patient surveillance, treatment decision-making, and family planning. Therefore, this study aimed to evaluate the use of functional data to support variant reclassification among patients referred to the RUNX1 Natural History Study.

Methods

Patients and samples

Patients with RUNX1 VUSs were referred for evaluation to the National Institutes of Health (NIH) RUNX1 Natural History Study entitled “Longitudinal Studies of Patients with FPDMM” (ClinicalTrials.gov identifier: NCT03854318), which was approved by the NIH Institutional Review Board. A subset of referred individuals was enrolled after providing written informed consent in accordance with the Declaration of Helsinki. Only authors involved in the study protocol had access to primary clinical data.

Plasmids and site-directed mutagenesis

RUNX1 complementary DNA (NM_001754.5) was cloned into pCMV or pmCherry-C1 vectors, whereas CBFB complementary DNA was cloned into pGEM or pEGFP-C1 vectors. RUNX1 variants were generated using the QuikChange II XL Site-Directed Mutagenesis Kit (Agilent Technologies, Santa Clara, CA) and confirmed by Sanger sequencing (ACGT Inc, Germantown, MD). Additional information about the vectors used in this study is described in the supplemental Methods.

Cell culture

HEK293T and HeLa cells were cultured in Dulbecco modified Eagle medium (Millipore Sigma, Burlington, MA) with 10% fetal bovine serum (Cytiva, Marlborough, MA) and 100 U/mL penicillin/streptomycin (Thermo Fisher Scientific, Waltham, MA) at 37°C in 5% CO₂.

CSF1R luciferase reporter assay

HEK293T cells were transiently transfected with pMCSF-R-luc,²⁴ pGL4.70[hRluc] (Promega Corporation, Madison, WI), pCMV-RUNX1 (wild-type [WT] or variant), and pGEM-CBFB using Lipofectamine LTX with PLUS Reagent (Thermo Fisher Scientific). Dual-luciferase activity was measured 24 hours after transfection (Promega) and normalized to WT RUNX1 with CBFβ.

EMSA

RUNX1 proteins were generated using the TnT Quick Coupled Transcription/Translation System (Promega), whereas purified CBFβ was generated by GenScript Biotech (Piscataway, NJ). Electrophoretic mobility shift assay (EMSA) was performed using IRDye700-labeled oligonucleotides (supplemental Methods). Binding reactions were analyzed via polyacrylamide gel electrophoresis and gels were imaged using an Odyssey CLx Imager (LI-COR, Lincoln, NE).

Western blot

Cytoplasmic and nuclear proteins were extracted using the Nuclear/Cytosol Fractionation Kit (Abcam, Waltham, MA), quantified using Qubit (Thermo Fisher Scientific), separated by sodium dodecyl sulfate–polyacrylamide gel electrophoresis, and transferred to a polyvinylidene difluoride membrane. Blots were probed for RUNX1, CBFβ, GAPDH (glyceraldehyde-3-phosphate dehydrogenase), and lamin A/C (supplemental Methods). Data were acquired using the Odyssey CLx imaging system (LI-COR), and band intensities were quantified using Empiria Studio software (LI-COR).

Immunofluorescence staining

HeLa cells were cotransfected with pmCherry-C1 RUNX1 (WT or variant) and pEGFP-C1 CBFβ.²⁵ Cells were fixed, permeabilized, blocked, and stained with an α-tubulin primary antibody (Abcam). DAPI (4′,6-diamidino-2-phenylindole) was used as a nuclear marker. Imaging was performed using a DeltaVision system with consistent acquisition settings (supplemental Methods).

Variant interpretation

Variant classifications were assigned using MM-VCEP version 2 RUNX1 specifications based on the American College of Medical Genetics and Genomics and the Association for Molecular Pathology guidelines.^22,26 Probands were defined as unrelated individuals with thrombocytopenia, qualitative platelet defects, and/or hematologic malignancy who carried the variant of interest. Segregation evidence was assessed using informative parent-offspring transmissions among genotype-positive/phenotype-positive individuals and obligate carriers in accordance with ClinGen recommendations.²⁷ Additional evidence included variant type, population frequency, localization within functional domains or clinically validated hot spots, and computational pathogenicity predictions (eg, REVEL).²²

Statistical analysis

Analyses were performed using Prism 10 (GraphPad, Boston, MA). For comparisons involving multiple variants, 1-way analysis of variance was used followed by Dunnett’s multiple comparisons test, comparing each RUNX1 variant with the WT control. Data are presented as mean ± standard deviation and P < .05 were considered statistically significant.

Results

RUNX1 variants in patients with FPDMM

The NIH RUNX1 Natural History Study (NCT03854318) has enrolled >200 individuals from >150 unrelated families carrying a range of germline RUNX1 variants. Individuals are frequently referred to the study with VUSs, and disease status is often unclear at the time of referral. To aid clinical variant interpretation, we evaluated these VUSs using well-established in vitro functional assays. Among the 38 variants tested (supplemental Table 1), we included 5 known benign variants (I22K, L56S, A60V, P275L, Q335H) and 6 known pathogenic variants (K110E, R166∗, R201∗, R201Q, R204∗, and R204Q) as controls. The likely pathogenic S388∗ variant, located in the C-terminus, was also included as the first hypermorphic variant characterized in FPDMM.²⁸ The tested VUSs were located throughout RUNX1, including 10 within the RHD, 11 in the TAD and ID, and 5 variants outside these regions (Figure 1). Table 1 summarizes the available clinical data for patients with VUSs analyzed in this study.

Table 1.

Summary of available clinical data and published reports on different germline RUNX1 VUSs investigated in this study

RUNX1 variant		Reference SNP report no.	Clinical data	Other reports
Nucleotide†	Amino acid‡
155T>A	M52K	200431130	1 proband with thrombocytopenia	NA
188C>T	A63V	NA	1 proband with pancytopenia	NA
280A>C#	S94R	1057523598	A family with 2 confirmed members (2 segregations), 1 with AML and thrombocytopenia, and another with just thrombocytopenia	NA
352G>A	V118M	NA	1 proband with thrombocytopenia	NA
367G>C	D123H	373498347	2 probands with platelet abnormalities	1 pedigree with thrombocytopenia (PMID 18723428) and 1 individual with AML (PMID 28933735)
422C>T	S141L	1601516118	1 proband who was found to have the variant as a secondary finding for other medical issues	1 proband with CMML diagnosed at age 67 years and antecedent thrombocytopenia (PMID 34028844) and another with myeloid disease/thrombocytopenia29
466G>A	A156T	NA	2 family members with easy bleeding and bruising but normal platelet counts; 1 with abnormal platelet aggregometry and 1 with normal platelet aggregometry	NA
477T>G	N159K	NA	1 family with 5 confirmed individuals and 2 obligate carriers with abnormal platelet function (4 segregations), 1 obligate carrier had AML	NA
482T>C	L161P	2057885248	Found to be somatic in the patient referred to our protocol	NA
484A>G#	R162G	1057519751	1 family with thrombocytopenia in 3 affected family members (2 segregations)	NA
508G>C	G170R	2146360175	Proband and child with thrombocytopenia and 2 additional family members with AML (3 segregations)	NA
590T>C	V197A	NA	De novo in 1 proband with aplastic anemia and prior thrombocytopenia	NA
645G>T	K215N	1312588693	1 proband with aplastic anemia and no prior history of abnormal bleeding, bruising or thrombocytopenia	NA
649G>A	G217R	749004431	1 proband with AML	NA
892C>T	P298S	746531841	1 proband with AML but no thrombocytopenia and the variant does not segregate with disease in the family	NA
967A>G	T323A	776703625	No clinical information available	One proband with B-ALL (PMID 34166225)
1098_1103dup	I366_G367dup	750495319	3 probands with thrombocytopenia	NA
1098_1103del	I366_G367del	750495319	1 proband with transient pancytopenia	NA
1154A>G	Y385C	2056456351	Proband with bleeding history, but platelets were normal upon evaluation, and the variant did not segregate in the family	NA
1159_1161del	G387del	975536871	1 proband with anemia but normal platelets	NA
1184C>T	P395L	868527382	2 probands with thrombocytopenia	NA
1196G>T	S399I	946348885	1 proband with aplastic anemia	Pediatric patient with B-ALL and a patient with primary myelofibrosis (PMID 34166225) and a Greek patient aged 58 years with ovarian cancer and a RAD51C PV (PMID 31135094)
1221C>G	Y407∗	914897271	No clinical information available	NA
1270T>C	S424P	2056451534	No clinical information available	Proband with thrombocytopenia that also carried RUNX1 c.1256T>G and a deletion including FLI1 (PMID 32935436), an individual with B-ALL, likely because of PAX5, and another proband with thrombocytopenia (PMID 35902733)
1289C>T	P430L	1253570532	1 proband with thrombocytopenia and abnormal bleeding	NA
1294T>C	C432R	1060499613	No clinical information available	NA

AML, acute myeloid leukemia; B-ALL, B-cell acute lymphoblastic leukemia; CMML, chronic myelomonocytic leukemia; NA, not applicable; PV, pathogenic variant; SNP, single nucleotide polymorphism.

^†The reference sequence used is NM_001754.5.

^‡The reference sequence used is NP_001745.2 (RUNX1c).

^#Variants reclassified by other groups during this study.

The importance of an appropriate control for the luciferase reporter assay

Internal controls are critical for luciferase reporter assays to account for variations in transfection efficiencies and sample handling. The Renilla luciferase gene under the HSV-1 TK promoter is among the most frequently used internal controls in such assays,³⁰ including in RUNX1 transactivation assays using firefly luciferase driven by RUNX1 target promoters, such as CSF1R.³¹ However, we observed a correlation between the presence of RUNX1 activity and the control Renilla activity, with WT and benign RUNX1 variants transactivating both firefly and Renilla luciferases, whereas known RUNX1 pathogenic controls exhibited reduced activities of both luciferases (Figure 2A). These findings suggest that Renilla expression from the TK promoter construct is under the control of RUNX1. If true, we posited that use of such an internal control would mask the real effects of hypo- or hypermorphic RUNX1 variants on CSF1R transcriptional regulation. In fact, we identified several potential RUNX1 motifs in the TK promoter sequence (Figure 2C). After screening other common promoters such as SV40 and CMV in the same backbone, and detecting RUNX1 motifs within these sequences as well, we opted to test the previously reported promoter-less Renilla vector.^32,33 This vector is based on the optimized pGL4 backbone (pGL4.7hRluc), which contains fewer regulatory elements and transcription factor binding sites. This internal control plasmid showed no significant effect on its activity by RUNX1 (Figure 2B), which allowed us to evaluate the effects of different RUNX1 variants on CSF1R luciferase activity and normalize these data.

Figure 2.

RUNX1 transactivation results for 38 variants including controls assessed by CSF1R luciferase reporter assay. (A) Relative luciferase activities for the test reporter (firefly luciferase) driven by the CSF1R promoter and the control reporter (Renilla luciferase) driven by the TK promoter, in the presence of the indicated effector plasmids, which were transfected into HEK293T cells. Data indicate mean ± 1 standard deviation (SD; n ≥ 3). Data are relative to the WT RUNX1c plus CBFβ control. (B) Relative luciferase activities for the test reporter (firefly luciferase) driven by the CSF1R promoter and the promoter-less control reporter (Renilla luciferase), in the presence of the indicated effector plasmids. Data indicate mean ± 1 SD (n ≥ 3). Data are relative to the WT RUNX1c plus CBFβ control. (C) Presence of multiple RUNX1 binding motifs (short bars in the middle of the plasmid map) in the pRL-TK Rluc vector. The consensus RUNX1 binding motif sequence is displayed at the top. (D) Firefly/Renilla luciferase activity ratios relative to WT RUNX1c plus CBFβ control (indicated by the dotted line), using the CSF1R reporter and the promoter-less control reporter in HEK293T cells. All RUNX1 variants were coexpressed with CBFβ. Data indicate mean ± 1 SD (n ≥ 3). Significant difference between different RUNX1 variants and WT with CBFβ was analyzed using 1-way analysis of variance using the Dunnett’s post hoc test. RUNX1 variants are colored according to their classification with benign/likely benign indicated in purple, pathogenic/likely pathogenic in pink, and VUS in teal. Statistical significance is defined as ∗P < .05, ∗∗P < .01, and ∗∗∗P < .001.

Pathogenic RUNX1 variants located in the RHD lead to loss of RUNX1 transcriptional activity

To functionally characterize the 26 RUNX1 VUSs, we started by evaluating their ability to induce transcriptional activity of the CSF1R promoter in HEK293T cells cotransfected with CBFβ. Validation of RUNX1 and CBFβ protein expression was performed by western blot (supplemental Figure 1). As expected, we observed reductions in RUNX1 transactivation for all pathogenic controls, whereas all benign controls showed transactivation levels comparable to WT (Figure 2D). Several VUSs located within the RHD, including S94R, R162G, and G170R, exhibited reduced transactivation activity relative to WT RUNX1 (P < .01), with S94R showing <20% of WT activity. Transactivation activity for variants S141L, L161P, and V197A trended lower, but did not reach statistical significance. Variants S94R and R162G, classified initially as VUSs, have recently been reported by other groups as being likely pathogenic³³ and pathogenic,³⁴ respectively. In our study, both were reclassified as likely pathogenic, consistent with our reporter assay results (Figure 2D).

In contrast, the known likely pathogenic variant S388∗ exhibited a trend (not reaching statistical significance) toward increased transactivation activity as previously reported,²⁸ although a similar trend was not observed for the respective S361∗ variant in the RUNX1b isoform.³³ The distal truncation VUS Y407∗, similarly located in the C-terminus, showed significantly increased transcriptional activity (Figure 2D). Moreover, none of the missense VUSs located in the C-terminus showed reduced transcriptional activities, and 2 of them, S424P and P430L, exhibited increased transcriptional activity compared to WT.

RUNX1 variants within the RHD impair its ability to bind DNA and/or CBFβ

RUNX1 binds to specific DNA elements through its RHD to regulate gene expression. To determine the functional impact of different VUSs on RUNX1 DNA binding ability, we performed EMSA (Figure 3). All benign controls exhibited DNA binding comparable to WT RUNX1. In contrast, all pathogenic controls within the RHD failed to bind the DNA probe, regardless of CBFβ presence, with the exception of R204Q. This variant could not bind the DNA probe on its own but was able to form a complex in the presence of CBFβ, as reported previously.^34,35

Figure 3.

DNA binding affinities of different RUNX1 variants for the CSF1R promoter sequence. EMSA results showing the binding affinities of RUNX1 WT and different RUNX1 variants with and without CBFβ. RUNX1 proteins were generated by coupled in vitro transcription and translation. Purified CBFβ protein was used for this assay, alongside a IRDye700-labeled 20-bp CSF1R promoter sequence as the DNA probe. + competitor represents an unlabeled 20-bp CSF1R promoter sequence which was used as cold competitor and added at 200 times in excess. + antibody represents a complex which was supershifted by adding an anti-RUNX1 antibody (ab23980). Purple color indicates benign RUNX1 variants, pink color indicates pathogenic and likely pathogenic RUNX1 variants, and teal color indicates RUNX1 VUS variants. Gels were imaged using the Odyssey CLx imager (LI-COR).

Absence of DNA binding was also observed for VUS R162G, consistent with previous reports of this amino acid being important for RUNX1-DNA interactions.¹⁰ VUSs S94R and S141L, both located within the RHD, showed reduced DNA binding in the presence or absence of CBFβ, with the effect being more pronounced in S94R. Both residues are known to interact with CBFβ because they lie in the interface with CBFβ.^4,36 In addition, 2 other VUSs in the RHD, L161P and G170R, bound DNA only in the presence of CBFβ, similar to R204Q. As expected, all C-terminal variants past the RHD exhibited normal DNA and CBFβ binding, including the likely pathogenic variant S388∗, suggesting that RHD integrity is critical for DNA and CBFβ binding.

Western blot with cell fractionation is an important secondary assay for variant interpretation

We next assessed RUNX1 and CBFβ subcellular localization by western blot with nuclear and cytoplasmic fractionation in HeLa cells coexpressing WT/variant RUNX1 and CBFβ (Figure 4). When the cells were transfected with WT RUNX1 and/or CBFβ, we readily detected RUNX1 in the nuclear fraction, whereas CBFβ alone was present in the cytoplasm and migrated into the nucleus when expressed with RUNX1. All benign RUNX1 controls behaved similar to WT, whereas most pathogenic controls showed decreased RUNX1 nuclear localization and consequently impaired CBFβ nuclear translocation, thus increasing their presence in the cytoplasm. In particular, early truncations such as R166∗, R201∗, and R204∗ showed the biggest effects on RUNX1 and CBFβ subcellular localization, consistent with their loss of the NLS and inability to interact with CBFβ. In contrast, the effect of R204Q variant on RUNX1 and CBFβ subcellular localizations was less significant. The reduced CBFβ heterodimerization by the S94R variant was also confirmed by the separation of RUNX1 S94R in the nucleus and CBFβ in the cytoplasm. Furthermore, VUSs V118M, L161P, and R162G exhibited lower percentages of RUNX1 nuclear localization at ∼63% to 65%, a similar value compared with the pathogenic variant R204Q (62%). These 3 variants are located within the RHD, with R162 residue being important for RUNX1-DNA binding. Although R204Q failed in transactivation, no significant impairment was observed for V118M and L161P.

Figure 4.

Subcellular localization of RUNX1 and different variants with CBFβ by western blot with cell fractionation. RUNX1 WT or variants were coexpressed with CBFβ by plasmid transfection into HeLa cells. After 24 hours, cytoplasmatic and nuclear proteins were extracted and analyzed by western blot using anti-RUNX1 and anti-CBFβ antibodies. GAPDH and lamin A/C were used as cytoplasmatic and nuclear loading controls, respectively. Blots were imaged using the Odyssey CLx system (LI-COR), and band intensities were quantified using Empiria Studio software (LI-COR). Each antibody image was cropped to show relevant bands and aligned per blot. The main truncated RUNX1 proteins are identified with an arrow and were used for quantification purposes. Percentages of RUNX1 and CBFβ proteins in the nucleus are indicated under each condition and were calculated from this single representative western blot.

Immunofluorescence imaging supports western blot with cell fractionation assay for subcellular localization

To complement the western blot analysis, we assessed RUNX1 and CBFβ subcellular localization through immunofluorescence using the same cells cotransfected with mCherry-RUNX1 (WT or variant) and eGFP-CBFβ plasmids. By itself, WT RUNX1 localizes to the nucleus, whereas CBFβ is found in both the cytoplasm and nucleus, translocating to the nucleus after its heterodimerization with RUNX1 (Figure 5). Overall, all benign variants showed normal colocalization of CBFβ with RUNX1 within the nucleus, as exemplified by L56S (Figure 5). In contrast, pathogenic variants with early truncations, such as R166∗, exhibited a strong mislocalization of both RUNX1 and CBFβ (Figure 5). Other pathogenic variants, including K110E, R201Q and R204Q, showed variable levels of RUNX1 and CBFβ mislocalization, with R201Q having more significant effect (supplemental Figure 2). Interestingly, apart from K110E, both R201Q and R204Q variants affect the NLS, which is located between amino acids 194 and 210. Among all other RUNX1 variants tested in this study, S94R showed the strongest mislocalization of CBFβ. This resulted from defective interaction between the 2 proteins because RUNX1 S94R remained in the nucleus (Figure 5). CBFβ mislocalization was also present in cells expressing the VUS R162G.

Figure 5.

Immunofluorescence results for CBFβ subcellular localization in the presence of different RUNX1 variants. HeLa cells were transiently transfected with plasmids coexpressing WT or variant RUNX1 along with CBFβ. After 24 hours, cells were fixed, permeabilized, and stained with α-tubulin (cytoplasmic) and DAPI (nuclear) markers. Images were acquired on a DeltaVision system (Leica Microsystems, Hesse, Germany) with consistent settings. GFP, mCherry, Alexa Fluor 633, and DAPI signals were collected with standard emission filters. Brightness of the mCherry channel (RUNX1) was uniformly increased by 20% across all panels to improve signal visibility without obscuring or misrepresenting any information in the original images. Representative images are shown separated by white lines. Scale bar, 25 μm.

SEUSS analysis and computational predictions of RUNX1 variants corroborate the in vitro functional studies

In our analyses, we also incorporated data from scalable functional screening by sequencing (SEUSS), a Perturb-seq–style single-cell transcriptomic assay assessing transcriptional impact of genetic variants.³⁷ A total of 115 RUNX1 variants were overexpressed in K-562 cells, including WT and loss-of-function (LOF) controls. S94R, S141L, L161P, R162G, and G170R were directly represented in the empirical SEUSS library, where all 5 variants were classified as LOF-like or hypomorphic, corroborating with our in vitro findings (Table 2). For the remaining variants, functional predictions were derived from the Random Forest classifier trained on SEUSS transcriptomic signatures (supplemental Methods). Table 3 shows the resulting prediction scores, with values <0.5 classified as WT-like and values >0.5 indicating a predicted functional impact. Notably, these predictions have higher confidence for RHD variants, the domain assessed by the SEUSS library. Overall, these data support our in vitro studies and proposed clinically relevant reclassification of variants S94R, S141L, L161P, R162G, and G170R.

Table 2.

Transcriptional effects of RUNX1 variants assessed by SEUSS

SEUSS data
Nucleotide	Amino acid change	Classification	Domain	Interactions	T2 score	P value	Phenotype
280A>C#	S94R	VUS	RHD	CBFB	847.250 697 6	1.324E-73	LOF-like
367G>C	D123H	VUS	RHD	CBFB	37.484 920 07	.026 785 1	WT-like
422C>T	S141L	VUS	RHD	CBFB	218.575 680 5	2.313E-19	Hypomorphic
482T>C	L161P	VUS	RHD	—	682.289 451 1	1.812E-63	LOF-like
484A>G#	R162G	VUS	RHD	DNA	852.073 313 3	4.345E-79	LOF-like
496C>T	R166∗	P	RHD	—	847.378 659 4	3.263E-93	LOF-like
508G>C	G170R	VUS	RHD	DNA	267.297 454 6	1.75E-23	Hypomorphic
602G>A	R201Q	P	RHD	DNA	452.462 324 3	5.458E-42	LOF-like
601C>T	R201∗	P	RHD	—	760.031 078	1.579E-66	LOF-like

T2 score is the score describing the functional impact of variant (low scores indicate WT RUNX1, whereas higher scores indicate LOF). Phenotype is a functional effect of the variant concluded based on SEUSS transcriptomic results (WT-like, hypomorphic, or LOF-like).

B, benign; P, pathogenic.

^#Variants reclassified by other groups during this study.

Table 3.

Computational predictions based on SEUSS library

Computational predictions based on SEUSS library
Nucleotide	Amino acid change	Category	Domain	Interactions	Prediction score	Prediction
155T>A	M52K	VUS	Not in RHD	—	0.228	WT-like
167T>C	L56S	B	Not in RHD	—	0.335	WT-like
179C>T	A60V	B	Not in RHD	—	0.382	WT-like
188C>T	A63V	VUS	Not in RHD	—	0.398	WT-like
328A>G	K110E	P	RHD	DNA	0.688	Functional
352G>A	V118M	VUS	RHD	—	0.356	WT-like
466G>A	A156T	VUS	RHD	—	0.664	Functional
477T>G	N159K	VUS	RHD	—	0.446	WT-like
590T>C	V197A	VUS	RHD	DNA	0.482	WT-like
611G>A	R204Q	P	RHD	DNA	0.811	Functional
645G>T	K215N	VUS	Not in RHD	—	0.197	WT-like
649G>A	G217R	VUS	Not in RHD	—	0.516	Functional
824C>T	P275L	B	Not in RHD	—	0.411	WT-like
892C>T	P298S	VUS	Not in RHD	—	0.419	WT-like
967A>G	T323A	VUS	Not in RHD	—	0.449	WT-like
1005G>T	Q335H	B	Not in RHD	—	0.279	WT-like
1154A>G	Y385C	VUS	Not in RHD	—	0.372	WT-like
1184C>T	P395L	VUS	Not in RHD	—	0.432	WT-like
1196G>T	S399I	VUS	Not in RHD	—	0.384	WT-like
1270T>C	S424P	VUS	Not in RHD	—	0.487	WT-like
1289C>T	P430L	VUS	Not in RHD	—	0.393	WT-like
1294T>C	C432R	VUS	Not in RHD	—	0.304	WT-like

Prediction score is a score ranging between 0 to 1, with <0.5 being classified as WT-like and >0.5 being classified as having a functional impact.

B, benign; P, pathogenic.

Correlation between functional studies to support clinical RUNX1 variant interpretation

According to the current MM-VCEP guidelines,²² transactivation assays exhibiting altered transactivation are considered as the gold-standard assay for RUNX1 variant interpretation, and on their own can meet PS3_moderate or PS3_supporting classification. PS3 classification can be further strengthened by using secondary assays in addition to RUNX1 transactivation, including EMSA, immunofluorescence, or western blot with cell fractionation. We correlated results from various functional assays and summarized them, along with additional classification criteria as outlined by the MM-VCEP in Figure 6. Using the current guidelines, 13 VUSs could be reclassified as benign/likely benign and 4 VUSs could be reclassified as pathogenic/likely pathogenic, providing clinically actionable interpretations for affected individuals and their families. However, 9 variants remained classified as VUSs even with the availability of our functional data. Detailed evidence codes applied to each variant according to the MM-VCEP guidelines are provided in supplemental Table 2. To further illustrate how variant classifications changed with the addition of functional evidence, we generated a Sankey diagram (Figure 7A). Based on these findings, we also propose a framework for RUNX1 variant evaluation (Figure 7B) that highlights a stepwise approach to functional testing and interpretation based on variant type and domain.

Figure 6.

Summary of the results obtained using different in vitro functional assays for the indicated RUNX1 variants compared with WT and relevant data for variant classification using MM-VCEP guidelines. Transactivation was assessed by CSF1R luciferase reporter assay, DNA binding and CBFB heterodimerization by EMSA, and subcellular localization by western blot with cell fractionation and immunofluorescence. Segregations denote passing down through generations of the variant together with FPD. One meiosis denotes 1 parent-offspring passage.

Figure 7.

Reclassification outcomes and proposed decision framework for germline RUNX1 variant interpretation. (A) Sankey diagram illustrating changes in the American College of Medical Genetics and Genomics and the Association for Molecular Pathology variant classification after integration of functional, clinical, and segregation data. Variants are shown transitioning from their initial VUS classification to their final classification after incorporation of in vitro functional assays. The first and second secondary assays include EMSA, western blot with cell fractionation, and immunofluorescence. For RHD variants, EMSA was prioritized as the first secondary assay because of its utility for this domain. Sankey diagram was created using SankeyArt. (B) Proposed decision framework for germline RUNX1 variant interpretation based on variant type and location. Functional assays are prioritized accordingly, with transactivation assays providing the primary evidence for PS3/BS3 classification and secondary assays (eg, EMSA, subcellular localization) contributing additional support, particularly for RHD variants. Variants predicted to undergo NMD are primarily interpreted using LOF criteria, whereas variants outside the RHD or within the C-terminal region may require cautious interpretation and integration of clinical and segregation data. Blue boxes indicate decision nodes, orange boxes indicate interpretation outcomes, and purple boxes indicate additional information. ∗For RHD variants, an additional secondary assay assessing CBFβ interaction may be incorporated to support BS3_strong classification.

Discussion

As of January 2026, 1846 germline RUNX1 variants have been reported in ClinVar, of which 984 are classified VUSs (53%). As RUNX1 is being increasingly included in clinical sequencing panels, the number of detected variants is expected to rise. Accurate classification of these variants is essential for proper diagnosis and effective clinical management of affected individuals. In this study, we provide evidence supporting the reclassification of 13 VUSs as benign/likely benign and 4 as pathogenic/likely pathogenic through systematic integration of functional, segregation, and clinical evidence within the MM-VCEP framework,^22,23 directly informing risk assessment and clinical management for affected individuals and their families. These findings demonstrate that a structured multiassay approach, tailored to variant location and type, can resolve a substantial proportion of RUNX1 VUSs and provide a replicable framework for ongoing variant curation as clinical sequencing expands.

Our study suggests that the heavily conserved RHD is more sensitive to missense variants, consistent with known function of this domain in DNA binding and CBFβ heterodimerization and the well-established 3-dimensional structure of the RHD.^3,4,10 Indeed, variants at structurally critical positions, such as those at the CBFβ interface or DNA-contacting residues, predictably showed functional impairment in our assays. High-throughput tools such as SEUSS, which are designed around these structural interfaces, further corroborate such findings by empirically measuring the functional impact of RHD variants across many variants simultaneously.³⁷ These findings are also supported by several previously published studies. For instance, the S94R variant has been shown to have reduced transcriptional activity compared with WT RUNX1 because it cannot heterodimerize with CBFβ.³⁸ Furthermore, this amino acid residue has been identified as a potential site for cyclin-dependent kinase (CDK) and extracellular signal-regulated kinase phosphorylation.³⁹ Similarly, S141L showed diminished CBFβ interaction and a twofold reduction in DNA binding affinity in previous studies.^4,36,40 Both S94 and S141 are located at the CBFβ interface⁴ and failed to show enhanced DNA binding in the presence of CBFβ in EMSA. R162 and G170, which are essential for DNA binding,⁴¹ were severely impaired in our study, with G170R also acting as a cryptic splice variant.¹⁹ These findings support the inclusion of splicing assays in future MM-VCEP guidelines. Furthermore, these variants (S94R, S141L, R162G, G170R), alongside L161P, were classified as functionally abnormal by SEUSS, which identified S94R, L161P, and R162G as LOF-like and S141L and G170R as hypomorphic. Notably, the variant L161P was found to be somatic in the patient who was first referred to our study as potentially germline. However, we hope that providing functional data will help support classification for future patients with germline variants. Nonetheless, high-throughput tools such as SEUSS may aid future RUNX1 variant classification and expand access to clinically actionable functional evidence. Several truncating variants were studied here, which may lead to nonsense-mediated decay (NMD). However, most of the truncating variants mentioned in this study such as R166∗, R201∗, R204∗, S388∗ were used as controls. Some of them may not produce the encoded truncated proteins in cells because of their more proximal locations. The only truncating VUS is Y407∗, which is likely to escape NMD given that it is located in the C-terminal region of the protein.

Importantly, we also identified a subset of C-terminal variants (Y407∗, S424P, P430L) that exhibited enhanced transactivation while maintaining normal DNA binding and subcellular localization with CBFβ. These residues lie within RUNX1’s ID, a region known to modulate transcriptional repression. Loss or disruption of this domain may therefore explain the observed increased activity for these variants. Previous studies support this finding, including data on S388∗, the first hypermorphic variant characterized in FPDMM,²⁸ and R320∗, which escapes NMD and drives transcriptional changes distinct from RUNX1 depletion.⁴² S424, a direct substrate of CDK1/6, has been shown to increase RUNX1 activity by ∼50% when mutated to S424D,³⁹ consistent with our findings. Y407∗, similarly to S388∗, also likely evades NMD and lacks the highly conserved terminal VWRPY domain, which normally recruits the TLE1 corepressor.^43,44 The ID also mediates homodimerization and RUNX1 autoregulation, and deletion of residues 399 to 480 (encompassing the ID and VWRPY) has been linked to increased CSF1R promoter activity in previous studies.^5,45,46 Therefore, the absence of these regions in Y407∗, while preserving the TAD, could cause enhanced transcriptional activity relative to WT.

To ensure that functional evidence supports clinical variant interpretation, each assay included a panel of 11 disease-relevant internal controls (5 benign and 6 pathogenic), selected based on variant type, disease mechanism, and key RUNX1 protein domains (RHD, TAD).⁴⁷ This exceeds the ClinGen Sequence Variant Interpretation guideline minimum required to support moderate-level evidence in the absence of rigorous statistical analysis. To our knowledge, this is the first study with such comprehensive controls, allowing all 4 assays to meet the criteria for variant classification under the MM-VCEP RUNX1 guidelines and increasing confidence in their clinical interpretability.

The MM-VCEP currently identifies transactivation assays showing altered RUNX1 activity relative to WT as essential for application of the PS3 and BS3 criteria in RUNX1 variant interpretation.²² Data from additional functional assays can further strengthen these classifications. Among these, the luciferase reporter assay (using the MM-VCEP–recommended CSF1R promoter) and EMSA provided the most clinically relevant data. CSF1R outperformed other promoters such as ETV1 and MYL9 in detecting variants with <20% activity relative to WT in HEK293T cells.^32,33 EMSAs offer a rapid assessment of both DNA binding and heterodimerization with CBFβ, potentially reducing the need for more laborious assays. We propose that EMSAs, together with the CSF1R luciferase reporter assay, be considered a minimal primary functional testing set to support PS3/BS3 classification under MM-VCEP guidelines, facilitating more timely variant interpretation in clinical research settings. However, EMSAs should not replace luciferase reporter assays, as they mainly evaluate RHD variants and are uninformative outside this domain. For example, VUSs S141L and L161P showed reduced transcriptional activity, which was <80% of WT activity, in the CSF1R luciferase reporter assay, although the reductions did not reach statistical significance. In these cases, their abnormal EMSA results provided additional support for pathogenicity, highlighting the value of this assay in classifying RHD variants.

Although luciferase reporter assays provide valuable functional insights, they may not adequately assess the pathogenicity of C-terminal variants, underscoring the need for complementary approaches.³³ C-terminal variants often retain normal DNA binding and localization, and their impact may arise from altered regulatory functions that are not captured by the MM-VCEP–approved functional assays. For example, the likely pathogenic variant S388∗ showed a 170% CSF1R transactivation increase without statistical significance and showed no abnormalities in secondary assays, illustrating the limitations of the assay despite clinical evidence. Therefore, novel assays addressing mechanisms such as loss of inhibitory regulation or disrupted cofactor interactions are urgently needed. Assessing the transcriptional effects of RUNX1 variants on the TCF4 promoter, a gene normally repressed by RUNX1, may offer a sensitive and relevant readout for evaluating the functional impact of C-terminal variants, particularly those affecting the ID.⁴⁸ Additionally, platelet transcriptomic profiling may also provide a complementary functional readout for C-terminal variants, capturing downstream effects of RUNX1 dysregulation in a disease-relevant cell type.^49,50

Our findings also underscore key limitations of the current MM-VCEP RUNX1 variant interpretation guidelines, particularly the strict PS3 and BS3 thresholds. Currently, only variants with luciferase activity <20% or >115% of WT are considered functionally abnormal or enhanced, respectively. However, many variants with statistically significant activity reductions between 20% and 80% demonstrate clear functional impairment but remain ineligible for PS3 support. We propose revising the guidelines to allow PS3_moderate evidence for variants with consistent, statistically significant functional impairment >20% of WT, provided at least 11 internal controls are used. This adjustment would increase the utility of functional assays and help reclassify variants that are currently stuck in the VUS category despite supportive experimental data.

Similarly, the use of a >115% threshold to define hypermorphic variants is problematic. This cutoff was originally based on a single variant, S388∗, which was classified as likely pathogenic without functional data and demonstrated increased transactivation. Our results show that this upper threshold lacks specificity, as several benign controls (I22K, P275L, Q335H) and many VUSs also exceeded 115% activity. This suggests that enhanced transcriptional activity alone is insufficient to infer pathogenicity and may lead to misclassification. We propose that PS3_supporting be applied only when increased transactivation is statistically significant and is accompanied by abnormalities in a second assay. Additionally, given that no single assay can assess all RUNX1 variants across different domains, MM-VCEP guidelines should consider variant location when applying PS3/BS3.

Beyond specific threshold adjustments, our findings more broadly emphasize that functional data should be interpreted in the context of other evidence. In this study, variant classification integrated functional results with clinical phenotype, segregation data, and previous reports, with concordant evidence across these domains supporting more confident reclassification. Functional assays were most informative when used to support or refute other lines of evidence, particularly when clinical data were limited or ambiguous.

Notably, under the current MM-VCEP version 2 criteria, benign functional evidence alone may be sufficient to support reclassification of a variant as likely benign. However, reliance on functional data in isolation raises concerns about potential overclassification of VUSs as benign/likely benign in the absence of additional supporting evidence, especially for variants located in a difficult to assess domain, such as the C-terminus. Future refinements to these guidelines, and development of additional functional assays, should better account for the biological complexity of RUNX1 function and improve the use of functional data in variant interpretation.

Although refining existing functional assays and updating MM-VCEP thresholds will improve variant interpretation, our study reveals deeper limitations in the current framework. Considering that RUNX1 is a multifunctional transcription factor whose activity varies across cell types, interacting partners, and developmental stages, current in vitro assays may not fully capture its diverse roles or the impact of specific variants. Furthermore, the overexpression systems used here do not fully recapitulate physiologic RUNX1 and CBFβ levels, and use of nonchromatinized DNA may not capture the full chromatin context in which RUNX1 normally operates. These constraints may partly explain why some variants fell just below functional evidence thresholds despite additional clinical data supporting pathogenicity, such as V197A and N159K. In cases where RUNX1 function was only mildly impaired, other supporting evidence, including phenotype and segregation, was also weaker. For example, N159K exhibited normal functional results but segregated with mildly reduced platelet counts in 4 family members, suggesting a subtle yet clinically relevant impact. These observations point to a spectrum of variant effects in FPDMM and raise the possibility of risk stratification or disease subgroups that may eventually justify tailored monitoring and surveillance approaches. They also underscore the need for additional functional approaches, such as platelet transcriptomics and proteomics, to better capture RUNX1’s diverse biological roles.^49,50

Moreover, given the limited access of functional testing in diagnostic settings, high-throughput approaches such as multiplexed assays of variant effect hold promise for generating preemptive evidence to guide variant classification.⁵¹ However, for these tools to be clinically informative for RUNX1, their design and readouts must be tailored to variant location, type, and relevant biological functions. Integration of such approaches with evolving curation guidelines will be essential to improve RUNX1 variant interpretation and ultimately enhance clinical management for individuals with FPDMM.

Conflict-of-interest disclosures: The authors declare no competing financial interests.