Splicing analysis of 24 potential spliceogenic variants in MMR genes and clinical interpretation based on refined ACMG/AMP criteria

Abstract

Lynch syndrome (LS) is a common hereditary cancer syndrome caused by heterozygous germline pathogenic variants in DNA mismatch repair (MMR) genes. Splicing defect constitutes one of the major mechanisms for MMR gene inactivation. Using RT-PCR based RNA analysis, we investigated 24 potential spliceogenic variants in MMR genes and determined their pathogenicity based on refined splicing-related American College of Medical Genetics and Genomics/Association for Molecular Pathology (ACMG/AMP) criteria. Aberrant transcripts were confirmed in 19 variants and 17 of which were classified as pathogenic including 11 located outside of canonical splice sites. Most of these variants were previously reported in LS patients without mRNA splicing assessment. Thus, our study provides crucial evidence for pathogenicity determination, allowing for appropriate clinical follow-up. We also found that computational predictions were globally well correlated with RNA analysis results and the use of both SPiP and SpliceAI software appeared more efficient for splicing defect prediction.

Lynch syndrome, splicing defect, ACMG criteria, RNA analysis

Introduction

Lynch syndrome (LS) (MIM#120435) is a common hereditary cancer syndrome characterized by an autosomal dominant predisposition to colorectal and endometrial cancer as well as, less frequently, cancers of other organs including ovary, intestinal and biliary track. It is caused by heterozygous germline pathogenic variants in DNA mismatch repair (MMR) genes namely: MLH1, MSH2, MSH6, PMS2, and by 3′ genomic deletion of the EPCAM gene. Tumors from LS patients show high microsatellite instability (MSI), owing to MMR deficiency (dMMR) following somatic hits. MSI is associated, in most cases, with a loss of expression of affected protein(s) evidenced by immunohistochemistry (IHC) [1].

Germline inactivation of MMR genes can be caused by different mechanisms. While protein-truncating variants lead clearly to loss of function including nonsense, small indels, large genomic rearrangements and splice site variants affecting the highly conserved canonical splice sites, biological consequence of exonic or intronic single nucleotide variants are often difficult to determine. Many of them remain as variants of uncertain significance (VUS). However, a number of such variants can have an impact on mRNA splicing. Actually, a wide range of computational tools have been developed, aiming to predict whether such exonic or intronic variants can affect splicing. These in silico tools are being routinely used in variant interpretation. Nevertheless, the use of computational prediction can still be challenging face to an increasing number of similar tools which give, sometimes, discordant predictions. More importantly, predicted effects do not always reflect real biological behaviors in vivo, even for some canonic splicing sites. As such, it is highly recommended to confirm predicted effects by functional testing, either from mRNA to detect altered transcripts or based on well-established in vitro assays like “mini-gene” system [2]. Indeed, functional testing provides strong evidence into biological interpretation of spliceogenic variants. Correspondingly, integration of functional data into currently used American College of Medical Genetics and Genomics/Association for Molecular Pathology (ACMG/AMP) guidelines [3] has recently been suggested as a refined recommendation by the ClinGen Sequence Variant Interpretation Splicing Subgroup [4].

The aim of the present study was to evaluate, by RT-PCR based RNA analysis, the spliceogenicity of 24 MMR gene variants predicted to affect splicing, and to apply the refined ACMG/AMP criteria to determine their pathogenicity.

Results

The spliceogenecity of variants detected in MMR genes was routinely evaluated using algorithms integrated in Alamut Visual Plus software including SSF, MES, NNSPLICE, GeneSplicer, and complemented by SPiP and SpliceAI. Variants with positive splicing prediction by at least 1 algorithm, and those affecting the first and the last nucleotide of their exons were considered as potential spliceogenic variants. They were subject to further confirmation by targeted mRNA analyses when RNA samples could be obtained. Here we reported the analysis of 24 variants. The Table 1 showed computational prediction for their spliceogenecity (Table 1).

Table 1

Potential spliceogenic variants selected for RNA analysis.

Gene	Variant	Alamut				SPiP		SpliceAI ∆ score
		SSF	MES	NNSPLICE	GeneSplicer	Interpretation	Risk [CI]	AG	AL	DG	DL
MLH1	c.208–12G > A	NSP	NSP	NSP	NSP	Alt	98.41% [91.47%—99.96%]	0.94	0.88	0	0
	c.399A > T	NSP	NSP	NSP	-	Alt	69.33% [61.29%—76.59%]	0	0.01	0	0
	c.589–10 T > A	NSP	NSP	NSP	NSP	Alt	98.41% [91.47%—99.96%]	0.94	0.88	0	0
	c.1559–1G > A	SSP	SSP	SSP	SSP	Alt	98.41% [91.47%—99.96%]	0.01	0.97	0	0
	c.1668–9_1668-4del	93.4- > 84.3	8.2—> 5.2	-	-	Alt	98.41% [91.47%—99.96%]	0.16	0.70	0	0
	c.1989+4_1989+5del	87.1- > 71.6	7.76- > 2.7	-	-	Alt	98.41% [91.47%—99.96%]	0	0	0.76	0.17
	c.2103+2_2103+21del	SSP	SSP	SSP	SSP	Alt	98.41% [91.47%—99.96%]	0	0	0.03	0.90
	c.2103+3A > G	SSP	8.7- > 4.9	0.9- > 2.8	SSP	Alt	98.41% [91.47%—99.96%]	0	0	0	0.42
MSH2	c.942+3A > C	82.5- > 72.6	8.6- > 2.4	-	-	Alt	98.41% [91.47%—99.96%]	0	0	0	0.68
	c.1861C > G	NSP	NSP	-	-	NTR	02.58% [00.84%—05.91%]	0.25	0	0	0
	c.2006–2A > G	SSP	SSP	SSP	SSP	Alt	98.41% [91.47%—99.96%]	0	0.99	0	0
MSH6	c.261–3C > T	79.3- > 73.1	5—> 3.8	0.7—> 0.8	-	Alt	98.11% [94.59%—99.61%]	0.01	0.01	0	0
	c.3173-2A > C	SSP	SSP	SSP	SSP	Alt	98.41% [91.47%—99.96%]	0.15	0.98	0	0
	c.3417C > T	NSP	NSP	NSP	-	CSP	98.41% [91.47%—99.96%]	0	0	0.99	0.66
	c.3439–1G > T	SSP	SSP	SSP	SSP	Alt	98.41% [91.47%—99.96%]	0.31	1	0	0
	c.4002–8A > G	NSP	NSP	NSP	NSP	NSP	98.54% [94.83%—99.82%]	0.93	0.91	0	0
PMS2	c.24–12 T > A	NSP	NSP	NSP	NSP	Alt	98.41% [91.47%—99.96%]	0.99	0.97	0	0
	c.163+1G > T	SSP	SSP	SSP	SSP	Alt	98.41% [91.47%—99.96%]	0	0	0	0.97
	c.164–1G > T	CSP	CSP	-	-	Alt	98.41% [91.47%—99.96%]	0.98	0.99	0	0
	c.353G > A*	SSP	SSP	SSP	SSP	Alt	98.41% [91.47%—99.96%]	0	0	0	0.84
	c.354C > T*	-	-	-	-	NTR	09.63% [05.81%—14.79%]	0	0	0	0
	c.353+4A > G	SSP	SSP	SSP	-	Alt	98.41% [91.47%—99.96%]	0	0	0.25	0.38
	c.803+5G > A	SSP	SSP	SSP	SSP	Alt	98.41% [91.47%—99.96%]	0	0	0.01	0.98
	c.2445G > A*	87- > 74,8	11.8- > 9.1	1.0- > 0.9	8.5- > 4.7	Alt	47.89% [39.44%—56.42%]	0	0	0.1	0

Note. NSP: novel splice site; SSP: suppression of splice site; CSP: Cryptic splicing site activation; Alt: alteration of consensus splicing site; NTR: nothing to report; -: no predicted splicing effect; ^*: first and last nucleotide of exons. Positive prediction values from splice AI software are in bold.

Clinicopathological features, mRNA splicing analysis and subsequent variant classification are summarized in the Table 2. For each assay, at least three normal controls were used to exclude common alternative transcripts. Lymphoblastoid cell lines were available only for patients carrying one of the following variants: MLH1 c.1559-1G > A, MLH1 c.1668–9_1668-4del, MSH6 c.3439-1G > T, PMS2 c. 164-1G > A. It was therefore possible to inhibit nonsense-mediated mRNA degradation (NMD) by puromycin treatment. For other patients, RNAs was extracted from a PAXgene tube without the possibility of NMD inhibition.

Table 2

Clinicopathological features, splicing outcome and ACMG/AMP clinical classification of 24 MMR variants.

Variant	Proband’s tumorsa/age (years)	Malignanciesa in FDR/age (years)	IHC assay	MSI status	Aberrant transcript	Protein effect	Elements for classification	Clinvar interpretationc	Classification in this study
MLH1 c.208–12G > A	CRC (42)	CRC (44)+SmbC (71), CRC (52)	MLH1/PMS2 lossb	MSI	r.207_208ins208–10_208–1	p.Lys70Tyrfs*12	PVS1_ (RNA)+PM2+PM (dMMR)	nr	P
MLH1 c.399A > T	Br (43)+Br (54)	/	na	na	r.399a > u	p.Gly133=	PM2+BP7_Strong+BP4	VUS	LB
MLH1 c.589–10 T > A	En (45)+CRC (50)	CRC (58), CRC (47), CRC (32)	na	na	r.588_589ins589-8_589–1	p.Gln197Phefs*8	PVS1_ (RNA)+PM2+PP1+PP4	VUS	P
MLH1 c.1559-1G > A  #	SmbC (63)+CRC (76)	/	normal	MSI	r.1559_1667del	p.Leu521Lysfs*34	PVS1_(RNA)+PM2+PS (dMMR)	LP or P	P
MLH1 c.1668-9_1668-4del  #	CRC (19)	CRC (52), CRC (43)	na	na	r.1668_1731del	p. Ser556Argfs*14	PVS1_(RNA)+PM2+PP4	nr	P
MLH1 c.1989+4_1989+5del	CRC (37)+CRC (38)	/	MLH1/PMS2 lossb	MSI	r.1897_1989del	p.Glu633_Glu663del	PVS1_(RNA)+PM2+PP (dMMR)	nr	P
MLH1 c.2103+2_2103+21del	CRC (34)+CRC (34)	/	MLH1/PMS2 lossb	MSI	r.1990_2103del	p.Val664_Gln701del	PVS1_(RNA)+PM2+PP (dMMR)	nr	P
MLH1 c.2103+3A > G	CRC (65)	/	MLH1 lossb	MSI	r.1990_2103del	p.Val664_Gln701del	PVS1_(RNA)+PM2+PP (dMMR)	VUS or LP or P	P
MSH2 c.942+3A > C	CRC (41)	Ov (35) + (CRC (49)	MSH2/MSH6 loss	MSI	r.793_942del	p.Val265_Gln314del	PVS1_(RNA)+PM2+PP (dMMR)	VUS or LP	P
MSH2 c.1861C > G	CRC (73)+CRC (73)	CRC (52)d	MSH2/MSH6 loss	na	r.1861c > g	p.Arg621Gly	PM2+PP3+PS3-Moderate +PM (dMMR)	VUS or LP or P	LP
MSH2 c.2006-2A > G	CRC (49)	Sto (52), CRC (43)	MSH2/MSH6 loss	MSI	r.2006_2210del	p.Pro670Leufs*7	PVS1_Strong (RNA)+PM2+PS (dMMR)	LP or P	P
MSH6 c.261-3C > T	Br (44)	/	na	na	no detectable effect	p.=	BP7_Strong(RNA)+PM2	LB or VUS	VUS
MSH6 c.3173-2A > C	Br (44)	/	na	na	r.[3173_3438del, 3173_3346del]	p.[Asp1058Glyfs*17, Asp1058_Ile1115del]	PVS1_(RNA)+PM2+PP (dMMR)	LP or P	P
MSH6 c.3417C > T	CRC (43)	/	normal	MSI	r.3416_3438del	p.Lys1140Trpfs16*	PVS1_(RNA)+PM2+PP (dMMR)	VUS or P	P
MSH6 c.3439-1G > T  #	CRC (47)	/	MSH6 Loss	MSI	r.3439_3556del	p.Ala1147Valfs*9	PVS1_(RNA)+PM2+PS (dMMR)	LP or P	P
MSH6 c.4002-8A > G	Ov (74)	/	MSH6 loss	MSI	r.4001_4002ins4002-7_4002–1	p.Glu1335Phefs*8	PVS1_(RNA)+PM2+PP (dMMR)	VUS	P
PMS2 c.24–12 T > A	Ov (58)	/	na	na	r.23_24ins24–10_24–1	p.Ser8Argfs*4	PVS1_(RNA)+PM2	nr	LP
PMS2 c.163+1G > T	CRC (56)	/	PMS2 loss	na	r.24_163del	p.Ser8Argfs*5	PVS1_(RNA)+PM2+PP (dMMR)	LP	P
PMS2 c.164-1G > A  #	Rb + CALMs (0.8)	/	na	cMSI	r.164_171del	p.Asp55Alafs*2	PVS1_(RNA)+PM2+PP (dMMR)	LP	P
PMS2 c.353G > A	CRC (44)	/	na	MSI	r.251_353del	p.Leu85Metfs*17	PVS1_Strong(RNA)+PM2+PS (dMMR)	VUS or P	P
PMS2 c.354C > T	CRC+polyposis (60)	/	normal	MSS	r.354c > u	p.Ser118=	PM2+BP7_Strong(RNA) + BP4	LB or VUS	LB
PMS2 c.353 + 4A > G	Br (59)	/	na	na	r.[=, 251_353del]	p.[=, Leu85Metfs*17]	PM2	VUS	VUS
PMS2 c.803 + 5G > A	Ov (73)	/	PMS2 loss	na	r.[706_803del, 762_803del]	p.[Leu236Hisfs*30, Gly256_Tyr268del]	PVS1_Strong (RNA)+PM2+PS (dMMR)	LP	P
PMS2 c.2445G > A	CRC (43)	/	normal	MSS	r.2445 g > a	p.Ser815=	PM2+BP7_Strong(RNA)+BP4	B or LB or VUS	LB

Note. cMSI: Constitutional microsatellite instability. FDR: first-degree relatives, MSI: microsatellite instability. MSS: microsatellite stable. dMMR: deficient MMR phenotype in tumor. na: not available; nr: not reported. ^#ARN isolated from lymphoblastoid cell line, B: benin. LB: likely benign, VUS: Variant of uncertain significance, LP: likely pathogenic, P: pathogenic.

^aCALMs: café-au-lait macules. Tumors of Br: breast; CRC: colorectal; En: endometrium; Ov: Ovary; Rb: retinoblastoma; SmbC: small bowel, Sto: stomach.

^bWithout MLH1 promoter methylation.

^cVariant classification reported in ClinVar database (https://www.ncbi.nlm.nih.gov/clinvar/, accessed on 13th September 2023).

^dRelative carrying the MSH2 c.1861C > G variant whose tumor displayed MSH2/MSH6 loss.

Among 24 variants tested, aberrant transcripts were able to be confirmed in 19. No aberrant splicing was detected in 5 cases, including an intronic variant (MSH6 c.261-3C > T), three exonic synonymous variants (MLH1 c.399A > T p.Gly133=, PMS2 c.354C > T p.Ser118=, PMS2 c.2445G > A p.Ser815=) and an exonic missense variant (MSH2 c.1861C > G p.Arg621Gly). For these samples, physiological expression from both wild type and mutant alleles were clearly detected (Supplementary Figs SS1 and SS2). Consequently, a BP7_Strong (RNA) code was attributed to intronic and synonymous variants, leading to a “likely benign” classification for synonymous variants and VUS for MSH6 c.261-3C > T (Table 2). For the variant MSH2 c.1861C > G, although no splicing defect was detected, its pathogenicity was still supported by other elements. Indeed, two siblings of the family developed colorectal cancers displaying loss of expression of MSH2/MSH6 (PM for two dMMR tumors in the same family). The variant is consistently predicted to be deleterious by in silico tools (PP3) and was reported to be functionally deleterious in a massively parallel cell-based functional assay (PS3-Moderate) [5]. Thus, it was classified as likely pathogenic due to the missense effect which need to be confirmed by another functional assay such as methylation tolerance-based [6].

Abnormal transcripts were detected in 19 samples which were individually detailed afterwards. The completion of abnormal splicing effect was estimated by visual inspection-based intensity comparison between aberrant and wild type transcripts, either on electrophoresis gel and/or on sequencing electropherograms (Fig. 1).

Figure 1

RNA analysis of splicing variants. Agarose gel electrophoreses are shown for cases with detectable additional band(s) compared to controls since some small sized alterations were failed to be detected due to limited resolution. RT-PCR product sequencing electropherograms for each case are displayed with altered transcript sequence and the reference sequences indicated below. On the right side, schematic presentation of altered splicing, with red letters indicating variants, orange-colored boxes denoting skipped/partially skipped exons or retained intron sequences, dotted lines denoting altered splice junction, and thick fleshes indicating primer locations/sense. + or −: with or without puromycin treatment. FL: full-length; P: patient; C: control; ^* Unknown band.

MLH1 c.208-12G > A

This variant was identified in a patient diagnosed with a colorectal cancer at the age of 42 whose family history fulfilled the Amsterdam criteria [7]. MSI and loss of MLH1/PMS2 expression with absence of MLH1 promoter hypermethylation were displayed in tumors of three affected members. Sequencing revealed an out-of-frame transcript: r.207_208ins208-10_208-1 p.Lys70Tyrfs^*12, resulting from the retention of a 10-bp intronic sequence downstream of the variant as it generated a de novo splicing acceptor (gg- > ag). The splicing defect appeared to be nearly complete with a similar intensity between mutant and wild type transcripts. This variant was classified as pathogenic with PVS1_(RNA), PM2 and PM (3 dMMR tumors in the same family).

MLH1 c.589-10T > A

This variant was detected in a patient diagnosed with an endometrial cancer and a colon cancer at the age of 45 and 50 with a family history fulfilling Amsterdam criteria. An aberrant transcript was detected showing a 8-bp intronic sequence retention caused by the creation of a de novo splice site (tg- > ag): r.588_589ins589-8_589-1 p.Gln197Phefs^*8. To note, the complete effect was demonstrated in another carrier with the help of an allelic discriminant polymorphism marker (unpublished data from French cancer genetic network). Lower expression intensity in our case was most likely due to NMD. This variant was co-segregated in four affected family members (Supplementary Fig. SS3). Thus, it was classified as pathogenic based on PVS1_(RNA), PM2, PP1 and PP4.

MLH1 c.1559-1G > A

It was identified in a patient diagnosed with a small bowel cancer at 63 and a colon cancer at 76 associated with MSI phenotype. Electrophoresis gel showed an abnormal band of 404 bp with comparable intensity with the wild type from patient’s RNA isolated from puromycin treated cells. Sequencing of mutant clones revealed an out-of-frame exon 14 skipping. The lower band of 340 bp detected by gel electrophoresis of the RT-PCR product from the patient and controls was characterized by Sanger sequencing (data not shown) as a transcript lacking exons 14 and 15, which may be a naturally occurring splicing event. This variant was previously reported in a LS patient with MSI tumor [8]. The variant was classified as pathogenic based on PVS1_(RNA), PM2 and PS (dMMR from two independent families).

MLH1 c.1668-9_1668-4del

This variant was identified in a patient diagnosed with colorectal cancer at the age of 19, associated with a family history fulfilling Amsterdam criteria. In addition to the wild-type band, electrophoresis gel showed two abnormal bands slightly less intense than wild type band from puromycin treated patient RNA. Sequencing of mutant clones of the lower abnormal band (404 bp) revealed an out-of-frame exon 15 skipping. Unfortunately, the upper abnormal band could not be sequenced. However, this variant was previously studied by an in-vitro minigene experiment, showing exon 15 skipping and suggesting a complete effect [9]. The variant was classified as pathogenic based on PVS1_(RNA), PM2 and PP4.

MLH1 c.1989+4_1989+5del

It was detected in a patient diagnosed at the age of 37 and 38 with metachronous colorectal cancers displaying MSI and loss of MLH1/PMS2 expression without promoter hypermethylation. An aberrant transcript corresponding to an in-frame exon 17 skipping (r.1897_1989del, p.Glu633_663del) was detected. An equal intensity with wild type was shown on electrophoresis gel and sequencing peaks. As the exon 17 lies in functionally important MLH1-PMS2 dimerization domain [10, 11], a PVS1 code was attributed despite its in-frame status. Thus, the variant was classified as pathogenic based on PVS1_(RNA), PM2 and PP for clinical and tumor phenotype.

MLH1 c.2103+2_2103+21del

This variant was identified in a patient diagnosed at the age of 34 with bifocal colon cancer displaying MSI and loss of MLH1 expression without promoter hypermethylation. Electrophoresis gel revealed additional band (494 bp) with an equal intensity compared to the wild type. Sequencing revealed an in-frame exon 18 skipping (r.1990_2103del p.Val664_Gln701del), affecting a functional domain of the protein [10]. An additional weak band was observed in controls cDNA (515 bp) corresponding to an in-frame exon 17 skipping (sequence not shown), a naturally occurring splicing event previously described in healthy controls [12, 13]. The variant was classified as pathogenic based on PVS1_(RNA), PM2 and PP for clinical and tumor phenotype.

MLH1 c.2103+3A > G

This variant was identified in a patient diagnosed at the age of 65 with colorectal cancer displaying MSI and loss of MLH1expression without promoter hypermethylation. An aberrant band was detected on electrophoresis gel showing nearly equal intensity with wild type which was revealed by sequencing to be an in-frame exon 18 skipping (r.1990_2103del p.Val664_Gln701del), affecting a functional domain of the protein. Our results confirm those previously published on the study of this variant using an RNA-seq approach [12] The variant was classified as pathogenic based on PVS1_(RNA), PM2 and PP for clinical and tumor phenotype.

MSH2 c.942+3A > C

It was identified in a patient diagnosed at the age of 41 with colorectal cancer displaying MSI and loss of MSH2/MSH6 proteins, associated with a family history fulfilling Amsterdam criteria. An in-frame skipping of the exon 5, a functional domain, was detected with an intensity consistent with a complete effect. This variant was classified as pathogenic with PVS1_(RNA), PM2 and PP for clinical and tumor phenotype.

MSH2 c.2006-2A > G

It was identified in a patient diagnosed at the age of 49 with colorectal cancer displaying MSI and loss of MSH2/MSH6 proteins, associated with a family history fulfilling Amsterdam criteria. This variant was previously reported in a LS-patient whose tumor displayed a loss of MSH2/MSH6 [14]. Sequencing revealed an out-of-frame skipping of the exon 13: r.2006_2210del; p.Pro670Leufs^*7, with an intensity nearly comparable to the wild type. This variant was classified as pathogenic with PVS1_Strong (RNA), PM2 and PS (dMMR in independent families).

MSH6 c.3173-2A > C

This variant was found in a patient diagnosed with a breast cancer at the age of 44 with no family history suggestive of LS. This variant affected a canonical site of the intron 4. Two forms of aberrant transcripts were detected by electrophoresis gel and by sequencing with a comparable intensity with wild type, corresponding respectively to a partial in-frame exon 5 skipping (r.3173_3346del p.Asp1058_Ile1115del) removing a part of the ATPase functional domain of the MSH6 protein [10], and, to a full out-of-frame exon 5 skipping: (r.3173_3438del p.Asp1058Glyfs^*17). This variant was previously reported in a LS patient whose tumor showed isolated loss of MSH6 [15]. It was classified as pathogenic with PVS1_(RNA), PM2, and PP for reported tumor phenotype.

MSH6 c.3417C > T

This synonymous variant was detected in a 43-year patient diagnosed with a colon cancer showing MSI-H with normal MMR protein expression. Sequencing revealed an out-of-frame transcript with a 22-bp deletion at the 3′ of the exon 5: r.3416_3438del, p.Lys1140Trpfs16^*, following the generation of a de novo exonic donor site (GC- > GT). Aberrant splicing effect was complete, evidenced by the absence of mutant allele’s physiological transcript (nucleotide T at c.3417). Reduced intensity of aberrant transcript may be caused by NMD. This variant was classified as pathogenic based on PVS1_(RNA), PM2 and PP for clinical and tumor phenotype.

MSH6 c.3439-1G > T

This variant was detected in a 47-year patient diagnosed with colorectal cancer displaying MSI and isolated loss of MSH6 expression. Sequencing of mutant clones revealed an out-of-frame exon 6 skipping (r.3439_3556del p.Ala1147Valfs^*9). The fact that the electrophoresis gel showed an equal intensity between abnormal and normal bands from puromycin treated RNA suggested a complete splicing effect. This variant was previously reported in two unrelated LS patients whose tumors displayed isolated loss of MSH6 expression [16, 17]. The variant was classified as pathogenic based on PVS1_(RNA), PM2 and PS (consistent tumor dMMR in three independent families).

MSH6 c.4002-8A > G

This variant was identified in a 74-year patient diagnosed with an ovarian cancer showing MSI and isolated loss of MSH6 expression. An out-of-frame transcript was detected corresponding to a 7-bp retention of downstream intronic sequence: r.4001_4002ins4002-7_4002–1, p.Glu1335Phefs^*8 caused by a de novo splicing site (aa—> ag). Higher intensity of its expression compared to wild type indicated a complete effect. It was classified as pathogenic: PVS1_(RNA), PM2 and PP for clinical and tumor phenotype.

PMS2 c.24-12T > A

This variant was found in a patient diagnosed at 58 with an ovary cancer with no familial history suggestive of LS. An aberrant transcript was revealed with the retention of last 10 bp of intron 1 sequence due to the creation of a de novo acceptor site (tg- > ag): r.23_24ins24-10_24-1 p.Ser8Argfs^*4. Complete splicing defect was evidenced by its higher intensity compared to the wild type. This variant was classified as likely pathogenic, based on PVS1_(RNA) and PM2.

PMS2 c.163+1G > T

This variant was identified in a patient who developed at the age 56 a colorectal cancer showing isolated loss of PMS2 expression. An out-of-frame transcript with exon 2 skipping was detected: r.24_163del p.Ser8Argfs^*5 with complete splicing effect since it showed a higher level of expression than wild type which may be explained by potential PCR bias as a result of preferential amplification of shorter fragments. It was classified as pathogenic with PVS1_(RNA), PM2 and PP for clinical and tumor phenotype.

PMS2 c.164-1G > A

This variant was identified in homozygous state in a 10-months-old patient diagnosed with a non-familial and unilateral retinoblastoma with negative germline RB1 screening. However, the presence of café-au-lait macules and a positive constitutional microsatellite instability (cMSI) assay [18] strongly suggested constitutional mismatch repair deficiency (CMMRD) syndrome. A 8-bp deletion of the 5′ of exon 3: r.164_171del, p.Asp55Alafs^*2 was detected which was caused by the activation of a downstream cryptic acceptor site. The absence of wild type transcript was consistent with a complete effect. The variant was classified as pathogenic with PVS1_(RNA), PM2 and PP for clinical and tumor phenotype.

PMS2 c.353G > A

This variant was detected in a patient who developed at 44 a colorectal cancer showing MSI. According to the Sanger electropherogram, an out-of-frame transcript with exon 4 skipping was detected, showing nearly comparable intensity with wild type. This variant was previously reported in an another LS patient whose tumors displayed dMMR phenotype [19]. It was classified as pathogenic with PVS1_Strong (RNA), PS (consistent tumor dMMR in two independent families) and PM2.

PMS2 c.353+4A > G

This variant was detected in a 59-year patient diagnosed with a breast cancer. Family history was unavailable. An out-of-frame of exon 4 skipping was revealed with a weak intensity while the exonic variant showed clearly an equal intensity with wild type. Given the minor impact on mRNA splicing and the absence of clinical phenotype, this variant remained as a variant of unknown significance.

PMS2 c.803+5G > A

This variant was detected in a patient who developed an ovarian cancer with isolated loss of PMS2 expression at the age 73. According to the Sanger electropherogram, two forms of aberrant transcripts were detected with high intensity: one with an in-frame deletion of 3′ part of the exon 7 (c.762_803del p.Gly256_Tyr268del) and the other, with an out-of-frame skipping of the exon 7 (r.706_803del p.Leu236Hisfs^*30). This variant was previously reported in other two unrelated LS patients whose tumors displayed consistent dMMR phenotype [19]. Since two aberrant transcripts were observed, an out-of-frame (PVS1) and the other in-frame leading to the deletion of less than 10% of the protein (PVS1-Moderate), the PVS1-Strong criterion was finally assigned to the overall interpretation of the splicing result. It was classified as pathogenic with PVS1_Strong (RNA), PS (consistent tumor dMMR in three independent families) and PM2.

Discussion

Determining whether an in silico predicted spliceogenic variant lead effectively to an altered mRNA splicing is crucial for its interpretation which represents a high importance for genetic counseling and for clinical follow-up of patients and families. Here we reported mRNA analysis of 24 potential spliceogenic variants in MMR genes. Most of these variants were previously reported in LS patients without mRNA splicing analysis and many were located outside of canonical splice sites for which mRNA study was essential to determine their pathogenicity. We confirmed the presence of aberrant transcripts in 19 of them and assessed their pathogenicity by using recently refined ACMG/AMP criteria which provided recommendations for spliceogenic variant interpretation. The assignment of PVS1_strength code for variants outside canonical splice sites appeared to be particularly beneficial. Indeed, 11 out of 18 pathogenic or likely pathogenic variants involved non-canonical intronic or exonic nucleotides. Their pathogenicity would not have been able to be determined without the attribution of PVS1_strength code even with highly consistent clinical and family feature and/or tumor dMMR phenotype. Indeed, coherent clinical and tumor dMMR phenotype were associated to all likely pathogenic and pathogenic variant carriers except for the variant MSH6 c.3173-2A > C which was detected in a patient with breast cancer at the age of 44 and in her family, only her grand-father had a colorectal cancer at the age of 85. The current lack of LS clinical phenotype in this family could be explained by MSH6 gene-related lower penetrance [20]. In fact, this variant was previously reported in a patient who developed an endometrial cancer with MSI and isolated loss of MSH6 expression [15].

Targeted RNA analysis remains actually the most direct approach to investigate RNA splicing alterations but could be challenging in some situations we encountered. The evaluation of whether the splicing effect was complete was still arbitrary for most of the cases, under the influence of NMD for out-of-frame transcripts since all but three of the RNAs were isolated from PAXgene tubes without NMD-blocking treatment. Still, complete splicing effect could be objectively assessed in 5 cases: disappearance of ‘physiological’ expression of mutant allele (MSH6 c.3417C > T, PMS2 c.164-1G > A) and a higher intensity than wild type in spite of NMD for out-of-frame transcripts (MLH1 c.1989+4_1989+5del; MSH6 c.4002-8A > G; PMS2 c.24-12 T > A). Although RT-PCR may favor the amplification of certain smaller sized mutant transcripts, it was apparently not the reason for these cases since two had larger size with intron retention and one had merely reduced size by 91 bp. Two variants (MSH6 c.3173-2A > C, PMS2 c.803+5G > A) induced, each, two forms of aberrant transcripts including an entire exon skipping and a partial exon skipping caused by the activation of nearby cryptic splicing sites. Of note, similar observations were previously described for variant PMS2 c.803+5G > A [21]. This finding demonstrated the complexity of RNA splicing process which can be altered differently in a dynamic way by the same variant. However, for these two cases, there was no ambiguity regarding to pathogenicity determination as both forms led to damaging consequences.

In silico spliceogenecity prediction was essential for routine diagnostic laboratories as it is not always possible to verify splicing effect by RNA analysis or other functional testing. It is particularly important for intronic and exonic variants outside canonical splicing sites. Our findings showed that, overall, all used algorithms provided consistent positive prediction despite some minor discordances. SPiP and SpliceAI are more recently developed tools with the advantage of online open access. For our variants, positive prediction with high scores was given by both software to all variants generating aberrant transcripts. For negative prediction, SPiP appeared less confident with the lack of a referable ‘negative threshold’ since low scores could reflect a ‘gray area’ for complex alterations [22]. Indeed, 47.89% (CI 39.44–56.42), 69.3% (CI 69.29–76.59) and 98.11% (CI 94.59%–99.61%) of risk were predicted for three variants in which aberrant transcripts were failed to be detected and for which negative prediction was given by SpliceAI (PMS2 c.2445G > A; MLH1 c.399A > T; MSH6 c.261-3C > T). Regarding SpliceAI, a borderline positive score (0.25) was given to a negative case (MSH2 c.1861C > G). Taken together, our results suggest that both algorithms showed comparable sensitivities while SpliceAI showed a better specificity. Certainly, this needs to be confirmed in larger series.

Limitations about this study included the lack of objective measurement for complete splicing effect partly due to the use of RNAs without NMD-blocking. Another limitation of our RT-PCR approach, is the difficulty to consider potential PCR bias as a result of preferential amplification of shorter fragments [23]. Certainly, it would be preferable to quantify aberrant transcripts by additional methods such as minigene experiments or capture RNA sequencing [24]. Unfortunately, such tests can hardly be accessible for a clinical diagnosis laboratory. Furthermore, clinical and tumor phenotype was not available for all samples tested. More generally, limitations included also the fact that this approach evaluated only a defined region of a transcript supposed to be involved, with the lack of an overall view on full-length transcript, in addition to the technical complexity with the need of optimizing experimental conditions for each variant to test.

Conclusions

In summary, we reported here targeted RNA analysis of 24 variants with potential spliceogenecity and confirmed splicing defect in 19 of them. This study allowed us to determine their pathogenicity with the use of refined ACMG/AMP criteria and to offer appropriate clinical surveillance for the patients and their families. Online in silico tools SPiP and SpliceAI showed good correlation with RNA results despite inconsistence for a few cases. Using both tools should increase the reliability for prediction. Nevertheless, RNA splicing analysis remains essential for pathogenicity determination. RT-PCR combined with Sanger sequencing is currently the most direct approach but may be limited by time-consuming experimental condition setup, complex splicing alterations and restricted analyses on small related fragments. Recently, whole mRNA analysis with RNAseq approach has been applied in routine practice which will certainly be of great help for spliceogenic variant’s assessment.

Materials and methods

Patients and spliceogenic variants

Patients suspected for cancer syndromes were identified through genetic consultation sessions ensured by clinical geneticists in health care centers throughout France between 2010 and 2023. Germline variant screening was performed using next-generation sequencing (NGS) or Sanger sequencing after informed written consent was obtained. MSI and immunostaining data were provided when available. Variants selection for RNA analysis was based on a positive prediction of splicing alteration, RNA sample availability and lack of reported RNA studies. Splicing effect prediction was based on Alamut Visual plus version 1.7.1 (SOPHiA GENETICS, Saint-Sulpice, Switzerland) including four prediction methods: SpliceSiteFinder-like (SSF), MaxEntScan (MES), NNSPLICE, and GeneSplicer, complemented by SPiP (https://sourceforge.net/projects/splicing-prediction-pipeline/) [22] and SpliceAI (https://spliceailookup.broadinstitute.org) [25]. A delta score ≥ 0.2 was considered as spliceogenic, and a max distance of 50 bp was used for SpliceAI predictions.

Transcription analysis

In total, 24 patients carrying potentially spliceogenic variants were selected for RNA analysis. Total RNAs were obtained from PAXgene tube collected peripheral blood without possibility of nonsense mediated mRNA decay (NMD) inhibition except four samples for which RNAs were isolated from EBV-immortalized lymphoblastoid cell lines with puromycin treatment to inhibit NMD (MLH1 c.1559-1G > A, MLH1 c.1668-9_1668-4del, MSH6 c.3439-1G > T, PMS2 c.164-1G > A). cDNA was synthesized using Superscript III First-Strand Synthesis SuperMix, (Invitrogen, Villebon sur Yvette, France) and RT-PCR were carried out by using specific primers (Supplementary Table S1) located in the upstream and downstream exons of each targeted variant using a HotstarTaq™ Master Mix Kit (Qiagen, GmbH, Germany). RT-PCR products were analyzed by electrophoresis followed by sequencing with ABI3730 automated sequencing apparatus using BigDye Terminator (Applied Biosystems, Foster City, CA) according to the manufacturers’ protocol. For variants MLH1 c.1559-1G > A, MLH1 c.1668-9_1668-4del, MLH1 c.2103+3A > G and MSH6 c.3439-1G > T, sequencing was conducted on cloned RT-PCR product as described elsewhere [26] since direct sequencing on RT-PCR products was not conclusive. The Human Genome Variation Society (HGVS) guidelines were used for variant nomenclature. Reference transcript sequences are following: MLH1 (NM_000249.3), MSH2 (NM_000251.2), MSH6 (NM_000179.3) and PMS2 (NM_000535.6).

Variant classification

Variants interpretation and classification were based on recently refined ACMG/AMP criteria proposed by The ClinGen Sequence Variant Interpretation (SVI) Splicing Subgroup [4], with the specification for spliceogenetic variant classification. Besides, tumor dMMR phenotype was considered as clinical evidence following ACMG/AMP-based French oncogenetic network (https://anpgm.fr/recommandations-professionnelles/), i.e. MSI-High and/or loss of MMR protein expression which was consistent with the affected genes was considered as a Pathogenic Supporting evidence (PP) when displayed in one tumor, as Pathogenic Moderate evidence (PM) if displayed in two tumors from the same family and as a Pathogenic Strong evidence (PS) if displayed in tumors from at least two independent families.

Acknowledgements

We thank the patients for their participation in this study. We thank geneticists Drs L Calavas, O Caron, C Colas, V Cusin, F Desseigne, S Fert-Ferrer, C Legrand, F Prieur, P Rochefort, J­C Saurin and genetic counselors for genetic consultations and clinical data collection.

Conflict of interest statement: The authors declare no conflict of interest.

Funding

None declared.

Ethical approval

Written informed consent was obtained for all patients who were tested and diagnosed within the frame of genetic counseling, in accordance with French law for diagnostic genetic testing. Samples were collected in the frame of care, from patients who consented to a research use of their samples. Testing was done in a hospital laboratory approved for genetic molecular diagnosis. The analyses were performed in accordance with French regulations and the principles of the Declaration of Helsinki.