I have this paper:
Single-cell	multi-omics	defines	the	cell-type	specific	impact	of	splicing
aberrations	in	human	hematopoietic	clonal	outgrowths
Federico	Gaiti1,2†,	Paulina	Chamely1,2†,	Allegra	G.	Hawkins1,2†,	Mariela	Cortés-López1,2†,	Ariel	D.	Swett1,2,	Saravanan
Ganesan1,2,	Tarek	H.	Mouhieddine3, Xiaoguang	Dai4,	Lloyd	Kluegel1,2,	Celine	Chen1,2,5,	Kiran	Batta6,	John	Beaulaurier7,
Alexander	W. Drong4,	Scott	Hickey7,	Neville	Dusaj1,2,5,	Gavriel	Mullokandov1,2,	Jiayu	Su1,8,	Ronan	Chaligné1,2,	Sissel	Juul4,
Eoghan	Harrington4,	David	A.	Knowles1,8,9,	Daniel	H.	Wiseman6,	Irene M. Ghobrial3,	Justin	Taylor10,	Omar	AbdelWahab11*,	Dan	A.	Landau1,2,12*
1New	York	Genome	Center, New	York,	NY,	USA.	 2Division	of	Hematology	and	Medical	Oncology,	Department	of	Medicine	and	Meyer	Cancer	Center,	Weill	Cornell	Medicine,	New	York,	NY,	USA. 3Department	of	Medical	Oncology,	Dana-Farber	Cancer	Institute, Boston,	MA,	USA. 4Oxford	Nanopore	Technologies Inc,	New	York,	NY,	USA. 5Tri-Institutional	MD-PhD	Program,	Weill	Cornell	Medicine,	Rockefeller	University,	Memorial	Sloan	Kettering	Cancer	Center, New	York,	NY,
USA.
6Division	of	Cancer	Sciences,	The	University	of	Manchester,	Manchester,	United	Kingdom.
7Oxford	Nanopore	Technologies Inc,	San	Francisco,	CA,	USA. 8Department	of	Systems	Biology,	Columbia	University,	New	York,	NY,	USA. 9Department	of	Computer	Science,	Columbia	University,	New	York,	NY,	USA.	 10Sylvester	Comprehensive	Cancer	Center,	University	of	Miami,	Miller	School	of	Medicine,	Miami,	FL,	USA. 11Human	Oncology	and	Pathogenesis	Program,	Memorial	Sloan	Kettering	Cancer	Center, New	York,	NY,	USA. 12Institute	for	Computational	Biomedicine,	Weill	Cornell	Medicine, New	York,	NY,	USA.
†Contributed	equally	to	this	work;	*Co-corresponding	authors.
Corresponding	authors contact	details: Dan	A.	Landau,	MD,	PhD - Weill	Cornell	Medicine Belfer	Research	Building,	413	East	69th	Street,	New
York,	NY	10021.	Email:	[email protected];	Omar	Abdel-Wahab,	MD	- Memorial	Sloan	Kettering	Cancer	Center,	1275	York	Ave,	New
York,	NY	10065.	Email:	[email protected].
Keywords: Single-cell,	RNA-seq,	multi-omics,	splicing,	long-read	sequencing,	genotyping,	clonal	hematopoiesis,	myelodysplastic	syndrome
ABSTRACT
RNA	splicing	factors	are	recurrently	affected	by	alteration-of-function	mutations	in	clonal	blood	disorders,	highlighting
the	importance	of	splicing	regulation	in	hematopoiesis.	However,	our	understanding	of	the	impact	of	dysregulated	RNA
splicing	has	been hampered	by	the	inability	to	distinguish	mutant	and	wildtype	cells	in	primary	patient	samples,	the	celltype	 complexity	 of	 the	 hematopoietic	 system,	 and	 the	 sparse	 and	 biased	 coverage	 of	 splice	 junctions	 by	 short-read
sequencing	 typically	 used	in	 single-cell	 RNA	 sequencing.	 To	 overcome	 these	limitations,	 we	 developed	 GoT-Splice	 by
integrating	Genotyping	of	Transcriptomes	(GoT)	with	enhanced	efficiency	long-read	single-cell	transcriptome	profiling,
as	well	as	proteogenomics	(with	CITE-seq).	This	allowed	for the	simultaneous	single-cell	profiling	of	gene	expression,	cell
surface	protein	markers,	somatic	mutation	status,	and	RNA	splicing.	We	applied	GoT-Splice	to	bone	marrow	progenitors
from	patients	with	myelodysplastic	syndrome	(MDS)	affected	by	mutations	in the	most	prevalent	mutated	RNA	splicing
factor	– the	core	RNA	splicing	factor	SF3B1.	High-resolution	mapping	of	SF3B1mut	vs.	SF3B1wt hematopoietic	progenitors
revealed	 a fitness	 advantage	 of	 SF3B1mut cells	 in	 the	 megakaryocytic-erythroid	 lineage,	 resulting	 in	 an	 expansion	 of
SF3B1mut erythroid	progenitor	(EP)	cells.	SF3B1mut EP	cells	exhibited	upregulation	of	genes	involved	in	regulation	of	cell
cycle	and	mRNA	translation.	Long-read	single-cell	transcriptomes	revealed	the	previously	reported	increase	of	aberrant
3’	splicing	site	usage	in	SF3B1mut cells.	However,	the	ability	to	profile	splicing	within	individual	cell	populations	uncovered
distinct	 cryptic	 3’	 splice	 site	 usage	 across	 different	 progenitor	 populations,	 as	well	 as	 stage-specific	 aberrant	 splicing
during	erythroid	maturation.	Lastly,	as	splice	factor	mutations	occur	in	clonal	hematopoiesis	(CH)	with	increased	risk	of
neoplastic	transformation,	we	applied	GoT-Splice	to	CH	samples.	These	data	revealed	that	the	erythroid	lineage	bias,	as
well	 as	 cell-type	 specific	 cryptic	 3’	 splice	 site	 usage	in	 SF3B1mut cells,	 precede	 overt	MDS.	 Collectively,	we	 present	 an
expanded	multi-omics	single-cell	toolkit	to	define	the	cell-type	specific	impact	of	somatic	mutations	on	RNA	splicing,	from
the	earliest	phases	of clonal	outgrowths	to	overt	neoplasia,	directly	in	human	samples.
INTRODUCTION
Genetic	diversity	in	the	form	of	clonal	outgrowths	has
been	 ubiquitously	 observed	 across	 normal	 and
malignant	 human	 tissues1–13.	 Likewise,	 phenotypic
diversity	is	a	hallmark	of	both	normal	and	malignant
tissues	in	human	samples,	as	has	been	observed	with
the	 widespread	 application	 of	 single-cell	 RNA
sequencing	 (scRNA-seq)14–20. These	 two	 axes	 of
available under aCC-BY-NC-ND 4.0 International license.
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
bioRxiv preprint doi: https://doi.org/10.1101/2022.06.08.495292; this version posted June 9, 2022. The copyright holder for this preprint (which
2 F.	Gaiti,	P.	Chamely,	A.	G.	Hawkins,	M.	Cortés-López	et	al.	(2022).	BioRxiv
cellular	 diversity	likely	exhibit	 complex	interplay,	as
cell	state	may	affect	the	phenotypic	impact	of	somatic
mutations21.	 Recent	 advances	 in	 single-cell	 multiomics	sequencing	have	allowed	us	to	reconcile	these
two	 aspects	 of	 cellular	 variability	 in	 human
tissues15,22,23,	 and	 link	 genetic	 variation	 and
transcriptional	 cell	 state	 diversity	 in	 somatic
evolution.	 For	 example,	 through	 the	 application	 of
Genotyping	 of	 Transcriptomes	 (GoT)15 technology,
which	 enables	 genotyping	 of	 somatic	 mutations
together	with	high-throughput	droplet-based	scRNAseq,	we	have	previously	demonstrated	that	the	effects
of	 somatic	 mutations	 on	 cellular	 fitness	 in	 blood
myeloproliferative	 disorders	 vary	 as	 a	 function	 of
progenitor	cell	identity15.
Mutations	 in	 genes	 encoding	 RNA	 splicing
factors	 serve	 as	 an	 informative	 example	 of	 the
challenge	of	linking	genotype	to	phenotype	in	complex
human	tissues.	Somatic	change-of-function	mutations
in	RNA	 splicing	 factors	are	 recurrent	in	 hematologic
malignancies24–26,	 highlighting	 the	 importance	 of
dysregulated	 RNA	 splicing	 in	 human	 hematopoietic
disorders.	SF3B1 (splicing	factor	3b	subunit	1),	a	core
component	 of	 the	 spliceosome	 complex,	 is	 a
commonly	mutated	splicing	factor	across	hematologic
malignancies	 and	 solid	 tumors, and	 is	 heavily
implicated	 in	 the	 pathogenesis	 of	 myelodysplastic
syndromes	(MDS)27,28. SF3B1 mutations	also	occur	in
subjects	with	clonal	hematopoiesis	(CH),	where	 they
confer	increased	risk	of	conversion	 to	overt	myeloid
neoplasms	compared	to	other	CH	driver	mutations1,2.
SF3B1 mutations	 result	 in	 incorrect	 branch	 point
recognition	 during	 RNA	 splicing,	 often	leading	 to	 an
increased	 usage	 of	 aberrant	 (or	 cryptic)	 intronproximal	 3’	 splice	 sites	in	 hundreds	 of	genes29.	Such
aberrant	3’	splice	site	recognition	typically	results	in
the	 inclusion	 of	 short	 intronic	 fragments	 in	 spliced
mRNA, which	most	commonly	alters	the	frame	of	the
transcript	 and	 renders it	 a	 substrate	 for	 nonsense
mediated	 mRNA	 decay	 (NMD)30.	 Prior	 work	 has
demonstrated	 that	 through	 mis-splicing, SF3B1
mutations	 lead	 to	 altered	 cell	 metabolism31 and
disruption	 of	 ribosomal	 biogenesis32,	 leading	 to	 the
aberrant	hematopoietic	differentiation	typical	of	MDS.
While	these	are	key	advances	in	our	understanding	of
the	role	of	SF3B1 mutations	in	MDS	development,	the
mechanisms	 through	 which	 mis-splicing	 leads	 to
disrupted	 hematopoietic	 differentiation	 in	 humans
remain	elusive.
To	 date,	 cell	 culture	 systems	 and	 murine
models	 have	 been	 critical	 for	 elucidating	 the	 role	 of
splicing	 factor	 mutations	 in	 disordered
hematopoiesis.	Nonetheless,	 these	methods	may	 not
fully	 recapitulate	 MDS	 development	 in	 the	 human
context.	For	example,	alternatively	spliced	genes	from
murine	 models	 of	 SF3B1mut MDS,	 which	 share	 some
phenotypic	 similarities	 with	 human	 MDS,	 show
limited	 overlap	 with	 those	 identified	 in	 human
samples33.	 The	 study	 of	 splice-altering	 mutations	 in
humans	 has	 been	 further	 hampered	 by	 three
important	 limitations.	 First,	 normal	 wildtype	 (WT)
and	aberrant	mutated	(MUT)	cells	are	often	admixed
without	 discriminating	 cell	 surface	markers	 that	are
required to	 uniquely	 isolate	 MUT	 cells,	 limiting	 the
ability	to	identify	signals	that	can	be	specifically	linked
to	the	SF3B1mut genotype.	This	obstacle	is	magnified	in
the	 context	 of	 CH	 where	 MUT	 cells	 commonly
constitute	a	minority	of	the	hematopoietic	progenitor
population.	Second,	the	hematopoietic	differentiation
process	yields	significant	complexity	of	cell	progenitor
types	that	 further	hinders	the	ability	to	link	mutated
genotypes	with	distinct	cellular	phenotypes.	SF3B1mut
MDS	 is	 indeed	 associated	 with	 a	 specific	 clinicomorphological	 phenotype	 of	 refractory	 anemia	 and
accumulation	 of	 ringed	 sideroblasts28,34,	 strongly
suggesting	that	the	interplay	between	cell	identity	and
SF3B1 mutations	is fundamental	in driving	disrupted
hematopoietic	differentiation.	Third,	scRNA-seq	by	3’
or	 5’	 biased	 short-read	 sequencing	 is	 limited	 in	 its
ability	 to	map	 full-length	 RNA	isoforms	 and	 splicing
aberrations.
To	 overcome	 these	 limitations	 and	 identify
cell-identity-dependent	 mis-splicing	 mediated	 by
SF3B1 mutations,	 we	 developed	 GoT-Splice	 by
integrating	 GoT15 with	 long-read	 single-cell
transcriptome	 profiling	 (with	 Oxford	 Nanopore
Technologies [ONT])	as	well	as	proteogenomics	(with
CITE-seq)35.	 This	 allowed	 for	 the	 simultaneous
profiling	 of	 gene	 expression,	 cell	 surface	 protein
markers,	 somatic	 mutation	 genotyping,	 and	 RNA
splicing	within	the	same	single	cell.	The	application	of
GoT-Splice	to	bone	marrow	progenitor	samples	from
individuals	with	SF3B1-mutated	MDS	and	CH	revealed
that,	 while	 SF3B1 mutations	 arise	 in	 uncommitted
hematopoietic	 stem	 progenitor	 cells	 (HSPCs),	 their
effect	 on	 fitness	 increases	 with	 differentiation	 into
committed	 erythroid	 progenitors	 (EPs),	 in	 line	 with
the	 SF3B1mut-driven	 dyserythropoiesis	 phenotype.
Importantly,	 the	 integration	 of	 GoT	 with	 full-length
isoform	 mapping	 via	 long-read	 sequencing	 showed
that	 SF3B1 mutations	 exert	 cell-type	 specific	 missplicing,	 already	 apparent	in	 CH	long	 before	 disease
onset.
RESULTS
available under aCC-BY-NC-ND 4.0 International license.
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
bioRxiv preprint doi: https://doi.org/10.1101/2022.06.08.495292; this version posted June 9, 2022. The copyright holder for this preprint (which
3 F.	Gaiti,	P.	Chamely,	A.	G.	Hawkins,	M.	Cortés-López	et	al.	(2022).	BioRxiv
GoT	 integrated	 with	 proteogenomics	 reveals
increased	fitness	of SF3B1mut cells	in	the	erythroid
lineage	linked	to	overexpression	of	cell-cycle	and
mRNA	translation	genes
As	we	have	recently	demonstrated	that	the	impact	of
somatic	 mutations	 on	 the	 transcriptome	 varies	 as	 a
function	 of	 underlying	 cell	 identity in
myeloproliferative	neoplasms15,	we	hypothesized	that
an	 interplay	 between	 cell	 identity	 and	 SF3B1
mutations	 may	 drive	 disrupted	 hematopoietic
differentiation	in	MDS.	To	test	this,	we	applied	GoT15
(Fig. 1a)	to	CD34+	bone	marrow	progenitor	cells	from
three	 untreated	 MDS	 patients	 with	 SF3B1 K700E
mutations	(discovery	cohort,	MDS01-03),	as well	as	a
distinct	 cohort	 consisting	 of	 three	 MDS	 patients
undergoing	treatment	(validation	cohort,	MDS04-06)
with	erythropoietin	(EPO)	and/or	granulocyte	colonystimulating	 factor	 (G-CSF;	 Fig.	 1b;	 Supplementary
Table	1).	As	normal	hematopoietic	development	has
been	 extensively	 studied	 using	 flow	 cytometry	 cell
surface	 markers,	 we	 further	 integrated	 GoT	 with
single-cell	 proteogenomics	 (CITE-seq35,36;	Fig.	1a).	A
total	of	24,315	cells	across	the	six	MDS	samples	were
obtained	after	sequencing	and	quality	control	filtering
(Extended	Data	Fig.	1a,	b;	MDS02	was	sequenced	in
two	technical	replicates).	To	chart	the	differentiation
map	of	the	CD34+	progenitor	cells,	we	integrated	the
data	across	the	primary	MDS	samples	(MDS01-03),	as
well	as	the	MDS	validation	samples	(MDS04-06),	and
clustered	based	on	transcriptomic	data	alone,	agnostic
to	 the	 genotyping	 and	 protein	 information	 (Fig.	 1c;
Extended	 Data	 Fig.	 1c,	 d).	 Using	 previously
annotated	 RNA	 identity	 markers	 for	 human	 CD34+
progenitor	cells37,	validated	via	Antibody-Derived	Tag
(ADT)	 markers	 in	 the	 CITE-seq	 panel
(Supplementary	 Table	 2, 3),	 we	 identified	 the
expected	 progenitor	 subtypes	 in	 the	 primary	 MDS
cohort,	along	with	a	population	of	mature	monocytic
cells	 characterized	 by	 CD14	 expression	 and	 lack	 of
CD34	expression	often	observed	in	CD34+	sorting	of
human	bone	marrow38 (Fig.	1c;	Extended	Data	Fig.
2a-c).	Cell	clustering	was	further	validated	using	RNA
and	ADT	multimodal	integration	(Extended	Data	Fig.
2d).	The	expected	progenitor	subtypes were	similarly
identified	 in	 the	 MDS	 validation	 cohort	 (MDS04-06;
Extended	Data	Fig.	3a-c).
Genotyping	 data	 were	 available	 for	 15,650
MDS	 cells	 (64.4%	 across	 MDS01-06)	 through	 GoT
(Fig.	1b; Extended	Data	Fig.	4a-d).	The	per-patient
mutant	 cell	 fractions	 obtained	 through	 GoT	 were
highly	 correlated	 with	 the	 variant	 allele	 frequencies
(VAFs)	obtained	through	bulk	sequencing	of	matched
unsorted	 peripheral	 blood	 mononuclear	 cells
(Pearson’s	r	=	0.81,	P-value	=	0.008;	Extended	Data
Fig.	 4a).	 Projection	 of	 the	 genotyping	 information
onto	the	differentiation	map	demonstrated	MUT	and
WT	 cells	 co-mingled	 throughout	 the	 differentiation
topology	(Extended	Data	Fig.	4c,	d),	highlighting	the
need	for	single-cell	multi-omics	to	link	genotypes	with
cellular	 phenotypes	in	SF3B1mut MDS.	Although	MUT
cells	 were	 found	 across	 CD34+	 progenitor	 cells,	 we
observed	 an	 accumulation	 of	 MUT	 cells	 along	 the
erythroid	 trajectory	 (Fig.	1d),	suggesting	 that	SF3B1
mutant	cell	frequency	(MCF)	varies	as	a	function	of	the
progenitor	subtype.	To	confirm	this,	we	evaluated	the
MCF	 across	 the	 different	 prevalent	 progenitor	 cell
types	(limited	to	progenitor	subsets	with	>	300	cells).
Of	 note,	 as	 cells	 may	 display	 variable expression	 of
SF3B1 itself,	 we	 performed	 amplicon	 UMIdownsampling	 to	 exclude	 sampling	 biases	 given	 the
heterozygosity	 of	 the	 mutated	 allele	 as	 a	 potential
confounder	 for	 observed	 differences	 in	 MCF	 (see
Methods).	Across	samples,	we	observed	a	significant
increase	 in	 MCF	 in	 the	 megakaryocyte-erythroid
lineage	 with	 the	 highest	 MCF	 observed	 in	 EPs
compared	 to	 HSPCs	 (P-value < 10-16; Fig.	 1e;
Extended	Data	Fig.	4e),	consistent	with	the	erythroid
lineage-specific	impact	of	mutated	SF3B139,40.
The	ability	to	layer	protein	measurements	on
top	 of	 GoT	 data	 further	 allowed	 us	 to	 identify
differentially	 expressed	 proteins	 between	 MUT	 and
WT	cells	within	each	progenitor	subset.	After	quality
control	 filtering	 for	 ADT	 markers	 with	 adequate
expression	in	at	least	two	major	progenitor	subtypes
(see	Methods),	protein	expression	was	highest	in	the
expected	 cell	 types,	 and	 correlated	 with	 mRNA
expression,	both	at	the	individual	cell	as	well	as	celltype	level,	comparable	 to	previous	data35 (Extended
Data	 Fig.	 5a,	 b).	 We	 directly	 compared	 protein
expression	between	MUT	and	WT	cells,	accounting	for
sample-to-sample	variability	in	mutated	cells	through
downsampling	 (see	 Methods),	 and	 observed
differential	expression	of	CD38,	CD99,	CD36	and	CD71
in	 at	 least	 one	 progenitor	 cell-type	 (Fig.	 1f;
Supplementary	Table	 4).	 CD38	is	 a	 known	marker
for	 the	 transition	 of	 primitive	 CD34+	 stem	 and
progenitor	 cells	 into	 more	 committed	 precursor
cells37,41,42.	Its	overexpression	in	SF3B1mut is	consistent
with	 the	 observed	 higher	 MCF	 in	 committed
progenitor	 subsets.	 CD99,	 over-expressed	 in	 MUT
immature	 myeloid	 progenitor	 cells	 (IMP)	 cells,	 was
previously	noted	to	be	overexpressed	in	both	AML	and
MDS	 stem	 cells,	 serving	 as	 a	 potential	 therapeutic
target	 of	malignant	 stem	 cells43,44.	 Finally, CD36	 and
CD71,	 erythroid	 lineage	 markers,	 were	 found	 to	 be
over-expressed	 in	 MUT	 EPs	 when	 compared	 to	 WT
available under aCC-BY-NC-ND 4.0 International license.
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
bioRxiv preprint doi: https://doi.org/10.1101/2022.06.08.495292; this version posted June 9, 2022. The copyright holder for this preprint (which
4 F.	Gaiti,	P.	Chamely,	A.	G.	Hawkins,	M.	Cortés-López	et	al.	(2022).	BioRxiv
EPs,	 consistent	 with	 the	 SF3B1mut-driven
dyserythropoiesis	 phenotype.	 We	 further	 leveraged
these	 erythroid	 maturation	 cell	 surface	 protein
markers	 to validate	 pseudo-temporal	 (pseudotime)
ordering	 of	 the	 continuous	 process	 of	 erythroid
maturation45 (Extended	Data	Fig.	5c).	This	analysis
revealed	an	increase	in	MCF	along	erythroid	lineage
maturation	 (Fig. 1g),	 confirming	 that	 SF3B1
Human CD34+ bone marrow
MDS01-03
No. of cells = 15,436
F.
UMAP1
UMAP2
WT (N = 2,498)
MUT (N = 9,996)
E/B/M
HSPC IMP
MkP
NP
MEP
EP
T/B cells
MonoDC
DC
Mono
PreB
G.
A.
C.
UMAP1
UMAP2
0.7
0.8
0.9
1.0
1.1
EP
MEP
MkP
IMP
HSPC
NP
Mono
Normalized mutant cell ratio
(only cell types with >300 cells in MDS01-03)
P < 2.2x10-16
H.
HSP90B1
CALR
SMC1A
LMNA
EIF3A
DYNLL1 SPTA1
CCNE1
DDX17
PRKDC
EIF5A
INCENP
DDX5
RPS29
CTNNB1
PSMD12 SPTB
EIF2S3
INO80D
RAD21
EIF3E
EIF4B
EIF3B
PSMD2 WSB1
DNMT1 TP53 MDM4
GALNT6
P4HB
CCDC42
DLK1
TIMP1
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
−1.0 −0.5 0.0 0.5 1.0
−log10(P−value)
# of genes = 2,000
Down-regulated
in EP SF3B1mut cells
Up-regulated
in EP SF3B1mut cells
log2
(EP SF3B1mut / EP SF3B1wt gene expression)
Translation
Cell cycle
CD71
CD36
CD99
CD38
Antibody-Derived Tag (ADT) markers
0.4 0.6 0.8 1.0 −0.2 −0.1 0.0 0.1
ADT log10FC (MUT vs. WT) ADT expression
* P < 0.05
*
*
**
**
** P < 0.01
NP IMP HSPC MkP MEP EP 1.0
2.0
3.0
0.8
0.85
0.9
0.95
0.0
0.1
0.2
0.3
Megakaryocyte-erythro differentiation
Pseudotime Early Late
EP
MEP
HSPC
Density ADT expression
ADT-CD36
ADT-CD71
IMP
D.
E.
I.
J.
Mutant cell fraction
P = 0.021 P = 0.013 P = 6.6x10-11
HSPC
IMP
MEP
EP
NP
−0.08
−0.06
−0.04
−0.02
HSPC
IMP
MEP
EP
NP
−0.08
−0.07
−0.06
−0.05
−0.04
Cell cycle
Module score
* *
*
Enzymes
Barcoded
Beads
Cells
(incubated with ABs)
1 Generate droplets
Cell
barcode
UMI
Full-length
cDNA
2 Amplify full length
cDNA library
5 Amplify locus of interest
(GoT)
3 Fragment and
prepare library
(Short read + CITE-seq)
4 Assess alternative splicing
in individual cells
(Long-read using ONT)
6 Integrate whole transcriptome with genotype,
protein markers and alternative splicing
Transcriptome +
Protein markers
Alternative Splicing Genotype
AAAAAAAAA AAAAAAAAA
+ AAAAAAAAA
AAAAAAAAA
AAAAAAAAA
AAAAAAAAA
B.
(MDS01-03)
MDS01-03
MDS01-03
SF3B1 Mutant Dataset
(CD34+ sorted hematopoietic progenitors)
Disease Type Patient ID Mutation Chromium 10X GoT Bulk VAF Additional
Cell # Cells Genotyped (%) Mutations (VAF)
Myelodysplastic
Syndrome
MDS01 SF3B1-K700E 757 47.2 0.395 TET2-M5331 (0.42)
DNMT3A-X556 (0.37)
MDS02 SF3B1-K700E 10676 80.9 0.376 None
MDS03 SF3B1-K700E 4003 87.3 0.386 None
Myelodysplastic
Syndrome
(Validation Cohort)
MDS04 SF3B1-K700E 1130 22.0 0.110 ASXL1-H633Qfs*2 (0.03)
IDH-R132G (0.50)
MDS05 SF3B1-K700E 5235 25.7 0.320 TP53-Y234C (0.39)
TYK2-S887W (0.53)
MDS06 SF3B1-K700E 2514 62.0 0.379 TET2-G1172Vfs*3 (0.39)
TET2-R1262P (0.49)
DNMT3A-G707Afs*72 (0.46)
Clonal
Hematopoiesis
CH01 SF3B1-K666N 2998 31.0 0.221 None
CH02 SF3B1-K700E 6009 45.1 0.154 None
MUT
WT
* P < 0.05
MDS01-03
TP53 pathway
Module score
available under aCC-BY-NC-ND 4.0 International license.
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
bioRxiv preprint doi: https://doi.org/10.1101/2022.06.08.495292; this version posted June 9, 2022. The copyright holder for this preprint (which
5 F.	Gaiti,	P.	Chamely,	A.	G.	Hawkins,	M.	Cortés-López	et	al.	(2022).	BioRxiv
mutational	 fitness	increases	with	differentiation into
committed	EPs.
To	 further	 explore	 SF3B1 driven
transcriptional	 dysregulation	 in	 committed	 EPs,	 we
performed	 differential	 gene	 expression	 analysis
between	 SF3B1mut and	 SF3B1wt cells.	 Mutated	 EPs
upregulated	 genes	 encoding	 important	 translation
and	ribosome	biogenesis	 factors	(FDR	<	0.2;	Fig.	1h;
Supplementary	 Table	 5,	 6),	 including	 a	 number	 of
eukaryotic	initiation	 factors	 (e.g., EIF3,	EIF5),	DEADbox	 helicases	 (e.g.,	 DDX5,	 DDX17),	 and	 ribosome
subunits	 (e.g., RPS29).	 This	 dysregulation	 of
translational	 activity,	 or	 ribosomal	 stress	 signal,	 is
evocative	 of	 studies	 showing	 that	 translational
regulation	 is	 critical	 during	 hematopoiesis46–49,	 and
may	lead	to	cell- and	tissue	type–restricted	activation
of	 TP53 signaling	 pathway	 in	 myeloid	 disease50–55.
Specifically,	 cells	 that	 require	 high	 levels	 of protein
synthesis,	such	as	erythroid	progenitors,	may	be	more
sensitive	 to	 changes	 caused	 by	 translational
dysfunction56.	 In	 line	 with	 this	 notion,	 TP53 gene
target	 upregulation	 in	 SF3B1mut cells	 was	 more
prominent	 in	 the	 megakaryocyte-erythroid	 lineage,
with	no	increased	expression	of	TP53-related	genes	in
earlier	 progenitors	 (HSPCs)	 or	 in	 neutrophil
progenitors	(NPs)	compared	to	WT	cells	(Fig.	1i).	Our
results	therefore	establish	a	molecular	phenotype	for
the	 SF3B1 mutation	 in	 human	 bone	 marrow
progenitors,	 potentially	 phenocopying	 translational
dysregulation	 typically	 observed	in	 ribosomopathies
driven	 by	 germline	 deleterious	 mutations	 of
ribosomal	subunits56.
Mutated	EPs	also	upregulated	genes	involved
in	 cell-cycle	 and	 checkpoint	 control	 (FDR	 < 0.2;	Fig.
1h,	j;	Supplementary	Table	5,	6).	 In	 particular,	we
observed	 an	 increase	 in	 expression	 of	 CCNE1,	 a
positive	 regulator	 of	 the	 G1/S	 transition	 of	 the	 cell
cycle57,	 and	 MDM4.	 The	 latter	 gene	 works	 together
with	TP53 during	the	G1/S	checkpoint	of	the	cell	cycle
to	determine	cell	fate	by	regulating	pathways	such	as
DNA	 repair,	 apoptosis,	 and	 senescence58.	 Increased
expression	 of	 MDM4 during	 ribosomal	 stress59
prevents	TP53	degradation	and	blocks	subsequential
inactivation	 of	 p21,	 resulting	 in	 a	 sustained	 cell
proliferation60. Together,	 the	combined	upregulation
of	 TP53,	 CCNE1,	 and	 MDM4 in	 mutated	 EPs	 may
therefore	lead	to	cell	survival	and	accumulation	rather
than	 cell	 death,	 supporting	 the	 finding	 that	 SF3B1
mutations	 impart	 a	 greater	 fitness	 advantage
specifically	in	the	erythroid	lineage.
GoT-Splice	 links	 somatic	 mutations,	 alternative
splicing,	 and	 cellular	 phenotype	 at	 single-cell
resolution
Figure	1.	Increased	fitness	advantage	of	SF3B1mut cells	in	the	megakaryocytic-erythroid	lineage.
(A) Schematic	 of	GoT-Splice	workflow.	The	combination	 of	GoT	with	CITE-seq	and	long-read	 full-length	cDNA	using	Oxford
Nanopore	Technologies	(ONT)	enables	the	simultaneous	profiling	of	protein	and	gene	expression,	somatic	mutation	status,	and
alternative	splicing	at	single-cell	resolution. (B) Summary	of	patient	metadata	and	GoT	data	(after	quality	control)	for	MDS	and
CH	samples	with	SF3B1 mutations.	(C) Uniform	manifold	approximation	and	projection	(UMAP)	of	CD34+	cells	(n	=	15,436	cells)
from	myelodysplastic	syndrome	patient	samples	with	SF3B1 K700E	mutations	(n	=	3	individuals),	overlaid	with	cluster	cell-type
assignments.	 HSPC,	 hematopoietic	 stem	 progenitor	 cells;	 IMP,	 immature	 myeloid	 progenitors;	 MkP,	 megakaryocytic
progenitors;	 MEP,	 megakaryocytic-erythroid	 progenitors;	 EP,	 erythroid	 progenitors;	 NP,	 neutrophil	 progenitors;	 E/B/M,
eosinophil/basophil/mast	progenitor	cells;	T/B	cells;	Mono,	monocyte;	DC,	dendritic	cells;	Pre-B,	precursors	B	cells;	Mono	DC,
monocyte/dendritic	cell	progenitors.	(D) Density	plot	of	SF3B1mut	vs.	SF3B1wt	cells.	Genotyping	information	(MDS01-03)	was
obtained	for	12,494	cells	(80.9	% of	all	cells). (E) Normalized	frequency	of	SF3B1 K700E	MUT	cells	in	progenitor	subsets	with
at	least	300	genotyped	cells.	Bars	show	aggregate	analysis	of	samples	MDS01-03	with	mean	+/- s.e.m.	of	100	downsampling
iterations	to	1	genotyping	UMI	per	cell.	Only	cell	types	with	>300	cells	were	used	in	the	analysis.	P-value	from	likelihood	ratio
test	 of	linear	mixed	model	 with	 or	 without	mutation	 status.	 (F) Differential	 ADT	marker	 expression	 between	 SF3B1mut	and
SF3B1wt	cells.	Red:	higher	expression	in	SF3B1mut cells;	blue:	higher	expression	in	SF3B1wt cells.	Size	of	the	dot	corresponds	to
the	average	expression	of	ADT	marker	across	cells	in	a	given	cell-type.	P-values	determined	through	permutation	testing. (G)
Mutant	 cell	 fraction	 and	 ADT	 expression	 levels	 of	 CD36	 and	 CD71	 as	 a	 function	 of	 pseudotime	 along	 the	 megakaryocyteerythroid	 differentiation	 trajectory	 for	 SF3B1mut	 and	 SF3B1wt	 cells	 in	 MDS01-03.	 Shading	 denotes	 95%	 confidence	 interval.
Histogram	shows	cell	density	of	clusters	included	in	the	analysis,	ordered	by	pseudotime.	P-values	were	calculated	by	Wilcoxon
rank	 sum	 test	 by	 comparing	mutant	 cell	 fraction	 between	 pseudotime	 trajectory	 quartiles.	 (H) Differential	gene	expression
between	SF3B1mut	and	SF3B1wt	EP	cells	in	MDS	samples.	Genes	with	an	absolute	log2(fold change)	>	0.1	and	P-value	<	0.05	were
defined	as	differentially	expressed	(DE).	DE	genes	belonging	to	cell	cycle	(red)	and	translation	(blue)	pathways	(Reactome)	are
highlighted	(BH-FDR	<	0.2). (I)	Expression	(mean	+/- s.e.m.)	of	TP53	pathway	related	genes	(Reactome)	between	SF3B1mut	and
SF3B1wt	cells	in	progenitor	cells	from	MDS01-03	samples.	Red:	module	score	in	SF3B1mut cells;	blue:	module	score	in	SF3B1wt
cells.	P-values	from	likelihood	ratio	test	of	linear	mixed	model	with	or	without	mutation	status. (J) Same	as	(I) for	expression	of
cell	cycle	related	genes	(Reactome)	between	SF3B1mut	and	SF3B1wt cells	in	progenitor	cells	from	MDS01-03	samples.
available under aCC-BY-NC-ND 4.0 International license.
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
bioRxiv preprint doi: https://doi.org/10.1101/2022.06.08.495292; this version posted June 9, 2022. The copyright holder for this preprint (which
6 F.	Gaiti,	P.	Chamely,	A.	G.	Hawkins,	M.	Cortés-López	et	al.	(2022).	BioRxiv
The	 integration	 of	 GoT	 with	 single-cell
proteogenomics	 data	 revealed	 that	 SF3B1 mutations
reshape	 hematopoietic	 differentiation	 and	 mediate
cell-identity-dependent	 transcriptional	 changes	 (Fig.
1).	Given	the	pivotal	role	of	SF3B1 in	mRNA	splicing,
we	next	explored	how	mis-splicing	may	serve	as	a	link
between	genotypes	and	 cellular	 phenotypes.	 Indeed,
SF3B1 mutations	 promote	 recognition	 of	 alternative
branch	 points,	 most	 commonly	 leading	 to	 increased
usage	of	aberrant	3’	splice	sites29.	However,	previous
studies	 in	 primary	 human	 samples	 have	 been
performed	 on	 bulk	 samples	 admixing	 MUT	 and	 WT
cells	 as	 well	 as	 progenitor	 subtypes30,32,61,62.
Conversely,	short-read	sequencing	typically	employed
in	 scRNA-seq	 does	 not	 adequately	 cover	 splice
junctions.	 Recent	 advances	 suggest	 that	 long-read
integration	 into	 scRNA-seq	 may	 overcome	 these
limitations63–67.	We	therefore	integrated	GoT	with	fulllength	ONT	long-read	 sequencing,	allowing	 for	highthroughput,	 single-cell	 integration	 of	 genotype,	 cell
surface	 proteome,	 gene	 expression,	 and	 mRNA
splicing	 information	 (GoT-Splice;	 Fig.	 1a). We	 note
that	 single-cell	 cDNA	 sequencing	with	ONT	 presents
unique	challenges,	as	cDNA	amplification	artifacts	are
still	 productively	 sequenced	 when	 using	 standard
ONT	ligation	chemistry. This	leads	to	a	high	fraction	of
uninformative	reads	in	the	highly	amplified	single-cell
libraries. To	enhance	ONT	efficiency,	we	incorporated
a	 biotin	 enrichment	 step	 using	 on-bead	 PCR	 to
selectively	amplify	full-length	reads	containing	intact
cell	 barcodes	 and	 unique	 molecular	 identifiers64
(UMIs;	Fig.	2a).	This	approach	increased	the	yield	of
full-length	 reads	 from	 50.4	 +/- 2.7	 to	 77.6	 +/- 2.0
(mean	+/- s.e.m.)	percent	of	all	sequenced	reads.	Thus,
GoT-Splice	 delivers	 high-resolution	 single-cell	 fulllength	 transcriptional	 profiles	 that	 are	 comparable
with	short-read	sequencing	(Fig.	2b,	c).	To	accurately
identify	 splice	 junctions	 using	 single-cell	 long-read
sequencing,	we	developed	an	analytical	pipeline	that
leverages	 the	recently	published	SiCeLoRe	pipeline64
(Extended	Data	Fig.	6a).	To	reduce	alignment	noise,
we	generated	a	splice	junction	reference	identified	in
single-cell	SMART-seq2	data	from	human	CD34+	cells
with	 no	 SF3B1 mutation	 (see	 Methods).	 Next,	 we
carried	 out	 intron-centric	 junction	 calling	 which
allows	for	the	independent	measurement	of	splicing	at
both	the	5’	and	3’	ends	of	each	intron.	This	allows	for
an	 unbiased	 assessment	 of	 junctions	 and	 a	 greater
accuracy	in	measuring	the	degree	of	mis-splicing	of	a
particular	transcript	when	compared	to	exon-centric
quantification	approaches68,	which	are	typically	used
for	 cassette	 exon	 usage	 profiling	 and	 rely	 on
predefined	transcript	models	or	splicing	events,	both
of	 which	 may	 be	 inaccurate	 or	 incomplete69,70.	 As
anticipated,	 we	 observed	 a	 4-fold	 increase	 in	 the
number	of	junctions	per	cell	detected	using	full-length
long-read	 sequencing	 over	 short-read,	despite	lower
absolute	 number	 of	UMI/cell	 (Fig.	 2d).	Additionally,
GoT-Splice	 afforded	 greater	 coverage	 uniformity
across	 the	 entire	 transcript,	 compared	 to	 3’-biased
coverage	in	short-read	sequencing	(Fig.	2e).
The	 most	 common	 mis-splicing	 events
observed	in	MDS	SF3B1mut cells	involved	alternative	3’
splice	sites,	accounting	for	57%	of	alternative	splicing
events	 (Fig.	 2f),	 consistent	 with	 prior	 reports29,71.
Notably,	 the	 usage	 of	 such	 alternative	 3’	 splice	 sites
was	not	observed	in	a	CD34+	sample	with	no	SF3B1
mutation	 (Extended	 Data	 Fig.	 6b).	 ONT	 long-read
sequencing	also	allowed	us	to	quantify	the	presence	of
different	 splicing	 events	 across	 the	 same	 mRNA
transcript.	While	only	one	aberrant	3’	splice	site	event
was	 observed	 for	 the	majority	 of	mRNA	transcripts,
we	identified	a	total	of	428	genes	(21.4%	of	the	total
number	of	genes	with	at	least	one	cryptic	3’	splice	site)
with	 more	 than	 one	 aberrant	 3’	 splice	 site	 event.
Interestingly,	 these	cryptic	3’	splicing	events	 tend	 to
appear	in	different	copies	of	the	transcript	(Extended
Data	 Fig.	6c),	 highlighting	 the	 unique	advantages	 of
long-read	sequencing	in	this	context.	Consistent	with
previous	MDS	bulk	sequencing	data72,73,	we	observed
a	 relative	 enrichment	 of	 purines	 upstream	 of	 the
aberrant	3’	splice	site	when	compared	to	the	canonical
3’	splice	site	(Extended	Data	Fig.	6d).
We	next	leveraged	 the	unique	ability	of	GoTSplice	 to	 resolve	 differential	 splice	 junction	 usage
between	SF3B1mut and	SF3B1wt cells	within	 the	same
primary	 human	 sample	 (see	 Methods).	 Of	 the
differentially	mis-spliced	cryptic	3’	splice	sites	(those
0-100bp	from	the	canonical	splice	site)	between	MUT
and	WT	cells,	87%	were	used	more	highly	in	SF3B1mut
cells	 (Fig.	 2f,	 inset),	 aligning	 with	 known
characteristics	of	SF3B1 mutations.	Furthermore,	we
observed	a	high	correlation	between	GoT-Splice	delta
PSI	(dPSI;	percent	spliced	in)	measurements	obtained
by	 comparing	 SF3B1mut and	 SF3B1wt cells,	 and	 dPSI
derived	 from	 bulk	 RNA-sequencing	 of	 CD34+ cells
from	SF3B1mut vs.	SF3B1wt MDS	samples32 for	shared
cryptic	3’	splice	sites	 (Fig.	2g).	 In	line	with	previous
work,	the	majority	of	these	cryptic	3’	splice	sites	were
found	to	be	~15-20	bps	upstream	of	the	canonical	3’
site29 (Fig.	3a;	Extended	Data	Fig.	7a-d).	GoT-Splice
enabled	 the	 visualization	 of	 cryptic	 3’	 splice	 sites	in
SF3B1mut vs.	 SF3B1wt cells,	 highlighting	 the	 striking
increased	 usage	 of	 cryptic	 3’	 splice	 sites	 specific	 to
SF3B1mut (Fig.	3b).	Altogether,	GoT-Splice	extends	the
ability	 to	 connect	 somatic	 mutations	 not	 only	 to
transcriptional	 and	 cell	 surface	 protein	 marker
available under aCC-BY-NC-ND 4.0 International license.
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
bioRxiv preprint doi: https://doi.org/10.1101/2022.06.08.495292; this version posted June 9, 2022. The copyright holder for this preprint (which
7 F.	Gaiti,	P.	Chamely,	A.	G.	Hawkins,	M.	Cortés-López	et	al.	(2022).	BioRxiv
phenotypes,	but	also	to	single-cell	mapping	of	splicing
changes.
GoT-Splice	 shows	 progenitor-specific	 patterns	 in
SF3B1mut- mis-splicing
An	important	advantage	of	GoT-Splice	is	the	ability	to
detect	splicing	changes	at	single-cell	resolution,	which
Figure	 2.	 Simultaneous	 profiling	 of	 gene	 expression,	 cell	 surface	 protein	 markers,	 somatic	 mutation	 status	 and
alternative	splicing	at	single-cell	resolution.
(A)	A	comparison	of	the	percentage	of	ONT	reads	with	either	incorrect	structure	(double	TSO,	no	adaptors,	single	R1	or	single
TSO)	or	correct	structure	(full-length	reads)	both	before	and	after	 the	inclusion	of	a	biotin	enrichment	protocol	step	during
preparation	for	sequencing.	Bars	show	the	aggregate	analysis	of	n = 5	samples	with	mean	+/- s.d.	of	the	percentage	for	each
category.	(B) Scatter	plot	of	the	correlation	between	the	number	of	UMIs/cell	detected	in	long-read	ONT	vs. short-read	Illumina
data	 for	cells	that	were	sequenced	across	both	platforms	 for	sample	MDS05.	(C) Density	plot	of	the	correlation	between	the
number	of	UMIs/gene	detected	in	long-read	ONT	vs.	short-read	Illumina	data	for	sample	MDS05.	(D) Number	of	splice	junctions
captured	in	the	full-length	long-read	ONT	data	compared	to	short-read	sequencing	data,	showing	that	GoT-Splice	allows	for	a
significant	 increase	 in	 the	 number	 of	 junctions	 captured	 per	 cell. (E) GoT-Splice	 provides	 greater	 sequencing	 coverage
uniformity	compared	to	inadequate	coverage	of	short-read	sequencing	over	splice	junctions,	as	exemplified	here	for	the	ERGIC3
gene. (F) Pie	chart	summarizing	the	distribution	of	different	alternative	splicing	events	detected	after	junction	annotation.	Inset:
Pie	chart	summarizing	 the	differences	in	 the	usage	of	cryptic	3’	and	5’	splice	site	events	between	SF3B1mut and	SF3B1wt cells
measured	with	a	dPSI	(SF3B1mut	PSI	- SF3B1wt PSI).	Associated	with	SF3B1mut:	+ve	dPSI;	associated	with	SF3B1wt:	-ve	dPSI. (G)
Comparison	 of	 delta	 percent	 spliced-in	 (dPSI)	 values	 of	 shared	 cryptic	 3’	 splicing	 events	identified	in	 the	MUT vs.	WT cell
comparison	from	GoT-Splice	of	SF3B1mut MDS01-03	samples	and	in	the	SF3B1mut vs.	SF3B1wt bulk	comparison	from	bulk	RNAsequencing	of CD34+	cells	of	MDS	samples	in	Pellagatti	et	al.
32.	Correlation	coefficient	ρ calculated	using	Spearman’s	correlation
and P-value	derived	from	Student's	t-distribution.
available under aCC-BY-NC-ND 4.0 International license.
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
bioRxiv preprint doi: https://doi.org/10.1101/2022.06.08.495292; this version posted June 9, 2022. The copyright holder for this preprint (which
8 F.	Gaiti,	P.	Chamely,	A.	G.	Hawkins,	M.	Cortés-López	et	al.	(2022).	BioRxiv
enables	 the	 comparison	 of	 alternative	 splicing
aberrations	between	MUT	and	WT	cells	within	specific
cell	 subsets	 (Fig.	 3c,	 Supplementary	 Table	 7).	 We
identified	both	shared	and	unique	SF3B1mut cryptic	3’
splice	 site	 events	 across	 progenitor	 subtypes.	 The
usage	of	cryptic	3’	splice	sites	was	highest	along	 the
megakaryocyte-erythroid	 lineage,	 with	 SF3B1mut
MEPs	and	EPs	accounting	for	the	majority	of cell-type
specific	 cryptic	 3’	 splice	 site	 events,	 highlighting	 the
specific	impact	 of	SF3B1 mutations	 on	 the	erythroid
lineage.	These	progenitor	specific	patterns	in	SF3B1mut
mis-splicing	 were	 further	 detected	 in	 the	 validation
cohort	of	MDS	patient	samples	(MDS04-06;	Extended
Data	 Fig.	 7e,	 f).	 In	 both	 MDS	 cohorts,	 progenitor
specific	cryptic	3’	splice	sites	involved	genes	related	to
cell	cycle	(e.g., CENPT)74,	RNA	processing	(e.g., CHTOP,
SF3B175,	SRSF11,	PRPF38A),	erythroid	differentiation
(e.g., CD36,	 FOXRED1,	 GATA134,76,77),	 and	 heme
metabolism	 (e.g., UROD,	 PPOX,	 CIAO1)	 (Fig.	 3c;
Extended	 Data	 Fig.	 7e,	f;	Supplementary	Table	 7,
8).	Many	of	these	genes	and	pathways	have	previously
been	reported	to	be	disrupted	by	alternative	splicing
in	 bulk	 studies	 of	SF3B1mut MDS	 samples32,	 but	 their
cell-type	specificity	was	unknown.	For	instance,	while
the	alternative	splicing	event	in	SF3B1 itself	has	been
suggested	before	as	being	neoplasm-specific,	here	we
narrowed	 down	 its	 erythroid-specific	 pattern.	 This
isoform	– SF3B1ins	– is	predicted	to	affect	splicing	by
impairing	U2	snRNP	assembly75,	likely contributing to
the	 enhanced	 mis-splicing	 dysregulation	 in	 the
megakaryocyte-erythroid	 lineage.	 In	 addition,	 cell
cycle	 plays	 a	 critical	 role	 in	 the	 terminal
differentiation	of	hematopoietic	stem	cells78 and	RNA
processing,	 erythroid	 differentiation,	 and	 heme
metabolism	 pathways	 are	 directly	 linked	 to	 the
regulation	 of	 erythropoiesis79–81.	 To	 further	 validate
cell-type	 specificity	 of	 mis-splicing	 events,	 we
compared	the	genes	with	cryptic	3’	splice	site	events
unique	 to	 MEPs	 and	 EPs	 in	 the	 two	 distinct	 MDS
cohorts	 and	 observed	 significant	 overlap	 of
megakaryocyte-erythroid	 lineage-specific	 aberrantly
spliced	 genes	 between	 the	 discovery	 and	 the
validation MDS	 cohorts	 (P-value =	 0.00029,	 Fisher’s
exact	test,	with	46.8%	of	the	cryptically	spliced	genes
in	MDS	also	aberrantly	spliced	in	the	MDS	validation
cohort).	 In	 contrast,	 no	 significant	 overlap	 was
observed	 when	 comparing	 the	 genes	 with	 cryptic	 3’
splice site	events	unique	to	MEPs	and	EPs	in	the	MDS
discovery	cohort	 to	genes	with	cryptic	3’	splice	sites
unique	 to	 earlier	 progenitor	 cells	 in	 the	 MDS
validation	 cohort	 (1.6%	 overlap;	 P-value =	 0.46,
Fisher’s	 exact	 test;	 Extended	 Data	 Fig.	 7f).	 These
findings reveal	 that	 alternative	 splicing	 is	 cell-type
and	differentiation-stage	dependent27,82–84.
Of	note,	erythropoiesis	occupies	a	continuum
of	 cell	 states	 and	 is	 dependent	 on	 a	 series	 of
transcriptional	changes	that	occur	along	a	continuous
trajectory45. Analyzing	the	SF3B1mut mis-splicing	along
this	continuum	(Fig.	4a)	revealed	that	some	erythroid
differentiation	 and	 heme	 metabolism	 genes	 can	 be
mis-spliced	more	 frequently	at	 the	earliest	 stages	 of
EP	 maturation	 (e.g.,	 UROD and	 FOXRED185),	 while
others	 display	 increased	 mis-splicing	 in	 the	 more
differentiated	EPs	(e.g.,	GYPA and	PPOX).	UROD is	part
of	 the	 heme	 biosynthesis	 pathway	 and	 not	 only	 is
heme	an	important	structural	component	of	erythroid
cells	 but	 it	 also	 plays	 a	 regulatory	 role	 in	 the
differentiation	 of	 erythroid	 precursors86. PPOX
encodes	 for	 an	 enzyme	 involved	 in	 mitochondrial
heme	biosynthesis	and,	as	such,	its	degradation	leads
to	ineffective	erythropoiesis	and	accumulation	of	iron
in	 the	 mitochondria	 typical	 of	 MDS	 with	 ring
sideroblast	clinical	phenotype87.	These	results	provide
evidence	 that	 disruptive	 and	 pathogenic	 SF3B1mutdriven	 mis-splicing	 impacts	 key	 mediators	 of
hemoglobin	synthesis	and	erythroid	differentiation	at
all	stages	of	erythroid	maturation88,89.
We	 further	 noted	 that	 the	 degree	 of	 missplicing	of	a	particular	transcript	(measured	via	PSI)
positively	 correlated	 with	 its expression	 across	 the
erythroid	differentiation	 trajectory in	some	cases.	 In
others,	 mis-splicing	 was	 anti-correlated	 with	 gene
expression,	 often	in	 cryptic	 3’	 splice	 site	events	 that
are	predicted	to	lead	to	transcript	degradation	by	the
NMD	pathway	(Fig.	4b for	representative	examples).
Cryptic	3’	splice	sites result	in	 the	inclusion	of	short
intronic	 fragments	 in mRNA	 and	 often	 introduce	 a
premature	 termination	 codon	 (PTC)90–92. mRNAs
harboring	 an	 NMD-inducing	 PTC	 located	 ≥50	 bps
upstream	of	the	last	exon–exon	junction	are	predicted
to	 undergo	 NMD,	 which	 in	 turn	 prevents	 the
production	 of	 potentially	 aberrant	 proteins.	 In
contrast,	 mRNAs	 harboring	 an	 NMD-neutral	 PTC,
which	 is	 generally	 located	 ≤50	 bps	 upstream	 of	 the
last	 exon–exon	 junction	 or	 in	 the	 last	 exon,	 fail	 to
trigger	NMD	and	produce	dysfunctional	proteins93,94.
We	classified	cryptic	3’	splice	sites	detected	in	the	MDS
samples	 into	 three	 major	 groups:	 (i)	 NMD-inducing
event	 (due	 to	 the	 introduction	 of	 a	 PTC);	 (ii)	 NMDneutral	with	a	frameshift	event;	and	(iii)	NMD-neutral
with	no	 frameshift	event	(Supplementary	Table	7).
In	 accordance	 with	 previous	 reports71,	 of	 the	 421
cryptic	3’	splice	sites	significantly	associated	with	the
SF3B1mut cells,	228	(54%)	of	 these	were	classified	as
NMD-inducing	events	while	the	remaining	193	(46%)
were	NMD-neutral	 (60	events	involving	a	 frameshift
and	 133	 events	 were	 in-frame).	 As	 expected,	 we
available under aCC-BY-NC-ND 4.0 International license.
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
bioRxiv preprint doi: https://doi.org/10.1101/2022.06.08.495292; this version posted June 9, 2022. The copyright holder for this preprint (which
9 F.	Gaiti,	P.	Chamely,	A.	G.	Hawkins,	M.	Cortés-López	et	al.	(2022).	BioRxiv
observed	 a	 significant	 decrease	 in	 the	 expression	 of
genes	harboring	NMD-inducing	events	compared	with
those	harboring	NMD-neutral	events	(P-value =	0.017,
Mann	Whitney	U	test;	Fig.	4c).
Figure	3.	Progenitor	cell-type	specific	mis-splicing	in	SF3B1mut MDS.
(A) Differential	splicing	analysis	between	SF3B1mut	and	SF3B1wt	cells	across	MDS	samples.	Junctions	with	an	absolute	dPSI	>	2
and	 BH-FDR	 adjusted	 P-value	 <	 0.2	 were	 defined	 as	 differentially	 spliced.	 Top: Bars	 showing	 the	 percentage	 of	 genes
differentially	spliced	in	SF3B1mut	and	SF3B1wt	cells	in	the	MDS	and	MDS	validation	cohorts.	Inset:	Expected	peak	in	the	number
of	identified	 cryptic	 3’	 splice	 sites	at	 the	anticipated	 distance	 (15-20 base	 pairs)	 upstream	 of	 the	 canonical	 3’	 splice	 site	in
SF3B1mut	cells. (B) Sashimi	Plot	of	METTL17 intron	junction	with	an	SF3B1mut associated	cryptic	3’	splice	site	showing	RNA-seq
coverage	in	SF3B1mut vs. SF3B1wt cells	within	MDS	samples.	Inset: Expected	marked	increase	in	the	PSI	value	for	the	usage	of	this
cryptic	3’	splice	site	in	SF3B1mut cells. (C) Representation	of	dPSI	values	between	SF3B1mut and	SF3B1wt cells	for	cryptic	3’	splicing
events	identified	in	 the main	 progenitor	 subsets	across	MDS	 samples.	Rows	 correspond	 to	 cryptic	 3’	junctions	 found	 to	 be
differentially	spliced	in	at	least	one	cell-type,	with	P-value	<=	0.05	and	dPSI	>=	2.	Columns	correspond	to	cell-type.	Genes	that
belong	 to	 pathways	 cell	 cycle (purple),	 heme	 metabolism	 (green),	 oxygen	 homeostasis	 (black),	 RNA	 processing	 (red)	 and
erythroid	differentiation	(yellow)	are	highlighted.	The	left	bar	plots	show	the	fraction	of	differentially	spliced	cryptic	3’ splice
sites	per	cell.	Top	bar	plots	quantify	the	total	number	of	cell	types	where	an	event	is	differentially	spliced,	with	the	cell-type
specific	events	located	to	the	right	side	of	the	plot.
available under aCC-BY-NC-ND 4.0 International license.
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
bioRxiv preprint doi: https://doi.org/10.1101/2022.06.08.495292; this version posted June 9, 2022. The copyright holder for this preprint (which
10 F.	Gaiti,	P.	Chamely,	A.	G.	Hawkins,	M.	Cortés-López	et	al.	(2022).	BioRxiv
NMD-inducing	events	affected	genes	including
UROD,	 GYPA,	 FOXRED1 and	 PPOX – key	 genes	 in
erythroid	development.	The	loss	of	 these	 transcripts
via	 NMD95,96 may	 thus	 contribute	 to	 disrupted
terminal	differentiation	of	EPs.	Notable	among	NMDneutral	affected	genes,	we	identified	BAX,	a	member	of
the	 Bcl-2	 gene	 family	 and	 transcriptional	 target	 of
TP53. BAX	 is	 a	 vital	 component	 of	 the	 apoptotic
Figure	4.	SF3B1mut-associated	mis-splicing	changes	along	the	continuum	of	erythropoiesis.
(A) Percent	spliced-in	(PSI)	of	junctions	in	SF3B1mut cells	along	the	hematopoietic	differentiation	trajectory	(HSPCs,	IMPs,	MEPs,
EPs).	Rows	(z-score	normalized)	correspond	to	cryptic	3’	splice	sites;	columns	represent	the	PSI	for	the	usage	of	a	given	cryptic
3’	splice	site	in	each	window	(size	of	3000 SF3B1mut	cells,	sliding	by	300	SF3B1mut	cells).	Only	junctions	found	to	be	differentially
spliced	in	at	least	one	cell-type	with	a	dPSI	>	2	were	used	in	the	analysis.	The	ADT	expression	of	erythroid	lineage	marker	CD71,
along	with	the	fraction	of	cell	types	in each	window,	is	shown.	Rows	are	ordered	according	to	the	peak	in	PSI.	Genes	that	belong
to	 pathways	 cell	 cycle	 (purple),	 heme	 metabolism	 (green),	 oxygen	 homeostasis	 (black),	 RNA	 processing	 (red),	 erythroid
differentiation	(yellow)	and	apoptosis	(blue)	are highlighted. (B) Examples	of	mis-spliced	genes	at	different	stages	of	erythroid
maturation.	Bars	represent	PSI	in	SF3B1mut cells.	Red	lines	represent	ONT	expression	of	the	given	junction	in	SF3B1mut	cells. (C)
Fold change	(log2)	of	gene	expression	between SF3B1mut	and	SF3B1wt	EP	cells	in	NMD-inducing	vs.	NMD-neutral	genes.	(D) Gene
model	of	BAX and	relevant	isoforms.	Characteristic	domains	and	their	location	are	highlighted	in	BAX-ɑ,	the	main	isoform.	The
cryptic	3’	splicing	event	on	the	terminal	exon defines	the	BAX-ω isoform,	characterized	by	the	disruption	of	the	transmembrane
domain	(TM)	as	a	result	of	a	frameshift.
available under aCC-BY-NC-ND 4.0 International license.
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
bioRxiv preprint doi: https://doi.org/10.1101/2022.06.08.495292; this version posted June 9, 2022. The copyright holder for this preprint (which
11 F.	Gaiti,	P.	Chamely,	A.	G.	Hawkins,	M.	Cortés-López	et	al.	(2022).	BioRxiv
cascade	 and	 in	 turn	 plays	 an	 important	 role	 in
balancing	 the	 control	 of	 survival,	 differentiation	and
proliferation	of	EPs	at	later	stages	of	erythropoiesis97
(Fig.	 4a).	 The	 identified	 BAX cryptic	 3’	 splice	 site,
though	 NMD-neutral,	 causes	 a	 frameshift	 in	 the	 last
exon,	 disrupting	 the	 C-terminus	 of	 the	 protein.	 This
BAX isoform,	previously	denoted	as	BAX-ω (Fig.	4d),
has	 been	 shown	 to	 protect	 cells	 from	 apoptotic	 cell
Figure	5.	SF3B1 mutation	promotes	accumulation	of	mutant	cells	along	the	erythroid	lineage	in	clonal	hematopoiesis.
(A) UMAP	of	CD34+	cells	(n	=	9,007	cells)	from	clonal	hematopoiesis	(CH)	samples,	one	with	SF3B1 K700E	mutation	and	one
with	 SF3B1 K666N	 mutation	 (n	 =	 2	 individuals),	 overlaid	 with	 cluster	 cell-type	 assignments.	 HSPC,	 hematopoietic	 stem
progenitor	cells;	IMP,	immature	myeloid	progenitors;	MEP,	megakaryocytic-erythroid	progenitors;	EP,	erythroid	progenitors;
MkP,	 megakaryocytic	 progenitors;	 NP,	 neutrophil	 progenitors;	 E/B/M,	 eosinophil/basophil/mast	 progenitor	 cells;	 Pre-B,
precursors	B	cells. (B) UMAP	of	CD34+	cells	 from	CH	samples	overlaid	with	genotyping	data.	WT,	cells	with	genotype	data
without	SF3B1 mutation;	MUT,	cells	with	genotype	data	with	SF3B1 mutation;	NA,	unassignable	cells	with	no	genotype	data. (C)
UMAP	of	CD34+	cells	from	CH	samples	overlaid	with	pseudotemporal	ordering.	Inset: Pseudotime	in	SF3B1mut	vs.	SF3B1wt	cells
in	the	aggregate	of	CH01-02.	P-value	for	comparison	of	means	from	Wilcoxon	rank	sum	test.	(D) Normalized	ratio	of	mutated
cells	 along	 pseudotime	 quartiles.	 Bars	 show	 aggregate	 analysis	 of	 samples	 CH01-CH02	 with	 mean	 +/- s.e.m.	 of	 100
downsampling	iterations	to	1	genotyping	UMI	per	cell.	Only	cell	types	with	>300	cells	were	used	in	the	analysis.	P-value	from
likelihood	 ratio	 test	 of	 linear	 mixed	 model	 with	 or	 without	 mutation	 status.	 Bottom:	 Fraction	 of	 cell	 types	 within	 each
pseudotime	quartile. (E) Differential	gene	expression	between	SF3B1mut	and	SF3B1wt	HSPC	cells	in	CH	samples.	Genes	with	an
absolute	log2(fold change)	>	0.1	and	P-value	<	0.05	were	defined	as	differentially	expressed	(DE).	DE	genes	belonging	to	the
translation	pathway	(red,	Reactome)	are	highlighted	(BH-FDR	<	0.2). (F) Gene	Set	Enrichment	Analysis	of	DE	genes	in	SF3B1mut
HSPC	cells	across	CH	samples.	Gene	sets	that	overlap	with	SF3B1mut	EP	cells	in	MDS	highlighted	(red). (G) Expression	(mean	+/-
s.e.m.)	of	mRNA	 translation-related	genes	(Reactome)	between	SF3B1mut	and	SF3B1wt	cells	in	progenitor	cells	 from	CH01-02
samples.	P-values	from	likelihood	ratio	test	of	linear	mixed	model	with	or	without	mutation	status.
available under aCC-BY-NC-ND 4.0 International license.
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
bioRxiv preprint doi: https://doi.org/10.1101/2022.06.08.495292; this version posted June 9, 2022. The copyright holder for this preprint (which
12 F.	Gaiti,	P.	Chamely,	A.	G.	Hawkins,	M.	Cortés-López	et	al.	(2022).	BioRxiv
death98,99.	 Interestingly,	 a	 recent	 study	 revealed	 Cterminal	BAX mutations	in	myeloid	clones	that	arise	in
chronic	 lymphocytic	 leukemia	 patients	 upon
prolonged	 exposure	 to	 venetoclax,	 demonstrating	 a
role	 for	 BAX	 c-terminal	 alterations	 in	 conferring	 a
survival	 advantage	 to	 myeloid	 cells	 with	 this	 proapoptotic	 treatment.	 Of	 note,	 early	 clinical
observations	 reported	lower	 response	 to	 venetoclax
in	 SF3B1mut AML100,101,	 consistent	 with	 a	 potential
anti-apoptotic	 effect	 of	 BAX-ω.	 Together,	 these
findings	 suggest	 a	 potential	 mechanism	 underlying
the	 erythroid-dysplasia	 phenotype	 of	 SF3B1mut MDS.
Despite	the	injury	to	translational	machinery	(Fig.	1hi),	SF3B1mut	EPs	may	gain	 some	 degree	 of	 protection
against	cell	death	due	to	the	presence	of	isoform	BAXω,	arising	from	aberrant	splicing.
Accumulation	 of SF3B1mut	 cells	 in	 the	 erythroid
progenitor	population	and	extensive	mis-splicing
in	clonal	hematopoiesis
While	SF3B1 mutations	are	the	most	common	genetic
alterations	in	MDS	 patients,	 they	are	also	associated
with	a	high-risk	of	malignant	transformation	in	clonal
hematopoiesis	 (CH)4–8,102,103.	 However,	 the	 study	 of
SF3B1 mutations	directly in	primary	human	samples
has	been	largely	limited	 to	MDS,	where	confounding
co-occurrence	of	other	genetic	alterations	is	common.
Thus,	CH	presents	a	unique	setting	to	interrogate	the
molecular	 consequences	 of	SF3B1 mutations	in	 nonmalignant	human	hematopoiesis.
We	therefore	isolated	viable	CD34+	cells	from
two	 CH	 samples	 with	 SF3B1 mutations	 (VAFs:	 0.15
and	0.22,	from	CD34+	autologous	grafts	collected	from
patients	 with	 multiple	 myeloma	 in	 remission)	 and
performed	 GoT-Splice.	 A	 total	 of	 9,007	 cells	 across
both	 samples	 passed quality	 filters	 (Extended	 Data
Fig.	8a)	and were	integrated	and	clustered	based	on
transcriptome	 data	 alone,	 agnostic	 to	 genotyping
information	 (Fig.	 5a;	 Extended	 Data	 Fig.	 8b).
Consistent	 with	 clinical	 data	 indicating	 normal
hematopoietic	production,	we	identified	the	expected
progenitor	 subtypes	 using	 previously	 annotated
progenitor	identity	markers	(Fig.	5a;	Extended	Data
Fig.	8c,	d).	Genotyping	data	were	available	for	3,642
cells	 of	 these	 9,007	 cells	 (40.4%)	 through	 GoT
(Extended	 Data	 Fig.	 9a).	 Finally,	 to	 exclude
additional	 genetic	 lesions	 in	 these	 CH	 samples,	 we
performed	copy	number	analysis	with	scRNA-seq	data
and	 identified	 no	 significant	 chromosomal	 gains	 or
losses	(Extended	Data	Fig.	9b).
Projection	of	the	genotyping	information	onto
the	differentiation	map	(Fig.	5b),	showed	no	novel	cell
identities	formed	by	the	SF3B1 mutations,	consistent
with	 the	 fact	 that	 patients	 with	 CH	 exhibit	 no	 overt
peripheral	 blood	 count	 or	 morphological
abnormalities.	However,	a	differentiation	pseudotime
ordering	 analysis	 showed	 that	 SF3B1mut cells	 are
enriched	at	later	pseudotime	points	when	compared
to	SF3B1wt cells	(Fig.	5c;	Extended	Data	Fig.	9c).	To
further	identify	differentiation	biases	in	SF3B1mut CH,
we	evaluated	the	mutated	cell	frequencies	across	the
different	 prevalent	 progenitor	 cell	 types,	 as
performed	 in	 MDS	 (Fig.	 1e).	 Mutated	 cells	 were
enriched	in	more	differentiated	EPs	compared	to	the
earlier	 HSPCs	 (P-value < 0.001,	 linear	 mixed	 model,
Fig.	 5d;	 Extended	 Data	 Fig.	 9d),	 showing	 that
SF3B1mut CH	 cells	 already	 demonstrate	 an	 erythroid
lineage	bias.
To	 further	 identify	 transcriptional
dysregulation	 in	 SF3B1mut HSPCs,	 we	 performed
differential	 gene	 expression	 analysis	 between
mutated	and	wildtype	cells.	We	observed	a	similar	upregulation	 of	genes	involved	in	mRNA	 translation	in
the	SF3B1mut HSPC	in	CH	(Fig.	5e,	f;	Supplementary
Table	 9,	 10),	 a	 pathway	 also	 observed	 to	 be
upregulated	 in	 our	 MDS	 analysis	 (Fig.	 1h).	 In	 CH,
upregulation	of	mRNA	translation	pathway	genes	was
observed	 across	 multiple	 cell	 subtypes	 along
erythroid	 differentiation,	 while	 absent	 in	 NPs	 (Fig.
5g). Thus,	 although	 no	 overt	 blood	 count
abnormalities	 are	 observed	 with	 SF3B1 mutation	 in
CH	individuals,	both	the	erythroid	differentiation	bias
and	 aberrant	 transcriptional	 profiles	 are	 already
apparent	at	this	early	pre-disease	stage.
The	analysis	of	differentially	used	alternative
3’	splice	sites	between	SF3B1mut and	SF3B1wt CH	cells
revealed	 a	 marked	 increase	 in	 cryptic	 3’	 splice	 site
usage	in	SF3B1mut cells,	as	observed	in	MDS	(Fig.	6a).
These	mutant-specific	 cryptic	 3’	 splice	 sites	 affected
genes	 including	 UROD,	 OXAIL,	 SERBP1,	 MED6 and
ERGIC3,	 which	 were	 also	 detected	 to	 be	 cryptically
spliced	 in	 the	 SF3B1mut MDS	 cells.	 Importantly,	 the
lower VAF	associated	with	pre-malignant	CH	samples
highlights	the	necessity	for	GoT-Splice	to	increase	the
detection	 of	 mis-splicing	 events	 occurring	 at	 low
frequencies,	and	that	may	otherwise be	missed	in	bulk
sequencing	 studies	 (Fig.	 6b;	 Extended	 Data	 Fig.
10a).
To	 compare	 mis-spliced	 transcripts	 between
CH	and	MDS,	we	compared	cryptic	3’	splice	sites	with
a	P-value	<	0.05	and	dPSI	of	>=	2	in	at	least	one	celltype	 along	 the	 erythroid	 differentiation	 trajectory
(HSPC,	IMP,	MEP	or	EP)	in	both	CH	and	in	MDS	cohorts
(Supplementary	 Table	 11).	 While	 the	 overall
number	of	significant	cryptic	3’	splice	sites	in	CH	was
lower	than	in	MDS,	we	observed	a	significant	overlap
available under aCC-BY-NC-ND 4.0 International license.
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
bioRxiv preprint doi: https://doi.org/10.1101/2022.06.08.495292; this version posted June 9, 2022. The copyright holder for this preprint (which
13 F.	Gaiti,	P.	Chamely,	A.	G.	Hawkins,	M.	Cortés-López	et	al.	(2022).	BioRxiv
in	shared	cryptic	events	(P-value < 10-16,	Fisher’s	exact
test;	 Fig.	 6c).	 Similarly	 to	 MDS,	 we	 identified	 misspliced	events	specific	to	different	stages	of	erythroid
maturation,	 the	 majority	 of	 which	 overlapped	 with
MDS	cryptic	3’	splice	sites	(Fig.	6d).	Notably,	CH	and
MDS	showed	similar	mis-splicing	dynamics	in	the	BAX
transcript	 along	 the	 erythroid	 differentiation
trajectory	(Fig.	6e).
Figure	6.	SF3B1mut	clonal	hematopoiesis	progenitor	cells	display	cell-type	specific	cryptic	3’	splice	site	usage.
(A) Differential	splicing	analysis	between	SF3B1mut	and	SF3B1wt	cells	across	CH	samples.	Junctions	with	an	absolute	delta	percent
spliced-in	(dPSI)	>	2	and	BH-FDR	adjusted	P-value	<	0.2	were	defined	as	differentially	spliced. (B) Sashimi	Plot	of	ERGIC intron
junction	with	an	SF3B1mut associated	cryptic	3’	splice	site	showing	RNA-seq	coverage	in	SF3B1mut vs. SF3B1wt cells	within	CH
samples,	as	well	as	compared	to	the	CH	samples	when	treated	as	bulk	(pseudobulk	of	all	cells	regardless	of	genotype).	PSI	values
showing	the	expected	marked	increase	in	the	usage	of	this	cryptic	3’	splice	site	in	SF3B1mut cells	alone	when	compared	to	both
SF3B1wt cells	 as	 well	 as	 all	 cells	 (pseudobulk	 of	 sample). (C) Venn	 Diagram	 of	 the	 overlap	 of	 genes	 with	 cryptic	 junctions
significantly	differentially	spliced	in	at	least	one	erythroid	lineage	cell	type	(HSPCs,	IMPs,	MEPs,	EPs)	with	a	dPSI	>	2	between
MDS01-03	and	CH	samples.	P-value	for	the	overlap	from	Fisher’s	Exact	test. (D) Percent	spliced-in	(PSI)	of	junctions	in	SF3B1mut
cells	 along	 the	 hematopoietic	 differentiation	 trajectory	 of	 erythroid	 lineage	 cells.	 Rows	 (z-score	 normalized)	 correspond	 to
cryptic	3’	splice	sites;	columns	represent	the	PSI	for	the	usage	of a	given	cryptic	3’	splice	site	in	each	window	(size	of	600	SF3B1mut
cells,	sliding	by	60	SF3B1mut cells).	Only	junctions	found	to	be	differentially	spliced	in	at	least	one	cell	type	with	a	dPSI	>	2	were
used	in	the	analysis.	Pseudotime	across	each	window	shown.	Rows	are	ordered	according	to	the	peak	in	PSI.	Cryptic	events	also
found	to	be	differentially	spiced	in	MDS	highlighted	(red). (E) Bar	plots	of	the	PSI	values	for	the	usage	of	the	BAX-ω isoform	across
each	window	of	SF3B1mut cells	in	the	MDS,	MDS	validation	and	CH	cohorts	along	the	hematopoietic	differentiation	trajectory	of
erythroid	lineage	cells.	Fraction	of	cell	types	in	each	window	shown	per	cohort	(MDS:	SF3B1mut cells	(n	=	6376)	ordered	by	CD71
expression,	MDS	validation:	SF3B1mut cells	(n	=	987)	ordered	by	pseudotime,	CH:	MUT	cells	(n	=	1021)	ordered	by	pseudotime).
available under aCC-BY-NC-ND 4.0 International license.
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
bioRxiv preprint doi: https://doi.org/10.1101/2022.06.08.495292; this version posted June 9, 2022. The copyright holder for this preprint (which
14 F.	Gaiti,	P.	Chamely,	A.	G.	Hawkins,	M.	Cortés-López	et	al.	(2022).	BioRxiv
DISCUSSION
Here,	we	present	GoT-Splice,	a	single-cell	multi-omics
integration	 that	 enables	 joint	 profiling	 of	 genotype,
gene	 expression,	 protein,	 and	 alternative	 splicing	 all
within	 the	 same	 cell.	GoT,	as	 previously	 described15,
allows	 for	 the	 comparison	 between	 somatically
mutated	and	wildtype	 cells	within	 the	 same	 sample,
for	genotype	to	phenotype	inferences.	Next,	by	further
optimization	 of long-read	 sequencing	 of	 scRNA-seq
libraries64,	 we	 were	 able	 to	 simultaneously	 capture
both	 short	 and	 long-read	 data	 within	 the	 same	 cell,
making	 it	 possible	 to analyze the	 impact	 of	 somatic
mutations	on	transcriptional	and	splicing	phenotypes.
To	date,	few	tools	are	available	to	process	and
analyze	 single-cell	 long-read	 data,	 especially	 for	 the
purpose	 of	 alternative	 splicing.	 To	 address	 existing
analytic	 gaps,	 we	 developed	 a	 long-read	 splicing
analysis	 pipeline	 that	 detects	 and	 quantifies
alternative	 splicing	 events	 within	 single	 cells	 and
highlights	 differential	 junction	 usage	 across	 cell
subpopulations.	For	processing	the	long-read	data, the
pipeline	 integrates	 SiCeLoRe64 to	 error-correct	 cell
barcodes	 and	 UMIs,	 followed	 by	 the	 generation	 of
consensus	reads.	Next,	unlike	other	isoform	detection
methods	 that	 perform	 exon-centric	 junction	 calling
(such	as	SiCeLoRe,	TALON63,	FLAME104),	we	opted	 for
an	 intron-centric	 approach	 followed	 by	 split	 five
prime	and	three	prime	PSI	measurements.	Calculating
the	rate	of	splicing	at	the	5’	and	3’	ends	of	the	intron
improves	the	detection	of	the	true	splicing	rate	of	each
individual	 intron,	 compared	 to	 exon-centric
approaches68. In	 addition,	 our	 pipeline	 detected
differential	 splicing	 patterns	 between	 MUT	 and	 WT
cells,	both	across	entire	samples	and	within	individual
cell	types,	with	sample-aware	permutation	testing	to
integrate	across	samples.	Finally,	the	pipeline	includes
a	functional	annotation	step	that	provides	information
regarding	 the	 translational	 consequences	 of	 the
alternative	spliced	isoforms. Altogether,	our	pipeline
provides	 a	 comprehensive	 toolkit	 to	 process	 and
analyze	differential	splicing	events	in	scRNA-seq	longread	data.
By	 applying	 GoT-Splice	 to	 the	most	 common
splice-altering	 mutation	 (SF3B1),	 we	 interrogated
differentiation	 biases,	 differential	 gene	 expression,
protein	 expression	 and	 splicing	 patterns,	 comparing
SF3B1mut vs.	SF3B1wt cells	co-existing	within	the	same
bone	 marrow.	 Importantly,	 while	 GoT	 revealed	 that
SF3B1mut cells	arise	early	 on	in	 uncommitted	HSPCs,
we	 observed	 a	 differentiation	 bias	 of	 SF3B1mut cells
toward	the	erythroid	progenitor	fate.	This	finding	is	of
particular	 interest	 given	 the	 clinical	 association
between	 SF3B1 mutations	 and	 dysplastic
erythropoiesis.	 Differential	 gene	 and	 protein
expression	 in	 erythroid	 progenitors	 revealed
signatures	 that	 may	 contribute	 to	 this	 observed
differentiation	 bias	 of	 SF3B1mut cells	 toward	 the
erythroid	 fate.	Notably,	 an	increase	in	 cell	 cycle	 and
checkpoint	gene	expression	(TP53,	MDM4 and	CCNE1)
as	 well	 as	 the	 over-expression	 of	 erythroid	 lineage
markers,	CD36	and	CD71,	specifically	in	SF3B1mut EPs,
suggest	a	fitness	advantage	for	SF3B1mut cells	along	the
erythroid	lineage.
CH	samples	likewise	showed	erythroid	biased
differentiation	with	higher	mutated	cell	 frequency	in
committed	 erythroid	 progenitors	 compared	 with
HSPCs.	 This	is	 one	 of	 the	 first	 phenotypic	 studies	 of
clonal	 mosaicism	 in	 human	 samples,	 and	 thus	 the
observation	of	a	somatic	mutation-related	phenotype,
which	aligns	with	the	more	advanced	MDS	phenotype,
is of	 particular	 interest.	 In	 our	 results,	 SF3B1mut CH
cells	 showed	 upregulation	 of	 genes	 in	 pathways
involved	in	translation	and	mRNA	processing,	similar
to	SF3B1mut cells	in	MDS.	This	finding	suggests	that	the
pervasive	 mis-splicing	 observed	 with	 SF3B1 														mutations	 may	 disrupt	 translation,	 reminiscent	 of
ribosomopathies,	 which	 often	 also	 result	 in
dyserythropoiesis105,106.	 Interestingly,	 it	 has	 been
shown	 that	 overexpression	 of	MDM4 prevents	 TP53
degradation	and	leads	to	TP53	complex	sequestration,
which	interferes	with	p21	activation	and	results	in	a
sustained	 cell	 proliferation.	 This	 finding aligns	 with
the	 observed	 upregulation	 of	 TP53 and	 other	 TP53-
related	pathway	genes	in	SF3B1mut EPs	in	MDS.	Thus,
in	addition	to	the	shared	erythroid	differentiation	bias
in	 MDS	 and	 CH,	 aberrant	 transcriptional	 profiles
linked	 to	 a	 dyserythropoiesis	 phenotype	 are	 also
already	apparent	at	the	pre-disease	CH	stage.
Leveraging	 the	 single-cell	 resolution	 of	 GoTSplice and	 differential	 splicing	 analysis	 between
SF3B1mut and	SF3B1wt cells	revealed	cell-type	specific
effects	of	SF3B1 mutations	on	patterns	of	mis-splicing.
First,	 key	 genes	involved	in	 pathways	important	 for
terminal	differentiation	of	hematopoietic	stem	cells	as
well	as	the	regulation	of	erythropoiesis	(namely	RNA
processing,	 erythroid	 differentiation,	 cell	 cycle	 and
heme	 metabolism)	 were	 found	 to	 be	 cryptically
spliced	across	distinct	SF3B1mut progenitor	cell	types,
many of	 which	 were	 previously	 reported	 to	 be
affected	in	bulk	studies	of	SF3B1mut MDS54,72,73.	While
some	cryptic	events	were	neutral	in	their	effect,	many
key	 genes	 important	 for	 erythroid	 differentiation
were	 found	 to	 be	 NMD-inducing	 (e.g., UROD,	 GYPA,
PPOX)	 or	 cause	 a	 frameshift	 event	 that	 may	 affect
protein	structure	and	function	(e.g., BAX)	in	both	the
primary	 and	 validation	MDS	 cohorts.	 Thus,	 our	 data
suggest	 that	 mis-splicing	 of	 erythroid	 specific	 genes
available under aCC-BY-NC-ND 4.0 International license.
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
bioRxiv preprint doi: https://doi.org/10.1101/2022.06.08.495292; this version posted June 9, 2022. The copyright holder for this preprint (which
15 F.	Gaiti,	P.	Chamely,	A.	G.	Hawkins,	M.	Cortés-López	et	al.	(2022).	BioRxiv
and	 pathways,	 together	 with	 the	 dysregulation	 of
apoptotic	 programs,	 may	 ultimately	 lead	 to	 the
accumulation	 of	 SF3B1mut EPs	 that	 fail	 to	 reach
terminal	 differentiation97,	 leading	 to	 the
dyserythropoiesis	 clinical	 phenotype.	 Importantly,
this	 SF3B1mut mis-splicing	 phenotype	 was	 already
evident	in	the	CH	samples,	suggesting	that	the	impact
of	 somatic	 CH	 driver	 mutations	 may	 be	 conserved
from	CH	to	overt	myeloid	neoplasia.
what is it about? what issue it tries to address, why is it important and what innovation it has? lastly, what can we learn from it?
please present your result in markdown.