Article Type : Research Article
Authors : Batchu S, Diaz MJ, Kleinberg G and Lucke-Wold B
Keywords : White matter lesions; RNA sequencing; Ligand binding; Clinical improvements
Multiple
sclerosis represents a debilitating disease. It has many different forms and
warrants further investigation. The purpose of this research paper is to like
at the available RNA sequencing for white matter lesion disorders in the brain.
We performed a comprehensive systematic review to correlate RNA sequence to
white matter lesion accuracy. In particular, we looked at ligand receptor
scoring. The results are highlighted in a series of tables showing key
significant findings. The paper will serve as a catalyst for further scientific
development.
Multiple sclerosis (MS) represents a
disabling autoimmune disease of the central nervous system characterized by
predominantly white matter lesions, or distinct areas of myelin loss and axonal
injury [1]. MS is distinguished from neurodegenerative mimics by primary
demyelination (with oligodendrocyte loss) and marked perivascular infiltration
[2,3]. At the time of writing, best
estimates suggest that north of 2.5 million persons worldwide (400,000+ in
United States alone) live with active MS and prevalence is on the rise [4,5]. MS
disproportionately affects females (3:1) and individuals of African-American
and European descent [4,6]. Owing to its heterogenous presentation, MS is
clinically divided into multiple disease phenotypes (e.g., relapsing-remitting
disease and progressive disease) and lesion types (e.g., active and chronic
lesions), both of which are becoming increasingly subdivided [7–10].
Approximately 85-90% of initial MS diagnoses are relapsing-remitting MS, which
presents clinically as periods of neurological deficit bridged by stretches of
remission [11]. Interestingly, however, a smattering of evidence suggests that
MS disability accumulation is an age-dependent process, independent of disease
phenotype at initial diagnosis (and related relapses) [12,13]. Novel investigation
of the variable MS lesion environment therefore represents a worthwhile
pursuit.
The hallmark lesioning contributes to
variable patterns of gliosis and inflammation [14]. Available science
recommends that non-mixed MS lesions be classified as follows (note: additional
lesion types have been proposed, but these are beyond the focus of the present
study). Active type lesions host significant immune cell filtrates that have
moved into the lesion from the blood [15]. Active lesions are characterized by a
wealth of T- and B-cells, macrophages, and microglia permeating the entire
lesion area [10]. Chronic active type lesions are slow-growing which evidence
greater accumulation of peripheral microglia and/or macrophages (compared to
active type lesions), consistent with greater differentially expressed gene
counts [9,16]. Inactive type lesions observe complete loss of oligodendrocytes
and a hypocellular lesion center which limits macrophage and microglia
populations to the lesion edge, attributing to negligible counts of each
[10,15]. Normal-appearing white matter (NAWM) type lesions are MRI-based
lesions characterized by their comparative indistinguishability in MRI imaging
from normal white matter tissue, save early reduction in T1/T2 ratios of
specific tissues [17,18]. Remyelinating type lesions are a distinct active
lesion subtype featuring thin myelin sheaths, shorter internodal lengths, and
oligodendrocyte recruitment [19]. The deleterious action of specific
ligand-receptor pairs in the context of MS pathophysiology coupled with its
described complexity supports lesion type-centric study of ligand-receptor
interaction [20,21].
Advances in spatial transcriptomic and
single-cell technologies have allowed for high resolution of the molecular
environment of numerous malignancies by understanding gene expression at the
lesion level [22–24]. However, less attention has been focused on multiple
sclerosis. Although bulk transcriptomics may not provide the resolution of
single-cell sequencing, this data can still provide valuable insight into
interactions in the lesion microenvironment. Bulk lesions are composed of
infiltrating immune and stromal components and this characteristic is reflected
in resulting gene expression data [25-28].
The present study uses a novel computational technique that leverages
this property to deconvolute the bulk gene expression data into theoretical
stromal and lesion compartments and estimates the relative strength of
ligand-receptor interactions within the multiple sclerosis lesion types. Since
it is now known that the microenvironment may play a critical role in
pathogenesis, it is crucial to elucidate any interactions amongst these stromal
and lesion compartments to guide future experiments and improve treatment
options. Therefore, the present in silico
exploratory study nominated novel and testable ligand-receptor interactions
between stromal and lesion compartments underlying different multiple sclerosis
molecular subtypes using pre-existing bulk gene expression data.
Data acquisition
RNA-sequencing data for different lesion
types in brain white matter in patients with progressive multiple sclerosis
were used for gene expression deconvolution [16]. The lesion types examined
included chronic active (n = 17), active (n = 16), inactive (n= 14), NAWM (n=
21), and remyelinating lesions (n= 5). White matter (n= 25) was used as normal
controls. The corresponding raw count matric was retrieved from Gene Expression
Omnibus GSE138614. A total of 98 samples that had complete data for
RNA-sequencing were used. Raw count data was converted to TPM space before
downstream analyses.
Lesion purity estimation
The ESTIMATE algorithm was used to estimate lesion purity (defined broadly as the proportion of lesion cells in the tissue sample). The ESTIMATE formula and pipeline uses known mRNA expression signatures of stromal and immune cells to infer lesion purity (Supplementary Figure 1) [29]
Lesion and stroma gene expression
deconvolution
The total mRNA expression for a specific gene in a bulk lesion sample s can be modeled as follows where represents the estimated lesion purity in sample s, represents the mean expression for the gene in the lesion compartment and denotes the mean expression for the gene in the stroma compartment [30]:
Therefore, the stroma and lesion compartment
expression levels were estimated using non-negative least-squares regression,
assuming that these expression levels are constant across the lesion samples.
Bootstrapping was used to derive 95% confidence intervals for the lesion and
stromal point estimates. TPM RNA-sequencing data was log2 transformed
before regression.
Ligand-receptor Interaction scoring
A combined database of 1380 ligand-receptor
pairs, previously curated by Ramilowski et al. and Ghoshdastider et
al. were used to annotate the inferred compartmental gene expression output [30,31].
The top-scoring ligand-receptor
interactions across all lesion subtypes for each signaling direction were
analyzed (Figures 1-5). For lesion-to-lesion signaling, active and chronic
active lesion types were enriched for AGRN interacting with LRP4 and CLCF1
interacting with CNTFR, among several others (Figure 1). Chronic active lesion
types evidenced consistently strong lesion-to-lesion signaling in the set of
top-scoring ligand-receptor pairs. Further, the VIP-VIPR1 ligand-receptor pair
reported high median relative crosstalk scores for lesion-to-lesion signaling
directionality across all lesion types. Unique results include stroma-to-lesion
signaling between UCN2 and CRHR2 in inactive and NAWM lesion types and the
reliance of CGA on stroma-to-lesion signaling to interact with FSHR in active
and remyelinating lesions.
Ligand-receptor pairs with preference for stroma-to-lesion signaling in multiple lesion types include AGT-MAS1, HGF-MET, INHBB-ACVR2A, LTF-LRP11, TNFSF15-TNFRSF25, WNT5A-MCAM, and WNT5A-ROR1 (Figure 2). Ligands EFNA1, INHBB, and WNT5A are implicated in 9 of the 15 top-scoring ligand-receptor interactions, indicating their importance in stroma-to-lesion signaling. In chronic active lesions, the set of top-scoring ligand-receptor pairs again reported consistently high relative crosstalk scores for stromal ligand and lesion receptor interactions. Active and remyelinating lesions were uniquely enriched for PROK2 interacting with PROKR1 via normal-to-normal and stroma-to-stroma signal directionality.
Analysis of the top 15 ligand-receptor pairs by highest median relative crosstalk score for lesion-stroma compartment interactions highlighted cytokine ligands (namely LTA)chemokine receptors (namely CXCR2) (Figure 3). Chronic active lesions displayed consistently relative preference for lesion-stroma signaling, which was not identified in other lesion types. Unique to the selected ligand-receptor pair set, PPBP predominantly interacted with CXCR2 via matched normal-to-normal ligand-receptor signaling. Remyelinating lesions reported distinct enrichment for stromal F2 and BMP3 ligand binding action with lesion receptors.Interaction directionality between top-scoring stromal ligands and stromal receptor pairs were similarly decisive (Figure 4). NAWM lesions showed uniquely predominant enrichment for the entire set of stromal ligand-receptor pairs. Remyelinating lesions enriched only ligand-receptor pairs SPP1-CD44, WNT5A-FZD7, and HGF-CD44 for stroma-to-stroma signaling. CD44 stromal receptors were involved in 3 of the 15 top-scoring ligand-receptor interactions. Interestingly, CD44 stromal receptors interacted primarily with lesion ligands in remyelinating lesions, despite high interaction preference for stromal ligands in all other lesion types.
Few multiple sclerosis lesion types showed
preferential enrichment for ligand-receptor pairs involved in matched normal-to-normal
tissue signaling (Figure 5). Of the selected ligand-receptor pairs, active and
chronic active lesions dedicated 73% (11/15) of compartment interactions to
strictly lesion-to-lesion and normal-to-normal signaling. Inactive, NAWM, and
remyelinating lesions were enriched for white matter PROK1 interacting with
white matter PROKR1. Only remyelinating lesions enriched the ARTN-RET
interaction for stroma-to-stroma signaling.
For each signaling direction,
interesting ligand-receptor interactions were identified within the
interactions with the most crosstalk across lesion subtypes. Of the 15
top-scoring ligand-receptor interactions for lesion-to-lesion signaling, 3
involved SLIT1 or SLIT2 (Figure 1). SLIT proteins are known predominantly for
their interactions with Robo receptors however there is growing evidence of
SLIT proteins engaging in interactions related to tumor cell migration and
metastasis as well as inflammatory cell and leukocyte chemotaxis [32,33]. This has potential implications in multiple
sclerosis as the inflammatory disease is often defined by the axonal damage,
demyelination, and inflammation occurring as a result of leukocyte infiltration
[34].
In stroma-to-lesion signaling,
3 out of the 15 top-scoring interactions between ligands and receptors involved
WNT5A (Figure 2). WNT5A is of particular interest as it has been shown to have
significant signaling alterations in amyotrophic lateral sclerosis (ALS)
patients [35]. Furthermore, there is evidence that WNT5A is upregulated in the
spinal cord dorsal horn of mice with experimental autoimmune encephalomyelitis
(EAE) [36]. EAE is widely used to study neurological complications related to
multiple sclerosis such as demyelination and motor impairments [36]. By better
understanding the role of WNT5A in such neurological complications, treatment
of these complications can be further researched.
For lesion-to-stroma signaling,
4 out of the top-scoring 15 ligand-receptor interactions involve either LTA or
LTB (Figure 3). Both cytokine proteins are known for tumor proliferation
regulation and immune regulations that could have serious implications for
therapeutic treatment of multiple sclerosis [37]. Furthermore, studies show LTB
as a considerable option for cancer therapies due to its abilities to regulate
apoptosis and immune responses which could also be applied to multiple
sclerosis [38].
Interestingly, 2 out of the 15
top-scoring ligand-receptor interactions for stroma-to-stroma signaling involve
the SPP1 protein (Figure 4). Of note, osteopontin (SPP1) has been linked
heavily to multiple sclerosis remissions and relapse in about two thirds of
multiple sclerosis patients [39]. Osteopontin functions as a binding partner to
the integrin primarily responsible for attracting lymphocytes to the brain
causing multiple sclerosis. By inhibiting apoptosis of the integrin,
osteopontin also has serious implications in multiple sclerosis. Interestingly,
studies injecting osteopontin into EAE mice (used as a model for multiple
sclerosis) causes relapse [39]. More research is warranted exploring the
prominence of SPP1 in stroma-to-stroma signaling.
Finally, in normal-to-normal
signaling, 3 out of 15 top-scoring ligand-receptor interactions involve
fibroblast growth factor (FGF) signaling (Figure 5). FGF signaling has been
shown to possibly regulate inflammation and myelination in multiple sclerosis
and is being researched further as a potential therapeutic option for induced
remyelination and decreasing inflammation in EAE as well as multiple sclerosis
[40] (Figure 6).
The above-described findings
provide support for the heterogeneity of the MS plaque environment. Here we
leveraged the composition properties of bulk lesions to estimate the relative
strength of 1380 databased ligand-receptor interactions, using mean gene
expression as a proxy for concentration. Our analysis theorized distinct lesion
and stroma compartments to base ligand-receptor signaling directionality
amongst a 98-sample set of active, chronic active, NWAM, and remyelinating
lesions with matched WM controls. Relative crosstalk scoring of each
compartment revealed highly variable signaling directionality between tested
lesion-receptor pairs and between MS lesion types. Notable results include: the
existence of 10-plus ligand-receptor pairs dedicated (~100%) for 4 of the 5
tested signaling paths in chronic active lesions; the unshared enrichment of
VIP interacting with VIPR1 for lesion-to-lesion signaling in all lesion types;
the importance of EFN family A ligands and Eph receptors in mediating
stroma-to-lesion signaling in chronic active, inactive, and remyelinating MS
lesions. The authors hope the present study serves to inspire future
exploration of ligand-receptor interaction in the context of spatially
heterogeneous disease environments.
The authors declare no
conflicts of interest.