Using human genetics to understand mechanisms in ischemic stroke outcome: from early brain injury to long-term recovery

Abstract

There is critical need to elucidate molecular mechanisms underlying brain injury, repair, and recovery following ischemic stroke—a global health problem with major social and economic impact. Despite five decades of intensive research, there are no widely accepted neuroprotective drugs that mitigate ischemic brain injury, or neuroreparative drugs, or personalized approaches that guide therapies to enhance recovery. We here explore novel reverse translational approaches that will complement traditional forward translational methods in identifying mechanisms relevant to human stroke outcome. Although Genome-Wide Association Studies (GWASs), have yielded over 30 genetic loci that influence ischemic stroke risk, only a few GWASs have been performed for stroke outcome. We discuss important considerations for genetic studies of ischemic stroke outcome—including carefully designed phenotypes that capture injury/recovery mechanisms, anchored in time to stroke onset. We also address recent GWASs that provide insight into mechanisms underlying brain injury and repair. There are several ongoing initiatives exploring genomic associations with novel phenotypes related to stroke outcome. To improve the understanding of the genetic architecture of ischemic stroke outcome, larger studies using standardized phenotypes embedded in standard-of-care measures are needed. Novel techniques beyond GWAS— including exploiting informatics, multi-omics, and novel analytics—promise to uncover genetic and molecular pathways from which drug targets and other new interventions may be identified.

Introduction

Stroke outcome is currently a critical focus of attention in stroke genetics because of the social and economic impact of long-term disability. This has increased the urgency to understand mechanisms involved in injury and repair. Demographics, comorbidities, and genetic factors influence acute and long-term stroke outcome.¹ Genomic strategies, such as Genome Wide Association studies (GWASs) have been useful for finding potential genetic risk factors for complex traits and diseases, and promise to identify new potential therapeutic targets. Despite the high number of genetic risk loci associated with stroke risk (>30 loci)² only few have been associated with stroke outcome.^3–5 Genetic factors related to stroke outcome are likely to be different from those influencing stroke risk, underscoring the need for more genetic studies on stroke outcome.⁶

The number of stroke survivors is increasing due to an aging population and increased survival rates.⁷ Worldwide, stroke remains the second leading cause of death and disability.⁸ Based on the global cost of stroke,⁹ even small improvements in stroke outcome/recovery may result in high impact in terms of health and economic cost reductions.

Despite more than five decades of research on ischemic brain injury mechanisms, there are no widely-accepted neuroprotective drugs in use for acute ischemic stroke. This “failure of forward translation” (from bench to clinical bedside) makes it necessary to reconsider the methods for neuroprotective trials and explore alternative approaches, including “reverse translational approaches” (arising from observations in stroke patients) that genomics and other omics techniques offer. Drugs enhancing longer term recovery after stroke are also lacking. Moreover, the clinical observation of heterogeneity of long term outcomes even in stroke patients with similar initial stroke severities, supports that hitherto unknown mechanisms influence stroke recovery and that genetic factors likely have an impact on recovery mechanisms. This makes genomic approaches attractive, as these studies are fueled by phenotypic and genetic diversity.

In this review, we describe current progress and challenges that face genetic studies of early and long-term ischemic stroke outcome. It is important to consider both stroke outcome which describes the degree of function at specific time points; and stroke recovery which comprises degree of improvement (or deterioration) over time and may better identify dynamic biological processes.¹⁰ We highlight the importance of dynamic phenotypes that may capture mechanisms involved in the injury-repair timeline following stroke. We discuss how genetic studies including GWASs can exploit these phenotypes to uncover genes and pathways that shape stroke outcome, from early brain injury to long term recovery. Finally, we examine novel approaches beyond GWAS for discovering treatments that mitigate brain injury, enhance brain repair, and ultimately improve recovery after ischemic stroke.

Mechanisms underlying Stroke Outcome

Outcome after ischemic stroke is a result of the sum-total of all pathomechanistic, reparative, and adaptive processes that occur after stroke onset. These processes are complex and multi-dimensional: 1) in scale/type, involving mechanisms at molecular and cellular, and ultimately at organ and systems levels; 2) in space/location, with different mechanisms involved in the infarct core, penumbra, peri-infarct, and interconnected regions throughout the brain; and 3) in time: the specific spatiotemporal evolution of these complex processes beginning with the acute ischemic injury, proceeding to subacute secondary injury, and ultimately to chronic brain repair and restorative processes including adaptation. For example, early ischemic brain injury involves macroscopic vascular mechanisms, including the occlusion of a blood vessel, with the possibility of recanalization and salvage of brain tissue. At the tissue level ischemic brain injury involves pathophysiological processes—CBF decline, energy debt, ischemia, that may ultimately be followed by irreversible infarction.¹¹ This is accompanied, at the cellular and molecular level, by excitotoxicity, inflammation, and free radical generation which contribute to neuronal death.¹² In acute and subacute phases during and after ischemia, blood brain barrier breakdown, edema formation, and hemorrhagic transformation, are injury mechanisms (at the macroscopic scale) that can also affect long-term outcome.¹³ Each of these macroscopic processes is associated with distinct and/or overlapping cellular and molecular mechanisms and may be modified by external influences, such as tPA treatment and thrombectomy.

During subacute to chronic phases after stroke, there is a transition from injury to repair and adaptation, bridged by inflammation that might trigger processes involved in brain repair.^{14, 15} Brain repair involves highly coordinated cellular and molecular processes that ultimately lead to salvage of function, manifesting clinically as recovery. These processes include functional restitution or remapping, which at the cellular level involves circuit repair (axonal sprouting, elongation, synaptogenesis, among other processes).^{16, 17} Additional mechanisms such as angiogenesis and neurogenesis have also been implicated. ¹⁸ These repair and restorative mechanisms are greatly influenced by environmental stimuli such as rehabilitative therapies. ^{17, 19}

Ultimately, long-term outcome after stroke depends on all these complex and interactive injury-repair processes. Studies in cellular and animal models of stroke and brain repair are conducive to invasive techniques that permit the examination of mechanisms across scale, space and time, involved in ischemic brain injury²⁰ and brain repair after infarction,²¹ but the validation of these specific mechanism in human stroke has not always been possible. Biological mechanisms that impact long-term outcome are also influenced by several factors, not directly associated with the original ischemic brain injury, such as rehabilitation treatment, reorientation in life (defined as adaption and finding strategies to address stroke consequences in daily life),²² recurrent stroke, and secondary complications like infections, falls, and social isolation.

Endogenous neuroprotective mechanisms during acute ischemic brain injury have been well-studied in cellular and animal models for over five decades.²³ Indeed, it has been estimated that more than 1,000 potential drug targets have been identified.²⁴ Of these, only a minority have been tested in human trials of acute ischemic stroke, and virtually all have been negative.²⁵ This has cast doubt on the validity of the mechanisms and/or animal models with regard to relevance to human pathobiology,^{26, 27} identifying the transition from preclinical studies to clinical trials as a bottleneck in drug development.²⁸

Thus, despite tremendous efforts, traditional forward-translational strategies (benchtop to bedside approaches) have not yielded a widely accepted neuroprotective drug. One reason for this failure may be that trials selected patients less likely to benefit from neuroprotective measures: a focused selection of patients with “intermediate” stroke severity might have resulted in a likelihood of success. However, alternative approaches are also needed to complement traditional forward-translation. Reverse-translational approaches—starting at the bedside using carefully crafted clinical phenotypes in combination with genomic methods can confirm known or identify novel mechanisms underlying human stroke outcome. Understanding the genetic architecture of various outcome phenotypes after ischemic stroke will provide insights into molecular mechanisms across the stroke injury-recovery spectrum (Figure). It should be stressed that forward and reverse translation are not mutually exclusive—instead the use of these approaches in concert is likely to increase research success.

Figure. From Phenotypes to Genotypes to Mechanisms.

Illustrated is a putative work flow using clinical and neuroimaging phenotypes at various times after stroke onset to perform GWAS. The figure only illustrates some examples of many possible phenotypes and measures that might be used to study genetics of outcomes after stroke. Genome-wide associated loci linked to genes for each phenotype provide insights into overlapping mechanisms involved at various phases of the injury-recovery time-line.

The genetic influence on stroke outcome is itself multifaceted. The activation or suppression of genes depends on several factors. Induced gene expression after stroke may have both stimulation and inhibitory effects that vary over time.²⁹ Extra-genomic influences on outcome are also important and discussed in more detail below. For example, epigenenetics—including environmentally induced DNA methylation—can influence long-term gene expression (even intergenerationally). Regulation of downstream processes, reflected in the transcriptome and proteome, can be altered by stroke (dependent on severity and time), and regulated by different cell types and is therefore more difficult to interpret compared to the fixed genome which remains constant throughout life.

Designing Phenotypes to capture Mechanisms Underlying Stroke Outcome

Genetic studies can capture mechanisms influencing recovery after ischemic stroke by using well-defined rich diverse phenotypes/biomarkers that can be quantified (Figure, Table 1). The situation is complicated by that outcome and recovery processes are complex and precise definitions are not well delineated. As discussed above, stroke recovery is the end-result of a series of complex interacting processes that evolve over time; the spatiotemporal evolution of these mechanisms are anchored to stroke onset and location, beginning immediately after stroke and continuing for days, months and possibly even years. These mechanisms may manifest in the evolution of clinical, neuroimaging, and/or fluid biomarker characteristics. For example, the secondary brain injury mechanism of cerebral edema may manifest with clinical deterioration and with midline shift on CT neuroimaging. These characteristics can be measured using clinical scales (e.g. change in NIHSS or Glasgow Coma Scale over time) and quantified by measuring midline shift on serial CT scans; providing quantitative phenotypes for genomic studies to query underlying genetic mechanisms. The ultimate goal is to identify genes, and pathways that influence the phenotypes of interest, thereby implicating mechanisms that are suitable treatment targets and leading to a greater understanding the numerous underlying molecular mechanisms involved in stroke injury and recovery (Figure). Thus, genetic studies can take advantage of macroscopic clinical/neuroimaging phenotypes and span the scale gap (macroscopic to molecular) of stroke outcomes.

Table 1.

Phenotypes that may be useful for studies of stroke outcome genetics.

	Biomarkers	Neurological measurements	Functional status	Social participation
Acute phase/Early phases (hours to days after stroke onset)	Imaging: penumbra, collateral flow, edema, infarct dynamics Blood samples: markers for injury, repair	Composite scales e.g. NIHSS. Specific domains e.g. motor, language, sensory Dynamic changes: dNIHSS	Less useful	Less useful
Subacute phase (days to weeks after stroke onset)	Imaging: edema, hemorrhagic transformation Blood samples: markers for injury, repair, inflammation (see text)	See above	Less useful	Less useful
Later phases (weeks to months after stroke onset)	Imaging: lesion volume, tractography, fMRI, functional connectivity MRI Blood samples: markers for repair, inflammation (see text)	See above Also: cognitive status, fatigue	mRS, Barthel index	Quality of life: EQ-5D

The value of phenotype assessment may vary depending on the time point after stroke onset. For detailed recommendations of clinical phenotype evaluations see.⁹

Time is a critical component in the definition of injury and recovery after stroke. Both neurological deterioration and functional recovery imply changes in neurological/ functional status over time. Thus, dynamic measures that capture the temporal evolution of phenotypes may be informative for genetic studies of stroke outcomes. An example of the different information that is conveyed by static vs. dynamic measures is a study of NIHSS changes during the first 24 hours after ischemic stroke. Neurological deficit, as measured by the NIHSS, is highly dynamic within the first 24 hours after stroke onset, with some patients showing rapid deterioration while others demonstrate rapid improvement.³⁰ These dynamic changes can be measured by differences in serial NIHSS score. For example, ΔNIHSS_24h can be defined as the difference between NIHSS measured at 6h minus that measured at 24 hours (positive ΔNIHSS_24h indicates improvement; negative ΔNIHSS_24h indicates deterioration). NIHSS at 24 hours captures a large portion of the variance of 90 day functional outcome.^{31, 32} However, when this 24 hour NIHSS is broken down into a 6 hour NIHSS and ΔNIHSS_24h, each of these measures independently contributes to 90 day functional outcome. Moreover, the variables that influence each of these measures are also independent of one another. TOAST etiologies have a major influence on static 6 hour NIHSS, but have little influence on ΔNIHSS_24h. On the other hand, early recanalization and hemorrhagic transformation have a major influence on ΔNIHSS, but lesser influence on 6 or 24 hour NIHSS.³⁰ Thus, static vs. dynamic measures of NIHSS capture different mechanisms. In a similar manner, the static measure of 90 day functional outcome may capture different mechanisms compared to the change in functional outcome between hospital discharge and 90 days.^{10, 33} Indeed, stroke recovery is defined by the regaining of function lost after stroke, and thus time is a critical component linked to change in function.

Another consideration in developing informative phenotypes for genetic studies is the use of neurological deficit vs. functional outcome measures. The International Classification of Function and disability (ICF) forms one platform for evaluating stroke outcome. The ICF categories can be divided into (1) neurological/physical deficit; (2) functional ability; and (3) social participation and these constitute different clinical qualities of stroke recovery.³⁴ Clinical trials often rely on functional ability scales (e.g. modified Rankin Scale, Barthel index, Glasgow outcome scale).³⁵ Some changes may be so small that they are not clinically meaningful for the individual patient³⁶ and from patients’ perspective, it could be argued that quality of life and social participation should be more highly valued. However, for genetic studies aiming to discover biological mechanisms involved in injury/recovery, high resolution measures are more likely to be informative. Thus, neurological deficit scales (e.g. Fugl-Meyer test and Action Research Arm Test (ARAT)) show more promise to relate to genetic biology. In addition, quantitative scales are more prone to detect smaller changes over time, and provide greater statistical power.³⁷

Genetic analyses exploiting endophenotypes, defined as intermediary phenotypes (e.g. neuroimaging findings), are an alternative approach to studying stroke outcome. Endophenotypes are more stable (homogeneous) traits useful for understanding the genetic architecture of a more complex, heterogeneous phenotype (e.g. clinical stroke outcome).³⁸ For example, hemorrhagic transformation is an endophenotype that can potentially explain acute and long-term outcome. Thus, genetic factors influencing endophenotypes can in turn be related to the more complex phenotype of stroke outcome (see below).

Neuroimaging are endophenotypes that can capture mechanisms involved in the injury/recovery.³⁹ Objective measures (e.g. infarct volume, edema, hemorrhage, white matter tractography, functional connectivity) brings the endophenotypes closer to the genetic biology, providing greater power to detect genetic associations.³⁷ Timing is important because of dynamic changes in infarct growth may occur during the first 24 hours after stroke onset.⁴⁰ Another example of a dynamic neuroimaging phenotype is cerebral edema⁴¹ where a commonly used clinical measure is “midline shift”. One drawback of this measure is that it occurs late, and has limited dynamic range. Another edema-related quantitative phenotype is change in total CSF volume over time after stroke (ΔCSF).⁴² The majority of the change in CSF volume occur within the first 24 hours, with a wide dynamic range, and the extent of ΔCSF_24h predicts future midline shift and “malignant edema”⁴³ Machine learning algorithms that automate quantification of ΔCSF_24h can rapidly assess large numbers of scans with high accuracy⁴⁴ and be useful for quantifying neuroimaging endophenotypes for future large-scale genomic studies.

Other promising functional neuroimaging approaches include core-penumbra mismatch, measures of collateral flow, and dynamic changes in cerebral blood flow, in the acute phase of stroke. The acute ischemic stroke algorithms used to select thrombectomy candidates rely on many of these functional imaging approaches⁴⁵ — providing rich neuroimaging endophenotypes in large numbers of patients. Other neuroimaging approaches that capture mechanisms of brain repair and reorganization include functional connectivity MRI, MR tractography, and measures of functional remapping.⁴⁶ These promise to be informative endophenotypes for future genetic studies of mechanisms involved in brain injury and repair.

Blood biomarkers can also be endophenotypes/phenotypes and promise to be informative for understanding mechanisms underlying stroke outcomes.⁴⁷ Genetic variations can influence the levels of blood biomarkers.^{48, 49} Areas of interest include markers of inflammation, coagulation, extracellular matrix, endothelial function, and blood brain barrier disruption.⁴⁷

Pheno-/endophenotypes should be derived from routine standard-of-care (SOC) assessments, without increasing the work burden for hospital staff, and standardized across multiple sites to be efficiently acquired across large populations. Recommendations for obtaining large harmonized data sets from prospective outcome studies have recently been suggested by the International Stroke Genetics Consortium (ISGC) Global Alliance for Acute and Long-Term Outcome stroke studies.¹⁰ The selection of specific phenotyping and outcome evaluations should depend on the type of neurological deficits are present in individual subjects with stroke (Braun et al. manuscript under review) as well as whether acute or long-term outcomes are to be examined in a specific study.¹⁰

Using Genetics to explore Mechanisms Underlying Stroke Outcome

Genetic research on ischemic stroke outcome and recovery can be used to answer several research questions, with clinically important treatment implications (Table 2) and to identify potential targets for intervention. Whether that target intervenes to mitigate ischemic brain injury, or stimulates brain plasticity to enhance stroke recovery, the goal is to improve clinically meaningful long-term outcomes after stroke. Past history has shown that the investment required to identify these targets can be costly, and the failure of numerous clinical trials has discouraged investment in some of these targets. However, new approaches such as combining rehabilitation therapy with vagus nerve stimulation may be of benefit.⁵⁰

Table 2.

Examples of treatment implications of genetic findings.

Genetic finding	Treatment implications
Biological pathways involved in stroke recovery	Development of drugs influencing biological pathways
Genes related to drug metabolism	Selection of most appropriate type and dose of drug
Genes related to response to treatment	Selection of specific treatments for certain patients e.g. Intense Rehabilitation Program, Transcranial magnetic stimulation
Genes related to recovery of specific domains of deficits e.g. motor deficit, language deficit, sensory deficit, fatigue, cognitive deficits (and endophenotypes)	Support for identifying patients with increased risk of having certain types of outcomes after stroke
Genes related to recovery in specific locations of cerebral injury caused by ischemic stroke, e.g. cerebral cortex, corticospinal tract, occipital lobe	Individualized treatment depending on potential for recovery

Recent reviews of drug pipelines from large pharmaceutical companies, covering a multitude of drug targets (including e.g. musculoskeletal, metabolic, inflammation, cardiovascular, and neurological/behavioural targets) have revealed that if a drug target is independently confirmed using human genetic data, the target is twice as likely to demonstrate clinical efficacy or attain FDA approval.^{51, 52} These improved odds can reduce investment in time and money for translation of targets to drugs. Several drug companies have invested in human genetics to enhance odds of detecting new pharmacological treatments, one example being GlaxoSmithKline using USD 300 million for co-operation with the gene testing company 23andMe, including a project on Parkinson’s disease (Hirschler 2018 https://www.reuters.com/article/us-gsk-results-idUSKBN1KF1AZ accessed on 9 Feb 2021).

Towards that end, human genetics can be a helpful adjunct to traditional forward translational approaches. To uncover genetic variations related to stroke outcome, many different approaches for genetic analyses can be considered: examination of monogenic disorders, candidate gene approaches, and GWAS.

Monogenic disorders may provide clues to mechanisms related to stroke outcome. For example, in CADASIL (Cerebral Autosomal Dominant Arteriopathy with Sub-cortical Infarcts and Leukoencephalopathy), impaired cognitive function has been related to decreased intranetwork connectivity and reduced local brain activity.⁵³ However, until now monogenic stroke disorders have shed more light on stroke risk rather than on stroke outcomes.

Hypothesis-testing candidate gene approaches—querying associations between variants in specific genes and clinical traits—have also been used to explore possible genetic associations. Genes that have received interest include BDNF (brain derived growth factor) rs6265 Val66Met variant (Supplement Table S1), APOE (Apolipoprotein E), VEGF (vascular endothelial growth factor), Cox-1 (cyclooxygenase-1) and Cox-2 (cyclooxygenase-2).^{34, 54, 55} However, candidate gene approaches, by their intrinsic design, are biased - the selection of the candidate genes is based on preconceived ideas of mechanisms and disease. Further, candidate gene studies have been shown to produce a high rate of false positives,⁵⁶ and are often difficult to replicate in much larger studies.⁵⁷ Candidate genes reported to be associated with complex diseases have been difficult to confirm by GWAS.⁵⁸ For example, the BDNF polymorphism has not been confirmed in GWASs as a variant influencing stroke outcome (GODS study, unpublished data). Publication bias may also confound these findings, as positive associations are more likely to be published than negative associations.³⁴ Candidate gene studies could be considered to confirm/replicate GWAS findings.

The GWAS approach has the advantage of being unbiased (searching hundreds of thousands/millions of SNPs throughout the genome for genetic associations). Thus, GWAS are agnostic and can yield unexpected findings indicating novel mechanisms; and confirm known mechanisms (often validating the approach). Additionally, GWASs are very reproducible and can be validated in independent laboratories with different populations. GWASs are driven by genetic and phenotypic diversity. However, GWASs often require large numbers of subjects (thousands to tens of thousands) to detect genetic associations with genome-wide statistical significance. This can only be achieved by combining large cohorts and multiple studies. But there is a risk that the phenotyping is not harmonized between the studies and may also be of insufficient detail to perform studies of quantitative endophenotypes.¹⁰ GWAS can be used to study copy number variations and one study reported that genetic imbalance is related to poor 3-month outcome after stroke.⁵⁹ Other genetic approaches such as Whole Exome Sequencing or Whole Genome Sequencing are promising but the implementation due to high costs is still limited.

Human disease GWASs have largely utilized case-control designs— comparing the frequency of genotypes in a cohort of cases compared to a matched cohort of controls.⁶⁰ Numerous case-control GWASs have examined stroke risk.² However, far fewer genetic studies have examined outcome after stroke. By design, these studies are different from the traditional case-control design because they involve only cases (no controls), and often examine quantitative outcome-related traits. Comparisons can be made over time in individual subjects and between subjects with different outcomes and phenotypes. These quantitative trait GWAS have the advantage of providing more power to detect a genetic effect compared to the binary case-control trait GWAS.⁶¹

Proof of Concept: Phenotypes, Genotypes, and Mechanisms involved in Stroke Outcome

We have discussed multiple clinical and neuroimaging traits related to stroke outcomes that might be used as phenotypes or endophenotypes in GWAS. But, can these phenotypes yield genes and pathways relevant to human stroke outcome?

This reverse translational approach may yield novel insights into underlying mechanisms contributing to human ischemic stroke outcome. Several groups have now examined distinct phenotypes within the injury-recovery spectrum after ischemic stroke onset, to capture genetic mechanisms using GWAS. Below we discuss four GWASs using phenotypes/endophenotypes from acute, subacute and chronic phases of stroke.

The Genetics of Early Instability after Ischemic Stroke (GENISIS) Study used GWAS to examine outcome in the acute phase of stroke³ and captured the evolution of neurological deficits (improvement/deterioration) during the first 24h after stroke, ΔNIHSS_24h, defined as the difference between NIHSS measured within six hours of stroke onset and NIHSS at 24h (see above).³⁰ Seven genome-wide significant loci were associated with ΔNIHSS_24h (Table 3).³ Two genes in these loci, ADAM23 and GRIA1 are expressed in neurons in the brain, specifically in excitatory synapses.^{62, 63} The two proteins (Adam23 and AMPA-receptor subtype 1) are part of a trans-synaptic protein complex that modulates neuronal excitability.⁶⁴ These results provide the first genetic evidence that excitotoxicity is relevant to acute ischemic stroke in humans. This is remarkable since acute ischemic stroke trials have failed to demonstrate efficacy of anti-excitotoxic drugs. Therefore, a re-examination of the earlier trial results may be of value, especially because many earlier clinical trials of anti-excitotoxic drugs may have lacked rigor, failing to confirm target engagement or to adhere to relevant therapeutic time windows.^65–67

Table 3.

Independent SNVs Leading the Most Significant Associations With Acute-/Long-Term Phenotypes From GWASs

Phenotype	Leading SNV	CHR	Location	Candidate Gene	Suggested Function of Candidate Gene	Other genes in the locus	p	Outcome	Reference
ΔNIHSS24h	rs58763243	2	intergenic	RSAD2	Interferon-inducible antiviral activity	RNF144A, LOC386597	LBF=6.51	Acute	3
ΔNIHSS24h	rs13403787	2	intergenic	DFNB59	Activity in auditory pathway neurons	AC074286.1	LBF=5.57	Acute	3
ΔNIHSS24h	rs72958644	2	downstream gene variant	ADAM23	Cell-matrix interactions.	CREB1, DYTN, NRP2, MDH1B	LBF=6.34	Acute	3
ΔNIHSS24h	rs12641856	4	intergenic	MGC45800	Unknown	LINC00290, MIR1205, TENM3	LBF=5.50	Acute	3
ΔNIHSS24h	rs114248865	5	intronic	GRIA1	Activity as excitatory neurotransmitter receptors	LOC101927134, FAM114A2, SAP30L, SAP30L-AS1, MFAP3, GALNT10, HAND1, MIR3141	LBF=5.29	Acute	3
ΔNIHSS24h	rs6930598	6	intronic	PARK2	Regulation of proteasomal degradation	PACRG, MAP3K4, AGPAT4, AGPAT4-IT1, LOC101929239	LBF=5.30	Acute	3
ΔNIHSS24h	rs10807797	7	intergenic	ABCB5	Transmembrane transport	TWISTNB, MACC1, TMEM196, RPL23P8	LBF=5.70	Acute	3
Parenchymal hematoma	rs76484331	20	intronic	ZBTB46	Unknown	LIME1, ABHD16B, ZGPAT	1.61x10−8	Acute	68
mRS at third month	rs76221407	1	intronic	PATJ	Mediation of protein-protein interactions	L1TD1, KANK4	1·72×10−9	Long-term	5
mRS 60–190 days after stroke	rs1842681	18	intronic	LOC105372028	Unknown	LOC729950	5.27×10−9	Long-term	4

CHR indicates chromosome; GWAS, Genome Wide Association Study; LBF, association value in MANTRA analysis; mRS, modified Rankin Scale; NIHSS, NIH Stroke Scale; ΔNIHSS_24h, NIHSS at 6 hours minus NIHSS at 24 hours; and SNV, single nucleotide variation.

The study also showed that common genetic variants accounted for 8.7% of the variance of ΔNIHSS_24h. The seven newly detected genetic loci only accounted for 2.1% of this variance.³ Therefore, many additional loci remain to be discovered.

Aside from the direct effects of acute brain ischemia, additional injury mechanisms such as hemorrhagic transformation, edema formation, and infectious complications may affect outcome after stroke. A GWAS of hemorrhagic transformation, defined by follow-up head CT images with parenchymal hematoma (PH), in acute ischemic stroke patients treated with recombinant tissue plasminogen activator (rtPA)⁶⁸ detected that a genome-wide significant variant (rs76484331) located in the ZBTB46 gene was associated with PH in the discovery cohort (n=1.324, p=1.61×10–8) and replicated in an independent population (n=580, p=0.01). The encoded protein is a transcription factor expressed in the brain. Interestingly a polygenic risk score with 3,506 genetic variants associated with PH was also associated with disability and mortality at three months. Disability at three months, remained significant after logistic regression including baseline NIHSS sex and age (p=0.01). This study confirms the strategy of using endophenotypes to study more complex variables such as acute and long-term outcome.^{68, 69}

Edema susceptibility post stroke – which can be studied by serial CT scans and automated machine-learning methods (see above) - could have a genetic predisposition, however more studies are needed and should ideally account for pathways important for edema formation, including initial stroke severity, collateral flow, and reperfusion status. For GWAS, large numbers of subjects are needed. Today, standard-of-care for patients with large vessel occlusion at greatest risk for edema—involves a wide array of neuroimaging (CT, and increasingly more often CTA, CTP, and sometimes MRI) which will provide this needed data.

Two GWASs have examined longer term outcome after ischemic stroke, using the modified Rankin scale (mRS) measured months after stroke onset. In the Genetic Contribution to Functional Outcome and Disability after Stroke (GODs) study, the authors examined 2,482 stroke patients applying liberal and stringent criteria.⁵ When the stringent criteria were applied, excluding patients with previous disabilities or minor strokes, a single nucleotide variants in the PATJ gene was associated with long-term outcome. The protein encoded by this gene localizes to tight-junctions at the apical membrane of epithelial cells.⁷⁰ PATJ, expressed in endothelial cells in the brain, might be associated with blood-brain barrier homeostasis.⁵ These findings emphasize the importance of carefully-defined phenotypes in stroke outcome studies. This concept has been adopted by the Global Alliance in Acute and Stroke Outcome initiative (https://genestroke.wixsite.com/alliesinstroke) as reflected in a review paper¹⁰ which recommends minimum variables needed for genetics studies in stroke acute and long-term outcome.

The second GWAS of long-term outcome after ischemic stroke⁴ enrolled 6,165 patients from the GISCOME (Genetics of Ischaemic Stroke Functional Outcome) network. The phenotype used in this study was a dichotomized mRS measured at 60 to 190 days after stroke onset. The authors found a genome-wide significant polymophism in LOC105372028. This locus is a trans-expression quantitative trait locus (eQTL) for PPP1R21, which encodes a regulatory subunit of protein phosphatase 1, and has been implicated in brain plasticity. Limitations included heterogeneous time-points for outcome measures (60–190 days), lack of data on t-PA treatment, and possible presence of disabilities before stroke occurrence. Thus, these findings should be replicated in independent cohorts. GISCOME data is available in the cerebrovascular portal for further study of long-term stroke outcome e.g.⁷¹

Though in early phases, the above described GWAS of stroke outcomes have demonstrated the feasibility of the approach for discovering genetic mechanisms involved in the injury-recovery spectrum. However, it remains to be proven that this approach will be fruitful for discovering molecular mechanisms relevant for stroke outcome. Towards this end, research “at the bench”, directed by GWAS results will be essential, highlighting the importance of fluid forward and reverse translation. Important considerations for future GWAS studies include: 1) carefully devised phenotypes and endophenotypes that remain close to genetic biology; 2) a stringent accounting of time after stroke onset and stroke severity; and 3) dynamic measures that capture evolution of pathophysiological or repair processes. Critical for feasibility of these studies will be the standardization of these phenotypes¹⁰ which should remain close to standard-of-care practices, to reach the large numbers of patients required for GWAS.

By examining genetic results of multiple phenotypes, it is likely that an overall genetic architecture across the injury-recovery spectrum will emerge. Many genetic associations may involve overlapping variants, genes or pathways that relate phenotypes in the complex web of mechanisms that likely tie them together with human stroke outcomes (Figure).

Beyond single Genetic Analyses/GWAS

Bioinformatics, epigenomics, transcriptomics among others are important tools that can be used in combination with genetics data to improve the understanding of diseases or to find potential drug targets. For example, the genetic information GWAS can be combined with powerful informatics tools to find genes and biological pathways associated with the trait or disease. These tools (e.g., Ingenuity Pathways (https://digitalinsights.qiagen.com/products-overview/discovery-insights-portfolio/analysis-and-visualization/qiagen-ipa/), Reactome⁷² among others) use curated information to determine biological pathways overrepresented in a set of genes. They can also match biological pathways with drug targets, to identify potential drugs that might intervene on relevant mechanisms.

In a similar manner that a single phenotype is used to discover multiple associated genetic variants, a single genetic variant can be used to discover multiple associated phenotypes. This so-called “PheWAS” (phenome-wide association studies) use well-annotated phenotyped populations (e.g., medical records from patients with well-delineated diagnostic codes) to find associations with a single genetic variant, with the potential to link disparate diseases or traits to a common genetic mechanism, and adding insight into disease processes.⁷³

The genetic correlation analysis also examines the proportion of variance that two traits or diseases share due to overlapping genetic contributions. Examples of genetic correlation tools include GNOVA and LDscore.⁷⁴ In one analysis of acute stroke patients treated with tPA, correlation genetics showed that spontaneous intracerebral cerebral hemorrhage (ICH) subtypes were associated with tPA-induced hemorrhagic transformation,⁶⁸ suggesting that risk factors associated with spontaneous ICH may be important for tPA-induced hemorrhagic transformation.

Other omics approaches such as epigenomics, transcriptomics, and proteomics promise to complement and extend the discovery of mechanisms, beyond that of genomics alone. Epigenomics—the study of specific DNA methylation sites (CpG sites)—has been applied to ischemic stroke risk using Epigenome Wide Association studies (EWAs).⁷⁵ These EWAs use DNA methylation chips that can evaluate up to 800,000 CpG sites. DNA methylation data can be used to calculate a “biological age” which may be distinct from chronological age and be associated with long term stroke outcome.⁷⁶ In an EWAs analysis of DNA methylation in acute and subacute phases of 643 acute ischemic stroke patients, neurological deterioration during hospitalization was associated with a hypomethylation of the EXOC4 gene,⁷⁷ component of the exocyst complex involved in vesicle trafficking at neuronal synapses.⁷⁸

Transcriptomic and Proteomic approaches are high-throughput methods that examine the complete mRNA or protein profile expressed in tissue or biofluids. Discussion of transcriptomic or proteomic studies is beyond the scope of this review and has been reviewed elsewhere.⁴⁷ Transcriptomic and proteomic techniques complement genomic approaches by expanding analysis to downstream gene products and their complex regulation in relation to stroke injury and recovery. The integration of big data across the omics, “multi-omics”,⁷⁹ promises to expand the array of biological pathways (across DNA, RNA, proteins, metabolites) that can be associated with the carefully defined phenotypes discussed above.

Confounding issues with transcriptomic studies include accessibility to relevant tissue (e.g. the brain) and the multitude of cell types from which the RNA is derived, making it difficult to interpret bulk transcriptomic data. While there are techniques capable of capturing individual cells (e.g., laser microdissection), reaching meaningful numbers is difficult. “Single cell transcriptomics” can quantify mRNA species from individual cells dissociated from bulk tissue samples, and bioinformatically “purify” clusters of transcriptionally-related cell types. This powerful technology allows for a deeper understanding of cellular reactions to traits or diseases at single-cell resolution. Techniques have also been developed to permit single cell transcriptomics to be performed on fixed human tissue.

Interpreting epigenomics, transcriptomics and proteomics data in the context of stroke is challenging. Unlike the genome, which is fixed throughout life, the epigenome, transcriptome, and proteome may change in response to the environment—e.g. in reaction to catastrophic events such as stroke. It is difficult to discern which epigenetic, transcriptomic, or proteomic changes are due to stroke risk vs. consequence of stroke.⁸⁰ Mendelian Randomization, a method that queries causal relations between modifiable risk factors and disease or condition (stroke outcome), can potentially resolve this. The risk alleles for modifiable factors (i.e. protein, or RNA levels) are randomly distributed in a population. Mendelian Randomization exploits this randomness to construct a “natural randomized clinical trial”. If the risk allele is associated with the disease, the modifiable factor may be causally related to disease. Applying Mendelian randomization to the GISCOME data showed that depression was related to poor recovery after stroke.⁷¹ Other Mendelian randomization studies focus on plasma proteins associated with stroke outcome, such as metalloproteinases (Carcel et al. Stroke in press) or coagulation/fibrinolytic factors.

Mendelian analyses have other utilities such as estimations of drug efficacy and safety.⁸¹ Using genetic information from genes that encode drug targets, Mendelian randomization can be used to offer insight into mechanism-based efficacy and adverse effects. Using a Phewas approach, it is possible to identify potential traits associated with the genetic variations in the drug target gene. Subsequently, MR analyses are useful for finding potential causal relationships among the genetic variations of the target gene and the traits found in the Phewas, thereby estimating potential drug effects.

Genetic databases can also be used for predicting the impact of drugs in disease. A previous study re-identified 70 of 600 known licensed targets by GWAS,⁸² demonstrating a pathway for the identification and priorization of drug targets. As discussed above, studies have shown that targets with genomic support have a higher rate of success.^{51, 83, 84}

Conclusions

Human genetics is a powerful approach to study mechanisms influencing both early brain injury and long-term recovery after ischemic stroke. Critical to the design of studies is the development of informative phenotypes and endophenotypes that capture mechanisms in the injury-repair time-line following stroke onset. GWAS is an established method to uncover genetic mechanisms of disease. Stroke outcome GWASs using these phenotypes/endophenotypes have yielded genetic loci and genes that are plausibly involved injury/repair mechanisms, supporting this reverse translational approach. Regardless of whether the associations can be linked to underlying molecular mechanisms they may be valuable as genetic biomarkers for prediction of treatment responses and prediction of prognosis. Larger sample sizes with harmonized data sets derived from standard of care measures will be needed to identify more genetic mechanisms. Indeed, the ISGC has promoted standardized phenotypes for use in genetic studies of stoke outcome. Beyond GWAS, novel methods that exploit informatics and multi-omics approaches hold promise for discovering even more novel molecular pathways related to stroke outcome and identification of potential drug targets, and genotypes that predict responses to specific therapeutic interventions.

Non-standard Abbreviations and Acronyms

CSF

Cerebrospinal fluid

GWAS

Genome Wide Association studies

tPA

Tissue-type plasminogen activator