Clinical spectrum, genetic complexity and therapeutic approaches for retinal disease caused by ABCA4 mutations

The ABCA4 protein (then called a “rim protein”) was first identified in 1978 in the rims and incisures of rod photoreceptors. The corresponding gene, ABCA4, was cloned in 1997, and variants were identified as the cause of autosomal recessive Stargardt disease (STGD1). Over the next two decades, variation in ABCA4 has been attributed to phenotypes other than the classically defined STGD1 or fundus flavimaculatus, ranging from early onset and fast progressing cone-rod dystrophy and retinitis pigmentosa-like phenotypes to very late onset cases of mostly mild disease sometimes resembling, and confused with, age-related macular degeneration. Similarly, analysis of the ABCA4 locus uncovered a trove of genetic information, including >1200 disease-causing mutations of varying severity, and of all types – missense, nonsense, small deletions/insertions, and splicing affecting variants, of which many are located deep-intronic. Altogether, this has greatly expanded our understanding of complexity not only of the diseases caused by ABCA4 mutations, but of all Mendelian diseases in general. This review provides an in depth assessment of the cumulative knowledge of ABCA4-associated retinopathy – clinical manifestations, genetic complexity, pathophysiology as well as current and proposed therapeutic approaches.

Numerous clinical studies have since been published adding to the growing body of knowledge. However, the "modern era" of our understanding of STGD1 was precipitated by parallel breakthroughs in the basic sciences beginning with the initial characterization of the ABCA4 protein in 1978, which was initially referred to as "rim" protein for its localization in rod photoreceptor outer segments and incisures (Papermaster et al., 1976). The genetic locus was mapped to 1p13 in mid-1990s (Anderson et al., 1995;Gerber et al., 1995;Kaplan et al., 1993) and, finally, the gene was cloned in 1997 (Allikmets et al., 1997). Taken together, cumulative advances over the last three decades have provided a defining foundation for understanding what we now know to be the most common inherited Mendelian eye disorder in the world.

Ophthalmic examination
In the "classic" presentation of STGD1, central vision loss typically becomes apparent between adolescence and young adulthood. However, the age of onset varies extensively, where a proportion of individuals start experiencing delayed vision loss between the 4th and 7th decades of life (Gerber et al., 1995;Lambertus et al., 2016;Runhart et al., 2018;Runhart et al., 2019;Westeneng-van Haaften et al., 2012;Yatsenko et al., 2001;Zernant et al., 2018). The exact age of disease onset is often difficult to determine, as many patients-particularly children-may be unaware of their visual impairment or have preserved central vision due to functional sparing of the fovea (Bax et al., 2019b;Fujinami et al., 2013b;Nakao et al., 2012;Runhart et al., 2019;van Huet et al., 2014). In general, ABCA4-associated retinopathy subtypes that manifest early in life tend to progress more rapidly, while a later age of onset is associated with a milder prognosis (Fujinami et al., 2015;Tanaka et al., 2018;Zernant et al., 2017). The initial symptoms of ABCA4-associated retinopathy typically begin with central or pericentral vision loss and may include difficulty with dark adaptation as the disease progresses in severity (Fishman et al., 1999;Kang Derwent et al., 2004;Klien and Krill, 1967;Salvatore et al., 2014;Scholl et al., 2002). The latter is seldomly reported by patients but generally recognized upon careful inquiry in the clinic. Other symptoms may include impaired color discrimination and photophobia (Klevering et al., 2002;Rotenstreich et al., 2003). Examination of the anterior segment and vitreous is generally unremarkable. As most patients present to the clinic in relatively early disease stages, fundoscopic findings may be subtle and rarely include typical features of advanced retinal degeneration such as a pale, disc pallor or extensive attenuation of the retinal vessels. al., 2005;Shankar et al., 2016;Sohocki et al., 1998). One should be aware of including relatives with age-related macular degeneration (AMD) in the assessment of family history due to its overlapping symptoms with late-onset STGD1 including central vision loss, especially as it has been reported that the prevalence of AMD is higher in families with STGD1 (Allikmets, 2000;Souied et al., 1999). Lastly, pseudodominant inheritance has been extensively reported in families segregating three Huckfeldt et al., 2016;Shroyer et al., 2000), or even four ABCA4 alleles within a single family (Klevering et al., 2002;Lee et al., 2016;Runhart et al., 2018). A further heightened awareness is warranted when working with patients of consanguineous families, mostly from culturally and geographically isolated populations due to the founder alleles resulting in frequent enrichment of homozygosity (Ducroq et al., 2006;Falfoul et al., 2018;Lee et al., 2017).

The diagnostic triad
Despite the breadth of clinical heterogeneity associated with ABCA4-associated retinopathy, all patients share a common genetic etiology caused by mutations in a single gene, and a set of near ubiquitous clinical features. The following are three diagnostic clinical findings that, when occurring together in a patient, are highly indicative of ABCA4-associated retinopathy: 2.3.1. Macular affection-Progressive deterioration of cellular layers originating in the central macula is a canonical feature of ABCA4-associated retinopathy and the principal cause of visual deterioration over time. The cellular origin of ABCA4 dysfunction and temporal sequence of cellular degeneration has been and remains a contentious issue Greenstein et al., 2017;Lenis et al., 2018;Song et al., 2015a;. Clinically, atrophic changes typically begin with loss of the outer retinal layers (retinal pigment epithelium -RPE and photoreceptor-attributable ellipsoid zone) and, in cases that progress to more advanced stages, invariably involve the choriocapillaris. Further deterioration of the underlying choroidal layers (Sattler and Haller) may occur in response to the rapid demise of RPE (Bertelsen et al., 2014;Muller et al., 2017;Tanaka et al., 2018) or progressively to a stage at which the underlying sclera may be visible on fundoscopy depending on the duration of disease ( Fig. 1)  .

Fundus flecks-
The augmented accumulation of RPE lipofuscin occurs throughout the retina; however, this process can also manifest locally giving rise to one of the most recognizable features of ABCA4-associated retinopathy-flecks. This feature is most conspicuous on short-wavelength autofluorescence (SW-AF) as intensely fluorescent foci distributed across the macula or extending far across the posterior pole at more advanced disease stages Cideciyan et al., 2015). Histopathological observations have attributed flecks to engorged RPE cells, stacked aggregations of RPE cells or remnants of dying RPE (Eagle et al., 1980;Lopez et al., 1990). However more recently, Sparrow et al. proposed that flecks may be extracellular accumulations of unphagocytized outer segments from longitudinal observations of their structural (SD-OCT) and autofluorescence (SW-AF and near infrared-AF) characteristics over time (Sparrow et al., 2015). The ubiquitous presence of flecks in ABCA4-associated retinopathy is an size and morphology, irregular patterns across the fundus and spatial evolution over time, may conceal crucial information about disease etiology and prognosis. The most frequent fleck patterns of ABCA4-associated retinopathy are illustrated in Fig. 2. Unfortunately,very few studies have comprehensively explored the significance of fleck characteristics to date. This is likely due to the community's preoccupation with lesion-centric characteristics and atrophy progression. Functionally, patients exhibit decreased visual sensitivity over flecked areas but their contribution to the centrally progressing atrophy is uncertain (Querques et al., 2006;Verdina et al., 2012). Flecks are generally regarded as a biomarker of disease severity based on their emergence along the central and peripheral axis of the ABCA4 associated retinopathy retina (Cukras et al., 2012). Teussink et al. sought to study the effect of light on the progression of ABCA4-associated retinopathy by comparing fleck accumulation in an eye of patients compared to the fellow eye that was continuously patched over one year, but the results were variable (Teussink et al., 2015). The causal role of flecks in the pathophysiology of ABCA4-associated retinopathy is thus far uncertain and much remains to be elucidated. Particular insight may lie in their longitudinal patternscentrifugal (Cukras et al., 2012) versus zonal Sparrow et al., 2015). Furthermore, flecks are highly dynamic and exhibit rapid changes, which may better define the "leading disease front" of ABCA4-associated retinopathy as compared to the central lesion of atrophy.

Peripapillary sparing-Perhaps
the most unusual feature amongst the triad is the observation that the proximal tissue surrounding the optic nerve is spared of disease changes in ABCA4-associated retinopathy ( Fig. 3) . Other diseases, most notably PRPH2and ROM1-associated pattern dystrophy and RDH12-associated Leber congenital amaurosis (LCA), have been reported to exhibit a similar manifestation although not as consistent as in ABCA4-associated retinopathy (Duncker et al., 2015c;Garg et al., 2017;Ma et al., 2019). Sparing of this region often persists in ABCA4-associated retinopathy, both structurally and functionally but progressively lost at later disease stages; however, its etiological basis is unknown. Several theories have been proposed including the disc membrane-load hypothesis, light-load hypothesis, lipofuscin-clearance hypothesis and neurotrophic factors hypothesis, although all are largely inferential .
Each feature of the ABCA4-associated retinopathy triad exhibits stage-dependent changes, which have been documented independently. Effective modeling of ABCA4-associated retinopathy should encompass the variability of all three features and their relationship to one another in order to acquire a deeper understanding of the condition.

Bull's eye maculopathy
A diagnostic exception to the triad is the Bull's eye maculopathy (BEM) stage which accounts for up to ~20% of ABCA4-associated retinopathy cases presenting to the clinic and is over-represented by the c.5882G>A, p.(G1961E) variant (Celia et al., 2009). BEM is defined as a circularly confined region of atrophy beginning in the central macula. The BEM stage precedes the development of all triad features making this stage the most challenging to diagnose. Furthermore, overlapping variations of the BEM phenotype are found in many inherited macular as well as non-genetic conditions including hydroxychloroquine toxicity, infectious diseases and acute injuries, although BEM observed in ABCA4-associated retinopathy can be often distinguished by quantifying macular levels of autofluorescence (RPE lipofuscin) (Duncker et al., 2015b). A detailed discussion of differential diagnoses is presented in section 3, Phenocopies of ABCA4-associated retinopathy. Representative manifestations of BEM in ABCA4-associated retinopathy are provided in Fig. 4.

Early perturbations in young patients
Studies to date indicate that visual function loss may precede readily detectable fundus features in young patients on clinical exam, although asymptomatic cases due to foveal sparing can be incidentally encountered by routine examination Lee et al., 2014). In a cohort of 50 young ABCA4-associated retinopathy patients (age ≤ 10 years), Lambertus and colleagues identified 10 individuals with visual function loss in the absence of discernible fundus abnormalities at the time of examination and reported the onset of visual acuity decline from as early as 3 years of age (Lambertus et al., 2015). Defects in color vision have also been reported in patients with early to no detectable fundus changes (Bax et al., 2019a;Vandenbroucke et al., 2015). The youngest documented case of ABCA4-associated retinopathy was an asymptomatic 5-year-old girl in a pseudodominant family harboring the c.5018 + 2T>C, p.(?) and c.5882G>A, p.(G1961E) alleles. At the time of examination, the girl had mildly decreased visual acuities and no fundus changes except a prominent thickening of the external limiting membrane (ELM) on OCT (Burke et al., 2013). A subsequent study  and several others thereafter (Bax et al., 2019a;Melillo et al., 2016;Pang et al., 2015;Park et al., 2015) corroborated the observation of ELM thickening to be a prominent feature of early stage ABCA4-associated retinopathy while more recently, this thickening has been attributed to the adjacent outer nuclear layer (ONL) . It is possible that structural changes occur prior to functional loss in patients. Considering the pathophysiology of STGD1, marginal increases in autofluorescence or microscopic perturbations in the cone or rod mosaic may be plausible outcomes to pursue in the future with advances in quantitative autofluorescence (qAF) imaging and adaptive optics-scanning laser ophthalmoscopy (AO-SLO), respectively. Doing so will shed light on the anatomic origin of STGD1 and ultimately shape therapeutic approaches.

Characteristics of advanced stages
Progression to the advanced stages of ABCA4-associated retinopathy varies in accordance with the age of disease onset. An earlier onset of disease is usually associated with a poorer prognosis in patients. In the most severe cases, generalized rod and cone function is unrecordable by ffERG, vision deteriorates to hand motion or light perception and chorioretinal atrophy extends as far as the equatorial regions of the eye and as deep as the underlying sclera. The emergence of pigmentary changes is also highly associated with advanced disease and often accompanies other indicators of retina-wide degeneration such as waxy optic disc pallor and severe attenuation of the retinal vasculature. The appearance of the pigment deposits ranges from nummular aggregations that co-localize with atrophic lesions to extensive bone spiculeshaped depositions. The appearance of the latter is identical to the pathognomonic bone-spicule pigment seen in the fundus of patients with retinitis pigmentosa (RP).

Phenocopies of ABCA4-associated retinopathy
The clinical expression of STGD1, an autosomal recessive disease, is extremely variable (see above). There have been three more loci, called STGD2-4, which describe genetically and phenotypically different diseases. Historically, the term "Stargardt-like macular dystrophy" was introduced in 1994 for a dominantly inherited maculopathy, which mapped to a locus on 6q (Stone et al., 1994). The ELOVL4 gene was later cloned from this locus, which is also called STGD3. Phenotypes caused by ELOVL4 mutations are clinically, genetically and pathophysiologically very different from the "real" Stargardt disease due to mutations in ABCA4 (STGD1). While the macular dystrophy phenotype of the dominant forms of the disease may somewhat resemble STGD1, the recessive forms express spinocerebellar ataxia 34 (Giroux and Barbeau, 1972;Turcotte Gauthier, 2010), ichthyosis, spastic quadriplegia and, mental retardation (Aldahmesh et al., 2011), among others. All forms of the disease are due to defects in fatty acid metabolism (Agbaga et al., 2008). Another suggested independent locus for STGD2 disease was eventually discarded and included in STGD3. The locus for another "Stargardt-like" phenotype, STGD4 (Kniazeva et al., 1999), contains the PROM1 gene. While the phenotype of patients caused by dominant PROM1 mutations sometimes resembles ABCA4-associated retinopathy (Wolock et al., 2019), the recessive form resembles RP (Maw et al., 2000;Zhang et al., 2007). These terms have, unfortunately, remained in the literature and even evolved into "dominant Stargardt disease", which is misleading and incorrect.
The common denominator for STGD1/ABCA4-associated retinopathy is maculopathy; i.e., the disease invariably begins in the central macula; however, as described below, the age of onset and progression are highly variable depending on the combination of specific disease-causing alleles and modifiers. Monogenic maculopathies collectively comprise a larger group; currently variants in ~38 genes are known to cause macular disease ( Table  1). Some of these, e.g., CRX, MT-TL1, PPRH2, RDH12 and RPGR-associated diseases, can be challenging to distinguish from various stages of ABCA4-associated retinopathy (Fig. 5). Moreover, some maculopathies are also caused by environmental factors, such as drug-toxicity (e.g. hydroxychloroquine) (Noupuu et al., 2016;Shroyer et al., 2001a), light damage, etc. Therefore, it is often practically impossible to determine the cause of a maculopathy without comprehensive genetic screening. Even at the most advanced retinal centers, about 10-15% of cases, who are clinically diagnosed with ABCA4-associated retinopathy, do not harbor disease-causing ABCA4 variants. Most of these cases, called phenocopies, are solved by more thorough clinical assessment, including a careful examination of family history, knowledge of environmental exposure and, eventually, by genetic screening, usually by whole exome sequencing (WES) (Wolock et al., 2019).
The gene that is most often carrying (dominant) variants, which are associated with phenotypes that are indistinguishable from ABCA4-associated retinopathy, is PRPH2. Variants in PRPH2 cause autosomal dominant pattern dystrophy and often exhibit the triad of ABCA4-associated retinopathy features. Since ABCA4-associated retinopathy is recessive, it should be quite straightforward to distinguish the two by family history. However, PRPH2 variants often exhibit variable penetrance across generations in a onset disease, which may also lead to an incorrect inference of AMD. Furthermore, pseudodominant families are frequent in ABCA4-associated retinopathy due to high allelic load of ABCA4 variants in the general population Huckfeldt et al., 2016;Lee et al., 2016;Maugeri et al., 1999;Shroyer et al., 2000;Tracewska et al., 2019) and late onset disease expression due to reduced penetrance of variants in both PRPH2 and ABCA4 is well documented Yatsenko et al., 2001;Zernant et al., 2017Zernant et al., , 2018. Therefore, also taking into account phenotype similarities, only comprehensive genetic screening can solve the causality in these cases.
In a recent study we investigated a cohort of cases where phenotypes were consistent with ABCA4-associated retinopathy but no disease-causing variants were found in ABCA4 even after complete locus sequencing (Wolock et al., 2019). While variants in PRPH2 were the most frequent cause of disease in this cohort (5-10% of all cases), and variants in another plausible gene, PROM1, came close second, this study also revealed variants in several other genes, usually not considered obvious causal candidates for phenocopies, including CDHR1, CERKL, CRX and RPE65 (Wolock et al., 2019).
These findings are not surprising, since genetic studies have recently significantly expanded phenotype heterogeneity in many retinal diseases in addition to those caused by ABCA4 variants. Variants in many genes, such as CRB1 and CRX, cause drastically different phenotypes depending on specific mutations and inheritance pattern, where some variants cause recessive disease while others cause dominant. The tiered list of genes (based on the diagnostic triad given above), mutations in which could lead to ABCA4 phenocopies, are given in Table 1. The important caveat, however, is the depth of clinical analysis, since some of these cases could have been distinguished from ABCA4-associated retinopathy already by comprehensive clinical assessment.

ABCA4 structure and function
The ABCA4 protein is an ATP-binding cassette (ABC) transporter in photoreceptor outer segments that functions in the visual cycle. More specifically, it is an N-retinylidene phosphatidylethanolamine and phosphatidylethanolamine importer (Quazi et al., 2012;Molday, 2013, 2014), the only known importer among mammalian ABC transporters. ABCA4 dysfunction results in accumulation of all-trans and 11-cis retinoids in photoreceptors (PRs), formation of A2E (and other bisretinoids) cumulatively called "lipofuscin", and their accumulation mostly in the RPE (Sparrow and Boulton, 2005;Sparrow et al., 2010;. This accumulation of cytotoxic products is a hallmark, and often also the cause, of most phenotypes resulting from dysfunctional ABCA4 . More recently, expression of ABCA4 has also been reported in the RPE, suggesting an additional role of the protein in this cell type that, when disturbed, could somehow contribute to ABCA4-associated retinopathy (Lenis et al., 2018).
The structure of ABCA4 at a high resolution is not yet described. Mammalian ABC transporters are notoriously difficult substrates for determining crystal structure (Dahl et al., 2004). The best published resolution of the native ABCA4 protein and some mutants, is 18 Å (Tsybovsky et al., 2010(Tsybovsky et al., , 2013Tsybovsky and Palczewski, 2014). The lack of high resolution ABCA4 structure makes functional studies challenging, limiting experimental systems to animal models (Makelainen et al., 2019;Molday et al., 2018;Molday and Molday RS, 2016;Zhang et al., 2015) and in vitro assays, including ATP binding, ATPase activity and vesicular transport studies (Ahn and Molday, 2000;Beharry et al., 2004;Sun et al., 1999). The current status of the structure and (biochemical) function of ABCA4 protein is outside of the scope of this review. These aspects have been described in depth in manuscripts from the laboratories of Robert Molday and Krzysztof Palczewski, and we direct the reader to these papers for excellent overviews of the status of structure and function correlations in ABCA4 (Molday, 2015;Tsybovsky et al., 2013;Tsybovsky and Palczewski, 2014). We will address the functional studies of ABCA4 variants affecting splicing in depth below.

Disease-causing variants in the ABCA4 locus
Disease-causing variation in the ABCA4 locus is extensive; there are currently >1200 disease-causing variants known and the number is rapidly growing as new cohorts, especially those of non-European descent, are screened (www.lovd.nl/ABCA4). When classified by a variant effect, the locus contains all classes of variants -missense, nonsense, indels, canonical and noncanonical splice site (NCSS) defects, deep-intronic variants and structural variants (SVs). These can be further grouped according to the proposed severity of the variant, including deleterious (i.e., complete null), severe, moderate, mild, and hypomorphic categories. The specific distinction between groups is not always unequivocal since, due to the lack of the high-resolution structure of the ABCA4 protein, the functional consequences derived from indirect assays (protein yield, ATPase and transport activity, etc.) do not always correlate exactly with resulting disease phenotypes and progression. For example, many missense alleles are deleterious (Molday et al., 2018;Zernant et al., 2017;Zhang et al., 2015), while other seemingly deleterious alleles (stop-gained and indel variants) sometimes do not result in a complete lack of function. Therefore, variant severity assignments are often not straightforward and these are constantly updated as new information becomes available from both genotype/phenotype analysis of extensive cohorts and functional studies, including those involving animal models. In section 6, we do classify variants based on pre-mRNA splicing defects.

Distribution of types of variants
As depicted in Fig. 6, most of the ABCA4 variants/alleles, both in number of unique variants/alleles (50%) (Fig. 6A) and total variant/allele numbers (61%) (Fig. 6B) are missense mutations. The relatively low contribution of protein truncating mutations (23% of total and 33% of unique alleles) probably can be explained through the genotype-phenotype correlation model in which all ABCA4-associated retinopathy cases, except those with early-onset disease, carry at least one non-truncating mutation. The latter may also explain the relatively high frequency of NCSS variants (~5%) as their effects range from mild to deleterious. As described in more detail below, 46 different structural variants (SVs) (in ~1% of all alleles) and 35 different causal deep-intronic (DI) variants (in 4% of all alleles), have been identified in ABCA4-associated retinopathy cases. As the majority of genotyped ABCA4-associated retinopathy cases has not yet been screened for the presence of SVs and DI variants, we estimate that ~2% of all ABCA4 alleles are SVs and ~10% are DI variants.
A significant fraction of ABCA4 alleles (~10%) consists of more than one variant (Shroyer et al., 2001b;Zhang et al., 2015). Two of these 'complex alleles' are conspicuously prevalent. The c.[2588G>C; 5603A>T], p.[Gly863A1a,Gly863del;Asnl868Ile] allele is present in ~50% of all complex alleles and its relevance only came to light after the significance of p.(Asn1868Ile) was appreciated as a single causal allele, when found in trans with a severe or moderately severe ABCA4 allele, see below Zernant et al., 2017). As shown previously, the c.2588G>C variant was always found in cis with c.5603A>T in ABCA4-associated retinopathy cases .
Both variants are required to render the complex allele fully penetrant  were also found in 43 and 78 alleles, respectively, in ABCA4-associated retinopathy cases (Cornells et al., 2017). Based on the combinations of variants identified in ABCA4 associated retinopathy cases and their allele frequencies in healthy individuals, they are considered to be (moderately) severe and mild, respectively (Cornells et al., 2017). The complex allele p.[Leu541Pro;Ala1038Val] is deleterious, a complete loss-of-function allele Zhang et al., 2015).

Founder mutations in different populations
The ABCA4 locus is informative with regard to multiple founder alleles; i.e. variants, which initially occurred in one geographical locale, a well-known phenomenon for geographically or culturally isolated populations (Table 2). What makes the ABCA4 locus especially interesting is that, in addition to expected significant differences of disease-causing ABCA4 alleles in various racial and ethnic groups, almost every nation in Europe has its "own" ABCA4 mutation, which is much higher in frequency than in neighboring countries.
Examples include the C.768G>T allele, which is very frequent in the Netherlands, but almost absent in neighboring Germany , the complex allele p.
[Leu541Pro;Alal038Val] in Germany , which has, however, spread especially throughout Eastern Europe due to geopolitical events (Sciezynska et al., 2016;Tracewska et al., 2019;Zolnikova et al., 2017), and the p.(Arg1129Leu) variant in Spain (Valverde et al., 2006). Most of these variants have occurred once; an interesting deviation from this is the severe p.(Asn965Ser) variant, which was first described as a founder mutation in Denmark  but subsequently also was found at a very high frequency in China . Haplotype analysis confirmed that the variant occurred independently in the two regions . The p. [Gly863Ala,Gly863del] variant is frequent in Western/Northern Europe  and the most frequent disease-causing ABCA4 allele, p.(Gly1961Glu), originates from Eastern Africa, where it is found in ~10% of Somalis Guymer et al., 2001). Subsequent population migration has spread p.(Gly1961Glu) throughout the world, but the allele frequency has dropped dramatically during evolution . The population frequency in Europe is ~0.4%, suggesting that the variant is causal in all, or at least most, populations. Other population-specific alleles are the deep-intronic c.4539 + 2001G>A variant, which is frequent in the Belgian population, comprising ~25% of all deep-intronic variants, but interestingly, is more rare in the Netherlands Bax et al., 2015). The p.(Ala1773Val) allele is a founder variant in Mexico (Chacon-Camacho et al., 2013;Lopez-Rubio et al., 2018). The Ashkenazi Jewish population has several founder alleles, including c. 4254-37_4254-15del (Beit-Ya'acov et al., 2007) and p.(Pro1380Leu) . In summary, the genetic screening for pathogenic ABCA4 alleles can often identify the ethnicity, and even the nationality, of a patient.
Although the founder alleles are already frequent within the population of European descent, they are much more prominent in various racial groups. While there is some overlap with the Caucasian population, possibly due to admixture, the disease-causing ABCA4 mutation spectrum is significantly different, for example, in African American  and in East Asian Jiang et al., 2016) populations. In both populations, the most prominent and frequent founder alleles, the c.101_106delCTTTAT, p. (Ser34_Leu35del), c.2424C>G, p.(Tyr808*), and c.6563T>C p.(Phe2188Ser), in China Jiang et al., 2016) and the p.(Val989Ala), p.(Gly991Arg) and p.(Arg2107His) variants in African Americans, are almost absent from European populations . The only exception is the p.(Asn965Ser) variant, which is an independent founder allele in both Denmark and China. The population frequency of the most frequent disease causing variant in African Americans, p.(Arg2107His), is ~2% in the general population of African American descent, suggesting that this allele can be considered hypomorphic, which is also supported by late-onset and mild disease in patients harboring this mutation . Another interesting observation is that the p.(Gly1961Glu) variant, which originates from East Africa and has a very high population frequency in Somalia, Kenya and Ethiopia , is almost absent in African Americans; i.e. in people of West African descent . Some populations, e.g., South Asian (Indian), have stronger admixture of European alleles . Most other racial and ethnic groups have not been screened in sufficient numbers of cases for valid statistical conclusions at this time.

Missing heritability
Recent advances in the genetic analysis of the ABCA4 locus, including complete locus sequencing, functional assays, and introduction of the concept of hypomorphic alleles, have significantly reduced the fraction of "missing alleles". Another important aspect in this regard is the much better clinical characterization of patients due to major advances in imaging technologies and increased experience of retinal specialists. In our centers in Nijmegen and New York, the fraction of "unsolved" cases (i.e., those with certain ABCA4 associated retinopathy diagnosis and one definite pathogenic allele) is <5%. The fraction of phenocopies, i.e., cases with ABCA4-associated retinopathy-like phenotypes but with causal mutations in genes other than ABCA4, is still 10-15% of all cases, even in the most advanced centers, but these are most often solved with WES, as described above (Wolock et al., 2019). So where are the remaining pathogenic ABCA4 alleles? There are several possible scenarios:

1.
Some variants can be in regulatory regions of ABCA4, promoter andenhancer sequences, which can affect ABCA4 expression Cherry et al., 2020). Some of the possibly regulatory variants have been identified, e.g., c.768 + 3223C>T and c.2919-383C>T , but more comprehensive searches and functional characterization are necessary for this category of possible mutations.

2.
Yet unidentified, and/or unconfirmed, deep-intronic variants. Sequencing of the entire ABCA4 genomic locus identifies in each patient, on average, 40 variants with an allele frequency <0.5% in population-matched control individuals, several of which could be considered causal even after thorough filtering for allele frequencies and comprehensive in silico analyses. Studies of monoallelic cases have identified and functionally proven 35 different deep-intronic variants to be causal, most of which are detected in more than one patient. However, since the ABCA4 locus is extremely heterogeneous and single cases of pathogenic variants are often identified in coding sequences, the comprehensive analysis of all possibly pathogenic deep-intronic variation remains a challenging task.
As described below, some deep-intronic variants cause retina-specific splicing defects , and almost all putative splicing variants thus far were tested in human embryonic kidney cells (HEK293T) which do not recapitulate the retina-specific splicing factors.

3.
Structural variants (SVs) are (very) rare in the ABCA4 locus (see below for deletions and duplications); however, it is likely that a small fraction of SVs is yet to be identified as short-read sequencing strategies will not identify inversions and insertions. Another class of very rare genetic events, such as uniparental isodisomy, has also been identified in three probands with ABCA4 associated retinopathy (Fingert et al., 2006;Khan et al., 2020;Riveiro-Alvarez et al., 2007).

4.
It has also been suggested that some of the ABCA4-associated retinopathy could be dominant, or di-or polygenic. While theoretically possible, there is currently no evidence for either scenario. While clinically dominant phenotypes, such as those caused by the p.(Gly1961Glu) mutation, are documented in ABCA4-associated retinopathy, there is no reason to expect any genetically dominant cases since, based on our current knowledge of the ABCA4 function, a dominant-negative effect is not expected for any ABCA4 allele. The entire ABCA4-associated retinopathy continuum is based on a loss-of-function mechanism, whether complete, or partial; i.e., haploinsufficiency, as in carriers of ABCA4 variants. Whether the latter mechanism results in a late-onset macular disease is still open for debate (Duncker et al., 2015a;Kjellstrom, 2015;Lee et al., 2019;Maia-Lopes et al., 2008). However, we postulate that, based on pre-mRNA splice assay data (see below), all recessive ABCA4-associated retinopathy cases (together from both alleles) have no more total residual ABCA4 activity than 40% .

In vitro splice assays in HEK293T cells
The analysis of putative splice defects ideally is performed using patient cells in which the gene of interest is endogenously expressed to perform reverse transcription-PCR analysis of the mRNA. In the absence of patient cells or when the gene of interest is not expressed in accessible human tissues, in vitro splice assays have traditionally been performed in human cell lines such as human embryonic kidney (HEK293T) cells. To this aim, small genomic fragments (<1 kb) were cloned in splicing vectors. As the ABCA4 gene is expressed at a very low level in non-ocular human tissues, we also employed minigenes to analyze NCSS variants previously identified in ABCA4-associated retinopathy cases. Due to the strong splice sites of vector exons that flank the cloned segments, splicing artefacts were observed . To systematically test the effect of NCSS ABCA4 variants, we cloned large wild-type genomic fragments (4.0-11.7 kb) of the ABCA4 gene into a Gateway splicing vector containing RHO exons 3 and 5 flanking the region of interest. The resulting splicing constructs were coined 'midigenes'. Apart from fragments containing parts of two very large introns (introns 6 and 11), all ABCA4 cloned segments contained at least 3 exons, enabling us to perform RT-PCR using primers annealing to ABCA4 exons, minimizing the occurrence of artefacts . As the first and last exon do not contain a splice acceptor or donor site, respectively, exon 1, intron 1, intron 49 and exon 50 are not or only partially represented in midigenes.

Causal noncanonical splice site variants in ABCA4
Upon testing all published NCSS variants, 64 showed a wide spectrum of splicing defects, including single exon skipping, multiple exon skipping, exon elongation, intron retention, and partial exon skipping (Table 3)  . The majority of these variants (35/64) resulted in 100% aberrantly spliced RNA and are considered deleterious. Another 12 variants showed >0 and ≤ 30% of normal splice products and can be classified as severe splice variants. Twelve variants showed >30% and ≤70% wild-type RNA and were classified to have a moderate effect, three variants were classified as mild (>70% and ≤80%) and one (C.3608G>A) was classified as benign as 95% of the RNA was correctly spliced. Finally, c.2588G>C resulted in a 3-nt deletion, p.(Gly863del), and a normally splice product that can be translated in ABCA4 protein carrying a missense variant, i.e., p.(Gly863Ala) Sangermano et al., 2018). In a subsequent study, this variant was found only to be causal (as a mild-moderate allele) when in cis with c.5603A>T, p.(Asn1868Ile) .

Causal near-exon variants in ABCA4
Seven variants located near exons were found to result in splicing defects that affect neighboring exons (Table 4). Variants c.1937 + 13T>G, c.1937 + 37C>G, c.3191-11T>A and c.4352+61G>A create or strengthen intronic splice sites. They thereby result in 12-nt, 36-nt, 9-nt and 57-nt exon elongations, respectively, leading to nonsense mutations in the corresponding RNA products (c.1937 + 13T>G, c.1937 + 37C>G, c.4352 + 61G>A) or a deletion of one amino acid and the insertion of four amino acids (c.3191-11T>A) Sangermano et al., 2018). Variants c.161-23T>G and c.4253 + 43G>A result in partial exon 3 and exon 28 skipping, respectively, rendering them mild variants based on genotype/phenotype analyses  and in vitro splice assays Sangermano et al., 2019). The c.4253 + 43G>A variant is the most frequent intronic ABCA4 variant that is not residing in NCSS sequences  resulted in a wildtype and three mutant cDNAs, one of which carried a pseudoexon (PE), one carried a partially overlapping PE and was missing exon 44, and one did not contain exon 45 ).

Causal deep-intronic variants in ABCA4
The first five causal deep-intronic variants in ABCA4 were discovered based on the hypothesis that they may strengthen cryptic splice sites flanking PEs that are present in a small fraction of ABCA4 transcripts within a normal retina .
In this way, small regions of the ABCA4 locus were sequenced in genetically unsolved ABCA4-associated retinopathy cases. Sequencing of the entire ABCA4 locus (Zernant et al., 2014b) in 114 monoallelic patients revealed another 16 possibly disease-associated variants, followed by variant-specific Bax et al., 2015;Schulz et al., 2017;Zernant et al., 2017), or complete locus analysis in several other ABCA4-associated retinopathy cohorts ( (Table 4).
In vitro splice assays were able to determine a partial or complete picture of the splicing defects for all deep-intronic variants except two. The effects of two neighboring variants in intron 30, i.e. c.4539+2001G>A and c.4539+2028C>T, were only shown using patient derived photoreceptor progenitor cells (PPCs) . Both variants do not affect the strength of splice sites flanking a 345-nt PE, but strengthen and/or create exonic splice enhancer motifs inside the pseudo-exon; i.e., have a different disease-causing mechanism than most other deep-intronic pathogenic variants. Based on the in vitro splice assays and PPC analysis, 7/35 deep-intronic and near-exon variants had a deleterious effect (i.e. no correct RNA), 13 showed a severe effect, 12 showed a moderate effect, one showed a mild effect, and two (c.1937+435C>G, c.1938-621G>A) were classified as benign as 95% of the RNA was correctly spliced. However, we do consider the latter variants to be causal and attribute the low percentage of mutant transcripts to the cell type used in the splice assay (HEK293T cells). It is difficult to estimate the frequency of these variants compared to all ABCA4 variants identified thus far, as not all ABCA4-associated retinopathy probands have been analyzed for variants in the entire genomic locus. We estimate that ~5% of all alleles (~10% of probands) carry causal deep-intronic or near-exon variants.

Structural variants in the ABCA4 locus
Structural variants (SVs) in the ABCA4 gene/locus are relatively rare based on Southern blot analysis Yatsenko et al., 2003), array-comparative genome hybridization (aCGH) assays (Zernant et al., 2014b) and multiplex ligation-dependent probe amplification (MLPA) analysis Bax et al., 2015;Sangermano et al., 2019;Zernant et al., 2014b). Table 5 lists all 46 reported SVs larger than 20 bp, including 35 deletions, 6 duplications, 2 deletions-insertions, 2 deletions with internal inversions, and 1 insertion. Of the 23 SVs for which the size is known, seven are smaller than 100 bp. However, he predicted effect is severe for all SVs, except for a 7-kb intron 1 duplication, for which the predicted effect is unknown. Only seven SVs have been found in more than one case. Based on nested RT-PCR studies of lymphoblast RNA of a homozygous proband, the intron 28 deletion c.4254-37_4254-15del resulted in the skipping of exons 29 or 28 and 29. The variant was found in homozygosity in 14 cases and in heterozygosity in one case in six families of an Arab-Muslim village in Israel (Beit-Ya'acov et al., 2007). The second most frequent SV is an exon 20-22 deletion that was reported in eight probands originating from Belgium, Germany and the Netherlands (Table 5). Based on these published data, we estimate that 1-2% of ABCA4-associated retinopathy probands carry a causal SV.

Genotype-phenotype correlations
The extensive clinical heterogeneity of ABCA4-associated retinopathy is primarily caused by the similarly significant genetic variability in the ABCA4 locus. Shortly after the discovery of the role of ABCA4 in various maculopathies, we proposed a somewhat simplistic genotype/phenotype association model suggesting a correlation between the continuum of disease phenotypes and residual ABCA4 activity/function Lewis et al., 1999;. According to that model, different combinations of "mild", "moderate", and "severe" ABCA4 mutant alleles were suggested to result in distinct phenotypes. While still valid in some, especially extreme, cases, this model underestimates the phenotypic heterogeneity of ABCA4-associated retinopathy.
The current model is derived from the analysis of large, comprehensively characterized cohorts of patients which have revealed an immensely complex landscape of independent disease trajectories that appear to be unrepresented in the ABCA4 genotype alone. The combinations of currently known >1200 disease-associated ABCA4 variants (Allikmets, 2007;Cornells et al., 2017;Maugeri et al., 1999) (www.lovd.nl/ABCA4) explain some, but definitely not all, disease phenotypes . The population frequency of potentially pathogenic ABCA4 alleles is 1:20 (Jaakson et al., 2003;Maugeri et al., 1999;Yatsenko et al., 2001), underscoring the substantial impact for the amount of retinal pathology attributable to ABCA4 variation. Some more recent developments in ABCA4 genetics, which have allowed us to make more precise genotype/phenotype correlations, are listed below.

Extremely hypomorphic and modifier variants
The emergence of 'extremely' hypomorphic and modifier alleles Runhart et al., 2018;Zernant et al., 2017Zernant et al., , 2018, most of which are still unknown, adds another layer of genetic and phenotypic heterogeneity. The most prominent of these variants is p.(Asn1868Ile), with population allele frequencies close to 7% in Europe. It had been shown in many studies to be more frequent in cases vs controls (Aguirre-Lamban et Maugeri et al., 2002;Webster et al., 2001) and speculated (Webster et al., 2001) or suggested , that it may be associated with the disease. While most of the association is due to linkage disequilibrium (LD) with highly penetrant pathogenic variants Zernant et al., 2017), we proved the pathogenicity of p.
(Asn1868Ile) under a specific condition -it is penetrant when in trans from a deleterious ABCA4 mutation   . The discovery of the p.(Asn1868Ile) pathogenicity when in trans with a deleterious ABCA4 allele solved ~50% of cases who were considered monoallelic at the time and, therefore, explained the missing heritability in ~10% of the entire ABCA4-associated retinopathy population . These studies also defined the frequent p.
[Gly863Ala,Gly863del] variant not as a pathogenic allele on its own, but rather as the first major modifier allele in the ABCA4 locus.

The c.5882G>A, p.(Gly1961Glu) allele is associated with mild disease and other specific subphenotypes
Patients harboring the p.(Gly1961Glu) allele in homozygosity or in compound heterozygosity have a substantially different phenotype. Age at onset (mean 22.7 years) is somewhat later than in ABCA4-associated retinopathy cases not carrying the p.
(Gly1916Glu) or p.(Asn1868Ile) alleles (mean 19.7 years) . Patients with the p.(Gly1961Glu) allele exhibit milder disease expression Celia et al., 2009), although not in the overall rate of disease progression but in a distinct phenotypic pattern that, interestingly, overlaps with patients harboring the p.(Asn1868Ile) allele. ABCA4-associated retinopathy invariably begins as a maculopathy with an enlarging lesion of outer retinal atrophy and accumulation of yellow foci, or flecks, at the level of the RPE. Patients with most other ABCA4 variants exhibit progressively severe fleck patterns from very few in the macula to a stage of "absolute confluence" across the posterior pole between the ages 30-40 years (Fig. 8).
Patients harboring p.(Gly1961Glu) or p.(Asn1868Ile) consistently exhibit milder spatio temporal fleck patterns even at advanced age, which never progresses to the absolute confluence stage as illustrated by fundus autofluorescence imaging which detects lipofuscin accumulation (Fig. 8). The resistance to fleck accumulation and lower rate of lipofuscin accumulation seen in bi-allelic cases carrying p.(Gly1961Glu) or p.(Asn1868Ile) , is further exemplified by the occurrence of a particular phenotype that is seen only in this patient group where a well-defined, unifocal lesion of dark atrophy with proximally-bordering, lesion-centric flecks appears at an early stage of the disease. Similar lesions of chorioretinal atrophy are typically numerous (multifocal) and observed in the background of more advanced fleck stages with most other ABCA4-associated retinopathy cases (Fig. 8). Additionally, the individual morphology of flecks in patients with p.(Gly1961Glu) or p.(Asn1868Ile) appears to be distinct in that they are predominantly larger in size and are well defined in shape and generally more sparsely distributed. Of specific interest is the fact that patients with the p.(Gly1961Glu) allele typically present with the 'bull's eye maculopathy' phenotype regardless of the allele in trans, thereby acting as 'clinically dominant' alleles (Celia et al., 2009). This is especially apparent in multi-generation families, where the other variables (genetic background, age, etc.) are largely the same . The pathophysiological reasons of this phenomenon remain to be elucidated; however, it is very likely that p.(Gly1961Glu) mutation selectively affects ABCA4 function of foveal cones, often resulting in the 'optical gap' phenotype at early disease stages .

Two deleterious ABCA4 alleles result in severe cone-rod dystrophy
Two deleterious alleles in ABCA4 consistently result in early, childhood-onset of disease symptoms and a rapid progression to advanced disease stages characterized by profound visual impairment and deep, retinal wide degeneration extending towards the equatorial limits of the retina. Such patients have been historically classified as cone-rod dystrophy (MIM# 604116) Maugeri et al., 2000), due to considerable attenuations of cone amplitude on ffERG, or retinitis pigmentosa (RP19, MIM# 601718) Martinez-Mir et al., 1998;Verbakel et al., 2018). Cases of the latter diagnosis indeed exhibit features associated with RP including deposition of bone spicule pigment, severe retinal vessel attenuation and extinguished ffERG responses, but also atypical features such as early macular involvement. Clinically distinguishing true RP (i.e. rod-cone dystrophy) from severe cone-rod dystrophy at such an advanced stage of disease can be challenging. Distinctions in the extent of residual rod or cone function would not be reliably detectable on ffERG due to widespread degeneration of both systems. Identifying indicators in the ocular history of night blindness and progressive constriction of the visual field may in such cases be more informative. An etiological connection between RP and ABCA4 thus remains to be established. Alternatively, the commonalities between pathophysiology of deleterious ABCA4 mutations and RP may result in the manifestation of overlapping features. Other characterizations of biallelic null ABCA4 phenotypes include the generalized choriocapillaris dystrophy (Bertelsen et al., 2014) and rapid-onset chorioretinopathy (ROC) (Tanaka et al., 2018). ROC is specifically characterized by disease onset within the first decade, a short interval of profoundly increased autofluorescence in the macula followed by the rapid development of degenerative lesions, which enlarge and coalesce across the posterior pole within the third decade of life (Fig. 8).

Variable expression and penetrance of ABCA4 alleles
As described above, ABCA4-associated retinopathy shows a broad spectrum of clinical expression, with onset ranging from as early as 5 years of age to as late as 70 years.
Although genotype-phenotype correlations are apparent, differences in clinical expression between individuals with the same combination of ABCA4 variants have often been observed, suggesting the involvement of both cisand trans-modifier alleles Runhart et al., 2018). Recently, the discussion, also between the authors of this review, has focused on late onset cases carrying a deleterious variant on one ABCA4 allele and the p.(Asn1868Ile) on the other allele.
Researchers at the Nijmegen center calculated the penetrance of the p.(Asn1868Ile) allele, when in trans with a deleterious or severe allele, to be ~5% in the general population (Cremers et al., 2018;Runhart et al., 2018) and documented unaffected male patients who carried the same combination of variants as their affected siblings, in three families . The time elapsed between the age at onset in the affected siblings and the current age of the unaffected siblings ranged between 14 and 37 years. However, it cannot be excluded that the unaffected bi-allelic individuals could still develop ABCA4-associated retinopathy later in life. In the same study, 22/34 (65%) affected persons were female and 12/34 (35%) were male, which was suggestive (p = 0.0615) for a sex imbalance; i.e. it seemed as if late-onset ABCA4-associated retinopathy affects more females than males (Cremers et al., 2018;Runhart et al., 2018). In a larger study of 125 international ABCA4 associated retinopathy cases carrying p. Researchers at Columbia argued against these observations  by suggesting that the disease prevalence (see below in the "Future research" section), and the "strength" of mutations (deleterious vs. severe, etc.) is largely unknown, so one has to be careful with using these variables in statistical calculations. Furthermore, the female-to-male ratio among all 1060 ABCA4-associated retinopathy cases at the Columbia center is 54:46 in the entire disease cohort and 57:43 in the sub-cohort with the p. (Asn1868Ile)  The observed differences in the sex data in two large ABCA4-associated retinopathy cohorts (with substantial statistical power) suggest further in-depth analysis of this phenomenon across populations. Of additional interest is the fact that differences between the two centers exist in both the overall sex balance in ABCA4-associated retinopathy and separately in the hypomorph subgroups.
In summary, these data strongly suggest our incomplete understanding of this, very important, issue, solving of which requires much more data and analysis. Specifically, this part of the ABCA4 genetic studies requires better understanding of both cisand trans-modifier alleles. We cannot exclude that there are thus far missed variants, especially in deep-intronic sequences, on the allele carrying the p.(Asn1868Ile) in ABCA4 associated retinopathy cases, which would render these alleles fully penetrant. A primary example for this scenario are the complex alleles harboring both p.(Asn1868Ile) and p.
[Gly863Ala,Gly863del] variants, which are fully penetrant . However, they would not explain the large differences of expression of ABCA4-associated retinopathy within families, in which trans-modifiers may play a larger role. Finding these, non-ABCA4, modifiers, genetic or non-genetic in nature, which influence the expression of ABCA4 associated retinopathy in a subset of the cases, remains an important task.

Animal models
The first animal model for ABCA4-associated retinopathy, the Abca4 knockout (KO) mouse (Abca4 −/− ), was introduced in 1999 ( Weng et al., 1999). Since then this mouse, which has also been generated independently by other groups (Kong et al., 2008), has been the main and only animal model for studying ABCA4-associated retinopathy. It is obvious that the mouse is not the best animal model for macular disease as it lacks the macula and has very few cones, which are the earliest target of ABCA4-associated retinopathy. Therefore, while the Abca4 KO mouse model does not fully replicate the human condition, its most prominent pathological feature, the extensive accumulation of lipofuscin/A2E, faithfully mimics the disease in humans and allows for precise quantification of the disease status, progression, and the therapeutic effect in pre-clinical studies (Charbel Issa et al., 2013;Sparrow et al., 2013). The rest of the disease features, photoreceptor degeneration (Bok, 2005), delayed dark adaptation (Mata et al., 2001;Weng et al., 1999) and ffERG defects are less robust and not always replicated from study to study.
More recently, two Abca4 knock-in (KI) animals have been generated and used to study specific protein defects and progression of the disease (Molday et al., 2018;Zhang et al., 2015). In addition to documenting the influence of several ABCA4 variants on the protein expression, misfolding, trafficking, ATPase activity, etc., these studies also defined the specific ABCA4 variants, p.(Asn965Ser) and p.[Leu541Pro;Ala1038Val], as mostly deleterious. The resulting phenotype did not differ from the KO mouse, i.e. the protein function was absent regardless of the specific mechanism and the phenotype outcome was the same.
In addition to the KO and KI animals, several other strains have been generated which represent dual or triple KOs, where other genes, in addition to Abca4, are knocked out. These include Abca4 −/− /Rdh8 −/− , Abca4 −/− /Rdh8 −/− /Nrl −/− and some others. Some of these, especially the Abca4 −/− /Rdh8 −/− strain, have been extensively used for modeling ABCA4 associated retinopathy and even AMD, although they do not, strictly speaking, represent models for these diseases. The Abca4 −/− /Rdh8 −/− strain presents with much more advanced phenotype, which is not surprising since the two consecutive proteins in the visual cycle are eliminated, thereby causing an earlier onset and fast progressing retinal degeneration (Maeda et al., 2008).
Anatomical differences between the mouse and human retina, namely the absence of a macula due to the spatial distribution of rods and cones in the former has been a long standing challenge in recapitulating disease features found in human patients. As such, extensive efforts over the last 20 years have been devoted to identify a more suitable model system, such as certain breeds of dogs, which have a macula-like "visual streak" that have been efficiently used to study severe retinal degenerations resulting from mutations in the RPE65 gene (Acland et al., 2001;Jacobson et al., 2005) and among others. While the existence of a dog with ABCA4-associated retinopathy was very likely due to the extensive genetic variability not only in the human ABCA4 locus but also in other mammals, the search was successful only very recently, when when several Labrador retrievers were identified and characterized as a KO for ABCA4 (Makelainen et al., 2019). The affected animals, homozygous for the deleterious, c.4176insC, p.(Phe1393Leufs*2), allele presented with a phenotype more closely resembling the ABCA4-associated retinopathy in humans (Makelainen et al., 2019), including visual impairment at 10 years of age, abnormal fundus images, a complete loss of ABCA4 protein, profound reduction of cone outer segments, and an approximately 50% reduction of photoreceptor nuclei in the affected retinae. The RPE autofluorescence in the affected animal, indicating lipofuscin accumulation, was ~7X higher compared to the unaffected dogs and a clear functional defect was detected by flash-electroretinography (Makelainen et al., 2019). In summary, the dogs with no functional ABCA4 closely resembled the human phenotype, although the disease severity was still less profound than in patients lacking ABCA4 (Tanaka et al., 2018).

Clinical trials
Given its relatively high prevalence as well as its disease course, ABCA4-associated retinopathy is an attractive target for therapeutic intervention, yet no approved therapy exists.
For several reasons, the eye (or retina) is an extremely suitable organ for the development and implementation of novel therapies, i.e. its easy accessibility, compartmentalized and immune-privileged nature, and the possibilities to measure potential therapeutic outcome non-invasively. Like for every other subtype of IRD, the chosen therapeutic strategy for ABCA4-associated retinopathy can range from mutation-specific approaches to more generally applicable cell replacement, mainly depending on the primary genetic defect, and the disease stage at the time of treatment (Vazquez-Dominguez et al., 2019). However, in particular for ABCA4-associated retinopathy, one of the most challenging aspects of implementing new treatments is the ability to measure therapeutic benefit. Below, we outline the various therapeutic strategies that are currently in clinical trials or in preclinical development, and discuss how disease progression can be accurately and quantitatively measured, with the ultimate goal to define useful clinical outcome parameters to prove therapeutic benefit.
As shown in Table 6, currently there are 16 clinical trials registered at http:// clinicaltrials.gov describing therapeutic intervention for ABCA4-associated retinopathy (combining the results of search queries 'ABCA4' and 'Stargardt' and selecting those containing a therapeutic intervention). One additional trial was only registered at https:// www.clinicaltrialsregister.eu/. These trials can roughly be divided into three categories, i.e. cell replacement, compound administration and gene augmentation (Fig. 9a). including (a combination of) retrobulbar, subtenon, intravitreal, subretinal or intravenous injection of these cells. In addition, the source of the stem cells varies, and includes bone marrow-derived cells as well as human embryonic stem cells (hESCs). The first study was started in 2011, with initial results reported in 2012, demonstrating interim safety and moderate efficacy in a single subject with ABCA4-associated retinopathy . Later, results of nine subjects in a dose-escalation study were published, again with no major safety issues, although some complications related to vitreoretinal surgery or immunosuppression were reported. In the majority of subjects, visual function appeared to improve in the treated compared to the contralateral eye . Delivery of hESC-derived RPE cells to the retina of ABCA4-associated retinopathy cases also was safe, yet no improvement in visual function could be measured (Mehat et al., 2018). For the other clinical trials aiming to assess the safety and efficacy of cell transplantation, no results have been published yet. In addition to the official clinical trial studies reported at http://www.clinicaltrials.gov, a number of other studies using cell replacement therapy for ABCA4-associated retinopathy have been published. In one study, four subjects received a graft of adipose tissue-derived mesenchymal stem cells that was bilaterally delivered between the choroid and the sclera (Oner et al., 2018). Improvements were found in visual performance as well as responses in multifocal ERG analyses, without any safety complications. Another study described the delivery of hESC-derived RPE cells, with safety and moderate efficacy reported up to one year after treatment (Song et al., 2015b). Overall, despite some moderate efficacy reported in a few studies, one should carefully take the pathophysiological mechanism of ABCA4-associated retinopathy into consideration when applying cell replacement. With ABCA4 being (mainly) expressed in photoreceptors, providing RPE cells (or cells destined to become RPE) will likely not have a long-term beneficial effect. Future research should thus be more directed towards the transplantation of cell sheets that contain both RPE and photoreceptor cells. Thus, although in essence, cell replacement therapy may hold great promise for the treatment of ABCA4-associated retinopathy, there are many variables that still need to be optimized, including selection of the optimal origin of stem cells, whether or not to differentiate cells ex vivo prior to transplantation and if so, until what stage, whether to deliver individual cells or cell sheets, and how to surgically deliver these cells. Finally, also the disease stage of the subject to be treated needs to be taken into account, and can have a major influence on the therapeutic outcome.

Compound administration therapy-
An attractive alternative to cell replacement therapy is the administration of compounds, each of which aims to modify either the physiological or the pathological pathways that are affected in ABCA4-associated retinopathy. In total, nine of these compounds have been or are currently being tested in human subjects (Table 6). Only for two of these, initial results have been published (MacDonald and Sieving, 2018;Piccardi et al., 2019). In one study, saffron was used, a compound harbouring carotenoid constituents that are able to counteract oxidative stress.
Oral administration of either saffron or placebo (in a crossover trial design) to a total of 31 cases with ABCA4-associated retinopathy revealed that the supplement was well tolerated but on the short term did not seem to give any measurable improvement in visual function (Piccardi et al., 2019). Long-term studies are needed to investigate the potential therapeutic efficacy of this drug. The other published study also made use of oral delivery of a compound, namely docosahexaenoic acid (DHA), in a cross-over trial design with 11 subjects with ABCA4-associated retinopathy. DHA is a major polyunsaturated fatty acid present in high concentrations in the retina (Fliesler and Anderson, 1983) and important for retinal structure and function. It is believed that in ABCA4-associated retinopathy cases, DHA metabolism is altered. Overall, the administration of DHA did not lead to visual improvements in this relatively small group. Small adverse events were noted but it was concluded that they were unrelated to the drug (MacDonald and Sieving, 2018). MADEOS is the abbreviation for Macular Degeneration Omega-3 Study in which the efficacy of omega-3 fatty acids is tested in subjects with ABCA4-associated retinopathy or AMD. This trial however is still in its recruiting phase.
Another potential drug used for the treatment of ABCA4-associated retinopathy is 4 methylpyrazole (4-MP), an alcohol dehydrogenase inhibitor that can delay dark adaptation, at least in laboratory animals. In healthy individuals, intravenous administration did not seem to have an effect on dark adaptation and thus the potential therapeutic efficacy was questioned (Jurgensmeier et al., 2007). However, 4-MP could potentially halt or delay the processing of vitamin A derivatives and thereby prevent the formation of toxic lipofuscin, and was therefore also tested in ten subjects with ABCA4-associated retinopathy.
No trial results however were published despite the fact that the study was completed more than a decade ago. ALK-001 is also a molecule that aims to prevent the formation of lipofuscin, yet by a slightly different mechanism. In fact, ALK-001 is a deuterated form of vitamin A that is less capable of forming vitamin A dimers, and thereby toxic lipofuscin. The therapeutic potential of ALK-001 was demonstrated in a murine model of ABCA4-associated retinopathy, by showing a reduced A2E dimer formation and lipofuscin accumulation compared to age-matched wild-type mice (Charbel Issa et al., 2015). A phase 1 study, in which ALK-001 was administered to 40 healthy adult volunteers, has been completed (NCT02230228), after which the phase II study, where 50 ABCA4-associated retinopathy cases were administered with either ALK-001 or placebo at a daily basis, was initiated. No results have been published to date. The last two compounds that are currently being tested in clinical trials both act at the level of the RPE. Emixustat hydrochloride (also known as ACU-4429) is a nonretinoid compound that can exert an inhibitory effect on one of the enzymes involved in the visual cycle, the RPE-specific 65 kDa protein isomerase, encoded by the RPE65 gene. A phase 1 placebo-controlled study exploring the safety of Emixustat that was administered to healthy individuals at a daily basis (NCT00942240) did not reveal any systemic adverse events, although ocular side effects were observed in the majority of participants (Kubota et al., 2014). However, these effects were considered mild and transient, disappearing after completion of the trial, and warranted further testing in subjects with impaired visual function. Initially, Emixustat was tested in subjects with geographic atrophy associated with dry AMD, with results that are supportive of the anticipated mode of action, yet not having led to spectacular results in terms of visual improvement (Dugel et al., 2015). As illustrated in Table 6, Emixustat is now also being evaluated as a drug for ABCA4-associated retinopathy, in two independent trials. The phase 2 trial that aimed to look at the potential short-term benefit has just been completed; however, no results have been reported. The multicentre phase 3 trial in which 162 subjects are planned to be enrolled is currently ongoing, with the aim to monitor the long-term effect of this new drug. One compound that is planned to be studied in human subjects is Zimura, an aptamer that can inhibit the activity of complement factor C5 (Drolet et al., 2016). Zimura was first tested in subjects with age-related macular degeneration but now also for cases with ABCA4-associated retinopathy. Finally, a compound called Soraprazan, a fast-acting inhibitor of H + ,K + -ATPase (Simon et al., 2007), is currently being investigated in a multi-national, multicenter, double-masked, placebo-controlled proof of concept trial. Previously, Soraprazan was used for the treatment of patients with gastroesophageal reflux disease. After the discovery that this compound could remove lipofuscin from the RPE in monkeys (Julien and Schraermeyer, 2012), its potential for the treatment of ABCA4 associated retinopathy is now also being studied, with no data reported so far.
Together, the various compounds that have been tested so far in subjects with ABCA4 associated retinopathy overall can be considered safe, yet none of them succeeded to demonstrate a high therapeutic effect. The advantage of many of these compounds, i.e. the fact that they can be administered orally, also can be considered a disadvantage, since this delivery route does not always allow the active compound to reach a sufficient concentration within the retina. Further research to identify the optimal route of administration, the concentrations needed to exert their effect without causing adverse events, and the ideal treatment regime, is needed to reveal the true therapeutic potential of the aforementioned compounds, as well as those that are currently in preclinical development.

Gene augmentation therapy-
The approval of gene augmentation therapy for another subtype of inherited retinal disease that is caused by bi-allelic mutations in RPE65, has provided hope for many visually impaired individuals, and is paralleled by the development of similar therapeutic strategies for other genes underlying these disorders (Vazquez-Dominguez et al., 2019). The vast majority of trials assessing gene augmentation for retinal diseases employs adeno-associated viruses (AAVs) to deliver the wild type cDNA of the gene that is mutated. The one and only exception is gene augmentation therapy for ABCA4 disease, foremost because the size of wild-type ABCA4 cDNA (6.8 kb) surpasses AAV's cargo capacity. A phase 1/2 clinical study employing lentiviral delivery of ABCA4 cDNA (therapeutic molecule SAR422459) has commenced in 2012, but recently has been terminated, without any data of efficacy published. However, in 2016, Parker et al. reported on test-retest variability for a number of clinical outcomes in subjects participating in this trial (Parker et al., 2016). The exact reasons for terminating this trial are unknown; a second trial employing the same therapeutic molecule however is still active and recruiting (Table  6).
Summarizing the clinical trials so far, one can conclude that there are several different treatments for ABCA4-associated retinopathy under development, yet none of them so far revealed itself to be the ideal therapeutic intervention. Not only the therapeutic drug itself needs to be further optimized, also the high degree of variability observed in different clinical tests (between and within a subject) warrants an improved trial design. This is further illustrated by the many pre-clinical studies that are currently ongoing, to identify novel therapeutic strategies as well as improved clinical diagnostics, as further outlined below.

Preclinical studies
Besides the clinical studies mentioned above, numerous therapeutic strategies are currently being assessed in preclinical models (examples are illustrated in Fig. 9b). Many of these studies focus on the replacement of stem cell-derived RPE cells, as summarized by Sachdeva et al. (Sachdeva and Eliott, 2016). In addition, novel compounds that e.g. interfere with lipofuscin accumulation, or affect intracellular trafficking of mutant ABCA4 protein are constantly being investigated, a few examples of which are provided below. Administration of BPN-14136, a non-retinoid antagonist of the retinol-binding protein 3 involved in the visual cycle, to Abca4 −/− mice inhibited bisretinoid synthesis while not altering the rate of the visual cycle, demonstrating its potential for the treatment of ABCA4-associated retinopathy (Racz et al., 2018). A systems pharmacological approach identified a number of G-protein coupled receptors that showed improved photoreceptor cell survival and function in Abca4 −/− mice (Sabirzhanova et al., 2015). VX-809, a molecule previously found to be efficacious for rescuing trafficking of the CFTR protein in cystic fibrosis, also increased the membrane localization of some mutant ABCA4 proteins in cultured HEK293T cells (Liu et al., 2019), although care is warranted when deciding which allele can be amenable for which type of therapy, as the pathogenic mechanism for many variants is still not entirely understood.
As no data on potential efficacy have been reported for lentiviral gene augmentation, alternative strategies to deliver ABCA4 cDNA to the retinal cells were developed. One delivery strategy uses nanoparticles, whereas others make use of a dual AAV approach. Han and colleagues developed non-viral nanoparticles, and demonstrated that upon sub-retinal delivery in Abca4 −/− mice, Abca4 transgene expression persisted up to eight months after injection, and resulted in reduced lipofuscin accumulation in the treated animals (Han et al., 2012). With dual AAV technology, cDNA fragments that exceed the cargo capacity of a single AAV can be split into two halves and each packaged into a separate AAV, with additional sequences that allow reconstitution of the complete cDNA once inside the target cell (Trapani, 2019). Dual AAVs have been used to deliver ABCA4 cDNA to the (cone enriched) porcine retina (Trapani et al., 2014(Trapani et al., , 2015, as well as the retina of Abca4 −/− mice, with reduced lipofuscin accumulation and/or correction of the autofluorescent phenotype measured in a number of separate studies (Dyka et al., 2019;McClements et al., 2019;Trapani et al., 2015). Although no clinical trials exploring the safety and efficacy of these gene augmentation strategies have been reported, it is expected that these will soon commence. Yet, it is important to realize that the transduction efficiency of both dual AAV vectors and non-viral vectors generally is lower when compared to classical AAV vectors.
Finally, there are also several mutation-specific therapies under development for ABCA4 associated retinopathy. These strategies mainly employ antisense oligonucleotides (AONs) and are focussed on those variants that affect pre-mRNA splicing of ABCA4. AONs are relatively small and versatile RNA molecules that can be synthesized in such a way that their sequence is complementary to their target pre-mRNA. So far, AONs have mainly been used to block PE inclusions caused by deep-intronic mutations. The first AON described for a retinal disease targets a recurrent deep-intronic mutation in CEP290, and was initially tested in lymphoblastoid and fibroblast cells derived from patients homozygously harbouring this mutation (Collin et al., 2012;Gerard et al., 2012). Following further demonstration of efficacy in a humanized mouse model  and in iPSC-derived retinal organoids (Dulla et al., 2018;Parfitt et al., 2016), a clinical trial was initiated, with recently reported positive interim results achieved by intravitreal delivery of AONs in subjects with CEP290-associated LCA (Cideciyan et al., 2019). As stated in section 6.4, an increasing number of deep-intronic ABCA4 variants that result in PE inclusion have been identified over the last few years. Using midigene splice assays and patient-derived cells, the ability of AONs to prevents the aberrant PE inclusions caused by several different ABCA4 variants has been demonstrated Bauwens et al., 2019;Garanto et al., 2019;Sangermano et al., 2019), although the therapeutic efficacy so far was only assessed at the RNA level. Besides deep-intronic variants, there are also several ABCA4 variants reported that are located in or near exons and result in altered pre-mRNA splicing, i.e. exon skipping or exon elongation. Also for these variants, AONs could be employed, e.g. by blocking splice silencer motifs that are created by such variants, or alternative cryptic splice sites that are used. Overall, the versatile nature of AONs render these attractive therapeutic molecules, yet the fact that some of the variants targeted by AONs are rare, or sometimes even ultra-rare, prevent a broad applicability of this therapeutic strategy for a large group of ABCA4-associated retinopathy cases. In addition to AON-based splicing correction, genome editing (or RNA editing) approaches are booming, and have the potential to correct mutations regardless of the size of the gene, or the position within the genome. By employing CRISPR/Cas9 technology with the homology-directed repair (HDR) pathway, one can replace e.g. single basepair substitutions with the wild-type nucleotide and in that way repair mutations (Hsu et al., 2014;Yanik et al., 2017). Other recently described strategies include e.g. base pair-editing systems (Billon et al., 2017;Rees and Liu, 2018). It is expected that these strategies will soon also be applied for the mutation-specific correction of ABCA4 variants.

Clinical outcome measures
The selection of appropriate clinical outcome measures for ABCA4-associated retinopathy should be predicated on a deep understanding of the factors underlying progression. The natural history of ABCA4-associated retinopathy consists of multiple trajectories that is largely determined by an individual's genotype. Effective outcome measures should be generalizable across patients but at the same time, account for their respective differences.

Understanding the parameters of atrophy growth
Monitoring the size of the atrophic lesion has been the primary outcome measure of most ABCA4-associated retinopathy clinical trials to date. At face value, atrophy progression appears to be a logical endpoint as it is the most discernible anatomic manifestation of cellular degeneration and historically, the primary endpoint in clinical trials for AMD (Klein et al., 2010;Knudtson et al., 2004;Lindblad et al., 2009). Although convenient, following the AMD model may not be an effective strategy for several reasons: (1) the two conditions share very little pathophysiological and demographic overlap, and (2) studies have already reported extensive differences in their respective rates and spatial correlation with the functional scotoma Lindner et al., 2017;Sunness and Steiner, 2008). Furthermore, the atrophic process in ABCA4-associated retinopathy is difficult to uniformly define, as it is an evolving entity that progressively spans multiple layers of the retina over time across the natural history of the disease (Fig. 1). The sensitivity of SW-AF imaging provides a reasonable assessment of cellular level involvement (Sunness et al., 2006) and as such, the investigators of ProgStar have proposed distinguishing lesions decreased autofluorescence (DDAF) or questionably decreased autofluorescence (QDAF) (ProgStar Report No. 9) (Strauss et al., 2017a(Strauss et al., , 2017b. Nevertheless, an analysis of reported rates of "DDAF" across the several prominent articles still uncovered a range of variability -0.94 ± 0.87) mm 2 /year (range 0.2-2.13 mm 2 /year (Chen et al., 2010) . Fujinami et al. stratified patients according to "AF type", or the background heterogeneity of flecks, and found that the rate of atrophy enlargement (RAE) (or "DDAF") in a more severe background exhibit significantly increased rates of growth (Fujinami et al., 2013a). More recent studies have looked to different modes of optical coherence tomography (OCT) to monitor the rate of lesion progression or even the delineation of new lesion types such as "dark atrophy" on OCT angiography (Pellegrini et al., 2016) allowing for a more restrictive definition of atrophy by observable changes in the anatomical layers visible on OCT scans (Arepalli et al., 2018;Cai et al., 2018;Kong et al., 2019;Park et al., 2015;Tanna et al., 2019). Increasing interest is being garnered in this area particularly with advancements in scan resolution, wide-field capture and analytical capabilities such as en face analysis (Alabduljalil et al., 2019;Greenstein et al., 2017;Melillo et al., 2016;Sodi et al., 2016).

Targeting the lipofuscin biomarker
Most ABCA4-associated retinopathy patients exhibit a spatially homogenous and localized increase in RPE lipofuscin because of ABCA4 dysfunction (Cideciyan et al., 2004).
Augmented RPE lipofuscin confers a Vermillion hue to the fundus under white light imaging and can obstruct fluorescence emanating from the underlying choroid in during fluorescein angiograms (FA) giving rise to the distinct "dark" or "silent" choroid in up to 62% of patients (Anmarkrud, 1979;Ernest and Krill, 1966;Fish et al., 1981). Despite the historical utility of FA, its role in the evaluation of STGD1 has been increasingly limited as it is no longer performed on a routine basis in favor of newer, less invasive imaging modalities. Further development of the confocal scanning laser ophthalmoscope (cSLO) introduced various modes of imaging that allow for the capture of an inherent autofluorescence emitted by photoreceptor and RPE fluorophores belonging to the family of bisretinoids that includes N-retinyl-N-retinylidene ethanolamine (A2E), its cis isomers and other related compounds (Fishkin et al., 2005;Kim and Sparrow, 2018;Parish et al., 1998;Wu et al., 2009;Yamamoto et al., 2011). This autofluorescence signal is emitted at wavelengths between 520 and 800 nm and can be captured by the standard SW-AF (448-nm excitation) and the more recently developed ultra-wide field AF (532-nm excitation). The components of photoreceptor/RPE lipofuscin contribute most predominantly to the 488-nm excitation wavelength (SW-AF) where its distribution and pattern in the normal retina has been well described (Delori et al., 1995;Delori et al., 2001;von Ruckmann et al., 1995von Ruckmann et al., , 1997. Deviations in SW-AF intensity and texture can be observed in nearly all inherited retinal degenerative diseases where a decrease or absence of SW-AF is generally indicative of a disruption in the tissue architecture in regions not obstructed by retinal vessels or luteal macular pigment in the fovea. Early attempts at measuring SW-AF (Cideciyan et al., 2004;Lois et al., 2004) lead to the development of quantitative autofluorescence (qAF) by Delori and colleagues (Delori et al., 2011) wherein the SW-AF intensities in non-normalized images (acquired without histogram stretching) were calibrated to the fluorescence intensities of an internal reference mounted within the cSLO and captured simultaneously to compensate for variations in laser power and detector gain. Using this method, Burke et al. verified the increase in autofluorescence and additionally, reported differences amongst genotypes Sparrow et al., 2020). In a similar study, the use of qAF was shown to be effective in differentiating ABCA4-associated retinopathy from non-ABCA4-associated retinopathy (masquerading) bull's eye maculopathy phenotypes (Duncker et al., 2015b). Near-infrared autofluorescence imaging (NIR-AF), which employs an excitation of signal at wavelength 787 nm, generates a signal corresponding to RPE and choroidal melanin (Keilhauer and Delori, 2006). This has also recently emerged as an effective modality for inherited retinal diseases and ABCA4-associated retinopathy (Cideciyan et al., 2007;Cukras et al., 2012;Duncker et al., 2013Duncker et al., , 2014Kellner et al., 2009;Sparrow et al., 2015). Interestingly, Paavo et al. reported quantitative increases in the NIR-AF (787-nm) signal in ABCA4-associated retinopathy as well as Abca4 −/− mice, corroborating results from an earlier study (Charbel Issa et al., 2013) prompting further revisions to either the interpretation of the anatomical origins of the NIR-AF signal or ABCA4-associated pathophysiology (e.g. the role of melano-lipofuscin) (Paavo et al., 2018). Additionally, NIR-AF has the practical advantage of being low luminance for ease in acquisition and possible safety consideration.
Effort should be allocated towards tracking other dynamic lipofuscin-related features such as the evolution of flecks patterns and the "leading disease front" .
However, as is the case with qAF, data acquisition and analysis may require a high-level expertise and computational proficiency that can impede its adoption as a routine method in the clinic or treatment trials. Nevertheless, the quantitation of autofluorescence holds promise as an effective outcome measure in ABCA4-associated retinopathy, particularly for monitoring the effects of lipofuscin-targeted therapies. Further studies evaluating the longitudinal sensitivity of qAF and a more precise understanding of the cellular origins of autofluorescence may further support its adoption as a primary outcome measure for clinical trials.

Mapping the range of functional loss
Impairment of visual function is the predominant symptom of ABCA4-associated retinopathy and all inherited retinal diseases and as such, developing effective ways to track it over time is essential to assessing the disease natural history and response to therapies. The most direct approach is measurement of best-corrected visual acuity (BCVA). Numerous systematic protocols have been developed and this is the standard used by all clinical trials (Table 6). Many studies have documented broad trends in BCVA progression and its relationship to other clinical features in ABCA4-associated retinopathy Collison and Fishman, 2018;Ergun et al., 2005;Kong et al., 2016;Parodi et al., 2015;Querques et al., 2008;Testa et al., 2012Testa et al., , 2014, however as a method, it is perhaps most susceptible to technical limitations such as measurement bias, repeatability as well as confounders that are specific to ABCA4-associated retinopathy such as the variable status of the fovea (Bax et al., 2019b;Collison et al., 2019;Nakao et al., 2012;van Huet et al., 2014) and shifting of the preferred retinal locus (PRL) into consideration (Bethlehem et al., 2014;Greenstein et al., 2008;Krishnan and Bedell, 2018;Schonbach et al., 2017aSchonbach et al., , 2018. The full-field electroretinogram (ffERG) is a powerful tool in the diagnostic repertoire of a retinal disease clinic. By measuring the summation of electrical activity generated by cones and rods across the entire retina (separately and combined), its application has been highly effective in broadly classifying severity and predicting the prognosis of individuals with ABCA4-associated retinopathy (Lois et al., 2001). The drawback to this generalized approach however, is an insensitivity to subtle changes in the macula, which disproportionately contributes to the aggregate electrophysiological response of the entire retina. The multifocal electroretinogram (mfERG) is more precise as local ERG responses can be recorded simultaneously from many defined regions of the retina. Few studies to date have examined its suitability for monitoring ABCA4-associated retinopathy progression (Kuniyoshi et al., 2014;Sisk and Leng, 2014;Tosha et al., 2010), which may be due to the requirement of fixation stability in a disease where most patients have profound central vision loss. The same issue is encountered with conventional methods of static and kinetic perimetry in the mapping of visual fields as they are performed under free-viewing conditions (fixation is not tracked) required but not monitored increasing the likelihood of obtaining false positives (Acton and Greenstein, 2013).
Methods circumventing the fixation requirement include full-field stimulus testing (FST) which provides a psychophysical measure of luminance. Although FST elicits a full field response, Collison et al. found that cone-mediated thresholds in ABCA4-associated retinopathy correlate well with locally defined changes such as visual acuity and macular thickness (Collison et al., 2014). Furthermore, the improvements on FST were instrumental in demonstrating the efficacy of voretigene neparvovec (AAV2-hRPE65v2, Luxturna) in RPE65-associated LCA (NCT00999609) supporting its suitability as an endpoint for aggressive therapies in patients with advanced phenotypes (Jacobson et al., 2009;Maguire et al., 2019). For a more localized assessment, microperimetry (MP) allows for the quantification of mesopic and scotopic visual sensitivity at user-defined points on the retina and fundus tracking which addresses unstable fixation. Studies assessing the scotoma of ABCA4-associated retinopathy conclude that it is reliable in tracking longitudinal changes and is highly correlated with changes in BCVA (Cideciyan et al., 2012;Schonbach et al., 2017b).

Future research
Despite significant advances in deciphering all aspects of ABCA4-associated retinopathy, there are still several areas requiring attention in the future. Despite its worldwide occurrence, the prevalence of ABCA4-associated retinopathy is not yet known. The 1:8000-10,000 estimate, which everybody cites, comes from the textbook chapter by Blacharski in 1988(Blacharski, 1988, which states: "We have seen this condition much more commonly than retinoblastoma, which has been estimated at 1 in 15,000 live births. Fundus flavimaculatus is not as common as retinitis pigmentosa, which has a prevalence of no more than 1 in 5000. We have roughly estimated the incidence to be between 1 in 8000 and 1 in 10,000." It is obviously not a scientific way to determine a disease prevalence. The actual prevalence of ABCA4 -associated retinopathy is very difficult to estimate due to enormous clinical and genetic heterogeneity, variable age of onset, and (still) incomplete genetic data. Most of the genetic information for ABCA4-associated retinopathy thus far has been collected from individuals of European descent. To fully appreciate the allelic heterogeneity for ABCA4-associated retinopathy and to better understand the differences in genetic background that may influence the expression of this disease, more emphasis should be put on the sequence analysis of ABCA4-associated retinopathy cases from non-Caucasian populations.
The wide functional spectrum of ABCA4 variants, from extremely hypomorphic to deleterious, leaves us with many ABCA4 variants for which the penetrance is still not known. This has a profound effect on genetic testing and genetic counseling of patients. For example, the disease causality for the p.(Asn1868Ile) allele is still not widely accepted, and this is complicated by the discussion about its penetrance. At least 10% of the entire ABCA4-associated retinopathy remains not genetically confirmed if the genetic testing laboratories do not detect and/or report the p.(Asn1868Ile) variant. Considering the p. (Asn1868Ile) allele "benign" has also a detrimental effect on family planning. Screening of large cohorts of familial cases of ABCA4-associated retinopathy will continue helping to decipher the penetrance of selected ABCA4 alleles on a specific genetic background. We estimate that ~10% of ABCA4-associated retinopathy probands carries a causal deep intronic variant and that ~2% of the cases carries a SV. Routine diagnostics therefore cannot only rely on the sequence analysis of the coding regions but also needs to sequence the entire ABCA4 genomic locus. To establish pathogenicity of novel deep-intronic variants, NCSS variants and rare synonymous coding variants should be tested using in vitro splice assays. The accuracy of these assays in the future can be improved by using cell lines that mimic the splicing processes in the normal retina. Another emerging important area is the concept of genetic modifiers in the ABCA4 locus and in the genome; i.e., both cisand trans-modifiers for ABCA4-associated retinopathy. We have started to determine some of the cis-modifiers (see the example of the p.[Gly863Ala,Gly863del] allele above), but most of these remain obscure. Trans-modifiers will be even harder to identify, but with the complete sequencing of the ABCA4 locus Sangermano et al., 2019;Zernant et al., 2014b) and the entire exome and genome, these are likely to be found through the analysis of very large familial cohorts, which are, fortunately, possible to obtain even for a relatively rare Mendelian disease. More than 50% of ABCA4 variants are missense mutations. For frequent missense mutations, we can predict their severity quite accurately based on genotype-phenotype correlations. Most of the rare missense variants however are classified as 'variants of unknown significance'. If a crystal structure would be determined at high resolution for the ABCA4 protein, proper functional studies could be conducted.
Currently, functional studies relying on animal models and in vitro data, can reach correct conclusions in some cases; however, in many cases these do not correlate with genetic and clinical studies. One of the best examples is the c. 2588G>C, p.[Gly863Ala,Gly863del] variant, which has been shown to result in a dual effect and also in significant defects in in vitro assays Sangermano et al., 2018), but is not penetrant based on clinical/genetic studies unless it is in cis with the p.(Asn1868Ile) allele. Determining the high-resolution protein structure would allow investigating the predicted effect of this and many other missense ABCA4 variants with great precision.
The relatively high prevalence of ABCA4-associated retinopathy as well as its progressive nature have led to an enormous attention from academia as well as industry with regard to the development of molecular and cellular therapies. Although ABCA4 is expressed in photoreceptors, its pathogenic effects are manifest primarily in the RPE (Cideciyan et al., 2004), where accumulation of toxic bisretinoids  occur from the perpetual shedding and subsequent phagocytosis of outer segments (Steinberg et al., 1977;Young, 1967;Young and Bok, 1969). This prevailing disease model, while in many ways consistent with general mechanism of ABCA4 dysfunction, may not be that straightforward. Rod discs are enclosed, self-containing structures and thus bisretinoids generated within would indeed eventually be phagocytosed, along with the rest of the shed outer segment, by RPE. Cone lamellae, however, are contiguous with its plasma membrane and therefore, accumulating bisretinoids may be retained in other cellular compartments, where their toxic effects can be expressed. Consistent with this model is the histopathological observation of lipofuscin-like autofluorescence in the cone inner segments of the retina of a patient with fundus flavimaculatus (ABCA4-associated retinopathy) (Birnbach et al., 1994). Clinical imaging studies in patients have largely demonstrated the trend that RPE cell death precedes photoreceptor cell death (Cideciyan et al., 2007;Duncker et al., 2014;Greenstein et al., 2017;Kellner et al., 2009); however most, if not all of these studies, report notable exceptions. For instance, in a study of 24 patients (45 eyes), Duncker et al. reported that zones of RPE atrophy (reduced AF) are statistically larger in NIR-AF compared to SW-AF in the same eye, although a subgroup of younger patients with relatively severe, early-onset disease exhibited larger areas of photoreceptor-attributable EZ loss compared to reduced NIR-AF . Similarly, AO-SLO imaging of two STGD1 patients revealed abnormal spacing of rods and cones in otherwise unaffected areas (Song et al., 2015a). Whether photoreceptor cell degeneration precedes or follows RPE loss in ABCA4 associated retinopathy remains unresolved. Given the complex role of ABCA4 dysfunction in both cell types, it is likely that the sequence of incapacitation varies in accordance with disease stage, location and overall severity of progression. There are still many challenges ahead, both in terms of identifying and optimizing the right therapeutic drug and finding the right clinical parameters to accurately measure disease progression as well as therapeutic benefit. In the end, there will likely not be a 'one-strategy-fits-all' treatment; rather every individual case may need its own tailor-made therapeutic intervention, based on their genetic profile as well as the stage of their disease.
In conclusion, clinical, molecular genetics and therapeutic studies have resulted in a large body of knowledge regarding the fascinating complex ABCA4-associated retinopathy. At the same time, new research avenues have opened to unravel the unexplained differences in disease expression between ABCA4-associated retinopathy cases, both within and between families. Long-read sequencing technologies will allow phasing of different genomic variants to establish the role of cis-modifiers and enable us to appreciate in great detail normal and abnormal RNA splicing. A full comprehension of the molecular genetic causes and molecular mechanisms of ABCA4-associated retinopathy will allow us to develop 'personalized' therapies to slow down or stop disease progression.
misfolding and the pathogenesis of ABCA4-associated retinal degenerations. Hum. Mol. Genet 24, 3220-3237. [PubMed: 25712131]             Location of deep-intronic variants in ABCA4. Left part shows variants located near exons that result in exon skipping or exon elongation. Right part shows deep-intronic variants that invariably result in the generation of pseudoexons (see Table 4).

@
The wild-type midigene shows 14% natural 3 exon skipping; # For variant c.2588G>C, a rough quantification was based on Sanger sequence traces as the two splice products (3-nt difference) could not be separated; $ Variants with in-frame small amino acid insertions that may not act deleterious in protein function.
The wild-type and a mutant (4-nt insertion) fragment co-migrate and together constitute 4% of the total RNA. For variants with multiple effects at the mRNA, the most prevalent product is listed first.
Prog Retin Eye Res. Author manuscript; available in PMC 2020 December 15.