What is developmental about developmental prosopagnosia?

Developmental prosopagnosia (DP) is characterised by difﬁculties recognising face identities and is associated with diverse co-occurring object recognition difﬁculties. The high co-occurrence rate and heterogeneity of associated difﬁculties in DP is an intrinsic feature of developmental conditions, where co-occurrence of difﬁculties is the rule, rather than the exception. However, despite its name, cognitive and neural theories of DP rarely consider the developmental context in which these difﬁculties occur. This leaves a large gap in our understanding of how DP emerges in light of the developmental trajectory of face recognition. Here, we argue that progress in the ﬁeld requires re-considering the developmental origins of differences in face recognition abilities, rather than studying the end-state alone. In practice, considering development in DP necessitates a re-evaluation of current approaches in recruitment, design, and analyses. ©

Developmental prosopagnosia (DP) is a neurodevelopmental condition affecting individuals' ability to recognise face identities (Hecaen & Angelergues, 1962;McConachie, 1976).Unlike acquired prosopagnosia (AP), where individuals develop good face recognition but, following brain damage, experience difficulties in recognising faces, DP manifests in the absence of brain damage.Individuals with DP struggle to identify people by their face and instead often recognise based on non-facial features, including voice, hairstyle, clothes, and gate (Fine, 2012).As face recognition is fundamentally social, DP could be disabling, resulting in social anxiety, feelings of embarrassment, guilt, and feeling of failure (Davis et al., 2011;Yardley et al., 2008).
The predominant research question in DP is the extent to which the condition affects recognition of just faces (domainspecific) or a broader range of visual categories (domaingeneral).In a large meta-analysis Geskin and Behrmann (2018) concluded that 80% of DP individuals had associated object agnosia, supporting a domain-general account.Importantly, recognition difficulties were heterogenous, affecting a range of diverse object categories, including cars, scenes, houses, flowers, glasses, shells, guns, horses, tools, and novel objects.However, studies have failed to find recognition difficulties on a group-level with the exact same object categories in DP (Barton, et al., 2019;Bate et al., 2019;Epihova et al., 2022;Esins et al., 2014;Palermo et al., 2017), but see (Gray et al., 2019;Rivolta et al., 2017).In other words, the categories DPs seem to find difficult are remarkably inconsistent across individuals.In the absence of group-level effects, the associations of difficulties within an individual might simply be viewed as noise.Garrido et al. (2018) pointed out that the estimated 80% co-occurrence between DP and object agnosia is likely overestimated; the same criteria would result in 50% of a control sample having object agnosia, which seems unlikely.Although we agree that the actual estimate of co-occurrence is likely overestimated, it is important to note that the estimate is substantially higher in DP compared to controls.While object recognition problems might not be as disruptive to the daily lives of DPs, from a theoretical perspective the extent to which difficulties with faces and objects tend to co-occur can inform cognitive and developmental theories.The increased rate of associated object agnosia in DP suggests theoretically important associations evident for an individual, but not at a group-level.We argue that the reason for the dissociation of findings on individual and group levels is the heterogeneity of DP itself.Associated object recognition difficulties are heterogenous, resulting in the lack of a single, universally affected, object category.Consistent with a heterogenous group profile, some DPs exhibit marked recognition difficulties, whereas others perform well above average with the same object categories (Barton et al., 2019;Bate et al., 2019).Similarly, even when an individual's recognition is affected for more than one visual category, the association between the magnitude of difficulties for the different categories is weak (Biotti et al., 2017;Zhao et al., 2016), suggesting the lack of a single domaingeneral mechanism responsible for all recognition difficulties both within and between DPs.The group-level characteristic of DP e higher rate of co-occurrence between face and objects agnosia, coupled with the absence of any one specific category of agnosia on a group-level e is a natural consequence of searching for group-level effects in a heterogeneous population.In short, investigating heterogenous difficulties with small samples will always yield inconsistent trends in group-level effects.
The face and object agnosia debate can be considered a part of a broader question: are neurodevelopmental difficulties ever really that specific?In addition to the increased rate of object agnosia, DP may be part of a broader heterogeneous profile of neurodevelopmental difference.For example, individuals with autism (Gray & Cook, 2018;Minio-Paluello et al., 2020;Ventura et al., 2023), aphantasia (difficulty visualising images) (Dance et al., 2023), and developmental coordination disorder (Maw et al., 2023) are more likely to experience face recognition difficulties compared to the general neurotypical population.A recent report with a relatively large sample (n ¼ 115) of self-reported DP individuals demonstrated that ~60% of DPs reported at least one cooccurring neurodevelopmental difficulty (Svart & Starrfelt, 2022).DPs' pattern of co-occurrence was itself diverse with numerous developmental difficulties, including navigation, memory, math difficulties and aphantasia.Mirroring the heterogeneity in object recognition abilities, the pattern of cognitive strengths and difficulties was heterogenous, with many individuals reporting difficulties within the same domains that others reported strengths.

Considering development when investigating cognitive mechanisms
The higher co-occurrence rate between DP and object agnosia, coupled with the heterogeneity of associated difficulties in those who experience DP, are intrinsic features of developmental conditions more generally, where co-occurrence of difficulties is the rule, rather than the exception (Gilger & Kaplan, 2001;Gillberg, 2010).There is a growing movement within developmental science towards understanding the mechanisms of neurodevelopmental differences, without tailoring the recruitment, assessments, and analyses to specific diagnostic categories (Astle & Fletcher-Watson, 2020).While transdiagnostic approaches have successfully been implemented to capture neurodiversity (Astle et al., 2019;Mareva & Holmes, 2019), DP and face recognition abilities more generally, have hitherto been omitted.Although there is a general consensus that DP is heterogenous, this has not been incorporated in the recruitment, design, and analyses of studies.A starting point is the consideration of heterogeneity in the inclusion and exclusion criteria of sample selection.Although currently there are no gold-standard inclusion criteria, the majority of studies select individuals with DP based on single-cases performing 2 SD of a control mean score on self-report questionnaires and at least one objective face recognition task.Furthermore, a commonly adopted exclusion criterion is the presence of any other neurodevelopmental condition, most prominently autism (Barton et al., 2019;Bate et al., 2019;Biotti et al., 2017;Epihova et al., 2022;Marsh et al., 2019;Ulrich et al., 2017).The theoretical justification behind these strict inclusion and exclusion criteria is that only the "purest" cases of face recognition difficulties are wanted, and that, in these pure samples, face processing difficulties are attributable to the same underlying mechanism.However, this inevitably exaggerates the internal homogeneity of the selected DP sample, while inflating differences to the broad population from which selected DPs were sampled from.Put simply, it does not reflect the real-life presentation of DP.More to the point, even in these so-called "pure" samples we are still unable to find universal shared mechanisms, suggesting that this approach fails even on its own terms.Within a transdiagnostic framework, the existence of "pure" face-selective difficulties would itself be an empirical question, rather than an inevitable consequence of the selection criteria.
Adopting a more transdiagnostic approach to recruitment would increase the heterogeneity of the sample, allowing a better characterisation of those who struggle to recognise faces, beyond testing specific visual recognition deficits.Within this framework, scores on face recognition tasks alongside developmentally relevant measures, including those sensitive to co-occurring difficulties, could be included as variables.Rather than operating strict exclusion criteria, we argue that individuals should initially be recruited based on self-reported face recognition difficulties.For example, the current cut-off on the PI20 (>65) is a well-validated, quick, and inexpensive way to diagnose DPs whose face recognition difficulties are impactful to their daily life, similarly to how mental health or neurodevelopmental conditions are diagnosed based on self-or others-report (Shah, Gaule, & Gaigg et al., 2015).There is a moderate correlation between self-report and face recognition performance (Gray et al., 2017;Livingston & Shah, 2018;Shah, Gaule, & Sowden et al., 2015;Shah, Sowden, & Gaule et al., 2015;Tagliente et al., 2023).This has been taken as evidence that self-report provides only moderate insight into face recognition abilities.However, one can easily reframe the apparent discrepancy, such that face recognition tasks only moderately explain real-world abilities.This point has recently been introduced by Burns et al. (2022) who provided an in-depth discussion of the pitfalls of relying on face processing task data over subjective reports, as the former may be a poor reflection of real-world abilities.Moreover, the authors highlighted that the imperfect testeretest reliabilities of the Cambridge Face Memory Test (CFMT) means that the diagnostic status of a DP might change on any given day that they are tested, suggesting that cognitive tests should not be used to exclude cases.Moreover, self-reported symptoms of face difficulties are identical between DPs who reached the diagnostic threshold on a face recognition task and those who did not (Burns et al., 2022).
When designing studies, it is valuable to include the heterogeneity of co-occurring difficulties as variables of interest, rather than nuisance variability.Future efforts should shift the focus to the variability of performance in DP by investigating the patterns of co-occurring difficulties within the sample, focusing on a broad range of assessments with a diverse set of visual categories and potential neurodevelopmental difficulties.A promising tool for uncovering neurodevelopmental profiles of difficulties is the use of datadriven clustering analyses to explore emerging groups based on profiles across measures (Astle et al., 2019).As a first step this can quantify the rate of co-occurrence of difficulties as well as explore the associations of difficulties beyond diagnostic categories.While characterising the diverse pattern of difficulties is a crucial step, a comprehensive exploration of DP will further involve investigating the mechanisms of how and when the pattern of difficulties emerges and changes throughout development.The constellation of difficulties within and beyond visual recognition may indicate shared vulnerabilities in the structural and functional brain organization that may serve as better future targets for identifying underlying mechanisms than "pure" diagnostic categories.

Considering development when investigating neural mechanisms
The occipital face area (OFA), fusiform face area (FFA) and superior temporal sulcus (STS) are thought to comprise the core face system responsible for visual processing of faces (Haxby et al., 2000).The face-selectivity found in these regions has focused the majority of research efforts to these facespecific areas as probable candidates of the neural basis of face recognition differences in DP (Manippa et al., 2023).This has produced highly inconsistent results e while some studies show reduced face-selectivity in DPs (Furl et al., 2011;Hadjikhani & De Gelder, 2002;Jiahui et al., 2018), others find face-selectivity comparable to that of neurotypical individuals (Avidan et al., 2005;Hasson et al., 2003;Rivolta et al., 2014).While the core face network is specialised for face perception, the multiplex task of recollecting and linking familiar identities and person-related knowledge requires the neural architectures to increasingly integrate signals from an extended network of regions across the brain (Davies-Thompson & Andrews, 2012;Gobbini & Haxby, 2007;Haxby et al., 2000).Indeed, DPs have difficulties with recognising face identities, rather than faces on a categorical level (recognising faces from other categories).Face recognition on a categorical level and identity recognition need not necessarily be co-localized.Some areas may not exhibit face selectivity, yet still play a role in face identity recognition (Anzellotti & Caramazza, 2014).The core and extended regions of the developed face network as well as their contribution for specific aspects of face processing are well-understood.For example, the interaction between the core and extended face processing regions has been linked to specific aspects of face perception, such as identity, expression, and biographical information (Collins & Olson, 2014).A promising hypothesis for poorer face recognition in DP is that it results from a disruption in the integration of the core with the extended regions of the face network (Fox et al., 2008;Sokolowski & Levine, 2023).Lower structural and functional integration between the core and extended face processing networks, particularly the anterior temporal lobe, have been linked with the identity recognition difficulties in DP (Avidan et al., 2014;Grossi et al., 2014;Herbet et al., 2018;Rosenthal et al., 2017;Thomas et al., 2009).
Despite containing "development" in its name, neural theories of developmental prosopagnosia and those aiming to explain the development of face recognition, rarely inform each other.However, phenotypic variation in face recognition abilities likely emerges as a function of developmental processes that impose or overcome constraints in the organization of brain circuits relevant for face recognition.We argue that a propitious way forward in understanding the neural basis of DP would see the integration of face recognition and developmental theories.Unlike AP, resulting from core localised damage to an already developed system (Barton, 2008;Davies-Thompson et al., 2014), face recognition difficulties in DP emerge over the course of development and are likely shaped via multiple different neural routes e the principle of equifinality (Bishop, 1997;Cicchetti & Rogosch, 1996;Luyten et al., 2008).The developmental plasticity of face recognition is illustrated by the extreme scenario in which patients whose entire hemisphere was removed in childhood can still achieve 80% of the average neurotypical face recognition performance (Granovetter et al., 2022), regardless of which hemisphere was removed.This suggests that even though in adult neurotypical participants the right hemisphere is dominant for face processing (Kanwisher et al., 1997;Pitcher et al., 2007), the functional processing of faces can be reorganized during development.Similarly, development of atypical function does not necessarily mirror the cause for the loss of that function in an otherwise typically developed brain (acquired cases) (Bishop, 1997).While there is a growing need for translating neural theories of face recognition to DP, and c o r t e x 1 7 3 ( 2 0 2 4 ) 3 3 3 e3 3 8 neurodevelopmental conditions more broadly, it is also becoming clear that this translational bridge must be raised on a solid understanding of how face recognition develops in both neurotypical and neurodivergent populations.In other words, a first step towards understanding altered function, is to understand how the organization of the face network changes throughout typical development.Although the face network is well understood, a comprehensive mapping of the course of its development is still missing.Behavioural results suggest that the specialization for human face recognition is a gradual experience-dependent process not fully present at birth (Pascalis et al., 2002(Pascalis et al., , 2005)).Instead, face recognition abilities improve rapidly throughout childhood and adolescence, reaching their peak at about 30 years of age (Germine et al., 2011).How these behavioural dynamics relate to the development of the face network is currently unknown.
Large-scale open multimodal neuroimaging datasets (Bookheimer et al., 2019;Harms et al., 2018;Somerville et al., 2018) provide ideal means to construct the normative trajectory of the anatomical and functional connectome of the face network over the whole lifespan and pave the way for exploring structureefunction relationships and brainbehaviour associations in face recognition.For example, using this approach we can track how the face network gradually emerges, integrates extended regions, and consolidates existing connections as it develops, and establish how this relates to changes in typical face recognition thought the lifespan.Moreover, studying DP individuals in the context of normative trajectories will aid understanding of how and when deviation from the typical organization of the face network relates to individual differences in face recognition abilities and whether different patterns of deviation can converge on the same perceptual profile of difficulties.Adopting this approach has further implications for exploring co-occurring difficulties in DP based on the network architecture.
In conclusion, progress towards understanding cognitive mechanisms and neural correlates of DP requires reconsidering the developmental origins of differences in face recognition abilities, rather than studying the end-state alone.We have highlighted how the co-occurrence and heterogeneity of difficulties in DP is an intrinsic feature of developmental conditions, where co-occurrence of difficulties is the rule, rather than the exception.We argue that considering development when investigating cognitive mechanisms in DP would require a re-evaluation of current approaches in recruitment, design, and analyses.In practice, quantifying heterogeneity will require recruiting samples that better reflect the real-world presentation of face recognition difficulties, without exclusion of co-occurring difficulties.Finally, we advocate for investigating neural correlates of DP in the context of normative developmental trajectories of neural circuits involved in face recognition.