Comparing the experimental results with a questionnaire based self-assessment, we found that all of the CPs tested in this study have provided more or less accurate estimates of the severity of their deficit. The only case with a strong deviation between self-assessment and behavioral measurements is VK, whose deficit is of an amnestic type and extends to the recognition of individual objects.

Out of the 15 participants with CP tested in this study, seven show perceptual deficits in face processing (HG, FP, MR, EB, HE, SE, HB). In all of the seven cases the deficits extend to associative and/or mnestic aspects of face recognition. Four of the CPs only show mnestic deficits (HW, MG, HS, VK). The remaining four CPs evinced a more diffuse pattern of deficits. Although these cases don't fall into one of the three subtypes, three of them (JM, JF, LL) display an overall face recognition performance clearly below control average. Testing recognition performance for individual objects we found deficits of an apperceptive type in four CPs (HE,SE,HB,RK) and deficits of an amnestic type in one CP (VK).

Combining the results of long-term memory with prior results on perceptual and associative aspects of face and object recognition [44] allowed a more comprehensive characterization of CP. Overall, individuals in the CP group displayed a large heterogeneity in test performance, which is in line with previous studies [35] , [54] . However, the observed heterogeneity is different from unstructured random variability but aligns with the separation of acquired prosopagnosia into distinct subtypes: apperceptive, associative, or amnestic.

First, we assessed deficits in long-term recognition memory for faces and objects in a controlled test with a retention interval of one year. Out of the eleven CPs that participated in this experiment, eight showed clear deficits in recognizing faces (FP, MR, EB, MG, VK, HE, SE, HB), and an additional two showed performance clearly below the control average (HS, JF). This decrease in face recognition performance was mainly due to an increased miss rate among CPs. With the exception of one CP participant (VK) the deficit was selective to faces and didn't extend to recognizing individual non-face objects. In addition to this controlled test, we conducted a famous face test as it is commonly used to assess long-term recognition memory. In the famous face test only four of the CPs (HW, HS, HE) performed significantly below control average. A more detailed analysis of differences in performance in famous face recognition between controls and CPs revealed a decrease in the positive influence of media consumption (i.e. prior exposure or training).

In real life situations, the recognition of another person is often a mutual process: both persons simultaneously try to recognize each other and tend to communicate their results to the counterpart, e.g. by greetings or changes in mimic. As a positive identification signal can be given by any of the two persons involved, a compensatory, evasive strategy for CPs would be to simply wait until such a signal is given by the opponent, and later on explain the delayed response by inattentiveness (28). In an experimental setting, this compensatory strategy would lead to more misses (false negatives) than false positives, which was what we observed in the controlled one-year recognition memory experiment.

In this study, we also present results on a controlled assessment of long-term recognition memory in CP for both faces and individual objects. The setup chosen provided the same degree of familiarization for each participant. We found that under this condition of equal familiarization long-term face recognition deficits in CP were more pronounced than in a test of famous face recognition. Moreover, by conducting an additional experiment with the same setup but using individual non-face object stimuli (Nike sneakers) we could rule out more general amnestic deficits unrelated to face processing.

So far, experiments of long-term recognition memory in CP have been mostly restricted to tests of familiar face recognition, e.g. using pictures of family members [55] , or tests of famous face recognition. Tests of familiar face recognition are necessarily adapted to each specific case, which renders comparisons of test results across different CP cases difficult. Tests of famous face recognition assume an apriori familiarization of participants with the faces through publicly available images and videos. The degree of familiarization depends to a large extent on each participants' social interest in the famous persons tested. Here, we used publicly available images of famous persons and found a strong influence of the degree of media consumption on recognition performance. This influence was also present among CPs, although to a lesser extent. Many CPs showed a normal recognition performance, which might be due to the presence of non-facial cues (hair, clothing, …) in the test images used.

A characterization of processing deficits

In acquired prosopagnosia, behavioral heterogeneity has been mostly explained by differences in the extent and location of the brain damage causing the deficit [5]–[7], [49]. In contrast, the deficit in CP is manifested as an endpoint of an inborn (endogenic), selective developmental impairment. Thus, while AP is caused by damaging a mature, functional system of face recognition, individuals with CP never evolve a fully functional face recognition system in the first place. Their deficit has to be interpreted in the context of a different developmental trajectory [56] into a mature but dysfunctional system.

Under normal developmental conditions, rudimentary abilities to discriminate between individuals can be observed already after the first days of age [57], [58] and continue to develop rapidly [59], [60]. Irrespective of the exact developmental processes underlying the subsequent functional specialization of cortical regions into a mature neural system for face recognition [61]–[63], this specialization presumably leads to an alignment between cortical location and functional process [52], [64]–[66]. Damage inflicted to a specific region can therefore lead to restricted deficits conditional on the interconnectedness and interdependence of the distributed processing [5]–[7], [49]. The spatial localization of specific functions in the ventral occipitotemporal cortex has enabled a characterization of processing deficits in AP based on the extent of the lesioned cortical regions.

Following the definitions proposed by Lissauer [2] AP has been divided into three subtypes:

apperceptive - caused by a dysfunctional perceptual encoding,

associative - resulting from deficits in the association of encoded percepts with individual objects, and

amnestic - the deficit is restricted to accessing semantic information for known objects,

see also Damasio et al. [5] who use slightly different denominations. This symptomatic categorization of deficits parallels a functional modularization proposed in conceptual models of intact face recognition [50]–[53], and it aligns with the location of underlying cortical lesions roughly along a caudal-rostral axis [5], [6], [49].

In contrast, a similar categorization of CP into subtypes based on neurophysiological differences has remained elusive so far. First indications of neuroanatomical differences point to a volumetric reduction of the anterior fusiform gyrus [67], a region involved in more associative and mnestic aspects of face recognition, and a reduced structural connectivity in the ventral occipito-temporal cortex [68], a region involved in more apperceptive aspects of face recognition. Contrasting CPs and controls, Garrido et al. [43] found a positive correlation between face identification performance and gray matter volume in the left superior temporal sulcus and right fusiform gyrus, and a negative correlation between object recognition and volume in the lateral occipital cortex. In a large sample study, Dinkelacker et al. [69] found widespread areas of diminished gray matter density in the bilateral lingual gyrus, correlated with face memory success, as well as in the the right middle temporal gyrus and the dorsolateral prefrontal cortex. Irrespective of possible structural differences, findings of differences in functional MRI activations in core regions of the face processing system have been mixed, both using classical localizer paradigms [25], [54], [70] as well as adaptation paradigms [54], [71], [72].

In the following, we discuss the phenotypical heterogeneity observed in this study in three different contexts. The proposed characterization of CP subtypes is, firstly, interpreted with respect to cognitive and computational models of face recognition and, secondly, contrasted with possible underlying differences in the neuroanatomical structure. Thirdly, behavioral heterogeneity is discussed in the more general context of compensatory processing and evasive coping strategies.

Apperceptive prosopagnosia. Apperceptive prosopagnosia refers to a deficit in the early perceptual encoding of face images [5], [6], [49]. During perceptual encoding visual information is extracted from different locations with a certain efficiency, and the total information is obtained by integrating spatially across different locations and temporally across inspection time (cf. featural and holistic processing, see below). Deficits in perceptual encoding can occur either at the level of local efficiency or at the level of spatial integration. The process of perceptual encoding proceeds more slowly over time and/or reaches a saturation level that is too low for successful recognition. Thus, it was assumed, that in order to encode a sufficient amount of information, CPs need to inspect images longer, either due to a decrease in encoding efficiency or because they engage in a deliberate, series of attentional or fixational shifts to extract information at different spatial locations [44]. This hypothesis of longer inspection times in CP is consistent with previous studies of an increase in reaction times [27], [33], and reports of a more pronounced deficit under tachystocopic presentation in acquired prosopagnosia [73], [74], dispersed gaze behavior in CP [42], and a face specific increase in the recruitment of frontal-areas in CPs compared to controls [54]. Here, we assessed purely perceptual aspects in a same-view recognition task using frontal images by measuring reaction times for correct responses and presentation times needed to perform at an 80% correct level. Apperceptive deficits in face recognition were clearly present for seven CPs. For three of these seven, deficits extended to shoe recognition, and for one participant with CP the deficits were present only for shoe recognition. In acquired prosopagnosia, apperceptive deficits are usually associated with damage to superior and inferior parts of the right posterior visual association areas [5]. It is assumed that under normal circumstances face-responsive regions in the inferior occipital gyrus are responsible for perceptual encoding and provide input to later areas of the core system of face processing [52]. Damage to early visual areas often also induces non-face related processing deficits, specifically deficits in within-object spatial coding after lesions to ventral occipito-temporal areas [17]. This decreased selectivity of apperceptive deficits was also observed in this study: The two CPs with clear overall deficits in object recognition (HE, SE) were cases of an apperceptive prosopagnosia, and conversely out of the seven apperceptive CPs, three also had apperceptive deficits with object recognition.

Associative prosopagnosia. In contrast to a purely perceptual deficit in perceptual encoding, associative deficits are characterized by a dysfunctional association of encoded percept and facial identity [5], [6], [49]. Under normal circumstances the information about the uniqueness of a face image, that is extracted during perceptual encoding, is associated with a specific facial identity such that future encounters of the same face lead to a recognition of this identity. To distinguish associative from perceptual deficits, that already occur at the level of matching identical images, we assessed recognition accuracy in a delayed recognition task using rotated and differently illuminated test images. In previous studies using rotated [27], or rotated, differently illuminated, and noised stimuli [40], CPs consistently evinced worse performance than controls. In this study, we observed associative deficits in face recognition for seven CPs, and borderline performance in four CPs. However, as already noted by Lissauer [2] the distinction between associative and apperceptive types of agnosia is anything but clear; he suspected that all observable cases of prosopagnosia would be rather a mixture between the two extremes. Comparing apperceptive and associative deficits, we found that all CPs with an apperceptive deficit also show deficits or below average performance in association. The reverse doesn't hold, as two CPs (LL, JF) show associative deficits but average performance w.r.t. perceptual aspects. This pattern of perceptual deficits leading to associative deficits aligns with models of hierarchical information processing [50]–[53]: Dysfunctional processing at lower areas can evince deficits in functions that are normally associated with processing in higher areas. Based on a classic model of face recognition and prosopagnosia [50]. Young and Burton [75] simulated associative prosopagnosia as a disconnection between face recognition units, where the encoded percept or facial memory is stored, and personal identity nodes. According to this model, if early perceptual encoding is performed in inferior occipital regions, e.g. the occipital and the fusiform face area (OFA and FFA), and facial memories are stored in anterior temporal regions, associative prosopagnosia might be caused by a disconnection of the tracts connecting the posterior occipitotemporal regions with more anterior temporal regions, e.g. the inferior longitudinal fasciculus (ILF). A contrary view posits that facial memories are already stored in the FFA and associative prosopagnosia results from a disconnection between OFA and FFA (see [49], for a discussion). Recent studies on brain abnormalities in CP have observed a reduction in the structural connectivity of ventral visual areas that is most prominent in the right ILF [68], and a decrease in grey matter volume in the right fusiform and inferior temporal gyri [43]. In both studies the reductions were correlated in magnitude with deficits in face recognition performance.