Data for this study were obtained from forty-six participants (16 PCA, 12 lvPPA, 18 EOAD; Table 1) in the Massachusetts General Hospital Frontotemporal Disorders Unit longitudinal cohort, including the Primary Progressive Aphasia program (Sapolsky et al., 2014) and Posterior Cortical Atrophy program (Putcha et al., 2018), which are affiliated with the Massachusetts Alzheimer’s Disease Research Center. All participants received a standard clinical evaluation comprising a comprehensive neurological and psychiatric history and exam and structured informant interviews following the Clinical Dementia Rating (CDR) protocol, and a separate neuropsychological battery including the National Alzheimer’s Coordinating Center (NACC) Uniform Data Set (UDS) version 2.0 or 3.0 battery. For each patient, clinical diagnostic formulation was performed through consensus conference, with each patient being classified based on all clinical information as having mild cognitive impairment or dementia (global clinical status), and then each patient’s cognitive-behavioral syndrome being diagnosed according to standard diagnostic criteria (Dickerson et al., 2017; Wong et al., 2019). Twenty-five patients met diagnostic criteria for PCA (Tang-Wai et al., 2004; Crutch et al., 2017), 23 patients met criteria for lvPPA (Gorno-Tempini et al., 2011), and 18 patients met criteria for EOAD with single- or multi-domain amnestic or non-amnestic syndromes (Ossenkoppele et al., 2015; Townley et al., 2020). Though having early onset of symptoms and having a PCA or PPA clinical syndrome are not mutually exclusive, in this study we have made them distinct. That is, if an individual patient had and early onset of symptoms (younger than age 65), and also met criteria for PCA or PPA, they were thus classified into those syndromic categories. If they had early onset of symptoms and did not meet criteria for PCA or PPA, they were categorized simply as “EOAD,” indicating likely multidomain impairment. The patient sample was further restricted to those participants who had imaging or cerebrospinal fluid (CSF) biomarker status consistent with AD. Some patients underwent neuroimaging sessions involving structural MRI, flortaucipir (FTP) PET, and amyloid (PiB or FBB) PET scans. Aβ positivity was determined by a combination of visual read and mean amyloid PET signal extracted from a cortical composite region of interest according to previously published procedures (Rabinovici et al., 2010; Villeneuve et al., 2015; Cho et al., 2023). Determination of tau and neurodegeneration positivity was conducted by visual read using internal methods similar to published work (e.g., Rabinovici et al., 2011; Fleisher et al., 2020; Sonni et al., 2020). Other patients underwent CSF sampling with results indicating abnormally low levels of CSF amyloid-β and abnormally high levels of CSF total tau and hyperphosphorylated tau (Shaw et al., 2009). This resulted in a final patient sample size of 16 A + T + N + PCA, 12 A + T + N + lvPPA participants, and 18 A + T + N + EOAD. Individuals were not recruited into this study if they had a primary psychiatric or other neurologic disorder including major cerebrovascular infarct or stroke, seizure, brain tumor, hydrocephalus, multiple sclerosis, HIV-associated cognitive impairment, or acute encephalopathy. This work was carried out in accordance with The Code of Ethics of the World Medical Association (Declaration of Helsinki) for experiments involving humans. All participants and their informants/caregivers provided informed consent in accordance with the protocol approved by the Mass General Brigham Human Research Committee Institutional Review Board in Boston, Massachusetts.

Auditory naming was assessed using the Auditory Naming Test (Hamberger and Seidel, 2003; Hirsch et al., 2016, 2021). The ANT is a 50-item test which provides oral descriptions to one word responses. The performance metrics of interest included: the total number of correct responses on immediate recall (IR), total correct given up to 20 s to respond, reflecting the “tip of the tongue” phenomenon (IR + TOT), and total correct with phonemic cues (IR + TOT + PC). The number of items correctly named were analyzed, as well as compared to cutoff scores (Hirsch et al., 2016) and normative age- and education- stratified data (Hirsch et al., 2021). Performance differences between AD syndromic groups were investigated using one-way analysis of variance, with post-hoc independent sample t-tests to verify between group differences. Statistical significance was set to a threshold of p < 0.05. Primary hypothesis-driven analyses were conducted with no corrections for multiple comparisons applied. Statistical analyses were conducted in IBM SPSS Version 29.0 (Armonk, NY, USA). In order to compare the ANT to other commonly used measures of word retrieval, we also administered select trials of the Controlled Oral Word Association Test (Spreen and Strauss, 1991), with the measures of interest being the total number of correct words produced in 1 min trials to the letter cues “F,” “A,” and “S,” and to the category cue of “Animals.”

Tests scores from the remainder of the neuropsychological battery are presented in Table 2 to describe the cognitive profile of the participants in the current study. This battery included Digit Span Forward and Backward (longest spans), Auditory Addition and Subtraction, the UDS3 Craft Story memory encoding and delayed recall task (Craft Story in UDS3), the Word Picture Matching test from the UDS3 FTLD Module, the California Verbal List Learning 2nd Edition Short Form (CVLT-II-SF) test, Spatial Span Forward and Backward (longest spans), the Birmingham Object Recognition Battery (BORB) Single and Overlapping Object Identification tests, and the Visual Object Spatial Perception (VOSP) Number Location test. To determine the mechanistic specificity of contributions to word retrieval performance observed on the ANT, we conducted bivariate correlations with digit span and verbal fluency, as well as the VOSP number location and Spatial Span Backward, with the a priori hypotheses that ANT performance would be related to performance on digit span and verbal fluency, and not to the latter two tests of visuospatial cognition. In order to establish that word retrieval difficulty observed on the ANT was not merely reflective of semantic knowledge loss, we also examined performance on a test of semantic association, as well as the relationship between the Word Picture Matching test and ANT performance.

All participants in the final sample received a structural 3D T1-weighted scan at MGH. Scans were acquired using a Siemens Trio 3T scanner (Siemens Medical Systems). T1 image volumes were examined qualitatively by a cortical surface-based reconstruction and analysis of cortical thickness using FreeSurfer version 6.0.¹ The general procedures for this processing method have been described in detail and applied and validated in a number of publications and presentations; the technical details can be found in select manuscripts (Dale et al., 1999; Fischl and Dale, 2000; Fischl et al., 2002, 2004).

All but 3 participants (1 PCA and 2 lvPPA) underwent ¹¹C- Pittsburgh Compound B (amyloid) and FTP PET scans, which were spherically registered to align each individual’s cortical surface between PET and MR scans. The ¹¹C-PiB PET scans were acquired with an 8.5 to 15 mCi bolus injection followed immediately by a 60-min dynamic acquisition in 69 frames (12 × 15 s, 57 × 60 s). FTP PET scans were acquired 80 to 100 min after the bolus injection of ∼10.0 mCi of FTP (4 × 5 min frames). All PET data were acquired using a Siemens/CTI (Knoxville, TN) ECAT HR + scanner (3D mode; 63 image planes; 15.2 cm axial field of view; 5.6 mm transaxial resolution and 2.4 mm slice interval). Data were reconstructed and attenuation corrected; each frame was evaluated to verify adequate count statistics; interframe head motion was corrected prior to further processing. Visual inspection confirmed accurate registration between anatomical and PET volumes. To evaluate the anatomy of PET binding, each individual’s PET data set was rigidly co-registered to the subject’s MPRAGE data using SPM8 (Wellcome Department of Cognitive Neurology, Function Imaging Laboratory, London). Similar to a previous report, ¹¹C-PiB PET data were expressed as the distribution volume ratio (DVR) with the cerebellar gray matter as a reference (Becker et al., 2011), where regional time-activity curves (TAC) were used to compute regional DVRs for each ROI using the Logan graphical method applied to data from 40 to 60 min after injection. FTP standard uptake value ratio (SUVR) images were derived with whole cerebellar gray matter as a reference region. PET data were not partial volume corrected and were performed using geometric transform matrix as implemented in FreeSurfer stable release version 6.0. The remaining three participants were deemed to be A + T + N + positive based on clinically obtained CSF samples as described above.

To determine if performance on the ANT was related to cortical atrophy in our hypothesized regions, we conducted whole cortical surface general linear models (GLM) for the effects of the task performance on cortical thickness at each vertex point on the cortical surface. Using methods we have previously published (Dickerson et al., 2008; Makaretz et al., 2017; Xia et al., 2017; Putcha et al., 2020), we used age-, education-, and sex- adjusted performance scores and thus did not control for these demographic factors again in our cortical thickness GLM analysis. Follow-up analysis ensured that cortical thickness was not related in any significant way to any of these demographic factors. GLM analyses was implemented using the mri_glmfit utility within FreeSurfer version 6. Given our specific a priori hypotheses, an uncorrected statistical threshold of p < 0.001 was set.

A total of 46 A + T + N + symptomatic individuals across the atypical syndromic spectrum of AD (16 PCA, 12 lvPPA, 18 EOAD) were included in this study. See Table 1 for complete sample characteristics. Consistent with the diagnostic criteria of “early onset” AD, we observed that our EOAD group were significantly younger than our PCA (t = 4.1, p = 0.0001) and lvPPA (t = 3.7, p = 0.001) groups. The average age of PCA and lvPPA individuals were comparable. There were no syndromic group differences in years of education or sex. All but one patient included in these analyses were rated at the stage of mild cognitive impairment or mild dementia. Some individuals were mildly symptomatic in their primary domains of impairment (i.e., language) though this was not captured on the traditional CDR scale thus giving them a CDR of 0, although their CDR supplemental language box score was 0.5 or 1. Participants from all groups were at the relatively same stage of cognitive decline; the mean MoCA scores for the whole sample was 15.3 out of 30, with no statistically significant difference among the three syndromic groups (p > 0.7). All individuals underwent comprehensive neuropsychological testing, including assessments of Attention and Executive Functioning, Language, Memory, and Visuospatial Skills. See Table 2 for a full neuropsychological characterization.

We measured three variables of the Auditory Naming Test: the number of correct items correctly named within 2 s (immediate recall or “IR”), the number of additional items correctly named given extra time (tip-of-the-tongue or “TOT”), and the number of items correctly named with a phonemic cue (“PC”). Comparing the IR to the IR + TOT scores, we observed that all groups benefitted from extra time. All benefitted further from phonemic cuing. On all three variables, the lvPPA group performed most poorly followed by the EOAD group and finally the PCA group (see Table 3). We focused the remainder of our analyses on the IR + TOT score, as previous publications (Hirsch et al., 2016) have identified this variable as most sensitive to measuring naming impairment in the dementia population. We found significant differences between groups on IR + TOT performance, such that lvPPA performed worse than EOAD (t = 1.9, p = 0.03) and worse than the PCA group (t = 4.7, p < 0.001). The EOAD group also performed worse than the PCA group (t = 2.3, p = 0.04; Figure 1). Notably, all groups performed in the normatively impaired range even with the extra time (Figure 2A), with z-scores falling below the 1st percentile (z < −3.0). Using the age- and education- stratified impairment cutoffs published by Hirsch et al. (2016), we discovered that 83.3% of individuals with lvPPA and EOAD, respectively, and more than half (56.3%) of individuals with PCA are considered impaired (Figure 2B). Finally, all participant groups benefitted from phonemic cuing, with PCA patients performing within normal limits on this fully cued condition of the ANT.

To better understand the cognitive processes contributing to the ability to perform an auditory naming task, we tested our a priori hypotheses that auditory naming would be related to auditory-verbal working memory as well as goal-directed speeded lexical retrieval. We identified the Digit Span Backward test as a reliable measure of working memory and tests of verbal fluency—to both letter and category cues—as measures of speeded lexical retrieval. Across our whole sample, we found that ANT IR + TOT performance was strongly correlated with Longest Digit Span Backward (r = 0.52, p = 0.0003; Figure 3A), letter fluency (FAS; r = 0.48, p = 0.006), and category fluency (Animals; r = 0.62, p = 0.000006; Figure 3B). Of note, letter and category fluency performance was highly correlated (r = 0.65, p = 0.0001), and letter fluency did not explain unique variance over and above category fluency in a multiple regression model (p > 0.1). We also found that ANT IR + TOT performance was correlated to a global screening measure of cognition (MOCA; r = 0.58, p = 0.00002). We then assessed the relationship between ANT performance and tests of visuospatial cognition, which should not be correlated, to determine the specificity of relationships to our hypothesized cognitive constructs. We chose two visuospatial tests from our battery, the VOSP Number Location Test and Longest Spatial Span Backward, to represent visual attention and space perception. We found that ANT IR + TOT performance was not correlated with either the Longest Spatial Span Backward (r = −0.2, p = 0.3; Figure 3C) nor the VOSP Number Location Test (r = −0.6, p = 0.7; Figure 3D). Lastly, we examined the association between ANT performance and a non-verbally mediated test of semantic knowledge, Word Picture Matching. All participant groups performed well on this test (Table 2) indicating intact semantic knowledge. Across the whole group, we did not observe a significant relationship between performance on the Word Picture Matching test and ANT IR + TOT (r = 0.4, p = 0.3), suggesting that the semantic word retrieval impairment we observe on the ANT is not explained by any loss of the actual semantic knowledge regarding the items being tested. These associations did not change in strength when demographic and clinical factors (e.g., disease severity) were controlled for.

We next tested our a priori hypotheses regarding the neuroanatomical correlates of auditory naming by conducting a whole-cortex general linear model predicting ANT IR + TOT performance. We combined our whole sample together for these analyses in an effort to capitalize on the heterogeneity in cognitive profile and neurodegeneration across syndromic groups. We found that ANT performance was correlated with circumscribed atrophy in the left anterior lateral temporal lobe extending posteriorly into the middle temporal gyri, as well as in the inferior frontal gyrus (Figure 4; p < 0.001). We did not observe any atrophy in the visual association cortices or frontoparietal networks as has been observed with traditional tests of visual confrontation naming.