Northwestern Anagram Test-Italian (Nat-I) for primary progressive aphasia

Abstract

Objectives

To test the ability of the Northwestern Anagram Test-Italian (NAT-I) to distinguish between the non-fluent/agrammatic (nfv-) and phonological/logopenic (lv-) variants of primary progressive aphasia (PPA), and to determine the relationship between NAT-I variables and brain integrity in PPA patients.

Methods

13 nfvPPA and 8 lvPPA patients underwent the 44-item-version of NAT-I and brain MRI. The NAT-I was also administered to six patients with the semantic variant (sv) PPA to sample performance in cases with no grammatical deficits. Performances were recorded and compared between patient groups. Receiver Operating Characteristic curve analysis assessed the ability of NAT-I to discriminate nfvPPA and lvPPA. The correlation between anatomical changes and NAT-I variables were assessed. A shortened (22-item)-version of NAT-I was also tested for classification ability.

Results

Participants with NfvPPA performed more poorly than lvPPA patients on canonical and non-canonical sentences. NAT-I non-canonical sentence and total scores achieved the highest diagnostic accuracy in discriminating the two patient groups (area under the curve: .93 and .91, respectively). SvPPA participants showed performances similar to lvPPA. NAT-I variables correlated with the integrity of the left inferior frontal gyrus and the body of the corpus callosum. The NAT-I 22-item-version total and non-canonical sentences scores reached diagnostic accuracy comparable to the full version.

Conclusions

The NAT-I, in particular the measure of non-canonical syntax, is an effective tool for distinguishing nfvPPA and lvPPA patients and correlated with the integrity of crucial brain regions implicated in syntactic processing. The 22-item-brief version of NAT-I is suitable for clinical practice and research.

1. Introduction

The current consensus guidelines for the diagnosis of primary progressive aphasia (PPA) recognize three variants: nonfluent/agrammatic (nfvPPA), mainly characterized by agrammatism and apraxia of speech; semantic (svPPA), with a main impairment of naming and single word comprehension; and logopenic/phonological (lvPPA), characterized by deficits in single word retrieval and in repetition of sentences and phrases (Gorno-Tempini et al., 2011). Beyond the impact on language features, these variants differ in terms of the pathological substrates (Spinelli et al., 2017) and distribution of cerebral damage (Gorno-Tempini et al., 2004). Aside from the available Alzheimer’s disease (AD) positron emission tomography (PET) and cerebrospinal fluid (CSF) biomarkers typically used in research studies, the heterogeneous neuropathological substrates (more frequently frontotemporal lobar degeneration [FTLD]-tau in nfvPPA, FTLD-TDP in svPPA, and AD in lvPPA) cannot be precisely diagnosed in vivo (Spinelli et al., 2017).

The clinical differentiation among the three variants is also often difficult. It becomes specially challenging between the nfvPPA and lvPPA cases, since both subtypes may present with frequent pauses and phonological paraphasias, although the nature of their disturbances affects different levels of the language system (Mack et al., 2015; Mesulam, Wieneke, Thompson, Rogalski, & Weintraub, 2012; Wilson et al., 2010). Together with apraxia of speech, the presence of agrammatism is the hallmark symptom of the nfvPPA profile, which is associated with post mortem tauopathy, and is crucial for the differential diagnosis of PPA variants (Gorno-Tempini et al., 2011; Grossman et al., 1996; Ogar, Dronkers, Brambati, Miller, & Gorno-Tempini, 2007; Thompson et al., 2013; Thompson & Mack, 2014). Grammatical competence is difficult to assess in the clinical setting, mainly when patients have severe impairments of speech production (Ash et al., 2013; Ballard et al., 2014; Mack et al., 2015; Mesulam et al., 2012; Wilson et al., 2010). Furthermore, although neuroimaging has been recognized to be powerful in supporting the PPA diagnosis, the initial distinctive functional and anatomical features (the left posterior fronto-insular involvement in nfvPPA cases; the left anterior temporal lobe in the svPPA; and the left posterior perisylvian or parietal involvement in the lvPPA) can be borderline in the early phase of the disease, or may be lost as degeneration progresses and converges over time.

In order to overcome patient difficulties in speech production, but also word-finding disturbances, word comprehension and/or reduced working memory, a testing method based on re-ordering of scrambled sentences, the Northwestern Anagram Test (NAT), was developed for the clinical assessment of production at the sentence (syntactic) level (Thompson, Weintraub, & Mesulam, 2012; Weintraub et al., 2009). The NAT requires the sequential ordering of individual word cards presented in scrambled order into meaningful simple and complex sentences that match action picture (Thompson et al., 2012; Weintraub et al., 2009). For the present study, we developed the NAT-Italian (NAT-I),based on an adaptation of the NAT for the Italian, a morphologically complex language. Such a tool was not available for testing Italian patients and was highly needed since Italian nfvPPA cases show prominent patterns of agrammatic production (Menn, Obler, & Miceli, 1990).

The main aim of this study was to assess the ability of the NAT-I to differentiate nfvPPA and lvPPA cases (compared to svPPA patients who show no grammatical impairments), and determine the relationship between NAT-I scores and brain cortical and white matter (WM) integrity in PPA patients. We also developed a shortened version of the NAT-I to be used in clinical practice and research. We hypothesized that, due to a selective presence of syntactic impairment in nfvPPA patients, these cases would perform more poorly than lvPPA on the NAT-I, mainly for the sentences with a complex structure. In contrast, lvPPA patients, whose difficulty with complex sentences reflects a working memory disorder, were expected to show impairment on tests of syntactic comprehension, but not on the NAT-I. These differences might improve accuracy in the diagnosis of these two PPA variants, and further, be related to the integrity of crucial cortical and WM regions implicated in syntactic processing.

2. Methods

We report how we determined our sample size, all data exclusions, all inclusion/exclusion criteria, whether inclusion/exclusion criteria were established prior to data analysis, all manipulations, and all measures in the study. This cross-sectional study was approved by the Local Ethical Committees on human studies and written informed consent from all subjects was obtained prior to their enrolment. The sample size was not determined a priori; however, we specified the power analysis in the Result section. No data were excluded according to inclusion/exclusion criteria reported below. Inclusion/exclusion criteria were established prior to data analysis and we report all manipulations, and all measures in the study. No part of the study procedures and analyses was pre-registered prior to the research being conducted.

2.1. Participants

A total of 27 PPA patients (13 nfvPPA, 8 lvPPA, 6 svPPA) were prospectively included into the study (Table 1). Inclusion criteria were: clinical diagnosis of PPA based on Mesulam’s criteria (Mesulam, 2001) and diagnosis of PPA subtypes (nfvPPA, lvPPA or svPPA) according to current international criteria (Gorno-Tempini et al., 2011); right-handedness; native Italian-speakers. Subjects were excluded if they had: known FTLD-related genetic mutations; significant medical illnesses or substance abuse that could interfere with cognitive functioning; any other systemic, psychiatric, or neurological illnesses; other causes of focal or diffuse brain damage, including cerebrovascular disorders at routine MRI. Patients received a comprehensive evaluation including structured history and neurological examination, neuropsychological testing, an extensive battery of language tests, and MRI. Clinical assessment was performed by experienced neurologists blinded to MRI results. Diagnosis was based on clinical judgement after considering history, all available neurologic, cognitive, and language data, and visual assessment of routine structural MRI. When available, clinical diagnoses were supported by FDG-PET. CSF biomarkers were obtained from 7 (54%) nfvPPA, 4 (67%) svPPA, and 7 (88%) lvPPA cases.

Table 1 –

Demographic, clinical and cognitive features of PPA patients.

	nfvPPA	lvPPA	svPPA	p nfvPPA versus lvPPA	p nfvPPA versus svPPA	p lvPPA versus svPPA
N	13	8	6
Age at visit, years	71.1 ± 5.3 (62–79)	72.6 ± 7.5 (60–81)	64.1 ± 2.2 (62–68)	.66	.03	.08
Sex, M/F	5/8	3/5	3/3	1.00	1.00	1.00
Education, years	10.3 ± 5.7 (5–22)	14.0 ± 2.6 (11–17)	10.0 ± 4.7 (5–18)	.16	.86	.16
Age at onset, years	68.6 ± 5.6 (59–78)	68.6 ± 7.9 (54–77)	60.8 ± 2.9 (58–65)	.61	.02	.06
Disease duration, years	2.5 ± 1.3 (1–5)	3.9 ± 1.5 (2–6)	3.2 ± 1.7 (1–4)	.15	.24	.37
ADL	5.8 ± .4 (5–6)	6.0 ± .0 (6–6)	6.0 ± .0 (6–6)	.45	.45	1.00
IADL	89.6 ± 20.5 (38–100)	96.4 ± 9.4 (75–100)	83.3 ± 21.9 (50–100)	.51	.51	.49
CSF, Aβ42*	775.9 ± 185.3 (516–972)	426.0 ± 95.7 (350–627)	816.3 ± 199.4 (557–1041)	.01	.71	.02
CSF, T-tau*	300.7 ± 189.6 (125–665)	521.7 ± 284.9 (313–999)	200.8 ± 56.4 (132–270)	.07	.34	.03
CSF, p-tau*	46.6 ± 6.8 (39–57)	89.9 ± 44.2 (46–154)	37.5 ± 2.9 (34–41)	.02	.02	.02
Global cognition
MMSE	23.7 ± 3.9 (17–29)	26.0 ± 1.4 (23–28)	24.7 ± 2.7 (20–28)	.46	.60	.46
Memory
Digit span, forward	4.3 ± .8 (3–6)	5.2 ± .8 (4–6)	5.5 ± .5 (5–6)	.04	.01	.42
Spatial span, forward	3.1 ± .8 (2–4)	4.3 ± .5 (4–5)	4.6 ± .9 (4–6)	.04	.03	.56
RAVLT, immediate recall	24.9 ± 10.4 (13–44)	21.3 ± 8.7 (13–31)	20.0 ± 3.4 (16–26)	.83	.83	.92
RAVLT, delayed recall	5.4 ± 4.4 (0–12)	3.8 ± 3.0 (0–7)	1.3 ± 2.0 (0–5)	.47	.22	.28
Rey’s figure, recall	9.3 ± 6.1 (4–24)	9.4 ± 3.0 (4–12)	7.7 ± 5.3 (4–16)	.46	.46	.46
Executive functions
Phonemic fluency	6.8 ± 5.6 (0–20)	20.3 ± 10.5 (9–34)	18.5 ± 11.2 (2–28)	.01	.06	.48
Semantic fluency	18.2 ± 9.1 (0–34)	27.3 ± 6.7 (18–40)	12.8 ± 6.0 (8–21)	.04	.14	.02
Clock Drawing Test	6.5 ± 3.5 (0–10)	7.0 ± 3.6 (0–10)	3.6 ± 4.5 (0–9)	.79	.46	.46
Attentive Matrices	29.5 ± 15.0 (1–52)	34.5 ± 10.1 (24–52)	49.2 ± 8.1 (36–58)	.57	.02	.04
TMT-BA	141.3 ± 37.5 (91–181)	179.0 ± 89.0 (78–280)	137.0 ± 34.7 (98–187)	1.00	1.00	1.00
Digit span, backward	2.4 ± 1.6 (0–4)	3.7 ± .6 (3–4)	3.8 ± .8 (3–5)	.25	.25	.87
RCPM	22.7 ± 7.3 (11–33)	25.9 ± 4.1 (21–31)	24.5 ± 7.2 (13–34)	.72	.72	.72
Visuospatial abilities
Rey’s figure copy	20.0 ± 8.1 (7–35)	26.3 ± 2.6 (23–30)	25.9 ± 4.0 (22–31)	.19	.19	.78
Language
Confrontation naming	46.2 ± 6.8 (38–63)	50.0 ± 14.5 (38–71)	33.2 ± 22.4 (12–74)	.74	.15	.16
Object knowledge	94.5 ± 6.7 (81–100)	95.2 ± 4.0 (89–98)	68.1 ± 19.3 (48–87)	.69	.01	.01
Single word comprehension	100.0 ± .0 (100–100)	100.0 ± .0 (100–100)	86.5 ± 16.1 (54–98)	1.00	.002	.003
Syntactic comprehension	74.4 ± 18.0 (48–97)	88.2 ± 2.9 (85–94)	86.7 ± 11.5 (65–96)	.20	.20	.48
Repetition	86.4 ± 19.3 (31–98)	87.7 ± 3.4 (83–93)	95.2 ± 3.0 (91–99)	.25	.25	.02
Written language	87.6 ± 14.4 (52–100)	93.3 ± 2.0 (90–95)	93.0 ± 7.7 (83–100)	.91	.91	.91

Values denote mean ± SD (range) or frequencies. IADL are expressed as mean percentage of intact abilities. Non-language cognitive variables are expressed as raw data. Language variables are expressed as mean percentage of correct answers obtained in all tests for each domain (see methods for further details). CSF cut off = Aβ42 > 500 ng/L (values below are considered abnormal); T-tau = 0–450 ng/L and p-tau = 0–61 ng/L (values above are considered abnormal).

^*Data available for 7 (54%) nfvPPA, 7 (88%) lvPPA, 4 (67%) svPPA. p values refer to Mann–Whitney U test or Fisher exact test (for continuous and categorical variables, respectively) and were adjusted for multiple comparisons controlling the False Discovery Rate at level .05, using Benjamini-Hochberg step-up procedure.

Abbreviations: Aβ42 = amyloid β42; ADL = Activities of Daily Living; CSF = cerebrospinal fluid; IADL = Instrumental Activities of Daily Living; lvPPA = logopenic variant of primary progressive aphasia; M/F = Males/Females; MMSE = Mini Mental State Examination; nfvPPA = nonfluent variant of primary progressive aphasia; RAVLT = Rey Auditory Verbal Learning Test; RCPM = Raven colored progressive matrices; svPPA = sematic variant of primary progressive aphasia; TMT-BA = Trail Making Test-BA; T-tau = total tau; p-tau = phosphorylated tau.

2.2. Neuropsychological assessment

An experienced neuropsychologist unaware of the MRI results performed the cognitive assessments (Table 1), which included the evaluation of the following: global cognitive functioning with the Mini-Mental State Examination (Folstein, Folstein, & McHugh, 1975); verbal memory with the digit span forward (Orsini et al., 1987) and the Rey auditory verbal learning test (Rey, 1964); visual-spatial memory with the spatial span (Orsini et al., 1987) and Rey’s figure delayed recall (Caffarra, Vezzadini, Dieci, Zonato, & Venneri, 2002); attention with the attentive matrices (Spinnler & Tognoni, 1987) and the trail making test (Giovagnoli et al., 1996), executive functions with the fluency (Novelli et al., 1986) and clock drawing (Manos, 1999) tests; abstract reasoning with Raven’s coloured progressive matrices (Basso, Capitani, & Laiacona, 1987); working memory with digit span backward (Monaco, Costa, Caltagirone, & Carlesimo, 2013); and visuospatial abilities with the Rey’s Figure Copy (Caffarra et al., 2002).

2.3. Language assessment

PPA patients underwent comprehensive language testing (Table 1), including the examination of: confrontation naming with the subtests of CaGi (Catricala et al., 2013) and Aachener Aphasie Test (AAT) (Luzzatti et al., 1994), and with the oral noun and verb confrontation naming subtests of BADA (Miceli, Laudanna, Burani, & Capasso, 1994); object knowledge with the Pyramids and Palm Trees Test (Gamboz, Coluccia, Iavarone, & Brandimonte, 2009); single-word comprehension with the word picture matching test of the CaGi (Catricala et al., 2013) and AAT (Luzzatti et al., 1994); comprehension of syntactically complex sentences with the auditory and visual comprehension subtests of BADA (Miceli et al., 1994), AAT (Luzzatti et al., 1994) and Token Test (De Renzi & Vignolo, 1962); repetition with the subtest of AAT (Luzzatti et al., 1994); and written language with the reading and writing subtests of AAT (Luzzatti et al., 1994). To evaluate the presence of motor speech, we recorded word repetition and speech samples from each of our patients while they described the image of the picnic picture subtest of the Western Aphasia Battery (Catricala et al., 2017; Kertesz, 1982). A standardized quantitative measure of motor ability is not available in Italian, thus performance of our participants is reported as no impairment or mild, moderate, severe impairment based on clinical judgment.

2.4. The 44-item Northwestern Anagram Test-Italian

All patients were tested using the 44-item NAT-I, an adapted version of the NAT (Thompson et al., 2012; Weintraub et al., 2009) for the Italian language (see Appendix for details on the test creation and administration; see E-Table 1 and E-Table 2 for the 44 test items and the scoring attribution, respectively). Briefly, patients were presented with a target drawing depicting three actors and an action. A word printed below the picture labeled the action in the picture. For each target sentence, single word cards constituting the correct sentence were printed. For each item, the examiner pointed to each picture, stating, for example, “The action in the picture is to push”. The examiner then placed the test cards in a random order under the picture and said, “Please build a sentence that describes this picture using these word cards’. The 44 items included four morphological structures (two canonical and two non-canonical) with eight verbs: eight simple active (canonical), eight passive (non-canonical), 16 complex active (canonical), 12 object-extracted (non-canonical) questions (E-Table 1). For the first two items, patients were provided feedback and corrected if wrong. Each item was administered with no time limit, both completion time and accuracy were the outcome measures. For each item, the accuracy scoring could be 0, .5 or 1. See E-Fig. 1 as an example of the test administration and E-Table 2 for scoring.

2.4.1. The 44-item Northwestern Anagram Test-Italian: main differences respect to the original (English) version

The typological differences between English and Italian (Haspelmath, Gil, & Comrie, 2005) dictated the need for an adaptation, rather than a translation, of the test. For example, word order can be expected to have a stronger role in sentence processing in English than in Italian (fixed vs relatively free), while morphology (for example, subject-verb agreement) plays a major role in the case of a highly inflected language, such as Italian (Bates, Wulfeck, & MacWhinney, 1991). Considering this crucial aspect, the substantial modifications that we applied to the NAT-I compared to the original version were the following (see Appendix for further details on the test creation):

1. in NAT-I, all active sentences and object questions include singular subject + plural object or viceversa; passive sentences include one singular and one plural character; questions used the interrogative adjective ‘quale’ (inflected)+ a noun rather than the interrogative pronoun ‘chi’(uninflected). These changes force the patient to focus on the correct morphological agreement (other than avoiding mistakes due to possible patient semantic disturbances). For this purpose, compared to the original version of NAT, words and arrows pointing at the characters were also excluded from the stimulus;

2. we avoided the inclusion of subject and object clefts since they are not frequently used in Italian. Instead, we added new complex active sentences with subject- or object-Prepositional Phrase Modifier (PPM).

2.5. Single- and multiple-combination, randomly selected, 22-item version of the Northwestern Anagram Test-Italian

From the 44 original items, and with items of grammatic complexity proportionally similar to the full test (four simple active, four passive, eight complex active, six object-extracted questions), we created and tested:

1. a single 22-item version: the following items were randomly selected one single time from the original version [see E-Table 1 for the full list of the original items]: 1-2-6-8-9-10-11-12-13-14-15-19-20-23-24-30-31-32-33-34-37-41.

2. multiple-combination 22-item versions: items were randomly selected multiple times from the original version. Specifically, they were resampled 5.837.860 times, which represent about .01% of the possible item combinations that can be obtained choosing a subset of 22 among 44 items without repetition and stratified by groups (i.e., 5.837.860 combinations among a total of possible 5.8 × 10¹⁰ combinations).

2.6. MRI acquisition

Brain MRI scans were obtained using a 3.0 T scanner (Intera, Philips Medical Systems, Best, the Netherlands). The following sequences were acquired from all subjects: T2-weighted spin echo (SE) [repetition time (TR) = 3500 msec; echo time (TE) = 85 msec; echo train length = 15; flip angle = 90°; 22 contiguous, 5-mm-thick, axial slices; matrix size = 512 × 512; 2 field of view (FOV) = 230 × 184 mm²]; fluid-attenuated inversion recovery (FLAIR; TR = 11 sec; TE = 120 msec; flip angle = 90°; 22 contiguous, 5-mm-thick, axial slices; matrix size = 512 × 512; FOV = 230 mm²); 3D T1-weighted fast field echo (FFE) (TR = 25 msec, TE = 4.6 msec, flip angle = 30°, 220 contiguous axial slices with voxel size = .89 × .89 × .8 mm, matrix size = 256 × 256, FOV = 230 × 182 mm²); and pulsed-gradient SE echo planar with sensitivity encoding (acceleration factor = 2.5; TR = 8986 msec; TE = 80 msec; 55 contiguous, 2.5-mm-thick axial slices; number of acquisitions = 2; acquisition matrix 96 × 96, with an in-plane 2 pixel size = 1.89 × 1.89 mm and a FOV = 240 × 240 mm ) and diffusion gradients applied in 32 non-collinear directions using a gradient scheme which is standard on this system (gradient over-plus) and optimized to reduce echo time as much as possible. The b factor used was 1000 sec/mm². Fat saturation was performed to avoid chemical shift artifacts. All slices were positioned to run parallel to a line that joins the most inferoanterior and inferoposterior parts of the corpus callosum.

2.7. MRI analysis

WM hyperintensities, if any, were identified on T2-weighted and FLAIR images. Voxel based morphometry (VBM) and DT MRI analyses were performed using standard procedures as implemented in Statistical Parametric Mapping (SPM) version 12 and the FMRIB software library (FSL), respectively.

2.7.1. Voxel-based morphometry

VBM was performed using SPM12 (https://www.fil.ion.ucl.ac.uk/spm/) and the Diffeomorphic Anatomical Registration Exponentiated Lie Algebra (DARTEL) registration method (Ashburner, 2007). Briefly, (i) T1-weighted images were segmented to produce grey matter (GM), WM and CSF tissue probability maps in the Montreal Neurological Institute (MNI) space; (ii) the segmentation parameters obtained from step (i) were imported in DARTEL; (iii) the rigidly aligned version of the images previously segmented (i) was generated; (iv) the DARTEL template was created and the obtained flow fields were applied to the rigidly-aligned segments to warp them to the common DARTEL space. Since the DARTEL process warps to a common space that is smaller than the MNI space, we performed an additional transformation as follows: (v) the aligned images from DARTEL were normalized to the MNI template using an affine transformation estimated from the DARTEL GM template and the a priori GM probability map. Prior to statistical computations, (vi) images were modulated using the Jacobian determinants and (vii) smoothed with an 8 mm FWHM Gaussian filter.

2.7.2. DT MRI preprocessing

DT MRI analysis was performed using the FSL tools (http://www.fmrib.ox.ac.uk/fsl/fdt/index.html) and the JIM5 software (Version 5.0, Xinapse Systems, Northants, UK, http://www.xinapse.com). The diffusion-weighted data were skull-stripped using the Brain Extraction Tool implemented in FSL. Using FMRIB’s Linear Image Registration Tool (FLIRT), the two diffusion-weighted scans were coregistered by applying the rigid transformation needed to correct for position between the two b0 images (T2-weighted, but not diffusion-weighted). The rotation component was also applied to diffusion-weighted directions. Eddy currents correction was performed using the JIM5 software. Then, the two acquisitions were concatenated. The DT was estimated on a voxel-by-voxel basis using DTIfit provided by the FMRIB Diffusion Toolbox. Maps of mean diffusivity (MD), fractional anisotropy (FA), axial diffusivity (axD) and radial diffusivity (radD) were obtained.

2.7.3. Tract based spatial statistics

Tract-based spatial statistics version 1.2 was used to perform the multisubject diffusion tensor MR imaging analysis. FA volumes were aligned to a target image by using the following procedure: (a) the FA template in standard space (provided by FSL) was selected as the target image, (b) the nonlinear transformation that mapped each subject’s FA to the target image was computed by using the FMRIB’s Nonlinear Image Registration Tool, and (c) the same transformation was used to align each subject’s FA to the standard space. A mean FA image was then created by averaging the aligned individual FA images and thinned to create a FA skeleton representing white-matter tracts common to all subjects. The FA skeleton was thresholded at a value of .2 to exclude voxels with low FA values, which are likely to include gray matter or cerebralspinal fluid. Individual FA, MD, axD and radD data were projected onto this common skeleton.

2.8. Statistical analysis

2.8.1. Demographic, clinical and cognitive data

Participant characteristics (i.e., demographic, clinical, cognitive and behavioural features) were reported as mean ± SD (range) or frequency for continuous and categorical variables, respectively. Pairwise comparisons between groups were performed using Mann–Whitney U test or Fisher exact test and P-values were adjusted for multiple comparisons controlling the False Discovery Rate at level .05, using Benjamini-Hochberg step-up procedure.

2.8.2. Discrimination analysis

For each NAT-I version (44-item, 22-item single and multiple-combination random selection), the diagnostic ability of each NAT-I measure to discriminate nfvPPA from lvPPA cases was assessed by Receiver Operating Characteristic (ROC) curve analysis retrieving the Area Under the ROC curve (AUC), along with its 95% confidence interval (CI) computed with 2000 stratified bootstrap replicates. Accuracy, sensitivity and specificity were calculated at the optimal cut-off of the ROC curve, which jointly maximizes sensitivity and specificity. Moreover, Random Forest analysis (Breiman, 2001) (with 100’0000 trees) was performed to provide a measure of “variable importance” (VIMP) along with the “relative variable importance” (RVIMP = VIMP/max VIMP × 100) of all items for each NAT-I version, separately. Both descriptive and discrimination analyses were performed using SAS Software, Release 9.4 (SAS Institute, Cary, NC, USA) and R (packages: random-ForestSRC, pROC) and p value < .05 was considered for statistical significance. Due to the small sample size, svPPA patients were not considered in the discrimination (and MRI – see below) analysis. NAT-I scores of svPPA patients were obtained to sample test performance in cases with no grammatical deficits.

2.8.3. MRI data: voxel based morphometry

Multiple regression models were performed in SPM12 to assess the relationship between GM density and the 44-item NAT-I variables. Total intracranial volume was added as a covariate. The statistical threshold was set at p < .001 uncorrected within 50 contiguous voxels.

2.8.4. MRI data: tract based spatial statistics

Multiple regression models were run using voxelwise statistics between DT MRI maps of MD, FA, axD and radD and NAT-I variables by a permutation-based inference tool for nonparametric statistical thresholding (the “randomize” tool; number of permutations, 5000). Statistical maps were thresholded at a p value of less than .05, family-wise error corrected for multiple comparisons at the cluster level using the threshold-free cluster enhancement option.

2.9. Data availability statement

The study participants did not grant explicit permission for sharing their data in open access platforms at the time of data collection. Due to these ethical constraints, we are therefore unable to publicly archive individual clinical, cognitive and MRI data. However, the raw, anonymized data will be made available by the corresponding author upon request to qualified researchers (i.e., affiliated to a university or research institution/hospital). Analysis codes are available on a public archive (https://osf.io/vnex7/). The owner of the above mentioned data and of the NAT-I material is Ospedale San Raffaele. This material will be made available upon precise request to the corresponding author.

3. Results

3.1. Demographic, clinical and cognitive data

Table 1 shows demographic, clinical and cognitive data of the patients. Patient groups were matched for global cognition, disease severity (in terms of independence in activities of daily living) and disease duration (Table 1). SvPPA patients were younger than nfvPPA cases at the time of the study visit and at disease onset. LvPPA patients showed reduced CSF Aβ42 and increased tau protein levels compared to the other two variants, while nfvPPA showed higher phosphorilated tau levels compared to svPPA cases. SvPPA patients showed poorer scores for tests of object knowledge and single word comprehension compared to both nfvPPA and lvPPA cases and better repetition compared to lvPPA patients. Only nfvPPA cases had qualitative motor speech disturbances with phonetic distortions present in two cases (15%; one with severe and one with moderate motor speech disturbances). The remaining nfvPPA cases presented with prevalent agrammatism. All patients performed similarly in verbal and spatial long-term memory, abstract reasoning, verbal working memory, attention shifting and in visuospatial domains (Table 1). Compared to the other patient groups, nfvPPA cases performed worse in phonemic fluency, verbal and spatial span. Finally, svPPA cases performed worse in semantic fluency and better in selective attention respect to the other PPA cases.

3.2. The 44-item Northwestern Anagram Test-Italian: performances and discriminatory power

Table 2 shows performances, in terms of accuracy and response times, on the 44-item version of the NAT-I. All patients required equal time to finish each section of the test. In terms of accuracy, compared to the other groups of patients, nfvPPA participants performed more poorly on the entire test except for the simple active sentences, where they had a lower score than svPPA only. SvPPA and lvPPA patients showed similar performances in each subtest. Table 3 shows the diagnostic accuracy of the NAT-I measures in discriminating between nfvPPA from lvPPA cases. NAT-I non-canonical and the NAT-I total scores best discriminated the two patient groups, achieving the highest discriminatory power: AUCs of .93 (95% CI: .82–1.00) and .91 (95% CI: .77–1.00), respectively (Table 3). A sample size of 8 lvPPA and 13 nfvPPA patients achieved 80% of statistical power to detect an AUC greater than or equal to .84 of any discrete (i.e., rating scale) response variable, under the null hypothesis of AUC equal to .50 and using a two-sided z-test at a significance level of .05. The AUC is computed between false positive rates of 0 and 1, assuming that the ratio between the SD of the responses in nfvPPA patients to the SD of the responses in the lvPPA patients is equal to 1.

Table 2 –

Accuracy and response time (mean, SD, and score ranges) of PPA patients (by variant) on the 44-item Northwestern Anagram Test-Italian.

	nfvPPA	lvPPA	svPPA	p nfvPPA versus lvPPA	p nfvPPA versus svPPA	p lvPPA versus svPPA
N	13	8	6
Simple active, score	6.5 ± 1.5 (3–8)	7.6 ± .9 (6–8)	8.0 ± .0 (8–8)	.11	.03	.20
Simple active, time	222.9 ± 130.7 (56–452)	182.6 ± 180.3 (43–565)	128.2 ± 80.0 (59–275)	.37	.29	.90
Passive, score	4.3 ± 2.0 (2–8)	7.4 ± 1.2 (5–8)	7.0 ± 1.7 (4–8)	.01	.03	.88
Passive, time	476.0 ± 255.2 (123–943)	317.3 ± 260.0 (67–868)	273.9 ± 203.1 (83–553)	.25	.16	1.00
Complex active, score	5.3 ± 5.0 (0–15)	13.5 ± 4.8 (2–16)	10.9 ± 2.9 (8–15)	.02	.03	.05
Complex active, time	1529.3 ± 764.2 (355–2879)	980.8 ± 674.4 (337–2294)	967.5 ± 700.1 (460–2336)	.12	.12	.80
Questions, score	4.2 ± 2.0 (1–7)	7.2 ± 1.9 (6–10)	7.5 ± 1.6 (6–11)	.02	.01	.32
Questions, time	527.3 ± 252.4 (135–1008)	456.9 ± 269.6 (186–1053)	359.9 ± 241.7 (143–721)	.35	.35	.35
Canonical, score	11.9 ± 6.1 (3–23)	21.1 ± 5.6 (8–24)	18.9 ± 2.9 (16–23)	.02	.03	.07
Canonical, time	1752.2 ± 867.0 (411–3331)	1163.4 ± 846.7 (386–2859)	1095.7 ± 775.9 (567–2611)	.19	.19	1.00
Non-canonical, score	8.5 ± 3.4 (4–15)	14.6 ± 2.6 (11–18)	14.5 ± 2.6 (11–19)	.003	.01	.95
Non-canonical, time	1003.3 ± 468.4 (265–1815)	774.2 ± 514.6 (265–1921)	633.8 ± 362.8 (226–1067)	.33	.24	.90
NAT-I, total score	20.3 ± 8.5 (6–36)	35.6 ± 7.7 (18–42)	33.8 ± 5.4 (27–42)	.01	.01	.27
NAT-I, total time	2755.5 ± 1271.4 (676–5146)	1937.7 ± 1340.7 (657–4780)	2342.4 ± 2508.3 (864–7355)	.31	.31	.80

Values denote mean ± SD. ‘Time’ refers to item completion time and is reported in seconds. p values refer to Mann-Whitney U test and were adjusted for multiple comparisons controlling the False Discovery Rate at level .05, using Benjamini-Hochberg step-up procedure.

Abbreviations: lvPPA = logopenic variant of primary progressive aphasia; nfvPPA = nonfluent variant of primary progressive aphasia; NAT-I = Northwestern Anagram Test-Italian; svPPA = sematic variant of primary progressive aphasia.

Table 3 –

Discrimination analysis (nfvPPA vs lvPPA) of the 44-item Northwestern Anagram Test-Italian. Items were sorted from the highest to the lowest relative variable importance (RVIMP) calculated by the Random Forest algorithm.

	RVIMP	AUC	AUC 95% lower	AUC 95% upper	Threshold	accuracy	sensitivity	specificity
Non-canonical, score	100.00%	.93	.82	1	12.75	90.50%	87.50%	92.30%
NAT-I, total score	89.51%	.91	.77	1	35.75	90.50%	75.00%	100.00%
Passive, score	31.95%	.87	.72	1	6.50	85.70%	87.50%	84.60%
Canonical, score	25.64%	.86	.67	1	18.75	85.70%	87.50%	84.60%
Complex active, score	25.18%	.87	.69	1	11.25	85.70%	87.50%	84.60%
Questions, score	20.00%	.83	.65	1	5.25	81.00%	100.00%	69.20%
NAT-I, total time	25.00%	.70	.45	.95	2556.46	71.40%	87.50%	61.50%
Complex active, time	−.04%	.73	.50	.96	1208.67	71.40%	75.00%	69.20%
Non-canonical, time	−.90%	.66	.40	.93	883.24	71.40%	87.50%	61.50%
Questions, time	−.94%	.63	.36	.89	554.50	66.70%	87.50%	53.80%
Simple active, score	−1.51%	.73	.50	.95	7.50	71.40%	75.00%	69.20%
Canonical, time	−2.50%	.70	.46	.95	1711.25	66.70%	87.50%	53.80%
Passive, time	−3.08%	.68	.42	.94	474.00	66.70%	87.50%	53.80%
Simple active, time	−5.48%	.65	.37	.93	77.70	76.20%	50.00%	92.30%

Abbreviations: AUC = area under the ROC curve; RIVMP = relative variable importance [i.e., variable importance (VIMP)/max VIMP × 100]; NAT-I = Northwestern Anagram Test-Italian. Accuracy, sensitivity and specificity were calculated at the optimal threshold of the ROC curve, which jointly maximizes sensitivity and specificity.

3.3. Relationship between the 44-item Northwestern Anagram Test-Italian and brain integrity

In nfvPPA and lvPPA patients, we observed a positive relationship between the NAT-I total and non-canonical scores and the GM density of the pars triangularis of the left inferior frontal gyrus (Fig. 1A; NAT-I total versus GM: left pars triangularis, cluster size = 683 voxels, x = −57, y = 20, z = 0, T value = 6.92; NAT-I non-canonical versus GM: left pars triangularis, cluster size = 113 voxels, x = −57, y = 18, z = 0, T value = 4.52). Furthermore, the NAT-I total score was positively related with the GM density of the pars triangularis and orbitalis of the right inferior frontal gyrus (Fig. 1A; NAT-I total versus GM: right pars triangularis, cluster size = 877, x = 58, y = 24, z = 8, T value = 5.48; right pars orbitalis, cluster size = 877, x = 46, y = 22, z = −6, T value = 4.39). Finally, NAT-I non-canonical score was positively related with the FA of the body of the corpus callosum (Fig. 1B).

Fig. 1 –

A) Positive relationship between the performance at the Northwestern Anagram Test-Italian and the gray matter volume in PPA patients. Results are overlaid on the Montreal Neurological Institute standard brain in neurological convention (right is right), and displayed at p < .001 uncorrected for multiple comparisons within 50 contiguous voxels. Colored bars denote T-values. B) Positive relationship between the performance at the Northwestern Anagram Test-Italian and the microstructural integrity (fractional anisotropy) of the corpus callosum in PPA patients. Results are overlaid on the Montreal Neurological Institute standard brain in neurological convention (right is right) and displayed at p < .05 threshold-free cluster enhancement.

3.4. The 22-item Northwestern Anagram Test-Italian: performances and discriminatory power

Table 4 shows performances, in terms of accuracy and completion time, on the single-time randomly selected 22-item NAT-I version (Table 4 legend reports the single-time randomly selected items). All patients required equal time to finish each section of the test. In terms of accuracy, compared to the other groups of patients, nfvPPA patients performed worse on the entire test, except for the simple active and total canonical sentences. SvPPA and lvPPA patients showed similar performances in each subtest. Table 5 shows the diagnostic accuracy of the single-time randomly selected 22-item NAT-I measures in discriminating nfvPPA from lvPPA cases. NAT-I total and non-canonical scores best discriminated the two patient groups, achieving the highest discriminatory power: AUCs of .94 (95% CI: .83–1.00) and .92 (95% CI: .81–1.00), respectively.

Table 4 –

Performances of patients at the single-time randomly selected 22-item version of the Northwestern Anagram Test-Italian.

	nfvPPA	lvPPA	svPPA	p nfvPPA versus lvPPA	p nfvPPA versus svPPA	p lvPPA versus svPPA
N	13	8	6
Simple active, score	3.1 ± 1.0 (1–4)	3.8 ± .5 (3–4)	4.0 ± .0 (4–4)	.12	.05	.20
Simple active, time	103.9 ± 49.7 (27–207)	96.4 ± 89.0 (20–236)	73.4 ± 52.5 (40–179)	.64	.34	.80
Passive, score	2.0 ± 1.3 (0–4)	3.7 ± .7 (2–4)	3.5 ± .8 (2–4)	.01	.03	.69
Passive, time	225.8 ± 117.8 (59–426)	147.0 ± 117.6 (33–388)	137.8 ± 102.4 (43–282)	.19	.19	.90
Complex active, score	3.2 ± 2.6 (0–8)	7.0 ± 2.1 (2–8)	4.7 ± 2.1 (3–8)	.01	.21	.07
Complex active, time	740.3 ± 380.8 (169–1642)	444.9 ± 314.6 (160–1083)	499.6 ± 403.0 (190–1274)	.15	.24	.61
Questions, score	1.9 ± 1.0 (1–4)	3.5 ± .9 (3–5)	3.7 ± .8 (3–5)	.01	.01	.50
Questions, time	290.2 ± 147.2 (71–516)	222.1 ± 133.8 (96–526)	202.5 ± 151.2 (76–456)	.54	.54	.61
Canonical, score	6.3 ± 3.3 (1–12)	10.8 ± 2.4 (5–12)	8.7 ± 2.1 (7–12)	.20	.08	.08
Canonical, time	844.2 ± 423.8 (196–1849)	541.3 ± 389.1 (200–1312)	573 ± 453.6 (230–1453)	.20	.20	.70
Non-canonical, score	3.9 ± 1.9 (1–8)	7.2 ± 1.3 (5–9)	7.2 ± 1.3 (6–9)	.004	.01	1.00
Non-canonical, time	519.9 ± 249.9 (135–924)	369.2 ± 248.1 (139–914)	340.3 ± 209.3 (120–553)	.29	.29	.90
NAT-I, total score	10.2 ± 4.7 (2–18)	17.9 ± 3.4 (10–21)	15.8 ± 3.1 (12–20)	.003	.03	.17
NAT-I, total time	1360.1 ± 624.0 (331–2773)	910.5 ± 629.9 (339–2226)	913.3 ± 621.4 (416–2006)	.28	.28	1.00

Values denote means ± SD (range). ‘Time’ is reported in terms of seconds. The following items were randomly selected from the original test: 1–2–6–8–9–10–11–12–13–14–15–19–20–23–24–30–31–32–33–34–37–41. p values refer to Mann–Whitney U test and were adjusted for multiple comparisons controlling the False Discovery Rate at level .05, using Benjamini-Hochberg step-up procedure.

Abbreviations: lvPPA = logopenic variant of primary progressive aphasia; nfvPPA = nonfluent variant of primary progressive aphasia; svPPA = sematic variant of primary progressive aphasia; NAT-I = Northwestern Anagram Test-Italian.

Table 5 –

Discrimination analysis of the single-time randomly selected 22-item version of the Northwestern Anagram Test-Italian for the classification of nfvPPA and lvPPA cases. Items were sorted from the highest to the lowest relative variable importance (RVIMP) calculated by the Random Forest algorithm.

	RVIMP	AUC	AUC 95% lower	AUC 95% upper	Threshold	Accuracy	Sensitivity	Specificity
NAT-I, total score	100.00%	.94	.83	1	16.75	90.50%	87.50%	92.30%
Non-canonical, score	62.77%	.92	.81	1	5.75	85.70%	87.50%	84.60%
Complex active, score	40.51%	.87	.70	1	6.50	85.70%	87.50%	84.60%
Passive, score	22.79%	.87	.72	1	3.25	81.00%	87.50%	76.90%
Canonical, score	22.32%	.86	.68	1	8.50	81.00%	87.50%	76.90%
Questions, score	18.47%	.88	.74	1	2.75	81.00%	87.50%	76.90%
Simple active, time	1.07%	.61	.30	.92	45.43	76.20%	50.00%	92.30%
Complex active, time	.11%	.76	.53	.99	499.49	76.20%	75.00%	76.90%
NAT-I, total time	−.02%	.72	.47	.97	1252.72	71.40%	87.50%	61.50%
Simple active, score	−.43%	.71	.52	.91	3.50	66.70%	75.00%	61.50%
Passive, time	−.46%	.70	.45	.95	107.00	76.20%	50.00%	92.30%
Canonical, time	−1.50%	.71	.46	.97	546.00	76.20%	62.50%	84.60%
Non-canonical, time	−3.85%	.67	.42	.93	489.50	66.70%	87.50%	53.80%
Questions, time	−4.21%	.65	.38	.92	259.78	66.70%	87.50%	53.80%

‘Time’ is reported in terms of seconds. Abbreviations: AUC = area under the ROC curve; nfvPPA = nonfluent variant of primary progressive aphasia; RIVMP = relative variable importance [i.e., variable importance (VIMP)/max VIMP × 100]; NAT-I = Northwestern Anagram Test-Italian.

Accuracy, sensitivity and specificity were calculated at the optimal threshold of the ROC curve, which jointly maximizes sensitivity and specificity.

After the multiple-combination random selection of the NAT-I 22-items, NAT-I total, non-canonical and passive scores revealed the highest diagnostic accuracy for the classification of nfvPPA from lvPPA patients. AUCs ranges from .83 to .97, from .84 to .98 and from .86 to .90 for NAT-I total, non-canonical and passive scores, respectively (Table 6).

Table 6 –

Minimum and maximum range of diagnostic measures for the classification of nfvPPA from lvPPA patients of multiple-combination, 22-item Northwestern Anagram Test-Italian.

	N Itema	Minimum				Maximum
		AUC	Accuracy	Sensitivity	Specificity	AUC	Accuracy	Sensitivity	Specificity
NAT-I, total score	22	.83	.81	.63	.69	.97	.95	1.00	1.00
Non-canonical, score	10	.84	.76	.63	.62	.98	.95	1.00	.92
Passive, score	4	.86	.76	.88	.62	.90	.91	1.00	.85
Complex active, score	8	.79	.81	.75	.77	.94	.91	.88	1.00
Canonical, score	12	.78	.81	.63	.77	.92	.91	.88	1.00
Questions, score	6	.68	.62	.50	.39	.94	.86	1.00	1.00
Simple active, score	4	.67	.57	.75	.31	.74	.71	1.00	.69
Passive, time	4	.63	.67	.38	.54	.75	.81	.88	1.00
Simple active, time	4	.62	.76	.50	.92	.64	.76	.50	.92
NAT-I, total time	22	.61	.62	.38	.46	.76	.81	.88	1.00
Complex active, time	8	.61	.67	.38	.54	.79	.81	.88	1.00
Canonical, time	12	.61	.67	.38	.54	.76	.76	.88	1.00
Non-canonical, time	10	.43	.57	.50	.46	.73	.76	.88	.85
Questions, time	6	.47	.48	.38	.15	.71	.81	1.00	.85

Abbreviations: AUC = Area Under the ROC Curve; NAT-I = Northwestern Anagram Test-Italian. Accuracy, sensibility and specificity measures are calculated with respect to the optimal threshold found at each possible combination.

^aRespect to the original 44-item Northwestern Anagram Test-Italian, 50% of the original items were randomly chosen within each sub-category.

4. Discussion

This study resulted in three major findings: 1) the adaptation of the NAT for the Italian language (i.e., NAT-I) is useful for distinguishing between nfvPPA and lvPPA in vivo, in particular on the basis of performance on non-canonical sentences; 2) patient performances on the NAT-I were associated with the integrity of the inferior frontal gyrus, which is implicated in syntactic processing, and the body of the corpus callosum, which has a role in interhemispheric integration; 3) the 22-item-brief version of NAT-I is suitable for clinical practice since it drastically reduced the time of the cognitive assessment and allows test randomization using different versions of the test for serial assessments.

Syntactic processing is a crucial language feature for PPA differential diagnosis, since deficits in this domain suggest an in vivo underlying tauopathy (Mesulam et al., 2008), which is often associated with the nfvPPA condition rather than with the other variants (Spinelli et al., 2017). In clinical practice, syntactic processing, in particular syntactic production, is often assessed by recording patient spontaneous speech samples or narrative speech while he/she is describing a picture. However, these tasks require several other language components, such as for instance word finding, verbal working memory and language motor programming. Furthermore, the analysis of the speech sample is laborious, time consuming and extremely dependent on the expertise of the language specialist, with the final high risk of misinterpreting the patient’s speech difficulties. Like the NAT (Thompson et al., 2012; Weintraub et al., 2009), the NAT-I overcomes the impairment of speech production and uniquely assesses the ability to compose syntactically correct sentences from their scrambled building blocks.

In this study, we demonstrated that the NAT-I was able to distinguish between nfvPPA and lvPPA cases, reaching the highest accuracies of classification for the total score of the test and for the score obtained from a subset of items measuring non-canonical sentence processing. This latter finding is in line with the well-known fact that the more complex the sentence structure, in terms of standard word order, as in the case of non-canonical structure (with patients having difficulty in determining the agent and recipient of the action in the sentences), the more difficult the sentence processing for nfvPPA patients (Saffran, Schwartz, & Marin, 1980; Thompson, Shapiro, Kiran, & Sobecks, 2003). These data also confirm previous findings obtained using the NAT and the Northwestern Assessment of Verbs and Sentences (NAVS) (Cho-Reyes & Thompson, 2012; Thompson, 2011; Weintraub et al., 2009), showing that lower performance on non-canonical sentences is specifically associated with agrammatic aphasia. In the original NAT study (Weintraub et al., 2009), although data were not presented separately for each PPA subtype, the authors observed that performances on the NAT were within the normal range in some PPA cases who, instead, had deficits in single word comprehension (likely svPPA) or motor speech (likely nfvPPA cases without agrammatism) (Weintraub et al., 2009). In a following study (Thompson et al., 2013), patients with agrammatic PPA were found to perform more poorly than those with lvPPA on non-canonical forms on the NAT. In addition, significantly poorer non-canonical sentence production was found for agrammatic compared to lvPPA patients on the NAVS (Thompson et al., 2013). Mesulam and colleagues found that 7 out of 9 early-stage nfvPPA cases were correctly classified and distinguished from the other variants (including lvPPA, svPPA and mixed PPA) based on a model which took into consideration the patients’ performances on the single word comprehension and on the NAT and NAVS (Mesulam et al., 2012). Impairment in repetition was present in most of the nfvPPA cases (likely due to the extension of atrophy to the left temporoparietal junction and superior temporal gyrus) and did not help in distinguishing nfvPPA from lvPPA cases (Mesulam et al., 2012). On the other hand, the absence of atrophy in the left inferior frontal gyrus and the preservation of grammaticality were core features for lvPPA correct identification (Mesulam et al., 2012).

NfvPPA patients can present with agrammatism and/or apraxia of speech. In our sample, only two nfvPPA cases (15%) had motor speech disturbances. These patients performed below the suggested thresholds in both NAT-I total and non-canonical scores. The remaining nfvPPA cases presented with prevalent agrammatism. On this purpose, our group observed that during connected speech samples, monolingual nfvPPA English speakers showed significantly high number of distortions and high motor speech rate, while nfvPPA Italian speakers had significantly reduced syntactic complexity. These findings occurred in patients with similar cognitive and anatomical cortical and subcortical features, suggesting the need to take into account the possible impact of the individual’s spoken language on the clinical presentation of PPA and the relevance of NAT when diagnosing Italian cases.

In all PPA patients, we observed an association between total score and non-canonical NAT-I sentences and atrophy of the anterior (pars orbitalis) and posterior (pars triangularis) regions of the bilateral (for the total score) and left (for the non-canonical score) inferior frontal gyrus. The role of the left inferior frontal gyrus in syntactic processing is supported by extensive evidence from clinical and neuroimaging studies (Friederici, 2018). In fact, these findings well reflect those of a previous study where authors used the NAT for assessing syntactic production in a PPA population and observed a relationship between the performance on the entire test and the cortical thickness of the left anterior and posterior inferior frontal gyrus in patients (Rogalski et al., 2011). One of the most interesting findings of this previous study is the neuroanatomical distinction between fluency (i.e., mean length of utterance in words) impairments from those of grammatical processing, as measured by sentence production ability (poor fluency was associated with regions dorsal to the traditional boundaries of Broca’s area in the inferior frontal sulcus and the posterior middle frontal gyrus, whereas grammatical processing was associated with more widespread atrophy, including the inferior frontal gyrus and supramarginal gyrus) (Rogalski et al., 2011). In other studies, authors observed a relationship between quantitative (Wilson et al., 2010) (such as the embedding frequency, the number of words in sentences and the syntactic errors) and qualitative (syntactic production rated on a 7-point scale by two researchers) (Wilson et al., 2011) syntactic production measures in connected speech and the volume loss in the left posterior (mainly pars triangularis) inferior frontal gyrus. Furthermore, some authors observed that patients with left inferior frontal atrophy experienced difficulties mainly in constructing sentences with embeddings (Deleon et al., 2012), made a greater number of syntactic errors (Deleon et al., 2012), and made more role reversal errors primarily in the passive sentences (non-canonical structures) (Tyler et al., 2011).The activation of the left posterior inferior frontal gyrus during a sentence production task has been also demonstrated in a healthy population in a PET study (Indefrey, Hellwig, Herzog, Seitz, & Hagoort, 2004).

Furthermore, in all PPA patients, we observed an association between production of non-canonical forms on the NAT-I and microstructural alterations of the body of the corpus callosum, which contains interhemispheric projections. Previous studies investigating white matter in PPA have been primary focused on tracts belonging to the dorsal and ventral language networks, such as the superior longitudinal fasciculus, and the inferior longitudinal and the uncinate fasciculi (Wilson, Galantucci, Tartaglia, & Gorno-Tempini, 2012). The corpus callosum involvement (Friederici, von Cramon, & Kotz, 2007), particularly in syntactic processing, has been less investigated. Here we speculate that integrating the information processing coming from the two hemispheres may play a central role in NAT-I performance.

In this study, we also demonstrated that a brief version of the NAT-I with only 22 items is able to reach similar high accuracies for distinguishing nfvPPA from lvPPA cases as the 44-item version, with non-canonical scores confirmed as the best variables to be used for classification. Furthermore, for the 22 item selection from the original version of NAT-I, we proposed a randomization procedure which respects the original proportion of canonical and non-canonical sentences. The randomization demonstrated high reliability ranges, thus promoting its use in clinical practice to reduce testing time and to allow repeated administrations of the measure devoid of learning effects.

There are some caveats to be considered when interpreting our findings. The major limitation of the study is the small sample size, in particular for the svPPA group, which was too small to allow a discriminant analysis. However, neuropsychological tests other than the NAT-I, such as those investigating object knowledge and single word comprehension, were sufficient to clinically differentiate this group from the other two variants. In addition, diagnoses were not pathology-proven, although CSF biomarkers in selected cases pointed to an underlying AD pathology in lvPPA and non-AD pathology in patients with the other variants. Future studies in larger and path-proven samples should test the performance of these measures for a correct classification at the single subject level. Third, although the validity of the NAT-I has been demonstrated in the differential diagnosis of PPA Italian cases, the gold standard for analyzing NAT-I performance was based upon the clinical PPA classification rather than upon objective grammar measures. Finally, we cannot rule out a role played by other cognitive domains, in particular by the executive functions, to the NAT-I and, consequently, to the present findings. However, this is unlikely since patients’ groups (nfvPPA and lvPPA) were similar for all investigated executive functions and test completion time (which can be primary affected by the executive dysfunction) was not a significant variable of the present work.

Abbreviations

AAT

Aachener Aphasie Test

AD

Alzheimer’s disease

AUC

area under the curve

axD

axial diffusivity

Aβ42

amyloid β42

BADA

Batteria per l’Analisi del Deficit Afasici (Italian battery for aphasia deficits)

CSF

cerebrospinal fluid

DARTEL

Diffeomorphic Anatomical Registration Exponentiated Lie Algebra

DTI

diffusion tensor imaging

FA

fractional anisotropy

FDG PET

fluorodeoxyglucose positron emission tomography

FFE

fast field echo

FLAIR

fluid-attenuated inversion recovery

FLIRT

FMRIB’s Linear Image Registration Tool

FOV

field of view

FSL

FMRIB software library

FTLD

frontotemporal lobar degeneration

GM

grey matter

lvPPA

logopenic variant of primary progressive aphasia

MD

mean diffusivity

MRI

magnetic resonance imaging

NAT

Northwestern Anagram Test

NAT-I

Northwestern Anagram Test-Italian

NAVS

Northwestern Assessment of Verbs and Sentences

nfvPPA

nonfluent variant of primary progressive aphasia

PET

positron emission tomography

PPA

primary progressive aphasia

PPM

Prepositional Phrase Modifier

radD

radial diffusivity

ROC

receiver operator characteristic

RVIMP

relative variable importance

SE

spin echo

SPM

Statistical Parametric Mapping

svPPA

semantic variant of primary progressive aphasia

TDP

TAR DNA-binding protein

TE

echo time

TR

repetition time

VBM

voxel-based morphometry

VIMP

variable importance

WM

white matter