Differential linguistic features of verbal fluency in behavioral variant frontotemporal dementia and primary progressive aphasia

Abstract

Frontotemporal dementia (FTD) is an early-onset neurodegenerative disorder with a heterogeneous clinical presentation. Verbal fluency is regularly used as a sensitive measure of language ability, semantic memory, and executive functioning, but qualitative changes in verbal fluency in FTD are currently overlooked. This retrospective study examined qualitative, linguistic features of verbal fluency in 137 patients with behavioral variant (bv)FTD (n = 50), or primary progressive aphasia (PPA) [25 non-fluent variant (nfvPPA), 27 semantic variant (svPPA), and 34 logopenic variant (lvPPA)] and 25 control participants. Between-group differences in clustering, switching, lexical frequency (LF), age of acquisition (AoA), neighborhood density (ND), and word length (WL) were examined in the category and letter fluency with analysis of variance adjusted for age, sex, and the total number of words. Associations with other cognitive functions were explored with linear regression analysis. The results showed that the verbal fluency performance of patients with svPPA could be distinguished from controls and other patient groups by fewer and smaller clusters, more switches, higher LF, and lower AoA (all p < 0.05). Patients with lvPPA specifically produced words with higher ND than the other patient groups (p < 0.05). Patients with bvFTD produced longer words than the PPA groups (p < 0.05). Clustering, switching, LF, AoA, and ND—but not WL—were differentially predicted by measures of language, memory, and executive functioning (range standardized regression coefficient 0.25–0.41). In addition to the total number of words, qualitative linguistic features differ between subtypes of FTD. These features provide additional information on lexical processing and semantic memory that may aid the differential diagnosis of FTD.

Keywords:

• Aphasia
• assessment
• cognitive
• language
• neurodegenerative

Introduction

Frontotemporal dementia (FTD) is an early-onset neurodegenerative disorder associated with behavioral and/or language impairment due to atrophy of specific areas of the frontal and temporal cortices in the brain (Rohrer & Rosen, 2013; Seelaar et al., 2011). Differentiating FTD subtypes can present a substantial diagnostic challenge because of the relatively low prevalence of the disease and the heterogeneous presentation of symptoms. Behavioral variant frontotemporal dementia (bvFTD) is characterized by profound and progressive behavioral changes including emotional blunting, disinhibition, and lack of insight (Rascovsky et al., 2011). bvFTD is also accompanied by deficits in different aspects of language processing, such as naming, single-word comprehension, and verbal semantic processing (Hardy et al., 2016). The language subtypes (primary progressive aphasia, PPA) are distinguished between the non-fluent variant (nfvPPA) characterized by halting, effortful speech (speech apraxia) and agrammatism, and the semantic variant (svPPA) characterized by fluent speech but impaired word comprehension, object recognition and/or surface dyslexia resulting from loss of semantic knowledge (Gorno-Tempini et al., 2011). The third subtype of logopenic variant PPA (lvPPA)—generally associated with Alzheimer’s pathology—is characterized by impaired word finding and sentence repetition, and phonological errors (Gorno-Tempini et al., 2011; Marshall et al., 2018).

Sensitive and specific cognitive markers can aid accurate and timely clinical diagnosis of FTD subtypes, allowing for appropriate care and treatment planning. Verbal fluency is a well-known, brief and sensitive measure of language ability, semantic memory, and executive functioning. It is frequently used in neuropsychological assessment of suspected neurodegeneration in general, and of bvFTD and PPA in particular (Jiskoot et al., 2018, Laisney et al., 2009; Staffaroni et al., 2021). The two major subtypes of verbal fluency are category fluency and letter fluency. These tasks require a person to name words that belong to a specific category (e.g., animals, clothing, supermarket items, professions) or start with a specific letter of the alphabet, respectively. Whereas category fluency relies heavily on semantic knowledge and memory functioning, letter fluency is additionally supported by executive processes. Performance is typically measured by the total number of generated words and differences have been reported among subtypes of FTD (Libon et al., 2009) and neurodegenerative parkinsonian syndromes (Parkinson’s disease dementia, dementia with Lewy bodies, progressive supranuclear palsy, corticobasal syndrome; Magdalinou et al., 2018), between bvFTD and Alzheimer’s disease (Rascovsky et al., 2007), and between Alzheimer’s disease and PPA (Marczinski & Kertesz, 2006).

In addition to the total number of words generated, qualitative aspects of verbal fluency performance may provide insight into the cognitive processes involved and may aid in differentiating subtypes of FTD. Previous work on such qualitative measures by our group showed that clustering of related words (e.g. dog, cat, mouse belong to the same subcategory) and switching between categories of related words differs between bvFTD and PPA subtypes (van den Berg et al., 2017). That is, patients with PPA generated (a) smaller (number of) clusters in both category and letter fluency compared with patients with bvFTD. Patients with svPPA also produced more switches than patients with nfvPPA and lvPPA. Clustering and switching are considered robust theory-based fluency measures, and differential impairments have been reported in patients with schizophrenia, HIV, Huntington’s disease, Parkinson’s disease, multiple sclerosis, and traumatic brain injury as well (Ho et al., 2002; Iudicello et al., 2008, Messinis et al., 2013; Troyer et al., 1998; Zakzanis et al., 2000, 2013).

In addition to clustering and switching, the words that are generated in a verbal fluency task also convey linguistic information at the level of the single word, e.g. by means of meaning or word form. Specific linguistic aspects that are distinguished at the single word level include lexical frequency (LF; how often a word occurs in a given language corpus), age of acquisition (AoA; the age at which a word is learned), neighborhood density (ND; the number of possible words in the lexicon that differ phonologically by one sound from the target word) and word length (WL). These variables are not (entirely) independent as, for example, high-frequency words are learned at a younger age and ND facilitates the retrieval of low-frequency words (Brysbaert & Ghyselinck, 2006). They do, however, differ in the relative association with the integrity of semantic memory at the conceptual level (Brysbaert et al., 2000) or word form processing (i.e. lexeme level) (Roelofs et al., 1996). As a consequence, these linguistic aspects of lexical retrieval are of particular interest in FTD as all PPA subtypes show impairments at the level of single words, albeit in different aspects, i.e. single word production (nfvPPA), retrieval (lvPPA), or understanding (svPPA). A previous study on lexical decision showed selective impairment in PPA subtypes associated with AoA, LF, and ND (Vonk, Jonkers, et al., 2019). Patients with svPPA responded faster and more accurately to early-acquired than late-acquired words in a lexical decision task compared with control participants and other subtypes of PPA. They also had a specific deficit in processing low-frequency—vs. high-frequency—words. ND is a linguistic feature that influences lexical-semantic processing through a word’s lexical label instead of its meaning (conceptual level). As word-finding difficulties in lvPPA and word production difficulties in nfvPPA are not primarily driven by impairment at the conceptual level we hypothesized a higher mean ND in lvPPA and nfvPPA compared with svPPA. Indeed, Vonk, Jonkers et al. (2019) reported that patients with nfvPPA and lvPPA showed an effect of ND, i.e. better performance on words with high vs. low ND, that was not present in svPPA. There is even evidence indicating that LF in verbal fluency is a specific marker for early degeneration of semantic networks in non-demented APOE ε4 carriers, the strongest genetic risk factor of Alzheimer’s disease (Vonk, Flores, et al., 2019). Language abilities in bvFTD remain relatively preserved compared with PPA—at least in the early stages of the disease—but semantic (Cousins & Grossman, 2017) and non-fluent/agrammatic changes have been reported (Geraudie et al., 2021).

This study aimed to examine qualitative, linguistic features of category and letter fluency and associated cognitive functions in a patient with bvFTD and PPA. Based on the differences in clinical subtypes of FTD and accompanying changes in the brain a disparate pattern of impairment in qualitative, linguistic features of verbal fluency is expected. Due to semantic impairment, we hypothesized words with a lower AoA and higher LF in svPPA compared with the other patient groups. In contrast, ND is expected to be higher and WL is expected to be lower in nfvPPA and lvPPA due to speech apraxia and word-finding difficulties, respectively.

Methods

Participants

This retrospective study included 137 patients who visited the Alzheimer Center Erasmus MC, Rotterdam, the Netherlands, between January 2012 and December 2019 and were diagnosed with bvFTD or PPA. A previous study by our group reported on a subsample (n = 79) of the present study population (van den Berg et al., 2017). The clinical diagnosis was made in a multidisciplinary consensus meeting using international diagnostic consensus criteria. For bvFTD (n = 51) the diagnostic criteria of Rascovsky et al. (2011) were used. The diagnosis of patients with PPA (27 svPPA, 25 nfvPPA, and 34 lvPPA) was based on the diagnostic criteria of Gorno-Tempini et al. (2011). As part of their diagnostic workup, all patients underwent a standardized clinical assessment including medical history, informant-based history, neurological examination, neuropsychological assessment (including Mini Mental-State Examination (MMSE; Folstein et al., 1975) and Frontal Assessment Battery (FAB; Dubois et al., 2000) and several other tests described in the section “Other measures of cognition and mood”), laboratory tests and MR imaging of the brain. Time since symptom onset was defined as the interval between patient and/or informant-reported first symptoms and diagnosis.

Control participants (n = 25) were community-dwelling older persons recruited through community centers and word of mouth from the greater Rotterdam area (Rotterdam, Schiedam, Barendrecht). Control participants were included when they had no self-reported history of neurological or psychiatric disorders, a depression score <11 on the Hospital Anxiety and Depression Scale (HADS; Bjelland et al., 2002), and a Mini Mental-State Examination (MMSE) score >25.

All participants were native Dutch speakers. The level of education was classified according to the system of Verhage ranging from one (less than primary school) to seven (university degree) (Duits & Kessels, 2014) and converted into years of education. The study was approved by the Medical and Ethical Review Committee of the Erasmus MC University Medical Center. Participants gave written informed consent.

Verbal fluency

All participants performed a category and letter fluency task as part of a standardized neuropsychological assessment (see below). For the category fluency task, participants were asked to generate as many exemplars as possible from the category animals in 60 s. For the letter fluency task, participants were asked to generate as many different words as possible beginning with the letter D, then A, and then T, with 60 s allowed for each letter in separate trials. The letters DAT are considered the Dutch equivalent to the letters FAS (Schmand et al., 2008). Participants were instructed not to generate proper names or a previously generated word with only a different suffix (e.g. shoe, shoelace, shoemaker). The total number of correct animals and the total number of correct words generated for the letters D, A, and T were recorded.

Clustering, switching, and linguistic variables

Clustering and switching in verbal fluency were assessed with the detailed scoring system reported in Ledoux et al. (2014) (adaptation of Troyer et al., 1997). A cluster contains words belonging to the same semantic or phonological (sub)category, such as “dog” and “cat” or share the first and last sound, such as “simple” and “sample.” The number of clusters, the mean cluster size, and the number of switches between clusters were calculated. All clustering and switching variables were scored by two independent raters (JD with EB or LJ). Interrater reliability for category and letter fluency was high (range 0.93–1.0).

Lexical frequency refers to the number of times a word occurs in a corpus of daily language. All correctly generated words were paired with the corresponding (log-transformed) lexical frequency value from the SUBTLEX-NL database (Keuleers et al., 2010), containing 44 million Dutch words from film and television subtitles, and calculated the average lexical frequency across the generated words for each participant. Words that were not found in SUBTLEX-NL were included with a frequency value of 0 (k = 28 words, 1.3% of all generated words).

Age of acquisition was calculated by the average across words for each participant after pairing correctly generated words with the Dutch AoA database (Brysbaert et al., 2014). When a word was not available in the database (1) a conjugation was used, (2) the main category was used (e.g. “bear” instead of “polar bear”), or (3) the highest AoA value recorded in the sample as a whole was imputed [k = 92 words, 1.5% of all generated words; this procedure was developed in consultation with an experienced clinical linguist (DS)].

Neighborhood density refers to the number of phonologically similar words in the lexicon and is calculated by determining the number of possible words created by addition, deletion, or substitution of a single sound (e.g. neighbors of “sit” include “it” and “hit”). Phonological neighborhood values were derived from the dutchPOND database (Marian et al., 2012), and averaged across the generated words for each participant.

Orthographic word length was calculated as the number of letters of correctly generated words and averaged across the generated words for each participant.

Other measures of cognition and mood

Memory was assessed with the Dutch version of the Rey Auditory Verbal Learning Test (RAVLT; immediate and delayed recall; van der Elst et al., 2005). The language was examined with the 60-item Boston Naming Test (BNT; Kaplan et al., 1978). Information processing speed was assessed with the Trail Making Test Part A (TMT A; Corrigan & Hinkeldey, 1987). Executive functioning was assessed with the TMT B/A ratio, the modified Wisconsin Card Sorting Test (mWCST; number of concepts; Nelson, 1976), and the Digit Span subtest of the Wechsler Adult Intelligence Scale—3rd edition (WAIS-III; total score; Wechsler, 1997). Standardized z-scores were calculated based on the raw mean and SD of the study population and averaged per cognitive domain to create composite scores. The Hospital Anxiety and Depression scale were used as a measure of symptoms of anxiety and depression (Spinhoven et al., 1997). Verbal fluency and the Trail Making Test were available in the control group, but the other additional measures were not.

Statistical analysis

Statistical analyses were performed using SPSS Statistics 25.0 (IBM Corp., Armonk, NY, USA). Between-group differences were analyzed with analysis of variance for continuous data, Jonckheere-Terpstra tests for ordinal data, and χ² tests for dichotomous data. Overall between-group analyses were followed by post-hoc pairwise comparisons, corrected for multiple comparisons. All analyses were adjusted for age and sex. Analyses of the qualitative fluency variables were additionally adjusted for the total number of generated words. In an exploratory analysis, linear regression was used to examine the extent to which the qualitative fluency variables were associated with performance in the domains of language, memory, processing speed, and executive functioning in the whole study population. Hierarchical linear regression analysis including age and sex as covariates in step 1 and the composite cognitive domain scores (language, memory, processing speed, executive functioning) in step 2.

Results

Demographics and general cognitive performance

Patients with bvFTD were younger and more often male than the other patient groups and control participants [age: F(4, 161) = 3.3, p < 0.05, ƞ_p² = 0.08; Sex: χ² = 10.3, p < 0.05; Table 1]. The level of education was comparable between the groups. Patients with lvPPA had a lower MMSE score than all other groups [F(4,152) = 10.3, p < 0.001, ƞ_p² = 0.22]. There was no difference in FAB score or time since symptom onset between the four patient groups (all p > 0.05). The patient groups had more symptoms of depression compared with the control group [F(4, 90) = 3.4, p < 0.05, ƞ_p² = 0.14], but average scores on the HADS Anxiety and Depression subscales were in the subclinical range for all groups. Table 1 also shows the unadjusted between-group differences in performance on the neuropsychological assessment. As expected, the patient groups performed lower than control participants on measures of language (BNT), processing speed (TMT A), and executive functioning (TMT B/A index, mWCST)—measures of memory were not available in the control group. Comparison between the patient groups also showed differences in the expected direction. In particular, patients with svPPA had the lowest naming performance [BNT F(3, 121) = 32.8, p < 0.001, ƞ_p² = 0.46] but spared mental flexibility [mWCST F(3, 66) = 8.8, p < 0.001, ƞ_p² = 0.32]. Patients with lvPPA had the lowest scores in memory functioning [F(3, 85) = 3.1, p = 0.03, ƞ_p² = 0.10].

Table 1. Characteristics of the study sample.

	Control	bvFTD	svPPA	nfvPPA	lvPPA	Statistics*	Difference
n	25	50	27	25	34	–	–
Age (years)	64.6 ± 1.8	60.4 ± 1.3	63.8 ± 1.8	65.0 ± 1.8	67.8 ± 1.6	F(4, 161) = 3.3, p < 0.05, ƞp^2 = 0.08	bvFTD < svPPA = nfvPPA = lvPPA = con
Sex (% male)	13 (52%)	36 (72%)	15 (56%)	12 (48%)	13 (38%)	χ^2 = 10.3, p < 0.05	bvFTD > svPPA = nfvPPA = lvPPA = con
Level of education (median, IQR)	5 (4–5)	5 (4–6)	6 (4–6)	5 (4–5)	5 (4–5)	TJT = 4997, z = −0.23, p = 0.82	–
MMSE	28.3 ± 1.3	25.4 ± 4.3	25.4 ± 3.5	26.6 ± 3.1	22.1 ± 5.0	F(4, 152) = 10.3, p < 0.001, ƞp^2 = 0.22	lvPPA < svPPA = bvFTD < nfvPPA = con
FAB	–	12.2 ± 4.1	14.7 ± 2.7	13.0 ± 3.9	11.6 ± 1.2	F(4, 102) = 2.3, p = 0.09, ƞp^2 = 0.06	–
Time since symptom onset (years)	–	3.3 ± 2.6	3.4 ± 1.8	2.7 ± 1.2	2.5 ± 1.4	F(4, 131) = 1.6, p = 0.20, ƞp^2 = 0.04	–
HADS—anxiety	3.6 ± 2.5	4.9 ± 3.4	6.1 ± 3.8	5.4 ± 3.8	7.1 ± 4.8	F(4, 90) = 2.4, p = 0.054, ƞp^2 = 0.10	–
HADS—depression	1.9 ± 1.9	4.0 ± 3.3	4.6 ± 3.1	5.1 ± 2.9	4.3 ± 4.4	F(4, 90) = 3.4, p < 0.05, ƞp^2 = 0.14	bvFTD = svPPA = nfvPPA = lvPPA > con
Neuropsychological assessment
Boston Naming Test	–	42.5 ± 10.8	17.0 ± 11.1	48.7 ± 7.3	38.7 ± 12.6	F(3, 121) = 32.8, p < 0.001, ƞp^2 = 0.46	svPPA < nfvPPA < lvPPA = bvFTD
RAVLT—immediate	–	29.2 ± 10.4	27.2 ± 9.8	35.0 ± 15.5	23.3 ± 11.5	F(3, 85) = 2.4, p = 0.07, ƞp^2 = 0.08	–
RAVLT—delay	–	5.4 ± 3.8	3.8 ± 2.6	7.8 ± 3.7	4.2 ± 3.4	F(3, 85) = 3.1, p = 0.03, ƞp^2 = 0.10	svPPA = lvPPA < nfvPPA = bvFTD
Digit span	–	19.4 ± 6.0	27.0 ± 2.9	17.5 ± 3.7	12.4 ± 5.6	F(3, 27) = 5.7, p < 0.01, ƞp^2 = 0.43	lvPPA < nfvPPA = bvFTD < svPPA
Trail making part A	41.1 ± 24.8	54.8 ± 23.8	42.2 ± 18.3	60.0 ± 26.5	87.3 ± 43.2	F(4, 1440) = 12.1, p < 0.001, ƞp^2 = 0.26	lvPPA < nfvPPA = bvFTD = svPPA = con
Trail making part B	98.6 ± 57.6	141.3 ± 95.1	90.5 ± 32.9	187.1 ± 111.8	223.2 ± 102.1	F(4, 109) = 8.8, p < 0.001, ƞp^2 = 0.25	lvPPA = nfvPPA < bvFTD = svPPA = con
Trail making B/A index	1.6 ± 0.8	1.8 ± 1.3	1.4 ± 0.6	2.2 ± 1.2	2.5 ± 1.0	F(4, 109) = 4.2, p < 0.01, ƞp^2 = 0.14	lvPPA = nfvPPA < bvFTD = svPPA = con
mWCST (concepts)	–	2.8 ± 2.1	5.6 ± 1.3	2.4 ± 1.5	2.4 ± 1.6	F(3, 66) = 8.8, p < 0.001, ƞp^2 = 0.32	lvPPA = nfvPPA = bvFTD < svPPA

Data are mean ± SD or n (%) unless otherwise specified.

*Statistical analysis shows unadjusted between-group differences.

Between group comparison of fluency variables

Category fluency

Table 2 shows the verbal fluency variables for the four patient groups and the control participants. For category fluency, between group differences were found in the total number of words [F(4, 159) = 23.1, p < 0.001, ƞ_p² = 0.38], mean LF [F(4, 159) = 4.9, p < 0.01, ƞ_p² = 0.11], and mean AoA [F(4, 159) = 3.8, p < 0.01, ƞ_p² = 0.09].

Table 2. Quantitative and qualitative measures of verbal fluency.

	control	bvFTD	svPPA	nfvPPA	lvPPA	Statistics*	Difference**
Category fluency
Total number of words	21.3 ± 5.6	13.3 ± 5.5	8.6 ± 4.8	11.4 ± 4.8	10.7 ± 4.6	F(4, 159) = 23.1, p < 0.001, ƞp^2 = 0.38	svPPA < lvPPA = nfvPPA = bvFTD < con
Switches	9.0 ± 3.0	5.8 ± 3.1	4.2 ± 3.1	4.8 ± 2.8	4.7 ± 2.5	F(4, 159) = 0.2, p = 0.93, ƞp^2 = 0.006	–
Clusters	5.8 ± 1.9	3.9 ± 1.6	2.7 ± 1.7	3.0 ± 1.6	3.1 ± 1.5	F(4, 159) = 1.2, p = 0.38, ƞp^2 = 0.03	–
Mean cluster size	3.1 ± 0.6	3.2 ± 1.1	2.6 ± 1.3	3.4 ± 2.0	2.8 ± 0.9	F(4, 159) = 1.5, p = 0.20, ƞp^2 = 0.04	–
Lexical frequency	3.8 ± 0.2	3.9 ± 0.3	4.2 ± 0.3	3.9 ± 0.3	4.0 ± 0.4	F(4, 159) = 4.9, p < 0.01, ƞp^2 = 0.11	bvFTD = nfvPPA = lvPPA < svPPA = con
Age of acquisition	6.0 ± 0.5	5.6 ± 0.7	4.9 ± 0.5	5.7 ± 0.7	5.3 ± 0.7	F(4, 159) = 3.8, p < 0.01, ƞp^2 = 0.09	svPPA < bvFTD = lvPPA = nfvPPA < con
Neighborhood density	8.8 ± 2.6	9.1 ± 3.7	11.1 ± 4.2	8.9 ± 3.9	9.9 ± 3.6	F(4, 159) = 1.2, p = 0.33, ƞp^2 = 0.03	–
Word length	5.3 ± 0.4	5.2 ± 0.7	5.1 ± 1.0	5.4 ± 0.7	4.9 ± 0.7	F(4, 159) = 1.2, p = 0.34, ƞp^2 = 0.03	–
Letter fluency
Total number of words	35.1 ± 10.5	18.6 ± 10.3	22.5 ± 11.6	14.0 ± 9.2	15.7 ± 8.9	F(4, 155) = 17.9, p < 0.001, ƞp^2 = 0.32	nfvPPA = lvPPA < svPPA = bvFTD < con
Switches	23.1 ± 7.5	13.0 ± 8.5	16.7 ± 9.3	8.2 ± 6.0	10.9 ± 7.0	F(4, 155) = 2.6, p < 0.05, ƞp^2 = 0.07	svPPA > con = nfvPPA
Clusters	8.4 ± 3.3	4.6 ± 3.6	4.1 ± 2.8	3.5 ± 3.0	3.6 ± 2.6	F(4, 155) = 2.7, p < 0.05, ƞp^2 = 0.07	svPPA < lvPPA = bvFTD = nfvPPA = con
Mean cluster size	2.5 ± 0.3	4.2 ± 2.6	3.5 ± 2.3	3.6 ± 2.4	4.0 ± 2.5	F(4, 155) = 6.1, p < 0.001, ƞp^2 = 0.14	con < svPPA = nfvPPA = lvPPA = bvFTD
Lexical frequency	3.9 ± 0.3	4.1 ± 0.5	4.4 ± 0.3	3.9 ± 0.6	4.2 ± 0.5	F(4, 155) = 5.5, p < 0.001, ƞp^2 = 0.13	svPPA > bvFTD = nfvPPA = con
Age of Acquisition	7.7 ± 0.8	7.2 ± 1.0	6.7 ± 0.6	7.0 ± 1.0	6.5 ± 1.0	F(4, 155) = 4.0, p < 0.01, ƞp^2 = 0.10	con = bvFTD = nfvPPA > svPPA = lvPPA
Neighborhood density	6.3 ± 2.1	7.3 ± 4.0	8.7 ± 2.8	8.4 ± 4.0	9.7 ± 4.7	F(4, 155) = 3.3, p < 0.05, ƞp^2 = 0.08	lvPPA > bvFTD = con
Word length	6.0 ± 0.7	5.9 ± 0.9	5.3 ± 0.7	5.6 ± 1.2	5.2 ± 1.1	F(4, 155) = 2.6, p < 0.01, ƞp^2 = 0.11	con = bvFTD > svPPA = lvPPA

Data are unadjusted mean ± SD.

*Analyses were adjusted for age, sex and total number of words. As a result of adjustment for these covariates apparent differences in raw scores (or lack thereof) not always directly correspond with the between-group differences indicated in the most right column.

**Pairwise analyses were adjusted for multiple comparisons.

Table 3. Relation between qualitative fluency variables and other cognitive domains.

	Language	Memory	Processing speed	Executive functioning
Category fluency
Total number of words	0.32 (0.12–0.52)	0.39 (0.19–0.59)	0.25 (0.03–0.45)	0.16 (−0.01–0.35)
Switches	0.21 (−0.05–0.46)	0.22 (−0.03–0.46)	0.15 (−0.11–0.41)	0.06 (−0.18–0.30)
Clusters	0.19 (−0.03–0.42)	0.33 (0.09–0.56)	0.23 (−0.02–0.47)	0.09 (−0.12–0.32)
Mean cluster size	0.12 (−0.15–0.38)	0.09 (−0.17–0.36)	0.11 (−0.17–0.38)	0.07 (−0.18–0.32)
Lexical frequency	0.41 (0.20–0.62)	0.27 (0.05–0.48)	0.11 (−0.11–0.33)	0.09 (−0.11–0.30)
Age of acquisition	0.39 (0.19–0.60)	0.32 (0.11–0.53)	0.12 (−0.10–0.33)	0.05 (−0.14–0.25)
Neighborhood density	0.26 (0.01–0.50)	0.12 (−0.12–0.37)	−0.02 (−0.28–0.24)	0.05 (−0.18–0.29)
Word length	0.19 (−0.06–0.44)	0.13 (−0.12–0.38)	0.01 (−0.25–0.27)	0.05 (−0.19–0.29)
Letter fluency
Total number of words	0.02 (−0.21–0.25)	0.28 (0.05–0.51)	0.28 (0.03–0.52)	0.29 (0.07–0.51)
Switches	−0.06 (−0.30–0.19)	0.20 (−0.04–0.45)	0.24 (−0.02–0.50)	0.25 (0.01–0.48)
Clusters	0.17 (−0.08–0.41)	0.20 (−0.04–0.44)	0.19 (−0.06–0.44)	0.25 (0.01–0.48)
Mean cluster size	0.17 (−0.09–0.43)	0.03 (−0.23–0.29)	0.13 (−0.15–0.40)	0.07 (−0.18–0.32)
Lexical frequency	0.38 (0.14–0.62)	−0.005 (−0.25–0.24)	−0.13 (−0.38–0.13)	−0.11 (−0.35–0.12)
Age of Acquisition	0.25 (−0.002–0.50)	0.10 (−0.15–0.35)	0.01 (−0,25–0.27)	0.11 (−0.13–0.35)
Neighborhood density	0.20 (−0.05–0.46)	0.02 (−0.24–0.28)	−0.12 (−0.39–0.14)	0.10 (−0.14–0.35)
Word length	0.25 (−0.004–0.50)	0.04 (−0.22–0.29)	−0.17 (−0.44–0.09)	−0.03 (−0.26–0.22)

Data are standardized regression coefficient Beta (95% confidence interval). All regression models were adjusted for age and sex. Significant associations are depicted in bold. Data for lexical frequency and neighborhood density were inversed.

Post-hoc pairwise comparisons showed that all patient groups produced fewer words and had a lower AoA than control participants. Patients with svPPA produced the lowest number of words with a higher mean LF and lower mean AoA compared with nfvPPA, lvPPA and bvFTD [(vs. nfvPPA: mean difference LF 0.25 (95% CI 0.11–0.40); AoA −0.54 (−0.83 to −0.25)); vs. lvPPA (LF 0.13 (0.001–0.27); AoA −0.30 (−0.57 to −0.03)); vs. bvFTD (LF 0.18 (0.05–0.31); AoA −0.30 (−0.57 to −0.04))]. The qualitative linguistic fluency variables of the patients with bvFTD were similar to the control participants, except for a higher LF and lower AoA.

There was no overall between group difference in the number of clusters [F(4, 159) = 1.2, p = 0.38, ƞ_p² = 0.03], switches [F(4, 159) = 0.2, p = 0.93, ƞ_p² = 0.006], ND [F(4, 159) = 1.2, p = 0.33, ƞ_p² = 0.03] or WL [F(4, 159) = 1.2, p = 0.34, ƞ_p² = 0.03].

Letter fluency

For letter fluency between group differences were found in the total number of words, clustering and switching, and all linguistic variables (Table 2). All patient groups produced fewer words [F(4, 155) = 17.9, p < 0.001, ƞ_p² = 0.32] than control participants. Patients with lvPPA and patients with nfvPPA produced fewer words than patients with svPPA and patients with bvFTD.

All patient groups had a lower mean cluster size [F(4, 155) = 6.1, p < 0.001, ƞ_p² = 0.14] than control participants. Patients with svPPA produced fewer clusters than the other patient groups (mean difference with nfvPPA −1.54 (95%CI −2.66 to −0.43); lvPPA (−1.21 (−2.36 to −0.07); bvFTD −1.39 (−2.34 to −.43). Patients with svPPA also produced smaller clusters [−0.64 (−0.89–0.61)] and used more switches [2.66 (0.66–4.67)] than patients with nfvPPA.

With regard to the linguistic variables patients with svPPA could be distinguished from the other groups by a higher LF [F(4, 155) = 5.5, p < 0.001, ƞ_p² = 0.13]. Patients with svPPA and patients with lvPPA produced shorter words with a lower AoA compared with patients with bvFTD and control participants (Table 2). Patients with lvPPA produced words with a higher ND than all other groups [F(4, 155) = 3.3, p < 0.05, ƞ_p² = 0.08]. Control participants and patients with bvFTD produced longer words than patients with PPA [F(4, 155) = 2.6, p < 0.01, ƞ_p² = 0.11]. No differences in the qualitative linguistic fluency variables between patients with bvFTD and control participants were observed.

Association with other cognitive functions

Linear regression analysis was used to examine the extent to which the qualitative fluency variables were predicted by performance in the domains of language, memory, processing speed, and executive functioning (Table 3). In category fluency, the number of clusters was associated with memory [standardized beta coefficient 0.33 (95% CI 0.09−0.56)]. LF and AoA were associated with language [0.41 (0.20−0.62); 0.39 (0.19−0.60)] and memory [0.27 (0.05−048); 0.32 (0.11−0.53)]. ND was exclusively associated with language [0.26 (0.01−0.50)]. In letter fluency the number of switches and clusters were associated with executive functioning [0.25 (0.01−0.48); 0.25 (0.01−0.48)]. LF was associated with language [0.38 (0.14−0.62)]. No other significant associations were found between the linguistic fluency variables and the cognitive domain scores.

Discussion

The present study examined differences in qualitative, linguistic features of verbal fluency in patients with bvFTD and PPA. The main results showed that in letter fluency patients with svPPA produce fewer clusters than all other groups and used smaller clusters and more switches than patients with nfvPPA. With regard to the linguistic variables patients with svPPA could be distinguished from the other patient groups as they produced words with a higher LF and a lower AoA. In contrast, patients with lvPPA specifically produced words with a higher ND than the other patient groups. Control participants and patients with bvFTD produced longer words than patients with any type of PPA. Clustering was associated with memory performance in category fluency, but both clustering and switching were associated with executive functioning in letter fluency. LF, AoA, and—to a lesser extent—ND were associated with language (naming) and memory in category fluency. In letter fluency, only LF was predicted by language. Word length was not associated with the other cognitive functions.

These findings corroborate the results of previous studies that showed a specific effect of LF and AoA in svPPA reflecting the decay of semantic (conceptual) processing (Marczinski & Kertesz, 2006; Vonk, Jonkers, et al., 2019). The absence of LF and/or AoA effects in nfvPPA underlines the relative sparing of semantic processing in this subtype of PPA, which has been previously reported as well (Carthery-Goulart et al., 2012). In contrast, ND was increased in lvPPA compared with the other groups, indicating a specific deficit at the word-form (lexeme) level as opposed to the conceptual level of lexical retrieval, similar to the finding by Vonk, Jonkers, et al. (2019). Activation of phonological neighbors is known to facilitate word-finding (Brysbaert & Ghyselinck, 2006), which is a key deficit in lvPPA. Higher ND in lvPPA may indeed facilitate the retrieval of early-acquired words (Karimi & Diaz, 2020), possibly resulting in the relatively low mean AoA in letter fluency observed in the present study. Although apraxia of speech has been shown to negatively impact the production of words with a lower ND (Laganaro et al., 2012), no effect of ND was observed in nfvPPA in the present study. These specific differences in qualitative linguistic fluency features between the subtypes of PPA were in line with the recent pathophysiological synthesis of PPA proposed by Ruksenaite et al. (2021) and may help elucidate critical issues around the diagnosis of PPA subtypes, including disease staging and diagnosis of atypical subtypes that do not fulfill current diagnostic criteria. The between-group differences in clustering and switching were similar to our previous report, albeit somewhat less pronounced (van den Berg et al., 2017). Moreover, although the qualitative linguistic variables of the bvFTD group were largely in line with the control group, a higher LF and lower AoA were found in category fluency corroborating previous deficits in verbal semantic processing in bvFTD (Hardy et al., 2016).

There are few studies that have reported on the cognitive processes associated with qualitative features of verbal fluency. The finding that clustering/switching is mainly supported by memory in category fluency and by executive functioning in letter fluency fits the disparity projected by the theoretical model of verbal fluency proposed by Troyer et al. (1997). In this model, category fluency is thought to rely on temporal lobe processes, such as verbal memory and word storage, whereas letter fluency relies more upon frontal lobe processes, such as strategic search processes, cognitive flexibility, and set-shifting ability (Troyer et al., 1997). Additional support for this theoretical distinction comes from MRI studies (Baldo et al., 2006; Vonk, Rizvi, et al., 2019) and clinical patient samples with frontal and/or temporal lesions (e.g. Okruszek et al., 2013). As would be expected, the linguistic fluency variables (LF, AoA, and ND) were associated with confrontation naming (language). This finding probably reflects the similarity in the process of lexical retrieval that is involved in both types of tasks and has been reported in previous studies as well (Libon et al., 2009). In addition, the (modest) association between AoA, LF, and (episodic) memory that is found in the present study, fits the semantic locus hypothesis (Brysbaert et al., 2000), which claims that late learned words are incorporated in semantic/memory representation already existing in early learned words. Similar associations between LF, AoA and short term memory and episodic memory have also been previously reported (Badham et al., 2017; Roodenrys et al., 1994). Of note, the associations with cognitive domains that were reported in the present study are generally modest in size (range regression coefficients ∼0.2 to 0.4) indicating that other explanatory variables should be considered as well.

Several strengths and weaknesses of the present study need to be addressed. The primary strengths of the present study were the substantial size of the patient sample—on average double the number of patients compared with previous studies on this subject, the inclusion of multiple qualitative, linguistic variables allowing for a detailed examination of between-group differences, and the high interrater reliability for clustering and switching. A limitation of the present study is the inherent heterogeneity of a clinical FTD sample and the lack of postmortem pathological confirmation of diagnosis. The patients included in the present study had a relatively mild disease severity (mean MMSE 22–26) and a moderate time since symptom onset (∼3 years). The qualitative, linguistic features of verbal fluency that were examined in the present study thus appear to be sensitive in discriminating subtypes of FTD early in the course of the disease, but whether these results could be generalized to more severe disease stages remains to be evaluated. Importantly, the absence of a significant difference in time since symptom onset between the groups does not necessarily rule out heterogeneity in the disease stage as the disease course differs between subtypes of FTD (Kertesz et al., 2007), which makes it difficult to truly match the patient groups. Since only a modest number of neuropsychological tests were available in all groups the analysis of the association between the qualitative fluency variables and other cognitive functions was exploratory in nature and more thorough analysis—additionally including brain imaging measures—is warranted.

The results of the present study underscore the clinical value of verbal fluency as a sensitive and specific diagnostic measure in neuropsychological assessment. Of note, the differential diagnosis of FTD/PPA subtypes is not (only) based on verbal fluency performance and relies on many other indices—such as blood and imaging biomarkers—besides verbal fluency. In addition to the total number of words that are traditionally recorded, our results encourage clinicians and researchers to evaluate qualitative features of verbal fluency as well. Databases summarizing LF, AoA, and ND are available in many languages and relatively easy to use, but even a simple visual inspection of the actual words individual patients generate during neuropsychological assessment (e.g. generating “bird, dog, cat” differs from “canary, sparrow, seagull”) in verbal fluency may aid discrimination between subtypes of neurodegenerative disorders and between neurodegenerative and primary psychiatric disorders as well. Our result also highlights the potential of developing linguistic interventions in PPA van bvFTD to support conversational speech in natural communication (Volkmer et al., 2020).

In sum, qualitative linguistic features of verbal fluency provide additional information that can be used to differentiate between subtypes of FTD and shed light on the cognitive processes involved.

Disclosure statement

The authors report no conflict of interest.

Additional information

Funding

This work was supported by the Dioraphte Foundation [grant number 09-02-03-00]; the Association for Frontotemporal Dementias Research Grant 2009; The Netherlands Organization for Scientific Research (NWO) grant HCMI [grant number 056-13-018]; ZonMw Memorabel (Deltaplan Dementie) [project numbers 733 051 042 and 733 050 103]; JPND PreFrontAls consortium project number 733051042; NIA K99/R00 award (K99AG066934); NWO/ZonMw Veni Grant (project number 09150161810017); Alzheimer Nederland and the Bluefield project. Several authors of this publication are members of the European Reference Network for Rare Neurological Diseases—Project ID No 739510.