Fluorodeoxyglucose F18 Positron Emission Tomography in Progressive Apraxia of Speech and Primary Progressive Aphasia Variants

Objectives  To determine patterns of hypometabolism on fluorodeoxyglucose F18 positron emission tomography (FDG-PET) in patients with progressive apraxia of speech (PAS) and primary progressive aphasia (PPA) variants and to use these patterns to further refine current classification.

Design  We identified all patients who had FDG-PET and PAS or PPA who were evaluated by an expert speech-language pathologist. Patterns of hypometabolism were independently classified by 2 raters blinded to clinical data. Three speech-language pathologists reclassified all patients into 1 of 7 operationally defined categories of PAS and PPA blinded to FDG-PET data.

Setting  Tertiary care medical center.

Patients  Twenty-four patients with PAS or PPA and FDG-PET.

Main Outcome Measure  Fluorodeoxyglucose F18 PET hypometabolic pattern.

Results  Of the 24 patients in the study, 9 had nonfluent speech output; 14, fluent speech; and 1 was unclassifiable. Twenty-one patients showed FDG hypometabolism; the remaining 3 did not. Among the patients showing hypometabolism, 8 had a prerolandic pattern of which 7 had nonfluent speech including progressive nonfluent aphasia (n = 3), PAS (n = 1), and mixed nonfluent aphasia/apraxia of speech (n = 3); the other patient had PPA unclassifiable. The remaining 13 had a postrolandic pattern, all with fluent speech (P < .001), including logopenic progressive aphasia (n = 6), progressive fluent aphasia (n = 6), and semantic dementia (n = 1). Patterns of hypometabolism differed between the nonfluent variants and between the fluent variants, including progressive fluent aphasia.

Conclusion  Patterns of FDG-PET hypometabolism support the clinical categorizations of fluency, the distinction of apraxia of speech from progressive nonfluent aphasia, and the designation of a progressive fluent aphasia category.

Progressive speech and language disorders are commonly referred to as primary progressive aphasia (PPA).¹ Patients with PPA are commonly categorized into those in whom speech output is nonfluent and those in whom speech output is fluent. We have previously shown that those with nonfluent speech output can be further subdivided based on the presence or absence of apraxia of speech (AOS).² Those with fluent output are typically classified as having either semantic dementia (SD)³ or logopenic progressive aphasia (LPA).⁴ However, we have evaluated many patients who meet criteria for PPA and are fluent yet do not meet published criteria for either SD or LPA. We have been classifying these fluent patients as having progressive fluent aphasia (PFA).

Fluorodeoxyglucose F18 positron emission tomography (FDG-PET) allows the indirect assessment of neuronal activity by measuring glucose metabolism in the brain. Fluorodeoxyglucose F18 PET has shown characteristic patterns of hypometabolism in some of the different PPA variants,^5,6 although the patterns of hypometabolism have not been well defined in those with AOS and only 1 study has systematically assessed FDG-PET patterns of hypometabolism in the LPA variant.⁷ Given that FDG-PET is sensitive to early changes in the brain, we used FDG-PET in this study to examine the pattern of hypometabolism in all variants of PPA, particularly in those with AOS and those with PFA.

The Mayo Clinic Medical Records Linkage system was queried to identify all patients seen in the Department of Neurology between January 1, 2005, and September 30, 2008, who had a clinical diagnosis of progressive aphasia or AOS and had an FDG-PET scan. To capture patients with progressive aphasia or AOS, the following search terms were used: nonfluent aphasia, anomic aphasia, anomia, semantic dementia, logopenic aphasia, fluent aphasia, and apraxia of speech. A total of 36 patients were identified. We then excluded patients who did not have a comprehensive speech and language assessment by a speech-language pathologist, leaving a total of 24 patients. A behavioral neurologist reviewed the clinical records of all 24 patients to extract clinical data, including age at onset, age at scan, sex, initial presenting symptoms, and Short Test of Mental Status⁸ scores at the time of PET scan. The study was approved by the Mayo institutional review board.

Detailed speech and language assessments were independently reviewed by 3 speech-language pathologists (J.R.D., T.R.F., and E.A.S.), blinded to FDG-PET data, to classify the 24 patients using operational definitions given later. There were a total of 35 judgment opportunities, with 24 initial evaluations, 8 patients with a second evaluation, and 3 patients with a third evaluation (Table 1 and Table 2). Classification was performed independently for each evaluation and was based on both qualitative and quantitative data. For any evaluation for which there was disagreement about final classification, records were re-reviewed, discussed, and consensus reached. If consensus could not be reached, classification was based on the majority opinion, ie, 2 of 3 judges agreed.

Language examination consisted of assessments of verbal comprehension and expression, reading, and writing. Specific tests most often included several subtests from the Minnesota Test for Differential Diagnosis of Aphasia,¹⁰ the Boston Naming Test (BNT),¹¹ part V of the Token Test,⁹ and a letter/word fluency task.¹² Both quantitative and qualitative data were used to estimate aphasia severity. The diagnosis of aphasia was based on the presence of behaviors common in aphasia (eg, cross-modality deficits, word-finding difficulties, agrammatism, semantic and/or phonological errors) that could not be explained by nonaphasic deficits (eg, confusion, disorientation). Perceptual characteristics used to diagnose AOS were consistent with current diagnostic criteria^13-16; diagnosis was based on the presence of at least 2 of the disorder's recognized features. The diagnosis of dysarthria was based on guidelines provided in Duffy.¹³ Dysarthria was diagnosed when the pattern of abnormal speech characteristics that occurs with a clinically recognized type of dysarthria was present. Dysarthria was identified as separate from AOS by the presence of speech characteristics not associated with AOS (eg, increased speaking rate, strained vocal quality, resonance abnormalities, reduced vocal loudness).

Aphasia is present. Verbal output characteristics contain evidence of agrammatism or telegraphic speech. Difficulties with verbal and reading comprehension and writing can be present, as can anomia. Target words are recognized when provided on the majority of BNT errors. Apraxia of speech is mild or not present. Dysarthria can be present.

Apraxia of speech is the sole or dominant feature of the communication disorder. Aphasia can be present but must be less prominent than the AOS. Anomia may be present but target words are recognized when provided on the majority of BNT errors. Dysarthria can be present.

Both AOS and nonfluent aphasia (NFA) are present and almost equal in severity. Anomia may be present but target words are recognized when provided on the majority of BNT errors. Dysarthria can be present.

Aphasia is present. Apraxia of speech and dysarthria are not unequivocally present. Verbal output is fluent (ie, grossly normal in grammar and syntax, average phrase length, and prosodically, excluding pauses for word retrieval). Anomia is evident with apparent loss of single-word meaning (inability to name objects and to recognize target words when provided for the majority of BNT items that could not be named with semantic and phonemic cues, or at least 1 comment by the patient during examination suggesting loss of word meaning [eg, “I don't know what ‘chin’ means.”]).

Aphasia is present. Apraxia of speech and dysarthria are not unequivocally present. Verbal output characteristics are not agrammatic or telegraphic, but syntax may be simple and output may be reduced (ie, brief and unelaborated). Speech rate may be slow because of frequent pauses for apparent word retrieval efforts (as opposed to motor deficits of AOS or dysarthria). Sentence repetition and sentence comprehension are typically impaired and phonemic paraphasias (ie, nondistorted sound substitutions within utterances that are normal in rate and prosody) may be evident. Anomia is present but target words are recognized when provided on the majority of BNT errors.

These patients with aphasia have normal grammar and syntax and average phrase length and prosody. They may have prominent anomia but have no obvious loss of word meaning (ie, not semantic dementia). Their verbal output is not brief or unelaborated, repetition may be relatively good, and phonemic paraphasias may or may not be evident. Basically, these patients are reliably classified as being “fluent” but they do not clearly fall into the category of SD or LPA.

Aphasia is present. Speech characteristics do not allow a clear-cut determination of whether the patient's speech is fluent or nonfluent. They do not reliably meet criteria for any other category.

Each patient was injected with 629 MBq of FDG (range, 555-740 MBq). After a 25- to 30-minute uptake period, the patient was positioned on the scanner bed with instructions to remain motionless. Five minutes prior to emission acquisition, a helical computed tomography image was obtained for attenuation correction. The FDG scan was acquired 30 minutes after injection for a period of 8 minutes using four 2-minute dynamic frames. The PET sinograms were reconstructed using a reprojection algorithm into a 256-mm field of view; pixel size, 1.0 mm; and slice thickness, 3.3 mm. The emission images were corrected for attenuation using the measured attenuation coefficients derived from the computed tomography images. Standard corrections were also applied. An absolute calibration correction, which converts the image intensity into activity concentration, was used. Individual frames of the FDG dynamic series were realigned if motion was detected and summed to create static FDG images for analysis.

We used the clinical tool of 3-dimensional stereotactic surface projections for the interpretation of the FDG-PET. This is a fully automated analysis. All scans were realigned and spatially normalized and underwent nonlinear warping. For the 3-dimensional stereotactic surface projections technique, the scans were sampled at 16 000 predefined cortical locations and projected on a 3-dimensional image. The activity in each subject's PET data set was normalized to the pons and compared with an age-segmented normative database, yielding a 3-dimensional stereotactic surface projections z score image. The image produced by this analysis produces a metabolic map using the z scores as calculated for each surface pixel.¹⁷ The software packages used to perform these analyses included CortexID (GE Healthcare, Waukesha, Wisconsin).

Patterns of hypometabolism were independently visually classified by 2 raters (K.A.J. and J.L.W.) blinded to speech and language classifications. The raters first categorized the patterns as predominantly prerolandic or postrolandic. The prerolandic patterns were then further classified as widespread, focal inferior frontal, focal superior frontal, or focal inferior and superior frontal, and the presence or absence of supplementary motor area hypometabolism was noted. The postrolandic patterns were further classified as anteromedial temporal or temporoparietal dominant. These classifications were based on previous magnetic resonance imaging or FDG-PET studies showing regions of the brain affected by progressive apraxia of speech (PAS) and PPA.^2,4,5 Interrater reliability was 100%.

To validate the visual assessment, automated average z scores generated from CortexID were calculated for the following regions of interest: right and left lateral frontal, medial frontal, temporal, lateral parietal, medial parietal, and occipital cortices. z Scores greater than 2 were considered significant.

Statistical analyses used JMP software, version 6.0.0 (SAS Institute Inc, Cary, North Carolina) with statistical significance set at P < .05. The Mann-Whitney U test was used to analyze differences across groups for all continuous variables and χ², for nominal data.

The speech and language features for each of the 24 patients are shown in Table 1 and Table 2. Of the 24 patients, 9 were classified as having nonfluent speech output, 14 were classified as having fluent speech output, and in 1 subject fluency was unclassifiable. One of the 9 patients with nonfluent speech (patient 3) was unclassifiable at his first speech-language evaluation but was classified as having NFA at the second and third evaluations. Three other patients with nonfluent speech output were classified as having NFA; 3, as having NFA-AOS; and 2, as having as PAS. Of the 14 patients with fluent speech output, only 1 was classified as having SD, 8 were initially classified as having PFA, and 5, as having LPA. Two patients initially classified as having PFA were later classified as having LPA at their second assessment. Distorted phonetic sound production errors were identified in 7 of the 9 nonfluent patients compared with only 2 of the 14 fluent patients (P = .003). In contrast, nondistorted phonological sound substitutions were identified in 3 of the 9 nonfluent patients but in 9 of the 14 fluent patients (P = .21).

For 2 of 3 judges, agreement was 100% (35 of 35) while for 3 of 3 judges, agreement was 71% (25 of 35). Of the 10 instances when 2 judgments agreed but the third did not, 6 (60%) involved the LPA variant vs PFA, 2 PPA unclassified vs LPA, and 1 each for PPA unclassified vs NFA and SD vs PFA. There were no disagreements that involved AOS variants. Disagreement typically centered on how frequently telegraphic/agrammatic or brief and unelaborated utterances had to occur to make the fluent vs nonfluent distinction; for example, sometimes a patient would produce long, complex sentences while at other times the patient might be brief and unelaborated. Three of the 10 instances in which 1 judge was not in agreement involved the PPA unclassified diagnosis because there was ambiguity about the fluent vs nonfluent distinction.

The FDG-PET patterns of hypometabolism at the time of speech and language assessment are shown in Table 3 and Figures 1, 2, and 3. Twenty-one of the 24 patients had some evidence of FDG-PET hypometabolism while 3 did not. The FDG-PET patterns of these 21 could be divided into prerolandic (n = 8) and postrolandic (n = 13) patterns of hypometabolism. Of the 8 patients with prerolandic patterns, the speech and language classification was nonfluent in 7, including NFA (n = 3), NFA-AOS (n = 3), and PAS (n = 1), while all patients with postrolandic patterns were associated with fluent speech (P < .001), including LPA (n = 6), PFA (n = 6), and SD (n = 1). The eighth case with a prerolandic pattern of hypometabolism had PPA unclassified. Of the 7 nonfluent patients with prerolandic patterns, those with NFA typically showed a widespread pattern of prerolandic hypometabolism, with particular emphasis on the inferior frontal lobe (Figure 1). Those with AOS, either NFA-AOS or PAS, typically showed a pattern of more focal superior frontal and supplementary motor area involvement (Table 3) (Figure 1). Of the 13 fluent patients with postrolandic patterns, 2 distinct patterns were visible. In 3 patients (2 with PFA and 1 with SD), hypometabolism was localized to the anteromedial temporal lobe (Figure 2) while temporoparietal hypometabolism was observed in the remaining 10 patients (6 with LPA and 4 with PFA) (Figure 2). A greater number of patients with LPA appeared to have medial parietal hypometabolism (5 of 6) compared with PFA (1 of 6). One patient with PFA with anteromedial temporal hypometabolism (patient 21) had a second FDG-PET scan 2 years later that showed progression of hypometabolism to the posterior left temporal lobe (Figure 2). Two patients were initially classified as having PPA unclassified (patients 3 and 24). Both showed a prerolandic pattern of hypometabolism (Figure 3). The FDG-PET at the second evaluation in patient 3, when classified as NFA, demonstrated a progressed and widespread pattern of prerolandic hypometabolism, similar to the pattern observed in patient 24 (Figure 3).

The quantitative regions-of-interest analysis using a z score cutoff of 2.0 essentially validated the visual assessments (Table 4). The NFA and NFA-AOS groups showed left lateral and medial frontal hypometabolism that was more widespread in the NFA group, while the PAS group showed very little hypometabolism. The patient with SD showed hypometabolism localized to the left anteromedial temporal lobe, as did the PFA group. The LPA group had bilateral temporoparietal and left lateral frontal and medial parietal involvement.

There was no difference in sex, age at onset, age at FDG-PET, or time from onset to scan between the LPA and PFA groups at the time of FDG-PET (Table 4). However, those with PFA performed significantly better on the Short Test of Mental Status⁸ (mean score, 27 of 38 vs 18 of 38; P = .02), a general measure of disease severity, and on part V of the Token Test (5 vs 10 errors; P = .03), which is sensitive to syntactic comprehension deficits, and there was a trend for fewer patients to make phonological errors in the PFA group (55% of patients made phonological errors) compared with the LPA group (100%; P = .08).

We have demonstrated that FDG-PET patterns of hypometabolism corresponded well to our designation of speech output as fluent vs nonfluent and that there are distinctive patterns of hypometabolism for specific nonfluent and specific fluent syndromes. We also observed that the patterns of hypometabolism in our patients with PFA were not homogeneous and that at the group level there are clinical and FDG-PET imaging differences between PFA and LPA.

When we subdivided our patients into those with nonfluent speech output vs those with fluent speech output we found 100% association with a prerolandic vs postrolandic pattern of hypometabolism, respectively. The group of patients with nonfluent speech output consisted of 3 speech-language syndromes that we have previously reported, including PAS, NFA-AOS, and NFA.^2,18 Those in whom AOS was dominant were associated with more superior frontal and supplementary motor cortex hypometabolism, as we previously reported using voxel-based morphometry² and was reported on FDG-PET in a single subject.¹⁹ We did not identify the insula as previously shown in stroke patients.²⁰ However, the syndrome of NFA, defined predominantly as a language disorder with little or no motor speech impairment (AOS), appears to be associated with more posterior inferior frontal lobe (the Broca area) involvement, an area that has previously been associated with NFA,^2,4,21 and then later with a more widespread pattern of frontal hypometabolism. These findings suggest that the AOS may be originating from the superior frontal lobe and supplementary motor cortex in progressive aphasia, while agrammatism and telegraphic speech is originating from the posterior inferior frontal cortex. These findings are in keeping with our previous suggestion that patients with AOS (either PAS or NFA-AOS) should not be lumped with those with NFA.² Given the patterns of hypometabolism that we observed in our nonfluent patients, we hypothesize a downward progression from the posterior superior frontal and supplementary motor cortex to the posterior inferior frontal lobe in patients presenting with dominant AOS who later develop nonfluent aphasia. On the contrary, in NFA where aphasia is the earliest presenting feature with little or no AOS, we hypothesize early posterior inferior frontal lobe impairment progressing to more widespread frontal lobe involvement, as was observed in 1 patient with 2 FDG-PET scans (patient 3).

The majority of the patients in our cohort showed a postrolandic pattern of hypometabolism and were classified as having fluent speech. Based on our operational definition, only 1 patient met criteria for SD and showed the expected pattern of left anteromedial hypometabolism.^5,22 Six patients met criteria for LPA based on our operational definition and showed a predominant pattern of lateral temporoparietal and medial parietal hypometabolism, with some patients also having left lateral frontal hypometabolism. This pattern has been reported for patients with LPA on magnetic resonance imaging,⁴ as well as with FDG-PET.⁷

As noted earlier, we have examined patients who are aphasic and predominantly fluent but who do not meet criteria for either SD³ or LPA⁴; they are referred to as having PFA since they all show a progressive fluent aphasia. The patients with PFA in this study seemed to separate into 2 groups based on the patterns of hypometabolism observed for each patient with PFA. Similar to SD,⁵ significant hypometabolism was located in the left anteromedial temporal lobe in 2 patients with PFA. However, the other patients with PFA had an FDG-PET pattern of hypometabolism that was more similar to the patients with LPA with involvement of more posterior temporoparietal regions.⁴ One could interpret this to mean that our patients with PFA are simply early in their disease course and hence will likely progress to either SD or LPA later. In support of this hypothesis, 2 of our patients with PFA did later convert to LPA. In addition, patients with PFA had less severe disease (as measured by the Short Test of Mental Status⁸), performed better on part V of the Token Test, and showed a trend for fewer phonological errors than patients with LPA. Furthermore, for 6 patients we had difficulty in differentiating PFA from LPA on the basis of the clinical data.

It is also possible that some patients classified as having PFA will not progress to either SD or LPA. Supporting this hypothesis is the fact that 2 patients with PFA were evaluated more than once and neither converted to LPA or SD. Furthermore, it is unlikely that many patients with PFA will convert to SD given that the patterns of hypometabolism in most patients with PFA were characterized by temporoparietal involvement, a feature that is not typical for SD. Also arguing against PFA always progressing to LPA is the fact that the severity of the temporoparietal hypometabolism for patients with PFA was not always less than the severity of temporoparietal hypometabolism for patients with LPA (for example, patient 20, who was evaluated twice, did not convert to LPA).

It remains to be determined whether PFA is different from LPA. This could have significant clinical implications. Based on amyloid imaging, it has been suggested that patients with LPA will likely show Alzheimer-type pathology.⁷ In keeping with that suggestion are autopsy-confirmed patients who met criteria for LPA who showed Alzheimer pathology.^2,23,24 Not surprisingly, therefore, is the fact that our patients with LPA had medial parietal hypometabolism, which is also a feature of Alzheimer disease.^25,26 However, medial parietal hypometabolism was only observed in 1 patient with PFA and was relatively spared in the regions-of-interest analysis. Therefore, it is possible that some, maybe even most, patients with PFA do not have underlying Alzheimer pathology. We speculate that the underlying pathology of such patients is likely frontotemporal lobar degeneration with TAR DNA binding protein 43 pathology,^27,28 as we have previously identified a group of patients with frontotemporal lobar degeneration with TAR DNA binding protein 43 pathology and a magnetic resonance imaging pattern of atrophy reminiscent of the FDG-PET pattern of hypometabolism of most patients with PFA in this study.^2,23

Correspondence: Keith A. Josephs, MD, MST, MSc, Department of Neurology, Divisions of Behavioral Neurology and Movements Disorders, Mayo Clinic, Rochester, MN 55905 (josephs.keith@mayo.edu).

Accepted for Publication: October 30, 2009.

Author Contributions:Study concept and design: Josephs and Whitwell. Acquisition of data: Josephs, Duffy, Strand, Claassen, and Peller. Analysis and interpretation of data: Josephs, Duffy, Fossett, Whitwell, and Peller. Drafting of the manuscript: Josephs and Whitwell. Critical revision of the manuscript for important intellectual content: Josephs, Duffy, Fossett, Strand, Claassen, Whitwell, and Peller. Statistical analysis: Josephs. Obtained funding: Josephs. Administrative, technical, and material support: Claassen. Study supervision: Duffy.

Financial Disclosure: None reported.

Funding/Support: This study was supported by NIH Roadmap Multidisciplinary Clinical Research Career Development Award grant (K12/NICHD)-HD49078.

Additional Contributions: We acknowledge all neurologists who cared for some of these patients clinically.