Neural dynamics of semantic categorization in semantic variant of primary progressive aphasia

  1. V Borghesani  Is a corresponding author
  2. CL Dale
  3. S Lukic
  4. LBN Hinkley
  5. M Lauricella
  6. W Shwe
  7. D Mizuiri
  8. S Honma
  9. Z Miller
  10. B Miller
  11. JF Houde
  12. ML Gorno-Tempini
  13. SS Nagarajan
  1. Memory and Aging Center, Department of Neurology, University of California, San Francisco, United States
  2. Department of Radiology and Biomedical Imaging, University of California, San Francisco, United States
  3. Department of Otolaryngology, University of California, San Francisco, United States
  4. Department of Neurology, Dyslexia Center University of California, San Francisco, United States

Abstract

Semantic representations are processed along a posterior-to-anterior gradient reflecting a shift from perceptual (e.g., it has eight legs) to conceptual (e.g., venomous spiders are rare) information. One critical region is the anterior temporal lobe (ATL): patients with semantic variant primary progressive aphasia (svPPA), a clinical syndrome associated with ATL neurodegeneration, manifest a deep loss of semantic knowledge. We test the hypothesis that svPPA patients perform semantic tasks by over-recruiting areas implicated in perceptual processing. We compared MEG recordings of svPPA patients and healthy controls during a categorization task. While behavioral performance did not differ, svPPA patients showed indications of greater activation over bilateral occipital cortices and superior temporal gyrus, and inconsistent engagement of frontal regions. These findings suggest a pervasive reorganization of brain networks in response to ATL neurodegeneration: the loss of this critical hub leads to a dysregulated (semantic) control system, and defective semantic representations are seemingly compensated via enhanced perceptual processing.

Introduction

Approaching a greenish, twisted object during a countryside walk, you might have two very different reactions: running away or simply stepping over it. Such a seemingly easy process, that is, telling a snake from a rope, requires the interplay of multiple cognitive processes relying on different neural substrates. First, the visual input must be analyzed, collecting information on all possibly relevant motor-perceptual features (e.g., color, sound, movement). Then, the extracted features must be merged into a unitary concept to allow proper identification (e.g., it’s a rope). Finally, one can select and perform an appropriate response (e.g., I’ll walk by it). All the neural computations supporting these processes occur within a few seconds. While the earliest perceptual processing takes place in the occipital cortex, the final stages (i.e., motor programming and execution) entail activation of frontal-parietal structures. The critical intermediate steps, involving the transformation from a visual input to a concept (and its semantic categorization as living vs. nonliving, dangerous vs. harmless), have been linked to the coordinated activity of multiple neural areas (Clarke and Tyler, 2015). Functional neuroimaging and neuropsychological research indicate that semantic knowledge is encoded within distributed networks (Huth et al., 2012; Fernandino et al., 2016), with a few key cortical regions acting as critical hubs (Ralph et al., 2017). However, many open questions remain as to the nature of neural representations and computations in these different areas, and how they dynamically interact.

Prior functional neuroimaging studies suggested that populations of neurons along the ventral occipito-temporal cortex (vOT) tune to ecologically relevant categories leading to a nested representational hierarchy of visual information (Grill-Spector and Weiner, 2014), where specialized cortical regions respond preferentially to faces (Gauthier et al., 2000; Kanwisher et al., 1997), places (Epstein and Kanwisher, 1998), bodies and body parts (Downing et al., 2007; Downing and Kanwisher, 2001), or objects (Lerner et al., 2001). Living stimuli appear to recruit lateral portions of vOT, while nonliving stimuli are highlighted in medial regions (Martin and Chao, 2001). Multiple organizing principles appear to be responsible for the representational organization of these areas, including agency and visual categorizability (Thorat et al., 2019). Overall, semantic representations appear to be processed in a graded fashion along a posterior-to-anterior axis: from perceptual (e.g., snakes are elongated and legless) to conceptual information (e.g., a snake is a carnivorous reptile) (Borghesani et al., 2016; Peelen and Caramazza, 2012). Notwithstanding this overall distributed view, different areas have been linked with specific computational roles: from modality-specific nodes in secondary motor and sensory areas to multimodal convergence hubs in associative cortices (Binder and Desai, 2011).

Neuropsychological findings corroborate the idea of a distributed yet specialized organization of semantic processing in the brain, supported by the interaction of a perceptual representational system arising along the occipito-temporal pathway, a semantic representational system confined to the anterior temporal lobe (ATL), and a semantic control system supported by fronto-parietal cortices (Ralph et al., 2017). For instance, focal lesions in the occipito-temporal pathway are associated with selective impairment for living items and spared performance on nonliving ones (Blundo et al., 2006; Caramazza and Shelton, 1998; Laiacona et al., 2003; Pietrini et al., 1988; Sartori et al., 1993; Warrington and Shallice, 1984) as well as the opposite pattern (Laiacona and Capitani, 2001; Sacchett and Humphreys, 1992). Moreover, acute brain damage to prefrontal or temporoparietal cortices in the semantic control system has been linked with semantic aphasia, a clinical syndrome characterized by deficits in tasks requiring manipulations of semantic knowledge (Jefferies and Lambon Ralph, 2006).

A powerful clinical model to study the organization of the semantic system is offered by the semantic variant primary progressive aphasia (svPPA or semantic dementia, Hodges et al., 1992; Gorno-Tempini et al., 2004). This rare syndrome is associated with ATL neurodegeneration as confirmed by the observation of gray matter atrophy (Collins et al., 2016), white matter alterations (Galantucci et al., 2011), and hypometabolism (Diehl et al., 2004), as well as neuropathological findings (Hodges and Patterson, 2007). Patients with svPPA present with an array of impairments (e.g., single-word comprehension deficits, surface dyslexia, impaired object knowledge) that can be traced back to a generalized loss of semantic knowledge, often affecting all stimuli modalities and all semantic categories (Hodges and Patterson, 2007). Conversely, executive functions and perceptual abilities are relatively preserved. Hence, these patients provide crucial neuropsychological evidence of the role played by the ATL in the storage of semantic representations, and can be leveraged to investigate the breakdown of the semantic system and the resulting compensatory mechanisms.

Pivotal steps forward in understanding the neurocognitive systems underlying semantic (as well as any other human) behaviors are enabled by the iterative, systematic combination of behavioral and neuroimaging data from both healthy controls (HC) and neurological patients (Price and Friston, 2002). However, task-based imaging in patients is hampered by specific difficulties (e.g., patients’ compliance) and limitations (e.g., performance is not matched and error signals can act as confounds) (Price et al., 2006; Wilson et al., 2018). To date, very few studies have attempted to deploy functional imaging in rare clinical syndromes such as svPPA, thus it is still not fully clear how structural damage and functional alterations relate to the observed cognitive and behavioral profile. Previous findings suggest that residual semantic abilities come from the recruitment of homologous and perilesional temporal regions, as well as increased functional demands on the semantic control system, that is, parietal/frontal regions (Maguire et al., 2010; Mummery et al., 1999; Pineault et al., 2019; Viard et al., 2013; Wilson et al., 2009). Recently, magnetoencephalographic (MEG) imaging has proven useful in detecting syndrome-specific network-level abnormalities (Ranasinghe et al., 2017; Sami et al., 2018) as well as task-related functional alterations (Kielar et al., 2018) in neurodegenerative patients. Critically, it has been suggested that imperfect behavioral compensation can be achieved via reorganization of the dynamic activity in the brain (Borghesani et al., 2020): owing to their damage to the ventral, lexico-semantic reading route, svPPA patients appear to over-recruit the dorsal, sublexical/phonological pathway to read not only pseudowords, but also irregular ones.

Here, we test the hypothesis that svPPA patients, burdened with ATL damage, thus lacking access to specific conceptual representations, overemphasize perceptual information as well as overtax the semantic control system to maintain accurate performance on a semantic categorization task (living vs. nonliving, see Figure 1a). Given the shallow semantic nature of the task, we expect comparable performance in patients with svPPA and a group of HC, with the critical differences emerging in neural signatures. Specifically, we expected patients to over-recruit occipital areas, supporting their greater reliance on visual processing.

Experimental paradigm, behavioral performance, and cortical atrophy.

(A) Cartoon representation of the experimental setting. Colored drawings were presented for 2 s, with an inter-stimuli interval jittered between 1.7 and 2.1 s. Subjects responded with a button press with their dominant hand. (B) Behavioral performance during the semantic categorization tasks in controls and semantic variant primary progressive aphasia (svPPA) patients, across the two stimuli conditions (living vs. nonliving items). There were no statistically significant effects of diagnosis, category, nor their interaction neither in percentage accuracy (healthy controls [HC]: living: 97.1 ± 6.6, nonliving: 96.8 ± 6.6; svPPA: living: 91.5 ± 6.2, nonliving: 95.9 ± 8.1) nor in reaction times (HC: living: 826.3 ± 112.5, nonliving: 856.9 ± 104.4; svPPA: living: 869.8 ± 179.8, nonliving: 911.1 ± 194.45). (C) Voxel-based morphometry (VBM)-derived atrophy pattern showing significantly reduced gray matter volumes in svPPA patients’ anterior temporal lobes, views from top to bottom: lateral, medial, ventral (thresholded at p<0.05 with family-wise error [FWE] correction, cluster threshold of 100 voxels).

Results

Behavioral data and cortical atrophy

Behavioral performance during the MEG scan neither differed between the two cohorts nor between the two stimulus categories. Statistically significant differences were not observed in reaction times (HC: living: 826.3 ± 112.5, nonliving: 856.9 ± 104.4; svPPA: living: 869.8 ± 179.8, nonliving: 911.1 ± 194.45), or accuracy (HC: living: 97.1 ± 6.6, nonliving: 96.8 ± 6.6; svPPA: living: 91.5 ± 6.2, nonliving: 95.9 ± 8.1). Overall, these results indicate that svPPA patients can perform the task as proficiently as healthy elders, an expected finding due to the relatively shallow semantic processing requirements and simple stimuli used in the task (see Figure 1b).

Distribution of cortical atrophy in the svPPA cohort is shown in Figure 1c. Patients present atrophy in the ATL, involving the temporal pole, the inferior and middle temporal gyrus. This pattern of neurodegeneration, bilateral yet strongly left-lateralized, is consistent with their clinical diagnosis and overall neuropsychological profile (see Table 1).

Table 1
Demographics and neuropsychological profiles.

Healthy controls and semantic variant of primary progressive aphasia (svPPA) patients, native English speakers, were matched for age, gender, and education. Scores shown are mean (standard deviation). * Indicates values significantly different from controls (p<0.05). MMSE = Mini-Mental State Exam; CDR = Clinical Dementia Rating; PPVT = Picture Vocabulary Test; WAB = Western Aphasia Battery; VOSP = Visual Object and Space Perception Battery.

ControlssvPPA
Demographic
N1818
Age, mean (SD)70.7 ± 6.567.1 ± 6.2
Education, mean (SD)17.5 ± 1.817.9 ± 3.2
Gender, n female129
Handedness, n right1515
MMSE (max. 30)29.0 ± 1.624.5 ± 3.8*
CDR score0.03 ± 0.10.7 ± 0.4*
CDR box score0.3 ± 1.24.0 ± 2.6*
Language production
Boston (object) naming test (15)14.7 ± 0.65.4 ± 3.7*
Phonemic (D-letter) fluency15.7 ± 5.89.1 ± 4.3*
Semantic (animal) fluency23.4 ± 3.99.3 ± 4.1*
Language comprehension
PPVT (max. 16)9.4 ± 3.2
WAB auditory word recognition (60)56.5 ± 4.2
WAB sequential command (100)70.7 ± 14.3
Digit span forwards7.1 ± 1.16.4 ± 1.2
Reading
Arizona reading total (max. 36)35.6 ± 0.530.5 ± 3.7*
Regular high-frequency words (9)9 ± 0.08.8 ± 0.4
Regular low-frequency words (9)8.9 ± 0.28.3 ± 1.2
Irregular high-frequency words (9)8.9 ± 0.37.7 ± 0.6
Irregular low-frequency words (9)8.8 ± 0.45.7 ± 2.3
Pseudowords (18)15.8 ± 2.715.2 ± 2.2
Spelling
Arizona spelling total (max. 20)18.1 ± 1.613.1 ± 4.0*
Regular high-frequency words (5)5 ± 0.04.4 ± 0.9
Regular low-frequency words (5)4.5 ± 0.64.1 ± 0.8
Irregular high-frequency words (5)4.1 ± 0.92.1 ± 1.6
Irregular low-frequency words (5)4.5 ± 0.52.6 ± 1.6
Pseudowords (10)8.8 ± 1.38.1 ± 2.6
Famous faces – spontaneous naming (max. 16)12.4 ± 3.42.9 ± 2.4*
Famous faces – face recognition (max 20)18.4 ± 2.012.8 ± 6.5*
Famous faces short triplets, pictures (max. 10)8.9 ± 1.06.6 ± 2.4
Famous faces short triplets, words (max. 10)9.7 ± 0.67.0 ± 2.0
Working memory/executive functions
Digit span backwards5.4 ± 1.14.5 ± 1.6*
Modified trials (total time)25.3 ± 13.641.9 ± 23.1*
Modified trials (# of correct lines)13.2 ± 3.213.2 ± 3.3
Design fluency (# of correct designs)11.7 ± 3.07.1 ± 3.4*
Visuospatial function
Benson figure copy (17)15.7 ± 0.715.3 ± 1.0
VOSP number location (30)9.3 ± 0.99.0 ± 1.5
Visual memory
Benson figure recall (17)12.1 ± 2.47.1 ± 4.9*

Time course of neural activity during visual semantic categorization

Within-group analyses of brain activity during the semantic categorization task, relative to pre-stimulus baseline activity levels, are presented for both controls and svPPA patients in Figure 2. In brief, following presentation of the images both cohorts showed posterior-to-anterior progression of functional activation across all five frequency bands. In the high-gamma band (63–117 Hz, see Figure 2a), we observed bilateral increases in synchronous power starting in the occipital cortex and progressively extending to temporal, parietal, and frontal regions. In the low-gamma band (30–55 Hz, see Figure 2b), subjects show heightened synchronization over bilateral occipital cortices, evident early in the svPPA group and only later in HC. Moreover, both groups showed reductions in activity over frontal cortices starting mid-trial. A similar progression of beta (12–30 Hz, see Figure 2c) and alpha (8–12 Hz, see Figure 2d) band activity revealed significant reductions in synchronous activity for both groups, extending from bilateral occipital cortices to temporal and parietal lobes, and involving progressively larger areas in precentral and superior frontal gyrus. A focus of increased alpha synchrony in anterior cingulate regions, mid-trial, is evident in both groups. Finally, induced theta band (3–7 Hz, see Figure 2e) activity revealed progressive increases in synchronous activity over bilateral occipital cortices, a similarly progressive pattern of increased synchronization within frontal regions at an onset window after that of occipital regions, and progressively reduced theta activity relative to baseline levels over parietal and temporal lobes.

Stimulus-locked (0 ms = stimulus onset) within-group analyses of task-related changes in oscillatory power.

(a) Rendering of the results in the high-gamma band for both controls (healthy controls [HC], upper row) and patients (semantic variant primary progressive aphasia [svPPA], lower row). Cold color = more desynchronization (vs. baseline). Warm color = more synchronization (vs. baseline). (c-e) Same as in (a) but for the low-gamma, beta, alpha, and theta band, respectively. Within-group analyses were performed, with no additional smoothing, on normalized reconstructions using statistical nonparametric mapping (SnPM one-sample, two-tailed t-test against baseline).

Taken together, these stimulus-locked task-induced changes indicate, in both cohorts and across all frequency bands, the expected pattern of visual processing followed by motor response preparation. Notwithstanding the overall similarity in spatiotemporal dynamics, specific activation differences were detected between svPPA patients and HC and are reported below.

Neural dynamics of semantic categorization in a faulty semantic system

We investigated when, where, and at which frequency svPPA patients differ from HC during semantic categorization of visual stimuli. While the overall pattern of activation across frequencies and time is similar, crucial differences between the two cohorts emerged in the between-group analyses performed in each frequency band. Table 2 summarizes the temporal windows, peaks of local maxima, and t-values of all clusters isolated by the direct comparison of the two cohorts. Figure 3 allows appreciation of the spatiotemporal distribution of these clusters at four exemplar time points.

Table 2
Local maxima in Montreal Neurological Institute (MNI) coordinates.

Time window, MNI coordinates, p- and t-value of the local maxima of the different magnetoencephalographic (MEG) whole brain contrasts performed. The spatiotemporal distribution of these clusters at four exemplar time points can be appreciated in Figure 3.

Time windowLocal maxima
t-test svPPA vs. HCmsMNI [x,y,z]p-valuet-value
Theta band [3–7 Hz]
Left lingual gyrus0–212−10.0 –100.0 −10.00.0053.7More ERD in svPPA
Left lingual gyrus412–612−8.5 –100.0 −8.10.0053.1More ERD in svPPA
Right medial and superior frontal gyrus187–38718.6 61.4 –14.70.001−3.92Less ERS in svPPA
Alpha band [8–12 Hz]
Right precentral gyrus212–61245.0 –15.0 40.00.0013.4More ERD in svPPA
Left middle temporal gyrus287–362−59.8 –41.6 −1.00.0052.8More ERD in svPPA
Bilateral medial and orbital frontal gyrus462–612−6.2 34.3 –24.90.0015.1More ERS in svPPA
Beta band [12–30 Hz]
Left cingulate cortex0–62−6.2–30.3 43.30.0052.9More ERS in svPPA
Right medial frontal gyrus137–2627.8 56.8 11.70.0013.6More ERS in svPPA
Left middle temporal gyrus237–362−65.0 –20.0 −5.00.001−3.4Less ERD in svPPA
Left superior frontal gyrus587–612−21.8 46.7 45.70.005−3.1Less ERD in svPPA
Low-gamma band [30–55 Hz]
Left lingual gyrus62–612−10.1–98.4 −8.90.0014.2More ERS in svPPA
Left inferior occipital gyrus362–612−34.8 –93.9 2.70.0014.1More ERS in svPPA
Right lingual gyrus212–43718.2 –89.1 8.30.0053.4More ERS in svPPA
Right medial frontal gyrus212–4129.3 63.0 2.20.0013.7Less ERD in svPPA
Left superior frontal gyrus262–462−3.8 62.8 14.00.0053.6Less ERD in svPPA
High-gamma band [63–117 Hz]
Left superior frontal gyrus62–137−36.5 26.6 48.80.0013.4More ERS in svPPA
Left superior temporal gyrus62–287−48.2 –22.3 13.30.0053More ERS in svPPA
Left parahippocampal gyrus212–312−15.5 –27.1 −6.50.0013.3More ERS in svPPA
Right medial frontal gyrus287–33713.2 70.7 0.60.0053.2More ERS in svPPA
Left superior frontal gyrus287–612−22 68.4 140.0013.6More ERS in svPPA
Right superior frontal gyrus462–61243.9 54.7 17.20.0013.9More ERS in svPPA
Stimulus-locked (0 ms = stimulus onset) between-group analyses of changes in oscillatory power.

Rendering of the results in the high-gamma (a), low-gamma (b), beta (c), alpha (d), and theta (e) bands. Purple color = more synchronization in semantic variant primary progressive aphasia (svPPA) (vs. healthy controls [HC]). Brown color = less synchronization in svPPA (vs. HC). Table 2 summarizes the temporal windows, peaks of local maxima, and t-values of all clusters isolated by the direct comparison of the two cohorts. Between-group analyses were performed, with no additional smoothing, on normalized reconstructions using statistical nonparametric mapping (SnPM two-sample, two-tailed t-test).

In the high-gamma band, we detected significantly higher synchronization in svPPA patients, relative to controls, over left superior temporal (at both early and late time points) and right frontal (at late time points) cortices (see Figure 3a). In the low-gamma band, we observed an extensive spatiotemporal cluster over bilateral occipital cortices with significantly higher synchronized activity in svPPA patients relative to controls. Similarly, small clusters of gamma activity, relatively more desynchronized in HC than svPPA, resulted in an increased gamma synchrony in medial frontal cortices at ~300 ms for the svPPA group (see Figure 3b). Overall, the results at high frequencies (30–117 Hz) suggest thus higher activity in svPPA over bilateral occipital and left superior temporal cortices throughout the trial, and right frontal cortices at late time points.

Between-group contrast in beta band revealed, in svPPA patients, more desynchronization (i.e., more beta suppression) over the left superior temporal gyrus at ~300 ms, while simultaneously displaying less desynchronization in a right middle-frontal cluster (see Figure 3c). In the alpha band, svPPA patients showed less desynchronization over left middle temporal gyrus at ~300 ms as well as in later clusters in the right precentral gyrus, left anterior cingulate, and left parahippocampal gyrus (see Figure 3d). Finally, in the theta band, significant differences over the left occipital cortex occurred at both early (~100 ms) and late (~500 ms) time points indicating higher synchronization in svPPA patients compared to HC, while the opposite pattern (i.e., higher activity for HC) is observed in a right frontal cluster at ~300 ms (see Figure 3e). Overall, the results at low frequencies (3–30 Hz) suggest thus higher activity in svPPA over bilateral occipital and left superior temporal cortices, while indicating less activity in left middle temporal and right frontal regions.

Taken together, these findings suggest that svPPA patients performed the semantic categorization tasks by over-recruiting bilateral occipital cortices and left superior temporal gyrus, while showing less reliance on left middle temporal regions and inconsistent engagement of frontal ones.

Post hoc region-of-interest analyses

Our first region-of-interest (ROI) post hoc analysis allows visualization, across all frequency bands, of the differences in temporal dynamics between the two cohorts (Figure 4). The three a priori defined ROIs cover the theorized perceptual-to-conceptual gradient of information processing along the ventral visual path (Borghesani and Piazza, 2017) and include the putative visual spoke (left occipital pole, OCC) and semantic hub (left ATL, Ralph et al., 2017). It appears clear that the main difference between svPPA patients and HC is heightened low-gamma activity over the occipital region. Such difference is evident around 100 ms post stimuli onset, peaks around 200 ms, and continues throughout the whole. These findings rule out an explanation of the observed whole brain differences as mere temporal shift or spreading, while highlighting the spatial specificity of the main results.

Results of the region of interest post hoc analysis.

Three regions-of-interest (ROIs) of 20 mm radius were centered on the occipital pole (OCC, Montreal Neurological Institute [MNI]: −10, –94, −16), left ventral occipito-temporal cortex (VOT, MNI: −50, –52, −20), and left ATL (MNI: −30, –6, −40). Pink color represents healthy controls data, light blue svPPA patients. Shaded areas indicate the standard deviation around the group average (i.e., solid line).

Our second ROI post hoc analysis allows characterization of the full time-frequency spectrum of both cohorts in two representative voxels (Figure 5). Critically, a broad and sustained increase in low-gamma band power is observed in svPPA patients and not in HC, with no traces of cross-spectral leakage between beta and low gamma, or low and high gamma. These findings rule out a possible interpretation of the observed effects in terms of frequency shift or spread, highlighting the spectral specificity of the main results.

Results of the post hoc regions of interest analysis of power changes.

Full time-frequency plot of power changes in two representative voxels centered in the peak of activation (as per group results, Montreal Neurological Institute [MNI]: −34.8, –93.9, 2.7) and on the occipital pole (OCC, MNI: −10, –94, −16).

Discussion

This is the first study investigating the spatiotemporal dynamics of semantic categorization of visual stimuli in a cohort of svPPA patients. We provide compelling evidence that, burdened with ATL damage, svPPA patients recruit additional perilesional and distal cortical regions to achieve normal performance on a shallow semantic task. As compared to healthy age-matched controls, svPPA patients showed greater activation over bilateral occipital cortices and superior temporal gyrus, indicating over-reliance on perceptual processing and spared dorsal language networks. Conversely, they showed inconsistent engagement of frontal regions, suggesting less efficient control responses.

These findings have important implications both for current neurocognitive models of the language systems and on the utility of MEG imaging in clinical populations. First, we detect over-recruitment of occipital and superior temporal regions paired with inconsistent engagement of frontal areas, where some spatiotemporal clusters suggest heightened activity in patients, others in controls. These results speak to the distributed and dynamic organization of the semantic system, where semantic representations are supported by occipito-temporal cortices and semantic control by fronto-parietal areas. Second, the observation that normal performance can be achieved via altered neural dynamics elucidates the neurocognitive mechanisms that support compensation in neurological patients. Specifically, we contribute to the body of literature illustrating how network-driven neurodegeneration leads to the reorganization of the interplay of various cortical regions.

Faulty semantic representations: compensating conceptual loss with perceptual information

Our key finding is that svPPA patients can achieve normal performance in a shallow semantic task by over-relying on perilesional language-related regions (STG), as well as on distal visual (occipital) and executive (frontal) networks. At frequencies spanning low and high gamma bands, svPPA patients show increased activity in occipital and superior temporal cortices relative to their healthy counterparts. Gamma oscillations have been associated with local computations (Donner and Siegel, 2011), promoting binding and selective long-range communication (Hagoort et al., 2004; Fries, 2015), including merging of multimodal semantic information (van Ackeren et al., 2014). Results at lower frequencies indicate greater neural activity in svPPA over bilateral occipital and left superior temporal cortices. Theta oscillations have generally been associated with operations over distributed networks, such as those required for lexico-semantic retrieval (Bastiaansen et al., 2005; Bastiaansen et al., 2008; Kielar et al., 2015), integration of unimodal semantic features (van Ackeren et al., 2014), and facilitating phase-specific coupling of selective communication between regions (Fries, 2015; Canolty et al., 2006). Therefore, in our patients, compensation for faulty semantic representations seems to rely primarily on local and distributed computations in networks associated with perceptual processing.

In principle, the semantic task employed in the current study (i.e., identifying a visually presented object as either a living or nonliving) can be performed by focusing on a few key, distinctive, motor-perceptual features: if it has eyes and teeth, it is a living being. Further processing steps, such as would be required for an object identification and naming (i.e., accessing the appropriate lexical label), require the integration of multiple motor perceptual as well as conceptual features (Borghesani and Piazza, 2017): a python is a nonvenomous snake that kills by constriction. Combining the behavioral data collected during the recordings and outside and the scanner (see Boston Naming Task performance, Table 1), it appears clear that HC can recognize (and likely inevitably mentally name) each item, while svPPA patients can only provide the categorical label. Patient data is thus critical in characterizing the division of labor between the distributed set of cortical regions involved in semantic processing. Our findings strongly suggest that ATL damage hampers operation of the semantic representation system, by shattering their conceptual components and thus forcing over-reliance on perceptual features coded in posterior cortices. This is consistent with a growing body of research. For instance, it has been shown that the ability to merge perceptual features into semantic concepts relies on the integrity of the ATL (Hoffman et al., 2014), and that ATL damage promotes reliance on perceptual similarities over conceptual ones (Lambon Ralph et al., 2010). Moreover, it appears that the more motor-perceptual information is associated with a given concept, the more resilient it is to damage, an advantage that is lost once the disease progresses from ATL to posterior ventral temporal regions (Hoffman et al., 2012).

Faulty semantic representations: overtaxing the semantic control network

Compared to HC, svPPA patients appear to have less activation in the left middle temporal gyrus and to inconsistently engage frontal regions, suggesting that increased demands to the semantic control systems are met by inefficient responses in prefrontal and superior frontal cortices. Comparing the two cohorts across frequency bands, it appears that an enhanced late high-frequency (local neural) response occurs in svPPA, vs. an earlier and lower frequency (long-range connection) response in controls. One speculation for this pattern is that in svPPA, an initial inefficient response in the (semantic) cognitive control network centered on frontal areas leads to a later higher reliance on local activity for (semantic) cognitive control and decision-making processes.

Previous studies demonstrated that object recognition in visual areas is facilitated by prior knowledge (Bannert and Bartels, 2013) received via feedback projections from both frontal (Bar et al., 2006) and anterior temporal (Coutanche and Thompson-Schill, 2015) cortices. Moreover, it has been observed that higher demands for feature integration entail more recurrent activity between fusiform and ATL (Clarke et al., 2011). Our study provides a direct contrast between subjects in which both frontal and ATL feedback inputs are preserved (HC), and those in which ATL neurodegeneration forces reliance exclusively on frontal inputs.

Interestingly, the observed temporal dynamics (with the detection of early frontal involvement) are not compatible with a strictly feedforward model of visual stimuli processing. This is in line with recent evidence that recurrent neural models are needed to explain the representational transformations supporting visual information processing (Gwilliams and King, 2019; Kietzmann et al., 2019).

Thus, taken together, our findings corroborate the idea that the conversion from percept to concept is supported by recurrent loops over fronto-parietal and occipito-temporal regions which have been implicated in, respectively, semantic control and semantic representations (Chiou et al., 2018).

Clinical implications

Our findings corroborate the idea that neurodegeneration leads to the dynamic reorganization of distributed networks (Agosta et al., 2014; Guo et al., 2013), and that task-based MEG imaging can be instrumental in deepening our understanding of the resulting alterations (Borghesani et al., 2020). Ultimately, these efforts will pave the way toward treatment options, as well as better early diagnostic markers as functional changes are known to precede structural ones (Bonakdarpour et al., 2017). For instance, our results support previous neuropsychological evidence suggesting that the origin of svPPA patients’ difficulties during semantic categorization tasks are linked to degraded feature knowledge rather than, as it happens in other FTDs (fronto-temporal dementias), to a deficit of executive processes (Koenig et al., 2006).

Our results are in line with prior studies relating svPPA patients’ performance on semantic tasks with respect to not only the expected hypoactivation of the left ATL and functionally connected left posterior inferior temporal lobe (Mummery et al., 1999), but also based on the patterns of hyperactivations observed in the current study. Heightened activity has been reported in periatrophic left anterior superior temporal gyrus as well as more distant left premotor cortex, and right ATL (Mummery et al., 1999; Pineault et al., 2019). Individual subject analyses have indicated that patients might attempt different compensatory strategies, which may vary in terms of efficiency and, crucially, would rely on the recruitment of different cortical networks (Viard et al., 2013; Viard et al., 2014). For instance, studies on reading have associated svPPA patients’ imperfect compensation of the semantic deficit (leading to regularization errors) with over-reliance on parietal regions subserving sublexical processes (Wilson et al., 2009). Consistently, task-free studies of intrinsic functional networks suggest that the downregulation of damaged neurocognitive systems can be associated with the upregulation of spared ones. In svPPA patients, recent fMRI evidence shows coupling of decreased connectivity in the ventral semantic network with increased connectivity in the dorsal articulatory-phonological one (Battistella et al., 2019; Montembeault et al., 2019). Additionally, svPPA has been linked with specific spatiotemporal patterns of neuronal synchrony alterations: alpha and beta hypersynchrony in the left posterior superior temporal and adjacent parietal cortices, and delta-theta hypersynchrony in left posterior temporal/occipital cortices (Ranasinghe et al., 2017). Our findings also align with the recent observation that, during reading, svPPA patients can (imperfectly) compensate for their damage to the ventral route by over-recruiting the dorsal one (Borghesani et al., 2020). The present findings corroborate thus the idea that neurodegeneration forces the reorganization of the interplay between ventral and dorsal language networks.

Critically, the present functional neuroimaging results and their interpretation rest on the fact that the task allowed engagement of semantic processing in patients in which the semantic system is, by definition, compromised. Contrary to a more challenging task such as naming, patients with svPPA were able to perform the semantic categorization as accurately and fast as HC. Hence, probing the semantic system at the proper level of difficulty (Wilson et al., 2018), we avoided the challenging interpretation of activation maps associated with failure to perform a task (Price et al., 2006). Our findings thus call for caution when evaluating studies comparing clinical cohorts based solely on behavioral data: failing to detect a difference in performance does not necessarily correspond to similar underlying neurocognitive resources.

Limitations and future perspectives

The nature of the clinical model we adopted constrains our sample. First, even if ours is the to-date largest cohort of svPPA patients assessed with task-based functional neuroimaging, our sample size is relatively small, owing to the rareness of the disease. We thus have limited statistical power, preventing us from, for instance, further exploring brain-behavior correlations. Second, our subjects (both HC and patients) are older than those reported in previous studies on semantic categorization, cautioning against direct comparisons. While it has been shown that the neural dynamics of visual processing are affected by aging, the reduced and delayed activity observed does not necessarily relate to poorer performance, but rather may be mediated by task difficulty (Bruffaerts et al., 2019). Moreover, previous evidence suggests that even if semantic processing remains intact during aging, its neurofunctional organization undergoes changes. For instance, Lacombe et al., 2015 found that, during a verbal semantic categorization task, older adults exhibited behavioral performance equivalent to that of young adults, but showed less activation of the left inferior parietal cortex and more activation of bilateral temporal cortex. Finally, our task design does not allow further investigation of potential categorical effects. Future studies wishing to investigate representations of living and nonliving items separately will require more trials and stimuli carefully controlled for psycholinguistic variables such as prototypicality and familiarity. Contrary to patients with damage to the vOT due to stroke or herpes simplex encephalitis, svPPA patients usually do not present categorical dissociations (Moss et al., 2005). However, deeper investigations of time-resolved neural activity in svPPA could shed light onto the debate on the nature of ATL representations: category-specific deficits might arise from lacunar (rather than generalized) impairment of graded representations (Lambon Ralph et al., 2007).

To date, the field lacks proper strategies to deal with tissue undergoing neurodegeneration while attempting to source-localize electrophysiological effects. The first and main issue is that of defining atrophy itself: atrophic tissue will vary subject by subject (and it is not an all-or-none phenomenon) yet a threshold would need to be established. In previous work, we took the parsimonious approach of masking out atrophic regions from group-level statistics to avoid uninterpretable results (e.g., Borghesani et al., 2020). In other settings, one option is that of correcting ROI statistics including gray matter volume as covariate (e.g., Ranasinghe et al., 2017). Finally, some would simply report whether electrophysiological differences and atrophy maps overlap or not (e.g., Kielar et al., 2018). Here, we decided not to mask the ATL and, as a consequence, an apparent signal coming from the atrophic region is observed in svPPA patients (mainly in beta and alpha). Clearly, the underlying issue of how tissue undergoing neurodegeneration affects source modeling is an open problem that requires further exploration.

Finally, our interpretation of higher low gamma as a sign of compensation is speculative and lies on twofold inference: that more low-gamma band in svPPA means more activity, and that more activity means compensation. Previous literature in healthy and pathological aging suggests that higher activation can be associated with compensatory effects or reflect neuropathology (e.g., Elman et al., 2014). Critically, when considering progressive phenomena such as neurodegeneration, one has to acknowledge that hyper- and hypoactivations might reflect different stages of disease and that their relation to behavioral performance might follow a nonlinear U-shaped trajectory (e.g., Gregory et al., 2017). Following Cabeza et al., 2018, we believe that greater activity can be interpreted as compensation if two criteria are met: (1) there has to be evidence of a ‘supply-demand gap’ and (2) there has to be evidence of a beneficial effect on cognitive performance. Our data clearly fulfills the first criterion as svPPA patients present insufficient neural resources (i.e., relatively focal ATL atrophy) and suffer from its behavioral consequences (i.e., pervasive semantic loss). Our findings also fulfill the second criterion as svPPA patients are able to perform the task with both accuracy and reaction times comparable with the HC. Thus, we believe that the neural and behavioral evidence we provide is enough to rule out alternatives to compensation such as inefficiency or pathology. While further studies are warranted to shed light onto the relation between neurodegeneration, neurophysiological markers, and behavior, we hold that the most appropriate (albeit speculative) interpretation of our findings is in terms of compensation.

Conclusions

Combining task-based MEG imaging and a neuropsychological model, we provide novel evidence that faulty semantic representations following ATL damage can be partially circumvented by additional processing in relatively spared occipital and dorsal stream regions. Our results thus inform current neurocognitive models of the semantics system by corroborating the idea that it relies on the dynamic interplay of distributed functional neural networks. Moreover, we highlight how MEG imaging can be leveraged in clinical populations to study compensation mechanisms such as the recruitment of perilesional and distal cortical regions.

Materials and methods

Subjects

Eighteen svPPA patients (13 females, 66.9 ± 6.9 years of age) and 18 healthy age-matched controls (11 females, 71.3 ± 6.1 years of age) were recruited through the University of California San Francisco (UCSF) Memory and Aging Center (MAC). All subjects were native speakers and had no contraindications to MEG. Patients met currently published criteria as determined by a team of clinicians based on a detailed medical history, comprehensive neurological and standardized neuropsychological and language evaluations (Gorno-Tempini et al., 2011). Besides being diagnosed with svPPA, patients were required to score at least 15 out of 30 on the Mini-Mental Status Exam (MMSE; Folstein et al., 1975) and be otherwise sufficiently functional to be scanned. HC were recruited from the UCSF MAC healthy aging cohort, a collection of subjects with normal cognitive and neurological exam and MRI scans without clinically evident strokes. Inclusion criteria required the absence of any psychiatric symptoms or cognitive deficits (i.e., Clinical Dementia Rating – CDR = 0 and MMSE ≥28/30). Demographic information and neuropsychological data are shown in Table 1. The study was approved by the UCSF Committee on Human Research and all subjects provided written informed consent (IRB # 11–05249).

Stimuli and experimental design

Request a detailed protocol

All subjects performed a semantic judgment task on visually presented stimuli (Figure 1a). Stimuli consisted of 70 colored drawings: 36 belonging to the semantic category of living items (e.g., animals, plants) and 34 belonging to the semantic category of nonliving items (e.g., tools, furniture).

To validate the set of stimuli, a behavioral study was conducted on a separate group of 54 age-matched healthy subjects (31 women; 47 right-handed; age = 74.21 years ± 8.63; education = 15 years ± 2.02). First, subjects had to report the most common name for each drawing (i.e., Identify the item in the image: what is the first name that comes to mind?). They were given the possibility of providing a second term if needed (i.e., If appropriate, write the second name that came to mind.). They were then asked to rate how familiar they are with the item on a 7-point scale from not at all familiar to very familiar. Finally, they were asked whether the item belongs to the category of living or nonliving items, and to rate how prototypical for that category the item is (i.e., How good is this picture as example of an item of that category?) on a 7-point scale from bad example to good example. Data were collected with Qualtrics software (Qualtrics, Provo, UT. https://www.qualtrics.com) and subjects recruited from the broad pool of subjects enrolled in the above described UCSF MAC healthy aging cohort. For each stimulus, we calculated the percentage of agreement with our pre-set categorization, average familiarity, average prototypicality, and then compared the living and nonliving categories. For living items, the average percentage of agreement with the assigned category was 96.86% ± 4.07, the lowest score was 75.93% for the item dinosaur. For nonliving items, the average percentage of agreement was 99.18% ± 1.20, the lowest score was 96.30% for the items pizza and hamburger. A two-tailed t-test revealed that the difference between the two categories was significant (p=0.002): the rate of agreement was higher for nonliving items than for living ones. The average prototypicality of living items was 6.24 ± 0.52 (range 6.74–4), while for nonliving items 6.47 ± 0.32 (range 6.85–5.19) for nonliving items. Again, a two-tailed t-test revealed a significant difference between the two categories (p=0.032): nonliving items were judged more prototypical of their category than living ones. As for familiarity, the average for living items was 6.15 ± 0.32 (range 6.8–4.81), while for nonliving items was 6.67 ± 0.21 (range 6.91–6.02). Even in this case the difference between the two categories was significant (two-tailed t-test, p<0.001): nonliving items were judged more familiar.

Images of the two categories were also compared in terms of visual complexity (calculated as Shannon entropy via the python package Scikit-Image, https://scikit-image.org/). No significant difference between living (3.04 ± 0.84) and nonliving (3.13 ± 0.96) items emerged. Finally, we compared stimuli in terms of the length (number of letter), imaginability, concreteness, and familiarity of their most common lexical label as extracted from the Medical Research Council (MRC, http://websites.psychology.uwa.edu.au/school/MRCDatabase/uwa_mrc.htm) Psycholinguistic Database, and word frequency was extracted from the Corpus of Contemporary American English (COCA, https://www.wordfrequency.info/). Consistent with our online questionnaire, the only statistically significant differences between the two categories were imageability (living: 613.19 ± 19.62, nonliving: 596.43.15 ± 28.08, p=0.03) and familiarity (living: 498.26 ± 69.32, nonliving: 547.96 ± 45.82, p<0.001). All the psycholinguistic variables characterizing the stimuli are shown in Table 3.

Table 3
Psycholinguistic characteristics of the stimuli.

Stimuli consisted of 70 colored drawings illustrating living items (n = 36) or nonliving items (n = 34). Length, imaginability, concreteness, and familiarity (norm) were extracted from the Medical Research Council (MRC) Psycholinguistic Database searching for the most common label for each item. Similarly, frequency was extracted from the Corpus of Contemporary American English (COCA). Category agreement, category prototypicality, and familiarity (quest.) were assessed with a behavioral study on separate age-matched healthy controls. As a proxy for visual complexity, we used Shannon entropy as computed with Scikit-Image. Values shown are mean (standard deviation). * Indicate values significantly different between the two categories (two-tailed t-test, p<0.05).

Living itemsNonliving items
N3634
ExamplesFish, flowerScissors, train
Frequency (log)3.69 (0.54)3.96 (0.65)
Length (# of letters)5.29 (1.58)5.61 (1.84)
Imageability613.19 (19.62)596.43 (28.08)*
Familiarity (norm)498.26 (69.32)547.96 (45.82)*
Familiarity (quest.)6.15 (0.32)6.67 (0.21)*
Concreteness608.27 (16.26)599.10 (25.94)
Category agreement96.86 (4.07)99.18 (1.20)*
Category prototypicality6.24 (0.52)6.47 (0.32)*
Visual complexity3.04 (0.84)3.13 (0.96)

Visual stimuli were projected into the magnetically shielded MEG scanner room via a system of mirrors mounted within the scanner room for this purpose, with the final mirror positioned roughly 61 cm from the subject’s face. Subjects were instructed to classify the pictures as living or nonliving by pressing one of two response buttons with their dominant hand. Stimuli were displayed for 2 s, with an inter-stimulus interval jittered between 1.7 and 2.1 s. A total of 170 trials were presented: each individual stimulus was repeated 2.5 times in a random order. E-Prime (https://pstnet.com/products/e-prime/) was used to present the stimuli; events from E-Prime and the response pad were automatically routed into the imaging acquisition software and integrated with MEG traces in real time.

Behavioral analyses

Request a detailed protocol

Subject performance, that is, reaction times and accuracy, was analyzed using an analysis of variance based on the two stimuli categories (living vs. nonliving) and two cohorts (controls vs. svPPA patients) using the Python statistical library (statsmodels – http://www.statsmodels.org). Data from one outlier in the svPPA cohort were excluded from the behavioral analyses (average reaction times were 1.35 s vs. 0.8 s in the whole cohort).

MRI protocol and analyses

Request a detailed protocol

Structural T1-weighted images were acquired on a 3 T Siemens system (Siemens, Erlagen, Germany) installed at the UCSF Neuroscience Imaging Center, equipped with a standard quadrature head coil with sequences previously described (Mandelli et al., 2014). MRI scans were acquired within 1 year of the MEG data acquisition.

To identify regions of atrophy, svPPA patients were compared to a separate set of 25 HC collected using the same protocol (14 females, mean age 66.2 ± 8.5) via voxel-based morphometry (VBM). Image processing and statistical analyses were performed using the VBM8 Toolbox implemented in Statistical Parametric Mapping (SPM8, Wellcome Trust Center for Neuroimaging, London, UK, http://www.fil.ion.ucl.ac.uk/spm) running under Matlab R2013a (MathWorks). The images were segmented into gray matter, white matter, and CSF, bias corrected, and then registered to the Montreal Neurological Institute (MNI). Gray matter value in each voxel was multiplied by the Jacobian determinant derived from the spatial normalization to preserve the total amount of gray matter from the original images. Finally, to ensure the data are normally distributed and compensate for inexact spatial normalization, the modulated gray matter images were smoothed with a full-width at half-maximum Gaussian kernel filter of 8 × 8 × 8 mm3. A general linear model was then fit at each voxel, with one variable of interest (group) and three confounds of no interest: gender, age, education, and total intracranial volume (calculated by summing across the gray matter, white matter, and CSF images). The resulting statistical parametric map was thresholded at p<0.05, with family-wise error correction, and a cluster extent threshold of 100 voxels.

MEG protocol and analyses

Request a detailed protocol

Neuromagnetic recordings were conducted using a whole-head 275 axial gradiometer MEG system (Omega 2000, CTF, Coquitlam, BC, Canada) at a sampling rate of 1200 Hz, under a bandpass filter of 0.001–300 Hz, while subjects performed the task. Subjects were lying supine, with their head supported near the center of the sensor array. Head position was recorded before and after each scan using three fiducial coils (nasion, left/right preauricular) placed on the subject. All subjects included in the current study exhibited movement under 1 cm, as measured pre- and post- experimental run. The two cohorts did not differ in average motion level: svPPA 0.28 cm (SD 0.11), HC 0.39 cm (SD 0.22) (p=0.12). Twenty-nine reference sensors were used to correct distant magnetic field disturbance by calculating a synthetic third-order gradiometer (Warrington and Shallice, 1984; Vrba and Robinson, 2001), which was applied to signal post-acquisition. Datasets were epoched with respect to stimulus presentation onset (stimulus-locked trials from −0.5 to 1.0 s) and artifacts rejected using a semi-automated process outlined as follows: noisy channels were identified as having more than 20 trials exceeding 1.5 pT amplitude under a temporary bandpass filter of 3–50 Hz, with no more than five channels in the sensor array removed. Epochs were then flagged and removed for any remaining artifacts exceeding the 1.5 pT threshold. Mean number of trials included in analyses for the two groups did not significantly differ (svPPA mean = 155 trials [SD = 20, range 121–170], control mean = 162 [SD = 16, range 103–172], two-tailed t[34]=1.059, p=0.297).

Alignment of structural and functional images was performed using three prominent anatomical points (nasion and preauricular points), marked in the individuals’ MR images and localized in the MEG sensor array using the three fiducial coils attached to these points during the MEG scan. A 3D grid of voxels with 5 mm spatial resolution covering the entire brain was created for each subject and recording, based on a multisphere head model of the coregistered structural 3D T1-weighted MR scan. Reconstruction of whole brain oscillatory activity within these voxels was performed via the Neurodynamic Utility Toolbox for MEG (NUTMEG; http://nutmeg.berkeley.edu, Hinkley et al., 2020), which implements a time-frequency optimized adaptive spatial filtering technique to estimate the spatiotemporal estimate of neural sources. The tomographic volume of source locations was computed using a 5 mm lead field that weights each cortical location relative to the signal of the MEG sensors (Dalal et al., 2008; Dalal et al., 2011). The beamforming algorithm choses was a variant of the synthetic aperture magnetometry (SAM) inverse solution (Vrba and Robinson, 2001) as implemented in NUTMEG (Hinkley et al., 2020). SAM is considered a scalar beamformer as it optimizes dipole orientation at the source to maximize signal.

We sought to focus on induced changes in brain activity, that is, to study modulations of ongoing oscillatory processes that are not necessarily phased-locked (Makeig et al., 2004). Moreover, we wished to explore both high- and low-frequency ranges as they bear different functional interpretations, in particular their association with different spatial scales: high- and low-frequency oscillations are associated with local and distributed computations, respectively (Donner and Siegel, 2011). Thus, we examined task-related modulations of ongoing oscillatory processes in five frequency bands: theta (3–7 Hz), alpha (8–12 Hz), beta (12–30 Hz), low-gamma (30–55 Hz), and high-gamma (63–117 Hz) (FIR filter having widths of 300 ms for theta/alpha, 200 ms for beta, 150 ms for low-gamma, and 100 ms for high-gamma; sliding over 25 ms time windows). Source power for each voxel location in a specific time window and frequency band was derived through a noise-corrected pseudo-F statistic expressed in logarithmic units (decibels, dB), describing signal magnitude during an ‘active’ experimental time window relative to an equivalently sized, static pre-stimulus baseline ‘control’ window (Robinson and Vrba, 1999). Single subject beamformer reconstructions were spatially normalized by applying each subject’s T1-weighted transformation matrix to their statistical map.

Group analyses were performed, with no additional smoothing, on normalized reconstructions using statistical nonparametric mapping (SnPM; Singh et al., 2003), both within-group and between-groups. Three-dimensional average and variance maps across subjects were calculated at each time point and smoothed with a 20 × 20 × 20 mm3 Gaussian kernel (Dalal et al., 2008; Dalal et al., 2011). From this map, pseudo-t statistics evaluated the magnitude of the contrast obtained at each voxel and time. Voxel labels were permuted to create a t-distribution map for within- and between-group contrasts (2N permutations, where N = number of subjects, up to 10,000 permutations). Each voxel’s t-value was evaluated using 2N degrees of freedom to determine the corresponding p-value associated with each voxel’s pseudo-F value (Singh et al., 2003). These cortical significance maps were spatially thresholded to include only voxels designated as ‘gray matter’ within the automated anatomical labeling atlas (Tzourio-Mazoyer et al., 2002), and the additional requirement for voxels with uncorrected p-values attaining a threshold of p<0.005 to include 26 adjacent gray matter voxels at p<0.005, effecting a cluster-based threshold of activity. We utilized these maps to examine the pattern of activation during semantic categorization separately for controls and svPPA patients (SnPM one-sample, two-tailed t-test against baseline) and directly compare svPPA patients and controls to highlight spatiotemporal clusters of differential activity between the two cohorts (SnPM two-sample, two-tailed t-test). It should be noted that while we are conducting nonparametric stats with a conservative cluster thresholding to reduce spurious findings from voxel to voxel (p-value threshold 0.005), no correction (nor additional test) is performed to account for multiple time windows or frequency bands.

Finally, we conducted two ROI post hoc analyses. The first one aimed at visualizing, across all frequency bands, the time course of the differential activation between the two cohorts. To avoid circularity and cherry-picking, ROI selection was based on anatomical references justified not only by the whole brain results, but also by the theoretical framework adopted, that is, the hub-and-spoke model (Ralph et al., 2017) and the idea of a perceptual-to-conceptual gradient of information processing along the ventral visual path (Borghesani and Piazza, 2017). Following previous investigations of the oscillatory dynamics of (visual) semantic processing (Mollo et al., 2017; Clarke et al., 2018), we draw three spheres of 20 mm radius along the vOT: left occipital pole (OCC, MNI: −10, –94, −16), left vOT (MNI: −50, –52, −20), and left ATL (MNI: −30, –6, −40). Figure 4 illustrates, for each frequency band, the evolution of single subjects’ values across the whole epoch. The second one aimed at characterizing the full time-frequency spectrum in two voxels centered in the OCC ROI (MNI: −10, –94, −16) and in the peak of activation observed at a cohort level (MNI: −34.8, –93.9, 2.7). For each subject, to extract activity at these specific locations, a broadband covariance matrix was first computed with all trial epoch data. This sample covariance matrix and the column-normalized lead field matrix specific to each voxel was used to calculate a linearly constrained minimum variance spatial filter (Van Veen et al., 1997). Broadband source activity for that voxel in each epoch was estimated by applying the spatial filter on the sensor data and projecting along the orientation with the maximum power. The estimated voxel time series was then subject to time-frequency analysis implemented in Fieldtrip (Oostenveld et al., 2011), using multi-taper spectral estimation methods. Event-related spectral power changes (2–120 Hz in 1 Hz steps) were estimated from the time-frequency decomposition, by scaling the length of the time window and the amount of frequency smoothing according to the frequency by a factor of 5 and 0.4, respectively (e.g., the time window at 10 Hz is 500 ms, the frequency smoothing 4 Hz). Once all subjects’ power spectra had been computed, group averages for both svPPA and HC were calculated and plotted (Figure 4).

Data availability and visualization

Request a detailed protocol

The sensitive nature of patients’ data and our current ethics protocol do not permit open data sharing. However, anonymized, pre-processed, group-level data used to generate the figures have been uploaded to NeuroVault [https://neurovault.org/collections/FTKQLDFP/]. The clinical and neuroimaging data used in the current paper are available from the senior author (SN), upon formal request indicating name and affiliation of the researcher as well as a brief description of the use that will be done of the data. All requests will undergo UCSF-regulated procedure thus require submission of a Material Transfer Agreement (MTA) which can be found at https://icd.ucsf.edu/material-transfer-and-data-agreements No commercial use would be approved. All images are rendered with the BrainNet Viewer (http://www.nitrc.org/projects/bnv/; Xia et al., 2013).

Data availability

The sensitive nature of patients' data and our current ethics protocol do not permit open data sharing. However, anonymized, pre-processed, group-level data used to generate the figures have been uploaded to NeuroVault [https://neurovault.org/collections/FTKQLDFP/]. The clinical and neuroimaging data used in the current paper are available from the Senior Author (S.N.), upon formal request indicating name and affiliation of the researcher as well as a brief description of the use that will be done of the data. All requests will undergo UCSF regulated procedure thus require submission of a Material Transfer Agreement (MTA) which can be found at https://icd.ucsf.edu/material-transfer-and-data-agreements No commercial use would be approved.

The following data sets were generated
    1. Borghesani V
    (2021) NeuroVault
    ID FTKQLDFP/. Neural dynamics of semantic categorization in semantic variant of Primary Progressive Aphasia.
    1. Borghesani V
    (2021) Material Transfer Agreement
    ID agreements. Incoming MTAs/Outgoing Human MTAs/Data Agreements.

References

    1. Kanwisher N
    2. McDermott J
    3. Chun MM
    (1997)
    The fusiform face area: a module in human extrastriate cortex specialized for face perception
    Journal of Neuroscience 17:4302–4311.
  1. Book
    1. Robinson S
    2. Vrba J
    (1999)
    Functional neuroimaging by synthetic aperture magnetometry (SAM)
    In: Yoshimoto T, editors. Recent Advances in Biomagnetism. Sendai, Japan: Tohoku University. pp. 302–305.

Decision letter

  1. Chris I Baker
    Senior and Reviewing Editor; National Institute of Mental Health, National Institutes of Health, United States
  2. Alex Clarke
    Reviewer; University of Cambridge, United Kingdom
  3. Aneta Kielar
    Reviewer; University of Arizona, United States

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

This study investigates how dysfunction in the anterior temporal lobe (ATL) alters dynamic activity during semantic categorization. Magnetoencephalography (MEG) responses were contrasted between patients with semantic variant Primary Progressive Aphasia (svPPA) and age-matched healthy controls. Despite similar profiles of behavioural performance on the categorization task, the svPPA patients showed enhanced γ synchronization in the occipital lobe compared to controls suggesting an increased engagement of early perceptual mechanisms for completing the task, as opposed to semantic identification of the picture.

Decision letter after peer review:

Thank you for submitting your article "Neural dynamics of semantic categorization in semantic variant of Primary Progressive Aphasia" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, and the evaluation has been overseen by Chris Baker as Reviewing and Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Alex Clarke (Reviewer #2); Aneta Kielar (Reviewer #3).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

As the editors have judged that your manuscript is of interest, but as described below that additional experiments are required before it is published, we would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). First, because many researchers have temporarily lost access to the labs, we will give authors as much time as they need to submit revised manuscripts. We are also offering, if you choose, to post the manuscript to bioRxiv (if it is not already there) along with this decision letter and a formal designation that the manuscript is "in revision at eLife". Please let us know if you would like to pursue this option. (If your work is more suitable for medRxiv, you will need to post the preprint yourself, as the mechanisms for us to do so are still in development.)

Summary:

Borghesani and colleagues aimed to understand how dysfunction in the anterior temporal lobe (ATL) alters dynamic activity during semantic categorization. They contrast MEG responses between 18 patients with semantic variant Primary Progressive Aphasia (PPA) and 18 age-matched healthy controls. Both groups show similar profiles of behavioural performance on the task, and broad similarities in MEG responses. Critically, however, svPPA patients show enhanced γ synchronization in the occipital lobe compared to controls. The authors interpret this as reflecting increased engagement of / reliance on early perceptual mechanisms for completing the task, as opposed to semantic identification of the picture.

Overall, the reviewers found the manuscript interesting. As svPPA is a rare (but scientifically informative) disorder, the sample size is impressive, and given that relatively few MEG studies exist in PPA at all, this is an interesting dataset. However, the general opinion is that the results could be more fully characterized, which would allow for more expansive interpretations and inferences.

Essential revisions:

1) Statistical thresholding

Using a high threshold prevents false positives, but may also lead to false negatives, and that may be the case here, with the high threshold contributing to an unrealistic impression of spatial specificity in MEG. It is obvious from the average responses in both groups that these oscillatory responses are widespread through the brain. Indeed, the α and β responses are significant in the majority of cortical voxels. This basic property of the responses should be presented clearly and prominently in the paper – not just in supplementary information where only a minority of readers will even see it. The authors then use an extremely high and conservative statistical threshold to contrast differences between the two groups. P<.005 uncorrected is a highly conservative threshold already, even before cluster-thresholding is added (although with data as smooth as MEG beamforming solutions, cluster-thresholding is unlikely to change anything). Essentially, this makes only the strongest part of the activation survive, and while it is valid to conclude that a significant group difference exists (protected from Type 1 error), this can also give a false impression that the difference is specific to that region. A more realistic characterization of the results would involve measuring differences in the strength of the responses between groups on a broader level, possibly the sensors or in large ROIs – and not ROIs pre-selected to show a dramatic difference by first searching the whole brain for the most significant effects – that is the classic "double-dipping" fallacy in neuroimaging.

2) Frequency bands

The ERD/ERS in each frequency band is treated as a separate entity, ignoring the fact that these bands are arbitrary and frequency is a continuous quantity. This matters because much is made of the fact that svPPA participants exhibited greater ERS in the low-γ range, and that this was correlated with reaction time. Supplementary Figure 1 shows that both groups had strong occipital ERS in the high-γ range, but only svPPA showed it in the low γ range as well. This suggests that the ERS in the svPPA group may simply have been shifted to a lower frequency range. A more fulsome characterization of these group differences via time-frequency analysis and/or power spectral analysis would help clarify what is going on here.

3) Decreased responses in svPPA?

It is surprising that svPPA participants only exhibited increased MEG responses compared to controls – assuming that both γ ERS and β ERD can be interpreted as increased neural activation, which is a reasonable assumption based on the literature. No decreases in the svPPA group are found, and thus the observed increases can be plausibly attributed to compensatory processes as framed by the authors. However, certain analysis choices may play a role in producing this data pattern. In particular, the authors state (line 611): "To remove potential artifacts due to neurodegeneration or eye movement (lacking electrooculograms), we masked statistical maps using patients' ATL atrophy maps (see section MRI protocol and analyses), as well as a ventromedial frontal mask."

It is not clear whether this masking was conducted in group space from average atrophy maps, or on an individual level. In either case, this is not well justified. What is the physical mechanism by which tissue undergoing neurodegeneration can be said to generate an artifactual signal? Atrophied tissue still contains living neurons with ionic currents; these are real signals not artifacts, and furthermore, atrophy is a continuous process with tissue further from the epicenter also undergoing similar neurodegenerative mechanisms. Atrophied tissue may well generate electromagnetic signals that are different from healthy tissue, and such differences should be included in this paper. There may be regions of hypoactivation as well as hyperactivation in this svPPA group. If the hypoactivation localizes to atrophied tissue and the hyperactivation to other regions, that will bolster the case that we are seeing compensatory processes, but it isn't certain with half the story masked. The statistical masking of the frontal region is also not really a valid solution to eye movement artifacts. The authors would have to present evidence that the region that they masked corresponds to the region potentially affected by eye movements. However, many studies have found that beamforming already does a pretty good job of removing ocular artifact from estimated brain signals, except for very close to the eyes.

4) RT correlation

The correlation with reaction time in the occipital cortex is consistent with the idea that the ERS there may reflect compensatory overreliance on perceptual information, but it isn't conclusive. The authors suggest that svPPA patients are able to categorize the stimuli correctly based on visual features, but are unable to name them. What about testing for correlations with the out-of-scanner behavioural measures that established that the patients have a naming deficit? It would strengthen the case if atrophy or hypoactivation (see comment above) correlated with the naming deficit.

5) Neural dynamics

As the paper is about 'Neural dynamics', this aspect could be developed, with the timing of the effects characterized further, and considered more in relation to the conclusions. For example, the main finding is the increased occipital γ response in svPPA compared to controls. Looking at Figure 3, there is a peak in the svPPA group near 200 ms, and very little synchronized activity in the control group. This is interesting as there are many ways we could have seen svPPA > controls, but this suggests that the γ synchronization response associated with compensation is specific to the svPPA group (and largely absent from controls – also from Supp Figure 1), and is distinguished from an initial visual evoked response (peaking ~100 ms). We recommend discussing and characterizing the dynamics of this effect more, such as what a later occipital effect could tell us about dynamics given ATL dysfunction? Is this increase a result of a lack of top-down effects from ATL?

6) Low-level vs. High-level

The occipital γ effect looks like the primary visual cortex, which might suggest the effects are not related to higher-level perceptual features (such as has eyes, teeth) as the authors suggest, but rather low-level visual effects. Do the authors perhaps think the effects could relate to enhanced processing of visual details (as related to the ideas of Hochstein and Asher's reverse hierarchy), or whether the effects relate to additional visual input following a visual saccade?

7) VBM

The VBM results for the svPPA patients were surprising given that all the atrophy appeared in the left hemisphere. There can be hemispheric differences in svPPA, but is this a true lateral pattern (meaning the right ATL is intact) or a product of VBM being run so that the most atrophied hemisphere is shifted to the left side? If the VBM maps are correct, and the svPPA patients are only showing left hemisphere atrophy, then what does this suggest about the role of the right ATL, and the bilateral nature of occipital increased in svPPA?

8) Task performance

Both svPPA patients and healthy controls achieved around 80% accuracy in the categorization task. This seems surprisingly low given, (1) the task (living vs. nonliving after seeing the image for 2 seconds), (2) that all the images were pretested and had high name agreement, and (3) that items were repeated on average 2.5 times. Is there something that explains this low performance for all individuals?

9) Compensation

One question for clarification is whether the recruitment of the occipital areas in svPPA is truly "compensatory", does it indicate a shift of resources due to the anterior temporal atrophy. Is the recruitment of the parieto-occipital regions associated with more accurate performance?

10) Other frequency bands (related to point 2 above)

The main results concentrate on the differences between patient and controls in the low γ range. There are also significant effects in the other frequency bands (e.g., high γ, β and α). What is the functional significance of these effects?

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for submitting your article "Neural dynamics of semantic categorization in semantic variant of Primary Progressive Aphasia" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, and the evaluation has been overseen by Chris Baker as the Reviewing and Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Jed A Meltzer (Reviewer #1); Alex Clarke (Reviewer #2).

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission. In general, the reviewers are still positive about the manuscript but think that the claims need to be tempered slightly and would like to see the time-frequency dynamics presented in more detail (as requested in the original reviews).

Essential Revisions:

1) Further analysis of the time-frequency dynamics is needed as laid out in the reviewers' comments below.

2) While the findings are consistent with a compensatory interpretation, especially given the equivalent performance in both groups, other interpretations are also possible. This should be discussed more fully, and the discussion could be grounded in earlier literature that has considered similar compensatory accounts e.g. age differences – for example many papers by Cheryl Grady show that older adults have more bilateral activation than younger. Those results were considered in the context of what kinds of findings constitute evidence of compensation vs. pathology.

Reviewer #1:

The revision by Borghesani et al., is much improved in terms of technical procedures and description, and most of the concerns raised by the reviewers have been adequately addressed. It is an interesting finding in a somewhat rare patient group.

I really only have one remaining concern that I still think should be addressed.

This paper puts a lot of emphasis on a particular interpretation of changes in oscillatory dynamics between the svPPA group and the control group. Based mainly on one particular finding – increased low-γ ERS in the occipital cortex for the svPPA group, the authors argue that svPPA patients compensate for their conceptual impairments by increasing their reliance on early perceptual processing implemented in occipital areas. Originally this interpretation was supported by both the increased low-γ ERS and also a correlation with performance. Since the changed analysis procedures resulted in dropping the claim of correlation, everything now rests on the shoulders of that low-γ finding. I think it needs to be unpacked a bit more.

If the increased low-γ finding were unambiguously interpretable as "activation" or "recruitment," this would be a straightforward story. But MEG data is complex and nuanced, more so than fMRI in my opinion, and there are some nuances here that are being overlooked. Both groups have robust activation in a higher band, high-γ, a band which is more strongly linked to increased neural firing and increased BOLD than the low-γ band is. On the other hand, the patients appear to have somewhat less ERD in the β band in this area, and β ERD is also strongly linked to neural firing and BOLD. The low γ band is kind of tricky – sometimes it goes up, sometimes it goes down. To understand this more, it would definitely help to see a real time-frequency decomposition of the activity, at least in this one key area.

We asked for this in the first round of review, and the authors declined to do it, citing concerns about time-frequency resolution tradeoff. That is not very convincing – there is ample resolution available in this data to characterize the effect in both time and frequency, and anyway in this case it is really frequency that raises the important questions – the group difference lasts for at least 400 ms so fine temporal resolution isn't so necessary. The authors argue that a lack of significant difference for the high γ band argues against a "frequency shift" interpretation – perhaps "spread" would be a more precise term than shift; in any case, it is clear that frequency is a key dimension in the difference of oscillatory response between these two groups, and it needs to be characterized better given the importance of this finding.

Perhaps a more practical concern is that the authors used optimized beamforming weights for specific frequency bands, precluding a traditional broad-band time-frequency analysis. However, they can still characterize time-frequency reactivity using an additional post-hoc analysis. This could be done on the sensor level, which I understand the authors do not prefer for legitimate reasons, but it could also be done in source space with non-frequency-optimized beamforming weights. This may not afford the same spatial resolution, but the blob of differential γ activity between groups is very large; precise spatial resolution isn't needed to answer this question.

I also think that given this ambiguity in the central finding, the authors should soften their conclusions somewhat and offer alternative interpretations. There is certainly a difference in the occipital lobe between groups, and that is interesting, but the idea that it's a compensatory increase in the patient group is somewhat speculative – consistent with the data, but not proven.

Reviewer #2:

I've read through all the comments and review responses, and think overall the manuscript is improved and several points made clearer.

I think there are a few points that remain for me:

1. The source analysis procedure is clear, along with thresholding and cluster extent. Yet, I didn't see any information on how the authors control for the effects over the sliding time windows, or for the frequency bands? We're these statistical contrasts taken into account?

2. New ROI data is presented showing the effects in 3 regions and across the frequency bands, with the authors claiming a difference in low γ activity around 100 ms. Yet stating the effect is around 100 ms doesn't seem to capture the data in the plot. It looks like difference may first appear around 100 ms, but peak nearer 200 ms, and continue throughout the epoch. I think a fuller description is warranted.

3. The ATL is no longer masked out from any of the analysis, and I would state this somewhere for clarity. There is also apparent signal coming from the atrophy region – mainly in β and α – it might be worth commenting on this.

4. Finally, to avoid switching back between Figures 2/3 and Table 3, I would consider adding if the effects relate to ERS or ERD in the table.

Reviewer #3:

I thank authors for addressing reviewers' comments. I think that the manuscript has improved. I don't have further comments.

https://doi.org/10.7554/eLife.63905.sa1

Author response

Essential revisions:

1) Statistical thresholding

Using a high threshold prevents false positives, but may also lead to false negatives, and that may be the case here, with the high threshold contributing to an unrealistic impression of spatial specificity in MEG. It is obvious from the average responses in both groups that these oscillatory responses are widespread through the brain. Indeed, the α and β responses are significant in the majority of cortical voxels. This basic property of the responses should be presented clearly and prominently in the paper – not just in supplementary information where only a minority of readers will even see it.

We thank the Reviewers for highlighting the value of the within-group whole brain activations. The information previously provided by our supplementary figure is now included as the first main figure in the manuscript (see updated Figure 2).

The authors then use an extremely high and conservative statistical threshold to contrast differences between the two groups. P<.005 uncorrected is a highly conservative threshold already, even before cluster-thresholding is added (although with data as smooth as MEG beamforming solutions, cluster-thresholding is unlikely to change anything). Essentially, this makes only the strongest part of the activation survive, and while it is valid to conclude that a significant group difference exists (protected from Type 1 error), this can also give a false impression that the difference is specific to that region.

While we agree that our approach has been rather conservative, we hold that this is warranted to avoid detection of false positives and focus on meaningful group-level effects. We would also like to note that the anonymized, pre-processed, group-level data used to generate the figures have been uploaded to NeuroVault [https://neurovault.org/collections/FTKQLDFP/]. This allows direct exploration of the results to further assess the regional specificity of the effects we here report and discuss.

A more realistic characterization of the results would involve measuring differences in the strength of the responses between groups on a broader level, possibly the sensors or in large ROIs – and not ROIs pre-selected to show a dramatic difference by first searching the whole brain for the most significant effects – that is the classic "double-dipping" fallacy in neuroimaging.

First, we would like to stress that our analyses do look for differences at the broadest level (whole-brain, all frequencies and time windows). The post-hoc region of interest analysis was conducted with the sole purpose of investigating the relation between subjects’ behavioral performance and neural activity in the two clusters that showed the strongest effects.

However, we agree with the Reviewers that our approach was inherently arbitrary and that generally speaking ROI-selection can lead to circularity and cherry-picking (not to mention multiple comparisons issues if one was to test out many different ROIs). We recognize that a better approach is that of showing group-level time plots across all frequencies in a-priori defined ROIs. Our final selection of three ventral occipito-temporal ROIs is consistent with the results of our whole brain analyses while being theory-driven. It builds on the theoretical framework we adopted (the hub-and-spoke model, Lambon-Ralph et al., 2016) and the previously put forward hypothesis of a perceptual-to-conceptual gradient of information processing along the ventral visual path (Borghesani and Piazza, 2017). Following, Mollo et al., 2011 and Clarke et al., 2018, we built spheres of 20 mm radius in three locations previously associated with visual and semantic processing: occipital pole (OCC, MNI: −10, −94, −16), left ventral occipitotemporal cortex (VOT, MNI:−50, −52, −20), and left ATL (MNI: −30, −6, −40), see updated Figure 4.

Finally, we would argue that a sensor level analysis approach will not yield (additional) meaningful insights. Any MEG sensor has broad sensitivity to signals arising from many brain regions. Therefore, it is very difficult to evaluate sensor level information in terms of brain topographies. In contrast, our analyses in brain source space following source reconstruction allows for clear interpretation of the findings in terms of brain topography.

2) Frequency bands

The ERD/ERS in each frequency band is treated as a separate entity, ignoring the fact that these bands are arbitrary and frequency is a continuous quantity. This matters because much is made of the fact that svPPA participants exhibited greater ERS in the low-γ range, and that this was correlated with reaction time. Supplementary Figure 1 shows that both groups had strong occipital ERS in the high-γ range, but only svPPA showed it in the low γ range as well. This suggests that the ERS in the svPPA group may simply have been shifted to a lower frequency range. A more fulsome characterization of these group differences via time-frequency analysis and/or power spectral analysis would help clarify what is going on here.

We agree with the Reviewers that frequency bands are inherently arbitrary. The ability to resolve time-frequency components in ERD/ERS are governed by time-frequency trade-off principles. To obtain, higher frequency resolution requires longer-time windows for analyses which in-turn will limit the temporal resolution of the source reconstructions. Conversely, to obtain high-temporal resolution one will forsake frequency resolution. The approach we have taken in relation to the time-frequency trade-off is three-fold: (1) Time-frequency optimized source reconstruction, where the source reconstruction adaptive filters are optimized for a particular time and frequency resolution, (2) Pre-specified frequency bands based on the literature, and (3) Variable time-windows for different frequency bands (100 ms for high-γ and 300 ms for α bands). A consequence of this approach indeed is that we are not able to examine the frequency content with higher resolution. Alternative approaches in the literature have been to forsake temporal resolution for higher frequency resolution, as suggested by the Reviewers. However, an unintended consequence of such an approach will result in non-time-frequency optimized source reconstruction with reduced sensitivity to higher frequency bands like γ band.

Critically, our results cannot be reduced to a shift in frequency range as specific spatio-temporal clusters are detected across the various bands. For instance, focusing on the occipital low γ effect mentioned by the Reviewers, arguably the strongest result, one should note that the svPPA > HC in low γ is not associated with a HC > svPPA in high γ. Rather than the interpretation suggested by the Reviewers, this observation suggests that computations in svPPA (but not HC) are occupying both low and high γ ranges.

Finally, as discussed above, our post-hoc ROI analyses now include all frequency bands, allowing further comparison of the full spectrum of time-frequency effects (see updated Figure 4).

3) Decreased responses in svPPA?

It is surprising that svPPA participants only exhibited increased MEG responses compared to controls – assuming that both γ ERS and β ERD can be interpreted as increased neural activation, which is a reasonable assumption based on the literature. No decreases in the svPPA group are found, and thus the observed increases can be plausibly attributed to compensatory processes as framed by the authors.

First of all, we would like to stress that all our analyses are two-tailed, allowing for effects in either direction to emerge. We have now clarified this aspect in the method section.

Second, we would like to point out that not all significant clusters denote increased activity in svPPA (vs. HC). We apologize for the poor visualization choices of the original Figure 2 (see now updated Figure 3), but few (smaller) clusters of HC>svPPA are detected, for instance, in the β band (late in the epoch, in left middle temporal gyrus and superior frontal gyrus, see negative t-values in Table 3).

Third, we would like to highlight that smaller, weaker, effects for HC > svPPA should not come as a surprise. As discussed in our manuscript, higher activations for svPPA (vs. controls) are the most common finding in the few previous functional neuroimaging studies comparing these populations, see for instance PET results from Mummery et al., 1999, fMRI results from Wilson et al., 2009, MEG results from Pineault et al., 2019.

However, certain analysis choices may play a role in producing this data pattern. In particular, the authors state (line 611): "To remove potential artifacts due to neurodegeneration or eye movement (lacking electrooculograms), we masked statistical maps using patients' ATL atrophy maps (see section MRI protocol and analyses), as well as a ventromedial frontal mask."

It is not clear whether this masking was conducted in group space from average atrophy maps, or on an individual level. In either case, this is not well justified. What is the physical mechanism by which tissue undergoing neurodegeneration can be said to generate an artifactual signal? Atrophied tissue still contains living neurons with ionic currents; these are real signals not artifacts, and furthermore, atrophy is a continuous process with tissue further from the epicenter also undergoing similar neurodegenerative mechanisms. Atrophied tissue may well generate electromagnetic signals that are different from healthy tissue, and such differences should be included in this paper. There may be regions of hypoactivation as well as hyperactivation in this svPPA group. If the hypoactivation localizes to atrophied tissue and the hyperactivation to other regions, that will bolster the case that we are seeing compensatory processes, but it isn't certain with half the story masked.

We agree with the Reviewers that the field currently lacks proper strategies to deal with atrophy while attempting to source localize effects in neurodegenerative patients. Please note that the first and main issue is that of defining atrophy itself: atrophic tissue will vary subject by subject (and it is not a all-or-none phenomenon) yet a threshold would need to be established.

In previous work, we took the parsimonious approach of masking out atrophic regions from group level statistics to avoid uninterpretable results (e.g. Borghesani et al., 2020). In other settings, one option is that of correcting region-of-interest statistics including GM volume as covariate (e.g. Ranasinghe et al., 2017). Finally, some would simply report whether electrophysiological differences and atrophy maps overlap or not (e.g., Kielar et al., 2018). Clearly, the underlying issue of how tissue undergoing neurodegeneration affects source modeling is an open problem that requires further exploration.

The statistical masking of the frontal region is also not really a valid solution to eye movement artifacts. The authors would have to present evidence that the region that they masked corresponds to the region potentially affected by eye movements. However, many studies have found that beamforming already does a pretty good job of removing ocular artifact from estimated brain signals, except for very close to the eyes.

We agree with the Reviewers that beamforming already minimizes the effect of ocular artifact (see Bardouille, Picton and Ross, 2006. Correlates of eye blinking as determined by synthetic aperture magnetometry. Clin. Neurophysiol. 117, 952–958.), and that in absence of proper electrooculograms there are no ideal procedures to verify whether ventro-medial frontal signals are spurious or not.

Thus, following Reviewers suggestion of minimizing data manipulation to avoid false negative results (while keeping a conservative approach minimizing false positive ones), we have repeated all our analyses removing the atrophy masking and frontal area masking steps. The current results are obtained simply ensuring that the statistical maps include only voxels designated as grey matter within the AAL atlas (Tzourio-Mazoyer et al., 2002). Overall, the current approach leads to the same results with the added advantage of being more parsimonious.

As our current visualization allows to appreciate (see updated Figure 3), the effects we observe are largely outside atrophied areas. Clearly, the issue of the electrophysiological effects of neurodegeneration is open and warrants further dedicated investigations.

4) RT correlation

The correlation with reaction time in the occipital cortex is consistent with the idea that the ERS there may reflect compensatory overreliance on perceptual information, but it isn't conclusive. The authors suggest that svPPA patients are able to categorize the stimuli correctly based on visual features, but are unable to name them. What about testing for correlations with the out-of-scanner behavioural measures that established that the patients have a naming deficit? It would strengthen the case if atrophy or hypoactivation (see comment above) correlated with the naming deficit.

We thank the Reviewers for this suggestion. However, to avoid circularity/cherry-picking in the ROI selection, as advised by the Reviewers, we have now drastically modified our approach. The current post-hoc analyses are conducted on a-priori regions centered around coordinates taken from the literature rather than those obtained from group differences in the current study (see updated figure 4). Within these ROIs, we do not observe any correlation between reaction times and neural signal (in either cohort) and thus have removed the correlation analysis from the Results section.

Unfortunately, further investigation of brain-behavior correlations are severely hampered by (1) the size of our sample (i.e., only 18 data points in each group); (2) the fact that patients were not explicitly tested for naming on the stimuli used in the MEG classification task; (3) the fact that controls are at ceiling in the Boston Naming Task while patients show (as expected) frank impairment but not sufficient variability (see Table 1). Consistent with these issues, we find that BNT performance does not correlate with the performance measures obtained in the semantic classification task (neither RTs nor accuracy).

5) Neural dynamics

As the paper is about 'Neural dynamics', this aspect could be developed, with the timing of the effects characterized further, and considered more in relation to the conclusions. For example, the main finding is the increased occipital γ response in svPPA compared to controls. Looking at Figure 3, there is a peak in the svPPA group near 200 ms, and very little synchronized activity in the control group. This is interesting as there are many ways we could have seen svPPA > controls, but this suggests that the γ synchronization response associated with compensation is specific to the svPPA group (and largely absent from controls – also from Supp Figure 1), and is distinguished from an initial visual evoked response (peaking ~100 ms). We recommend discussing and characterizing the dynamics of this effect more, such as what a later occipital effect could tell us about dynamics given ATL dysfunction? Is this increase a result of a lack of top-down effects from ATL?

We thank the Reviewers for this comment. We believe that the major changes to the analytical steps now implemented, see major #1, address this issue as well. Namely, a better characterization of the full breath of neural dynamics, across the epoch and the frequencies, is now enabled by (1) the removal of the atrophy mask, (2) having put within-group findings as main result, see Figure 2, (3) the choice of 3 a-priori ROIs, and (4) the presentation of the group results in those regions across all frequencies.

The result section “Post-hoc region-of-interest analyses” now highlights how the main difference between svPPA patients and healthy controls is the heightened low γ activity over the occipital region, evident around 100 ms post stimuli onset but maintained across the whole epoch. These findings rule out an explanation of the observed whole brain differences as mere temporal (or frequency) shift while highlighting the spectral (and spatial) specificity of the main results. For a discussion on how these findings can be interpreted in terms of functional compensation, please see major point #9.

6) Low-level vs. High-level

The occipital γ effect looks like the primary visual cortex, which might suggest the effects are not related to higher-level perceptual features (such as has eyes, teeth) as the authors suggest, but rather low-level visual effects. Do the authors perhaps think the effects could relate to enhanced processing of visual details (as related to the ideas of Hochstein and Asher's reverse hierarchy), or whether the effects relate to additional visual input following a visual saccade?

We thank the Reviewers for giving us the opportunity of clarifying this aspect of our results. We would like to point out that given the regional specificity of MEG source reconstructions, our findings are not specific to primary visual cortex and most certainly include non-primary visual areas. We did not measure eye movements during the task and trials with very large eye-movement artifacts were removed by our preprocessing procedure. Therefore, we cannot quantify (nor compare) the number of saccades across cohorts. Please see also point # 1 and revised manuscript section “Post-hoc region-of-interest analyses” for a discussion on the regional specificity of our findings.

We appreciate the reference to the Reverse Hierarchy Theory (RHT), which was mainly proposed for visual perceptual learning, framing it as a top-down guided process, which begins at high-level areas of the visual system, and progresses backwards to the input levels as/if needed. If, as a “leap-of-faith, we view the changes in processing of visual semantic information in svPPA during the course of the disease as a maladaptive learning process, we would speculate that RHT would predict results that are consistent with our observation – namely that when higher-level areas of the visual system are impaired, the processing shifts to lower-levels along the visual cortical hierarchy. However, given our paradigm it is impossible to contrast heighten bottom-up information (i.e., additional visual input) vs. top-down enhanced processing. Dedicated empirical paradigms are needed to directly investigate the role of feedback connections in modulating information processing. Furthermore, recent empirical studies investigating top-down vs. bottom-up flow of information along the ventral visual path (Dijkstra et al., 2020) question the static nature of visual processing (in either direction) which is the basis of the RHT theory, Dijkstra’s findings associated perception with cycles of recurrent processing while detecting only one feed-back flow during imagery. Interestingly, the 11 Hz oscillation associated with the iteratively update of precepts along the visual hierarchy is consistent with the idea that γ-band activity is linked to bottom-up information while α and β bands with top-down (Fries, 2015). Generally speaking, we do not deem appropriate to further expand the discussion of our results in a top-down vs. bottom-up framework as our findings cannot really speak to this debate.

7) VBM

The VBM results for the svPPA patients were surprising given that all the atrophy appeared in the left hemisphere. There can be hemispheric differences in svPPA, but is this a true lateral pattern (meaning the right ATL is intact) or a product of VBM being run so that the most atrophied hemisphere is shifted to the left side? If the VBM maps are correct, and the svPPA patients are only showing left hemisphere atrophy, then what does this suggest about the role of the right ATL, and the bilateral nature of occipital increased in svPPA?

We apologize if our choice of threshold and colormap suggested a stronger-than-expected left lateralization of svPPA patients’ atrophy. As it can be appreciated now in Figure 1c [as well as in the unthresholded SPM T map loaded NeuroVault, ID 441824] the atrophy is bilateral yet clearly asymmetrical. As a matter of fact, at least 13 svPPA patients present marked left-lateralized atrophy with only 5 of them showing a bilateral or right-lateralized pattern. This percentage of L vs R and the overall pattern is in line with multiple previous evidence from our group as well as others (Binney et al., 2016; Snowden et al., 2018; Woollams and Patterson, 2018).

A comparison of L vs R predominant ATL damage, investigating if (and how) that would affect the observed bilateral occipital hyperactivation would be of interest. Unfortunately, in our sample we don’t have enough R-predominant cases to perform a sound comparison. Given the current understanding of the graded, bilateral, nature of the ATL semantic hub, with the R hemisphere appearing specialized for nonverbal, socio-emotional material, it could be speculated that predominantly R svPPA would have a worst performance, possibly linked to higher (attempt to compensate via) occipital activation. However, we believe that such (perhaps heightened) occipital activation would still have a bilateral presentation: to our knowledge, there are no reasons to believe that low-level visual analysis of the stimuli would show any lateralization.

8) Task performance

Both svPPA patients and healthy controls achieved around 80% accuracy in the categorization task. This seems surprisingly low given, (1) the task (living vs. nonliving after seeing the image for 2 seconds), (2) that all the images were pretested and had high name agreement, and (3) that items were repeated on average 2.5 times. Is there something that explains this low performance for all individuals?

We are very grateful to the Reviewers for pointing this inconsistency out. While total percentage accuracy was correctly computed out of the 170 trials, an error in our code was incorrectly computing the percentage accuracy condition-wise (living vs. nonliving). The error has been fixed and both the text and figure updated with the correct values. Please note that the mistake did not change the main observation of no statistical difference between cohorts (nor condition, nor their interaction): HC: living: 97.1±6.6, nonliving: 96.8±6.6; svPPA: living: 91.5±6.2, nonliving: 95.9±8.1.

9) Compensation

One question for clarification is whether the recruitment of the occipital areas in svPPA is truly "compensatory", does it indicate a shift of resources due to the anterior temporal atrophy. Is the recruitment of the parieto-occipital regions associated with more accurate performance?

We agree with the Reviewers that an interpretation of our findings as functional compensation is hampered by the disclosed and discussed limitations of our paradigm and sample. Namely, (1) the size of our sample; (2) the similar, almost at ceiling, performance during the classification task of patients and HC; and (3) the fact that patients were not explicitly tested for naming on the same stimuli. Please note that, with the novel ROIs, we do not observe any correlation between reaction times and neural signal, thus cannot provide any explicit evidence of higher occipital activation being associated with better performance.

We would still hold that our patients’ semantic deficit is well established (Table 1) and its origin in ATL neurodegeneration is clear (Figure 1c): the link between svPPA patients neuropsychological and neuroanatomical profiles is undisputed. It is also clear from our data that svPPA patients can perform the semantic classification task as well as healthy controls (Figure 1b), thus somehow overcoming their semantic loss. This behavioral effect is matched, at the neural level, with the spatiotemporal clusters of differential activity we describe (and discuss).

10) Other frequency bands (related to point 2 above)

The main results concentrate on the differences between patient and controls in the low γ range. There are also significant effects in the other frequency bands (e.g., high γ, β and α). What is the functional significance of these effects?

We thank the Reviewers for highlighting the breath of our results: indeed, the occipital low γ findings are the strongest, but other spatiotemporal clusters of interest emerge across all frequencies.

For further discussions of our approach and the resulting frequency-specific results, please see also major point #2 (i.e., ruling out simple time or frequency shift) and #6 (i.e., speculations on top-down vs. bottom-up effects), as well as minor point #4 (i.e., speculation on the local vs. global integration effects).

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission. In general, the reviewers are still positive about the manuscript but think that the claims need to be tempered slightly and would like to see the time-frequency dynamics presented in more detail (as requested in the original reviews).

We thank the Editor and the Reviewers for their feedback on our revised manuscript. We appreciate the time they invested and their thoughtful comments. We have now included a thorough time-frequency analysis to rule out potential confounds from frequency shift effects that may manifest as power changes. We have now also better framed our speculative interpretation of the results in the discussion. We thus believe that the attached revised paper responds to the few remaining concerns and, thanks to their suggestions, is greatly improved.

Essential Revisions:

1) Further analysis of the time-frequency dynamics is needed as laid out in the reviewers' comments below.

We acknowledge that the characterization of the full time-frequency dynamics over the area of interest would allow better appreciation of the observed effect. We now provide in Figure 5 the full time-frequency spectrum in two voxels: one located at the centroid of the OCC ROI (MNI: −10, −94, −16) and a second one located at the peak of activation observed at a cohort level, (MNI: -34.8 -93.9 2.7). For each subject, to extract activity at these specific locations, a broad-band covariance matrix was first computed with all trial epoch data. This sample covariance matrix and the column-normalized lead field matrix specific to each voxel was used to calculate a linearly constrained minimum variance spatial filter (LCMV, Van Veen et al., 1997). Broadband source activity for that voxel in each epoch was estimated by applying the spatial filter on the sensor data and projecting along the orientation with the maximum power. The estimated voxel time-series was then subject to time-frequency analysis implemented in fieldtrip (www.fieldtriptoolbox.org) using multi-taper spectral estimation methods. Event-related spectral power changes (2 to 120 Hz in 1-Hz steps) were estimated from the time–frequency decomposition, by scaling the length of the time window and the amount of frequency smoothing according to the frequency by a factor of 5 and 0.4 respectively (so for instance the time window at 10 Hz is 500ms, the frequency smoothing 4 Hz). Group averages for both svPPA and HC were calculated (using ft_freqgrandaverage) and plotted (using ft_singleplotTFR). The results, shown in Figure 5 , clearly rule out the possibility that observed effects (reported in Figure 2 and 3) can arise from frequency shift or spread. Instead, we observe broad and sustained increase in low-γ band power in svPPA patients that is not seen in healthy controls.

2) While the findings are consistent with a compensatory interpretation, especially given the equivalent performance in both groups, other interpretations are also possible. This should be discussed more fully, and the discussion could be grounded in earlier literature that has considered similar compensatory accounts e.g. age differences – for example many papers by Cheryl Grady show that older adults have more bilateral activation than younger. Those results were considered in the context of what kinds of findings constitute evidence of compensation vs. pathology.

We thank the Reviewers and the Editor for pointing out previous work on age-related differences and for suggesting a more thorough framing of our results within the broader “maintenance, reserve, and compensation” spectrum. We have now revised our discussion to include the following observation.

“Finally, our interpretation of higher low-γ as a sign of compensation is speculative and lies on two-fold inference: that more low-γ band in svPPA means more activity, and that more activity means compensation. Previous literature in healthy and pathological aging suggests that higher activation can be associated with compensatory effects or reflect neuropathology (e.g., Elman et al., 2014). Critically, when considering progressive phenomena such as neurodegeneration one has to acknowledge that hyper- and hypo- activations might reflect different stages of disease and that their relation to behavioral performance might follow a nonlinear U-shaped trajectory (e.g., Gregory et al., 2017). Following Cabeza and colleagues (2018), we believe that greater activity can be interpreted as compensation if two criteria are met: (1) there has to be evidence of a “supply-demand gap”, and (2) there has to be evidence of a beneficial effect on cognitive performance. Our data clearly fulfills the first criterion as svPPA patients present insufficient neural resources (i.e., relatively focal ATL atrophy) and suffer from its behavioral consequences (i.e., pervasive semantic loss). Our findings also fulfill the second criterion as svPPA patients are able to perform the task with both accuracy and reaction times comparable with the healthy controls. Thus, we believe that the neural and behavioral evidence we provide is enough to rule out alternatives to compensation such as inefficiency or pathology. While further studies are warranted to shed light onto the relation between neurodegeneration, neurophysiological markers, and behavior, we hold that the most appropriate (albeit speculative) interpretation of our findings is in terms of compensation.”

Reviewer #1:

The revision by Borghesani et al., is much improved in terms of technical procedures and description, and most of the concerns raised by the reviewers have been adequately addressed. It is an interesting finding in a somewhat rare patient group.

I really only have one remaining concern that I still think should be addressed.

This paper puts a lot of emphasis on a particular interpretation of changes in oscillatory dynamics between the svPPA group and the control group. Based mainly on one particular finding – increased low-γ ERS in the occipital cortex for the svPPA group, the authors argue that svPPA patients compensate for their conceptual impairments by increasing their reliance on early perceptual processing implemented in occipital areas. Originally this interpretation was supported by both the increased low-γ ERS and also a correlation with performance. Since the changed analysis procedures resulted in dropping the claim of correlation, everything now rests on the shoulders of that low-γ finding. I think it needs to be unpacked a bit more.

If the increased low-γ finding were unambiguously interpretable as "activation" or "recruitment," this would be a straightforward story. But MEG data is complex and nuanced, more so than fMRI in my opinion, and there are some nuances here that are being overlooked. Both groups have robust activation in a higher band, high-γ, a band which is more strongly linked to increased neural firing and increased BOLD than the low-γ band is. On the other hand, the patients appear to have somewhat less ERD in the β band in this area, and β ERD is also strongly linked to neural firing and BOLD. The low γ band is kind of tricky – sometimes it goes up, sometimes it goes down. To understand this more, it would definitely help to see a real time-frequency decomposition of the activity, at least in this one key area.

We asked for this in the first round of review, and the authors declined to do it, citing concerns about time-frequency resolution tradeoff. That is not very convincing – there is ample resolution available in this data to characterize the effect in both time and frequency, and anyway in this case it is really frequency that raises the important questions – the group difference lasts for at least 400 ms so fine temporal resolution isn't so necessary. The authors argue that a lack of significant difference for the high γ band argues against a "frequency shift" interpretation – perhaps "spread" would be a more precise term than shift; in any case, it is clear that frequency is a key dimension in the difference of oscillatory response between these two groups, and it needs to be characterized better given the importance of this finding.

Perhaps a more practical concern is that the authors used optimized beamforming weights for specific frequency bands, precluding a traditional broad-band time-frequency analysis. However, they can still characterize time-frequency reactivity using an additional post-hoc analysis. This could be done on the sensor level, which I understand the authors do not prefer for legitimate reasons, but it could also be done in source space with non-frequency-optimized beamforming weights. This may not afford the same spatial resolution, but the blob of differential γ activity between groups is very large; precise spatial resolution isn't needed to answer this question.

Please see the additional analysis and figure 5.

I also think that given this ambiguity in the central finding, the authors should soften their conclusions somewhat and offer alternative interpretations. There is certainly a difference in the occipital lobe between groups, and that is interesting, but the idea that it's a compensatory increase in the patient group is somewhat speculative – consistent with the data, but not proven.

This has been explicitly addressed in the limitations section.

Reviewer #2:

I've read through all the comments and review responses, and think overall the manuscript is improved and several points made clearer.

I think there are a few points that remain for me:

1. The source analysis procedure is clear, along with thresholding and cluster extent. Yet, I didn't see any information on how the authors control for the effects over the sliding time windows, or for the frequency bands? We're these statistical contrasts taken into account?

We’d like to clarify that we are conducting non-parametric stats with cluster thresholding to reduce spurious findings from voxel to voxel, adopting a more conservative p-value threshold (.005) than is typically reported, but with no correction of p values (nor additional tests) to account for multiple time windows or frequency bands. This information has been made explicit in the method section.

2. New ROI data is presented showing the effects in 3 regions and across the frequency bands, with the authors claiming a difference in low γ activity around 100 ms. Yet stating the effect is around 100 ms doesn't seem to capture the data in the plot. It looks like difference may first appear around 100 ms, but peak nearer 200 ms, and continue throughout the epoch. I think a fuller description is warranted.

This has been explicitly added to the Results section.

3. The ATL is no longer masked out from any of the analysis, and I would state this somewhere for clarity. There is also apparent signal coming from the atrophy region – mainly in β and α – it might be worth commenting on this.

This has been explicitly addressed in the limitations section.

4. Finally, to avoid switching back between Figures 2/3 and Table 3, I would consider adding if the effects relate to ERS or ERD in the table.

The suggested change has been implemented.

https://doi.org/10.7554/eLife.63905.sa2

Article and author information

Author details

  1. V Borghesani

    Memory and Aging Center, Department of Neurology, University of California, San Francisco, San Francisco, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing - original draft, Writing - review and editing
    For correspondence
    valentina.borghesani@ucsf.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-7909-8631
  2. CL Dale

    Department of Radiology and Biomedical Imaging, University of California, San Francisco, San Francisco, United States
    Contribution
    Data curation, Formal analysis, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
  3. S Lukic

    Memory and Aging Center, Department of Neurology, University of California, San Francisco, San Francisco, United States
    Contribution
    Conceptualization, Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
  4. LBN Hinkley

    Department of Radiology and Biomedical Imaging, University of California, San Francisco, San Francisco, United States
    Contribution
    Resources, Data curation, Software, Formal analysis, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
  5. M Lauricella

    Memory and Aging Center, Department of Neurology, University of California, San Francisco, San Francisco, United States
    Contribution
    Resources, Methodology
    Competing interests
    No competing interests declared
  6. W Shwe

    Memory and Aging Center, Department of Neurology, University of California, San Francisco, San Francisco, United States
    Contribution
    Resources, Methodology
    Competing interests
    No competing interests declared
  7. D Mizuiri

    Department of Radiology and Biomedical Imaging, University of California, San Francisco, San Francisco, United States
    Contribution
    Resources, Methodology
    Competing interests
    No competing interests declared
  8. S Honma

    Department of Radiology and Biomedical Imaging, University of California, San Francisco, San Francisco, United States
    Contribution
    Resources, Methodology
    Competing interests
    No competing interests declared
  9. Z Miller

    Memory and Aging Center, Department of Neurology, University of California, San Francisco, San Francisco, United States
    Contribution
    Resources
    Competing interests
    No competing interests declared
  10. B Miller

    Memory and Aging Center, Department of Neurology, University of California, San Francisco, San Francisco, United States
    Contribution
    Conceptualization, Funding acquisition, Project administration
    Competing interests
    No competing interests declared
  11. JF Houde

    Department of Otolaryngology, University of California, San Francisco, San Francisco, United States
    Contribution
    Conceptualization, Funding acquisition, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared
  12. ML Gorno-Tempini

    1. Memory and Aging Center, Department of Neurology, University of California, San Francisco, San Francisco, United States
    2. Department of Neurology, Dyslexia Center University of California, San Francisco, San Francisco, United States
    Contribution
    Conceptualization, Supervision, Funding acquisition, Writing - original draft, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared
  13. SS Nagarajan

    1. Department of Radiology and Biomedical Imaging, University of California, San Francisco, San Francisco, United States
    2. Department of Otolaryngology, University of California, San Francisco, San Francisco, United States
    Contribution
    Conceptualization, Supervision, Funding acquisition, Writing - original draft, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared

Funding

National Institutes of Health (R01NS050915)

  • ML Gorno-Tempini

National Institutes of Health (K24DC015544)

  • ML Gorno-Tempini

National Institutes of Health (R01NS100440)

  • JF Houde

National Institutes of Health (R01DC013979)

  • SS Nagarajan

National Institutes of Health (R01DC176960)

  • SS Nagarajan

National Institutes of Health (R01DC017091)

  • SS Nagarajan

National Institutes of Health (R01EB022717)

  • SS Nagarajan

National Institutes of Health (R01AG062196)

  • SS Nagarajan

Larry L. Hillblom Foundation

  • ML Gorno-Tempini

Global Brain Health Institute

  • ML Gorno-Tempini

University of California (MRP-17-454755)

  • SS Nagarajan

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

The authors thank the patients and their families for the time and effort they dedicated to this research. Funding: This work was funded by the following National Institutes of Health grants (R01NS050915, K24DC015544, R01NS100440, R01DC013979, R01DC176960, R01DC017091, R01EB022717, R01AG062196). Additional funds include the Larry Hillblom Foundation, the Global Brain Health Institute, and UCOP grant MRP-17–454755. These supporting sources were not involved in the study design, collection, analysis, or interpretation of data, nor were they involved in writing the paper or the decision to submit this report for publication.

Ethics

Human subjects: The study was approved by the UCSF Committee on Human Research and all subjects provided written informed consent (IRB # 11-05249).

Senior and Reviewing Editor

  1. Chris I Baker, National Institute of Mental Health, National Institutes of Health, United States

Reviewers

  1. Alex Clarke, University of Cambridge, United Kingdom
  2. Aneta Kielar, University of Arizona, United States

Publication history

  1. Received: October 10, 2020
  2. Accepted: June 21, 2021
  3. Accepted Manuscript published: June 22, 2021 (version 1)
  4. Version of Record published: June 29, 2021 (version 2)

Copyright

© 2021, Borghesani et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 839
    Page views
  • 130
    Downloads
  • 4
    Citations

Article citation count generated by polling the highest count across the following sources: PubMed Central, Crossref, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. V Borghesani
  2. CL Dale
  3. S Lukic
  4. LBN Hinkley
  5. M Lauricella
  6. W Shwe
  7. D Mizuiri
  8. S Honma
  9. Z Miller
  10. B Miller
  11. JF Houde
  12. ML Gorno-Tempini
  13. SS Nagarajan
(2021)
Neural dynamics of semantic categorization in semantic variant of primary progressive aphasia
eLife 10:e63905.
https://doi.org/10.7554/eLife.63905

Further reading

    1. Genetics and Genomics
    2. Neuroscience
    Carolyn Elya, Danylo Lavrentovich ... Benjamin de Bivort
    Research Article Updated

    For at least two centuries, scientists have been enthralled by the “zombie” behaviors induced by mind-controlling parasites. Despite this interest, the mechanistic bases of these uncanny processes have remained mostly a mystery. Here, we leverage the Entomophthora muscae-Drosophila melanogaster “zombie fly” system to reveal the mechanistic underpinnings of summit disease, a manipulated behavior evoked by many fungal parasites. Using a high-throughput approach to measure summiting, we discovered that summiting behavior is characterized by a burst of locomotion and requires the host circadian and neurosecretory systems, specifically DN1p circadian neurons, pars intercerebralis to corpora allata projecting (PI-CA) neurons and corpora allata (CA), the latter being solely responsible for juvenile hormone (JH) synthesis and release. Using a machine learning classifier to identify summiting animals in real time, we observed that PI-CA neurons and CA appeared intact in summiting animals, despite invasion of adjacent regions of the “zombie fly” brain by E. muscae cells and extensive host tissue damage in the body cavity. The blood-brain barrier of flies late in their infection was significantly permeabilized, suggesting that factors in the hemolymph may have greater access to the central nervous system during summiting. Metabolomic analysis of hemolymph from summiting flies revealed differential abundance of several compounds compared to non-summiting flies. Transfusing the hemolymph of summiting flies into non-summiting recipients induced a burst of locomotion, demonstrating that factor(s) in the hemolymph likely cause summiting behavior. Altogether, our work reveals a neuro-mechanistic model for summiting wherein fungal cells perturb the fly’s hemolymph, activating a neurohormonal pathway linking clock neurons to juvenile hormone production in the CA, ultimately inducing locomotor activity in their host.

    1. Neuroscience
    Flavia Venetucci Gouveia, Jurgen Germann ... Clement Hamani
    Research Article Updated

    Deep brain stimulation targeting the posterior hypothalamus (pHyp-DBS) is being investigated as a treatment for refractory aggressive behavior, but its mechanisms of action remain elusive. We conducted an integrated imaging analysis of a large multi-centre dataset, incorporating volume of activated tissue modeling, probabilistic mapping, normative connectomics, and atlas-derived transcriptomics. Ninety-one percent of the patients responded positively to treatment, with a more striking improvement recorded in the pediatric population. Probabilistic mapping revealed an optimized surgical target within the posterior-inferior-lateral region of the posterior hypothalamic area. Normative connectomic analyses identified fiber tracts and functionally connected with brain areas associated with sensorimotor function, emotional regulation, and monoamine production. Functional connectivity between the target, periaqueductal gray and key limbic areas – together with patient age – were highly predictive of treatment outcome. Transcriptomic analysis showed that genes involved in mechanisms of aggressive behavior, neuronal communication, plasticity and neuroinflammation might underlie this functional network.