Abstract
Alzheimer’s disease (AD) is known to be associated with disruption in semantic networks. Previous studies examining changes in spreading activation in AD have used a lexical decision task paradigm. We have used a paradigm based on average word frequencies obtained from the words generated on the Controlled Oral Word Association Test (COWAT) and the Animal Naming (AN) test. The COWAT and AN tests were administered to a group of 25 patients with AD and 20 control participants. We predicted that the patients with AD would have higher average word frequencies on the COWAT and AN tests than the control participants. The results indicated that the AD group generated words with a higher average word frequency on the AN test but a lower average word frequency on the COWAT. The reasons for the discrepancy in average word frequencies on the AN test and COWAT are discussed.
Degradation of semantic networks in Alzheimer’s disease (AD) is a well-known phenomenon and gives rise to many of the memory and cognitive deficits associated with AD. Naming to confrontation, verbal fluency, and semantic priming paradigms have been the most commonly used methods to demonstrate deficits in semantic networks in AD. Regarding naming to confrontation, patients with AD exhibited impaired naming, as indexed by performance on the Boston Naming Test (BNT), when compared to individuals with mild cognitive impairment 1,2 and normal, healthy controls. 1,2 A number of other investigations have demonstrated impaired naming performance in patients with AD, 3 –5 even with items that have high frequency. 6 An investigation by Smith et al 7 examined the types of errors committed by patients with AD on a naming task and found that although these patients were unable to name the objects, they were able to recognize the stimuli, being able to identify the semantic class to which the object belongs. Hence, naming deficits in patients with AD were attributed to disrupted semantic networks as opposed to visuoperceptual deficits. Balthazar et al 1 found that the performance of patients with AD on the BNT improved and matched the level of individuals with mild cognitive impairment and normal controls when provided with phonemic cues. The authors argued that this finding indicates that the impaired performance of patients with AD cannot be solely explained by a breakdown in semantic networks. This finding suggests that the semantic networks of patients with AD are also impaired at accessing or activating the phonological lexicon.
Performance on tests of semantic fluency has also been widely used to investigate the breakdown of semantic memory in patients with AD. The results of these investigations have found patients with AD to be significantly impaired on tests of semantic or category fluency. 3,6,8,9 A number of different semantic categories have been used in these investigations, including both living and nonliving categories. Graham et al 4 used 8 different categories, 4 categories of living things and 4 categories of man-made things. The results indicated significantly impaired category fluency in patients with AD, although the analyses did not address any category-specific impairments. Moreno-Martinez et al 10 examined the performance of patients with AD on semantic fluency using 14 different semantic categories, 7 from living categories (flowers, insects, trees, vegetables, fruits, animals, and body parts) and 7 from nonliving categories (buildings, vehicles, tools, furniture, musical instruments, kitchen utensils, and clothing). Although a number of sex differences were found in their study, the results generally indicated reduced semantic fluency in patients with AD across all semantic categories. Patients with AD also appear to perform significantly worse on measures of semantic fluency than on measure of lexical or phonemic fluency. 11 –13
Semantic priming paradigms have been used quite extensively to investigate the spreading activation in semantic memory networks. The theory of spreading activation proposed by Collins and Loftus 14 proposes that the semantic networks that represent specific semantic memories (eg, maple) are organized into a larger network that comprises concepts (eg, trees). Semantic memories are then represented as nodes within these conceptual networks, and the semantic nodes (eg, maple and oak) within a conceptual network (eg trees) are more strongly interconnected through associative, bidirectional links than are semantic nodes from different conceptual networks (eg, maple and car). The strength of the connections between nodes within a conceptual network varies, with some connections being quite strong and others being relatively weaker. Activation of any given node within a conceptual network is then proposed to spread, in a parallel fashion, along the associative links to other related nodes that also comprise the network. The strength of the associative links between nodes is determined by the production frequency norms or the frequency of use of the links.
The use of semantic priming paradigms differs greatly from the aforementioned methods of investigating semantic networks in that priming studies involve implicit memory as opposed to explicit memory. Also, unlike investigations using naming and fluency, the results of studies using semantic priming paradigms have been far from equivocal (for a review see. 15 ). Specifically, some have reported reduced priming or hypopriming in patients with AD16-18 and others have reported equivalent priming between patients with AD and normal controls. 19 –21 Still other investigations have found increased or hyperpriming in AD. 22,23 Differences between living and nonliving stimuli have been reported. Specifically, Hernandez et al 24 reported reduced priming for artifacts but normal priming for animals. However, as reviewed by Giffard et al, 15 a number of methodological limitations exist with semantic priming paradigms. These include the possibility that a partial degradation of semantic representations may exist even with apparently normal semantic priming. Further, semantic priming effects are quite sensitive to a number of different experimental manipulations. 15 Finally, unlike paradigms using naming and semantic fluency, which primarily use a top-down approach, in semantic priming paradigms the stimuli are provided and thus use a bottom-up approach.
We have used a different paradigm to investigate spreading activation in lexical and semantic networks, a paradigm based on average word frequencies from the responses to the Controlled Oral Word Association Test (COWAT) and Animal Naming (AN) test. The COWAT requires patients to generate as many words as possible that begin with a specified letter, and the AN test requires patients to generate as many names of animals as possible. Typically, patients are given 60 seconds to generate the words. To measure spreading activation, the word frequency of each word generated is obtained and then averaged. According to the Hebbian principle that neurons that fire together wire together, the connections between nodes within a semantic network should be strengthened with increased frequency of use. Thus, relatively higher frequency words should have stronger and more numerous connections with other words in the network, because they are accessed and used more frequently. Conversely, words that occur less frequently should possess weaker neural representations and fewer links to other words within the network. As a result, greater spreading activation is required to activate lower frequency words. Taken together, when computing the average word frequency for the words generated on the COWAT and AN test, increasing spreading activation will lower the overall average word frequency, because the greater extent of spreading activation will activate relatively more lower frequency words. Conversely, decreasing spreading activation will result in a higher overall average word frequency. We have used this method of measuring spreading activation to investigate the effects of Parkinson’s disease, 25 depression, 26 and acetylcholinesterase inhibitors (AChEIs) in patients with dementia 27 on spreading activation in lexical/semantic memory networks.
Measuring spreading activation using word frequencies taken from tests of verbal fluency has the advantage of assessing performance when patients are using a top-down strategy for accessing their lexical and semantic networks. Unlike the semantic priming paradigms that use stimuli controlled by the experimenter, the paradigm using word frequencies from verbal fluency permits an assessment of spreading activation that is not as constrained by experimental methodologies. As a result, ecological validity may also be enhanced. Thus, we sought to investigate spreading activation in patients with AD using this method of calculating average word frequencies from the COWAT and AN tests. Based on the previous research of degradation in semantic memory networks in AD, our hypothesis was that patients with AD would exhibit a significantly higher average word frequency (reduced spreading activation) for the AN test when compared to healthy controls. However, lexical networks might be relatively preserved in AD. 28 As mentioned previously, research has consistently found worse performance on tests of semantic fluency as compared to tests of phonemic fluency in patients with AD. Hence, our second hypothesis was no difference would emerge between patients with AD and healthy controls when examining the average word frequencies from the COWAT.
Methods
Participants
A total of 25 patients (5 men and 20 women) diagnosed with probable AD participated in this investigation. Patients meet criteria for probable AD based on the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association (NINCDS-ADRDA) criteria 29 and were recruited from the Drexel University College of Medicine. The ages of the patients with AD ranged from 71 to 89 years (mean [M] = 79.64, standard deviation [SD] = 5.12), with an average education of 12.52 years (SD = 2.02). The average Mini-Mental State Examination (MMSE) score for the patients with AD was 22.52 (SD = 2.85). A sample of 20 control participants was also used (5 men and 15 women), with an age range of 51 to 82 years (M = 67.20, SD = 8.88) and an average 14.35 years of education (SD = 2.48). The average MMSE score for the controls was 29.15 (SD = 1.27). The controls were recruited from the community.
Apparatus
Mini-Mental State Examination
The MMSE is a screening test used to assess general cognitive functioning. Areas of functioning assessed include orientation, registration, attention, recall, working memory, language, and construction or drawing ability. The range of possible scores is from 0 to 30.
Controlled Oral Word Association Test
The COWAT requires the patient to name as many words as possible that begin with a specified letter (F, A, and S) within 60 seconds. However, they cannot use proper nouns, they cannot count, and they cannot use a stem word and then simply add different endings.
Animal Naming
The AN test requires the patient to name as many different animals as possible, with no restrictions based on beginning letter or any other characteristic. They are permitted 60 seconds to generate as many names of animals as possible.
Procedure
The study was approved by the Institutional Review Board of Drexel University, and all participants were treated in accordance with the ethical principles of the American Psychological Association. There are no financial or other conflicts of interest associated with this study. The MMSE, COWAT, and AN tests were administered to the patients with AD and controls using standard procedures. Following administration of the COWAT and the AN test, the frequency of each word generated was obtained. To ensure reliability in the findings, we used 2 different corpuses of word frequencies, the Francis-Kucera 30 and the American Heritage. 31 The Francis-Kucera 30 corpus was based on over 1 million graphic words from numerous different sources, including periodicals, novels, newspapers, technical writings, philosophical essays, and writings of fiction. The American Heritage corpus 31 was based on over 5 million graphic words from several different sources, including textbooks, workbooks, novels, poetry, nonfiction books, encyclopedias, and magazines. The average word frequency for the COWAT was calculated by first averaging the word frequencies for each letter used. The average word frequency across the 3 letters was then calculated, based on the average obtained for each letter. The average word frequency for the AN test was calculated by obtaining the word frequency for each animal named and then averaging across all the names of animals.
Results
Initial analyses indicated that the patients with AD and controls differed significantly in age, F 1,43 = 34.76, P < .0001, and education, F 1,43 = 7.45, P = .009. Expectedly, the MMSE score was also significantly different, F 1,43 = 93.43, P < .0001. Additional analyses indicated that the AD group generated significantly, F 1,43 = 33.87, P < .0001, fewer words on the FAS (M = 24.80, SD = 10.73) than the control group (M = 43.90, SD = 11.20). Likewise, the AD group also generated significantly, F 1,43 = 78.64, P < .0001, fewer words on the AN test (M = 9.44, SD = 3.98) than the control group (M = 21.10, SD = 4.84). Given the significant difference in age, education, and number of words generated on the FAS and AN tests, all subsequent analyses were conducted using these variables as covariates.
The word frequency data for the FAS and AN tests obtained from the Kucera-Francis and American Heritage were first analyzed using a multivariate analysis of covariance to control for experimentwise error rate. The result indicated that a significant difference existed between at least one of the means, F 4,36 = 5.40, P = .002. Examination of individual group comparisons based on the word frequency data obtained from the Francis-Kucera corpus was conducted first. The results indicated that the AD group obtained a significantly, F 1, 40 = 5.58, P = .023, R 2 = .132, lower FAS average word frequency (M = 129.79, SD = 167.82) than the control group (M = 205.99, SD = 146.12). No significant difference emerged between the AD (M = 41.42, SD = 17.24) and the control (M = 23.95, SD = 8.68) groups in AN average word frequency (see Figure 1). Data from the American Heritage corpus were then examined. The results indicated that the AD group obtained a significantly, F 1,40 = 4.07, P = .05, R 2 = .206, lower FAS average word frequency (M = 1389.99, SD = 2895.27) than the control group (M = 1700.55, SD = 1888.51). Additionally, the AD group obtained a significantly, F 1,40 = 5.18, P = .028, R 2 = .575, higher AN average word frequency (M = 400.46, SD = 139.44) than the control group (M = 268.11, SD = 98.14; see Figure 2).
Differences between patients with Alzheimer’s disease (AD) and controls in word frequencies from the Francis-Kucera corpus.
Differences between patients with Alzheimer’s disease (AD) and controls in word frequencies from the American Heritage corpus.
Discussion
The results of our study provide support for the hypothesis that patients with AD would exhibit reduced spreading activation in semantic memory networks as evidenced by significantly higher average word frequency on the AN test. However, the difference between our patients with AD and the controls was found only when using data obtained from the American Heritage corpus. The fact that no significant difference was found using the Francis-Kucera corpus may be due to the effects of a more restricted range for the Francis-Kucera corpus. As indicated by the means, the total possible word frequencies from the American Heritage corpus were much higher than that for the Francis-Kucera corpus. The result for the data based on the American Heritage corpus is in agreement with an extensive literature demonstrating impaired semantic memory networks in AD as evidenced by naming, 1,3 –5 fluency 3,6,8,9 and semantic priming. 16 –18
The finding that the patients with AD have a reduction in spreading activation in their semantic networks but relative preservation of the lexical networks was expected, since it is known since the work of Braak and Braak 32 that in the neocortex it is the supramodal and polymodal cortex that are the first to deteriorate. Delacourte 33 has written that the tauopathy associated with AD, “always progresses in the brain along a very precise and invariable pathway, from the entorhinal then hippocampal formation to polymodal association areas to end in primary regions … The cognitive impairment follows exactly the progression of the affected brain regions.” Patients with AD most often first have impaired episodic memory, and it has been well established that the entorhinal cortex and the hippocampal formation are area critical in mediating episodic memory. The inability to name and comprehend with a preserved ability to repeat implies a destruction of the semantic networks. Studies of patients with strokes who have transcortical sensory aphasia, with impaired naming and comprehension but intact repetition, have revealed that most often it is the posterior parietal cortex that is injured. 34 In addition, studies of patients with semantic dementia who also have impaired comprehension and naming but intact repetition has revealed degeneration of the left anterior temporal lobe. 35 According to Delacourte et al, 36 these are 2 of the polymodal association areas that are often affected in the course of AD. Within neuronal networks, connectivity is very heavily dependent upon dendritic arborizations, and Terry 37 has revealed that the one of the first changes in the neurons in those areas affected by AD is a loss of dendritic arborizations, and it is a loss of these extensive dendritic arborizations in these polymodal areas that reduces the spread of activation.
In contrast, to the semantic networks, the phonological lexicon is thought to be stored primarily in Wernicke’s area, which is localized to the dominant superior and middle temporal gyri. This area is modality-specific auditory association cortex and degrades relative late in this disease. 33 The relative preservation of this area in AD until very late stages may help explain the relative preservation of the letter-phoneme word fluency that is dependent on the frontal lobe and the phonological lexicon.
Although Glosser and colleagues 16,28 also found that AD was associated with relatively intact lexical-associative networks, our findings indicate something more. A somewhat unexpected finding was that the patients with AD demonstrated significantly lower average word frequency on the COWAT than did the controls. Our findings that the patients with AD used lower frequency words than did control participants suggest that at the same time as these patients with AD had decrease in spreading activation in their semantic networks, they had an increase in spreading activation in their lexical networks. The dynamic interplay between the lexical and the semantic networks may account for the greater spreading activation within the lexical network in AD. That is, degradation of the semantic system may consequently facilitate spreading activation within the lexical system in AD.
This increase in spreading activation cannot be explained by neuronal degeneration in the phonological lexicon, and thus some other mechanism must be inducing this change. Although the cause of this change is not known, there are several possibilities that will be briefly discussed.
John Hughlings Jackson, 38 one of the founders of modern neurology, wrote about the evolutionary organization of the nervous system. He posited that more highly evolved brain areas provide the human with a greater understanding of their environment and a greater repertoire of behaviors, but in order to function these higher areas have inhibit lower areas that program more stereotypic behaviors. In regard to the language-speech system, the semantic networks, found in polymodal cortex, is the higher system and semantics normally controls the lexical system. According to Hughlings Jackson, damage to a higher area induces both negative symptoms and a loss of function and positive symptoms caused by the release of the lower centers. Hughlings Jackson calls this process, “dissolution.” It is possible that the phenomena we observed, a decrement in the spreading activation of the semantic system and an enhancement or release of the phonological lexicon are examples of Hughlings Jackson principle of “dissolution.”
The neurophysiological basis for increased spreading activation in lexical networks may be due to the effects of acetylcholine. It is well known that AD is associated with reduced levels of acetylcholine, giving rise to the treatment of AD using AChEIs. The cholinergic system has been associated with not only memory but also attentional systems. 39 Administering donepezil, an AChEI, to healthy individuals has been found to increase sustained attention 40 and voluntary attention. 41 Klinkenberg et al 42 have proposed that in the regions of the brain important for attentional processing, including the prefrontal cortex, acetylcholine has a vital role in the top-down control of attention. Other investigators have also proposed that cholinergic innervations of the frontal cortex from the nucleus basalis of Meynert are involved in attentional processes. 43,44 Scopolamine, a muscarinic antagonist, is associated with reduced activity in the left superior and left middle frontal regions during attentional processes. 45 Additionally, the cholinesterase inhibitor physostigmine is associated with improved attention during an emotional task and increased activity in the dorsolateral and medial prefrontal cortices, among other brain regions. 46
Given the role of acetylcholine in enhancing attentional processes, the possibility exists that this effect of acetylcholine in producing an attentional spotlight has the effect of constricting lexical networks. Hence, given that AD is associated with the depletion of acetylcholine, the opposite effect may exist for patients with AD. Specifically, AD may be associated with more of an attentional floodlight and the expansion of lexical networks. This expansion in lexical networks would be consistent with our finding of increased spreading activation in lexical networks as evidenced by the significantly lower COWAT average word frequency for patients with AD. The reason this same effect might not exist for semantic networks may be due to the task demands and functional neuroanatomy of performance on the COWAT. As mentioned previously, acetylcholine innervates regions of the frontal cortex important for attentional processes. Research has also indicated that performance on the COWAT is sensitive to left frontal lobe functioning. 47 –49 However, semantic fluency tasks are associated with left temporal lobe functioning. 48,50 Thus, given the aforementioned research regarding the cholinergic innervations of the frontal cortex and the effects of acetylcholine in producing an attentional spotlight, it is consistent that AD would be associated with increased spreading activation in lexical networks but decreased spreading activation in semantic networks.
We recently investigated this potential role of acetylcholine in constricting lexical networks by examining the effects of AChEIs on spreading activation. 27 Specifically, we compared matched groups of patients with dementia who were or were not taking an AChEI in their average word frequency on the COWAT and AN test. We found that patients taking AChEIs exhibited significantly higher COWAT average word frequency, indicating significantly less spreading activation and a constriction of semantic networks. The sample for this study, though, was comprised of a mixed group of patients with dementia, including AD, vascular dementia, and Lewy body dementia. Further research should be conducted to extend the present findings and those of our previous investigation by examining the effects of AChEIs on spreading activation in patients with AD. As noted previously, AD is associated with category-specific deficits with reduced priming for artifacts but normal priming for animals. 24 Thus, future research could examine whether differences exist in these separate semantic networks by calculating word frequencies from semantic fluency tests that use living and nonliving categories. Differential effects of AChEIs on spreading activation within living versus nonliving networks could also be investigated in future research. Hopefully, the results of the present study will stimulate future research to address these additional questions.
Footnotes
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.