Procedural and Declarative Memory Brain Systems in Developmental Language Disorder (DLD)

Abstract

The aim of the current study was to examine microstructural differences in white matter relevant to procedural and declarative memory between adolescents/young adults with and without Developmental Language Disorder (DLD) using diffusion tensor imaging (DTI). The findings showed atypical age-related changes in white matter structures in the corticostriatal system, in the corticocerebellar system, and in the medial temporal region in individuals with DLD. Results highlight the importance of considering the age factor in research on DLD. Future studies are needed to examine the developmental relationship between long-term memory and individual differences in language development and learning.

1. Introduction

There is a long history of research attempting to characterize the relationship between language development and different forms of memory, with a presumption that language development is not dependent solely on processes specifically tied to language, but it also relies on domain-general cognitive processes, such as memory acquisition, processing, and retrieval (Adams & Gathercole, 2009; Baddeley, 2003; Kidd, 2012; Ullman & Pierpont, 2005; but see Chomsky, 2011; van der Lely, 2005). Developmental language disorder (DLD) is a neurodevelopmental disorder in language development and learning, which is not associated with any known condition, such as autistic spectrum disorders (ASD), intellectual disabilities, or congenital hearing impairment (Bishop, Snowling, Thompson, Greenhalgh, & CATALISE-2 consortium, 2017). Thus, it provides a valuable testing ground for hypotheses regarding the role of memory in language development and learning. Previous studies have consistently shown a role of procedural and declarative memory in DLD at the behavioral level (for a review, see Lum & Conti-Ramsden, 2013; Lum, Conti-Ramsden, Morgan, & Ullman, 2014; Ullman, Earle, Walenski, & Janacsek, 2020). In the current study, we aimed to look at whether there were brain structural differences in these two memory brain systems in individuals with and without DLD, and whether these differences varied across age.

1.1. Development of Long-term Memory Brain Systems

Procedural and declarative memory differ with regard to their supporting brain structures as well as the cognitive processes involved in memory and learning (Eichenbaum & Cohen, 2004; Gabrieli, 1998; Squire, 1992). We used the terms “procedural memory brain system” and “declarative memory brain system” to indicate that the focus of the current study was on the neural systems involved in learning, representation, and use of the procedural and declarative information respectively.

1.1.1. Procedural Memory Brain System.

The procedural memory system supports incremental and implicit learning of rule- or pattern-based relations via repetitive exposures (Cohen & Bacdayan, 1994; Packard & Knowlton, 2002). Learning that relies upon this system requires relatively few cognitive resources, and can even occur without having the intention to do so. Although the learning process is gradual and slow, rules/patterns can be applied quickly and automatically once they are acquired (Squire, 1992).

Results pertaining to development of procedural memory are mixed (see Zwart et al., 2019, for a review). Reber (1993) hypothesized that procedural learning should demonstrate early ontogenetic maturation because it mainly recruits evolutionarily primitive brain regions (e.g., the basal ganglia and the cerebellum). However, this hypothesis has been well challenged. For example, Thomas et al. (2004) showed age-related improvement in procedural learning on the serial reaction time (SRT) task, a commonly used paradigm to investigate procedural learning. Janacsek et al. (2012) further argued that the period between birth and adolescence is critical for development of procedural learning, and at around the age of 12, the cognitively controlled processes are more mature and thus are useful for targeted explicit learning at the cost of becoming less sensitive to probabilities of sequences/events during procedural learning. This suggests a developmental shift in the interaction between simple probability-based procedural learning and explicit higher-order learning in early adolescence (Nemeth, Janacsek, & Fiser, 2013).

From a neuroanatomical perspective, the procedural memory system is composed of a network of interconnected brain structures, including the corticostriatal and the corticocerebellar systems. The corticostriatal system is a collective term describing the reciprocal connections between the cortex and the basal ganglia, and some of the connections are via the thalamus (Alexander, DeLong, & Strick, 1986; Lehericy et al., 2004). The prominent white matter structures in the corticostriatal system include (but are not limited to) the internal capsule, which contains both ascending and descending axons between the cortex and the basal ganglia (Schmahmann & Pandya, 2006). The corticocerebellar system is comprised of several major white matter structures, including the superior and middle cerebellar peduncles (Habas & Cabanis, 2007; Leitner, Travis, Ben-Shachar, Yeom, & Feldman, 2015; Naidich et al., 2009). The superior cerebellar peduncle contains primarily efferent fibers that emerge from the deep cerebellar nuclei, traveling to the thalamus, and then to the cerebral cortex. The middle cerebellar peduncle is a major afferent projection that carries input fibers from the contralateral cerebral cortex via pontine nuclei across the midline of the cerebellum to the cerebellar cortex.

Recent DTI studies showed that the corticostriatal tracts and the corticocerebellar tracts reached maturation during adolescence/early adulthood, as revealed by increased fractional anisotropy (FA) with age (Asato et al., 2010; Lebel et al., 2012; Simmonds et al., 2014; Tamnes et al., 2010). In addition, growth continues into adulthood in regional termination zones (i.e., white matter regions adjacent to gray matter) in cortical and basal ganglia regions, suggesting a critical role of the cortical-subcortical network in cognitive development beyond adolescence (Simmonds et al., 2014).

1.1.2. Declarative Memory Brain System.

The declarative memory system supports acquisition and flexible use of context-free knowledge about the world (i.e., semantic memory) as well as context-specific information unique to an individual (i.e., episodic memory) (Tulving, 1972). In the classic definition, learning that relies on the declarative memory system typically requires conscious effort to encode and retrieve information, and therefore, in contrast with procedural memory, declarative memory can be directly measured by recognition or recall of previously learned stimuli (Eichenbaum & Cohen, 2004; Gabrielli, 1998; Squire, 1992; but see Dew & Cabeza, 2011; Henke, 2010).

From a neuroanatomical perspective, the declarative memory system is primarily supported by the medial temporal region (Eichenbaum & Lipton, 2008; Squire, Stark, & Clark, 2004). The hippocampus, located in the medial temporal lobes, plays a critical role in binding together different inputs to form representations of the relations among the constituent elements of scenes or events. Two white matter structures have been consistently shown to be associated with declarative memory functioning, including: 1) the cingulum bundle, which forms a ring from the orbitofrontal cortex, along the dorsal surface of the corpus callosum, then down the temporal lobe towards the pole while connecting the thalamic nuclei with the cingulate gyrus, and 2) the fornix, one of the major output structures of the hippocampus (Bubb, Metzler-Baddeley, & Aggleton, 2018; Sepulcre et al., 2008; Schmahmann & Pandya, 2006). DTI studies have shown a prolonged development of the cingulum bundle and the fornix (Lebel et al, 2012; Simmonds et al., 2014), which are consistent with behavioral findings showing continuous development of declarative memory until early adulthood (Ronnlund et al., 2005). Simmonds et al. (2014) also found a distinct phase of growth trajectory in the cingulate part of the cingulum bundle, with rapid growth in childhood, followed by a slowing of growth in early-middle adolescence, and acceleration of growth again in late adolescence/early adulthood.

1.2. Procedural and Declarative Memory Systems in DLD

While little is known about the fundamental processes underlying DLD, many attempts have been made to look at the role of memory in DLD (e.g., Archibald & Gathercole, 2006; Lum & Conti-Ramsden, 2013). One of the predominant hypotheses concerning the memory-language connection in DLD is the Procedural Deficit Hypothesis (PDH; Ullman & Pierpont, 2005). According to the PDH, individuals with DLD have an altered procedural memory system, resulting in impaired acquisition of rule- or pattern-based aspects of language (e.g., grammar or phonotactics) as well as impaired non-linguistic function mediated by the same brain system. In contrast, individuals with DLD are predicted to have an intact (or even strengthened) declarative memory system, which not only supports lexical-semantic aspects of language development and learning, but also compensates for poor grammatical learning by using explicit rules or chunking (Ullman & Pullman, 2015).

A growing body of behavioral research has been conducted to test and further specify the PDH (see Ullman et al., 2020, for a review). Evidence concerning the procedural memory system in DLD is relatively consistent, showing that individuals with DLD have poor procedural learning, and this deficit is associated with their grammatical ability (Hedenius et al., 2011; Lee & Tomblin, 2015; Tomblin, Mainela-Arnold, & Zhang, 2007; but see Gabriel et al., 2011). However, results are mixed with regard to the relationship between the declarative memory system and DLD. Individuals with DLD were reported to have poor declarative memory and learning in the verbal domain (Bishop & Hsu, 2015; Lukacs et al., 2017; Lum, Gelgic, & Conti-Ramsden, 2010). Lum et al. (2012, 2015) argued that this poor performance is indeed associated with comorbid poor working memory capacity. Disagreement also exists in terms of the nonverbal domain of the declarative memory system in DLD. Some researchers showed typical or even enhanced declarative memory for non-verbal information in DLD (Lukacs et al., 2017; Lum et al., 2010; Riccio, Cash, & Cohen, 2007), whereas others found that individuals with DLD have difficulty memorizing paired associates in nonverbal declarative learning tasks (Bishop & Hsu, 2015; Collisson et al., 2014; Lee, 2018; Poll, Miller, & van Hell, 2015). The discrepancy in previous studies may be attributed to different task-dependent mnemonic processes underlying different tests/tasks. Therefore, an alternative to using behavioral tasks to test these memory processes is to examine the brain structures that are known to subserve them.

To sum up, behavioral findings support the notion that both procedural and declarative memory systems may play a role in language learning difficulties of individuals with DLD. However, no studies have directly examined the procedural and declarative memory brain systems in DLD with a hypothesis driven approach (i.e., brain regions of interest were selected based on the PDH), although a few studies included substructures of the basal ganglia as ancillary analyses. For example, previous studies showed structural alterations in the caudate nucleus in individuals with DLD (Badcock et al., 2012; Soriano-Mas et al., 2009; but see Jernigan et al., 1991), but reports in Badcock et al. (2012) and Soriano-Mas et al. (2009) differed with respect to the direction of volumetric difference (i.e., smaller vs. larger volumes in the caudate nucleus in DLD). This difference may be, at least partially, attributed to the discrepancy in subject age, as well as imaging methods and analyses across studies (see Liegeois et al., 2014; Mayes et al., 2015, for a review). Lee, Nopoulos & Tomblin (2013) showed that young adults with DLD had larger relative volumes of the putamen, the nucleus accumbens, and the hippocampus than those without DLD; however, they did not find significant difference in the structures of the caudate nucleus after the intracranial volumes were controlled.

White matter structures pertaining to procedural and declarative memory were not yet examined in individuals with and without DLD. In addition, DLD is a neurodevelopmental disorder that persists across the lifespan, and thus the age factor needs to be taken into account in data interpretation. As far as we know, there is only one exploratory study in the literature examining age-related brain structural changes in children with DLD across a wide age range (Soriano-Mas et al., 2009).

1.3. The Current Study

The aim of the current study was to examine the differences in white matter structures relevant to procedural and declarative memory between adolescents/young adults with and without DLD. This wide age range provided an opportunity to examine whether group differences varied across age. We hypothesized that individuals with DLD would show a significant difference in white matter structures within the corticostriatal system as well as in the medial temporal region, which is reflective of poor functioning of procedural and declarative memory in DLD respectively. However, we did not expect to find a group difference in the corticocerebellar system given that previous behavioral studies showed intact cerebellar-dependent learning (i.e., eyeblink conditioning) in children and adolescents with DLD (Hardiman, Hsu, & Bishop, 2013; Steinmetz & Rice, 2010).

2. Material and Methods

2.1. Participants

Participants were recruited at two different time points, with the gap of two years. The recruitment methods were different with respect to subject sources (i.e., from longitudinal database vs. from school-wide screening at local public schools) and test batteries. However, the diagnostic criteria for each group (i.e., DLD vs. Comparison) was comparable between the first and the second recruitment (see Section 2.1.3 for statistical analyses). Table 1 summarizes the demographic information and test scores of the participants.

Table 1.

Demographics and test scores for the DLD and the Comparison group.

	DLD (n=14)		Comparison (n=12)
	M	SD	M	SD	P
Cohort 1
Age (year)	22.42	2.02	22.12	.53	.62
Sex (Male:Female)	5:9		4:8		.90
Handedness (Right:Left)	13:1		12:0		.35
Language Composite Scores	69.51	13.01	100.13	11.80	<.001
Nonverbal IQ	89.36	13.40	112.50	10.03	<.001
	DLD (n=17)		Comparison (n=51)
	M	SD	M	SD	P
Cohort 2
Age (year)	16.98	1.54	16.77	1.45	.61
Sex (Male:Female)	7:10		20:31		.89
Handedness (Right:Left)	11:6		47:4		.01
Language Composite Scores	78.67	5.27	102.99	9.96	<.001
Nonverbal IQ	85.82	8.67	101.88	15.26	<.001

2.1.1. Recruitment for Cohort 1.

Participants were recruited from a longitudinal study (see Tomblin & Nippold, 2014, for details of the longitudinal study). Their language ability at the time of their participation was based on three language tests: 1) Word Derivations, a subtest from The Test of Adolescent and Adult Language, Fourth Edition (TOAL-4; Hammill, Brown, Larsen, & Wiederholt, 2007) to assess knowledge of derivational morphology, 2) Peabody Picture Vocabulary Test, Fourth Edition (PPVT-4; Dunn & Dunn, 2007) to assess receptive vocabulary, and 3) a modified version of the Token Test (de Renzi & Faglioni, 1978; Morice & McNicol, 1985) to assess sentence comprehension. The diagnosis of language disorders was reconfirmed if participants with a history of DLD had language composite scores below one standard deviation below the mean. All participants were required to have nonverbal IQ above 70, as measured on two nonverbal subtests (i.e., Block Design and Matrix Reasoning) of Wechsler Abbreviated Scale of Intelligence (WASI, Wechsler, 1999).

2.1.2. Recruitment for Cohort 2.

Participants were recruited from high schools and colleges in the Midwest with the age range of 14 to 20 years. None of them had a diagnosis of neurodevelopmental disorders (e.g., intellectual disabilities, ASD) or hearing loss. This recruitment method was different from the one used in first recruitment. The sample of adolescents was obtained via a school-wide screening of language ability. Adolescents were then drawn from the screened pool to provide a sample that had a roughly rectangular distribution, with oversampling of adolescents in the high and low ends of the normal distribution with regard to language. Those selected were then given a battery of language tests, including: 1) PPVT-4 (Dunn & Dunn, 2007), 2) Expressive Vocabulary Test, Second Edition (EVT-2; Williams, 1997) to assess expressive vocabulary, 3) Recalling Sentences in Clinical Evaluation of Language Fundamentals, Fourth Edition (CELF-4; Semel et al., 2003) to assess the ability to recall and reproduce sentences of syntactic complexity, 4) Understanding Spoken Paragraphs in CELF-4 to assess passage comprehension, and 5) Word Derivations in TOAL-4 (Hammill et al., 2007). All participants had nonverbal IQ above 70 on Block Design and Matrix Reasoning from WASI (Wechsler, 1999). The language tests used in the second recruitment were carefully selected based on Tomblin, Freese, and Records (1992) and Fidler, Plante, and Vance (2011), with the aim to assess multiple domains of language in adolescents and young adults.

2.1.3. Establishing Comparability of Language Measures Used in Two Cohorts.

Measurement of language ability at different points in development often requires use of different measures. Nevertheless, it has been shown that oral language in childhood and adolescence consists largely of a uni-dimensional latent trait (Anthony, Davis, Williams, & Anthony, 2014; Bornstein, Hahn, Putnick, & Suwalsky, 2013; Language and Reading Research Consortium, 2015; Tomblin & Zhang, 2006). In this study, there were two tests in common across the cohorts (i.e., PPVT and Word Derivations in TOAL-4) along with the tests that were unique in each cohort. Results showed that the common measures were strongly correlated with the unique measures employed for Cohort 1, r=.60, p=.001, and for Cohort 2, r=.81, p<.001. The Fisher’s r-to-z transformation showed that these two correlations were not significantly different, z=−1.9, p=.06, suggesting that language measures used in different cohorts indeed represent a similar latent language ability trait, and thus the diagnostic label for each group (i.e., DLD vs. Comparison) is comparable between the first and the second recruitment. The standard scores based on the test norms were averaged for each subject and used to assign individuals to groups. Individuals with DLD were required to have average language scores below one standard deviation below the mean. Those with scores equal to or greater than one standard deviation below the mean were assigned to the Comparison group.

2.2. MRI Acquisition

All image scans were obtained at the University of Iowa Hospital and Clinics using the Siemens 3T Trio scanner.

2.2.1. Anatomical Image.

The T1 weighted images were acquired in the coronal plane using a 3D MP-RAGE sequence with the following parameters: echo time (TE)=2.86 ms, repetition time (TR)=2300 ms, inversion time=900 ms, flip angle=10°, number of excitations (NEX)=1, field of view (FOV)=256×256×256 mm, and slice thickness=1 mm. The PD/T2 weighted images were acquired with the following parameters: TE=430 ms, TR=4800 ms, NEX=1, slice thickness=1.4 mm, FOV=256×256×256 mm, and turbo factor=137.

2.2.2. Diffusion Tensor Image.

The diffusion weighted images were collected using a 2D twice refocused echo-planar spin-echo sequence with the following parameters: TE=81 ms, TR=9000 ms, FOV=256×256 mm, Matrix=128×128, slice thickness=2.0 mm, the number of diffusion directions=64, and b-values=1000 s/mm².

2.3. Image Preprocessing and Processing

2.3.1. Anatomical Image Processing.

The anatomical data were preprocessed using AutoWorkup (Magnotta et al., 2002). The primary steps of the standard pipeline include: 1) anterior commissure (AC)-posterior commissure (PC) alignment of T1 volume, 2) co-registration of T2 weighted images to AC-PC aligned T1, defining of Talairach parameters to warp the Talairach grid onto the raw space of each subject, 3) tissue classification of white matter, gray matter, and cerebrospinal fluid, and 4) skull stripping using an artificial neural network (Magnotta et al., 1999). After completion of AutoWorkup, all scans were individually inspected for quality control.

2.3.2. Diffusion Tensor Image Processing.

The DTIPrep (Liu et al., 2010) was used to perform several quality assurance steps as well as removing volumes within a scan that did not meet its minimal quality criteria before tensor image estimation. The automatic pipeline in the software includes: 1) protocol verification, 2) slice-wise checking, and 3) detection and removal of artifacts caused by eddy-currents, head motion, bed vibration and pulsation, venetian blind artifacts, as well as intensity inconsistencies (Oguz et al., 2014). The final dataset contained an averaged baseline image and diffusion-weighted images that passed all the quality assurance tests. Next, we used the Guided Tensor Restore Anatomical Connectivity Tractography (GTRACT; Cheng et al., 2006) software to process the output files from the DTIPrep, including estimation of tensor images and subsequent computation of fractional anisotropy (FA) images. Johns Hopkins University (JHU) white matter tractography atlas (Hua et al., 2008; Wakana et al., 2007) was then used to label white matter structures of interest, as shown below.

1. Major white matter structures in the corticostriatal system (see Figure 1): bilateral anterior limbs of internal capsule (ALIC), and bilateral posterior limbs of internal capsule (PLIC)

2. Major white matter structures in the corticocerebellar system (see Figure 2): bilateral superior cerebellar peduncle (SCP), and the middle cerebellar peduncle (MCP)

3. Major white matter structures in the medial temporal region (see Figure 3): bilateral fornix, bilateral cingulate gyrus part of the cingulum bundle (CGC), and bilateral hippocampal part of the cingulum bundle (CGH)

Figure 1.

JHU white matter tractography atlas of regions of interest in the corticostriatal system. ALIC: Anterior limb of internal capsule; PLIC: Posterior limb of internal capsule; RLIC.

Figure 2.

JHU white matter tractography atlas of regions of interest in the corticocerebellar system. CP: Cerebellar peduncle.

Figure 3.

JHU white matter tractography atlas of regions of interest in the medial temporal region. CGC: Cingulum_cingulate gyrus; CGH: Cingulum_hippocampus.

2.4. Statistical Analysis

FA was used as the primary diffusion measure in the analyses; it is one of the primary indices of white matter microstructural integrity (Alexander et al., 2007). Recent studies have consistently demonstrated developmental increases in FA from childhood into early adulthood (e.g., Barnea-Goraly et al., 2005; Schmithorst et al., 2002).

An analysis of covariance (ANCOVA) within the general linear model was used to compute the effects of Age, Group, and Age-by-Group interactions on FAs in the white matter structures of interest, with sex, handedness, and nonverbal IQ entered as covariates. All statistical analyses were carried out using SPSS version 25. In addition, we used false discovery rate (FDR) to correct for multiple comparisons in each hypothesis testing (Benjamini & Hochberg, 1995). The p-values in each table (i.e., Table 3, Table 4, and Table 5 respectively) defined the domain of the correction.

Table 3.

Fractional anisotropy (FA) of the white matter structures in the corticostriatal system. Multivariate analyses were carried out, with sex, handedness, and nonverbal IQ as covariates. An asterisk indicates that the p-value remains significant after false discovery rate (FDR) adjustment. η_p²: partial eta-squared; L: Left; R: Right.

			F	P	ηp2
Anterior Limb of Internal Capsule	R	Age	5.31	.02*	.06
		Group	10.11	.002*	.10
		Age × Group	3.49	.07	.04
	L	Age	5.05	.027*	.06
		Group	8.52	.004*	.09
		Age × Group	8.00	.006*	.08
Posterior Limb of Internal Capsule	R	Age	7.02	.01*	.08
		Group	5.26	.02*	.06
		Age × Group	2.95	.09	.03
	L	Age	7.33	.008*	.08
		Group	2.59	.11	.03
		Age × Group	6.46	.01*	.07

Table 4.

Fractional anisotropy (FA) of the white matter structures in the corticocerebellar system. Multivariate analyses were carried out, with sex, handedness, and nonverbal IQ as covariates. An asterisk indicates that the p-value remains significant after false discovery rate (FDR) adjustment. η_p²: partial eta-squared; L: Left; R: Right.

			F	P	ηp2
Middle Cerebellar Peduncle		Age	.68	.41	.01
		Group	7.18	.009*	.08
		Age × Group	6.91	.01*	.07
Superior Cerebellar Peduncle	R	Age	.20	.66	< .01
		Group	10.91	.001*	.11
		Age × Group	12.52	.001*	.13
	L	Age	.96	.33	.01
		Group	7.49	.008*	.08
		Age × Group	6.01	.016*	.07

Table 5.

Fractional anisotropy (FA) of the fiber tracts in the medial temporal region. Multivariate analyses were carried out, with sex, handedness, and nonverbal IQ as covariates. An asterisk indicates that the p-value remains significant after false discovery rate (FDR) adjustment. η_p²: partial eta-squared; L: Left; R: Right.

			F	P	ηp2
Fornix	R	Age	2.88	.09	.03
		Group	13.89	<.001	.14
		Age × Group	5.76	.02*	.06
	L	Age	4.08	.05	.05
		Group	11.00	.001*	.11
		Age × Group	6.28	.01*	.07
Cingulate gyrus part of the cingulum bundle	R	Age	13.49	<.001*	.13
		Group	4.77	.03	.05
		Age × Group	3.35	.07	.04
	L	Age	2.57	.11	.03
		Group	2.93	.09	.03
		Age × Group	2.03	.16	.02
Hippocampal part of the cingulum bundle	R	Age	8.76	.004*	.09
		Group	6.62	.01*	.07
		Age × Group	3.40	.07	.04
	L	Age	8.06	.006*	.09
		Group	4.22	.04	.05
		Age × Group	1.08	.30	.01

3. Results

The mean and standard deviations of the FA values in the brain structures of interest as well as in the control structures for each group were summarized in Table 2. Levene’s test showed that variances in the dependent variables were not significantly different (.78 > ps > .054), except for the MCP, F(1, 92)=5.59, p=.02.

Table 2.

Summary of fractional anisotropy (FA) values in the fiber tracts of interest as well as in the control tracts. DLD: Developmental language disorder; L: Left; R: Right; SD: Standard deviation.

Fiber Tracts		Group	Mean	SD
Corticostriatal System
Anterior limb of internal capsule	R	Comparison	.39	.03
		DLD	.37	.03
	L	Comparison	.39	.03
		DLD	.37	.03
Posterior limb of internal capsule	R	Comparison	.53	.03
		DLD	.52	.03
	L	Comparison	.54	.03
		DLD	.53	.03
Corticocerebellar System
Superior cerebellar peduncle	R	Comparison	.34	.03
		DLD	.32	.03
	L	Comparison	.35	.03
		DLD	.32	.03
Middle cerebellar peduncle		Comparison	.36	.02
		DLD	.35	.03
Medial Temporal Region
Fornix	R	Comparison	.34	.02
		DLD	.32	.02
	L	Comparison	.36	.02
		DLD	.34	.02
Cingulate gyrus part of the cingulum bundle	R	Comparison	.30	.05
		DLD	.28	.05
	L	Comparison	.32	.05
		DLD	.30	.04
Hippocampal part of the cingulum bundle	R	Comparison	.22	.03
		DLD	.20	.03
	L	Comparison	.22	.03
		DLD	.21	.03

3.1. Corticostriatal System in DLD across Age

The results of ANOVAs concerned with the white matter structures in the corticostriatal system were shown in Table 3. Significant main effects involving Age and Group without significant interactions were found in two structures of interest, including the right ALIC and the right PLIC. In these cases, the DLD group had significantly lower FA values than the Comparison group, and the Age effect reflected an increase in FA with age.

In addition, a significant interaction between Age and Group was found in both brain structures of interest (i.e., ALIC and PLIC), particularly in the left hemisphere. These findings suggest atypical age-related microstructural changes in the corticostriatal system in individuals with DLD.

Follow-up tests on these significant interactions showed that across the brain structures of interest, the same pattern was evident: comparison participants showed gradually increased FA across age in the left ALIC, r=.45, p<.001, and in the left PLIC, r=.45, p<.001, whereas individuals with DLD did not show a significantly increased FA across age in the left ALIC, r=−.13, p=.51, or in the left PLIC, r=−.05, p=.78. The left ALIC was used as an example to illustrate the interaction pattern (see Figure 4).

Figure 4.

Age-by-group interaction effect in the left anterior limb of the internal capsule (ALIC). FA: Fractional anisotropy.

3.2. Corticocerebellar System in DLD across Age

Table 4 summarizes the ANOVA results performed on the measures of the white matter structures in the corticocerebellar system. The Age-by-Group interaction effect was significant in the MCP and in the bilateral SCP. We used the right SCP as an example to show different developmental trajectories of FA changes between the DLD and the Comparison group (see Figure 5). In Figure 5, comparison participants showed gradually increased FA in the right SCP across age, r=.36, p=.004, whereas individuals with DLD showed a significant decrease in FA across age, r=−.43, p=.02. These findings suggest atypical age-related microstructural changes in the corticocerebellar system in individuals with DLD.

Figure 5.

Age-by-group interaction effect in the left superior cerebellar peduncle (SCP). FA: Fractional anisotropy.

3.3. White Matter Structures in the Medial Temporal Region in DLD across Age

Table 5 provides the results of the ANOVA for the brain structures in the medial temporal region. The Age-by-Group interaction effect was only significant in the right and left fornix. Age-related change in the left fornix in both groups was plotted in Figure 6, showing that FA increased with age for the Comparison group, r=.44, p<.001, whereas this was not evident in the DLD group, r=−.10, p=.59. In terms of the cingulum bundle, while the Age-by-Group interaction was not significant, the Group main effect was found in the right CGH. The DLD group showed significantly lower FA in the right CGH than the Comparison group, suggesting poor white matter microstructure in the right CGH independent of age.

Figure 6.

Age-by-group interaction effect in the left fornix. FA: Fractional anisotropy.

4. Discussion

The aim of the current study was to examine differences in white matter microstructures relevant to procedural and declarative memory in individuals with and without DLD. The findings showed atypical age-related changes in white matter structures in the corticostriatal system, in the corticocerebellar system, and in the medial temporal region in individuals with DLD. Specifically, FA increased across age in the comparison participants with normal language abilities, but did not increase in individuals with DLD.

4.1. Corticostriatal System in DLD

The corticostriatal system is one of the major neural networks supporting procedural learning. Within this neural network, the basal ganglia modulate cognition and behavior by gating or selecting a subset of representations or movements activated by the cortex (Seger, 2006). It has been well established that the corticostriatal system mediates a wide variety of learning behaviors, including habit learning, motor skill learning, sequential learning, categorization learning, and reinforcement learning (see Koziol & Budding, 2009, for a review). In addition, growing studies have shown a critical role of the corticostriatal system in different domains of language learning and processing, syntax and phonology in particular (Chan, Ryan, & Bever, 2013; Crosson et al., 2003; Dominey, Inui, & Hoen, 2009; Tettamanti et al., 2005; see Bohsali & Crosson, 2016, for a review).

Previous studies showed that most corticostriatal white matter tracts reach 90% of the maximum FA values during adolescence/early adulthood (Barnea-Goraly et al., 2005; Lebel et al., 2012). In the current study, we found atypical age-related microstructural changes in the white matter structures in the corticostriatal system in individuals with DLD. That is, individuals with DLD did not have a significant increase in FA in the corticostriatal white matter structures across time, whereas comparison participants did. Animal and human studies have shown that maturation of corticostriatal connectivity supports developmental change in goal-directed as well as value-guided behavior (Somerville & Casey, 2010). Lee (2017) showed that adolescents/young adults with DLD had difficulty coding the motivational value of prospective incentives, and thus were not able to choose the most optimal actions in order to maximize future outcomes. Thus, future studies are suggested to conduct seed-based tractography analysis to examine white matter integrity of the executive and motivational corticostriatal loops in DLD (Seger, 2006; Lawrence et al., 1998; see Alexander et al., 1986, for five corticostriatal loops), which will provide information on the specificity of preserved and/or altered behavioral functioning in individuals with DLD and thus provide new ideas for intervention.

4.2. Corticocerebellar System in DLD

The corticocerebellar system is the other major neural network supporting procedural learning. Unlike the basal ganglia, which primarily involve selective gating of cortical representations, the cerebellum engages in online amplification and refinement of behaviors, serving as an error detection mechanism to detect deviations of timing or sequential events so that behavior can be adjusted to cope with environmental changes (Ide & Li, 2011). It is also proposed that the cerebellum involves updating of predictions about action consequences via error-driven learning (Koziol et al., 2014). Recent studies showed that this domain-general mechanism mediated a variety of language processing and learning (Timmann & Daum, 2007; Mariën et al., 2014). For example, Moberget et al. (2014) found an association between increased cerebellar activations and processing of predictive sentences (i.e., the initial part of a sentence gives clues to what will be at the end of the sentence), implicating a generalized role of the cerebellum in processing of prediction error signals. In addition, the corticocerebellar system is also shown to engage in grammatical, semantic, and phonological processing (Adamaszek & Kirkby, 2015; Stoodley & Schmahmann, 2009). Despite mounting evidence, the exact role of the corticocerebellar system in language remains to be elucidated.

Cerebellar white matter reaches maturation during adolescence/young adulthood (Lebel et al., 2012; Simmonds et al., 2014). In the current study, we showed that individuals with DLD had structural alteration in white matter structures in the corticocerebellar system, and this alteration varied across age. If structural alterations at the brain level are likely to be reflected on the behavioral level, the current data seem contrary to prior behavioral findings showing typical learning in eyeblink conditioning in children and adolescents with DLD (Hardiman et al., 2013; Steinmetz & Rice, 2010; see Ullman et al., 2020, for a review). This discrepancy may be attributed to the age factor. The age range in our current sample is between 14 to 28 years, which is older than that in Hardiman et al. (2013; age: 7–11 years) and Steinmetz and Rice (2010; age: 9–20 years). In other words, it is possible that behavioral differences in eyeblink conditioning between individuals with and with DLD cannot be sensitively detected until later years.

The other possible explanation to the discrepancy between our brain imaging finding and prior behavioral finding may have to do with the degree of task demands placed on the cerebellum as well as its connection with the cortex. Eyeblink conditioning is a primitive learning process, which places relatively less demand on the cerebral cortex. Thus, it is likely that information processing within the cerebellar macrostructure is spared in individuals with DLD, which is then able to support eyeblink conditioning behavior. However, if we use a learning paradigm that strongly involves the reciprocal connection between the cerebellum and the cerebral cortex, we might expect to see impaired learning in DLD.

4.3. Medial Temporal Region in DLD

The medial temporal region comprises a neural system of anatomically related structures that is necessary for declarative memory (Squire et al., 2004). Recently, researchers have attempted to make a connection between declarative memory and language. For example, Ullman (2001, 2004) proposed the Declarative/Procedural model, wherein the division between grammar and vocabulary in language is closely tied to the division between procedural and declarative memory. Several studies have shown that the mental lexicon uses the brain systems that also subserve the acquisition of episodic information in healthy individuals (e.g., Bartha-Doering et al., 2018; Breitenstein et al., 2005; Davis & Gaskell, 2009).

Duff and her colleagues extend the contribution of declarative memory to language even further, proposing that the same processes underlying the declarative memory system, such as relational binding of items into integrated representations as well as flexible use and online maintenance of those representations, are indeed the same processes essential for various aspects of language processing and use, in addition to word learning (Duff & Brown-Schmidt, 2012). By studying patients with hippocampal amnesia, they showed that the hippocampal declarative memory system involves language processing and use beyond the lexical domain, including reported speech (Duff et al., 2007), verbal play in social interactions (Duff et al., 2009), and the use of definite references during online processing (Duff et al., 2011). Duff’s work mainly focused on the role of the hippocampus in language in adults with brain lesions, thus it remains unclear with respect to the associations between other structures of the medial temporal region and language development and learning.

In the current study, individuals with DLD showed atypical age-related changes in the white matter microstructure of the fornix. The fornix is the major output structure of the hippocampus, arcing around the thalamus and connecting the medial temporal lobes to the hypothalamus (Thomas, Koumellis, & Dineen, 2011). Integrity of this white matter structure plays an important role in the formation and consolidation of declarative memory (Mabbott et al., 2009; Sepulcre et al., 2008). Therefore, our findings suggest that individuals with DLD may also have declarative memory deficits, and this poor functioning may be more easily detected during adolescence/young adulthood. In the literature, results are mixed regarding declarative memory functioning in DLD. Some researchers showed impaired declarative memory for verbal and nonverbal information in DLD, whereas others found no evidence of such mnemonic difficulty (Bishop & Hsu, 2015; Kuppuraj, Rao, & Bishop, 2016; Lum et al., 2012; Poll et al., 2015; Records et al., 1995). This discrepancy may lie in different task-dependent processes that rely on distinct neural circuits in the medial temporal regions. Moreover, the age factor may also, at least in part, account for different results in different studies.

At this point, we are not able to falsify the declarative memory compensation hypothesis, an extension of the PDH stating that declarative memory is relatively spared and therefore is able to compensate for procedural memory and other dysfunctions in DLD (Ullman & Pierpont, 2005). A better understanding of declarative memory is important in yielding new ways to treat people with neurodevelopmental disorders (Ullman & Pullman, 2015). Future work is needed to examine how declarative memory-related brain regions play a role in functional compensation when the procedural memory brain system is impaired along development in DLD.

4.4. Domain-General Cognitive Processes and Individual Differences in Language

DLD is a label given to individuals who have poor language ability with unknown causes, and this label is created based on an arbitrary cutoff language score. However, if we treat language as a continuous variable, DLD simply represents the lower end of a normal distribution with regard to language ability, instead of a qualitatively distinct clinical group (Dollaghan, 2011; Leonard, 2014). And thus, findings of DLD studies not only provide clinical implications for language intervention, but they also inform us the neurobiological mechanisms underlying individual differences in language development and learning (Tomblin & Christensen, 2009).

The corticostriatal system, the corticocerebellar system, and the medial temporal region are not dedicated to language behaviors only, as they are also involved in memory function. Our current results suggest that language development and learning is associated with domain-general cognitive processes; however, it remains unclear about the causality. An intriguing question raised is that the DLD and the Comparison group do not appear to differ in the diffusivity measure at younger age. In other words, the discrepancy in language ability between individuals with and without DLD continues to exist across age, whereas the difference in the microstructure of the brain region of interest gradually emerges. One explanation is that structural alterations in the memory brain systems are driven by poor language ability along development. It is also possible that language and cognitive deficits in DLD are comorbid, developing in parallel. Another explanation is that the relationship between language and memory follows a nonlinear developmental trajectory. Thus, our current findings simply reflect the brain-behavior relationship in the tail of the developmental process. These hypotheses await testing, results of which will shed light on the role of domain-general cognitive ability in language development and learning.

5. Conclusion

In sum, individuals with DLD showed atypical age-related changes in the microstructure of white matter in the corticostriatal system, the corticocerebellar system, and the medial temporal region. These findings support and further specify the PDH, providing preliminary biological evidence of an impaired procedural memory system in DLD. At the same time, questions remain concerning the effect of altered white matter microstructure of the fornix on declarative memory functioning in DLD.

Despite significant findings, the study has several limitations. First, we used JHU brain atlas to label the main regions of interest. Thus, we were not able to specifically delineate the connections between different parts of the cortex and different parts of the basal ganglia. Seed-based tractography will be a necessary next step to parcellate different corticostriatal loops. Second, this is a cross-sectional study, and thus caution needs to be taken when interpreting the results pertaining to the trajectory of brain development. Longitudinal designs will be more appropriate to directly address this issue. Third, it is unclear whether these atypical FA findings could be accounted for by system-wide poor white matter integrity in the brain. Exploratory whole-brain analyses are needed to identify brain structures that are relatively intact and thus can serve as control regions for comparison. Last, no behavioral measures on procedural and declarative memory were included in the current study, and thus we were not able to further explore the brain-behavior association in DLD. Future studies are needed to examine the developmental relationship between memory performance and individual differences in language.

Highlights.

• Atypical age-related change in WM structures in corticostriatal system in DLD

• Atypical age-related change in WM structures in corticocerebellar system in DLD

• Atypical age-related change in WM structures in medial temporal region in DLD

• Importance of considering the age factor in research on DLD