Mesial Temporal Lobe and Memory Function in Autism Spectrum Disorder: An Exploration of Volumetric Findings

Abstract

Studies have shown that individuals with autism spectrum disorder (ASD) tend to perform significantly below typical developing individuals on standardized measures of memory, even when not significantly different on measures of IQ. The current study sought to examine within ASD whether anatomical correlates of memory performance differed between those with average-to-above-average IQ (AIQ Group) compared to those with low average to borderline ability (LIQ group) as well as in relations to typically-developing comparisons (TDC). Using automated volumetric analyses, we examined regional volume of classic memory areas including the hippocampus, parahippocampal gyrus, entorhinal cortex, and amygdala in an all-male sample AIQ (n = 38) and LIQ (n = 18) individuals with ASD along with 30 typically-developing comparisons (TDC). Memory performance was assessed using the Test of Memory and Learning (TOMAL) compared among the groups and then correlated with regional brain volumes. Analyses revealed group differences on almost all facets of memory and learning as assessed by the various subtests of the TOMAL. The three groups did not differ on any ROI memory-related brain volumes. However, significant size-memory function interactions were observed. Negative correlations were found between the volume of the amygdala and composite, verbal, and delayed memory indices for the LIQ ASD group indicating larger volume related to poorer performance. Implications for general memory functioning and dysfunctional neural connectivity in ASD are discussed.

Lower performance on memory tasks, particularly episodic memory, has been consistently observed in autism spectrum disorders (ASD) when compared with typically developing individuals (Ben Shalom, 2003; Hill & Frith, 2003; Boucher & Bowler, 2008; Narzisi, Muratori, Calderoni, Fabbro, & Urgesi, 2013; Russell, Jarrold, & Henry, 1996; Southwick et al., 2011). Yet, areas of preserved memory processing have also been found (e.g. Assouline, Foley Nicpon, & Dockery, 2012; Boucher & Lewis, 1989; Boucher & Warrington, 1976; O’Shea, Fein, Cillessen, Klin, & Schultz, 2005) including exceptional memory skills (Bennett & Heaton, 2012; Mottron, Belleville, Stip, & Morasse, 1998). Southwick et al. (2011) used a comprehensive test battery, the Test of Memory and Learning (TOMAL; Reynolds and Bigler, 1994), to broadly assess memory performance in ASD and found that the individuals with ASD performed significantly below age- and education-matched typically-developing comparisons (TDC), 5–19 years of age, across all domains of memory function.

The Southwick et al. (2011) study did not examine whether there were any neuroanatomical differences between the ASD and TDC groups. Is part of the explanation for differences in memory related to any size-function differences between the ASD and TDC groups? Since well-known mesial temporal lobe structures participate in memory, could a difference in their size relate to memory function? Size-function relations tend to be positive and their evolutionary basis is well documented in the mammalian brain (Koscik & Tranel, 2012). Volume of a given structure or region of interest (ROI) has been a traditional measure of size where developmentally, neuroimaging based volumetric methods have long been used as a proxy of brain health and growth (Brown & Jernigan, 2012; Stiles & Jernigan, 2010), including volumetric analyses of developing mesial temporal lobe structures critical for memory like the hippocampus (Hu, Pruessner, Coupe, & Collins, 2013). Nonetheless, the issues are complex because with aberrant development other regions may be recruited to participate in a given function associated with more typical development. For example, Brunnemann et al. (2013) found smaller hippocampal volume in preterm children with uncomplicated neonatal courses (<34 weeks of gestation, birth weight <2,000 g) compared to controls (7–11 years). However, only in the controls did hippocampal volume positively relate to episodic memory function based on neuropsychological testing (see also Omizzolo et al., 2013). Whether impaired memory functioning in ASD compared with TDC is associated with any gross volumetric differences in brain regions known to participate in memory is the focus of the current investigation.

Neuroanatomy and Memory Functioning

Key anatomical regions associated with memory and underlying neural networks have been well-established and investigated for decades, including critical mesial temporal lobe structures like the parahippocampal, entorhinal and perirhinal cortices and the hippocampus (Aggleton et al., 2010; Clark & Boutros, 1999; Miller, Li, & Desimone, 1991). The amygdala, a subcortical mesial temporal lobe structure, also plays an important role, particularly in learning contingencies between internal and external cues of emotion (Aggleton, 1993; Aggleton et al., 2010; Clark & Boutros, 1999) including social stimuli, processing of which may be more aberrant in autism (Barnea-Goraly et al., 2013; Green et al., 2013). All of the aforementioned brain structures critical for memory have some overlap with networks important for social-emotional processing and therein may be an important relation between ASD and impaired memory performance (Boucher, Mayes & Bigham, 2012).

Cognitive Deficits Across the Autism Spectrum

Higher frequency of intellectual disability is associated with ASD when compared to typically developing individuals (Charman et al., 2011). Additionally, even when individuals with ASD without intellectual disability are demographically matched to neurotypical comparisons, individuals with ASD generally score significantly lower on measures of IQ (Jou, Frazier, Keshavan, Minshew, & Hardan, 2013). Intellectual ability constitutes an important issue in the study of memory functioning in those with ASD because of the important relation between memory and IQ (see Lezak, Howieson, Bigler, & Tranel, 2012). However, in a study by Southwick et al. (2011), even when statistically comparable on IQ, individuals with ASD displayed significantly lower memory performance across most domains in comparison to TDC.

Since IQ also relates to brain morphology (Eliez, Blasey, Freund, Hastie, & Reiss, 2001; Freitag et al., 2009; Haier, Jung, Yeo, Head, & Alkire, 2004; Lange, Froimowitz, Bigler, & Lainhart, 2010; Reiss, Abrams, Singer, Ross, & Denckla, 1996), examination of neuroanatomical correlates of memory performance would require controlling in some fashion for intellectual level of the participants being examined. The need to control for IQ would also apply to brain structures associated with memory, where lower IQ may be uniquely associated with morphology differences. Boucher et al. (2012) provided an extensive review of memory in ASD and discussed the potential differences in memory function that may occur along the dimension of general cognitive ability as well as brain regions associated with memory that may differ in ASD. Operationally, they differentiated ASD in terms of level of functioning, with high functioning defined as “individuals with ASD and intellectual and linguistic abilities within the normal range, regardless of whether language was initially delayed” (p. 460). In their schema, individuals with verbal IQ scores of 85 and above would be categorized as high functioning. Boucher et al. (2012) discussed lower functioning autism with several qualifiers depending on borderline or below intellectual ability scores and degree of language impairment. By categorizing along the dimension of intellectual ability, Boucher et al. (2012) showed differences in memory performance associated with ASD.

For example, Boucher et al. (2012) summarized a number of studies contrasting high functioning versus lower functioning ASD groups on norm-referenced memory tasks. Unimpaired non-declarative memory in the high functioning ASD groups (14/15 findings) as well as intact declarative memory when tested by recognition (29/35 findings) was observed. However, with free recall paradigms, the declarative memory of high functioning individuals with ASD was often, yet inconsistently, found to be impaired (21/42 findings). Additionally, in the lower functioning ASD groups, declarative memory was impaired regardless of whether assessed by recognition (10/14 findings) or by free recall (11/18 findings). Both groups generally had intact cued recall (15/18 findings for high functioning and 11/12 findings for lower functioning individuals). Studies on source memory were mixed, with both impaired and unimpaired findings for both higher (3/5 impaired findings) and lower functioning groups (3/5 impaired findings). The Boucher et al. review documented the variability in memory performance associated with ASD along with the importance of examining memory function across the dimension of intellectual ability.

Working Memory in ASD

The Boucher et al. (2012) review also highlighted particular features of memory function that may be central to ASD. For example, working memory may be particularly interesting in relation to ASD inasmuch as Barendse et al. (2013) argued that working memory deficits are central to symptoms of ASD. Working memory operates during processing of complex information and executive functions, including during social cognition and interpersonal interactions, all areas of impairment in ASD (Barendse et al., 2013). Boucher et al. (2012) showed mixed results with working memory, with some investigations finding impairment in both higher and lower functioning groups (11/26 and 8/15 impaired findings respectively). The Boucher et al. review found relatively intact performance for both groups during working memory tasks that simply keep basic information online for further processing. However, some studies found that on tasks that require executive control in order to manipulate and process the online information (e.g., digits backward), individuals with ASD exhibited greater deficits with working memory, regardless of level of intellectual ability.

Size- Function Relations in Memory Impairments in ASD

Similar to Barendse et al. (2013), Minshew and Goldstein (1998) have suggested that the pattern of memory performance observed in ASD relates to disordered information processing of complex information, leaving simple information processing intact. In other words, individuals with autism have greater impairment on tasks that require organization, integration, or a high processing load when compared to individuals with neurotypical development (Minshew & Goldstein, 1998). From this perspective, the impairment in information processing in ASD may reflect neural disconnectivity of a “generalized dysfunction of the association cortex, with sparing of the primary sensory and motor cortex” (Minshew & Williams, 2007, p. 946). Boucher et al. (2012) noted that impaired information processing in autism is consistent with models that view autism as a disconnection syndrome, which would affect memory functioning (Belmonte et al., 2004; Courchesne, 2004; Rippon, Brock, Brown, & Boucher, 2007). Indeed, Casanova and Trippe (2009) and Casanova (2007) have shown that increases in cortical gray matter may relate to aberrant increased white matter projections necessary to maintain the connectivity of the increased number of cortical cells in the brain in ASD. Lewis, Theilman, Townsend and Evans (2013) argue that larger intracranial size in ASD is associated with diminished network efficiency. The size-function relationships posited by these investigators suggests that it may be possible to use volumetric measurements as a proxy for assessing brain connectivity.

Current Study

The current investigation is an extension of the study by Southwick et al. (2011) and explores the anatomical correlates of TOMAL-assessed memory function in ASD as well as potential relations when the ASD sample is differentiated into functional groups based on level of measured intellectual ability. In a technical sense, because of the IQ exclusion rules for participants in the current investigation, the lower functioning schema operationalized by Boucher et al. (2012) could not be applied to this group of ASD in whom TOMAL testing had been performed. Accordingly, for the current investigation, the dimension of intellectual ability was divided into two groups, one of average IQ (AIQ), defined as verbal IQ (VIQ) ≥ 85, and the other with low IQ (LIQ), defined as VIQ ≤84.

Memory performance was examined broadly using the TOMAL (Bigler & Reynolds, 1994) comparing AIQ and LIQ ASD groups to TDC participants. Age, education and head circumference were not significantly different at the p < .05 level between the groups. To examine region of interest (ROI) volumes of traditional areas associated with memory function based on magnetic resonance imaging (MRI) findings, the automated FreeSurfer (http://surfer.nmr.mgh.harvard.edu/) method was used to compute volumes of mesial temporal lobe structures associated with memory including overall temporal lobe along with mesial temporal lobe volumes including volumes of the hippocampus, parahippocampal gyrus, entorhinal cortex, and the amygdala. Much of this investigation was focused on descriptive findings of potential TOMAL differences between AIQ and LIQ functioning individuals with ASD and sought to (1) distinguish memory performance of ASD both within the disorder by differentiating AIQ and LIQ, and from TDC participants, (2) explore ROI findings in relation to memory in both ASD AIQ and LIQ groups and TDC participants, and (3) examine the relations of memory impairment to volumetric differences in classic memory ROIs of the temporal lobe in ASD. It was hypothesized that TOMAL scores would be significantly higher for TDC subjects than for the AIQ ASD group, and that within ASD, scores for AIQ would be significantly higher than LIQ. We also hypothesized that ROI volume relations by TOMAL memory index performance would be different between the ASD and TDC groups.

Method

Ascertainment

Subjects were recruited predominantly from community sources, including parent support groups, youth groups, and schools, and from clinic social skills groups. The participants in this study are a subset of individuals in a longitudinal investigation of late brain development from three years of age through early adulthood, with the initial findings of TOMAL performance previously reported by Southwick et al. (2011). The subset for this investigation was selected from the larger sample based on age (5–19 years of age) within the reference norms of the TOMAL, having complete TOMAL data from the time of initial assessment, and complete neuroimaging data. Additionally, MRI studies had to have been completed within a few months of when TOMAL testing was performed to be included in the current investigation (M = 2.1, SD = 0.26). All facets of this investigation were undertaken with the understanding and written consent and assent of each subject or legal guardian, with the approval of the University of Utah and Brigham Young University Institutional Review Boards, where testing was performed, and in compliance with national legislation and the Code of Ethical Principles for Medical Research Involving Human Subjects of the World Medical Association.

Participant groups

All participants were males, 5–19 years of age and were part of the Time 1 phase of the longitudinal study. The ASD group had a total of 56 subjects (38 AIQ, 18 LIQ) and the TDC group a total of 31 subjects with complete neuropsychological and neuroimaging datasets. Subjects were selected from the larger neuroimaging study if they met the following criteria: male; between age 5–19 years; complete neuroimaging and neuropsychological datasets. Forty-seven participants were excluded from the larger dataset due to missing or incomplete IQ data; 8 were excluded for missing or incomplete imaging data; 24 were excluded for missing or incomplete TOMAL data. Subject characteristics are summarized in Table 1.

Table 1.

Subject Characteristics

	TDC n= 31			AIQ n= 38			LIQ n= 18
	Mean	SD	Range	Mean	SD	Range	Mean	SD	Range	F	p
Age in years	12.0	4.2	5.3–19.4	13.2	4.1	5.0–19.7	10.4	4.1	5.7–18.4	2.9	0.06
Head Cir (cm)	55.4	2.2	51.8–60.5	55.8	2.7	50.7–60.5	54.5	1.8	51.5–57.9	1.8	0.17
TICV (cm3)	1679.0	178.6	1400.0–2170.0	1674.4	164.0	1270.0–2060.0	1667.9	148.5	1380.0–2010.0	0.03	0.98
Handed	64.8	45.7	−80–100	73.2	45.9	−100–100	49.0	68.5	−93.3–100	1.4	0.26
Education	5.6	3.7	1–12	6.8	3.7	1–14	4.6	3.4	1–12	2.5	.09
FIQ	116.3	14.9	93–152	106.7	12.0	85–137	79.9	8.5	61–99	46.0abc	0.00
PIQ	116.5	15.6	90–155	107.2	12.1	83–131	92.3	18.7	66–138	14.7abc	0.00
VIQ	112.2	14.9	87–140	106.8	14.8	85–145	71.8	7.4	55–83	54.8bc	0.00

^a= TDC is significantly greater than AIQ at p<.05.

^b= TDC is significantly greater than LIQ at p<.05.

^c= AIQ is significantly greater than LIQ at p<.05; FIQ = Full Scale IQ, PIQ = Performance IQ and VIQ = Verbal IQ; TDC = Typical Developing Comparisons, AIQ = Average IQ, LIQ = Low IQ

Idiopathic autism sample

Autism was diagnosed rigorously. For individuals in the autism group, subject’s mother was interviewed using the Autism Diagnostic Interview–Revised (ADI-R; (Lord, Rutter, & Le Couteur, 1994) a semi-structured, investigator-based interview with good reliability and validity. Subjects with autism were also directly assessed using the Autism Diagnostic Observation Schedule–Generic (ADOS-G; Lord et al., 2000) a semi-structured play and interview session designed to elicit social, communication, and stereotyped repetitive behaviors characteristic of autism. All subjects in the autism group met ADI–R, ADOS–G, and the Diagnostic and Statistical Manual of Mental Disorders –Fourth Edition (DSM–IV; American Psychiatric Association, 1994) criteria for autistic disorder (see Table 2 for greater characterization of the autism sample). History, physical exam, fragile X gene testing, and karyotype, performed on all subjects, excluded medical causes of autism. Eighteen subjects in the autism sample were prescribed medications (9 subjects were on selective serotonin reuptake inhibitors, 2 subjects were on tricyclic antidepressants, 4 subjects were on stimulant medications, 2 subjects were on antipsychotic medications, 1 subject was on a proton pump inhibitor). No subjects had a history of seizures, severe head injury, bipolar disorder, schizophrenia, drug, or alcohol abuse at the time of participation. While their primary diagnosis was autism, within the autism sample several had secondary co-morbid diagnoses as follow: 7 subjects were diagnosed with anxiety disorders, 3 subjects were diagnosed with depression, and 5 were diagnosed with attention deficit hyperactivity disorder. Using the verbal IQ metric (see Boucher, Mayes, and Bigham, 2012), the ASD group was divided into two groups, an average (VIQ≥ 85) to above average IQ (AIQ) group and a low (VIQ ≤84) intellectual (LIQ) group.

Table 2.

Characterization of the Autism and Control Sample

	LIQ			AIQ			TDC
Test	n	Mean (SD)	Range	n	Mean (SD)	Range	n	Mean (SD)	Range
ADOS S+C Module 1	2	14 (1.4)	13–15	1	18 (−)	–	0	–	–
ADOS S+C Module 2	6	18.7 (3.3)	15–23	4	17.3 (4.6)	12–22	1	2 (−)	−
ADOS S+C Module 3	9	15.7 (2.1)	14–19	26	13.6 (3.6)	7–20	18	1.4 (1.4)	0–4
ADOS S+C Module 4	1	18 (−)	–	7	15.4 (4.5)	8–20	10	1.1 (1.5)	1.5
ADI-R Soc	18	20.2 (6.3)	8–29	35	18.8 (5.6)	6–28
ADI-R Com	18	16.8 (4.9)	8–25	35	15.8 (4.2)	8–24
ADI-R RSB	18	7.4 (2.4)	2–11	35	6.7 (2.3)	0–10

Note. ADOS S+C = Autism Diagnostic Observation Schedule: Social and Communication Total. The ADOS consists of four modules, and the individual being evaluated is given just one module, depending on expressive language level and chronological age. Each module has different cutoff scores, and as such they should not be considered equivalent. Three AIQ participants were not administered the ADI-R. Two comparison participants had incomplete ADOS data, which is not reported above. ADI–R Soc = Autism Diagnostic Interview, Revised: Reciprocal Social Interactions; ADI–R Com = Autism Diagnostic Interview, Revised: Language/Communication; ADI–R RSB = Autism Diagnostic Interview, Revised: Restricted, Repetitive, and Stereotyped Behaviors and Interests.

Typically-developing sample

Typically-developing comparison (TDC) subjects had no developmental, neurological, or clinical history of major developmental, learning, cognitive, neurological, or neuropsychiatric disorders, with the exception of one subject with a history of seizure. One subject was prescribed an antihistamine medication. Comparison subjects likewise completed an assessment with the ADOS-G and were assessed rigorously for ASD to ensure none met any criterion, including no first-degree relative with autism. Level of intellectual functioning was not stratified in the comparison subjects.

IQ

Different versions of intellectual tests were used over the 10 years of subject accrual to the parent project. IQ findings are reported (for descriptive purposes) that included summary measures from one of the following: Wechsler Intelligence Scale for Children–Third Edition (WISC–III; Wechsler, 1991); Wechsler Adult Intelligence Scale– Third Edition (WAIS-III; Wechsler, 1997); Wechsler Abbreviated Scale of Intelligence (WASI; Wechsler, 1999; VIQ and PIQ indexes), or Differential Ability Scales (DAS; Elliott, 1993). IQ was not used as a covariate since IQ was used as an independent selection variable for AIQ and LIQ designation.

Handedness and Head Circumference

Handedness was measured using the Edinburgh Handedness Inventory (Oldfield, 1971). A score of +100 signifies complete right handedness and −100 indicates complete left handedness. Standard occipitofrontal head circumference was obtained on all participants as previously described (see Bigler et al., 2003). Head circumference was initially done at the time of recruitment as a quick proxy for brain volume (Tate, Bigler, McMahon, & Lainhart, 2007) and because of the role of head size as a factor in autism and the importance of reporting head circumference values in autism research (Lainhart & Lange, 2011). However, a more precise measure of brain volume (TICV) was controlled for during analysis.

Memory

Memory was assessed using the TOMAL (Reynolds & Bigler, 1994). Details of this memory battery in autism have been previously published in Trontel et al. (2013) and Southwick et al. (2011). The updated TOMAL had not been released at the time that this sample was tested. Briefly, the TOMAL samples various domains of memory in children and adolescents, ages 5 years 0 months through 19 years 11 months. The TOMAL is composed of a core battery of 10 subtests, including five verbal and five nonverbal, as well as supplementary subtests (three verbal, one nonverbal). Four TOMAL subtests assess retrieval both immediately upon stimulus presentation and following a 30-min filled delay. Among the 10 core subtests, Memory for Stories involves immediate and delayed free recall of short verbal narratives; Word Selective Reminding is a verbal list-learning task that includes a delayed free recall condition; Object Recall requires immediate verbal recall of paired verbal-visual stimuli; Digits Forward involves repetition of a number series; and Paired Recall involves learning verbal paired associates. Among the 5 core nonverbal subtests, Facial Memory presents arrays of pictured faces which must be recognized and selected among distractors immediately and following a delay; Visual Selective Reminding is a test of spatial learning with a delayed recall condition; Abstract Visual Memory involves immediate recognition and discrimination of abstract geometric figures; in Visual Sequential Memory a set of abstract figures must be recalled sequentially; and Memory for Location is a spatial recall task. Supplementary subtests consist of three additional auditory span and working memory tasks, Digits Backward, Letters Forward, Letters Backward, and Manual Imitation, which involves serial repetition of basic hand gestures. The TOMAL has been shown to have high reliability using standard methods for estimating the internal consistency of the subtests and composites (Reynolds & Bigler, 1994). The TOMAL assesses declarative memory for novel information that was encountered within a specific context. Thus, in the present study these results are broadly described as measures of episodic memory functioning.

Neuroimaging

Volumetric studies were based on magnetic resonance images acquired on a Siemens Trio 3.0 Tesla scanner at the University of Utah. At time point 1, an 8-channel, receive-only RF head coil was used to acquire sagittal 3D MPRAGE T1-weighted images (inversion time=1100ms, echo time=2.93ms, repetition time=1800ms, flip angle=12degrees, field of view=56mm, slice thickness=1.0mm, 160 slices). No subjects in the current study underwent sedation for scanning. No complications were encountered in the scanning process. Additional neuroimaging details previously have been published (see Prigge et al., 2013a; 2013b).

Volumetric image analysis

All analyses were performed with FreeSurfer image analysis suite, version 5.1 (http://surfer.nmr.mgh.harvard.edu/). The established FreeSurfer image pipeline was followed wherein individual DICOM image files were conformed and saved in native MGZ format. All analyses were performed on identical nodes at the Fulton Supercomputing Lab at Brigham Young University. Skull striping, normalization, segmentation and classification were all performed as part of the normal FreeSurfer pipeline. The FreeSurfer Query, Design, Estimate, Contrast (QDEC) function was used for quality inspection of the classified images. Within the QDEC function a ROI may be identified plotting all volumes for that ROI across all participants. This method permits identification of outliers where the segmentation/classification can then undergo visual inspection. All participant scans were visually inspected in this manner before being included in the study. No operator-controlled editing was performed.

Volume calculations of the following FreeSurfer identified ROIs were selected: amygdala and hippocampal, parahippocampal gyrus, entorhinal cortex along total temporal lobe or medial temporal lobe volume (summed volumes of amygdala, hippocampal, parahippocampal, entorhinal cortex) with total intracranial volume (TICV). TICV was used as a covariate.

Statistical analysis

Given the descriptive nature of this investigation, group means were calculated and compared for autism and comparison subjects for ROIs, using MANOVA and p values were Bonferroni corrected for multiplicity. TOMAL composite, index, and subtest scores were compared using MANOVA. TOMAL scores were then correlated with neuroanatomically defined ROIs controlling for TICV and age for each group.

Results

Sample characteristics

As shown in Table 1, no significant differences at the p < .05 level were found between groups on demographic variables (age, head circumference, Total Intracranial Volume (TICV), handedness index, or education). However, there was a significant group difference for IQ, which was expected. Post-hoc comparisons using Bonferroni correction revealed that all three groups were significantly different on PIQ, F(83, 2) = 14.74, p < .001, and FIQ, F(83, 2) =45.96, p<.001, with the TDC group having the highest scores, followed by the AIQ and then LIQ. However, VIQ was not significantly different for the TDC group when compared with AIQ but VIQ for both TDC and AIQ were significantly higher when compared with LIQ (both p<.001).

Memory Performance

Results for TOMAL composite, index, and subtest scores are reported in Table 3. Post hoc analyses revealed that all three groups were significantly different from one another on the Composite, Verbal, Nonverbal, and Delayed memory indices of the TOMAL (p < .01), with TDCs having the highest scores, AIQ having the next highest, and LIQ having the lowest scores. This pattern was also true for the majority of the subtests, including Memory for Stories, Object Recall, Digits Forward, Visual Sequential Memory, and Memories for Stories Delayed (p < .01). The TDC and AIQ groups performed significantly better than LIQ while not differing from each other for Paired Recall, Letters Forward, Memory for Location, and Word Selective Reminding Delayed. No differences were found between AIQ and LIQ, while performance for both was significantly worse than TDC, for Digits Backward, Letters Backward, Facial Memory, Abstract Visual Memory, and Facial Memory Delayed. Visual Selective Reminding was significantly different only between the TDC and LIQ groups (p < .01), with the LIQ group performing significantly worse. No significant differences were found between any of the three groups on Manual Imitation and Visual Selective Reminding Delayed. Table 4 shows the relationship between IQ and TOMAL performance for each group.

Table 3.

Results of Test of Memory and Learning (TOMAL)

	TDC		AIQ		LIQ
	n= 31		n= 38		n= 18
	Mean	SD	Mean	SD	Mean	SD	F	P	ρη2
Composite Memory Index	107.8	9.6	90.8	14.6	70.8	12.3	35.90	.000abc	.51
Verbal Memory Index	105.1	11.5	89.2	15.6	66.1	13.2	33.97	.000abc	.50
Memory for Stories	11.7	3.2	8.3	3.4	4.4	2.5	23.58	.000abc	.41
Word Selective Reminding	11.3	2.3	9.7	4.5	4.9	3.4	14.24	.000bc	.29
Object Recall	9.4	2.7	7.1	3.5	4.4	3.3	11.32	.000abc	.25
Digits Forward	9.7	3.4	7.4	3.5	4.6	2.5	12.22	.000abc	.26
Paired Recall	11.4	2.5	9.8	2.6	6.9	3.8	11.12	.000bc	.24
Letters Forward	9.0	3.0	7.5	3.1	4.4	2.6	11.81	.000bc	.26
Digits Backward	10.7	2.4	8.4	2.8	6.9	2.7	11.76	.000ab	.25
Letters Backward	10.5	2.3	8.3	2.7	6.2	3.3	12.91	.000ab	.27
Nonverbal Memory Index	109.4	11.6	93.0	15.2	77.9	13.2	21.35	.000abc	.38
Facial Memory	10.4	2.6	7.8	2.3	6.6	2.6	13.06	.000ab	.27
Visual Selective Reminding	10.0	2.7	8.2	3.2	6.6	3.2	6.25	.003b	.15
Abstract Visual Memory	12.7	2.7	9.8	3.0	8.2	3.5	12.27	.000ab	.26
Visual Sequential Memory	11.3	2.7	9.1	2.5	6.9	2.4	14.43	.000abc	.29
Memory for Location	11.9	3.6	9.8	4.5	6.6	4.0	8.04	.001bc	.19
Manual Imitation	12.0	2.9	10.7	2.7	10.5	3.2	1.84	.163	.05
Delayed Recall Index	102.5	9.3	91.5	10.9	78.4	11.7	21.87	.000abc	.39
Memory Stories Delayed	11.3	3.5	8.1	3.3	4.7	2.9	18.65	.000abc	.35
Facial Memory Delayed	9.9	2.2	8.0	2.7	6.7	3.5	6.68	.002ab	.16
Word Selective Reminding	10.0	2.5	9.8	2.3	7.2	2.9	6.19	.003bc	.15
Visual Selective Reminding	10.1	1.6	9.1	2.1	8.6	2.0	3.27	.040	.09

Note. Word Selective Reminding was not normally distributed did not pass tests of normality, however an identity transformation was optimal; Word Selective Reminding Delayed did meet the assumption of normality.

^abcIndicates ^ap<.01 for TDC and AIQ, ^bp<.01 for TDC and LIQ, ^cp<.01 for AIQ and LIQ.

ρη² = partial eta squared is an estimate of the effect variance and error variance that is attributable to the effect.

Table 4.

Correlations between IQ and TOMAL performance

	Performance IQ			Verbal IQ			Full Scale IQ
	TDC	AIQ	LIQ	TDC	AIQ	LIQ	TDC	AIQ	LIQ
Index Score
Verbal Memory Index	.13	.57**	−.04	.36	.56**	.66**	.26	.66**	.09
Nonverbal Memory Index	.57**	.63**	.60*	.45*	.21	.06	.60**	.49**	.50
Delayed Memory Index	−.06	.38*	−.16	.18	.26	.50	.07	.37*	.35
Composite Memory Index	.47*	.67**	.35	.52**	.43*	.38	.55**	.64**	−.12

^*= p<.05

^**= p<.01. TDC=typically developing comparisons; AIQ=above average IQ autism; LIQ=below average IQ autism

Effect sizes for individual contrasts (TDC vs. AIQ, TDC vs. LIQ, AIQ vs. LIQ) were calculated for the Composite Memory Index and Verbal Memory Index, as the overall effect size of the MANOVA for these indices was greater than or equal to a moderate effect (Cohen’s d > .5). The effects between the individual groups for the Composite Memory Index were quite large. The effect size for the Composite Memory Index between TCD and AIQ was d = 1.39, d = 3.52 between TDC and LIQ, and d = 1.48 between AIQ and LIQ. Similarly, the effect between the individual groups on the Verbal Memory Index was d = 1.20 between TDC and AIQ, d = 3.27 between TDC and LIQ, and d = 1.61 between AIQ and LIQ.

Volumetric Differences in Memory ROIs

Volumetric findings are summarized in Table 5. MANOVAs comparing the two ASD groups with TDC controlling for age and TICV revealed no significant volume differences in the left, right, or total entorhinal cortex, parahippocampal gyrus, hippocampus, or amygdala (p > .05). Additionally, no group differences were found for total temporal lobe, total mesial temporal lobe (sum of parahippocampal, entorhinal, hippocampus, and amygdala), total gray matter, or total brain volume (p > 0.05).

Table 5.

Comparisons of Structure Volume by Group controlling for TICV and Age

Structure Volume	TDC Mean (SD)	AIQ Mean (SD)	LIQ Mean (SD)	F	p
Temporal Lobe	134.8 (13.6)	132.8 (13.1)	132.8 (14.8)	0.95	0.39
Left	67.2 (7.5)	66.3 (6.9)	66.1 (7.7)	0.71	0.49
Right	67.5 (6.4)	66.5 (6.7)	66.8 (7.6)	1.05	0.36
Entorhinal	4.3 (1.1)	4.0 (0.8)	4.1 (0.6)	1.18	0.31
Left	2.3 (0.6)	2.1 (0.4)	2.2 (0.4)	1.65	0.20
Right	2.1 (0.6)	1.9 (0.5)	2.0 (0.4)	0.36	0.70
Parahippocampal	4.9 (0.8)	4.9 (0.8)	4.9 (0.5)	0.02	0.98
Left	2.6 (0.4)	2.6 (0.5)	2.5 (0.3)	0.55	0.58
Right	2.3 (0.5)	2.3 (0.4)	2.4 (0.4)	0.36	0.70
Hippocampus	8.9 (1.0)	9.3 (1.2)	9.1 (1.2)	1.29	0.28
Left	4.4 (0.7)	4.7 (0.6)	4.5 (0.7)	2.40	0.10
Right	4.6 (0.5)	4.6 (0.7)	4.6 (0.6)	0.29	0.75
Amygdala	3.4 (0.4)	3.6 (0.6)	3.5 (0.4)	1.33	0.27
Left	1.7 (0.2)	1.8 (0.3)	1.8 (0.2)	1.79	0.17
Right	1.7 (0.2)	1.8 (0.3)	1.7 (0.2)	0.97	0.38
Mesial Temporal Lobe	21.59 (2.6)	21.7 (2.6)	21.6 (1.8)	0.10	0.90
Total GM	596.5 (53.8)	587.0 (63.4)	613.0 (71.9)	0.09	0.92
Total Brain	1345.7 (123.9)	1338.0 (130.1)	1332.7 (115.2)	0.92	0.40

Note. Volumes are in cm³. Controlling for Age and TICV.

Relation between ROI Volume and TOMAL Performance

As shown in Table 6, ASD individuals’ performance on the Composite Memory Index, Verbal Memory Index, and Nonverbal Memory Index of the TOMAL was negatively correlated only with the amygdala in the LIQ group, indicating that larger amygdala volume related to lower memory performance for that group. Additionally, total gray matter volume was negatively correlated with the Verbal Memory Index for both AIQ and LIQ, but not for TDC, indicating that greater gray matter volumes were related to lower verbal memory performance in the ASD groups. None of the other classic memory ROIs was correlated with TOMAL performance on the Composite Memory Index, Verbal Memory Index, Nonverbal Memory Index, or Delayed Recall Index for the three groups.

Table 6.

Correlations between brain volumes and TOMAL performance controlling for TICV

Structure	Composite Memory Index			Verbal Memory Index			Nonverbal Memory Index			Delayed Recall Index
	TDC	AIQ	LIQ	TDC	AIQ	LIQ	TDC	AIQ	LIQ	TDC	AIQ	LIQ
Temporal Lobe	−0.01	−0.17	−0.25	0.07	−0.26	−0.34	−.09	−0.05	−0.13	0.05	−0.13	−0.15
Entorhinal	−0.11	−0.12	−0.06	−0.06	−0.10	−0.12	−0.11	−0.13	−0.02	−0.03	−0.08	−0.14
Parahippocampal	−0.17	−0.01	−0.18	−0.10	0.09	−0.12	−0.16	−0.11	−0.18	−0.15	0.08	−0.11
Hippocampal	0.08	−0.07	−0.20	0.04	−0.07	−0.08	0.10	−0.05	−0.27	0.09	−0.16	0.14
Amygdala	−0.12	0.26	−.69**	−0.17	0.28	−0.59*	−0.02	0.18	−0.60*	0.05	0.33	−0.44
Mesial Temporal Lobe	−0.09	0.01	−0.39	−0.07	0.05	−0.29	−0.06	−0.06	−0.39	−0.01	0.03	−0.13
Tot Gray Matter	0.28	−0.27	−0.37	0.32	−.37*	−0.51*	0.11	−0.11	−0.17	0.16	−0.23	−0.43
Total Brain	0.26	−0.07	0.06	0.24	−0.10	−0.07	0.18	−0.03	0.15	0.29	−0.15	0.02

^*= p<.05

^**= p<.01, Controlling for Age and TICV. Note. Only the correlation between the amygdala and the composite memory index for the LIQ group remained significant after the Bonferroni correction.

Discussion

Memory Performance Differences

This investigation revealed, consistent with our hypotheses, that individuals with ASD in the AIQ or LIQ group performed significantly worse on memory tasks than did their TDC counterparts. Furthermore, AIQ and LIQ groups differed in memory performance as expected; AIQ performed significantly better than LIQ. Interestingly, in comparison to the TOMAL normative standard, the AIQ group performed within one standard deviation (average) on all four memory index scores, while the LIQ group performed one standard deviation below (low average) on the composite, nonverbal, and delayed recall indices but two standard deviations (impaired) below on the verbal memory index. This suggests that verbal intellectual abilities likely have the greatest influence on verbal-based memory impairments. These findings are consistent with previous literature implicating memory impairments as part of the ASD cognitive phenotype (Ben Shalom, 2003; Boucher & Bowler, 2008) and suggest that memory performance is related to level of intellectual ability. Additionally, the relationship between IQ and memory performance for each group may suggest that despite the overall lower TOMAL performance in the AIQ ASD group, their relation with intellectual ability is similar to TDC, whereas with LIQ an even greater level of impaired memory function is present.

Overall, the ASD AIQ and LIQ groups appear to perform more similarly to one another on nonverbal tasks (with the exception of visual sequential memory and memory for location). However, for verbal memory tasks that do not place prominent demands on working memory and for delayed memory tasks (with the exception of facial memory), AIQ appears to perform significantly better than LIQ individuals. Given that the LIQ group is defined by lower levels of VIQ and ASD is associated with language deficits (DePape, Hall, Tillmann, & Trainor, 2012), the memory impairments in the LIQ group likely reflect a negative interaction on linguistic and mnestic deficits from the core language deficits that are part of ASD (see Tyson et al., 2013). How level of general cognitive ability in ASD relates to memory performance requires further study.

Working memory

In the current investigation, the AIQ group performed similarly to TDC on a verbal working memory task (e.g. letters forward), which is consistent with research by Rumsey and Hamburger (1988) and Minshew and Goldstein (1993) who failed to find any memory deficits in higher-functioning individuals with autism. However, as the task becomes more complex (i.e., recitation of numbers/letters backwards), AIQ and LIQ groups appear to perform more similarly (e.g., digits and letters backward). This is consistent with Barendse et al.’s (2013) review of the working memory literature in ASD that found greater deficits in working memory with heavier demands and with Boucher et al.’s (2003) review which found that tasks that require executive control in order to manipulate and process the online information (e.g. digits backward), were more difficult for individuals with ASD. Likewise, a difficulty with processing of complex information in ASD is consistent with the disconnection model posited by Minshew and Goldstein (2007).

Intact Memory Performance in ASD

Two subtests revealed no impairments in ASD participants regardless of whether AIQ or LIQ when compared with TDC: manual imitation (a sequential motor memory task) and visual selective reminding delayed (recall of visual information previously learned over trials). While motor proficiency is a commonly reported deficit in ASD (Duffield et al., 2013; Minshew, Goldstein, & Siegel, 1997), the current study suggests that memory for a motor sequence (with no measure of motor fluency) remains intact in ASD. Additionally, intact performance on delayed visual selective reminding in both groups is consistent with the Task Support Hypothesis (Bowler, Matthews, & Gardiner, 1997; Bowler, Gardiner, & Berthollier, 2004), which posits that the absence of adequate “supports” during testing explains the deficits seen on recall tasks. The importance of supports has been found to be particularly true for multiple trial learning (Bowler, Gaigg, & Gardiner, 2008; Bowler, Limoges, & Mottron, 2009).

Heterogeneity in ASD

In the newest version of the Diagnostic and Statistical Manual-5^th Edition (DSM-5; American Psychiatric Association, 2013), individuals meet criteria for a diagnosis of ASD if they demonstrate characteristics in two core areas (communication and social deficits and fixed or repetitive behaviors). The criteria reflect an attempt to acknowledge the spectrum of abilities and deficits in the disorder. The current memory performance findings underscore a dimensional model approach to autism and reflect the vast differences in memory across the spectrum, not simply between TDC and ASD. TOMAL Composite Memory Index ranged from profoundly impaired to superior (41 to 124) for ASD, while the TDC group ranged from low average to superior (81 to 126). Clearly, much greater variability exists within the ASD group for memory performance (see Table 3). Likewise, full scale IQ in ASD ranged from 61–137, while TDC FSIQ scores ranged from 93–152. This neurocognitive heterogeneity emphasizes the complexity of the neurobiological underpinnings of autism. Further, the lowest functioning participants in the ASD group were excluded from this study, so it is unknown how substantial the memory impairment is or how broad the within group differences may be that are not reliably captured with only two severity levels of comparison for ASD (i.e., AIQ versus LIQ).

No Structural Volumetric Differences in Mesial Temporal Lobe Structures

There were no gross volumetric differences between the ASD and TDC groups in mesial temporal lobe structures typically thought to be involved in memory processes. The finding of “normal” ROI structural volumes in ASD in the current study is consistent with several volumetric studies (Bigler et al., 2003; Saitoh et al., 1995; Piven, Bailey, Ranson, & Arndt, 1998) that have not found unique differences in gross temporal lobe morphology in ASD. Since memory differences in both AIQ and LIQ ASD were shown compared to TDC individuals, the absence of gross morphological differences implies that memory function in ASD is likely due to disrupted neural connectivity within memory networks rather than the absolute size of mesial temporal lobe structures.

Relation of Memory Performance and ROI

Contrary to our hypotheses, hippocampal, parahippocampal gyral, and entorhinal cortex volumes were not significantly related to memory performance for any of the groups. However, for the LIQ group only, memory performance was negatively correlated with amygdala volume. In other words, larger amygdala volumes in the LIQ functioning ASD group were related to poorer memory performance in our sample. Although this relationship did not withstand correction for family-wise error, this finding is interesting given research suggesting that learning and memory processes may be specifically linked to the amygdala in autism (Howard et al., 2000; O’Keefe & Nadel, 1978).

The relation between larger amygdala and poorer memory possibly relates to the overgrowth hypothesis of neural development in ASD where larger volume is associated with less efficient networks (Amaral, Geschwind, & Dawson, 2011). Casanova and Trippe (2002) and Casanova (2007) have shown that increases in cortical gray matter may relate to aberrant increased white matter projections necessary to maintain the connectivity of the increased number of cortical cells in the autistic brain. This suggests that an increased number of white matter connections in autism may not result in greater and more efficient functional connectivity, but rather less efficient networks.

Limitations

The two autism groups were differentiated using a cutoff of one standard deviation below average verbal intelligence. Although this differentiation proved fruitful in delineating memory differences between the groups, it was arbitrary and not necessarily a clinically derived criterion for differentiating within autism. More clinically-meaningful phenotypes may allow better delineation of cognitive factors associated with impaired memory function in ASD. Small (LIQ = 18) and modest (AIQ = 38; TDC = 31) group sizes limited power to detect meaningful differences in important demographic variables. The LIQ group had, on average, younger participants with fewer years of education than the other two groups (approximately 3 years difference), which might suggest that memory performance was related to age and education in these individuals with ASD. Head circumference also favored the TCD group, although total brain volume did not. It could be that these variables contribute to group differences given sufficient sample sizes as direct or indirect predictors. Although all children were in educational programs, it was impossible to control for the kinds and types of educational experiences that had occurred. Likewise, there were potential differences in the kinds of supportive therapies and interventions received by participants. There are also potential age, education and family interaction variables that could influence memory function as measured by the TOMAL that could not be controlled in this study. Accordingly, given the limited power of group cell sizes, caution is recommended in the genrealizability of these findings. Replication with larger sample sizes matched on identified key variables related to memory performance will be critical for future research to confirm present findings. Further, participants who were not amenable to testing due to behavioral and/or cognitive factors (low functioning participants) were not included in the current study, which may also limit generalizability of the findings to the autism spectrum as a whole.

The inability to find consistent volumetric differences correlated with memory function in ASD may suggest that the neuropathology of ASD is not expressed at the level of gross morphology within these structures. It may be that more refined measures that assess neural connectivity and networks such as fMRI and diffusion tensor imaging (DTI) will better clarify the neurobiological basis of memory impairment in autism. A review of the literature on memory in ASD by Goh and Peterson (2012) reveals that the bulk of the existing research in this involves structural imaging rather than functional imaging, so there is an absence of such research to address the neural connectivity of memory in ASD. Future studies investigating the structure and functional neuroimaging of neural pathways in memory regions may provide insight regarding abnormalities that contribute to memory and learning impairments in this population. It should be noted that automated image analysis methods used in this investigation have limitations and it remains possible that more refined structural image analysis methods may prove successful in defining brain structure-function relations in memory and ASD (Hanson et al., 2012).

Conclusion

Generally, lower performance on measures of intellectual functioning was associated with worse TOMAL performance in ASD and when compared to TDC. The AIQ group performed similarly to TDC on less complex verbal working memory tasks. AIQ and LIQ ASD groups appear to perform more similarly on nonverbal tasks and more complex verbal working memory tasks. However, for verbal memory tasks that do not require working memory and for delayed memory tasks (with the exception of facial memory), AIQ performed significantly better than LIQ ASD (i.e., greater within group variability) and more similar to TDC.

Although amygdala volume was found to be negatively related to general memory performance in LIQ and total gray matter volume was found to be negatively related to verbal memory performance for both ASD groups, these relations were not robust enough to withstand correction for multiple comparisons and unlikely to represent a major explanation for poor memory performance in ASD.