Childhood leukemia survivors exhibit deficiencies in sensory and cognitive processes, as reflected by event-related brain potentials after completion of curative chemotherapy: A preliminary investigation.

Abstract

Objective:

The purpose of this study was to characterize post-chemotherapy sensory, memory, and attention abilities in childhood survivors of acute lymphoblastic leukemia (ALL) to better understand how treatment affects cognitive functioning.

Methods:

Eight ALL survivors and eight age-matched, healthy children between the ages of 5-11 years participated in the study. Among the ALL survivors, a median of 63 days (range 22-267 days) elapsed between completion of chemotherapy and this assessment. Sounds were presented in an oddball paradigm while recording the electroencephalogram in separate conditions of passive listening and active task performance. To assess different domains of cognition, we measured event-related brain potentials (ERPs) reflecting sensory processing (P1 component), working memory (mismatch negativity [MMN] component), attentional orienting (P3a), and target detection (P3b component) in response to the sounds. We also measured sound discrimination and response speed performance.

Results:

Relative to control subjects, ALL survivors had poorer performance on auditory tasks, as well as decreased amplitude of the P1, MMN, P3a, and P3b components. ALL survivors also did not exhibit the amplitude gain typically observed in the sensory P1 component when attending to the sound input compared to when passively listening.

Conclusions:

Atypical responses were observed in brain processes associated with sensory discrimination, auditory working memory, and attentional control in pediatric ALL survivors indicating deficiencies in all cognitive domains compared to age-matched controls.

Significance:

ERPs differentiated aspects of cognitive functioning, which may provide a useful tool for assessing recovery and risk of post-chemotherapy cognitive deficiencies in young children. The decreased MMN amplitude in ALL survivors may indicate (N-methyl D-aspartate) NMDA dysfunction induced by methotrexate, and thus provides a potential therapeutic target for chemotherapy-associated cognitive impairments.

Introduction

Childhood acute lymphoblastic leukemia (ALL) is typically curable with intensive chemotherapy (Hunger & Mullighan, 2015), although the treatment has been linked to neurotoxicity that brings about cognitive deficits even when cranial radiation is omitted from the treatment regimen (van der Plas et al., 2015; Iyer et al., 2015; Kanellopoulos et al., 2016; Jacola et al., 2016; Pierson et al., 2016). Children treated for ALL with systemic and intrathecal chemotherapy exhibit increased rates of impairment in discrete neurocognitive and behavioral domains, including attention and working memory, processing speed, and executive functioning (Ashford et al., 2010; Conklin et al., 2012; Duffner et al., 2014; Edelstein et al., 2011; Harila, Winqvist, Lanning, Bloigu, & Harila-Saari, 2009; Hodgson, Hutchinson, Wilson, & Nettelbeck, 2013; Jansen et al., 2008; Kadan-Lottick et al., 2009; Krull et al., 2013; Lofstad, Reinfjell, Hestad, & Diseth, 2009; Reinfjell, Lofstad, Veenstra, Vikan, & Diseth, 2007), which may reflect interference with normal brain development and/or accelerated central nervous system aging (Schuitema et al., 2013). Of all the chemotherapeutic agents used in typical curative regimens for children with leukemia, methotrexate is thought to be the one most responsible for causing persistent cognitive dysfunction, presumably by increasing concentrations of homocysteine, leading to oxidative damage of neurons and vascular endothelium as well as altered neurotransmission through glutamate receptors (Duffner et al., 2014; Cole & Kamen, 2006). However, the specific mechanisms by which the combination of chemotherapy regimens for childhood ALL causes neurotoxicity associated with cognitive impairments is poorly understood. Neuroimaging studies have sought to determine which neural substrates are affected in ALL survivors, but results reporting structural changes have been mixed and correlate poorly with severity of cognitive impairment (Hearps et al., 2017; Reddick & Conklin, 2010).

The negative impact that cognitive impairments can have on the neural development of the child, and the cascading effects on academic performance, long after treatment has ceased (Buizer et al., 2006; Peckham & Meadows, 1988) has a significant impact on quality of life for these cancer survivors. Thus, it is crucial to gain a more detailed understanding of the specific effects of ALL treatment on cognitive functioning. Previous studies using psychometric testing have shown impairments in working memory (Harila et al., 2009; Jansen et al., 2008; Kingma et al., 2001, 2002; Lofstad et al., 2009; Raymond-Speden, 2000; Rodgers, Marckus, Kearns, & Windebank, 2003) and attention (Kingma et al., 2001; Lofstad et al., 2009; Rodgers et al., 2003) but have not connected working memory performance with brain activity. The goal of the current study was to characterize cognitive functions associated with sensory processing, working memory, and attention abilities by recording electrophysiological and behavioral measures during task performance that engages those processes.

Electroencephalography (EEG) is a powerful tool because it can be used to assess multiple domains of cognitive functions in ALL survivors (Järvelä et al., 2011; Schroger, 1998). Event-related potentials (ERPs), which are extracted from the continuous EEG record, have been shown to correlate well with cognitive functions in normal individuals, as well as in individuals with cognitive impairments related to illness (Evans et al., 2011; Foster et al., 2013; Light & Braff, 2005; Light et al., 2008; Polich & Herbst, 2000). ERPs provide high temporal resolution that can be used to differentiate where in the information processing stream cognition is affected. For example, the auditory P1 component represents the physiologic detection of sound (Alain & Tremblay, 2007; Picton et al., 1999) and can be used to assess basic cortical sensory auditory processes in ALL survivors (Anderson et al., 2010). The P1 is evoked whether or not participants are actively attending to a stimulus, though attention to the sound increases the P1 amplitude (Giuliano et al., 2014; Sanders et al., 2006). The P1 latency evolves during childhood in a predictable manner (becoming earlier with increasing age), and can thus be used to assess auditory system development (Sharma et al., 2013; Sussman et al., 2008). In this way, the P1 component provides a metric of sensory processing that can be assessed with regard to more complex cognitive deficits that may be affected by subtleties in processing incoming stimuli.

The mismatch negativity (MMN) component of the ERPs provides a measure of auditory working memory. It is elicited by oddball sounds whether or not attention is directly focused on the sound input (Naatanen & Alho, 1995; Näätänen et al., 1978; Näätänen et al., 2001; Sussman et al., 2005; Sussman, 2007). MMN elicitation is dependent upon open N-methyl-D-aspartate (NMDA) channels (Javitt et al., 1996; Tikhonravov et al., 2008) and is thus associated with cellular mechanisms involved in memory processes (Compte et at., 2000; Kauer et al., 1988). This suggests a link between memory and MMN. The MMN component is an important tool for investigation because it has been hypothesized that over-activation of NMDA receptors by glutamate to be one of the causes of chemotherapy induced neurotoxicity (Vijayanathan, Gulinello, Ali, & Cole, 2011). Thus, abnormal MMN elicitation in leukemia survivors compared to controls would be consistent with this observation in animal models.

In contrast to the sensory-based P1 and memory-based MMN components, the P3 component of ERPs provides an index of attention-based functions. The P3 has multiple subdivisions (Friedman et al., 2001), reflecting higher-level cognitive processes involved in orienting attention, target discrimination, and context updating (Polich, 2009). The P3a component represents involuntary orienting to salient environmental sounds, and can be elicited during an active or passive listening tasks (Friedman et al., 2001), whereas the P3b component represents voluntary target responses (Knight & Scabini, 1998) and is only elicited when participants are actively listening to the sounds. Therefore, these two components represent different aspects of attention. P3a can indicate difficulties in orienting attention, while P3b can represent difficulty maintaining attention needed to focus on a task and select a relevant stimulus. P3b latency has been found to peak later, and to have a smaller amplitude in survivors of childhood cancer, suggesting slower and more effortful target detection (Lähteenmäki et al., 2001; Überall et al., 1996).

We identified, in the current study, which cognitive processes were most affected in ALL survivors compared to healthy controls by obtaining behavioral responses of sound discrimination accuracy (hit rate and false alarm rate) and processing speed (reaction time), and by recording and measuring the timing and amplitude of ERP components that reflect sensory, memory, and these attention-based functions. Sensory processing was indexed using the P1 component, working memory was indexed by the MMN component, orienting attention by the P3a component, and volitional attention was reflected in the P3b component. We hypothesized that methotrexate treatment negatively impacts brain activity associated with specific sensory and cognitive functions. Because previous studies using conventional neurocognitive tests have detected deficits in attention ability, we predicted that the P3 components would have smaller amplitude and longer latency responses compared to healthy controls. We also expected longer response times along with longer latency components. Additionally, in light of previous studies showing excessive stimulation of NMDA-receptors with methotrexate treatment, we predict a decrease or absence of the MMN component in ALL survivors compared to healthy controls. Understanding how these brain responses are affected post-treatment will elucidate the nature of the cognitive deficits and provide insights into new targets for intervention or prevention strategies.

Methods

Participants

Eight children age 5–11 years (median age 7 years, 3 males) who underwent chemotherapy treatment for ALL participated in the study. Participant data for the ALL survivors is shown in Table 1. Children were treated with one of three cooperative group protocols: Dana Farber Cancer institute Acute Lymhoblastic Leukemia Consortium protocol 05–001 (n=6); Children’s Oncology Group (COG) protocol AALL0433 (n=1); or COG protocol AALL0232 (n=1). One child was treated with protocol AALL0433, another single child was treated with protocol AALL0232, and the other 6 children were treated with protocol DCFI 05–001 These three protocols all consist of repeated cycles of multi-agent chemotherapy administered systemically and intrathecally over a period of two to three years. The details of the timing and individual drug doses have been previously published (Larsen et al., 2016; Lew et al., 2014; Place et al., 2015), along with outcome results indicating similar results in terms of both toxicity and relapse risk. A median of 63.5 days (range 22–267 days) elapsed since the completion of leukemia therapy at time of testing. All subjects recovered from any acute toxicities related to treatment. Subject #5 had T-lineage ALL and received 12 Gy cranial radiation as part of his leukemia therapy, as dictated by his chemotherapy regimen (Place et al., 2015).

Table 1.

Participant information for ALL survivors

ID	Gender	Race	Ethnicity	Immuno-phenotype	Treatment	Radio-therapy	Age at Testing (years)	Days Since Treatment End
1	M	C	H	preB	AALL0433	No	9	29
2	M	C	H	preB	DFCI 05–001	No	11	80
3	F	C	H	preB	AALL0232	No	11	57
4	F	AA	H	preB	DFCI 05–001	No	7	70
5	M	C	N	T	DFCI 05–001	Yes	7	30
6	F	C	H	preB	DFCI 05–001	No	6	122
7	F	C	H	preB	DFCI 05–001	No	6	22
8	F	C	H	preB	DFCI 05–001	No	5	267

M=male, F=female, C= Caucasian, AA=African American, H= Hispanic, N= Non-hispanic.

Eight healthy children (6–11 years, median age 8 years, 4 males) participated in the study and served as controls. Healthy, control participants were recruited through flyers displayed in the local schools, hospitals, and community centers, and were age-matched as closely as possible to the ALL survivors (within 6 months). Parents of healthy control participants were pre-screened by phone interview to exclude past or present diagnoses of learning, speech/language, hearing, emotional/behavioral, or neurological disorders. All participants were in their age-appropriate grade in school, with no grade retention and no report of special education services. Recruits were then scheduled for a 2-h screening session with a licensed psychologist for psychometric testing. All eight participants serving as healthy controls had standard scores of at least 85 on the following instruments testing for cognitive and language function, reading, and phonological processing abilities: Wechsler Abbreviated Scale of Intelligence (WASI) for cognitive function; Woodcock–Johnson-III Tests of Achievement (WJ-III) Letter/Word Identification and Word Attack for reading/decoding; WJ-III Understanding Directions and Children’s Essentials of Language Fundamentals-3 (CELF-3) Sentence Repetition for language function; The Phonological Awareness Test (PAT) Rhyming; and The Comprehensive Test of Phonological Processing (CTOPP) core screening tests for phonological awareness, phonological memory, and rapid naming. In addition, participants had to report (parent report for children) fewer than six symptoms of inattention and hyperactivity/impulsivity on a DSM-IV based checklist.

Protocols followed the Declaration of Helsinki, with oral assent was obtained from all of the participants and written consent was obtained from the parents of participants, after the study was explained to them. Participants were told that they could stop the session at any time for any reason. The protocol and informed consent documents were available in Spanish and English and were approved by the Institutional Review Board at Albert Einstein College of Medicine where the study was conducted. All participants passed a hearing screening (20 dB HL or better bilaterally at 500, 1000, 2000, and 4000 Hz).

Stimuli

Stimuli were 50 ms duration pure tones with a 5 ms rise/fall time, created using Adobe Audition, and presented binaurally using insert headphones (E-a-rtone® 3A, Indianapolis, IN). Tones were presented with a stimulus onset asynchrony (SOA, onset-to-onset) of 300 ms. An auditory oddball paradigm was used, in which a frequent tone was presented 88% of the time (the “standard”) and an infrequent tone was presented 12% of the time (the “deviant”). The deviant tone differed from the standard tone in frequency value or in intensity level, but not in both. The deviant tones were presented randomly so their occurrence could not be anticipated across the stimulus block. In addition, presentation of the deviant stimuli was constrained so that no two deviant tones would occur successively. Thus, deviants occurred singly and unpredictably throughout each stimulus block.

There were two conditions of intensity, one in which the louder intensity tone was the standard and the softer intensity tone was the deviant (soft-intensity deviant condition) and the other in which the softer intensity tone was the standard and the louder intensity was the deviant (loud-intensity deviant condition). Each intensity condition was presented using two types of attention to the sounds: “Attended” (Att) and “Ignored” (Ign). For the loud-intensity deviant conditions (hereafter called “Att-Loud” and “Ign-Loud”) tones had a frequency value of 880 Hz, a standard intensity of 50.5 dBA, and a deviant intensity of 65.5 dBA. For the soft-intensity deviant conditions (hereafter called “Att-Soft” and “Ign-Soft”), tones had a frequency value of 880 Hz, a standard intensity of 65.5 dBA, and a deviant intensity of 50.5 dBA.

There was one ‘ignore’ condition for the frequency oddball (hereafter called Ign-Freq), in which the intensity value of all the tones was 50.5 dBA, the standard frequency was 880 Hz and the deviant frequency was 2093 Hz. Deviants were presented pseudorandomly within each condition, with the restriction that at least four standard tones be presented between any two deviant tones, and each sequence begins with at least five consecutively presented standard tones. All stimuli were calibrated using a sound level meter (calibrated with a Brüel & Kjær® sound level meter with an artificial ear). Data for aged-matched controls were obtained from a previous study using similar parameters (Sussman & Steinschneider, 2011). The stimuli differed only for the frequency oddball condition, in which the standard tone had a frequency of 440 Hz and the deviant frequency was 2637 Hz) with all other experimental parameters identical.

Procedures

Participants were seated in a comfortable chair in a sound attenuated booth (Industrial Acoustics Corp., Bronx, NY). A research assistant or parent was present in the booth during the experiment to facilitate cooperation and to reduce anxiety. The passive listening conditions (Ign-Loud, Ign-Soft, and Ign-Freq) were presented first, in random order across participants. During passive conditions, participants watched a silent close-captioned video of their choosing while sounds were presented through insert headphones. EEG was monitored for eye saccades. In the second half of the experiment, the active listening conditions (Att-Loud and Att-Soft) were presented randomly across participants, and the movie was turned off. Participants were instructed to listen to the sounds and press a response button when they heard the louder (or softer) intensity tones in their respective conditions. A short instruction and practice was provided before each new condition. The passive condition contained 1650 stimuli, presented in 3 blocks of 550 stimuli, which was 8.25 minutes in total (2.75 mins per block). The active condition contained 1000 stimuli in total, presented in 2 blocks of 500 stimuli, which was 5 minutes in total (2.5 mins per block). When attention is directed away from the stimuli (passive listening), a greater number of stimuli are needed to increase signal-to-noise ratio for the sound responses.

Cap placement took approximately 30 minutes, during which time the child watched a self-selected movie with the sound on. A brief practice session was provided prior to the start of the active listening conditions, on sound sequences that were not part of the experimental protocol, to make sure the participant understood the task and had a chance to ask questions. The recording session was approximately half an hour (including instructions and practice time), with ten-to-fifteen minutes of watching a silent (closed-captioned) movie during EEG recording, and approximately five-to-ten minutes of active listening and button presses during EEG recording. Short breaks were given as needed at any time during testing, and all participants took a longer break around the middle of the session in which they were unhooked from the amplifiers, had a snack and talked to the experimenter. All of the participants completed the full session and seemed to enjoy interacting with the experimenter.

EEG recording and data processing

The EEG recordings were obtained using a 32 channel electrode cap in the international 10–20 system, plus the left and right mastoid electrodes. The reference electrode was affixed to the bulb of the nose. Electro-oculogram (EOG) was monitored with an external electrode placed on the left cheek in a bipolar configuration with the F1 electrode. Impedance of each electrode was kept below 5Ω. Continuous EEG was acquired at a rate of 500 Hz with 0.05–100 Hz bandpass filtering. Offline filtering was performed using a .1–30 Hz bandpass filter with 24 dB rolloff and a zero phase shift. Filtered EEG was transformed into epochs of 600 ms in length, which included a 100 ms pre-stimulus period and 500 ms post-stimulus period. Artifact rejection criteria was −100/+100 μV, performed after baseline correction on the entire epoch. The remaining epochs were then averaged according to stimulus type (standard and deviant) separately in each condition, for each individual and then baseline corrected to the pre-stimulus period. The amplitude was measured as the mean voltage in the post-stimulus time window, relative to the mean voltage during the 100-ms pre-stimulus baseline period. Thus, the pre-stimulus period served as the reference (`biological zero’), from which the stimulus-elicited electrical responses were measured. Individual grand-mean waveforms obtained within each group were averaged across participants to produce the grand-mean group waveforms for each condition (ALL survivors and healthy controls separately).

The P1 component is elicited by sound onsets in both the standard and deviant waveforms at this stimulus rate (Sussman et al., 2008). The P1 component was measured in the standard waveforms to eliminate potential overlap with the oddball response (which is obtained only in the deviant tones). The endogenous components (MMN, P3a, and P3b), which are elicited only by the oddball stimuli, reflect sound discrimination. To display responses to the oddball deviants, difference waveforms were calculated by subtracting the standard ERP from the deviant ERP from the grand-mean waveforms in each condition, separately. This subtracts out the obligatory waveforms that are common to both standard and deviant and leaves the mean peak response of the change detection response in the MMN, P3a, and P3b components.

Quantification and Statistics

A response was counted correct if it fell between 100–900 ms post-target onset. Hit rate (HR) was calculated as the number of correct button presses divided by the total number of target stimuli. A false alarm was any button press outside the window of a target stimulus. The false alarm rate (FAR) was calculated as the number of false button presses divided by the total number of non-target stimuli. Reaction time (RT) was calculated from the onset of the target stimulus. To statistically verify the presence of the ERP components, the peak latency of the MMN, P3a, and P3b components were visually identified in the group mean waveforms for each group and condition. The Neuroscan software was used to objectively determine peak latency of each ERP component in the grand mean difference waveform. Each component peak was identified in the grand-mean waveforms at the electrode with its maximum signal-to-noise ratio. The P1 was measured from the Fz electrode, the MMN from the Fz, P3a at the Cz, and P3b at the Pz electrode. An interval centered on the peak latency of the component was used to calculate the interval used to measure the amplitude of the ERP waveform elicited by each stimulus type in each individual grand-mean waveform. We used a 50 ms interval for P1, MMN, and P3a components, and 60 ms for the P3b component. Three midline electrodes covering frontal, central, and parietal scalp locations (Fz, Cz, Pz) were included in statistical analyses to assess the scalp distribution of the ERP components (where Fz is largest for P1, Fz is largest for MMN, Cz is largest for P3a, and Pz is largest for P3b).

Mixed model repeated measures analysis of variance (rmANOVA) was used to statistically compare amplitude and latency of the components for within-subjects and between-subjects comparisons. The within-subjects factors were stimulus type (standard/deviant), deviant type (frequency/loud intensity/soft intensity), attentional listening state (passive/active), and electrode (Fz/Cz/Pz). The electrode factor was included to confirm the scalp distribution of each ERP component as consistent with its known topography. The between-subjects factor was group (ALL survivors/healthy controls). Where data violated the assumption of sphericity, Greenhouse-Geisser corrections were applied and corrected p values are reported, along with epsilon values. Partial eta squared was used to measure effect size. For post hoc analyses, Tukey HSD for repeated measures was conducted on pairwise contrasts only when main effects or interactions of the omnibus ANOVA were significant. Contrasts were reported as significantly different at p < 0.05. All statistical analyses were performed using Statistica 12 software (Statsoft, Inc., Tulsa, OK). Effect sizes are also reported.

Results

Behavior

Behavioral results are displayed in Figure 1 and in Tables 2–3. At the group level, the soft deviants were more difficult to identify than loud deviants (main effect of intensity deviant type: F_1,14=31.10, p<0.01, η_p²=0.70), with no main effect of disease status (F_1,14=2.80, p=0.12, η_p²=0.17), and no interaction between intensity deviant type and disease status on HR (F_1,14<1, p=0.94). Similarly, error rates did not distinguish ALL survivors from healthy controls (main effect of intensity deviant type on FAR: F_1,14=6.68, p=0.02, η_p²=0.32), with no main effect of disease status (F_1,14<1, p=0.67), and no interaction between intensity deviant type and disease status on FAR (F_1,14<1, p=0.83). The mean FAR was less than 5% across all participants.

Figure 1. Behavioral results.

Individual data points are plotted for ALL survivors (left column, grey circles) and healthy control participants (right column, grey triangles). The mean of all participants in each group is represented with a black mark. A) Hit rate was calculated as the number of correct button presses divided by the total number of targets. The group mean hit rate was significantly lower for the ALL group, and there was greater variability among participants compared to the control group. B) Reaction time was calculated by the mean reaction time (in milliseconds) of all of the correct responses to target stimuli. There was no group mean difference in reaction time but there was greater variability among the ALL group compared to healthy controls.

Table 2.

Behavioral response to attended tasks.

Condition	Hit Rate		Reaction Time (in milliseconds)		False Alarm rate
	Survivors	Control	Survivors	Control	Survivors	Control
Att-Loud	0.68 (0.27)	0.86 (0.10)	381 (110)	365 (70)	0.018 (0.010)	0.022 (0.038)
Att-Soft	0.47 (0.30)	0.64 (0.14)	454 (106)	462 (70)	0.027 (0.010)	0.033 (0.029)

Att=attend. Standard deviation is in parentheses.

Table 3.

Behavioral results for individual ALL survivors, with age-matched controls displayed in the same row.

Subject	Age	Hit Rate		Reaction Time (in milliseconds)
		ALL Survivors	Matched Control	ALL Survivors	Matched Control
1	9	0.89	0.88	269	377
2	11	0.72	0.82	355	322
3	11	0.82	0.83	364	349
4	7	0.19	0.84	492	434
5	7	0.79	0.69	344	470
6	6	0.42	0.70	525	449
7	6	0.54	0.58	415	395
8	5	0.25	0.65	578	511

Reaction time was significantly longer to the softer than to the louder intensity deviants (main effect of deviant type on RT: F_1,14=38.88, p<0.01, η_p²=0.74), and no main effect of disease status (F_1,14<1, p=0.92), and no interaction between deviant type and disease status (F_1,14<1, p=0.41). Individual differences in performance were noted within the ALL group. Three of the eight children (subjects 4, 6, and 8) performed particularly poorly on the identification tasks compared to their age-matched controls. The other ALL survivors performed comparably as well as the healthy controls.

Sensory Processing

The P1 component elicited by the standard sounds in each condition is displayed in Figure 2, with a summary of the latencies and amplitudes provided in Tables 4 and 5. The intensity of the stimulus did not affect the amplitude of the P1 (no main effect of condition, F_1,14=2.98, p=0.11, η_p²=0.16), and there was no interaction between intensity of the standard (loud/soft) and disease status, F_1,14=1.52, p=0.24, η_p²=0.06). There was a main effect of electrode (F_2,28=81.18, p<0.01, η_p²=0.61, ε=0.68), which post-hoc calculations showed there was typical scalp distribution of the P1 component, with amplitude largest at Fz (Fz>Cz>Pz). The P1 amplitude was larger in the healthy control group (main effect of disease status: F_1,14=5.54, p=0.03, η_p²=0.28). This smaller P1amplitude in ALL survivors is suggestive of a deficiency in sensory processing.

Figure 2. P1 component.

The P1 component evoked by standard tones is shown at the Fz electrode in the Attended conditions (solid line) and the Ignore conditions (dashed line) for all conditions in the ALL survivors (left column) and healthy controls (right column). The standard tone in the Frequency deviant condition (top row) had a frequency of 880 Hz and an intensity of 50.5 dBA. The standard in the Loud intensity deviant condition (middle row) had a frequency of 880 Hz and an intensity of 50.5 dBA. The standard in the Soft intensity deviant condition (bottom row) had a frequency of 880 Hz and intensity of 65.5 dBA. The response amplitude is plotted (in μV) along the y-axis and time (in milliseconds) along the x-axis. Epochs displayed are 50 ms pre-stimulus and 250 ms post-stimulus onset. The gray bar denotes the area of measurement around the peak of the P1.

Table 4.

Mean peak latency (in milliseconds) for the ERP components.

Condition	P1		MMN		P3a		P3b
	ALL Survivors	Healthy Controls	ALL Survivors	Healthy Controls	ALL Survivors	Healthy Controls	ALL Survivors	Healthy Controls
Att-Loud	92 (14)	91 (14)	137 (27)	166 (31)	-----	-----	270 (26)	260 (20)
Att-Soft	106 (14)	99 (12)	125 (12)	119 (13)	-----	-----	400 (23)	466 (20)
Ign-Freq	100 (13)	106 (14)	98 (23)	98 (23)	1.61 (1.99)	4.47 (2.81)	-----	-----
Ign-Loud	105 (8)	106 (8)	130 (20)	166 (29)	−0.23 (2.70)	1.71 (2.06)	-----	-----
Ign-Soft	100 (14)	105 (14)	165 (9)	141 (15)	1.21 (1.67)	−0.61 (1.49)	-----	-----

Att=attend; Ign=ignore; Freq=frequency. Standard deviation is in parentheses.

Table 5.

Mean amplitude (in μV) for P1 component elicited by the standard tones at midline electrodes.

Condition	ALL Survivors			Healthy Controls
	Fz	Cz	Pz	Fz	Cz	Pz
Attn-Loud	3.74 (2.15)	3.05 (2.05)	0.66 (0.94)	7.08 (2.09)	5.91 (1.97)	2.47 (1.40)
Attn-Soft	4.14 (1.84)	3.78 (2.06)	0.69 (1.38)	6.84 (2.16)	5.80 (1.96)	2.35 (0.96)
Ign-Freq	4.15 (1.87)	4.07 (1.96)	1.02 (1.02)	5.93 (1.66)	5.40 (1.45)	2.31 (0.89)
Ign-Loud	3.97 (1.94)	3.78 (2.23)	1.09 (1.24)	5.09 (1.67)	4.92 (1.32)	1.81 (1.07)
Ign-Soft	4.50 (1.55)	4.82 (1.74)	1.64 (0.72)	5.65 (1.83)	5.24 (1.36)	1.94 (0.96)

Att=attend; Ign=ignore; Freq=frequency. Standard deviation in parentheses.

Overall, P1 amplitude did not differ as a function of attention (no main effect of attention, F_1,14=1.02, p=0.33, η_p²=0.07). However, there was a significant interaction between disease status and attention on P1 amplitude (F_1,14=22.06, p<0.01, η_p²=0.61). The interaction was due to a larger P1 amplitude in the active compared to passive listening conditions in the control group but not in the ALL group. There was no difference in P1 amplitude between active and passive listening conditions in the ALL group. This is an important result because the absence of an attention effect on P1 amplitude indicates a poorer attention functions in the ALL group. There was no interaction between attention and intensity (F_1,14=1.91, p=0.19, η_p²=0.12), and no interaction between attention, intensity, and disease status (F_1,14<1, p=0.78).

A separate one-way ANOVA was performed to assess the P1 amplitude in the Ign-Freq condition to compare scalp topography (within-subject factor of electrode) and groups (between-subjects factor of disease status). The ALL survivors had a smaller P1 amplitude than controls (main effect of disease status F_1,14=4.08, p<0.05, η_p²=0.25). A main effect of electrode (F_2,28=68.53, p=<0.01, η_p²=0.830, ε=0.70) showed typical P1 scalp distribution for children overall, with post hoc calculations showing largest amplitude at the Fz electrode, consistent with previous work (Sussman et al., 2008). There was no interaction between disease status and electrode (F_2,28<1, p=0.70), suggesting no difference in P1 topography on the basis of disease status.

P1 latency is shown in Table 4. There was no main effect of P1 latency by disease status (F_1,14<1, p=0.91) or condition (F_1,14=1.77, p=0.21, η_p²=0.11). There was a main effect of attention (F_1,14=9.11, p<0.01, η_p²=0.38). P1 peak latency was shorter when elicited by attended compared to unattended intensity deviants. An interaction between attention and condition (loud/soft) (F_1,14=5.61, p=0.03, η_p²=0.29) showed a shorter latency for the louder intensity deviants when they were attended. There was no interaction between attention and disease status (F_1,14=2.14, p=0.17, η_p²=0.13), between task and disease status (F_1,14<1, p=0.81), or between attention, task, and disease status (F_1,14<1, p=0.45).

Auditory Memory

Intensity oddball conditions.

The MMN component is displayed in Figure 3. Latency and amplitude measurements are summarized in Tables 4 and 6.. There was a main effect of intensity deviant type (F_1,14=23.24, p<0.01, η_p²=0.62), in which MMN amplitude was larger to the louder intensity deviants. There was an interaction between stimulus type and group (F_1,14=5.43, p<0.05, η_p²=0.28). Post hoc analysis showed that the control children had a larger amplitude difference between deviant and standard tones but not ALL survivors, suggesting that MMN was elicited in all conditions in the controls but not in the ALL survivors. There was a main effect of electrode (F_1,28=7.89, p<0.01, η_p²=0.36, ε=0.82) with post hoc analysis showing largest amplitude at Fz, which is typical for MMN topography (Fz>Cz>Pz). There was a significant interaction between stimulus and electrode (F_2,28=17.95, p<0.01, η_p²=0.56, ε=0.75), with post hoc calculations showing that the amplitude difference between standard and deviant was largest at the Fz electrode, which is typical for MMN scalp topography. There was a significant three-way interaction between stimulus type, attention, and electrode (F_2,28=11.45, p<0.01, η_p²=0.45). Post hoc analysis demonstrated that the MMN amplitude difference was largest in attended conditions and at Fz. There was also a three-way interaction between stimulus, intensity deviant type, and electrode (F_2,28=11.04, p<0.01, η_p²=0.44, ε=0.91), with post hoc showing that MMN amplitude was larger to the louder than to the softer deviants at Fz. This is consistent with greater difficulty for detecting a quieter sound. There was a trend toward MMN amplitude being larger overall for controls than ALL, however, that effect did not reach significance (no main effect of disease status (F_1,14=3.96, p=0.07, η_p²=0.22). There was no main effect of stimulus type (F_1,14=1.76, p=0.21, η_p²=0.11), and no main effect of attention (F_1,14=1.38, p=0.26, η_p²=0.09). There were no interactions between attention and group (F_1,14<1, p=0.36,), stimulus and intensity deviant type (F_1,14<1, p=0.38), attention and intensity deviant type (F_1,14<1, p=0.34), attention and electrode (F_2,28=2.98, p=0.07, η_p²=0.18), or deviant type and electrode (F_2,28=2.25, p=0.12, η_p²=0.14).

Figure 3. Mismatch negativity (MMN) component.

Event-related potentials (ERPs, left two columns) evoked by standard tones (gray, thin solid line) and deviant tones (black, thick solid line) are overlain for the Ignore (top three panels) and the Attend (bottom two panels) conditions, for the three deviant types (Frequency, Intensity-Loud, and Intensity-Soft) in both groups (ALL survivors left column, healthy controls right column). The response amplitude is plotted (in μV) along the y-axis and time (in milliseconds) along the x-axis. The peak P1 response is indicated in the ERPs with an arrow (left column). The difference between the standard and deviant ERP responses (deviant-minus-standard) delineates the MMN component. MMN is displayed in the difference waveforms (right two columns) at Fz (solid, black lines) and at LM (dashed, black lines) electrodes. Epochs displayed are 50 ms pre-stimulus and 250 ms post-stimulus onset. Significant MMN components are labeled and denoted with an arrow, along with a voltage map displaying the scalp distribution at the peak of the MMN response, with blue for negative polarity and red for positive polarity.

Table 6.

Mean amplitude (in μV) for the mismatch negativity (MMN) component at midline electrodes.

Condition	Stimulus	ALL Survivors			Healthy Controls
		Fz	Cz	Pz	Fz	Cz	Pz
Att-Loud	Dev	0.64 (2.30)	2.73 (4.17)	3.07 (3.59)	−2.20 (4.13)	2.83 (4.60)	2.68 (4.76)
	Std	3.04 (2.21)	2.88 (2.09)	0.78 (1.03)	2.90 (1.49)	2.63 (1.47)	1.05 (1.27)
	Δ	−2.41 (3.00)	−0.14 (3.22)	2.29 (2.77)	−5.10 (3.60)	0.19 (3.82)	1.63 (4.37)
Att-Soft	Dev	1.03 (2.28)	1.84 (2.43)	1.61 (2.73)	−3.15 (3.07)	−1.34 (4.67)	−1.05 (2.88)
	Std	0.86 (0.80)	0.86 (0.75)	0.07 (1.19)	0.12 (1.21)	−0.07 (1.57)	−0.11 (1.32)
	Δ	0.18 (2.25)	0.99 (2.28)	1.53 (2.45)	−3.27 (3.25)	−1.26 (5.41)	−0.94 (3.70)
Ign-Freq	Dev	1.40 (1.78)	1.80 (2.60)	−0.23 (2.03)	3.70 (1.85)	3.69 (1.89)	1.35 (2.01)
	Std	4.20 (1.91)	4.17 (1.99)	1.09 (1.06)	6.02 (1.55)	5.61 (1.39)	2.52 (0.95)
	Δ	−2.80 (2.44)	−2.37 (2.52)	−1.32 (2.62)	−2.32 (2.31)	−1.92 (2.45)	−1.17 (1.87)
Ign-Loud	Dev	2.27 (2.51)	3.78 (3.19)	2.74 (2.72)	1.58 (2.06)	3.75 (2.81)	2.55 (2.31)
	Std	3.11 (1.45)	3.04 (1.81)	0.76 (1.31)	2.65 (1.06)	3.30 (1.13)	1.70 (0.77)
	Δ	−0.85 (1.79)	0.74 (2.61)	1.97 (2.07)	−1.07 (2.04)	0.45 (2.38)	0.85 (2.30)
Ign-Soft	Dev	1.02 (1.03)	1.24 (0.98)	−0.21 (1.98)	−1.97 (3.85)	−0.93 (2.06)	−0.97 (1.24)
	Std	0.61 (0.90)	1.33 (1.27)	0.58 (1.03)	0.38 (0.98)	0.80 (0.73)	0.42 (0.84)
	Δ	0.41 (1.78)	−0.09 (1.95)	−0.79 (2.41)	−2.35 (3.44)	−1.74 (2.01)	−1.39 (1.33)

Att=attend; Ign=ignore; Freq=frequency. Std=Standard, Dev=Deviant, Δ=standard-minus-deviant. Standard deviation is in parentheses.

MMN latency is summarized in Table 6. There was no main effect of disease status on the latency of the component (F_1,14=2.33, p=0.15, η_p²=0.14) or attention (F_1,14<1, p=0.35), but there was a significant main effect of intensity deviant type (F_1,14=133.22, p<0.01, η_p²=0.91). The mean peak of the MMN component was later for the soft deviant. This indicates greater difficulty to respond and is consistent with a longer reaction time in the softer oddball task condition. There was a significant interaction between intensity deviant type and group (F_1,14=16.31, p<0.01, η_p²=0.54). The ALL survivors had a shorter MMN latency to the loud intensity deviant in both ignore and attend conditions.

Frequency oddball condition.

There was a significant main effect of disease status (F_1,14=9.20, p<0.01, η_p²=0.40), showing that the overall amplitude of the ERPs were larger for the controls. There was a main effect of stimulus type (F_1,14=13.33, p<0.01, η_p²=0.49), showing that the deviant response was more negative than the standard response; denoting the presence of MMN. There was a main effect of electrode (F_2,28=48.86, p<0.01, η_p²=0.78, ε=0.90). Post hoc calculations showed that the difference between the standard and deviant amplitude was largest at Fz, which is typical scalp distribution for auditory evoked MMN (Fz>Cz >Pz). There was no interaction between disease status and stimulus type for frequency (F_1,14<1, p=0.75), showing no group difference for the frequency MMN amplitude. There was no mean latency difference in the frequency MMN as calculated by an independent Student’s t-test comparing the ALL and control children (p=0.97).

Orienting

The P3a component is displayed in Figure 4, and latencies and amplitude are summarized in Tables 4 and 7. The P3a was quantified only in passive listening conditions because the P3a indexes an involuntary orienting response to the sounds. Thus, while attention was directed toward watching the movie, we can observe attentional capture by salient sounds. P3a was elicited. There was a main effect of stimulus type (F_1,14=9.60, p<0.01, η_p²=0.41), showing that overall, the amplitude of the ERP response to the deviant tones was larger than that to the standard tones. A main effect of deviant type (F_2,28=14.96, p<0.01, η_p²=0.52, ε=0.94) showed that frequency deviants elicited the largest P3a (Fig 4, Ign-Freq condition, top row). This is consistent with our previous findings showing greater salience of frequency deviants than intensity deviants when sounds are in the background and children are watching a movie and passively listening to the sounds (Sussman & Steinschneider, 2011). A main effect of electrode (F_2,28=13.25, p<0.01, η_p²=0.49, ε=0.834) showed that the P3a component was had the expected scalp distribution, with its largest amplitude at the Cz electrode (Cz>Fz>Pz). There was a significant three-way interaction between stimulus, deviant type, and group (F_2,28=5.42, p<0.01, η_p²=0.28, ε=0.93), with post hoc calculations showing that the P3a amplitude was larger in control children than ALL survivors only in the Ign-Freq condition. The P3a was elicited by the loud deviant only in the control group. There was no P3a elicited by soft deviants in either group. Thus, P3a was not elicited by intensity deviants soft or loud in the ALL group. There was a no main effect of disease status (F_1,14<1, p=0.38).

Figure 4. P3a component.

Difference waveforms (deviant-minus-standard) are displayed at the Cz electrode, showing the P3a response elicited by deviant tones in the Ignore conditions (frequency deviant, top row; loud intensity deviant, middle row; and soft intensity deviant, bottom row), for ALL survivors (thick, solid black lines) and healthy controls (thin, gray solid lines). The response amplitude is plotted (in μV) along the y-axis and time (in milliseconds) along the x-axis. Epochs displayed are 100 ms pre-stimulus and 500 ms post-stimulus onset. The arrows denote a significant P3a component.

Table 7.

Mean amplitude (in μV) for the P3a component at midline electrodes.

Condition	Stimulus	ALL Survivors			Healthy Controls
		Fz	Cz	Pz	Fz	Cz	Pz
Ign-Freq	Dev	0.30 (1.60)	2.54 (2.01)	1.30 (2.04)	2.80 (2.72)	4.69 (2.41)	3.45 (2.24)
	Std	0.75 (0.74)	0.93 (0.87)	0.32 (0.59)	0.22 (1.05)	0.22 (0.78)	−0.01 (0.65)
	Δ	−0.45 (1.62)	1.61 (1.99)	0.97 (1.83)	2.58 (3.07)	4.47 (2.81)	3.46 (2.59)
Ign-Loud	Dev	−1.73 (1.38)	−0.22 (1.55)	0.04 (1.93)	−0.19 (3.45)	1.64 (2.22)	1.52 (1.37)
	Std	−0.23 (0.74)	−0.23 (0.69)	−0.18 (0.61)	−0.09 (0.72)	−0.07 (0.64)	0.15 (0.61)
	Δ	−1.50 (1.30)	0.01 (1.89)	0.22 (2.34)	−0.11 (3.33)	1.71 (2.06)	1.37 (1.20)
Ign-Soft	Dev	0.29 (1.23)	1.22 (1.45)	0.41 (0.79)	−2.49 (2.92)	−1.23 (1.64)	−0.08 (1.80)
	Std	0.05 (0.52)	−0.18 (0.63)	0.09 (0.71)	−0.34 (0.41)	−0.62 (0.50)	−0.73 (0.49)
	Δ	0.25 (1.63)	1.40 (1.94)	0.32 (1.21)	−2.15 (2.87)	−0.61 (1.49)	0.65 (1.94

Std=Standard, Dev=Deviant, Δ=standard-minus-deviant. Ign=ignore. Standard deviation is in parentheses.

P3a mean peak latency was not modulated by any factors: no main effect of disease status (F_1,14=1.35, p=0.26, η_p²=0.09), or condition (F_1,14<1, p=0.68), and no interaction between disease status and condition on peak latency (F_1,14=3.07, p=0.10, η_p²=0.18).

Attention

The target detection P3b component is shown in Figure 5, with latency and amplitude summarized in Tables 4 and 8.

Figure 5. P3b component.

Difference waveforms (deviant-minus-standard) are displayed at the Pz electrode, showing the P3b response elicited by deviant tones in the Attend conditions (loud intensity deviant, top row; soft intensity deviant, bottom row), for ALL survivors (thick, solid black lines) and healthy controls (thin, gray solid lines). The response amplitude is plotted (in μV) along the y-axis and time (in milliseconds) along the x-axis. Epochs displayed are 100 ms pre-stimulus and 500 ms post-stimulus onset. The arrows denote significant P3b components.

Table 8.

Mean amplitude (in μV) for the P3b component at midline electrodes.

Condition	Stimulus	ALL Survivors			Healthy Controls
		Fz	Cz	Pz	Fz	Cz	Pz
Att-Loud	Dev	2.04 (5.12)	3.98 (9.23)	4.03 (8.19)	1.43 (5.12)	6.72 (9.23)	8.56 (8.19)
	Std	3.50 (0.61)	3.04 (0.86)	0.44 (1.13)	0.57 (0.61)	−0.31 (0.86)	−0.61 (1.13)
	Δ	−1.46 (1.60)	0.94 (2.39)	3.59 (2.53)	0.86 (5.08)	7.03 (9.27)	9.16 (8.03)
Att-Soft	Dev	3.14 (4.07)	4.98 (5.24)	4.59 (4.19)	3.27 (4.07)	7.36 (5.24)	6.25 (4.19)
	Std	0.53 (2.03)	0.20 (2.55)	−0.38 (1.58)	5.63 (2.03)	4.25 (2.55)	1.61 (1.58)
	Δ	2.61 (2.10)	4.79 (2.98)	4.98 (2.82)	−2.36 (3.00)	3.11 (5.64)	4.64 (3.81)

Std=Standard, Dev=Deviant, Δ=standard-minus-deviant. Att=attend. Standard deviation is in parentheses.

P3b amplitude.

P3b was elicited for all deviant types in both groups (main effect of stimulus type: F_1,14=12.38, p<0.01, η_p²=0.47). Overall, the amplitude elicited by the deviant tones was larger (more positive amplitude) than the amplitude to the standard tones. There was a main effect of electrode (F_2,28=3.61, p<0.05, η_p²=0.21, ε=0.86), showing typical scalp voltage distribution for the P3b, with maximal amplitude at the Pz electrode (Table 8). A significant three-way interaction between task, electrode, and disease status (F_2,28=4.04, p<0.05, η_p²=0.22, ε=0.74) showed that the P3b amplitude was larger in the control group than the ALL survivors in the Att-loud condition (Fig 5, top row). There was no main effect of disease status (F_1,14=1.33, p=0.27, η_p²=0.09) and no main effect of intensity deviant type (F_1,14<1, p=0.34).

P3b latency was modulated by disease status and deviant type. There was a main effect of disease status (F_1,14=11.88, p<0.01, η_p²=0.46). P3b latency shorter, overall, in the ALL group. There was a main effect of deviant type (F_1,14=484.96, p<0.01, η_p²-0.97). The mean P3b peak latency was longer when elicited by the soft intensity deviants compared to the loud intensity deviants. The longer mean P3b latency is consistent with the longer mean reaction time to the softer deviants (458 ms) than the louder deviants (373 ms), and indicates that detecting the softer deviant was a more difficult task. Figure 5 shows the broader, more sustained P3b peak to the softer deviants suggesting more individual variability in the harder task, compared to the earlier and sharper peak for the louder intensity deviant. There was a significant interaction between disease status and deviant type (F_1,14=24.73, p<0.01, η_p²=0.64). Post hoc analyses showed that the P3b component had a shorter mean peak latency for ALL survivors in the attended soft condition.

Discussion

Childhood ALL survivors differed significantly from healthy controls in distinct components of the ERPs, including decreased amplitude of the P1, MMN, P3a, and P3b components and absence of amplitude gain in the P1 component typically observed when attending to the sounds. These differences suggest that treatment for leukemia induces changes in brain processes associated with sensory discrimination, auditory working memory, and attentional. Behavioral measures of sound discrimination did not distinguish ALL survivors from controls on the whole (by mean HR and RT). However, younger ALL survivors in the current sample were less accurate and took longer to respond than their age-matched peers. In contrast, brain responses in the ALL survivors showed considerable differences in sound discrimination processing overall in both in passive and active listening. These results indicate dysfunction in several different cognitive domains in the ALL survivors through neurophysiological measures of auditory discrimination, and are described in more detail below.

Behavior

There was no mean difference in HR, RT, or FAR between ALL survivors and control children. However, there were individual differences among the ALL survivors (Figure 1, Table 3). Three ALL survivors (subjects 4, 6, and 8) were inaccurate at pressing the response key to target deviants, but the performance of the other five ALL survivors were comparable to the control children. It is interesting to note that the individual who received cranial radiation was not one of the poor performers. 30–60% of patients demonstrate post-chemotherapy cognitive impairments, which is consistent with the current results, in which three out of eight survivors had poor performance (Buizer, de Sonneville, & Veerman, 2009). The three survivors with the worst behavioral results were also younger and female, consistent with the observation that being younger and female is a risk factor for developing cognitive decline after chemotherapy (Buizer et al., 2009). The poor performing participants in our study were tested at a later time after treatment than previously reported (e.g., Buizer et al.), suggesting long-term lasting effects of chemotherapy treatment. The younger ALL survivors had poorer performance (both longer RTs and lower HR) than their age-matched counterparts, suggesting that poorer performance was due to treatment and not age. However, because of the small number of participants used, it was not possible to stratify the groups for statistical analysis of the younger ages or by sex.

Sensory Processing

While the P1 latency was age-appropriate for all participants, the P1 amplitude was reduced in ALL survivors. In addition P1 typically reflects an increase in sensory gain when sounds are attended (Hillyard, Hink, Schewent, & Picton, 1973). This was absent in the ALL group (Fig 2). Within-subjects comparison of the passive and actively elicited P1 component in the healthy control group showed a 1.17 μV difference in mean P1 amplitude (Table 5). This amplitude difference is an expected increase due to attentional gain of the ERP component. In contrast, in the ALL survivors, no such amplitude gain with attention was observed for the active conditions: there was no significant amplitude difference between passive and active conditions. This is an important result because the absence of an attention effect on P1 amplitude indicates a poorer attention functions in the ALL group. This may be due to the fidelity of the information the brain is receiving is reduced compared to that of healthy children or that attentional ability is impaired. Degraded sensory information could have cascading effects on auditory perception and could explain a wide range of cognitive deficits seen in post-chemotherapy children. This is the first evidence to our knowledge of an effect of attentional gain control at the level of P1 potentially due to neurotoxic effects of treatment.

Auditory Memory

The MMN amplitude was smaller in ALL survivors. The reduction in MMN amplitude in the ALL survivors was consistent with previous ERP studies in ALL survivors (Järvelä et al., 2011). Smaller MMN amplitude suggests impairment in working memory of survivors (Harila, Winqvist, Lanning, Bloigu, & Harila-Saari, 2009; Jansen et al., 2008; Kingma et al., 2001, 2002; Lofstad, Reinfjell, Hestad, & Diseth, 2009; Raymond-Speden, 2000; Rodgers, Marckus, Kearns, & Windebank, 2003). One proposed mechanism of methotrexate neurotoxicity is through increased glutamate release and over-activation of NMDA receptors (Vijayanathan et al., 2011). Thus, the smaller MMN amplitude may indicate NMDA dysfunction induced by methotrexate.

A remarkable result was the absence of MMN in the ALL survivors to both attend-soft and ignore-soft deviants. We and others have previously shown that the auditory system undergoes significant refinement and development throughout childhood and into adolescence (Krizman et al., 2015; Litovsky, 2015; Mcmahon, Wintermark, & Lahav, 2012; Sussman & Steinschneider, 2011; Sussman & Steinschneider, 2009; Werner, Marean, Halpin, Spetner, & Gillenwater, 1992). We have shown that the ability to automatically discriminate frequency deviants is fairly robust from a very young age, whereas the ability to automatically detect intensity deviants is less robust in children under 11 years of age (Sussman & Steinschneider, 2011). Our findings thus suggest that the deviance detection system is immature in ALL survivors compared to age expectations in healthy children. The current dataset cannot determine if this difference is a developmental delay in memory functions that will resolve over time or if the auditory discrimination process is chronically damaged and will persist through adulthood.

In addition to memory-related sound discrimination differences between groups, the MMN also showed attention effects between groups. ALL children had smaller amplitude MMNs overall, whereas healthy control children had larger MMN amplitude when attending the sounds compared to when ignoring the sounds. The attention effect may be inherited from lower sensory processing areas in the brain, as the P1 component also showed no gain in sensory processing as would have been expected with attention to the stimuli.

Orienting

The P3a amplitude elicited by the frequency and intensity deviants during passive listening was larger in the healthy control children than the ALL children. This suggests that healthy children had more robust attentional orienting responses to irrelevant, unexpected sounds than ALL children. One possibility is that the sensory input had higher fidelity in the healthy controls, having more salience and evoking a larger involuntary orienting response. For the frequency deviants, there was a paradigmatic difference; the frequency separation between deviant and standard was larger in the control group than the ALL group. However, the loudness deviants had exactly the same magnitude in ALL and control groups, and the P3a amplitude was still larger in the control children. Thus, it is unlikely that the stimulus difference was the sole contributor to the magnitude difference of the orienting response in controls compared to ALL. Considering that the P3a response reflects attentional capture by a salient stimulus, it seems that impaired ability to properly orient to environmental stimuli may be part of the cognitive profile in ALL. This finding is consistent with previous ERP studies of ALL survivors (Järvelä et al., 2011).

Attention

The P3b reflects ability to perform a sound discrimination task. The P3b amplitude was smaller in the ALL children, indicating greater difficulty with the task overall. Smaller amplitude P3b amplitude would be consistent with increased difficulty in performing the task, which is consistent with previous studies assessing attention in ALL survivors (Kingma et al., 2001; Lofstad et al., 2009; Rodgers et al., 2003). It is not clear from these data if the attention deficit is inherited from difficulties in lower sensory areas, indicated by deficits in P1 and MMN, or is more localized to attention-directed activity, such as vigilance or target detection.

Clinical Significance

Using an animal model of chemotherapy-induced cognitive deficits, we are beginning to identify pharmacologic interventions that can prevent deficits when administered concurrently with chemotherapy (Vijayanathan et al., 2011). However, the selection of a specific protective strategy to test in the clinic depends on a more complete delineation of the key components of pathophysiology as well as of the primary loci responsible for functional deficits. These preliminary data demonstrate the utility of ERP analysis in directing the selection of a therapeutic strategy. The MMN component, for example, is dependent on open N-methyl-d-aspartate (NMDA) channels. Differences in elicitation of the MMN component between cancer survivors and controls are consistent with altered neurotransmission through NMDA receptors. This in turn would suggest that an NMDA modulating agent such as dextromethorphan or memantine may benefit childhood leukemia survivors, ameliorating the cognitive deficits induced by therapy. These NMDA antagonists have shown promising effects among adults treated with cranial radiation for brain tumors(Brown et al., 2013). These results, therefore, support a larger investigation of children with ALL to confirm the differences observed here and better define the relative contributions of sensory processes, memory and attentional mechanisms in order to differentiate the level at which processing deficits occur.

Limitations

A caveat of this study is that our patient group was somewhat heterogeneous, having been treated on a variety of treatment protocols, including one participant who received cranial radiotherapy. However, for this preliminary investigation, we intentionally chose to study patients reflecting the diversity of childhood leukemia survivors. In addition, secondary analyses were limited due to the small sample size. Although we were sufficiently powered to detect differences within and across groups, we consider the data and the interpretations to be preliminary and therefore are unable to draw conclusions regarding risk factors for cognitive dysfunction, including treatment-related factors (e.g., use of radiation, dose intensity) or patient related factors (e.g., age, sex, race or ethnicity, genetic polymorphisms)(Cole et al., 2015; Cole & Kamen, 2006). Finally, this study was limited by the lack of a baseline assessment, prior to the start of leukemia treatment. Such a baseline evaluation would be necessary to conclusively demonstrate that the observed differences between survivors and controls were due to chemotherapy exposure, pathology induced by leukemia itself, the experience of prolonged hospitalization, or repeated exposure to anesthetic agents, and not to inherent differences between these individual subjects. However, children are typically acutely ill when they present with acute lymphoblastic leukemia. Therapy is initiated immediately to prevent deterioration or death, making it impractical to conduct a reliable baseline assessment.

Conclusions

Post-treatment sensory and cognitive deficiencies were identified in childhood leukemia survivors through behavioral and electrophysiological indices of sensory, working memory, and attentional processes via a sound discrimination task. Brain activity associated with cognitive functioning was less robust in ALL survivors compared to healthy, age-matched controls. This was demonstrated by smaller amplitude and longer latency ERP components (P1, MMN, P3a, and P3b) in all cognitive domains. Atypical brain responses were consistent with impaired behavioral performance. Younger ALL survivors were less accurate than healthy age-matched controls on the auditory detection tasks. Taking into account the results from all analyzed ERP components, there are impairments at all measured levels of processing, including basic sensory and memory processes, as well as attention networks involved in orienting and attentional control. Overall, our results are consistent with the varied set of data on childhood ALL that show post-treatment deficits in various domains of cognitive functions obtained through psychometric testing (Williams, Zent, & Janelsins, 2016). ERPs distinguished the type of cognitive deficit, and thus may provide a useful non-invasive tool to assess recovery and risk of cognitive dysfunction in children who have undergone chemotherapy treatment.