An Empirical Mediation Analysis of Mechanisms Underlying HIV-1-Associated Neurocognitive Disorders

Abstract

HIV-1-associated neurocognitive disorders (HAND), characterized by alterations in the core components of cognitive function and age-related disease progression, persist in the post-cART era. However, the neurobehavioral mechanisms that mediate alterations in the core components of cognitive function and the progression of neurocognitive impairments have yet to be systematically evaluated. To address this knowledge gap, statistical mediation analysis was assessed, providing a critical opportunity to empirically evaluate putative neurobehavioral mechanisms underlying HAND. Neurocognitive assessments, conducted in HIV-1 transgenic (Tg) and control animals across the functional lifespan (i.e., Postnatal Day (PD) 30 to PD 600), tapped multiple cognitive domains including preattentive processes, learning, sustained attention, and long-term episodic memory. Three longitudinal mediation models were utilized to assess whether deficits in preattentive processes mediate alterations in learning, sustained attention and/or long-term episodic memory over time. Preattentive processes partially mediated the relationship between genotype and learning, genotype and sustained attention, and genotype and long-term episodic memory across the functional lifespan, explaining between 44% and 58% of the HIV-1 transgene effect. Understanding the neurobehavioral mechanisms mediating alterations in HAND may provide key targets for the development of a diagnostic biomarker, novel therapeutics, and cure/restoration strategies.

1. Introduction

The prevalence of neurodegenerative diseases is rapidly increasing, largely due to the increasing population of elderly individuals (Heemels, 2016). Most broadly, neurodegenerative diseases are characterized by deterioration of neurons in the brain and spinal cord; deterioration which leads to progressive alterations in neurocognitive function (Katsuno et al., 2018). Diseases such as Alzheimer’s disease, Parkinson’s disease, and multiple sclerosis are some of the most commonly identified neurodegenerative diseases. However, emerging evidence suggests that HIV-1-associated neurocognitive disorders (HAND) may also be considered a neurodegenerative disease (Cohen et al., 2015; McLaurin et al., 2019a).

The great success of combination antiretroviral therapy (cART) has dramatically increased the life expectancy for HIV-1 seropositive individuals, leading to an increased prevalence (i.e., 30–50%) of older individuals (> 50 years of age) living with HIV-1 (UNAIDS, 2014). While cross-sectional studies have been able to characterize the primary neurocognitive domains affected by HAND (e.g., speed of information processing, attention, working memory, executive function; Cysique et al., 2004; Heaton et al., 2011), longitudinal studies are vital for defining the progression of HAND. Examination of neurocognitive decline in HIV-1 seropositive individuals has revealed that a significant proportion (i.e., 13–30%) of individuals exhibit overall decline across time (e.g., 35 months: Heaton et al., 2015; 48 months: Sacktor et al., 2016; 4 years: Rubin et al., 2017; 18 months: Gott et al., 2017). Longitudinally, neurocognitive impairments in some domains (e.g., learning, attention, executive function) are persistent across time, whereas impairments in other domains (e.g., motor skills, memory) are progressive (Rubin et al., 2017; Maki et al., 2018). Furthermore, profound sex differences have been observed in neurocognitive function, with HIV-1 seropositive women exhibiting significantly greater impairments relative to HIV-1 seropositive men (Maki et al., 2018).

In a seminal manuscript, Reid et al. (2001) reported the derivation of the HIV-1 Tg rat, a biological system that has permitted the translational modeling of HAND in the post-cART era. The HIV-1 Tg rat contains a gag-pol-deleted HIV-1 provirus regulated by the viral promotor, rendering the HIV-1 Tg rat noninfectious. However, the expression of 7 of the 9 HIV-1 genes constitutively throughout development (Peng et al., 2010; Abbondanzo and Chang, 2014) resembles HIV1 seropositive individuals on cART. The contemporary HIV-1 Tg phenotype, on the F344/N background, exhibits similar growth rates to F344/N controls (Peng et al., 2010; Moran et al., 2012; Roscoe et al., 2014) and intact sensory (Peng et al., 2010; Bertrand et al., 2018; McLaurin et al., 2018a) and motor system function (McLaurin et al., 2018a) through advanced age. Furthermore, preclinical studies in the HIV-1 Tg rat have helped define the nature (Vigorito et al., 2007; Lashomb et al., 2009; Moran et al., 2014; Repunte-Canonigo et al., 2014; McLaurin et al., 2017a) and progression (McLaurin et al., 2018a; McLaurin et al., 2019a) of neurocognitive impairments in the post-cART era. Longitudinal experimental designs have revealed profound age-related disease progression across the functional lifespan (McLaurin et al., 2018a; McLaurin et al., 2019a), with sex-dependent expression of neurocognitive impairments observed at advanced ages (McLaurin et al., 2019a). Despite our improved understanding of the progression of HAND in both clinical and preclinical studies, the neurobehavioral mechanisms mediating alterations in the core components of cognitive function across the functional lifespan remain a critical knowledge gap.

Statistical mediation analysis examines how the effect of a third-variable conveys the influence of an independent variable onto a dependent variable. In psychology, one of the earliest mediation theories was the stimulus-organism-response model, outlined by Woodworth (1928), which described the translation of a stimulus into a behavior via mediating mechanisms such as mental processes (Baron and Kenny, 1986; MacKinnon, 2008; MacKinnon and Fairchild, 2009; Vanderweele, 2015; Fairchild and McDaniel, 2017). Neurocognitive functions, which are generally componential (Keeler and Robbins, 2011), may serve as one example of the stimulus-organism-response model (Fig. 1). Specifically, sensory input (i.e., stimulus) is transformed to motor output (i.e., response) via preattentive processes and core components of cognitive function endogenous to the organism. Despite the valuable information that statistical mediation can provide, these analyses remain underutilized in basic science and/or HIV-1 research, with the exception of a few recent manuscripts (e.g., Fazeli et al., 2017; Woods and Sullivan, 2019). The present study, therefore, addresses this critical knowledge gap by conducting a mediation for explanation (MacKinnon, 2008; Fairchild and MacKinnon, 2014; Fairchild and McDaniel, 2017) study to examine potential neurobehavioral mechanisms underlying HAND in the HIV-1 Tg rat.

Fig. 1.

The stimulus-organism response model was one of the earliest mediation theories in psychology (Woodworth, 1928). Neurocognitive functions, which are generally componential (Keeler and Robbins, 2011), may serve as one example of the stimulus-organism-response model. Specifically, sensory input (i.e., stimulus) is transformed to motor output (i.e., response) via preattentive processes and the core components of cognitive function, including attention, learning, and long-term episodic memory (i.e., organism).

Prepulse inhibition (PPI) of the auditory startle response (ASR) is a translational experimental paradigm that was introduced and popularized by Hoffman and Ison for the assessment of preattentive processing (e.g., Hoffman and Searle, 1965; Ison and Hammond, 1971). In PPI, a discrete prestimulus is presented 30–500 msec prior to a startling stimulus, producing a dramatic reduction in ASR (Hoffman and Ison, 1980). Alterations in preattentive processes due to experimental condition (e.g., presence or absence of the HIV-1 transgene) are delineated via the systematic manipulation of the interstimulus interval (ISI; i.e., the time interval between the discrete prepulse and startling stimulus), which affords an opportunity to infer temporal processing (McLaurin et al., 2019b). Most notably, HIV-1 seropositive individuals (Minassian et al., 2013) display prominent alterations in preattentive processes; deficits which have been translationally modeled across multiple biological systems used to study HAND (e.g., HIV-1 Tg rat: Moran et al., 2013a, McLaurin et al., 2017b, 2018a; stereotaxic injections of HIV-1 viral proteins: Fitting et al., 2006a,b; gp120 transgenic mice: Henry et al., 2014, Bachis et al., 2016; Tat transgenic mice: Paris et al., 2015) and progress across the functional lifespan (McLaurin et al., 2016; McLaurin et al., 2018a). Although deficits in preattentive processes have been implicated as a potential elemental dimension of HAND (e.g., Chao et al., 2004; Matas et al., 2010; Moran et al., 2013a), to date, no study has systematically evaluated the hypothesis that deficits in preattentive processing underlie the effect of the HIV-1 transgene on the core components of cognitive function (i.e., learning, sustained attention, long-term episodic memory) across the functional lifespan.

In preclinical studies, learning and sustained attention can be assessed using a signal detection operant task, developed and validated by McGaughy and Sarter (1995). During the task, animals are trained to attend to the presence or absence of a visual stimulus, indicating which response to make (i.e., which lever to press). Establishing a criterion (e.g., 70% accuracy for 5 consecutive or 7 non-consecutive days) affords an opportunity to evaluate the number of days required to learn the task. Furthermore, manipulating the stimulus duration provides an opportunity to assess sustained attention, which relies upon the detection of rare, unpredictable and/or weak stimuli over a long period of time (Sarter et al., 2001). Long-term episodic memory is commonly assessed using measures of novelty and habituation (e.g., Eacott et al., 2005; Barker et al., 2007; Moran et al., 2013b; Chao et al., 2016). Specifically, locomotor activity can be utilized to measure habituation, one of the simplest forms of learning characterized by a decrease in response following repeated stimulation (Harris, 1943; Thompson and Spencer, 1966; Rankin et al., 2009). Notably, alterations in learning (Moran et al., 2014; McLaurin et al., 2019a), sustained attention (Moran et al., 2014; McLaurin et al., 2019a), and long-term episodic memory (Moran et al., 2013b; McLaurin et al., 2018a) have been previously reported in the HIV-1 Tg rat; deficits which progress across the functional lifespan (McLaurin et al., 2018a; McLaurin et al., 2019a).

In light of previous work, the present study utilized statistical mediation analysis to assess whether preattentive processing was a neurobehavioral mechanism underlying alterations in the core components of cognitive function across the functional lifespan. To address this research question, three longitudinal mediation models, capturing the majority of an animal’s functional lifespan and multiple cognitive domains, were analyzed. The guiding hypothesis was that deficits in preattentive processes are a putative mediator of alterations in learning, sustained attention, and/or long-term episodic memory across the functional lifespan in the HIV-1 transgenic rat. Understanding the neurobehavioral mechanisms mediating alterations in HAND may provide key targets for the development of diagnostic screening tools, novel therapeutics, and cure/restoration strategies.

2. Results

2.1. Descriptive statistics

Outcome measurements for assessing the neurobehavioral mechanisms underlying HAND were derived from the measured functions associated with each neurocognitive outcome (Fig. 2). Specifically, PPI, a measure of preattentive processing, was derived by calculating the area of inflection of the ASR amplitude response curve (McLaurin et al., 2016; Fig. 2a). Learning was assessed by calculating the number of days to criteria in the signal detection operant task at each wave. Area of the signal detection operant curve, tapping sustained attention, was derived by calculating the area of the curve (AUC) where the number of hits was greater than the number of misses (Fig. 2b). Long-term episodic memory was assessed by calculating the plateau of the one-phase decay, a function which provided a well-described fit for the effect of genotype at all waves assessed (McLaurin et al., 2018a; Fig. 2c). Descriptive statistics for all outcome measurements, by genotype, are presented in Table 1.

Fig. 2.

Diagrams of the methodology used to derive preattentive processing, sustained attention, and long-term episodic memory. (A) Prepulse inhibition, a measure of preattentive processing, was derived by calculating the area of the peak inflection, shaded in orange. Adapted from McLaurin et al. (2016). (B) Sustained attention was derived from the signal detection curve by calculating the area under the hits and above the misses, shaded in orange. (C) Long-term episodic memory was assessed by calculating the plateau of the one-phase decay function.

Table 1.

Descriptive Statistics.

	Control	HIV-1 Transgenic
Variable	M(SD)	M(SD)
Preattentive Processing: Visual PPI
Wave 1	337641.15 (34296.97)	100596.81 (44911.78)
Wave 2	362117.62 (43094.53)	251855.40 (80250.64)
Wave 3	440690.53 (54478.14)	303503.55 (58576.28)
Wave 4	480067.82 (58286.31)	252881.21 (45157.80)
Wave 5	370160.95 (43178.24)	224591.86 (37122.96)
Learning: Days to Criteria
Wave 3	6.65 (0.48)	9.64 (0.93)
Wave 4	6.88 (0.45)	9.45 (0.90)
Wave 5	6.18 (0.23)	8.80 (0.86)
Sustained Attention: AUC
Wave 3	4094.82 (362.42)	3082.75 (308.92)
Wave 4	4183.47 (317.45)	2812.30 (365.45)
Wave 5	4491.50 (341.57)	3167.42 (406.48)
Long-Term Episodic Memory: Plateau
Wave 1	268.50 (14.70)	305.09 (16.48)
Wave 2	226.80 (15.77)	297.06 (22.74)
Wave 3	155.30 (22.87)	278.22 (22.98)
Wave 4	176.38 (16.91)	295.59 (21.18)
Wave 5	152.12 (11.94)	230.853 (21.46)

2.2. Longitudinal mediation model

Three lag-1 autoregressive mediation models were analyzed. A generalized figure reflecting the longitudinal models is illustrated in Fig. 3. The models examining visual PPI as a putative mediator underlying learning (i.e., days to criteria) and sustained attention (i.e., AUC) utilized three measurement waves of the mediator and outcome variables, beginning at wave 3, to reflect the point at which control animals reached asymptotic performance for these outcomes. The model examining visual PPI as a putative mediator underlying long-term episodic memory (i.e., one-phase decay plateau) utilized five measurement waves of the mediator and outcome variables, given continued changes in performance in both HIV-1 Tg and control animals across all waves for this outcome. The HIV-1 transgene was a timeinvariant predictor in all models, and animal sex was controlled for in the analyses.

Fig. 3.

Three lag-1 autoregressive longitudinal mediation models were examined. For all three models, the time-invariant independent variable was genotype (i.e., HIV-1 Tg vs. Control) and the mediator was preattentive processing, assessed using visual prepulse inhibition (PPI). The first model examined the outcome of stimulus-response learning in Waves 3, 4, and 5, and the second model examined the outcome of sustained attention in Waves 3, 4, and 5. The third model examined the outcome of long-term episodic memory using all five measurement waves of data. Solid lines in the figure represent paths that were estimated across all three models examined; dashed lines in the figure represent paths that were only estimated in the third model that included five data waves. Orange arrows in the diagram indicate component pathways of the mediation model, blue arrows represent the assessment of stability of the measures across time, and double-headed pink arrows indicate correlational relationships.

Model fit indicated tenability of assuming stationarity in the relationship between genotype and visual PPI, as well as in relationships between genotype and the varying outcome variables over time in all three mediation models. Thus, the lag-1 autoregressive paths associated with these relationships were constrained to equality across time points in each model. All other paths were freely estimated.

2.2.1. Model 1: Learning

Overall model fit statistics for the three-wave, lag-1 autoregressive mediation model predicting the effect of genotype on days to criteria through visual PPI indicated strong support for the proposed model, such that the observed covariance matrix of the data did not significantly differ from the model-implied covariance structure: χ²(12) = 14.061, p=.297; CFI = 0.987; SRMR = 0.063. The overall effect of genotype on days to criteria in the model was 2.842 (0.918), p = .002, 95% bias-corrected bootstrap CI [1.263, 4.842], demonstrating a sustained impact of the HIV-1 transgene on learning. The overall indirect effect of genotype on days to criteria through visual PPI was 1.254 (0.460), p = .006, 95% bias-corrected bootstrap CI [0.521, 2.438], indicating that preattentive processing represents a plausible mediating mechanism underlying the HIV-1 transgene’s influence on learning over time. The direct effect of the HIV-1 transgene on days to criteria once accounting for the mediator remained significant, 1.588 (0.565), p=.005, 95% bias-corrected bootstrap CI [0.625, 2.777], demonstrating that preattentive processing is a partial mediator and explains 44.1% of the observed HIV-1 transgene effect. All parameter estimates for the model are provided in Table 2.

Table 2.

Model Parameter Estimates: Learning (Days to Criteria) as outcome.

Variable	Estimate (SE)	p-value	95% Bootstrapped CI
Visual PPI (T5)
Visual PPI (T4)	0.164 (0.102)	0.109	[−0.016, 0.398]
Genotype	−1.443 (0.394)	< 0.001	[−2.202, −0.627]
Visual PPI (T4)
Visual PPI (T3)	0.164 (0.102)	0.109	[−0.016, 0.398]
Genotype	−1.443 (0.394)	< 0.001	[−2.202, −0.627]
Visual PPI (T3)
Genotype	−0.1.443 (0.394)	< 0.001	[−2.202, −0.627]
Learning (T5)
Learning (T4)	0.499 (0.116)	< 0.001	[0.292, 0.724]
Visual PPI (T4)	−0.104 (0.108)	0.334	[−0.360, 0.074]
Genotype	1.588 (0.565)	0.005	[0.625, 2.777]
Learning (T4)
Learning (T3)	0.499 (0.116)	< 0.001	[0.292, 0.724]
Visual PPI (T3)	0.152 (0.158)	0.336	[−0.099, 0.516]
Genotype	1.588 (0.565)	0.005	[0.625, 2.777]
Learning (T3)
Genotype	1.588 (0.565)	0.005	[0.625, 2.777]
Total Effect of Genotype on Learning Across Time	2.842 (0.918)	0.002	[1.263, 4.842]
Total Mediated Effect of Genotype on Learning through Visual PPI Across Time	1.254 (0.460)	0.006	[0.521, 2.438]

2.2.2. Model 2: Sustained attention

Overall model fit statistics for the three-wave, lag-1 autoregressive mediation model predicting the effect of genotype on sustained attention through visual PPI indicated moderate support for the proposed model, such that the observed covariance matrix of the data significantly differed from the model-implied covariance structure, but comparative fit indices showed the degree of misfit to be minor: χ²(12) = 23.442, p=.024; CFI = 0.948; SRMR = 0.054. The overall effect of genotype on sustained attention in the model was −13.879 (5.648), p = .014, 95% bias-corrected bootstrap CI [–23.448, −1.298], demonstrating a sustained negative impact of the HIV-1 transgene on sustained attention. The overall indirect effect of genotype on sustained attention through visual PPI was −8.005 (3.024), p = .008, 95% biascorrected bootstrap CI [−14.547, −1.294], indicating that preattentive processing represents a possible mediating mechanism underlying the HIV-1 transgene’s influence on sustained attention over time. The direct effect of the HIV-1 transgene on sustained attention once accounting for the mediator remained significant, −5.874 (2.813), p = .037, 95% bias-corrected bootstrap CI [−11.383, −0.280], demonstrating that preattentive processing is a partial mediator and explains 57.7% of the observed HIV-1 transgene effect. All parameter estimates for the model are provided in Table 3.

Table 3.

Model Parameter Estimates: Sustained Attention (AUC) as outcome.

Variable	Estimate (SE)	p-value	95% Bootstrapped CI
Visual PPI (T5)
Visual PPI (T4)	0.169 (0.102)	0.096	[−0.016, 0.398]
Genotype	−1.363 (0.408)	0.001	[−2.202, −0.627]
Visual PPI (T4)
Visual PPI (T3)	0.169 (0.102)	0.096	[−0.016, 0.398]
Genotype	−1.443 (0.394)	< 0.001	[−2.202, −0.627]
Visual PPI (T3)
Genotype	−0.1.363 (0.408)	0.001	[−2.202, −0.627]
Sustained Attention (T5)
Sustained Attention (T4)	0.750 (0.095)	< 0.001	[0.556, 0.934]
Visual PPI (T4)	0.768 (0.559)	0.170	[−0.545, 1.768]
Genotype	−5.874 (2.813)	0.037	[−11.383, −0.280]
Sustained Attention (T4)
AUC (T3)	0.750 (0.095)	< 0.001	[0.556, 0.934]
Visual PPI (T3)	−0.908 (0.644)	0.159	[−2.181, 0.330]
Genotype	−5.874 (2.813)	0.037	[−11.383, −0.280]
Sustained Attention (T3)
Genotype	−5.874 (2.813)	0.037	[−11.383, −0.280]
Total Effect of Genotype on Sustained Attention Across Time	−13.879 (5.648)	0.014	[–23.448, −1.298]
Total Mediated Effect of Genotype on Sustained Attention through Visual PPI Across Time	−8.005 (3.024)	0.008	[−13.272, −1.294]

2.2.3. Model 3: Long-term episodic memory

Overall model fit statistics for the five-wave, lag-1 autoregressive mediation model predicting the effect of genotype on long-term episodic memory through visual PPI indicated strong support for the proposed model, such that the observed covariance matrix of the data did not significantly differ from the model-implied covariance structure: χ²(42) = 52.683, p = .125; CFI = 0.953; SRMR = 0.072. The overall effect of genotype on long-term episodic memory in the model was 100.090 (18.813), p < .001, 95% bias-corrected bootstrap CI [63.008, 136.831], demonstrating a sustained impact of the HIV-1 transgene on long-term episodic memory. The overall indirect effect of genotype on long-term episodic memory through visual PPI was 48.422 (11.633), p < .001, 95% bias-corrected bootstrap CI [29.858, 77.145], indicating that preattentive processing represents a plausible mediating mechanism underlying the HIV-1 transgene’s influence on long-term episodic memory over time. The direct effect of the HIV-1 transgene on long-term episodic memory once accounting for the mediator remained significant, −5.874 (2.813), p = .037, 95% bias-corrected bootstrap CI [−11.383, −0.280], demonstrating that preattentive processing is a partial mediator and explains 48.4% of the observed HIV-1 transgene effect. All parameter estimates for the model are provided in Table 4.

Table 4.

Model Parameter Estimates: Long-Term Episodic Memory (Plateau) as outcome.

Variable	Estimate (SE)	p-value	95% Bootstrapped CI
Visual PPI (T5)
Visual PPI (T4)	0.151 (0.066)	0.022	[0.016, 0.267]
Genotype	−1.560 (0.305)	< 0.001	[−2.138, −0.940]
Visual PPI (T4)
Visual PPI (T3)	0.151 (0.066)	0.022	[0.016, 0.267]
Genotype	−1.560 (0.305)	< 0.001	[−2.138, −0.940]
Visual PPI (T3)
Visual PPI (T2)	0.151 (0.066)	0.022	[0.016, 0.267]
Genotype	−1.560 (0.305)	< 0.001	[−2.138, −0.940]
Visual PPI (T2)
Visual PPI (T1)	0.151 (0.066)	0.022	[0.016, 0.267]
Genotype	−1.560 (0.305)	< 0.001	[−2.138, −0.940]
Visual PPI (T1)
Genotype	−1.560 (0.305)	< 0.001	[−2.138, −0.940]
Long-Term Episodic Memory (T5)
Long-Term Episodic Memory (T4)	0.380 (0.078)	< 0.001	[0.244, 0.552]
Visual PPI (T4)	−6.680 (3.010)	0.026	[−11.786, −0.247]
Genotype	51.667 (12.358)	< 0.001	[28.088, 76.514]
Long-Term Episodic Memory (T4)
Long-Term Episodic Memory (T3)	0.380 (0.078)	< 0.001	[0.244, 0.552]
Visual PPI (T3)	−4.579 (4.795)	0.340	[−14.000, 5.225]
Genotype	51.667 (12.358)	< 0.001	[28.088, 76.514]
Long-Term Episodic Memory (T3)
Long-Term Episodic Memory (T2)	0.380 (0.078)	< 0.001	[0.244, 0.552]
Visual PPI (T2)	−6.803 (7.111)	0.339	[−26.805, 0.155]
Genotype	51.667 (12.358)	< 0.001	[28.088, 76.514]
Long-Term Episodic Memory (T2)
Long-Term Episodic Memory (T1)	0.380 (0.078)	< 0.001	[0.244, 0.552]
Visual PPI (T1)	−2.617 (5.587)	0.639	[−0.026, 0.267]
Genotype	51.667 (12.358)	< 0.001	[28.088, 76.514]
Long-Term Episodic Memory (T1)
Genotype	51.667 (12.358)	< 0.001	[28.088, 76.514]
Total Effect of Genotype on Long-Term Episodic Memory Across Time	100.090 (18.813)	< 0.001	[63.008, 136.831]
Total Mediated Effect of Genotype on Long-Term Episodic Memory through Visual PPI Across Time	48.422 (11.633)	< 0.001	[29.858, 77.145]

3. Discussion

Statistical mediation analyses revealed a putative neurobehavioral mechanism underlying alterations in the core components of cognitive function. Progressive neurocognitive impairments, characterized by alterations in preattentive processing, learning, sustained attention, and long-term episodic memory (McLaurin et al., 2018a, 2019a), were observed in the HIV-1 Tg rat using a longitudinal experimental design across the functional lifespan. Three lag-1 autoregressive mediation models revealed that preattentive processing partially mediated learning, sustained attention, and long-term episodic memory in the HIV-1 Tg rat, explaining 44% to 58% of the observed HIV-1 transgene effect. Understanding the neurobehavioral mechanisms mediating alterations in HAND may provide key targets for the development of diagnostic screening tools, novel therapeutics, and cure/restoration strategies.

A biomarker, or mediator, has been defined by the National Institutes of Health as “a characteristic that is objectively measured and evaluated as an indicator of normal biological processes, pathogenic processes, or pharmacologic responses to a therapeutic intervention” (Biomarkers Definitions Working Group, 2001). Findings that PPI, tapping preattentive processing, partially mediated the relationship between genotype and alterations in learning, sustained attention, and long-term episodic memory in the HIV-1 Tg rat support its utility as a key target for the development of a biomarker for HAND. Given that currently available diagnostic tools lack the sensitivity and specificity needed for an accurate diagnosis for HAND (e.g., Haddow et al., 2013; Sakamoto et al., 2013; Zipursky et al., 2013), the development of an innovative biomarker has the potential for great clinical significance (Zipursky et al., 2013; Chan and Brew, 2014). Stringent criteria, including both high specificity, high sensitivity, and a high likelihood of causation, must be considered in the development of a novel biomarker (Aronson, 2005). Discriminant function analyses (e.g., McLaurin et al., 2016; McLaurin et al., 2017b) and receiver operator characteristic curves, complementary statistical techniques, support the utility of PPI as a powerful biomarker with both high sensitivity (i.e., 89.3–100%) and high specificity (i.e., 79.5–94.1%) across the functional lifespan (McLaurin et al., 2019c). Furthermore, statistical mediation analysis, as assessed in the present study, elucidated the possible role of PPI as an intervening variable in causal pathways between HIV-1 and several neurocognitive markers. Thus, evidence from multiple statistical techniques points toward PPI as a potentially innovative biomarker.

Despite the potential clinical significance of deficits in preattentive processing, results indicate that the construct of PPI does not completely explain alterations in the core components of cognitive function. Other behavioral alterations commonly observed in HAND, including apathy (e.g., Kamat et al., 2012; Bertrand et al., 2018) and depression (e.g., Ciesla and Roberts, 2001; Nemeth et al., 2014), may also underlie alterations in the core components of cognitive function. Indeed, an examination of the relationship between apathy and neurocognitive function in HIV-1 seropositive individuals, revealed a significant association between higher levels of apathy and greater impairments in neurocognitive function (Shapiro et al., 2013). Furthermore, a comprehensive review revealed a significant association between depression and neurocognitive impairment in HIV-1 seropositive individuals (Rubin and Maki, 2019). Depression has also been associated with greater impairment in speed of information processing (Rubin et al., 2019), suggesting it may underlie alterations in preattentive processing. Identifying the most effective biomarker for HAND may benefit from using multiple assessments, tapping both preattentive processing and behavioral alterations (i.e., apathy and depression). The potential of this identification lies in testing novel therapeutics and cure/restoration strategies for the HIV-1 virus.

The neural circuitry underlying PPI, which has been well-established using lesioning (e.g., Leitner and Cohen, 1985) and electrical stimulation studies (e.g., Li and Yeomans, 2000), may offer a key target for the development of adjunctive therapeutics and cure/restoration strategies in HIV-1. Neurotransmitter systems, including dopamine (DA; e.g., Kumar et al., 2011), glutamate (e.g., Melendez et al., 2016) and acetylcholine (ACh; e.g., Ramos et al., 2016), implicated in the pathogenesis of HAND, play a critical role in the neural circuitry underlying PPI (Koch et al., 1993; Fendt et al., 2001). Specifically, DA afferents from the ventral tegmental area (VTA) innervate the nucleus accumbens, which subsequently relays information to the pedunculopontine tegmental nucleus (PPTg); activity which is likely influenced by glutamate projections from the medial prefrontal cortex to the VTA (Taber et al., 1995). Subsequently, ACh projections from the PPTg to the caudal pontine reticular nucleus are instrumental in eliciting a startle response (Koch et al., 1993; Fendt et al., 1994; Koch and Schnitzler, 1997; Fendt et al., 2001). Targeting these neurotransmitter systems, either alone or in combination, may aid in the development of adjunctive therapeutics and cure/restoration strategies in HIV-1; a critical need given the continued prevalence of HAND in the post-cART era.

Utilization of a longitudinal experimental design across the functional lifespan is one of the undeniable strengths of the present study. Longitudinal studies, which rely on repeated measures across time, afford a critical opportunity to examine mediational processes. First, longitudinal data more accurately represent the temporal ordering of variables (i.e., X precedes M which occurs before Y), satisfying a critical assumption of mediation (i.e., temporal precedence; MacKinnon, 2008). Second, the use of repeated measures dramatically decreases the sample size required to achieve adequate statistical power (Pan et al., 2018). For example, when cross-sectional data are analyzed using the single mediator model, a sample size between 34 and 148 is needed for 80% power (Fritz and MacKinnon, 2007) when effect sizes are comparable to those observed in the present study (i.e., η_p² = 0.022–0.076 corresponding to Cohen’s d of 0.30–0.57). In sharp contrast, with three repeated observations and a small intraclass correlation (ICC), a sample size between 15 and 69 is required to achieve 80% power in longitudinal data; a sample size savings that was observed even when the ICC was high (Pan et al., 2018).

Longitudinal experimental designs, however, are not without potential caveats. Specifically, “practice effects” or “learning effects” are observed in many neuropsychological assessments (Dikmen et al., 1999); an effect that is strongest during the first two retest assessments (e.g., Thorndike, 1922; Falleti et al., 2006). Given “practice effects” in longitudinal data, it may be more difficult to ascertain a singular effect at any given timepoint relative to investigating an overall cumulative effect over time. However, if an experiment focuses on one developmental timeframe (e.g., adolescence, adulthood, or geriatrics) it may be arguable that a cumulative effect over time would be of greater interest than a relationship at a single time point.

Despite the strengths of the present study, low statistical power precluded the use of a moderated mediation model to investigate potential interactive effects of biological sex. Investigations of the role of biological sex in HAND have elucidated prominent sex differences with women exhibiting greater neurocognitive impairments than men (e.g., Clinical: Royal et al., 2016; Maki et al., 2018; Preclinical: Rowson et al., 2016; McLaurin et al., 2019a). Pronounced sex differences in HAND suggest that the factor of biological sex may be a moderator, influencing the strength or direction (MacKinnon, 2008), of the mediating processes revealed in the present study. However, the sample size needed to accurately estimate a moderated mediation model with statistical power of 0.80 is significantly greater than was available (i.e., N = 65) in the present study. Specifically, research has revealed that the power to detect a mediated effect in a cross-sectional moderated mediation model with latent variables with N = 200 was only 0.37 (Thoemmes et al., 2010). Thus, while the present study controlled for the factor of biological sex, there remains a critical need to conduct a study with increased sample size to investigate possible moderation effects within the mediation model.

Elucidating a potential neurobehavioral mechanism underlying the core components of executive function in HAND affords an opportunity to similarly investigate potential neuroanatomical mechanisms mediating HAND. Multiple neural mechanisms, including synaptic dysfunction (e.g., Gelman et al., 2012; Desplats et al., 2013; Roscoe et al., 2014; Festa et al., 2015; McLaurin et al., 2018b; McLaurin et al., 2019a), neurotransmitter alterations (e.g., Kumar et al., 2011; Javadi-Paydar et al., 2017; Sinharay et al., 2017; Denton et al., 2019), and neuroinflammation (e.g., Royal et al., 2012, 2016), are likely involved in the pathogenesis of HAND. A clinical, multilevel analysis examining genetic and histopathological markers reported that markers of synaptodendritic integrity (i.e., microtubule associated protein 2, synaptophysin) were most strongly related to HAND (Moore et al., 2006; Levine et al., 2016). Inflammatory cytokines (i.e., sCD14; Royal et al., 2016) and decreased dopaminergic system function (Chang et al., 2008) have also been associated with neurocognitive impairment. Statistical mediation analyses, therefore, may provide a critical opportunity to empirically evaluate if, and to what extent, synaptic dysfunction, neurotransmitter alterations, and neuroinflammation underlie HAND.

In conclusion, the present study utilized statistical mediation analysis to establish preattentive processing as a partial mediator of alterations in the core components of cognitive function. Statistical mediation is a powerful statistical technique for the examination of key mechanisms underlying neurodegenerative diseases. PPI, which exhibits high sensitivity, high specificity, and a high likelihood of causation, may provide a key biomarker for milder forms of HAND. Furthermore, the neural circuitry underlying PPI is fairly well-established, supporting an opportunity to develop adjunctive therapeutics and cure/restoration strategies for HIV-1 in the post-cART era.

4. Experimental procedure

4.1. Animals

Fischer HIV-1 Tg (F344/N) and control animals were procured from Harlan Laboratories Inc. (Indianapolis, IN) across 12 months. Animals arrived at the animal vivarium, housed with their biological dam, between postnatal day (PD) 7 and PD 9. At weaning (i.e., ~PD 21), animals were sampled from each litter (HIV-1 Tg: N = 20 litters, male, n = 37, female, n = 33; Control: N = 17 litters, male n = 34; female, n = 33) and pair- or group-housed with animals of the same sex for the duration of experimentation.

Food restriction was implemented at approximately PD 60, one week prior to the beginning of the original acquisition of signal detection, to maintain animals at 85% body weight. Rodent food (Pro-Lab Rat, Mouse, Hamster Chow #3000) was provided ad libitum following the successful acquisition of the signal detection task (PD 100-PD 277) and throughout the duration of experimentation. Water was provided ad libitum.

HIV-1 Tg and control animals were maintained in AAALAC-accredited facilities at the University of South Carolina utilizing recommendations published by the National Institutes of Health in the Guide for the Care and Use of Laboratory Animals. Environmental conditions in the animal vivarium were targeted at 21° ± 2 °C, 50% ± 10% relative humidity and a 12-h light:12-h dark cycle with lights on at 0700 h (EST). The project protocol was approved by the Institutional Animal Care and Use Committee (IACUC) at the University of South Carolina under federal assurance (# D16–00028).

4.2. Experimental design

The longitudinal experimental design for neurocognitive assessments, including cross-modal PPI, tapping temporal processing, signal detection, tapping learning and sustained attention, and locomotor activity, tapping long-term episodic memory, is illustrated in Fig. 4.

Fig. 4.

Experimental design for all neurocognitive assessments. Adapted from McLaurin et al. (2019a)

4.3. Neurocognitive assessments

4.3.1. Prepulse Inhibition: preattentive processing

4.3.1.1. Apparatus.

A 10 cm-thick double-walled, 81 × 81 × 116-cm isolation cabinet (external dimensions) (Industrial Acoustic Company, INC., Bronx, NY) enclosed the startle platform (SR-Lab Startle Reflex System, San Diego Instruments, Inc., San Diego, CA), providing over 30 dB(A) of sound attenuation relative to the external environment. The ambient sound level in the chamber, in the absence of any stimuli, was 22 dB(A). A high-frequency loudspeaker of the SR-Lab system (Radio Shack model #40–1278B) was mounted inside the isolation cabinet 30cm above the Plexiglas test cylinder for the presentation of all discrete auditory prepulses and startling stimuli (frequency range of 5 k-16 k Hz). A microphone was placed inside the Plexiglas cylinder for the measurement and calibration of sound levels (Sound Level Meter: model #2203, Bruël and Kjaer, Norcross, GA). A 22 lx white LED light (Light meter model #840006, Sper Scientific, Ltd, Scottsdale, AZ) was affixed on the isolation cabinet wall in front of the Plexiglas test cylinder for the presentation of discrete visual prepulses. The animal’s whole body startle response to the auditory startle stimulus produced deflection of the Plexiglas test cylinder; a deflection that was converted into analog signals by a piezoelectric accelerometer integral to the bottom of the cylinder. Response signals were digitized (12 bit A to D) and saved to a hard disk. The SR-LAB Startle Calibration System was utilized to calibrate all response sensitivities.

4.3.1.2. Procedure: cross-modal prepulse inhibition.

Cross-modal PPI was assessed every 30 days from PD 90 to PD 180 and every 60 days from PD 240 to PD 540. Methodological details for cross-modal PPI are presented in our prior publications (e.g., McLaurin et al., 2018a). In brief, PPI of the auditory startle response (ASR) was assessed using both auditory and visual prepulse stimuli during a 30-minute test session. Following a 5-minute acclimation period with 70 dB(A) background noise, six pulse-only ASR trials, with a fixed 10 s intertrial interval (ITI), were used for habituation. Seventy-two testing trials were subsequently presented, including an equal number of auditory (85 dB(A)) and visual (22 lx) prepulse trials (20 msec duration), arranged using an ABBA counterbalanced order of presentation. The auditory startle stimulus (100 dB(A)) also had a 20 msec duration. Testing trials were presented in 6-trial blocks, interdigitated using a Latin-square experimental design, with interstimulus intervals (ISI) of 0, 30, 50, 100, 200, and 4000 msec and a variable ITI (15–25 sec). Control trials, including both the 0 and 4000 msec ISI trials, provided a reference ASR within the test session. Mean peak ASR amplitude values were collected for analysis. Cross-modal PPI data, assessing sensory system integrity and alterations in the progression of temporal processing, was reported in McLaurin et al. (2018a).

4.3.2. Signal detection operant task: sustained attention

4.3.2.1. Apparatus.

A signal detection operant task, tapping sustained attention, was conducted in sixteen operant chambers, located inside sound-attenuating chambers (Med Associates, Inc., VT). The front wall of the operant chamber included two retractable levers, a pellet dispenser (45 mg), and three panel lights. The pellet dispenser was located between the two retractable levers. One incandescent panel light (22 lx), located above the pellet dispenser, was utilized for signal presentation during the present experiment. A house light, utilized for the shaping response protocol, was located at the top of the rear wall of the operant chamber. Signal presentation, lever operation, reinforcement delivery, and data collection were control by a PC and Med-PC for Windows software (V 4.1.3; Med Associates, Inc., Fairfax, VT).

4.3.2.2. Procedure.

Methodology for the signal detection operant task, tapping sustained attention, is detailed in McLaurin et al. (2019a).

In brief, at approximately 2 months of age, HIV-1 Tg and control animals were initially trained to lever-press using a standard shaping response protocol. Following the successful acquisition of shaping, three vigilance programs, initially described by McGaughy and Sarter (1995), were employed to train animals to discriminate between signal (i.e., central panel illumination) and non-signal (i.e., no illumination) trials. Specifically, in the first vigilance program, the stimulus light remained illuminated until the animal responded or for 6 s. In the second vigilance program, the stimulus light was illuminated for 1 s. Following an incorrect response, an animal was given up to three correction trials (i.e., a repetition of the trial). A forced-choice trial was employed when an animal failed to respond correctly during the correction trials. In the third vigilance program the length of the stimulus was manipulated (i.e., 1000, 500, 100 ms) across trials using a block randomized experimental design. Correction trials and force-choice trials were removed in the third vigilance program.

HIV-1 Tg and control animals were required to meet criterion of at least 70% accuracy for 5 consecutive or 7 non-consecutive test sessions during each vigilance program. After successfully meeting criterion, animals were promoted to the next program. Percent accuracy was calculated as follows: (Total Number of Hits and Correct Rejections)/ (Total Number of Correct and Incorrect Responses × 100).

After successfully acquiring the signal detection task, HIV-1 Tg and control animals were assessed in the third vigilance program with varying signal durations (i.e., 1000, 500, 100 msec) every 60 days through approximately 18 months of age. Animals were given up to 60 days to successfully complete each retest assessment and were assessed in up to 6 retest assessments. Data from the signal detection operant task, tapping sustained attention, was presented in McLaurin et al. (2019a).

4.3.3. Locomotor activity: long-term episodic memory

4.3.3.1. Apparatus.

Long-term episodic memory was assessed using locomotor activity in 16 square (40 × 40 cm) activity chambers (Hamilton Kinder, San Diego Instruments, San Diego, CA), which, using perspex inserts, were converted into round (~40 cm diameter) compartments. Free movement was detected using infrared photocell (32 emitter/detector pairs) interruptions. To maintain the infrared photocell sensitivity with the additional layer of perspex, photocells were tuned by the manufacturer.

4.3.3.2. Procedure.

Locomotor activity was assessed every 30 days from PD 90 to PD 180 and every 60 days from PD 240 to PD 540. A 60-min test session was conducted between 700 and 1200 h (EST) in an isolated room under dim light conditions (< 10 lx). The number of photocell interruptions within the 60-min test session were collected for analysis. Locomotor activity data, assessing motor system function integrity and alterations in the progression of long-term episodic memory, was reported in McLaurin et al. (2018a).

4.4. Statistical analysis

Given the nested experimental design, (i.e., rats sampled from within litters), statistical analyses were conducted on litter means and standard errors, dependent upon biological sex, to preclude the violation of the independence assumption (Denenberg, 1993; Wears, 2002). Censored data, either due to euthanization, an equipment malfunction, or failure to complete the retest assessment was handled using full information maximum likelihood estimation in line with currently recommended methods (e.g., Enders, 2010). Data previously reported in McLaurin et al. (2018a, 2019a) were reanalyzed in a novel manner to investigate mediation.

4.4.1. Statistical mediation analysis

All analyses were conducted in Mplus Version 8.2 (Muthén and Muthén, 1998). Full information maximum-likelihood was utilized to estimate model parameters, as this technique has been shown to yield the most asymptotically unbiased and efficient estimates across a variety of circumstances (West et al., 1995). Unstandardized parameter estimates and standard errors were reported, alongside p-values and bias-corrected bootstrapped confidence intervals. Both absolute and incremental fit indices were utilized to assess adequacy of model fit, evaluating whether, and to what extent, the observed covariance matrix of the data differed from the model-implied covariance structure (e.g., Hoyle, 2012). The χ² goodness-of-fit test was utilized to ascertain absolute model fit, where the expected value of the statistic is the degrees of freedom for the model. Non-significant χ² values indicate support for model fit, such that the covariance structure of the observed data does not significantly differ from the model-implied covariance structure defined by the model. Comparative model fit indices were also considered to evaluate the extent of model misfit in the presence of a significant χ² statistic. Specifically, the comparative fit index (CFI) and standardized root mean square residual (SRMR) were examined (Hu and Bentler, 1999). The root mean square error of approximation (RMSEA) was not reviewed, as previous methodological work has recommended not computing the measure in small sample size studies with small df values (e.g., Kenny et al., 2015).

The CFI ranges from 0 to 1, with higher scores reflecting better model fit. Methodological resources have indicated that CFI values ≥0.90 (Bollen, 1989) or ≥0.95 (Hu and Bentler, 1999) demonstrate close fit of the observed data to the model. The SRMR provides a standardized measure of the average difference between the observed and model-implied covariance matrices. Thus, lower SRMR values indicate better model fit. Hu and Bentler (1999) have suggested that values ≤0.08 indicate close fit of the observed data to the theoretical model.

Estimates of overall total effects of the HIV-1 transgene on each outcome and overall mediated (i.e., indirect) effects through visual PPI were estimated in line with conventional path tracing rules for structural equation models (Bollen, 1989). Statistical significance testing of the mediated effects was conducted using the bias-corrected bootstrap to form asymmetric confidence intervals, in line with recommended practice for testing mediation (e.g., MacKinnon et al., 2007; MacKinnon, 2008; MacKinnon and Fairchild, 2009; Fairchild and McDaniel, 2017).

HIGHLIGHTS.

• Progressive neurocognitive impairments were observed in the HIV-1 Tg rat.

• Statistical mediation was used to evaluate mechanisms underlying HAND.

• Preattentive processes partially mediate alterations in cognitive function.

• Prepulse inhibition is a key target for the development of a diagnostic biomarker.