Beyond t test and ANOVA: applications of mixed-effects models for more rigorous statistical analysis in neuroscience research

Yu, Zhaoxia; Guindani, Michele; Grieco, Steven F.; Chen, Lujia; Holmes, Todd C.; Xu, Xiangmin

doi:10.1016/j.neuron.2021.10.030

Cited by 232 publications

(168 citation statements)

References 55 publications

Supporting

Mentioning

167

Contrasting

Unclassified

Order By: Relevance

“…For cases in which the same brain slice was stimulated multiple times, and other repeated measures, linear mixed modeling (LMM) was used to examine the effects of genotype or drug treatment. This approach estimates the effect size of each factor while accounting for intra- and inter-animal variability (Aarts et al, 2014 ; Boisgontier and Cheval, 2016 ; Yu et al, 2021 ) and is gaining wide acceptance (Lau et al, 2017 ; Huang et al, 2018 ; Hanson et al, 2019 ; Koenig et al, 2019 ; Grieco et al, 2020 ; Kurucu et al, 2021 ). LMMs were fitted with random intercepts to assess for the correlation between repeated measurements on the same mouse, and experiment-specific effects were analyzed for statistical significance.…”

Section: Methodsmentioning

confidence: 99%

Neuronal Loss of the Glutamate Transporter GLT-1 Promotes Excitotoxic Injury in the Hippocampus

Rimmele

Andersen

et al. 2021

Front. Cell. Neurosci.

View full text Add to dashboard Cite

GLT-1, the major glutamate transporter in the mammalian central nervous system, is expressed in presynaptic terminals that use glutamate as a neurotransmitter, in addition to astrocytes. It is widely assumed that glutamate homeostasis is regulated primarily by glutamate transporters expressed in astrocytes, leaving the function of GLT-1 in neurons relatively unexplored. We generated conditional GLT-1 knockout (KO) mouse lines to understand the cell-specific functions of GLT-1. We found that stimulus-evoked field extracellular postsynaptic potentials (fEPSPs) recorded in the CA1 region of the hippocampus were normal in the astrocytic GLT-1 KO but were reduced and often absent in the neuronal GLT-1 KO at 40 weeks. The failure of fEPSP generation in the neuronal GLT-1 KO was also observed in slices from 20 weeks old mice but not consistently from 10 weeks old mice. Using an extracellular FRET-based glutamate sensor, we found no difference in stimulus-evoked glutamate accumulation in the neuronal GLT-1 KO, suggesting a postsynaptic cause of the transmission failure. We hypothesized that excitotoxicity underlies the failure of functional recovery of slices from the neuronal GLT-1 KO. Consistent with this hypothesis, the non-competitive NMDA receptor antagonist MK801, when present in the ACSF during the recovery period following cutting of slices, promoted full restoration of fEPSP generation. The inclusion of an enzymatic glutamate scavenging system in the ACSF conferred partial protection. Excitotoxicity might be due to excess release or accumulation of excitatory amino acids, or to metabolic perturbation resulting in increased vulnerability to NMDA receptor activation. Previous studies have demonstrated a defect in the utilization of glutamate by synaptic mitochondria and aspartate production in the synGLT-1 KO in vivo, and we found evidence for similar metabolic perturbations in the slice preparation. In addition, mitochondrial cristae density was higher in synaptic mitochondria in the CA1 region in 20–25 weeks old synGLT-1 KO mice in the CA1 region, suggesting compensation for loss of axon terminal GLT-1 by increased mitochondrial efficiency. These data suggest that GLT-1 expressed in presynaptic terminals serves an important role in the regulation of vulnerability to excitotoxicity, and this regulation may be related to the metabolic role of GLT-1 expressed in glutamatergic axon terminals.

show abstract

Section: Methodsmentioning

confidence: 99%

Neuronal Loss of the Glutamate Transporter GLT-1 Promotes Excitotoxic Injury in the Hippocampus

Rimmele

Andersen

et al. 2021

Front. Cell. Neurosci.

View full text Add to dashboard Cite

show abstract

“…The statistical analysis of most types of electrophysiological datasets is complicated by both the repeated nature as well as the hierarchical relationships between measurements. Traditional approaches to analyze these datasets typically rely on the implicit assumption that individual measurements are statistically independent [Aarts et al, 2014; Yu et al, 2022]. Our demonstration that the interevent interval and amplitude of EPSCs is clustered around the mean of individual neurons demonstrates that this assumption is violated and illustrates the importance of using statistical models that reflect the hierarchical nature of electrophysiological datasets.…”

Section: Discussionmentioning

confidence: 93%

“…All of these approaches can record the electrophysiological behavior of individual neurons and do so by performing (many) repeated measurements of each neuron. The repeated nature of electrophysiological measurements does however complicate the analysis of these recordings and commonly used analysis approaches are often overly conservative or unreliable due to pseudoreplication [Aarts et al, 2014; Yu et al, 2022].…”

Section: Introductionmentioning

confidence: 99%

“…Due to the normalizing at the level of individual cells, an outsized weight will thus be attached to these more extreme values. Finally, it is important to recognize that the distribution of electrophysiological values typically resembles a Poisson distribution, thus requiring either a non-parametric statistical test or transformation of the data to ensure that the residuals are normally distributed [Limpert et al, 2001; Yu et al, 2022].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

EPHierStats: a statistical tool to model the hierarchical relationships in electrophysiological data

Kolpakova

Marvel-Zuccola

Futai

et al. 2022

Preprint

View full text Add to dashboard Cite

Electrophysiological datasets are typically analyzed under the assumption that repeated measurements of the same unit of analysis (i.e. neuron or animal) can be treated as statistically independent. Recently, this assumption has been questioned and our data confirms and quantifies this skepticism using ex vivo slice recordings of synaptic currents in D1R+ medium spiny neurons in the nucleus accumbens. We therefore present EPHierStats as a statistical framework to analyze electrophysiological datasets with large numbers of measurements (>100) per unit of analysis. This novel analysis framework enables encoding of the full hierarchical relationships between measurements in a mixed-effects general linear model while also analyzing the distribution of values in assessed variables. Our method can easily be adapted to analyze a wide range of repeated-measures electrophysiological experiments. Implementation of the EPHierStats tool will aid the adaption of modern statistical approaches that prevent pseudoreplication and its associated false discovery rate while enabling statistical assessments of the complex relationships inherent to the field of neuroscience.

show abstract

“…The purpose of the current study was to 1) determine if the effects of dietary protein restriction on body weight, food consumption, and preference for protein solution could be demonstrated in individual C57BL/6J mice and whether such effects would be absent in Fgf 21‐KO mice and 2) illustrate how multilevel model analyses could be used with SCED data to quantify the effects of dietary protein restriction on those outcomes. An in‐depth discussion of the benefits of multilevel analyses is beyond the scope of the current manuscript, but their use has been advocated for the analysis of within‐subject repeated measures data for multiple reasons, including their ability to handle unbalanced and missing data (for detailed discussions see Boisgontier & Cheval, 2016; DeHart & Kaplan, 2019; Kaplan et al, 2021; Young, 2017; Young, 2018; Yu et al, 2021).…”

mentioning

confidence: 99%

Dynamic effects of dietary protein restriction on body weights, food consumption, and protein preference in C57BL/6J and Fgf21‐KO mice

Torres

Khan

Fernandez-Kim

et al. 2022

J Exper Analysis Behavior

View full text Add to dashboard Cite

Single‐case experimental designs (SCEDs) are rarely used in behavioral neuroscience despite their potential benefits. The current study used a SCED to evaluate the effects of dietary protein restriction in C57BL/6J and Fgf21‐knockout (KO) mice on body weight, food consumption, and protein preference and changes in those outcome measures were quantified using multilevel linear models. In C57BL/6J mice, rate of weight gain was lower and food consumption and protein preference higher during periods of low (4% kcal) protein diet feeding compared to periods of normal (18% kcal) protein diet feeding. In Fgf21‐KO mice, who do not produce the liver‐derived hormone FGF21, rate of weight gain and protein preference were not substantially affected by diet although food consumption was slightly higher during periods of low protein diet than periods of normal protein diet. These results demonstrate that protein restriction dynamically regulates physiological and behavioral responses at the individual mouse level and that FGF21 is necessary for those responses. Further, the current results demonstrate how a SCED can be used in behavioral neuroscience research, which entails both scientific and practical benefits.

show abstract

Beyond t test and ANOVA: applications of mixed-effects models for more rigorous statistical analysis in neuroscience research

Cited by 232 publications

References 55 publications

Neuronal Loss of the Glutamate Transporter GLT-1 Promotes Excitotoxic Injury in the Hippocampus

Neuronal Loss of the Glutamate Transporter GLT-1 Promotes Excitotoxic Injury in the Hippocampus

EPHierStats: a statistical tool to model the hierarchical relationships in electrophysiological data

Dynamic effects of dietary protein restriction on body weights, food consumption, and protein preference in C57BL/6J and Fgf21‐KO mice

Contact Info

Product

Resources

About