Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2021 Apr 20.
Published in final edited form as: Psychopharmacology (Berl). 2014 Jan;231(1):231–242. doi: 10.1007/s00213-013-3230-6

Mining Mouse Behavior for Patterns Predicting Psychiatric Drug Classification

Neri Kafkafi 1, Cheryl L Mayo 2, Greg I Elmer 3
PMCID: PMC8056474  NIHMSID: NIHMS517218  PMID: 23958942

Abstract

Rationale

In psychiatric drug discovery, a critical step is predicting the psychopharmacological effect and therapeutic potential of novel (or repurposed) compounds early in the development process. This process is hampered by the need to utilize multiple, disorder-specific and labor intensive behavioral assays.

Objectives

This study aims to investigate the feasibility of a single high-throughput behavioral assay to classify psychiatric drugs into multiple psychopharmacological classes.

Methods

Using Pattern Array, a procedure for data-mining exploratory behavior in mice, we mined ~100,000 complex movement patterns for those that best predict psychopharmacological class and dose. The best patterns were integrated into a classification model that assigns psychopharmacological compounds to one of six clinically-relevant classes – antipsychotic, antidepressant, opioids, psychotomimetic, psychomotor stimulant and α-adrenergic.

Results

Surprisingly, only a small number of well-chosen behaviors were required for successful class prediction. One of them, a behavior termed “universal drug detector” was dose-dependently decreased by drugs from all classes, thus providing a sensitive index of psychopharmacological activity. In independent validation in a blind fashion, simulating the process of in vivo pre-clinical drug screening, the classification model correctly classified 9 out of 11 “unknown” compounds. Interestingly, even “misclassifications” match known alternate therapeutic indications, illustrating drug “repurposing” potential.

Conclusions

Unlike standard animal models, the discovered classification model can be systematically updated to improve its predictive power, and add therapeutic classes and subclasses with each additional diversification of the database. Our study demonstrates the power of data-mining approaches for behavior analysis, using multiple measures in parallel for drug screening and behavioral phenotyping.

Keywords: animal model, behavioral phenotyping, SEE, open field, spatial behavior

4. Introduction

There is a growing consensus that psychiatric CNS drug discovery is largely failing to provide new medications alternatives (Conn and Roth 2008; Schoepp 2011). The lack of a clear neuropathology and the complexity of the emotional and cognitive disturbances make the use of behavioral animal models particularly challenging (Agid et al. 2007; Pangalos et al. 2007; Nestler and Hyman 2010). The in vivo capacity to screen novel chemical entities is further hampered by the need to utilize multiple, often labor intensive behavioral assays, each limited to identifying a narrow drug class or psychiatric disorder using a small set of endpoints. It seems reasonable to expect that some properties in the free movement of a mouse would indicate whether it was injected with, e.g., an antidepressant, an antipsychotic or an opioid compound. Complex locomotor patterns have been suggested for classification of compounds in animals and humans (Geyer et al. 1986; Henry et al. 2010), but no standard behavioral assay has been shown to consistently identify and differentiate multiple psychopharmacological classes. Such an assay, especially if it is high throughput and reliable, would be very useful as an in vivo screening procedure for identifying the therapeutic potential of novel compounds, as well as for discovering new uses for existing compounds (repurposing).

In recent years, data mining classification strategies has been successfully applied at the molecular and physiological level. For example, large gene-expression data sets are mined for profiles that are predictive of a toxicological or carcinogenic response (“class predictors”, e.g. Golub et al. 1999). This in vitro approach was used to discover gene-expression patterns that classify psychoactive drugs (Gunther et al. 2003). A similar approach for in vivo mining at the level of behavior was proposed in previous reviews (Brunner et al. 2002; Tecott and Nestler 2004). Rihel et al. (2010) recently applied this approach to construct a high throughput behavioral assay in zebra fish, but their typical prediction success of a compound’s therapeutic value was only slightly better than chance level. The difficulty seems to be that the natural units of behavior are not as well-defined and understood as those measured at the molecular and physiological levels, such as, e.g., highly annotated genes in gene-expression profiling. Therefore, the key to successful behavior mining is in the design of a useful “behavioral chip”, i.e., a proper categorization of the data into multiple types of behavior that can be mined.

Such a categorization of “open field” behavior in mice and rats was proposed in the PA method (Kafkafi et al. 2009), by using different combinations of 10 well-studied, ethologicaly-relevant attributes (Drai and Golani 2001; Kafkafi and Benjamini et al. 2005; Kafkafi and Elmer 2005; Benjamini et al. 2010) to define ~100,000 complex movement patterns. PA patterns range from general, e.g., “moving near the wall of the arena”, to more specific definitions, e.g., “moving near the wall while braking hard and sharply turning away from the wall”. This last pattern was actually discovered by PA to diagnose SOD1 rats, an animal model of Amyotrophic Lateral Sclerosis, at much earlier age than any standard behavioral measure, or even a human observer (Kafkafi et al. 2008). The large number and rich complexity of patterns used by PA increases the probability that some of them can serve as reliable predictors of any particular drug-effect of interest. The most predictive and reliable behaviors can be mined in a behavior database of animals injected with drugs of known psychopharmacological class and mechanism of action, and integrated into a classification model. As we demonstrate here, this classification model provides for the first time a single behavioral assay capable of predicting, with high rate of success, the therapeutic application of a drug among multiple, therapeutically-relevant psychopharmacological classes. This study also tests the feasibility of systematically updating and increasing the power of the classification model, simply by adding compounds and classes to the PA database.

5. Materials and Methods

Testing and algorithmic methods were described in detail in previous PA studies (Kafkafi et al. 2008; Kafkafi and Elmer 2009), and are summarized here in brief, except where differences and improvements were applied.

Animals, Drugs and Experimental Procedures

Animals were all 60- to 80-day-old C57BL/6J male mice all 60- to 80-day-old C57BL/6J male mice (Jackson Laboratories), acclimated to the animal facility for at least 7 days before testing. They were housed five per cage in standard conditions of 12:12 light cycle, 22°C room temperature, water and food ad libitum. 41 drugs representing 6 drug classes (psychomotor stimulant, opioid, psychotomimetic, antidepressant, antipsychotic, and α-adrenergic) were investigated in this study (Table 1). Open field testing took place during the light phase of the cycle. Each animal was injected once, and immediately introduced into a 2.503m diameter circular open-field arena, where its location was video-tracked for 603min using Noldus EthoVision®. The {Time, X, Y} coordinates of the path were exported and analyzed using the standard procedure in SEE (Software for the Exploration of Exploration; see Drai and Golani 2001; Kafkafi and Benjamini et al. 2005). The experimental protocols followed the “Principles of Laboratory Animal Care” (NIH publication no. 86–23, 1996). The animals used in this study were maintained in facilities fully accredited by the American Association for the Accreditation of Laboratory Animal Care.

Table 1. List of Drugs.

The 41 drugs in the study, divided into 6 classes and 3 datasets.

Drugs Doses (mg/kg) Data set
I II III
Psychomotor
Cocaine 3.0, 5.6, 10.0, 17.0, 30.0 X
Methamphetamine 0.3, 1.0, 1.7, 3.0 X
Methylphenidate 1.7, 5.6, 10.0, 17.0 X
Mazindol 1.0, 3.0, 5.6, 10.0 X
Modafinil 30, 56, 100, 170 X
Apomorphine 0.17, 0.3, 0.56, 1.0, 3.0 −
Nomifensine 3.0, 5.6, 10.0, 20.0 +
Opioid
Morphine 1.0, 3.0, 5.6, 10.0 X
Oxycodone 1.0, 3.0, 5.6 X
Fentanyl 0.056, 0.1, 0.17, 0.3 X
Codeine 5.6, 17.0, 30.0, 56.0 X
BW373U86 20.0 X
Buprenorphine 0.17, 0.3, 0.56, 1.0, 3.0 ±
Hydromorphone 0.3, 1.0, 3.0 +
Psychotomimetic
PCP 3.0, 5.6 X
SDZ220851 1.7, 3.0, 5.6 X
Ketamine 5.6, 10.0, 17.0 X
Salvinorin A 1.0, 3.0, 10.0 X
MK801 0.1, 0.3, 1.0 X
Memantine 10.0, 20.0, 30.0 X
α-Adrenergic
Yohimbine 0.3, 3.0, 5.6 X
Izadoxan 3.0, 5.6, 10.0 X
Atipemazole 10.0, 17.0, 30.0 X
Antidepressant
Fluoxetine 10.0, 20.0, 30.0 X
Paroxetine 10.0, 20.0, 30.0 X
Imipramine 10.0, 20.0, 30.0 X
Desipramine 5.0,10.0, 20.0, 30.0 X
Nortriptyline 3.0, 5.6, 10.0 X
Bupropion 10.0, 17.0, 30.0 X
Venlafaxine 30.0, 40.0 −
Maprotiline 10.0, 20.0, 30.0 +
Amitriptyline 10.0, 20.0 +
Antipsychotic
Haloperidol 0.1, 0.3 X
Chlorpromazine 0.1, 0.17, 0.3 X
Amoxapine 3.0, 5.6 X
Clozapine 1.0, 1.7, 2.3, 3.0 X
Olanzapine 0.3, 1.0 X
Risperidone 0.25, 0.56, 1.00 +
Droperidol 0.1, 0.3, 1.0 +
Loxapine 1.7, 3.0, 5.6 +
Eticlopride 0.03, 0.17, 0.3 +
Open in a new tab

Red: Psychomotor stimulants; blue: opioids; green: psychotomimetic; purple: antidepressant; cyan: antipsychotic; orange: adrenergic. X: included in this dataset; +: correctly classified by PA; −: incorrectly classified by PA. ±: correctly limited by PA into two classes.

Datasets

Four datasets were used in the analysis. Two of which were gathered especially for the present study, while the other two were gathered in previous studies (Table 1). Data set I was recorded for both training and validation phases of Kafkafi et al (2009), and included animals injected with 13 drugs, each including several dose groups, belonging to the psychomotor stimulant, opioid and psychotomimetic drug classes. Data set II was recorded for the mining phase of the present study, and included 17 additional drugs, each including several dose groups, belonging to the α-adrenergic, antidepressant and antipsychotic drug classes. Data set III was recorded for the validation of the classification model in the present study, and included 11 additional drugs belonging to the antidepressant, antipsychotic, psychomotor stimulant and opioid classes. Data set VI included drug naïve animals from 10 inbred strains across 3 laboratories, recorded in the frame of a previous study (Kafkafi and Benjamini et al. 2005) and used here to estimate heritability and replicability of identified behavioral patterns across laboratories.

Drugs

In total, 41 drugs representing six drug classes (psychomotor stimulant, opioid, psychotomimetic, antidepressant, antipsychotic, and α-adrenergic) were investigated in this study (Table 1). Nomifensine maleate salt, apomorphine, morphine, hydromorphone hydrochloride, MK801, memantine hydrochloride, nortriptyline hydrochloride, venlafaxine hydrochloride, maprotiline hydrochloride, amitriptyline hydrochloride, risperidone, droperidol, loxapine succinate salt, eticlopride, and atipamezole were all purchased from Sigma-Aldrich (St. Louis, MO). Nomifensine, apomorphine, hydromorphone, maprotiline, amitriptyline, loxapine, eticlopride and atipamezole were dissolved in deionized water vehicle. Morphine, MK801, memantine, nortriptyline, buprenorphine and venlafaxine were all dissolved in 0.9% saline. Risperidone was dissolved in acetic acid, and then brought to the appropriate concentration with saline (acetic acid (0.2%)). Droperidol was dissolved in Tween and ethanol, and then brought to the appropriate concentration with saline (Tween (20%)/ethanol (2%)). Vehicle solutions were used as control. Apomorphine, morphine, oxycodone, fentanyl, codeine, salvinorin A and hydromorphone were given subcutaneously (s.c.); all other drugs were given intraperitoneally (i.p.). All drugs were given at an injection volume of 10.0 ml/kg.

The classification of the drugs (Table 1) could be based on a pharmacological structure-activity basis, therapeutic application or both. Since our goal was to predict a drugs clinical relevance, we have classified drugs based largely upon their clinical application in the case of known therapeutic drugs, or their affective psychopharmacological effects in the case of experimental or abused drugs. Several notable examples of the discrepancy between known pharmacological and known classified therapeutic or commonly characterized affective psychopharmacological effect were buproprion (DA uptake inhibitor in antidepressant class) and Salvinorin A (kappa opioid agonist in psychotomimetic class). We left Psychomotor Stimulants and α-adrenergics as pharmacological classes since their utility in the therapeutic setting did not cover most of the tested drugs in the case of the Psychomotor Stimulants and their utility in psychiatry was limited, nevertheless a factor in some drug profiles, in the case of the adrenergics.

In addition to the potential classification conflicts cited above, it is recognized that many of the drugs are used for more than one indication. In all cases we have utilized the primary indication for classification.

PA analysis

The PA algorithm was described in detail in Kafkafi et al. (2008) and in Kafkafi and Elmer (2009). Briefly, each path coordinate out of the progression of the animal was represented by a 10 dynamical attributes, such as speed, acceleration, direction of movement and direction change. The range of each attribute is partitioned into several indexed bins, and patterns of movement are then defined as combinations of bins from one or more attributes. We code these patterns using the same order as in the list of attributes, and using asterisks (standing for “wildcards”) to denote attributes that can accept any value, and are therefore irrelevant to the specific pattern definition. As more attributes are added to the definition of a pattern it becomes more and more specific, e.g. the four-attribute pattern P{*,*,1,2,*,1,5,*,*,*} means “moving very slowly while slightly decelerating in the direction of the arena wall but turning sharply away from it”. As in previous work we do not consider patterns of more than 4 attributes, because this would amount to an astronomical number of combinations, most of them so over-specified that they rarely occur in normal behavior. All possible bin combinations of up to 4 attributes amount to a total of 73,042 different behavior patterns. The animal use of each pattern is computed as the total time it spent in this pattern, expressed as percentage out of its total progression time during the session, using a logit transformation, commonly used to transform ratios. Mining of patterns is performed by testing the difference between experimental groups in each pattern, employing the animal use as the dependent variable.

Mining and validation strategy

In the study we use data mining to establish a classification model (for comprehensive introduction see Tan et al, 2006) separating six psychopharmacological classes of clinical importance: antidepressants, antipsychotics, α-adrenergics, psychomotor stimulants, opioids and psychotomimetics. We used data from a previous study (Table 1, dataset I; Kafkafi et al., 2009) as well as new data (Table 1, dataset II). We also utilized another previously-collected dataset of naïve mice of 10 inbred strains across three laboratories (Kafkafi and Benjamini et al., 2005) to further screen discovered predictor behaviors for high heritability and replicability. The data, consisting of path coordinates of the mice in the arena, are first quantified into a large number of behavior patterns, and the frequency of using each pattern by each animal is measured. The classification model is then “trained” by mining for patterns that best discriminate these drug classes. Finally we validate the model by its ability to correctly classify additional drugs that it did not encounter during the training process (Table 1, dataset III), a simulation of novel compound classification in drug discovery.

A hierarchical strategy is adopted in this study in order to navigate the larger number of drug classes. In the first level we mine for general predictors, capable of overall dose-response detection and good separation of all classes in the database. As seen in Fig. 1, just two predictor patterns were sufficient to limit a compound into one, two or (rarely) three likely classes. These remaining ambiguities are then resolved in the second level using “discriminator patterns”, each mined for the best discrimination of just two classes. As in the previous study, patterns were mined mainly by their p-values in standard statistical tests, but the specific tests used in the present study were slightly different (see Results).

Fig. 1.

Fig. 1

Open in a new tab

Separation of classes in the plane of the Universal Drug Detector (horizontal axis) vs the General Class Identifier (vertical axis), in the mining datasets I and II (left) and in the validation dataset III (right). Each arrow denotes a dose-response curve, going out from vehicle through several doses. Each dose group is represented by its median. Axes units are given in both (logit-scaled) percentage of pattern use out of total progression time of the animal in the session (primary axes), and in vehicle standard deviations (secondary axes). Doses are detailed in Table 1, using the same class color coding. The antidepressant, antipsychotic and α-adrenergic drugs are further separated in Fig. 2.

As in previous studies we addressed the statistical issue of multiple comparisons (e.g., Benjamini and Yekutieli, 2005) by employing the most conservative multiple comparisons criterion, the Bonferroni criterion. This criterion is a corrected significance level of α/n, where n is the number of comparisons, thus ensuring that the probability of even a single “false positive” is less than α. The comparison of 73,042 different patterns using α=0.05 yields a Bonferroni criterion of p<0.05/73042=6.8 × 10−7. Except for a few cases specifically noted (see Table S1) the p-values scored by discovered patterns in the mining datasets of this study were more significant than this criterion, often by many orders of magnitude.

Additional mining criteria

Since a large number of patterns usually passed the Bonferroni criterion in the mining process, we used six additional criteria for screening patterns that are likely to prove more reliable: i) frequent use of the pattern by most animals; ii) well-defined dose-response in most drugs; iii) fewer outliers; iv) consistency of vehicle group results over time; v) high broad-sense heritability in naïve animals of different genotypes; vi) high replicability in naïve animals across different laboratories. In some cases patterns that scored higher in these criteria were preferred even when they were not the best scoring patterns in p-value significance. Criterion i was applied by visualizing graphs of pattern significance vs. their average (logit scaled) proportion out of the total progression time, as demonstrated in Kafkafi et al. (2009). Criteria ii to vi were applied by visualizing the graphs of the results over the whole database for each considered pattern. Broad-sense heritability for criterion v was estimated in dataset IV, which included 10 inbred strains across 3 laboratories, using the proportion of variance attributed to the genotype out of the total variance (Kafkafi and Benjamini et al. 2005; Kafkafi et al. 2009). This is a conservative estimation of heritability, since it is measured in a multi-laboratory study while completely excluding the Laboratory×Strain interaction (Kafkafi et al. 2005). Replicability for criterion vi was also estimated in dataset IV, using the proportion of variance attributed to Laboratory×Strain interaction (Kafkafi et al. 2005).

Pattern combination

As demonstrated in our previous study (Kafkafi et al., 2009), we frequently combine together several similar patterns that were isolated in the mining process into more inclusive patterns, providing they were explicitly non-overlapping, had significant p-values in the relevant mining test, and were “neighbors” or “symmetric”. These requirements ensure that patterns are combined only if they were arbitrarily separated by the PA definition procedure to begin with. Thus inclusive patterns still function as concrete coordinations of the original movement attributes, in every sense like their component patterns.

Vehicle correction

Appropriate vehicles were run for all drugs. No significant differences were found between vehicle groups that were run within less than a year from each other. However, in a minority of the 73,042 patterns we detected some “drift” of vehicle results over the six years represented in the database. Since such patterns tend to come up as false positives during mining, we calculated the yearly mean of all vehicle animals in each pattern, and corrected any pattern result in a certain year by subtracting the difference between this year vehicle mean and the first year vehicle mean. This correction was applied only for mining, and was very small in all the patterns isolated in the present study, so for simplicity none of the results shown in the figures include any correction. For drugs of datasets I and II, dose-response curves are shown with the pooled vehicle mean of dataset I and dataset II. In independent validation results of dataset III each dose-response is shown with its own vehicle result.

Drug classification model

A classification model including six classes was constructed, based only on the small set of patterns isolated in the mining process (Results section, Table 2). This model was then independently validated with the 11 compounds of dataset III (see Results). The classification model was not formalized into a quantitative decision algorithm, but rather based on visual examination of the dose response curves of the “unknown” compounds in Figs 1, 2 and 3, and their comparison with the dose response curves of the known drugs in these figures. This is because several properties of these curves are difficult to quantify in a decision algorithm of any kind. Moreover, there is no absolute consensus over definition of psychopharmacological classes, including the classes used in the present study, and each drug has some unique properties and indications. We therefore expect that psychopharmacologists using the model for preclinical drug screening will want to visually compare the dose response curve of each novel compound with specific drugs they already know, rather than trust the decision of a formal algorithm.

Table 2.

Movement patterns included in the classification model in this study. PA notations are according to the system detailed in Kafkafi et al., 2008 and Kafkafi et al., 2009. h2: broad-sense heritability in dataset IV; L×G: Laboratroy×Genotype interaction in dataset IV. For visualization of patterns see animations in the Supplementary Information.

Pattern PA notation Literal description h2
L×G 		Notes
Universal Drug Detector (UDD) P{*, *, *, 5, *, *, *, *, 1, *}
P{*, *, *, 1, *, *, *, *, *, 1}
P{*, *, *, 5, *, *, *, *, 2, *}
P{*, *, *, 1, *, *, *, *, *, 2}		 Strong acceleration during the first second of progression segments OR strong deceleration during the last second of progression segments 49.0%
4.9%		 Dose dependently decreased by all drugs but one, regardless of class. (Fig. 1, 3).
General Class Identifier (GCI) P{*, *, *, *, *, *, 5, 4, 1, 1}
P{*, *, *, *, *, *, 5, 4, 1, 2}
P{*, *, *, *, *, *, 5, 4, 2, 1}
P{*, *, *, *, *, *, 5, 4, 2, 2}
P{*, *, *, *, *, *, 5, 4, 3, 1}
P{*, *, *, *, *, *, 5, 4, 3, 2}
P{*, *, *, *, *, *, 5, 4, 1, 3}
P{*, *, *, *, *, *, 5, 4, 2, 3}
P{*, *, *, *, *, *, 1, 2, 1, 1}
P{*, *, *, *, *, *, 1, 2, 1, 2}
P{*, *, *, *, *, *, 1, 2, 2, 1}
P{*, *, *, *, *, *, 1, 2, 2, 2}
P{*, *, *, *, *, *, 1, 2, 3, 1}
P{*, *, *, *, *, *, 1, 2, 3, 2}
P{*, *, *, *, *, *, 1, 2, 1, 3}
P{*, *, *, *, *, *, 1, 2, 2, 3}		 Very sharp turn in the animal length scale, during less sharp turn in the medium distance scale, during the first OR last second of progression segments 58.1%
5.8%		 Intermittent behavior mainly reflecting ground sniffing. Most increased by α-adrenergic. Most decreased by psychomotor stimulants and opioids (Fig. 1)
Antidepressant/Antipsychotic Discriminator P{1, 2, 1, *, *, 5, *, *, *, *}
P{1, 2, 1, *, *, 1, *, *, *, *}
P{2, 2, 1, *, *, 5, *, *, *, *}
P{2, 2, 1, *, *, 1, *, *, *, *}		 Slow progression towards the wall or away from it, near the wall but not touching it, during the first 40 minutes of the session 45.8%
2.6%		 Small, slow forays from the wall typical to antipsychotics. The time limit avoids the late session inactivity common in high doses of these classes (Fig. 2).
Antidepressant/α-Adrenergic Discriminator P{*, *, *, *, 3, *, 3, *, 3, 3} Straight, regular movement in the middle of progression segments 65.3%
5.3%		 Smooth and regular movement, both in time and space, is decreased only by α-Adrenergics (Fig. 2)
Psychomotor/Opioid Discriminator P{*, *, 1, 2, *, *, *, *, *, *}
P{*, *, 1, 4, *, *, *, *, *, *}		 Medium acceleration or deceleration during slow speed 66.5%
5.0%		 Decreased by opioids (Fig. 3). See also Kafkafi et al (2009)
Open in a new tab

Fig. 2.

Fig. 2

Open in a new tab

Separation of antidepressant, antipsychotic and α-adrenergic classes in the plane of the Antidepressant/α-Adrenergic discriminator pattern (horizontal axis) vs the Antidepressant/Antipsychotic discriminator pattern (vertical axis), in the mining datasets I and II (left) and in the validation dataset III (right). Symbols, units and color coding are as in Fig. 1.

Fig. 3.

Fig. 3

Open in a new tab

Separation of opioid (blue) and psychomotor stimulant (red) in the plane of the Universal Drug Detector (horizontal axis) vs the Opioid-Psychomotor Discriminator (vertical axis), in the mining datasets I and II (left) and in the validation dataset III (right). Symbols, units and color coding are as in Fig. 1.

The classification model has two decision levels. The dose response curve of the unknown compound is first compared to those in Fig. 1, left. If it is closest to the antidepressant/antipsychotic/adrenergic regime, then its dose response curve in Fig. 2 is used for resolving which of these three classes is the more suitable prediction. If the dose response curve in Fig. 1 is closest to the opioid/psychomotor regime, then Fig. 3 is used to decide which of these two classes is the more suitable prediction. If the dose response curve in Fig, 1 the drug is closest to the psychotomimetic regime, then this class is the suitable prediction.

At any level, if the dose response curve of the unknown drug covers several class regime, or if it falls in the middle between several class regimes, then a unanimous prediction is not possible, as in the case of buprenorphine in this study. However, it is usually still possible to limit the prediction to a subset of the six classes.

6. Results

We used PA to develop a classification model separating six psychopharmacological classes of clinical importance: antidepressants, antipsychotics, α-adrenergics, psychomotor stimulants, opioids and psychotomimetics (Table 1). The model is based on newly-collected data as well as a previously-collected dataset (see methods section and Table 1). In the training phase these data, consisting of path coordinates of the mice in the arena, are used to generate a classification model, by mining for patterns that best discriminate the drug classes. This is done in a hierarchical fashion: at the first level we mine for general predictors, capable of overall dose-response detection and good separation of all classes in the database. As seen in Fig. 1, just two predictor patterns were sufficient to limit most compounds into one, two or at most three likely classes. These remaining ambiguities are then resolved at the second level using “discriminator patterns”, each mined for the best discrimination between just two classes (Figs 2 and 3). In the “validation” phase we test the classification model by its ability to correctly classify drugs that it did not encounter during the training phase, a simulation of novel compound classification in drug discovery.

Detection of General Dose Response

An interesting question that is almost impossible to address using traditional psychopharmacological assays is whether there is such a thing as a universal “drugged” behavior, irrespective of the particular class and drug. PA mining (ANOVA by dose within each drug) isolated several similar patterns that detected dose-sensitive changes across all tested drug classes. We combined four non-overlapping such patterns into a single one (Table 2), with a PA definition that literally means “very strong acceleration when starting to move and very strong deceleration when stopping”. The animal use of this behavior, in percentage out of its total progression time in the arena, was dose-dependently decreased by all 41 drugs in our database but one (Fig 1, horizontal axes). The sole exception was haloperidol, in which no significant change was detected (Table S1), but this drug had a very limited dose range due to its tendency to induce strong inactivity even in medium doses. We therefore refer to this pattern as the Universal Drug Detector (UDD).

The decrease in UDD was usually monotonically dose-dependent, except in psychomotor stimulants (Fig. 1 left), which tend to exhibit a U-shape function that is also typical in traditional assays and measures, most likely due to stereotypic behavior at high doses. The strongest UDD effect was found in certain opioids, which decreased it about 10 fold, equivalent to more than 15 vehicle standard deviations (Fig, 1, horizontal axes). The vehicle-injected level of performing UDD, about 25% of the total progression time, replicated that of naïve animals of the same genotype, C57BL/6J, in data set IV across three other laboratories. It was also very similar to that of naïve BALB/cByJ, 129S/SvlJ and A/J, although these genotypes were considerably less active. In comparison, C3H/HeJ, DBA/2J, FVB/NJ and the wild-derived CAST/EiJ and CZECHII/EiJ inbred strains all had higher levels of UDD, about 35% or more. Interestingly, none of the inbred strains in our database performed UDD in the drug-induced range, again suggesting that this can only be achieved by drugs. However, it should be noted that the naïve mouse data cannot measure drug effect in these inbred strains, which may be different than in C57BL/6J (Crowley et al. 2005; Petit-Demouliere et al. 2005).

Its universality makes UDD the preferred pattern for a priori detection of pharmacologically-relevant threshold dosing, and for determining the proper dose range while evaluating any unknown compound. UDD dose-ranges overlapped well with the pharmacologically-relevant dose ranges as published in each drugs class-relevant assay(s) (e.g., analgesia for opioids, swim-test for antidepressants). When combined with additional patterns that are mined for class discrimination (see next sections), UDD ensures that the resulting classification model will be responsive in a pharmacologically-relevant, dose-related manner (Fig. 1 and 3).

Because PA results are concrete coordination patterns of movements, they can readily be visualized in the behavior of any animal in the database. The movement of 7 mice, injected with representative drugs from the 6 classes and with saline, is shown in animation videos (Online Resource, Section 5), in which occurrences of the UDD are highlighted. They illustrate how drugs producing very different behaviors can all decrease UDD.

General Identification of Class

We mined for behaviors that best classified the drugs into classes using ANOVA by class, collapsed over drug and dose. A combination of 16 similar patterns performed best at differentiating most of the classes from each other, especially in combination with UDD (Table 2). Common to all of them is that they are performed during the first or last second of the progression segment, and include small sharp turns overlaying a straighter path in a larger distance scale (see Kafkafi and Elmer 2005a for detailed discussion of the path texture concept). Visualization of specific instances in animations and videos indicate that this meandering, interrupted locomotion (see animations in Online Resource 5) reflects increased floor sniffing and similar scanning movements. We denote this pattern here as the General Class Identifier (GCI).

As seen in Fig. 1, the combination of UDD and GCI separated most classes in a dose-dependent manner, especially psychomotors and opioids, which decreased GCI (vertical axes) by 3 to 10 vehicle standard deviations (SDs), and psychotomimetics, which left it unchanged, while ambiguities remained between antidepressants, antipsychotics and α-adrenergics, all of which increased the use of GCI by 2 to 7 vehicle SDs.

Only four drugs could not be fit into their correct, strictly pharmacological classification in this scheme, or indeed in any other scheme based on two PA patterns that we were able to find (Fig. S4): Salvinorin A, a kappa opioid agonist, and memantine, an NMDA antagonist, are both considered to be psychotomimetic, yet they fit better into the antidepressant/antipsychotic regime. BW373U86, a δ-opioid agonist, fit into the antidepressant regime as well. Bupropion, a therapeutically designated antidepressant, usually fit better in the psychomotor stimulant regime. However, for all these compounds there is indeed evidence for therapeutic utility in the model-identified classes. Memantine, BW373U86 and salvinorin A all have support for utility as antidepressants (Broom et al. 2002; Quan et al. 2011; Braida et al. 2009; but also see Carlezon et al. 2006 for Salvinorin A) and buproprion has utility in ADHD (Pataki et al 2004).

Class Discrimination

The combination of UDD and GCI did not separate antidepressants, antipsychotics and α-adrenergics well (Fig. 1), and thus might prove indecisive when classifying unknown compounds. We therefore also mined for “discriminator” patterns that separate any two of the six classes (Figs. 2, 3). The best antidepressants-antipsychotic discriminator was a combination of 4 patterns (Table 2), together defined as “slow movement to or from the wall, at a distance of 5–15 cm from the wall, during the first 40 minutes of the session”. Antipsychotics dose-dependently increased the use of this pattern by 1 to 6 vehicle SDs, while antidepressants left it unchanged or slightly decreased it (Fig. 2, vertical axis). Visualization of specific instances belonging to this pattern suggests that antipsychotics tend to generate many small and slow incursions away from the wall and back. An increase in this behavior serves as nearly exclusive antipsychotic predictor, while naïve mice from 10 inbred strains (dataset IV) did not even reach vehicle-injected level.

The best Antidepressant/α-Adrenergic discriminator discovered was a single pattern defined as “straight smooth movement in the middle of progression segments”. This behavior included regular, straight and smooth locomotion, which was disrupted mainly by α-adrenergics (Fig. 2, horizontal axes). The combination of the above two discriminator patterns was effective at separating the antidepressant, antipsychotic and α-adrenergic classes (Fig. 2).

The best discriminator between psychomotor stimulants and opioids (Table 2 and Fig. 3) was the same pattern discovered in our previous study (Kafkafi et al. 2009) to separate psychomotor stimulants from opioids and psychotomimetics. However, it was reaffirmed in the present study using several additional drugs.

Pharmacokinetics is an obvious concern in assessing and differentiating drug effects. Interestingly, despite large differences in the rate of onset and duration of action across these 41 drugs, 4 of the 5 discriminating patterns did not utilize the PA attribute ”time from beginning of session” as a basis for classification. The only discriminating pattern to incorporate this attribute was the Antidepressant/Antipsychotic discriminator (Table 2). In this instance, the time constraint avoids the late session inactivity common to high dose antipsychotics. The time constraint thus focuses the mining procedure to the pharmacologically relevant window that reveals differences in active behavior, not sedation.

Validation in Unknown Compounds

While behaviors singled out in PA studies all pass the most conservative multiple comparisons criterion (see methods: “mining and validation strategy”), the true rate of correct prediction is most reliably estimated by the ability to classify “unknown” compounds, i.e., that were not used for the mining process. We tested 11 unknown compounds as independent validation of the classification model (Fig. 1 right, 2 right and 3 right). The person performing the PA analysis and classification (N.K.) was blind to the identity of these compounds, and was frequently requesting to increase or decrease dosage based on the shaping of results for UDD. The predictions reported here (Table 1) do not deviate in any way from this blind classification. As expected, all validation compounds significantly decreased UDD (Table S2). Out of the 11 compounds, 8 were correctly classified into a single class (Table 1, last column). One opioid compound (buprenorphine) was classified as “opioid OR psychomotor stimulant” because its dose-response curve overlapped both the opioid and psychomotor regimes (Fig. 3 right). Only 2 validation compounds were incorrectly classified: in Fig. 1, right, the antidepressant venlafaxine fits the opioid regime, far from the antidepressant domain, and the psychomotor stimulant apomorphine fits the α-adrenergic regime.

Interestingly, while venlafaxine’s mechanism of action is predominantly NE/5HT uptake inhibition, it has shown classic analgesic effects mediated by kappa-opioid, δ-opioid and α2-adrenergic mechanisms, despite the fact that it does not bind to opioid receptors (Schreiber et al. 1999). While tricyclic antidepressants are known to provide some analgesic relief when given alone and supplemental relief when combined with an opioid, it was the only antidepressant compound tested with demonstrable indirect opioid action. A “misclassification” such as this could prove useful given that venlafaxine may be considered unique in its efficacy against treatment-resistant or psychopathic depression (Dubovsky and Thomas 1992; Coryell 1996). Novel antidepressants that provide an opioid signal may identify similarly efficacious antidepressants. Apomorphine may have been incorrectly classified because it is a direct dopamine agonist (D1 and D2), whereas the classification model was trained only with indirect dopamine agonists (dopamine uptake inhibitors). The fact that the classification model could discriminate this difference may prove useful in future studies as subclasses of direct and indirect DA agonists are investigated.

7. Discussion

Our results show that the established classification model is able to achieve versatility far beyond those of standard pre-clinical and clinical behavioral assays. We are not aware of any other single behavioral assay, in animal models or humans, that can consistently and dose-dependently discriminate six different psychopharmacological classes. This was achieved using small sample sizes, in a high-throughput setup that does not require manual scoring or training sessions. While the classification model does not yet include other mouse genotypes, our results imply high reliability, since the isolated patterns were also mined for high broad-sense heritability and high replicability across laboratories in a dataset of naïve animals from 10 inbred strains (Table 2). Thus, the classification model offers significant potential as a first stage in vivo screening paradigm in psychiatric drug discovery.

Several important outcomes of the approach should be emphasized, especially in comparison with both standard assays and other behavioral data mining methods. First, only a small number of behaviors were required for successful class prediction. Moreover, due to the way PA was designed, these predictors are not abstract mathematical combinations of general behavioral endpoints (e.g., 11). Rather they are concrete coordination patterns of concurrent movements. Therefore they can readily be visualized in the behavior of any animal in the database (see videos in Online Resource section 5), and potentially identified in other experimental setups, in order to study them and gain greater insight into their meaning.

Second, a striking example of the power of the approach is the discovered UDD pattern, a behavior that was dose-dependently decreased by 40 out of 41 drugs regardless of class. We are not aware of any known indicator of psychopharmacological drug effect that is so general. Visualizing UDD under the influence of drugs that produce very different behaviors (e.g. methylphenidate, morphine or fluoxetine) suggests a deficiency in acceleration performances, detected by a failure to apply strong forces exactly when they are required, such as while initiating a progression movement or while stopping. This was true even in very fast animals under the effects of psychomotor stimulants. In contrast, drug naïve animals from even the most inactive inbred strains had higher UDD than drug-injected animals, and the wild-derived inbred strains had the highest measured UDD. Thus, it appears that UDD detects any deviation from a chemical balance of the CNS that is in some sense “optimal” for the mouse. Due to its generality, UDD is valuable for assessing dose range and threshold in novel compounds even without apriori knowledge of their effect, and can be employed to compare dose responses of different drugs on the same scale.

Third, when compared with our previous PA study (Kafkafi et al. 2008), the present study demonstrates the feasibility of systematically updating and increasing the power of the classification model, simply by adding compounds and classes to the database. The approach is thus an open-ended strategy, in contrast with standard behavioral assays and animal models, and will continue to improve its performance. Further increase of the number and diversity of compounds belonging to current classes may enable the discrimination of subclasses. The 9 antipsychotic drugs in this study are already sufficient for proposing behaviors capable of discriminating 1st generation from 2nd generation antipsychotics (see Fig. S5). PA can also be tailored to address many specific research questions. Classification models can be updated based upon new data, and configured to specific research questions. For example, it could be configured to discern drug effects based upon on recent discoveries using NMDA antagonist in depression, or to reveal commonalities in drugs with abuse liability.

PA’s utility and conceptual implementation extends beyond that explored in this report. A direct extension of PA application is in the area of repurposing – searching for latent therapeutic applications of established drugs. In general, the approach has a special advantage over standard assays when apriori knowledge of the expected effect is lacking. For example, PA may prove useful to a) explore novel genotype-dependent patterns predicting specific beneficial or adverse drug effects (Crowley et al. 2005, Petit-Demouliere et al. 2005, Kafkafi and Elmer 2005b), thus availing the database to genomics interrogation, b) explore subtle differences in genetically modified animal models (Crawley 2007, Kafkafi et al. 2008) and c) explore unique reliable patterns associated with disease-induction models such as adult and prenatal chronic unpredictable stress (in depression, Willner et al. 2005; and schizophrenia, Koenig et al. 2005). Once unique and specific patterns are identified in disease models, these patterns could be used to ascertain therapeutic efficacy of known, repurposed and novel compounds and determine the consequences of chronic drug administration especially in drug thought to require repeated administration to obtain therapeutic efficacy.

Psychopharmacology, neurophysiology, behavior genetics and additional fields all suffer from chronic methodological difficulties in identifying and measuring behavioral effects (Rihel et al. 2010; Benjamini et al. 2010). The power to identify these effects may be substantially improved by radically expanding the “field of view”, that is, by increasing the number of alternative behavioral measures by several orders of magnitude. Data mining methods for searching multiple endpoints in parallel are well-established at the molecular and cellular levels, however their application at the behavioral level is far from trivial. Our study demonstrates a powerful approach to neuropsychopharmacological drug assessment using a data mining approach that could facilitate drug discovery in the complex field of psychiatric disorders.

Supplementary Material

213_2013_3230_MOESM1_ESM

Fig. S1: 4 drugs in datasets I and II that could not be fitted into the classification in Fig. 1, superimposed over the results shown in Fig. 1.

Fig. S2: Separation of 1st generation (red) and 2nd generation (blue) antipsychotics in the plane of two of the best discriminator patterns between them.

Table S1: UDD effect size and significance in each of the 30 mining drugs (datasets I and II), using one-way ANOVA by dose. Identical vehicle groups were pooled within each dataset. NS: no significance; *: p < 0.05; **: p < Bonferroni criterion = 6.85×10−7

Table S2: UDD effect size and significance in each of the 11 validation drugs (dataset III), using one-way ANOVA by dose. Identical vehicle groups were pooled.

Acknowledgments

This work was funded by NIDA/NIH research grant #DA025647. We would also like to acknowledge Drs. Ilan Golani, Yoav Bemjamini, Daniel Yekutieli and R.W. Buchanan for their helpful comments.

Footnotes

The authors declare no conflict of interest.

Contributor Information

Neri Kafkafi, Department of Zoology, Tel Aviv University, Israel

Cheryl L. Mayo, Maryland Psychiatric Research Center, Department of Psychiatry, University of Maryland, Maryland, Baltimore, USA

Greg I. Elmer, Maryland Psychiatric Research Center, Department of Psychiatry, University of Maryland, Maryland, Baltimore, USA

References

  1. Agid Y, Buzsáki G, Diamond DM, Frackowiak R, Giedd J, Girault JA, Grace A, Lambert JJ, Manji H, Mayberg H, Popoli M, Prochiantz A, Richter-Levin G, Somogyi P, Spedding M, Svenningsson P, Weinberger D. How can drug discovery for psychiatric disorders be improved? Nat Rev Drug Discov. 2007;6:189–201. doi: 10.1038/nrd2217. [DOI] [PubMed] [Google Scholar]
  2. Benjamini Y, Lipkind D, Horev G, Fonio E, Kafkafi Neri, Golani I. Ten ways to improve the quality of descriptions of whole-animal movement. Neurosci Biobehav Rev. 2010;34(8):1351–1365. doi: 10.1016/j.neubiorev.2010.04.004. [DOI] [PubMed] [Google Scholar]
  3. Braida D, et al. Potential anxiolytic- and antidepressant-like effects of salvinorin A, the main active ingredient of Salvia divinorum, in rodents. Br J Pharmacol. 2009;157:844–853. doi: 10.1111/j.1476-5381.2009.00230.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Broom DC, et al. Convulsant activity of a non-peptidic delta-opioid receptor agonist is not required for its antidepressant-like effects in Sprague-Dawley rats. Psychopharmacol (Berl) 2002;164:42–48. doi: 10.1007/s00213-002-1179-y. [DOI] [PubMed] [Google Scholar]
  5. Brunner D, Nestler E, Leahy E. In need of high-throughput behavioral systems. Drug Discov Today. 2002;7(18 Suppl):S107–S112. doi: 10.1016/S1359-6446(02)02423-6. [DOI] [PubMed] [Google Scholar]
  6. Carlezon WA, Jr, et al. Depressive-like effects of the kappa-opioid receptor agonist salvinorin A on behavior and neurochemistry in rats. J Pharmacol Exp Ther. 2006;316:440–447. doi: 10.1124/jpet.105.092304. [DOI] [PubMed] [Google Scholar]
  7. Conn PJ, Roth BL. Opportunities and challenges of psychiatric drug discovery: roles for scientists in academic, industry, and government settings. Neuropsychopharmacol. 2008;33:2048–2060. doi: 10.1038/sj.npp.1301638. [DOI] [PubMed] [Google Scholar]
  8. Coryell W. Psychotic depression. J Clin Psychiatry. 1996;57(Suppl 3):27–31. discussion 49. [PubMed] [Google Scholar]
  9. Crawley JN. What’s Wrong With My Mouse?: behavioral phenotyping of transgenic and knockout mice. 2. John Wiley & Sons; NJ: 2007. [Google Scholar]
  10. Crowley JJ, Blendy JA, Lucki I. Strain-dependent antidepressant-like effects of citalopram in the mouse tail suspension test. Psychopharmacology. 2005;183:257–264. doi: 10.1007/s00213-005-0166-5. [DOI] [PubMed] [Google Scholar]
  11. Drai D, Golani I. SEE: a tool for the visualization and analysis of rodent exploratory behavior. Neurosci Biobehav Rev. 2001;25:409–426. doi: 10.1016/s0149-7634(01)00022-7. 0.1016/S0149–7634(01)00022–7. [DOI] [PubMed] [Google Scholar]
  12. Dubovsky SL, Thomas M. Psychotic depression: advances in conceptualization and treatment. Hosp Community Psychiatry. 1992;43:1189–1198. doi: 10.1176/ps.43.12.1189. [DOI] [PubMed] [Google Scholar]
  13. Geyer MA, Russo PV, Masten VL. Multivariate assessment of locomotor behavior: pharmacological and behavioral analyses. Pharmaco Biochem Behav. 1986;25(1):277–88. doi: 10.1016/0091-3057(86)90266-2. [DOI] [PubMed] [Google Scholar]
  14. Golub TR, et al. Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science. 1999;286:531–537. doi: 10.1126/science.286.5439.531. [DOI] [PubMed] [Google Scholar]
  15. Gunther EC, Stone DJ, Gerwien RW, Bento P, Melvyn PH. Prediction of clinical drug efficacy by classification of drug-induced genomic expression profiles in vitro. Proc Natl Acad Sci USA. 2003;100:9608–9613. doi: 10.1073/pnas.1632587100. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Henry BL, Minassian A, Young J, Paulus MP, Geyer MA, Perry W. Cross-species assessments of motor and exploratory behavior related to bipolar disorder. Neurosci Biobehav Rev. 2010;34:1296–1306. doi: 10.1016/j.neubiorev.2010.04.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Kafkafi N, Elmer GI. Texture of locomotor path: a replicable characterization of a complex behavioral phenotype. Genes Brain Behav. 2005a;4(7):431–443. doi: 10.1111/j.1601-183X.2005.00126.x. [DOI] [PubMed] [Google Scholar]
  18. Kafkafi N, Elmer GI. Activity density in the open field: a measure for differentiating the effect of psychostimulants. Pharmacol Biochem Behav. 2005b;80(2):239–249. doi: 10.1016/j.pbb.2004.11.004. [DOI] [PubMed] [Google Scholar]
  19. Kafkafi N, Benjamini Y, Sakov A, Elmer GI, Golani I. Genotype-environment interactions in mouse behavior: a way out of the problem. Proc Natl Acad Sci U S A. 2005;102(12):4619–4624. doi: 10.1073/pnas.0409554102. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Kafkafi N, Yekutieli D, Yarowsky P, Elmer GI. Data mining in a behavioral test detects early symptoms in a model of amyotrophic lateral sclerosis. Behav Neurosci. 2008;122(4):777–787. doi: 10.1037/0735-7044.122.4.777. [DOI] [PubMed] [Google Scholar]
  21. Kafkafi N, Yekutieli D, Elmer GI. A data mining approach to in vivo classification of psychopharmacological drugs. Neuropsychopharmacol. 2009;34(3):607–623. doi: 10.1038/npp.2008.103. [DOI] [PubMed] [Google Scholar]
  22. Koenig JI, Elmer GI, Shepard PD, Lee PR, Mayo C, Joy B, Hercher E, Brady DL. Prenatal exposure to a repeated variable stress paradigm elicits behavioral and neuroendocrinological changes in the adult offspring: potential relevance to schizophrenia. Behav Brain Res. 2005;156(2):251–61. doi: 10.1016/j.bbr.2004.05.030. [DOI] [PubMed] [Google Scholar]
  23. Nestler EJ, Hynan SE. Animal models of neuropsychiatric disorders. Nature Neurosci. 2010;13:1161–1169. doi: 10.1038/nn.2647. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Pangalos MN, Schechter LE, Hurko O. Drug development for CNS disorders: strategies for balancing risk and reducing attrition. Nat Rev Drug Discov. 2007;6(7):521–532. doi: 10.1038/nrd2094. [DOI] [PubMed] [Google Scholar]
  25. Pataki CS, Feinberg DT, McGough JJ. New drugs for the treatment of attention-deficit/hyperactivity disorder. Expert Opin Emerg Drugs. 2004;9:293–302. doi: 10.1517/14728214.9.2.293. [DOI] [PubMed] [Google Scholar]
  26. Petit-Demouliere B, Chenu F, Bourin M. Forced swimming test in mice: a review of antidepressant activity. Psychopharmacology. 2005;177:245–255. doi: 10.1007/s00213-004-2048-7. [DOI] [PubMed] [Google Scholar]
  27. Quan MN, Zhang N, Wang YY, Zhang T, Yang Z. Possible antidepressant effects and mechanisms of memantine in behaviors and synaptic plasticity of a depression rat model. Neurosci. 2011;182:88–97. doi: 10.1016/j.neuroscience.2011.03.026. [DOI] [PubMed] [Google Scholar]
  28. Rihel J, et al. Zebrafish behavioral profiling links drugs to biological targets and rest/wake regulation. Science. 2010;327(5963):348–351. doi: 10.1126/science.1183090. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Schopp DT. Where will new neuroscience therapies come from? Nat Rev Drug Discov. 2011;10:715–716. doi: 10.1038/nrd3559. [DOI] [PubMed] [Google Scholar]
  30. Schreiber S, Backer MM, Pick CG. The antinociceptive effect of venlafaxine in mice is mediated through opioid and adrenergic mechanisms. Neurosci Lett. 1999;273:85–88. doi: 10.1016/s0304-3940(99)00627-8. [DOI] [PubMed] [Google Scholar]
  31. Tan PN, Steinbach M, Kumar V. Classification: basic concepts, decision trees and model evaluation. Addison–Wesley; Boston: 2006. Introduction to Data Mining. [Google Scholar]
  32. Tecott LH, Nestler EJ. Neurobehavioral assessment in the information age. Nat Neurosci. 2004;7:462–466. doi: 10.1038/nn1225. [DOI] [PubMed] [Google Scholar]
  33. Willner P. Chronic mild stress (CMS) revisited: consistency and behavioural-neurobiological concordance in the effects of CMS. Neuropsychobiology. 2005;52(2):90–110. doi: 10.1159/000087097. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

213_2013_3230_MOESM1_ESM

Fig. S1: 4 drugs in datasets I and II that could not be fitted into the classification in Fig. 1, superimposed over the results shown in Fig. 1.

Fig. S2: Separation of 1st generation (red) and 2nd generation (blue) antipsychotics in the plane of two of the best discriminator patterns between them.

Table S1: UDD effect size and significance in each of the 30 mining drugs (datasets I and II), using one-way ANOVA by dose. Identical vehicle groups were pooled within each dataset. NS: no significance; *: p < 0.05; **: p < Bonferroni criterion = 6.85×10−7

Table S2: UDD effect size and significance in each of the 11 validation drugs (dataset III), using one-way ANOVA by dose. Identical vehicle groups were pooled.

NIHMS517218-supplement-213_2013_3230_MOESM1_ESM.docx (236.8KB, docx)

RESOURCES