Working memory capacity of crows and monkeys arises from similar neuronal computations — TEI-tagged view

HahnpersName, BalakhonovpersName, FongaropersName, NiederpersName, RosepersName, RichpersName, FrankpersName, ColombopersName (2021date). “Working memory capacity of crows and monkeys arises from similar neuronal computations.” eLifetitle·j.

Object of study — whether the neuronal computation that limits working memory capacityterm·dom is the same in a birdterm and a monkeyterm, despite entirely different brain architecturesterm

Research question — whether modelsterm of working memory capacityterm·dom built and tested only in mammals describe working memory as such, or only its mammalianterm implementation

Contribution — divisive normalizationterm·def is found in a non-layered avian associative area, with loadterm effects virtually identical to monkeyterm prefrontal cortex

Full text — tagged

The reference list is not part of the text we hold for this paper. Our copy was extracted from the publisher's page and the bibliography did not survive that extraction, so the article below ends before its references. Nothing has been removed by us.

Abstract Complex cognition relies on flexible working memory, which is severely limited in its capacityterm·dom. The neuronal computations underlying these capacityterm·dom limits have been extensively studied in humans and in monkeysterm, resulting in competing theoretical modelsterm. We probed the working memory capacityterm·dom of crowsterm ( Corvus corone ) in a change detection taskterm, developed for monkeysterm ( Macaca mulatta ), while we performed extracellular recordings of the prefrontal-like area nidopallium caudolaterale. We found that neuronal encoding and maintenance of informationterm were affected by item loadterm, in a way that is virtually identical to results obtained from monkeyterm prefrontal cortex. Contemporary neurophysiological modelsterm of working memory employ divisive normalizationterm·def as an important mechanism that may result in the capacityterm·dom limitation. As these modelsterm are usually conceptualized and tested in an exclusively mammalianterm context, it remains unclear if they fully capture a general concept of working memory or if they are restricted to the mammalianterm neocortexterm. Here, we report that carrion crowsterm and macaque monkeysterm share divisive normalizationterm·def as a neuronal computation that is in line with mammalianterm modelsterm. This indicates that computational modelsterm of working memory developed in the mammalianterm cortex can also apply to non-cortical associative brain regions of birdsterm.

Research organism: Other eLife digest Working memory is the brain’s ability to temporarily hold and manipulate informationterm. It is essential for carrying out complex cognitive tasksterm, such as reasoning, planning, following instructions or solving problems. Unlike long-term memory, informationterm is not stored and recalled, but held in an accessible state for brief periods. However, the capacityterm·dom of working memory is very limited. Humans, for example, can only hold around four items of informationterm simultaneously.

There are various competing theories about how this limitation arises from the network of neurons in the brain. These modelsterm are based on studies of humans and other primates. But memory limitations are not exclusive to mammals. Indeed, the working memory of some birdsterm, such as crowsterm, has a similar capacityterm·dom to humans despite the architectureterm of their brains being very different to mammals. So, how do brains with such distinct structural differences produce working memories with similar capacities?

To investigate, HahnpersName et al. probed the working memory of carrion crowsterm in a change detection taskterm developed for macaque monkeysterm. Crowsterm were trained to memorize varying numbers of colored squares and indicate which square had changed after a one second delayterm when the screen went blank. While the crowsterm performed the taskterm, HahnpersName et al. measured the activity of neurons in an area of the brain equivalent to the prefrontal cortex, the central hub of cognition in mammals.

The experiments showed that neurons in the crowterm brain responded to the changing colorsterm virtually the same way as neurons in monkeysterm. HahnpersName et al. also noticed that increasing the number of items the crowsterm had to remember affected individual neurons in a similar fashion as had previously been observed in monkeysterm.

This suggests that birdsterm and monkeysterm share the same central mechanisms of, and limits to, working memory despite differences in brain architectureterm. The similarities across distantly related species also validates core ideas about the limits of working memory developed from studies of mammals.

Introduction Working memory (WMterm·dom) can hold informationterm for a short period of time to allow further processing in the absence of sensory input ( Cowan, 2017 ; Oberauer et al., 2018 ). By bridging this gap between the immediate sensory environment and behavior, WMterm·dom is a cornerstone for complex cognition. It is a very flexible memory system, yet severely limited in its capacityterm·dom. While this capacityterm·dom is often seen as a general cognitive bottleneck, for simple stimuli, like colorsterm, the capacityterm·dom is very similar between humans, monkeysterm, and crowsterm ( BalakhonovpersName and RosepersName, 2017 ; Buschman et al., 2011 ; Cowan, 2001 ; Luck and Vogel, 1997 ). Different modelsterm have been proposed to conceptualize how this capacityterm·dom limit arises. This work motivated many psychophysical and electrophysiological experiments that in turn led to a spectrum of more refined modelsterm of WMterm·dom ( Ma et al., 2014 ). ‘Discrete modelsterm·def’ of WMterm·dom argue that a fixed number of items can be stored. Once this capacityterm·dom is reached, an additional item can only be maintained if it replaces a previous item ( Awh et al., 2007 ; Fukuda et al., 2010 ; Luck and Vogel, 1997 ; Vogel and Machizawa, 2004 ; Zhang and Luck, 2008 ). ‘Continuous modelsterm·def’ describe WMterm·dom as a flexible resourceterm·dom that is allocated to individual items. A minimum amount of this resourceterm·dom has to be allocated to each item for successful retention, thereby resulting in a capacityterm·dom limit ( Bays and Husain, 2008 ; van den Berg et al., 2012 ; Wilken and Ma, 2004 ). On the neurophysiological level, modelsterm of WMterm·dom capacityterm·dom suggest that interference between memory representations (‘ items ’) within the neuronal network is a source of informationterm loss and capacityterm·dom limitation ( Bouchacourt and Buschman, 2019 ; Lundqvist et al., 2016 ; Lundqvist et al., 2011 ; Schneegans et al., 2020 ). Interference may arise due to divisive normalizationterm·def that appears as competition between items, related to oscillatory dynamics ( Lundqvist et al., 2016 ; Lundqvist et al., 2011 ), WMterm·dom flexibility ( Bouchacourt and Buschman, 2019 ), and neuronal informationterm sampling ( Schneegans et al., 2020 ). Divisive normalizationterm·def is a computational principle that acts upon neurons when presenting multiple stimuli simultaneously, it normalizes neuronal responses by creating ‘a ratio between the response of an individual neuron and the summed activity of a pool of neurons’ ( Carandini and Heeger, 2011 , p. 51). An effect related to divisive normalizationterm·def can be observed when two stimuli are presented either individually or simultaneously within the receptive field of a visual sensory neuron. The neuron’s firing rate when the stimuli are presented simultaneously becomes normalized by the populations’ responses to each individual stimulus ( Carandini et al., 1997 ; Heeger, 1992 ). This effect also occurs in relation to attentive processes and has been formalized into the ‘normalizationterm·dom modelterm of attentionterm·dom’ ( Reynolds et al., 1999 ; Reynolds and Heeger, 2009 ). Normalizationterm·dom of neuronal responses is commonly observed in many species throughout the animal kingdom, not just in sensory, but also in cognitive domains ( Carandini and Heeger, 2011 ). Investigations into WMterm·dom capacityterm·dom and modelterm predictions focus mostly on humans and monkeysterm. By extending this work to include birdsterm, one can gain a unique comparative perspective. Crowsterm have a similar limit in WMterm·dom capacityterm·dom and neuronal correlates of WMterm·dom are comparable to monkeysterm’ ( BalakhonovpersName and RosepersName, 2017 ; NiederpersName, 2017 ). But while the neuronal architectureterm of sensory areas is similar between birdsterm and mammals, higher associative areas, critical for WMterm·dom, do not share a common architectureterm between the species ( Stacho et al., 2020 ). Therefore, an outstanding question is whether modern modelsterm of WMterm·dom such as the ‘flexible modelterm’ capture WMterm·dom capacityterm·dom in general, or if their predictions (e.g., divisive normalizationterm·def) are confined to the mammalianterm neocortexterm. To resolve this, it is crucial to investigate the avian brain to understand how its different organization can produce such similar behavioral and neurophysiological results. While the neuronal correlates of WMterm·dom maintenance in birdsterm have been investigated in some detail ( Diekamp et al., 2002a ; Hartmann et al., 2018 ; Rinnert et al., 2019 ; RosepersName and ColombopersName, 2005 ; Veit et al., 2014 ), a neurophysiological investigation of WMterm·dom capacityterm·dom limitation is still lacking. The avian forebrain structure, nidopallium caudolaterale (NCL)term·def is a critical component of avian WMterm·dom. The NCLterm is considered functionally equivalent to the mammalianterm prefrontal cortex (PFCterm) ( Güntürkün and Bugnyar, 2016 ; NiederpersName, 2017 ), as it receives projections from all sensory modalities ( Kröner and Güntürkün, 1999 ), projects to premotor areas ( Kröner and Güntürkün, 1999 ), and is a target of dopaminergic innervation ( Waldmann and Güntürkün, 1993 ). To investigate the neurophysiology of WMterm·dom capacityterm·dom in birdsterm, we adopted a taskterm design developed for monkeysterm ( Buschman et al., 2011 ) to use it with carrion crowsterm ( Corvus corone ). Our animals were trained to memorize an array of colorsterm and to indicate which colorterm had changed after a short memory delayterm, while we performed extracellular recordings of individual neurons in the NCLterm using multi-channel probes. We expected to find a clear correlate of WMterm·dom representations in NCLterm neurons and a load-dependent response modulation based on divisive normalizationterm·def of neuronal responses. This would allow us to evaluate if the behavioral WMterm·dom capacityterm·dom observations of crowsterm fit a ‘discrete’ or ‘continuous’ WMterm·dom resourceterm·dom modelterm. If the neuronal responses also fit the contemporary neurophysiological modelsterm of WMterm·dom capacityterm·dom limitations ( Bouchacourt and Buschman, 2019 ; Lundqvist et al., 2016 ; Lundqvist et al., 2011 ; Schneegans et al., 2020 ), it would further suggest that crowsterm and monkeysterm have convergently evolved a similar neurophysiological basis for WMterm·dom capacityterm·dom despite a different architectureterm of the critical forebrain structures.

Results The WMterm·dom capacityterm·dom of crowsterm is similar to that of monkeysterm The behavioral performance was influenced by the number of colored squaresterm·var on the screen. It significantly decreased with an increasing number of ipsilateralterm squares (median performances, loadterm 1: 95.88%, loadterm 2: 78.31%, loadterm 3: 58.21%; Friedman testterm·stat: Χ ² = 92.00, p < 0.001, Figure 1B ). We ran a generalized linear modelterm·stat with ipsilateral loadterm·var (i.e., loadterm of hemifieldterm where a colorterm changed), contralateral loadterm·var (i.e., loadterm of hemifieldterm without a colorterm change) and their interactionterm as predictors for performance ( R 2 adj 0.78, F (460,456) = 555.00, p < 0.001). We found that the number of ipsilateralterm colorsterm significantly reduced performance ( β ipsi –0.177, t (459) = –18.00, p < 0.001), whereas the number of contralateralterm colorsterm did not ( β contra –0.021, t (459) = –1.77, p = 0.0772; Figure 1C ). There was also a significantterm interactionterm between ipsilateralterm and contralateral loadterm·var ( β = –0.024, t (458) = –4.28, p < 0.001). We compared this modelterm to a reduced modelterm, where we omitted the non-significant β contra and found that this reduced modelterm ( R 2 adj 0.78, F (460,457) = 828.00, p < 0.001) explained the performance equally as well (|ΔLLR| = 0.0102). Therefore, we conclude that contralateral loadterm·var by itself did not significantly affect performance. We calculated the capacityterm·dom K (see Materials and methods) for all full WMterm·dom loadsterm (i.e., two to five items). The capacityterm·dom K peaks at four items (mean ± SEM: 3.05±.038, Figure 1—figure supplement 1 ). These observations are very similar to observations made in primates ( Buschman et al., 2011 ) and fully reproduce our earlier behavioral findings ( BalakhonovpersName and RosepersName, 2017 ).

Figure 1. Behavioral overview.

( A ) Behavioral paradigm (reproduced from BalakhonovpersName and RosepersName, 2017 ). The birdsterm had to center and hold their gaze for the duration of the sampleterm and delayterm period, and subsequently indicate which colored square had changed. ( B ) Boxplot of performance for different ipsilateral loadsterm·var (i.e., on the side where the change occurred). Horizontal lines indicate significantterm differences between loadsterm, box indicates the median, 1st, and 3rd quartile (whiskers extend to 1.5 times the inter-quartile range). ( C ) Mean performance matrix for ipsi- and contralateral loadterm·var combinations (values are rounded to the nearest integer). Additional contralateralterm items at an ipsilateral loadterm·var of 1 barely affected performance (bottom row). At higher ipsilateral loadsterm·var additional contralateralterm items reduced performance more clearly (middle and top row). Statistical modeling revealed an interactionterm at these higher loadsterm (see text).

Figure 1—figure supplement 1. Capacityterm·dom of crowterm working memory (WMterm·dom).

Line indicates capacityterm·dom K at different loadsterm. The peak at four items indicates the capacityterm·dom. Dashed lines indicate maximum capacityterm·dom and fixed capacityterm·dom of 1. Error bars indicate the standard error of the mean.

Neurons of the NCLterm encode the colorterm identity and maintain it in WMterm·dom We recorded 362 neurons from the NCLterm of two crowsterm performing the WMterm·dom taskterm (delayed change localization). All reported effects were also present in each individual birdterm ( Figure 3—figure supplements 1 – 2 ), we, therefore, pooledterm·mth the data for population analysis. A large subset of neurons responded to the presence of a colorterm (i.e., at loadterm 1) by substantially increasing or decreasing their firing rate relative to baseline. This change in firing rate occurred selectively, depending on the presented colorterm either in the sampleterm ( Figure 2A ) or the delayterm period ( Figure 2—figure supplement 1 ). For most neurons, this difference in firing rate between the two possible colorsterm became attenuated when the loadterm increased from one to two colorsterm, and it was further attenuated from two to three colorsterm. To quantify this effect, we calculated the amount of informationterm about the colorterm identity at a neuron's favorite locationterm·def as the percent explained varianceterm·var (PEV) during a memory loadterm of one, two, or three items in bins of 200 ms (see Materials and methods for details). Most neurons did not sustain informationterm about colorterm (measured as a significantterm PEV, henceforth ‘ informationterm ’) throughout the entire sampleterm or memory delayterm but rather had shorter periods in which the informationterm was significantterm ( Figure 2A bottom).

Figure 2. Colorterm discrimination in the neuronal response (informationterm, percent explained varianceterm·var [PEV]) generally decreases with loadterm, but some neurons show the opposite effect.

Shown are the three ipsilateral loadterm·var conditions (i.e., loadterm increases on the same side as the neuron’s favorite locationterm·def). Ipsilateral loadsterm·var are one (blue), two (yellow), and three (red). The labels ‘colorterm A’ and ‘colorterm B’ always refer to the same pair of colorsterm at the neuron’s favorite locationterm·def, irrespective of the loadterm condition. ( A ) Example of a sampleterm neuron with colorterm informationterm decline at loadterm 1 (blue), loadterm 2 (yellow), and loadterm 3 (red). Top: raster plot, where every dot represents a single spike during the individual trials (rows of dots); middle: peri-stimulus-time histogram (PSTH) of average firing rate (solid line for colorterm ID 1, dashed line for colorterm ID 2) with the standard error of the mean (shaded areas); bottom: percent explained varianceterm·var of colorterm identity (a measure of informationterm about colorterm) along the trial, the line at the top of the y-axis indicates significantterm bins. ( B ) Same as in ( A ) for an example of a delayterm neuron with informationterm gain at a higher loadterm.

Figure 2—figure supplement 1. Colorterm discrimination in the neuronal response (informationterm, percent explained varianceterm·var [PEV]) decreases with loadterm.

Example of a delayterm neuron with colorterm informationterm decline, at loadterm 1 (blue), loadterm 2 (green), and loadterm 3 (red). Top: raster plot, where every dot represents a single spike during the individual trials (rows of dots); middle: peri-stimulus-time histogram (PSTH) of average firing rate (solid line for colorterm ID 1, dashed line for colorterm ID 2) with the standard error of the mean (shaded areas); bottom: percent explained varianceterm·var of colorterm identity (a measure of informationterm about colorterm) along the trial, the line at the top of the y-axis indicates significantterm bins.

Figure 2—figure supplement 2. Colorterm discrimination in the neuronal response (informationterm, percent explained varianceterm·var [PEV]) increases with loadterm.

Example of a sampleterm neuron with colorterm informationterm gain; at loadterm 1 (blue), at loadterm 2 (green), and loadterm 3 (red). Top: raster plot, where every dot represents a single spike during the individual trials (rows of dots); middle: peri-stimulus-time histogram (PSTH) of average firing rate (solid line for colorterm ID 1, dashed line for colorterm ID 2) with the standard error of the mean (shaded areas); bottom: percent explained varianceterm·var of colorterm identity (a measure of informationterm about colorterm) along the trial, the line at the top of the y-axis indicates significantterm bins.

To better capture the time points when the individual neurons carried informationterm, we performed a hierarchical clustering analysisterm·stat of the PEV values of the individual neurons at loadterm 1 (see Materials and methods for details). We found a total of seven clusters that were organized into two overarching groups ( Figure 3A ). Group 1 contained neurons ( n = 227) that showed peak informationterm during the sampleterm and early delayterm phase, while group 2 contained neurons ( n = 135) that showed peak informationterm during the delayterm phase. For each neuron, we then calculated if it carried a significantterm amount of colorterm informationterm by applying a permutation testterm·stat (for all bins at loadterm 1, see Materials and methods). The individual neurons were then further classified into three groups depending on the phase in which they had a significantterm amount of informationterm ( Figure 3B ). Overall, 37.57% ( n = 136) of neurons were significantterm during the sampleterm phase, 9.39% ( n = 34) of neurons were significantterm during the memory delayterm, and 14.64% ( n = 53) of neurons were significantterm during both the sampleterm phase and the memory delayterm (all proportions of neurons were significantly higher than expected by chance [binomial testterm·stat, see Materials and methods, all p < 0.001]). Refer to Figure 2A for an example neuron, significantterm at loadterm 1 with a large differentiation in firing rate between colorterm identities (a large PEV) and a loss of differentiation with increasing ipsilateral loadterm·var. Further inspection of individual neuronal activity revealed, however, that a substantial number of neurons responded differently. Instead of losing informationterm at higher loadsterm, many neurons gained informationterm (e.g., Figure 2B , Figure 2—figure supplement 2 ). Thus, we additionally performed the permutation testing for loadsterm 2 and 3 to determine which neurons had significantterm informationterm (see Materials and methods). We found that many of the neurons that did not have significantterm informationterm at loadterm 1 did have significantterm informationterm at loadterm 2 and loadterm 3 ( Figure 3C ). For the memory delayterm, more than half of the significantterm neurons we detected were only significantterm for either loadterm 2 or loadterm 3, compared to only 36% of neurons that were significantterm at loadterm 1 ( Figure 3C middle). By including the higher loadsterm in our analysis, we found a total of 249 (68.78%) sampleterm neurons and 94 (25.97%) delayterm neurons. For the population analyses, we subsequently pooledterm·mth all significantterm neurons into three groups (one per loadterm). These pooledterm·mth groups were then each subdivided into sampleterm and delayterm neurons (i.e., ‘sample-load1’, ‘delay-load1’, ‘sample-load2’, etc., see Table 1 in the Materials and methods for an overview).

Figure 3. Overview of recorded neurons.

( A ) The neuronal population can be best described by seven individual clusters. ( B ) Percentages of neurons (total n = 362) with significantterm colorterm informationterm at loadterm 1, during the sampleterm and the delayterm. ( C ) Percentages (rounded) of significantterm neurons in individual loadterm conditions for sampleterm ( n = 249), delayterm ( n = 94), and sampleterm and delayterm ( n = 133). The pieces of the pies depicting significance at a specific loadterm relate to the number of significantterm neurons in the respective phase (e.g., 36% of the 94 delayterm neurons (i.e., 34 neurons) are significantterm at loadterm 1 (all pieces contain blue), these are the same neurons that make up the 9.36% of the total 362 neurons depicted in B).

Figure 3—figure supplement 1. Overview of analyses for birdterm 1.

( A ) The neuronal population can be best described by seven individual clusters. ( B ) Percentages (rounded) of significantterm neurons in individual loadterm conditions for sampleterm ( n = 68), delayterm ( n = 19), and sampleterm and delayterm ( n = 37). ( C ) On correct trials (left) colorterm is represented during the early and late phase of the sampleterm and, to a lesser degree, during the early and late delayterm. On error trialsterm (right), colorterm informationterm can be found in the early sampleterm phase at loadterm 2, and in the late sampleterm phase at loadsterm 2 and 3 (asterisks). Analysis of loadterm 1 error trialsterm was omitted due to their very low abundance. Statistical comparisons of correct vs. error trialterm informationterm were performed on sub-sampled correct trials. Early and late sampleterm each 400 ms, early and late delayterm each 500 ms, error bars indicate the standard error of the mean. ( D ) Divisive normalization-like regularization was observable for neuronal responses of neurons losing informationterm (top) but not for neurons gaining colorterm informationterm at loadterm 2 (bottom). Selectivity (SE) indicates how much the neuronal response is influenced by a colorterm, relative to a second colorterm when either is presented alone. Sensory interactionterm (SI) indicates how much the neuronal response is influenced by either colorterm when both were displayed simultaneously. Slopesterm close to 0.5 indicate an equal influence of both colorsterm. Slopesterm <0.5, or >0.5 indicate a weighted influence of a colorterm. (Top) Information-carrying neurons in the sampleterm ( n = 35; left) and delayterm ( n = 15; right) population. (Bottom) Informationterm gaining neurons in the sampleterm ( n = 10; left) and delayterm ( n = 3; right) population. The red line indicates the regressionterm·stat fit.

Figure 3—figure supplement 2. Overview of analyses for birdterm 2.

( A ) The neuronal population can be best described by seven individual clusters. ( B ) Percentages (rounded) of significantterm neurons in individual loadterm conditions for sampleterm ( n = 181), delayterm ( n = 75), and sampleterm and delayterm ( n = 96). ( C ) On correct trials (left) colorterm is represented during the early and late phase of the sampleterm and, to a lesser degree, during the early and late delayterm. On error trialsterm (right), colorterm informationterm can be found in the late sampleterm phase at loadsterm 2 and 3 (asterisks). Analysis of loadterm 1 error trialsterm was omitted due to their very low abundance. Statistical comparisons of correct vs. error trialterm informationterm were performed on sub-sampled correct trials. Early and late sampleterm each 400 ms, early and late delayterm each 500 ms, error bars indicate the standard error of the mean. ( D ) Divisive normalization-like regularization was observable for neuronal responses of neurons losing informationterm (top) but not for neurons gaining colorterm informationterm at loadterm 2 (bottom). Selectivity (SE) indicates how much the neuronal response is influenced by a colorterm, relative to a second colorterm when either is presented alone. Sensory interactionterm (SI) indicates how much the neuronal response is influenced by either colorterm when both were displayed simultaneously. Slopesterm close to 0.5 indicate an equal influence of both colorsterm. Slopesterm <0.5 or >0.5 indicate a weighted influence of a colorterm. (Top) Information-carrying neurons in the sampleterm ( n = 70; left) and delayterm ( n = 28; right) population. (Bottom) Informationterm gaining neurons in the sampleterm ( n = 46; left) and delayterm ( n = 5; right) population. The red line indicates the regressionterm·stat fit.

Table 1. Overview of significantterm groups.

The ‘+’ denotes that a neuron of the respective group had a significantterm percent explained varianceterm·var (PEV) in the respective loadterm condition. The ‘-’ denotes that a neuron of the respective group did not have a significantterm PEV in the loadterm respective condition. The pooledterm·mth groups contained only neurons with a ‘+’ for the respective loadterm condition.

Loadterm 1 Loadterm 2 Loadterm 3 Group name - - Loadterm 1 neurons Group I - Loadterm 2 neurons Group II - - Loadterm 3 neurons Group III - Loadterm 1 and 2 neurons Group IV Loadterm 1 and 3 neurons Group V Loadterm 2 and 3 neurons Group VI Loadterm 1, 2, and neurons Group VII Pooledterm·mth group 1 Pooledterm·mth group 2 Pooledterm·mth group 3 The neuronal population has gradually less informationterm with increasing loadterm The clustering analysis indicated that the population of neurons as a whole did sustain the colorterm informationterm throughout the entire trial ( Figure 3A ). Plotting the informationterm averages of each of the three ‘sampleterm populations’ and the ‘delayterm populations’ over time confirmed this result ( Figure 4A , Figure 4—figure supplement 1 ). After the onset of the stimulus array, the average informationterm exhibited a sharp increase that peaked roughly 400 ms after stimulus onset and remained at an elevated level throughout the memory delayterm, until the choice array appeared. Results obtained from neurons of the lateral PFCterm of monkeysterm indicated distinct hemispheric independence of WMterm·dom capacityterm·dom ( Buschman et al., 2011 ). This means that increasing ipsilateral loadterm·var (i.e., loadterm in the hemifieldterm containing the target for which informationterm is assessed) should affect neuronal processing while increasing contralateral loadterm·var should not. This effect might be further emphasized in birdsterm due to the full decussation of their optic nerve ( Husband and Shimizu, 2001 ). Parallel to the behavioral results and in line with the results from monkeysterm, we found a strong effect of ipsilateral loadterm·var on the informationterm maintenance, as there was a sharp drop in informationterm when the loadterm increased from one item to two items ( Figure 4A , blue and yellow curves). The addition of a third item only slightly decreased the maintained informationterm further ( Figure 4A , red curve). The loadterm dependence was much more pronounced during the sampleterm period than during the memory delayterm where the informationterm remained at a lower elevated level. Notably, the loadterm effect was only present for ipsilateralterm manipulations. If the number of items on the contralateralterm side was increased, the informationterm encoded about the colorsterm at the favorite locationterm·def did not change ( Figure 4A , right). To compare our results to the results obtained in monkeysterm we also applied the method of Buschman et al., 2011 for testing the ipsilateral loadterm·var effect during the sampleterm and delayterm phase, by splitting each phase into an early and a late portion (first and second 400 ms of the sampleterm, and first and second 500 ms for the delayterm). We did find a significantterm drop in informationterm with an increase in loadterm from one through three in the early ( F (2,537) = 18.73, p < 0.001, ω ² = 0.0616) and late ( F (2,536) = 20.07, p < 0.001, ω ² = 0.0661) sampleterm period and the early ( F (2,267) = 6.88, p = 0.0012, ω ² = 0.0417) and late ( F (2,267) = 3.85, p = 0.0225, ω ² = 0.0207) delayterm period ( Figure 4B ). There was a large and significantterm drop between one and two items (post hoc Bonferroni corrected multiple comparisonsterm·stat: early and late sampleterm p < 0.001, early delayterm p < 0.001, late delayterm p = 0.0198) and one and three items (post hoc Bonferroni corrected multiple comparisonsterm·stat: early and late sampleterm p < 0.001, early delayterm p = 0.019, late delayterm p > 0.05) but no difference between loadsterm 2 and 3 (post hoc Bonferroni corrected multiple comparisonsterm·stat: all p > 0.05). The maintenance of a significantterm amount of informationterm at higher loadsterm (even for three items, early sampleterm t (156) = 7.55, p < 0.001; late sampleterm t (156) = 8.73, p < 0.001; early delayterm t (87) = 3.84, p < 0.001; late delayterm t (87) = 4.73, p < 0.001) and its gradual reduction when items were added to the corresponding hemifieldterm are indicative of a flexible resourceterm·dom allocation and not an all-or-nothing slot-like WMterm·dom. Furthermore, if there is a flexible resourceterm·dom, in error trialsterm a small but insufficient amount of resourceterm·dom might still be allocated to an item. Indeed, error trialterm analysis (applying correct trial sub-sampling, see Materials and methods) for the loadterm 2 and 3 conditions further supported this interpretation. The amount of informationterm in the early and late sampleterm phase remained above zero (loadterm 2: early, t (186) = 3.25, p = 0.0014; late, t (186) = 5.33, p < 0.001; loadterm 3: t (156) = 4.21, p < 0.001; Figure 4B asterisks), and was significantly smaller than in correct trials (loadterm 2: late, t (186) = 2.81, p = 0.0055, d = 0.26; loadterm 3: late t (156) = 2.55, p = 0.0117, d = 0.23). Additionally, there was no further maintenance throughout the memory delayterm at any loadterm ( Figure 4B , PEV at loadsterm 2 and 3 in error trialsterm delayterm, all non-significant). This indicates that a failure to report which colorterm had changed at higher loadsterm (two and three ipsilateralterm items) resulted from a smaller amount of informationterm encoding during the sampleterm phase that was not maintained throughout the delayterm. A possible alternative ‘slot-model’ explanation would be that, on error trialsterm, the colorterm informationterm was completely lost after the sampleterm phase, because it was not successfully transferred into a slotterm·dom (or that a slotterm·dom was not available to take on informationterm). The graded amount of informationterm on correct trials is not compatible with the simple (all or none) slotterm·dom modelterm, but could fit the ‘slotsterm·dom and averaging modelterm’ ( Zhang and Luck, 2008 ).

Figure 4. Informationterm encoding at the population level.

( A ) Colorterm informationterm (percent explained varianceterm·var [PEV]) decreases with an increasing ipsilateral loadterm·var (i.e., on the same side as the neuron’s favorite locationterm·def) but not with an increasing contralateral loadterm·var (i.e., on the opposite side to the neuron’s favorite locationterm·def). ( B ) On correct trials (left) colorterm is represented during the early and late phase of the sampleterm and, to a lesser degree, during the early and late delayterm. On error trialsterm (right), colorterm informationterm can be found in the early sampleterm phase at loadterm 2, and in the late sampleterm phase at loadsterm 2 and 3 (asterisks). Analysis of loadterm 1 error trialsterm was omitted due to their very low abundance. Statistical comparisons of correct vs. error trialterm informationterm were performed on sub-sampled correct trials. Early and late sampleterm each 400 ms, early and late delayterm each 500 ms, shaded areas, and error bars indicate the standard error of the mean.

Figure 4—figure supplement 1. Sampleterm population ( A ) and delayterm population ( B ), same as Figure 4A with full time axis.

Notably, the ‘delayterm populations’ ( B ) also showed an elevated level of informationterm during the sampleterm, whereas the ‘sampleterm populations’ ( A ) did not show an elevated level of informationterm during the delayterm.

Figure 4—figure supplement 2. Same as Figure 4B , after applying a more stringent criterion on neuronal significance (see text).

On correct trials (left) colorterm is represented during the early and late phase of the sampleterm and, to a lesser degree, during the early and late delayterm. On error trialsterm (right), colorterm informationterm can be found in the early sampleterm phase at loadterm 2, and in the late sampleterm phase at loadsterm 2 and 3 (asterisks). Analysis of loadterm 1 error trialsterm was omitted due to their very low abundance. Statistical comparisons of correct vs. error trialterm informationterm were performed on sub-sampled correct trials. Early and late sampleterm each 400 ms, early and late delayterm each 500 ms, error bars indicate the standard error of the mean.

Higher loadsterm produce divisive normalization-like neuronal responses We next wanted to understand the neuronal mechanisms behind the informationterm loss at higher WMterm·dom loadsterm. For that, we analyzed how the responses of individual sampleterm and delayterm neurons changed when the loadterm increased from one colorterm to two colorsterm. For the ‘sampleterm populations’ and the ‘delayterm populations’, an increasing number of items reduced the amount of encoded informationterm about the colorterm identity ( Figure 4 ). This effect was due to neurons that had a large difference of firing rates between the colorterm A and colorterm B at loadterm 1 (high PEV, i.e., informationterm about colorterm), and reduced differentiation at loadterm 2 (small PEV, no or little informationterm about colorterm, e.g., Figure 2A ). ‘Divisive-normalization-like regularizationterm·def’ (DNRterm·mth; Carandini and Heeger, 2011 ) can explain this effect. DNRterm·mth describes the computation that takes place when two stimuli are presented simultaneously. In a simplified case a neuronal response becomes normalized, analogous to vector normalizationterm·dom, with a normalizationterm·dom factor consisting of the simultaneous stimuli ( Carandini and Heeger, 2011 ). Applied to our context, a consequence of DNRterm·mth would be a reduced differentiation between two colorterm identities at loadterm 2 because differences in firing rate (for each stimulus by itself) at loadterm 1 would be normalized at loadterm 2 (resulting in informationterm loss, e.g., Figure 2A ). We, therefore, hypothesized that DNRterm·mth was observable for neurons with significantterm informationterm at loadterm 1. We tested for DNRterm·mth in the NCLterm by calculating a selectivity index (SE)term·def and a sensory interaction index (SI)term·def for each neuron for the sampleterm phase and the memory delayterm phase ( Reynolds et al., 1999 , see Materials and methods for details). SE indicates how strongly the neuronal response is driven by a colorterm at the favorite locationterm·def of the neuron (reference) in relation to a selected probe colorterm (when either is presented alone). SI indicates how the probe colorterm interacts with the reference colorterm when both are presented simultaneously. Values of both indices, SE and SI, lie between –1 and +1. The addition of a probe colorterm influences the response to the reference colorterm by either suppressing the firing rate of the reference colorterm (if the reference elicits a higher firing rate than the probe, i.e., SE <0), or increasing the firing rate for the reference colorterm (if the probe elicits a higher firing rate than the reference, i.e., SE >0). If DNRterm·mth was present, this influence to suppress or enhance neuronal responses should be an even mixture at the population level, resulting in a significantterm regressionterm·stat between SE and SI with a slopeterm of around 0.5 ( Bouchacourt and Buschman, 2019 ). We compared regressionsterm·stat for the sampleterm and delayterm phase (each as one bin, see Materials and methods for details) for two groups of neurons: information-carrying neurons (significantterm informationterm at loadterm 1), and non-informative neurons (no informationterm at loadterm 1 or at loadterm 2; Figure 5—figure supplement 1 ). We found that DNRterm·mth was present in both sampleterm and delayterm phases ( Figure 5A ). Information-carrying sampleterm neurons had a fitted slopeterm of 0.47 ( R 2 adj 0.39, F (1,838) = 547.69, p < 0.001, CI = [0.43 0.51]) and delayterm neurons had a slopeterm of 0.50 ( R 2 adj 0.34, F (1,342) = 175.60, p < 0.001, CI = [0.43 0.58]). As the slopesterm were not significantly different from 0.5, this indicates that reference and probe colorterm had an equal influence on neuronal responses. We thus show that DNRterm·mth was observable in the neuronal population, and as a consequence of this computation, neurons had generally less informationterm about the colorterm identity at loadterm 2.

Figure 5. Divisive normalization-like regularization was observable for neuronal responses of neurons losing informationterm ( A ) but not for neurons gaining colorterm informationterm at loadterm 2 ( B ).

Selectivity (SE) indicates how much the neuronal response is influenced by a colorterm, relative to a second colorterm when either is presented alone. Sensory interactionterm (SI) indicates how much the neuronal response is influenced by either colorterm when both were displayed simultaneously. Slopesterm close to 0.5 indicate an equal influence of both colorsterm. Slopesterm <0.5, or >0.5 indicate a weighted influence of a colorterm. ( A ) Information-carrying neurons in the sampleterm phase (as one bin; n = 105; left) and delayterm phase (as one bin; n = 43; right) population. ( B ) Informationterm gaining neurons in the sampleterm phase (as one bin; n = 56; left) and delayterm phase (as one bin; n = 8; right) population. The red line indicates the regressionterm·stat fit.

Figure 5—figure supplement 1. Divisive normalization-like regularization was observable for neuronal responses of neurons without significantterm informationterm.

Both phases contain the same neurons ( n = 171). Selectivity (SE) indicates how much the neuronal response is influenced by a colorterm, relative to a second colorterm when either is presented alone. Sensory interactionterm (SI) indicates how much the neuronal response is influenced by either colorterm when both were displayed simultaneously. Slopesterm close to 0.5 indicate an equal influence of both colorsterm. Slopesterm <0.5, or >0.5 indicate a weighted influence of a colorterm. The red line indicates the regressionterm·stat fit. Non-informative sampleterm neurons had a fitted slopeterm of 0.38 ( R 2 adj = 0.16, F (1,1366) = 258.08, p < 0.001), significantly smaller than 0.5 (CI = [0.33 0.42]). Delayterm neurons had a slopeterm of 0.40, also significantly smaller than 0.5 ( R 2 adj = 0.13, F (1,1366) = 197.51, p < .001, CI = [0.35 0.46]). This indicates that for these neurons the reference colorterm influenced firing rate more than the probe colorterm. This smaller slopeterm is not related to the amount of informationterm encoded for the individual colorsterm (which determined the favorite locationterm·def). It does however indicate that those non-informative neurons were influenced by any colorterm at their favorite locationterm·def and thereby might have been informative about if the favorite locationterm·def had a colorterm but not about what colorterm.

Figure 5—figure supplement 2. Example for informationterm gain due to unequal interactionterm at loadterm 2.

Notice how a lack of fire rate differentiation at loadterm 1 (blue curves, left plot) turns into a differential firing rate at loadterm 2 (purple curves, right plot). Depicted is one typical neuron. Black box insets depict the colorterm and loadterm condition during the sampleterm (there were no colorsterm present during the delayterm), and indicate which firing rate curve (mean ± SEM) the neuron had. Source Data File 1: This file contains all details of the reported statistics. The file consists of one MATLAB struct. The struct contains three major fields, corresponding to the different statistical analyses conducted in the Results section. ‘Behavior’ contains statistical results reported in the section ‘The WMterm·dom capacityterm·dom of crowsterm is similar to that of monkeysterm’. ‘smpDlyEarlyLate’ contains statistical results reported in the section ‘The neuronal population has gradually less informationterm with increasing loadterm’. Divisive-normalization-like regularizationterm·def (‘DNRterm·mth’) contains statistical results reported in the sections ‘Gain of informationterm at loadterm 2 can be explained by neuronal normalizationterm·dom’ and ‘Higher loadsterm produce divisive normalization-like neuronal responses’.

Gain of informationterm at loadterm 2 can be explained by neuronal normalizationterm·dom Some neurons showed encoding of colorterm identity at higher loadsterm, instead of loss of informationterm. These neurons were abundant in both the sampleterm phase and the delayterm phase ( Figure 3C ). For example, the neuron shown in Figure 2B did not differentiate between colorterm identities at loadterm 1 but did so for loadterm 2, thus, representing a case of informationterm gain (instead of loss) at a higher loadterm. We wanted to understand if DNRterm·mth, the mechanism that we found reduced colorterm informationterm at loadterm 2, could also produce colorterm differentiation. The ‘normalizationterm·dom modelterm of attentionterm·dom’ ( Reynolds and Heeger, 2009 ) incorporates divisive normalizationterm·def, and can explain how attentionterm·dom can modulate neuronal responses. By attending a preferred (or non-preferred) second colored square in the loadterm 2 condition the neuronal response of a neuron to the target location (i.e., to colorterm A and to colorterm B at the favorite locationterm·def) might be altered. As a result a difference between colorterm A and B may arise even though each colorterm by itself elicited a similar response. In other words, the interactionterm between the additional colorterm and the target colorterm is unequal. Neurons without a colorterm differentiation at loadterm 1 that gained differentiation at loadterm 2 through this process (e.g., if the interactionterm of probe colorterm A with reference colorterm A is larger than the interactionterm of probe colorterm A with reference colorterm B, see Figure 5—figure supplement 2 for an example) should have a population regressionterm·stat slopeterm smaller than 0.5. We thus hypothesized that the population of neurons showing informationterm at loadterm 2, but not at loadterm 1 (e.g., Figure 2B ), would have a smaller slopeterm than the neurons that lost informationterm ( Figure 5A ). Sampleterm neurons had a slopeterm of 0.19 ( R 2 adj 0.05, F (1,446) = 23.0, p < 0.001, CI = [0.11 0.27], Figure 5B ), and delayterm neurons had a slopeterm of –0.04 ( R 2 adj –0.015, F (1,62) = 0.08, p = 0.78, CI = [–0.29 0.22], Figure 5B ). Both slopesterm were significantly smaller than 0.5 and smaller than the slopesterm of the non-informative neurons ( Figure 5—figure supplement 1 ). This indicates that these neurons were influenced more strongly by the reference colorterm, and that the addition of the probe colorterm at loadterm 2 resulted in an unequal interactionterm. Therefore, DNRterm·mth was also computationally responsible for a gain of informationterm at loadterm 2, in a specific subset of neurons.

Discussion Neuronal resourcesterm·dom of WMterm·dom capacityterm·dom are hemifieldterm independent and gradually allocated Our results confirm behavioral findings that have been discussed in detail in an earlier study ( BalakhonovpersName and RosepersName, 2017 ). In brief, we found that the WMterm·dom capacityterm·dom of crowsterm is limited to about four items, and that the two visual hemifieldsterm are largely independent (i.e., the number of items on one side does not affect change detection performance on the other side). Within each hemifieldterm, performance dropped gradually with the addition of a second and third item but remained above chance. Fittingly, on the neuronal level, we found a markedly reduced amount of colorterm informationterm when the number of colored squaresterm·var was increased from one to two (roughly 50% reduction in correct trials). This suggests that WMterm·dom could be conceptualized as a continuous resourceterm·dom that has to be divided between the two items ( Bays and Husain, 2008 ; van den Berg et al., 2012 ; Wilken and Ma, 2004 ), rather than two ‘simple’ slotsterm·dom that would each have the same amount of informationterm irrespective of the memory loadterm. This is also consistent with results of human neuroimaging that report decreased signal amplitude and precision with increasing memory loadterm ( Emrich et al., 2013 ; Sprague et al., 2014 ). In contrast, the hemispheric independence we observed would fit a slot-like modelterm, in which the hemispheres as a whole act like discrete slotsterm·dom. A more nuanced version of the slotterm·dom modelterm (‘slotsterm·dom and averaging’; Zhang and Luck, 2008 ) could also account for graded amounts of informationterm within a limited number of slotsterm·dom ( Fukuda et al., 2010 ; Zhang and Luck, 2008 ), as we found here. The mix of discrete and independent hemispheres with a graded allocation of informationterm between items that we found is comparable to results by Buschman et al., 2011 , observed in monkeyterm PFCterm. On the neuronal level, recurrent connections between neurons within a hemisphere may reduce item differentiation when multiple items are present simultaneously, creating capacityterm·dom limitations within the hemisphere ( Matsushima and Tanaka, 2014 ). A lack of interhemispheric recurrent connections would make processing in the other hemisphere independent. Like in monkeysterm, WMterm·dom capacityterm·dom in crowsterm may therefore result from neuronal activity patterns governed by multiple individual items. We probed the WMterm·dom capacityterm·dom of crowsterm using colored squares, based on the taskterm design of Buschman et al., 2011 . Using the identical taskterm allowed us to directly compare our neuronal results of WMterm·dom capacityterm·dom from NCLterm to results from PFCterm of monkeysterm. Taskterm similarity is very important for such cross species comparisons as even small changes in taskterm parameters may introduce substantial differences in neuronal responses, leading to potentially different conclusions. In a taskterm similar to the one used here, Lara and Wallis, 2014 , have found that neurons in the PFCterm of monkeysterm encoded nearly no informationterm about colorterm, but instead about location. In their taskterm monkeysterm had to memorize the colorterm of squares at two locations on a screen, and were again confronted with a colored square at one of the two locations after a delayterm. The monkeysterm then had to indicate if the colorterm at that location had changed. Lara and Wallis, 2014 , discuss the absence of colorterm informationterm in the neurons they recorded in relation to the taskterm of Buschman et al., 2011 , who like us, did find colorterm informationterm. In brief, the exact taskterm design may determine the neuronal encoding of taskterm relevant informationterm ( Lara and Wallis, 2014 ). Similar to the complex contribution of PFCterm neurons to WMterm·dom, neurons of NCLterm can also encode a wide range of very different taskterm relevant aspects, like colorterm (this study), spatial locations ( Rinnert et al., 2019 ; Veit et al., 2017 ), and more abstract items like rules ( Veit and NiederpersName, 2013 ) and numerosities ( Ditz and NiederpersName, 2015 ).

Attentional processes guide WMterm·dom allocation and maintenance One way to circumvent WMterm·dom failure when item loadterm increases is to allocate attentionterm·dom. Our results suggest that attentionterm·dom may play an important role in crowterm WMterm·dom. Capacityterm·dom limitation became apparent during encoding, as the amount of informationterm at the end of the sampleterm period was affected by the stimulus loadterm. Adding a second and third item to the ipsilateralterm stimulus array reduced the amount of colorterm informationterm encoded by NCLterm neurons that carried over into the memory delayterm. Furthermore, neuronal activity in trials in which the birdsterm made an incorrect response showed only weak encoding during the sampleterm phase without informationterm maintenance during the memory delayterm. This fits studies of human WMterm·dom that have shown attentive filtering during encoding of stimuli influencing WMterm·dom capacityterm·dom ( Bays and Husain, 2008 ; Vogel et al., 2005 ; Vogel and Machizawa, 2004 ), and neuronal correlates of this have been reported for monkeysterm as well ( Buschman et al., 2011 ). Beyond the domain of sensory signals, attentionterm·dom and WMterm·dom may be directly linked. Neuronal correlates of WMterm·dom and attentionterm·dom overlap in PFCterm neurons, for example, Lebedev et al., 2004 , found that a substantial amount of PFCterm neurons encode either an attentional signal, or a memory signal, and some (hybrid) neurons do both. A purely mnemonic function of PFCterm thereby seems unlikely. Indeed, very recently, Panichello and Buschman, 2021 , have reported that at the population level neurons of PFCterm encode ‘both the selection of items from working memory and attentionterm·dom to sensory inputs’ (p. 2), rather than just memory content. The independence of hemifieldsterm that we observed on the behavioral level (this study and BalakhonovpersName and RosepersName, 2017 ) and found in the neuronal responses could also be related to attentionterm·dom. Adding stimuli in the contralateralterm hemifieldterm affected neither performance nor informationterm maintained by NCLterm neurons, whereas additional ipsilateralterm stimuli strongly reduced both. This fits the influence of attentionterm·dom on WMterm·dom and hemifieldterm independence, which is consistently accentuated in studies in which attentionterm·dom had to be divided between the two hemifieldsterm ( Alvarez and Cavanagh, 2005 ; Buschman et al., 2011 ; Cavanagh and Alvarez, 2005 ; Delvenne, 2005 ; Delvenne et al., 2011 ). Finally, the DNRterm·mth computation may explain the responses of the neurons that gained informationterm at loadterm 2 through attentional processes predicted by the ‘normalizationterm·dom modelterm of attentionterm·dom’ ( Reynolds and Heeger, 2009 ). This may appear counter-intuitive and contradictory, considering that the same process is also responsible for the loss of informationterm. However, when attentionterm·dom is overtly directed to a specific (preferred or non-preferred) item within the receptive field of a neuron, the DNRterm·mth computation shifts its weighting of the normalized response toward the response of the attended item ( Reynolds et al., 1999 ; Reynolds and Heeger, 2009 ). This weighted normalizationterm·dom can produce a difference in the neuronal response to both colorterm identities at loadterm 2, even if the neuronal response was non-informative at loadterm 1. At the population level we were able to observe such an effect as the reduced slopeterm of the selectivity/interactionterm fit. Thus, an attentive process might have enhanced informationterm in WMterm·dom at higher loadsterm. It is important to clarify that, as we did not use any form of attentional cueing in our study, we cannot explicitly test for such an attentionterm·dom effect. However, we do know that the animals participating in this study can use attentional cues to enhance their WMterm·dom ( FongaropersName and RosepersName, 2020 ). The attentionterm·dom cues used by FongaropersName and RosepersName, 2020 , positively affected not only encoding but also the maintenance and retrieval of the informationterm held in WMterm·dom, comparable to results from monkeysterm and humans ( Brady and Hampton, 2018 ; Souza and Oberauer, 2016 ). We, therefore, want to emphasize that our data is in line with the interpretation that the birdsterm possibly attended a loadterm 2 stimulus array differently than a loadterm 1 stimulus array in order to enhance their performance in trials with higher loadsterm.

Modern modelsterm of mammalianterm WMterm·dom capacityterm·dom are applicable to crowsterm Our neuronal recordings offer a mechanistic explanation for the behavioral effects, as we found clear evidence of DNRterm·mth governing the neuronal responses tied to WMterm·dom capacityterm·dom that is in accordance with mammalianterm modelsterm of WMterm·dom capacityterm·dom ( Bouchacourt and Buschman, 2019 ; Lundqvist et al., 2016 ; Lundqvist et al., 2011 ). The loss of informationterm about colorterm identity (i.e., neuronal response differentiation between colorsterm) can be accounted for by DNRterm·mth when an item is added to a neuron’s receptive field. The normalizationterm·dom of neuronal firing rate diminishes the differentiation between colorterm identities. As such it is analogous to neurophysiological responses from visual areas ( Carandini et al., 1997 ; Reynolds et al., 1999 ) and to the PFCterm during spatial WMterm·dom ( Matsushima and Tanaka, 2014 ). The WMterm·dom modelterm of Bouchacourt and Buschman, 2019 , is based solely on data from monkeyterm electrophysiologyterm·mth, and thus implicitly tied to the layered columns of the neocortexterm. The results we report here show that the modelterm also fits the neurophysiology of WMterm·dom in crowsterm. However, the picture is incomplete since important aspects of monkeysterm’ WMterm·dom are still not investigated in crowsterm. Oscillations of local field potentials are relevant for how informationterm enters WMterm·dom and how it is maintained ( Miller et al., 2018 ), and have been tied to normalizationterm·dom and competition between items in WMterm·dom ( Lundqvist et al., 2018b ). Thus, the oscillatory interplay of the layers and different regions of the mammalianterm neocortexterm are important fields of research to further our understanding of WMterm·dom. Such aspects are so far completely unknown in crowsterm and their non-layered associative areas. This encourages further investigation into the neuronal circuits of WMterm·dom in birdsterm. There is also ongoing debate about the role of sustained activity during delayterm periods and how it relates to WMterm·dom ( Constantinidis et al., 2018 ; Lundqvist et al., 2018a ; Miller et al., 2018 ). We cannot report of any neuron that showed persistentterm activity comparable to those reported by classical WMterm·dom studies in PFCterm (e.g., Funahashi et al., 1989 ; Fuster and Alexander, 1971 ), or in NCLterm ( Diekamp et al., 2002b ; Veit et al., 2014 ; Veit and NiederpersName, 2013 ). This may be reconcilable with some other contemporary modelsterm of WMterm·dom. One major type of those modelsterm implements ‘synfire chains’, where individual neurons fire sequentially (and transiently) to bridge temporal gaps and maintain taskterm relevant contents ( Rajan et al., 2016 ). This has, for example, been reported to be the case in posterior parietal cortex of mice performing a T-maze taskterm that required WMterm·dom for cued spatial locations to be maintained ( Harvey et al., 2012 ). The transient activity of neurons that we report ( Figures 2 and 3A ) might fit into such modelsterm. However, our results can only be compared very cautiously to this (since even small changes in taskterm design significantly alter neuronal responses, e.g., Lara and Wallis, 2014 ). Therefore, while we cannot, yet, fully equate crowterm and monkeyterm WMterm·dom, our results raise two important questions about how WMterm·dom is implemented on the level of neuronal networks that have implications for our comparative view of crowterm WMterm·dom. The first regards the neuronal computations underlying WMterm·dom. Is there a common canonicalterm computation governing WMterm·dom, or are there different solutions based on different neuronal architecturesterm? Recent work has shown that the sensory areas of mammals and birdsterm show remarkably similar circuit organization ( Stacho et al., 2020 ). However, higher-order associative areas involved in WMterm·dom, like the LPFC in mammals and the NCLterm in birdsterm, have distinctly different architecturesterm ( Stacho et al., 2020 ). The fact that differently organized areas like LPFC and NCLterm produce strikingly similar physiological responses points to shared computational principles. Modeling work already suggests that the competing WMterm·dom capacityterm·dom modelsterm can be accommodated into a unifying framework based on theoretical neuronal informationterm sampling, where stochastic informationterm sampling (assumed for continuous resourceterm·dom modelsterm) can account for item limitations better than fixed informationterm sampling (assumed by the slotsterm·dom and averaging modelsterm) ( Schneegans et al., 2020 ). Similarly, DNRterm·mth is already considered to be a general, canonicalterm computation of the nervous system, present in evolutionarily distant phyla, for example, fruit flies and monkeysterm ( Carandini and Heeger, 2011 ). The second question regards the tradeoff between WMterm·dom flexibility and capacityterm·dom ( Bouchacourt and Buschman, 2019 ). Is the WMterm·dom of a crowterm as flexible as that of a monkeyterm? Our results show that the computations by individual neurons that result in WMterm·dom capacityterm·dom limitations are virtually the same in crowsterm and monkeysterm, highlighting a further aspect of WMterm·dom that is similar between these animal groups ( NiederpersName, 2017 ). Ultimately, our results were in line with different modern modelsterm of WMterm·dom that implement DNRterm·mth to explain capacityterm·dom ( Bouchacourt and Buschman, 2019 ; Lundqvist et al., 2016 ; Lundqvist et al., 2011 ; Schneegans et al., 2020 ). However, the data we presented cannot carry a definitive conclusion about which of the different modelsterm fits best. For example, a tradeoff between flexibility and capacityterm·dom ( Bouchacourt and Buschman, 2019 ) might be present, but further investigation into the modelsterm’ predictions is required. We do, however, show that mammalianterm modelsterm of WMterm·dom are in line with WMterm·dom in birdsterm, which implies that fundamental aspects of WMterm·dom are shared between these animal groups.

Conclusion Together, all these facets of crowterm WMterm·dom capacityterm·dom suggest that the different intricate neuronal architecturesterm that carry out the computations in monkeysterm and crowsterm have likely been shaped by convergentterm·dom evolution – into systems that yield similar cognitive performances. The systems may share the same basic mechanisms and thus limitations. Further investigation into the oscillatory dynamics of WMterm·dom in the avian brain may elucidate if birdsterm also share the prominent limitation of a tradeoff between flexibility and capacityterm·dom.

Materials and methods Subjects Two hand-raised carrion crowsterm ( C. corone ) of 2 years of age served as subjects in this study. The birdsterm were housed in spacious aviaries in social groups. During the experimental procedures, the animals were held on a controlled food protocol with ad libitum access to water and grit. All experimental procedures and housing conditions were carried out in accordance with the National Institutes of Health Guide for Care and Use of Laboratory Animals and were authorized by the national authority ( LANUV ).

Experimental setup We used operant training chambers (50 × 50.5 × 77.5 cm 3 , width × depth × height) equipped with an acoustic pulse touchscreen (22’’, ELO 2200L APR, Elo Touch Solutions Inc, Milpitas, CA) and an infrared camera (Sygonix, Nürnberg, Germany) for remote monitoring. The birdsterm sat on a wooden perch so that the distance between the birdterm’s eye and the touchscreen was 8 cm. Food pellets were delivered as a reward via a custom-made automatic feeder (plans available at http://www.jonasrose.net/ ). The position of the animal’s head was tracked online during the experiment by two open-source computer vision cameras (‘Pixy’, CMUcam5, Charmed Labs, Austin, TX) that reported the location and angle between two LEDs. For tracking, we surgically implanted a lightweight head-post and used a lightweight 3D-printed mount with LEDs that was removed after each experimental session. The system reported the head location at a frame rate of 50 Hz and data was smoothed by integrating over two frames in MATLAB using custom programs on a control PC. All experiments were controlled by custom programs in MATLAB using the Biopsychology ( RosepersName et al., 2008 ) and Psychophysics toolboxes ( Brainard, 1997 ). Digital input and output of the control PC were handled by a microcontroller (ODROID C1, Hardkernel co. Ltd, Anyang, South Korea) connected through a gigabit network running custom software (available at http://www.jonasrose.net/ ).

Behavioral protocol The behavioral protocol was identical to the one described in BalakhonovpersName and RosepersName, 2017 . We trained the birdsterm to perform a delayed change localization taskterm that had previously been used to test the performance under different WMterm·dom loadsterm in primates ( Buschman et al., 2011 ). Each trial started after a 2 s inter-trial-interval, with the presentation of a red dot centered on the touchscreen (for a maximum of 40 s). The animals initiated the trial by centering their head in front of the red dot for 160 ms. This caused the red dot to disappear and a stimulus array of two to five colored squares to appear ( Figure 1A , ‘sampleterm’). The colored squares were presented for a period of 800 ms, during which the animals had to hold their head still and center their gaze on the screen (‘hold gaze’, no more than 2 cm horizontal or vertical displacement, and no more than 20 degrees horizontal or vertical rotation). Failure to hold the head in this position resulted in an aborted trial. This sampleterm phase was followed by a memory delayterm of 1000 ms after which the stimulus array reappeared with one colorterm exchanged. The animal had to indicate the location of the colorterm change by pecking the respective square. Correct responses were rewarded probabilistically (BEO special pellets, in 55% of correct trials, additional 2 s illumination of the food receptacle in 100% of correct trials). Incorrect responses to colorsterm that had not changed or a failure to respond within 4 s resulted in a brief screen flash and a 10 s timeout. The stimuli were presented at six fixed locations on the screen (1–6, Figure 1A ). In each session, one pair of colorsterm was assigned to each of the six locations. Each location had its own distinct pair. These pairs were randomly chosen from a pool of 14 colorsterm (two colorterm combinations were excluded since the animals did not discriminate them equally well during a pre-training). Let us consider Figure 1A as an example. The colorterm change occurs in the middle-left where turquois (T) is presented during the sampleterm and orange (O) during the choice. In this particular session the middle-left could thus show either of the following colorsterm during the sampleterm and choice: T-O (shown in Figure 1A ); O-T; O-O; T-T; None-None. On the next session a new random pair of colorsterm was displayed at this location.

For identification and analysis an arbitrary label was assigned to each of the randomly drawn colorsterm at the start of each session (i.e., left middle location: ‘colorterm A’, or ‘colorterm B’; left top location: ‘colorterm A’, or ‘colorterm B’; etc.). These indices do not refer to the order of presentation of the colorsterm at any time, but were held constant for neuronal analysis. The order of presentation of colorsterm within a pair, the target location (where the colorterm change occurred), and the number of stimuli in the array (two to five) were randomized and balanced across trials so that each condition had an equal likelihood to appear. The order of presentation of colorsterm within a pair, the target location (where the colorterm change occurred), and the number of stimuli in the array (two to five) were randomized and balanced across trials so that each condition had an equal likelihood to appear. The colorterm squares had a width of 10 degrees of visual angle (DVA) and were placed on the horizontal meridian of the screen and at 45.8 DVA above or below the meridian at a distance of 54 and 55.4 DVA from the center. This arrangement in combination with the head tracking ensured that all stimuli appeared outside of the binocular visual field of crowsterm (37.6 DVA; Troscianko et al., 2012 ).

Surgery Both animals were chronically implanted with a lightweight head-post to attach a small LED holder during the experiments. Before surgery, animals were deeply anesthetized with ketamine (50 mg/kg) and xylazine (5 mg/kg). Once deeply anesthetized, animals were placed in a stereotaxic frame. After attaching the small head-post with dental acrylic, a microdrive with a multi-channel microelectrode was stereotactically implanted at the craniotomy (Neuronexus Technologies Inc, Ann Arbor MI, DDrive). The electrode was positioned in NCLterm (AP 5.0, ML 13.0) of the left hemisphere (coordinates for the region based on histological studies on the localization of NCLterm in crowsterm; Veit and NiederpersName, 2013 ). After the surgery, the crowsterm received analgesics.

Electrophysiological recordings Extracellular single neuron recordings were performed using chronically implanted multi-channel microelectrodes. The distance between recording sites was 50 µm. The signal was amplified, filtered, and digitized using Intan RHD2000 headstages and a USB-Interface board (Intan Technologies LLC, Los Angeles, CA). The system also recorded digital event codes that were sent from the behavioral control PC using a custom IO device (details available at http://www.jonasrose.net/ ). Before each recording session, the electrodes were advanced manually using the microdrive. Recordings were started 20 min after the advancement, and each recording site was manually checked for neuronal signals. The signals were recorded at a sampling rate of 30 kHz and filtered with a band-pass filter at recording (0.5–7.5 kHz). The recorded neuronal signals were not pre-selected for taskterm involvement. We performed spike sorting using the semi-automatic Klusta-suite software ( Rossant et al., 2016 ), which uses the high electrode count and their close spacing to isolate signals of single neurons. For spike sorting, we filtered with a high pass of 500 Hz and a low pass of 7125 Hz. The software utilizes the spatial distribution of the recorded signal along the different recording sites to untangle overlapping signals and separate signals with similar waveforms but different recording depths.

Data analysis All statistical analyses were performed in MATLAB (2018b, Mathworks Inc) using commercially available toolboxes (Curve Fitting Toolbox Version 3.5.3, Statistics and Machine Learning Toolbox Version 10.2) and custom code. For all statistical tests, we assumed a significance level of α = 0.05, unless stated otherwise. Trials were classified as error trialsterm if the birdterm chose a location where no change of colorsterm had appeared. Trials in which the birdterm did not choose any location or failed to maintain head fixation were not analyzed. All correct trials were included in the analysis of neural data. Depending on the analysis we refer to different ‘loadterm conditions’ relative to referential sides of the screen which are either ipsilateralterm (same side) or contralateralterm (opposite side), each with a possible loadterm between one and three items. For the behavioral analyses the terms ipsilateralterm and contralateralterm refer to the location of the change that had to be detected. For the neurophysiological analyses the terms ipsilateralterm and contralateralterm refer to the respective neuron’s favorite locationterm·def (described in the following section).

Because there were only very few error trialsterm in the loadterm one condition, we performed error trialterm analysis only for the loadterm 2 and loadterm 3 conditions. The behavioral data were analyzed as described in our previous study ( BalakhonovpersName and RosepersName, 2017 ), estimating the WMterm·dom capacityterm·dom K for each loadterm by Equation 1 .

K = n * p (1) where p is the percentage correct and n is the number of items in WMterm·dom. This estimate has been applied to similar primate data and in studies with humans ( Johnson et al., 2013 ; Kornblith et al., 2016 ).

Informationterm about colorterm Identity Based on a one-way ANOVAterm·stat of colorterm identity at a given location, we calculated a PEV statistic to measure the effect sizeterm·stat of neuronal modulation. Its main parameter ω ² is a measurement for the percentage to which the tested factor can explain the variance of the data, and it is calculated from the sum of squares of the effect ( SS effect ) and the mean squares of the within-group (error) variance ( MS error ) ( Equation 2 ).

ω 2 = S S e f f e c t - d f * M S e r r o r S S t o t a l + M S e r r o r (2) For each neuron, we determined a ‘favorite locationterm·def’, which was defined as the location with the highest cumulative PEV, of the three possible locations on the right half of the screen, that is, opposite to the implanted hemisphere, across four non-overlapping bins during the sampleterm phase (bin size 200 ms, advanced in steps of 200 ms, from start till the end of the sampleterm phase). The significance of calculated effect sizeterm·stat values was determined by a permutation testterm·stat. We ran the permutation to calculate the likelihood of getting an explained variance value bigger than the one calculated from the actual distribution of the data by randomly permuting the colorterm identity labels and calculating the PEV 1000 times. The test thereby does not assume any distribution of the data and returns an unbiased estimate of the likelihood of generating an effect sizeterm·stat within the data randomly. The measured value of explained variance from the actual dataset was assumed to be significantterm if the likelihood of randomly generating a bigger value was below 5%. We chose to not correct for the multiple comparisons at this level, as we reasoned that if we were to only include those neurons that had the most informationterm at the individual loadsterm (i.e., those with the highest PEV values that are significantterm even under very stringent statistical criteria) we would have artificially inflated the amount of informationterm present at each loadterm. By including neurons that encoded less informationterm too (i.e., at the uncorrected p-value) our analysis population was more resistant to such outlier effects. We additionally performed our analyses using a more stringent statistical criterion (significance of two consecutive non-overlapping bins) and found the same results ( Figure 4—figure supplement 2 ). We tested the proportions of significantterm neurons we found for the different trial phases by performing a binomial testterm·stat, assuming a significance level α = 0.05 ( Equation 3 ).

P X = i = B p 0 , n = n k p 0 i 1 - p 0 n - i (3) Calculating the probability p, of finding X significantterm neurons, given a total amount of i (362) neurons, and a probability p 0 of 5% finding a significance by chance.

Population analyses We considered neuronal significance (i.e., significantterm PEV as determined above) for each loadterm independently. This means, we tested if the PEV of a neuron was significantterm three times with the permutation method described above: once for each of the three loadterm conditions. Therefore, we can report seven groups of significance ( Table 1 , Figure 3C ). Subsequently, we created three pooledterm·mth groups ( Table 1 ) from all neurons with a significantterm PEV at each individual loadterm. We used these pooledterm·mth groups for the population analyses ( Figures 4 and 5 ). Neurons of these pooledterm·mth groups, with a significantterm PEV during the sampleterm phase were assigned to the ‘sampleterm population’, and neurons with a significantterm amount of informationterm during the memory-delay phase were assigned to the ‘delayterm population’ (significance criterion: one significantterm 200 ms bin, at α = 0.05, see above for our reasoning not to correct for multiple comparison at this point). Thus, neurons with significantterm PEV during both the sampleterm and delayterm phase were included in both subpopulations. We corrected for the unequal amount of correct and error trialsterm when comparing informationterm about colorterm (PEV) between the trial conditions, by sub-sampling correct trials with the number of error trialsterm 1000 times for each neuron. The resulting PEV values of correct trials were then averaged for each neuron, this population of averaged PEV values was then statistically tested against the PEV values of error trialsterm (of the same neurons) using a dependent t-testterm·stat.

Divisive normalization-like regularization We tested for the presence of divisive normalizationterm·def using the method of Reynolds et al., 1999 . Three conditions were considered: (1) neuronal response to stimulus A, (2) neuronal response to stimulus B, and (3) neuronal response to the simultaneity of stimuli A and B. As we wanted to relate this to the informationterm about colorterm identity, we selected subsets of the favorite locationterm·def and the additional two ipsilateralterm locations. To test how the neurons altered their response when multiple stimuli were presented simultaneously, we calculated the colorterm selectivity index (SE)term·def and the sensory interaction index (SI)term·def of each neuron. SE i was calculated by subtracting the normalized firing rate for the chosen reference colorterm i ( REF i ) at the neuron’s favorite locationterm·def, from a second colorterm j ( PROBE j ) at a different location (ipsilateralterm to the favorite locationterm·def, Equation 4 ).

S E i = P R O B E j - R E F i (4) The resulting selectivity index lies between –1 (completely selective for the reference colorterm) and 1 (completely selective for the probe colorterm). SI was calculated ( Equation 5 ) by subtracting the normalized firing rate for REF i from the normalized firing rate of the combination of REF i and PROBE j ( PAIR i , j ).

S I i , j = P A I R i , j - R E F i (5) This interactionterm index also lies between –1 (full suppression of reference stimulus by the probe stimulus) and 1 (full enhancement of the reference stimulus by the probe stimulus). As each of the three locations had two possible colorsterm, we calculated eight SE and SI indices per neuron and performed a linear regressionterm·stat for all indices. This is required as each stimulus combination is informative about the normalizationterm·dom. The effects of divisive normalizationterm·def were compared between the sampleterm and the delayterm phase. Therefore, SE and SI indices were calculated across the entire sampleterm (800 ms) and memory delayterm (1000 ms) phase. Neurons with significantterm informationterm were accordingly identified over the entire sampleterm and delayterm as one bin, using the permutation testterm·stat described in the section ‘ informationterm about colorterm identity ’. We considered the entire sampleterm and delayterm phase because we wanted to analyze the population response as a whole, irrespective of highly diverse response profiles of individual neurons.

Hierarchical clusteringterm·stat To visualize the different groups of neurons that encoded and maintained informationterm about the colorterm identity during different phases of the trial, we performed a hierarchical clustering analysisterm·stat in MATLAB on the normalized PEV values of individual neurons throughout the trial. We used a (1 − correlation) distance metric and an average distance linkage function for a maximum of seven clusters. The maximum number of clusters was first determined by calculating the clustering for different amounts of clusters (1–10) and subsequently calculating the within-cluster sum-of-squares. This resulted in a graph that allowed us to visually inspect the tradeoff between cluster number and fit improvement, from which we estimated the inflection point (elbow method). A cluster number of seven presented the best tradeoff that allowed visualization of the different groups at an acceptable clustering success. We then ordered the neuron clusters to minimize the average distance between the clusters in the dendrogram.

Acknowledgements We would like to thank Mikael Lundqvist for helpful comments on an earlier version of the manuscript.

Funding Statement The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Contributor Informationterm Lukas Alexander HahnpersName, Email: lukas.hahnpersName@ruhr-uni-bochum.de.

Dmitry BalakhonovpersName, Email: balakhonovpersName.ds@gmail.com.

Erin L RichpersName, Icahn School of Medicine at Mount Sinai, United States.

Michael J FrankpersName, Brown University, United States.

Funding Informationterm This paper was supported by the following grants: Volkswagen Foundation Freigeist Fellowship 93299 to Jonas RosepersName.

Deutsche Forschungsgemeinschaft Project B13 of the collaborative research center 874 (122679504) to Jonas RosepersName.

Additional informationterm Competing interests No competing interests declared.

Author contributions Data curation, Formal analysis, Investigation, Methodology, Software, Visualization, Writing - original draft, Writing - review and editing.

Conceptualization, Data curation, Methodology.

Data curation.

Project administration, Resourcesterm·dom, Writing - review and editing.

Conceptualization, Funding acquisition, Methodology, Project administration, Resourcesterm·dom, Supervision, Visualization, Writing - review and editing.

Ethics All experimental procedures and housing conditions were carried out in accordance with the National Institutes of Health Guide for Care and Use of Laboratory Animals and were authorized by the national authority (LANUV protocol no. 84-02.04.2017.A001).

Additional files Transparent reporting form elife-72783-transrepform1.pdf (816KB, pdf) Source data 1. Reported statistical results and numerical values.

elife-72783-supp1.zip (204.8KB, zip) Data availability All details of statistics reported in the manuscript is provided as a supporting file. Source data files of all figures are publicly available via dryad https://doi.org/10.5061/dryad.0k6djhb1q .

The following dataset was generated: HahnpersName LA, BalakhonovpersName D, RosepersName J. 2021. Working memory capacityterm·dom of crowsterm and monkeysterm arises from similar neuronal computations. Dryad Digital Repository.