We also explored whether the strength of early visual monotonic effects correlated with the strength of numerosity-tuned response effects, given that tuned responses might be derived from monotonic responses. We tested for a correlation between early visual suppression (proportional reduction in response slope) and the reduction in explained response variance of the tuned response model between low and high adaptor conditions in the numerosity maps. We observed marginal correlations specifically between the numerosity map NTO and visual field maps V1-hV4 (Supplementary Figs. 6b and 7b). However, most of these few significant correlations did not survive correction for multiple comparisons, and no significant correlations were found among other tested visual field or numerosity maps. We tested for a correlation between early visual suppression and the changes in numerosity preferences of the tuned responses between low and high adaptor conditions in the numerosity maps. While comparing numerosity preferences to early visual response suppression was complex due to bidirectional shifts in numerosity preferences44, some marginal correlations again appeared between NTO and V1-hV4. (Supplementary Figs. 6b and 7b). In short, these between-hemisphere correlations of monotonic and tuned effects were underpowered and are complicated by notable interhemispheric differences within participants in numerosity maps beyond NTO22.
In the current study, we asked whether numerosity adaptation affects the responses of the early visual cortex. First, we found that the monotonically increasing neural response to numerosity occurred regardless of numerosity adaptation, from V1 through the early visual hierarchy to V2, V3, hV4, V3A/B, and LO1-LO2. Second, in all these visual field maps, the amplitude of this monotonic increase (slope) was reduced when the adapting numerosity was higher. Third, the proportion by which the response slope was reduced during higher compared to lower numerosity adaptation (i.e., the magnitude of the adaptation effect) increased hierarchically from V1 onward. Fourth, the magnitude of this adaptation effect generally correlated among the visual field maps V1, V2, V3 and hV4.
In this study, we focus on the early visual neural response that monotonically increases with numerosity. We have explained these findings by the close relationship between numerosity and contrast energy in the spatial frequency domain. Image contrast refers to the overall variation in brightness or color distribution across the entire image, typically reflecting the range of luminance values within the image. In the spatial frequency domain, this variation is measured as contrast energy, expressed in Fourier power. At a fixed image contrast, this aggregate Fourier power follows numerosity closely but nonlinearly, with little effect of size or spacing, and predicts population responses in V1 and computational models more closely than numerosity does. Recent modelling studies also show how image filters specifically modelling the properties of early visual neurons could capture these contrast properties to yield early visual monotonic responses. This provides a response in the early visual cortex from which numerosity itself may be straightforwardly derived. Indeed, responses in the numerosity-tuned populations of the association cortices are more closely predicted by numerosity than aggregate Fourier power. Therefore, we describe the monotonic responses in the current study as responses to contrast and the tuned responses as responses to numerosity. However, we found a very similar pattern of results if we model the early visual responses as functions of the log(numerosity) (Supplementary Figs. 1b, 2, 4, 5 and 7), rather than as functions of the logarithm of the aggregate Fourier power of our displays.
Adaptation effects on numerosity perception have always been assumed to reflect changes in the responses of numerosity-tuned neurons. This is a reasonable assumption for several reasons. First, when the adaptation of numerosity perception was first described, numerosity-tuned neurons had recently been found in macaque parietal and frontal cortices, and tuned effects of repetition suppression were found in human parietal cortex. Then, the monotonic response was shown in the parietal lobe and early visual monotonic responses to numerosity were only described years later. Second, adaptation effects are often found for image features with tuned neural representations, like orientation and motion direction. Third, adaptation to a low numerosity has been shown to increase perceived numerosity, as well as adaptation to high numerosity decreasing perceived numerosity. The bidirectionality of this repulsive effect seems likely to reflect effects on numerosity-tuned neural populations with different numerosity preferences. Specifically, adaptation to a numerosity below the numerosity preference of a numerosity-tuned neuron should suppress responses to lower numerosities more than responses to higher numerosities. This should thereby increase the numerosity yielding the largest response (the numerosity preference). Accordingly, we have recently shown (using the present data set) that tuned neural numerosity preferences are affected by adaptation.
However, converging evidence also suggests that the neural effects of perceptual numerosity adaptation begin at early visual processing stages, with spatially specific responses to image contrast. First, perceptual numerosity adaptation is highly spatially specific (limited to the adapted location), while numerosity-tuned neurons have large spatial receptive fields and their response to numerosity does not depend on the stimulus falling within that receptive field. Second, perceptual numerosity adaptation is weaker when the adaptor and test displays differ in color or other low-level visual features. Different low-level features activate distinct neural populations in early visual processing, so an adaptation effect on the population responding to one feature is unlikely to affect populations responding to other features. Conversely, similar numerosity-tuned responses are found regardless of item color, so adaptation effects working on these populations should generalize across low-level features. Third, compelling recent results show that perceived numerosity is affected by adaptation to gratings with no numerosity but a spatial frequency matching that of the numerosity display. Fourth, recent results show that the strength of the numerosity adaptation effect is greater when the positions of the individual items in the adaptor and test displays overlap. Again, different positions activate distinct neural populations in early visual processing, but similar numerosity-tuned responses are found regardless of item position. Fifth, the increase in perceived numerosity after low numerosity adaptation is far weaker than the decrease after high numerosity adaptation. The asymmetry of this bidirectional effect may reflect an additional effect of adaptation at the monotonic response stage for high numerosity displays. Finally, the numerosity adaptation effect becomes weaker as contrast decreases, though it remains clear even at low contrasts. Together with the present results, these results suggest that perceptual numerosity adaptation at least partly originates in early visual processing stages with spatially specific responses to contrast.
Importantly, none of these findings show that perceptual numerosity adaptation arises only through early visual contrast adaptation and indeed several results speak against this interpretation. First, we found that effects on monotonic responses become progressively stronger through the early visual hierarchy, suggesting additional neural numerosity adaptation effects at many stages of numerosity processing. Second, recent results show that responses to numerosity in more anterior areas of the association cortices depend progressively more on the context of recently presented numerosities. Third, the effects on monotonic responses that we see are only correlated with effects on tuned responses in the most posterior numerosity map. All of these results suggest progressively increasing neural adaptation effects throughout the numerosity processing hierarchy, not effects at an early stage alone. Furthermore, adaptation effects on visual numerosity perception can also be produced by adapting to quantities in other sensory modalities, though these cross-modal adaptation effects are weaker than the effects of adaptation to visual numerosity itself. Finally, beyond adaptation, numerosity estimation is reduced when individual items are connected by bars. This effect is not present in the earlier visual responses to numerosity and cannot be explained by changes in the spatial frequency domain contrast of the displays, so at least some effects on numerosity perception depend on later stages. We therefore propose that neural effects at many stages of numerosity processing contribute to perceptual numerosity adaptation effects. Neural populations in many areas represent information about numerosity in either their monotonic or tuned responses, with hierarchical processing of each response across many stages and tuned responses likely being derived from monotonic responses. As adaptation may be best understood as a property of all neural responses, we can expect adaptation effects at all of these stages, with effects at one stage likely being inherited by the next.
Our results do not convincingly demonstrate that adaptation effects on early visual monotonic responses ultimately cause adaptation effects on numerosity-tuned responses. Indeed, it is not yet clear that early visual monotonic responses are required to produce a numerosity-tuned response. Nevertheless, several findings suggest that adaptation effects on numerosity-tuned responses are inherited in part from effects on early visual contrast representations. First, almost all visual inputs to the cortex come through the primary visual cortex, which represents image features by contrast-driven responses in the spatial frequency domain. There is no other pathway through which numerosity-tuned neurons could be activated by visual stimuli. Second, computational models for the derivation of numerosity-tuned responses generally rely on an intermediate stage with monotonic responses to numerosity. We have previously shown that the monotonic responses to numerosity shown by two very different neural network models are better predicted by early visual responses to contrast. Changing the early visual contrast representation seems likely to change any response derived from this representation.
As such, our findings of multi-level neural adaptation impact on the current understanding of numerosity processing in three crucial ways. First, while numerosity processing is often considered a high-level cognitive function linked to mathematics and decision-making, with neural adaptation correspondingly shown in association cortices, recent studies increasingly point to lower-level processes in early visual cortices. Our work confirms this by revealing neural numerosity adaptation effects at these lower levels. Second, we demonstrate that neural adaptation effects accumulate hierarchically through multiple stages of early visual processing. This suggests that numerosity adaptation may build on (rather than being solely explained by) effects on early visual contrast processing. Finally, by suggesting that perceptual numerosity adaptation may be partly rooted in early visual mechanisms, our study offers a potential neural basis for recent behavioural findings demonstrating the influence of low-level visual features on perceptual numerosity adaptation.
We have previously used this data set to reveal numerosity adaptation effects on the numerosity preferences of numerosity-tuned neural populations in the parietal, frontal and lateral occipital lobes. We tested whether the strength of this tuned neural adaptation effect was correlated with the strength of the monotonic adaptation effect described here. Unfortunately, this analysis (Supplementary Fig. 6b, c) lacked the statistical power to show such correlations because the data set only included eight participants or 16 hemispheres. Some trends suggest that hemispheres with larger reductions in monotonic response slope in V1-hV4 may also show greater suppression of tuned responses and larger changes in numerosity preferences between high and low adaptor conditions, specifically in numerosity map NTO. Beyond NTO, the tuned numerosity preferences exhibit a strong hemispheric lateralization, which was not found in the monotonic early visual responses.
All our conditions only present the adaptor very briefly (and typically once) before each presentation of a changing numerosity, although many times over different presentations of changing numerosities. Is this sufficient to produce repulsive numerosity adaptation effects in perception? Or does this instead produce attractive serial dependence effects that occur when single presentations of a particular numerosity bias perception of the numerosity in the next presentation? We have previously shown that the stimulus timing used here produces a clear repulsive adaptation effects. Previous results also showed repulsive adaptation effects with brief adaptor presentations. Again, here, these brief but frequent presentations, although separated by changing numerosities, would be expected to affect the average level of recent activity in the early visual cortex that we propose underlies the effects we observe.
Demonstrating perceptual adaptation with our stimulus timing requires participants to attend to the test numerosity, though not necessarily the adaptor. While judging the test numerosity necessitates attention to numerosity, which might influence its perception, our fMRI participants attended the items without judging the changing numerosities. Attention to numerosity itself enhances tuned neural responses to numerosity, but we have repeatedly shown responses to numerosity without attention specifically to numerosity. Attention to numerosity may potentially enhance early visual response amplitudes, so adding explicit attention to numerosity might strengthen the effects we describe, but we do not expect it to qualitatively change these effects.
While we have previously shown that a lack of any attention to the dots (in a non-numerical task) strongly affects responses to numerosity, our fMRI participants did attend to the dots in a non-numerical task. Although neither our behavioural nor fMRI experiments required attention to the numerosity of the adaptor, they may differ in the attention allocated to the dots or their location. In our fMRI data, adaptors were always presented at fixation, and we compared the effects of the different adaptors. However, in our behavioural experiment, an adaptor was presented on one side of the screen with nothing on the other side. In this situation, the appearance of the adaptor on one side only is likely to implicitly attract attention to the adaptor location and numerosity and so increase the neural response to the adaptor and its adaptation effects. Therefore, perceptual adaptation effects in our fMRI experiment may be weaker than in our behavioural experiment. Nevertheless, in our fMRI experiment, the appearance of the adaptor alone and its presentation at fixation are also likely to attract attention. Further work would be needed to address how attention affects neural numerosity adaptation effects.
Functionally, adaptation is usually proposed to adapt perception to the context of recently seen sensory stimuli, thereby increasing sensitivity in the stimulus range we are currently working with by increasing discriminability around the adapted range. Seeing contrast adaptation as a fundamental contributor to numerosity adaptation instead suggests numerosity adaptation's functional role may be to help separate numerosity from contrast. Both numerosity and the contrast between items and their background (i.e., item contrast) similarly affect the image's total Fourier power in the spatial frequency domain (i.e., image contrast). An image can have greater total Fourier power because it contains more items or greater item contrast. To determine numerosity, we need to normalize the image contrast for item contrast. Indeed, responses in V1 are strongly contrast-dependent, while responses in the first areas showing numerosity-tuned responses (visual field maps TO1 and TO2, i.e., area hMT+) are minimally affected by item contrast. Therefore, under normal circumstances, contrast adaptation may serve to normalize item contrast by considering the contrast of recently viewed items, and thereby yield a contrast-invariant representation to numerosity. However, during the unusual circumstances of numerosity adaptation, numerosity affects image contrast while item contrast is held constant. This may thereby disrupt this normalization process, leading to inaccurate numerosity perception. This view sees mechanisms of numerosity adaptation as inherent to the process of numerosity estimation itself, rather than an adaptive aspect of numerosity perception. These views are not mutually exclusive.