Sensory cortex often undergoes population activity fluctuations due to attentional modulations and arousal changes, but activity fluctuations can also be generated intrinsically. The mechanisms that underlie these spontaneous activity fluctuations are unknown, in particular whether they arise from common inputs or more active neuronal processes. We measured the activity of tens of V1 neurons simultaneously in monkeys stimulated with oriented visual stimuli, and found that the tuning curves (TCs) of orientation selective neurons spontaneously underwent shifts and multiplicative scalings proportional to population activity while their widths remained constant Based on this observed TC activity dependence, we constructed a precise statistical model featuring Poisson-like firing, shifts and gain modulation. The model not only accounted for neuronal co-variability but also approached the performance of state-of-art decoding methods. Surprisingly, we found that decoding performance on sensory stimuli increased with population activity, despite the fact that the stimulus was identical. Therefore, spontaneous population activity fluctuations display highly non-random features, boosting TCs by shifts and multiplicative factors that gate information processing.

The observed boosting properties of TCs motivated us to build a statistical model with Poisson-like neurons that includes a multiplicative scaling factor (PS). We observed that the effect of population activity on TCs is not purely multiplicative, and we introduced therefore a shift (PSS). Our models approached the performance of state-of-art decoding techniques (SVM, and logistic regression -LR-) and provided higher accuracy than other tested decoders based on population vector and independent Poisson neurons. Experimental methods are as in [1].