Spike-timing patterns - crucial for synaptic plasticity and neural computation - are often modeled as Poisson-like random processes, log-normal distribution or gamma-distribution patterns, each with different underlying assumptions that may or may not be biologically true. However, it is not entirely clear whether (and how well) these different models would or would not capture spike-timing statistical patterns across different neurons, regions, animal species and cognitive states. Here, we examine statistical patterns of spike-timing irregularity in 13 different cortical and subcortical regions from mouse, hamster, cat and monkey brains. In contrast to the widely-assumed Poisson or log-normal distribution patterns, we show that spike-timing patterns of various projection neurons - including cortical excitatory principal cells, hippocampal pyramidal cells, inhibitory striatal medium spiny neurons and dopaminergic neurons, as well as fast-spiking interneurons - all invariantly conform to the gamma-distribution model. While higher regularity in spike-timing patterns are observed in a few cases, such as mouse DA neurons and monkey motor cortical neurons, there is no clear tendency in increased firing regularity from the sensory and subcortical neurons to prefrontal or motor cortices, as previously entertained. Moreover, gamma shapes of spike-timing patterns remain robust over various natural cognitive states, such as sleep, awake periods, or during fearful episodic experiences. Interestingly, ketamine-induced general anesthesia or unconsciousness is associated with the breakdown of forebrain spike patterns from a singular gamma distribution into two distinct subtypes of gamma distributions, suggesting the importance of this spike-timing pattern in supporting natural cognitive states. These results suggest that gamma-distribution patterns of spike timing reflect not only a fundamental property conserved across different neurons, regions and animal species, but also an operation crucial for supporting natural cognitive states. Such gamma-distribution-based spike-timing patterns can also have important implications for real-time neural coding and realistic neuromorphic computing.