The Neural Underpinnings of Functional MRI Networks

可能对我们理解人脑运作产生最大影响的技术进步是功能磁共振成像(FMRI)。它非凡的能力可以观察大脑内部，不仅观察大脑的结构，而且观察它的功能，为基础和临床研究项目服务。就其核心而言，功能磁共振成像代表了血流局部变化的读数，最常见的是神经活动的局部变化。由于血液流动和神经活动的运作原理完全不同，准确地确定它们之间的具体联系一直难以捉摸，而且似乎取决于许多因素。这并不令人惊讶，因为人们如何在一个体素中数百万个神经元之间的特定活动模式与作为局部血流动力学信号测量的单个标量值的缓慢变化之间建立一对一的映射？尽管这个问题令人沮丧，但这个话题非常重要，因为任何关于与局部神经活动或上升神经调节有关的线索，都可能对解释人类的结果产生广泛的影响，包括精神疾病患者。虽然我们的实验室并不专注于神经血管耦合本身的研究，但我们确实进行了一些实验，为解释血流动力学fMRI信号带来了新的见解。例如，我们正在研究不同类型信号的尖峰反应中局部神经多样性的性质，以及这如何影响来自同一体素或区域的血流动力学反应。我们还在调查整个大脑的大规模功能核磁共振网络中的活动与在单一位置测量的局部神经活动之间的关系。 在过去的一年里，我们在理解单一神经元放电模式和整个大脑测量的基于血液的活动之间的联系方面取得了进展。在最近发表的一项研究中(Zaldivar等人，2022，PNAS)，我们研究了自发单位活动与同时测量的fMRI信号的关系。实际上，我们能够使用全脑功能磁共振成像来表征局部人群中神经元的自发放电。这需要同时收集磁共振扫描仪内的单个单元和功能磁共振反应。由于与这类工作相关的技术障碍是巨大的，我们需要开发和获得与MR兼容的电极、微型驱动器、合适的射频线圈、前置放大器、电缆和过滤器来实现这种同时记录。在过去的一年里，我们发现，与第一项研究中描述的自然主义视觉反应相比，自发活动的fMRI映射更加均匀，跨皮质和皮质下区域的限制更多。 在一项与自然视觉更直接相关的研究中，我们还结合了单单元和全脑fMRI来研究局部神经群体的功能组成，在这种情况下，是在观看自然电影期间(Park等人，2022，Sci ADV)。在那项研究中，我们得出了一个令人惊讶的结论：通常认为，在分析视觉场景时，功能不同的视觉皮质区域具有类似的神经元混合，其功能专门化程度差异很大。我们的研究偏离了将孤立的图像显示在空白屏幕上的标准模式后得出了这一结论。相反，我们在免费观看自然主义电影的过程中对受试者进行了测试。我们发现，在这些条件下，所有名义上对面部有选择性的相邻神经元对非常不同类型的特征做出反应。此外，通过将它们的反应曲线与整个大脑中的体素的反应曲线进行比较，我们发现局部人群中的神经元与整个大脑的网络显示出不同的对应范围。这些发现的含义是，在自然视觉模式下，视觉皮质不是由以循序渐进的方式处理刺激的离散的、功能相同的区域组成的。相反，更广泛的专门化区域被并行的功能子网络所弥漫，这些子网络有助于多个大脑区域的功能。这一发现与普遍持有的观点相矛盾，即视觉大脑的基本组织是以严格的功能分离为标志的，这是许多实验的关键假设。总之，这两项研究为fMRI反应的性质提供了新的见解，包括局部神经血管关系，以及高层视觉的网络布局原则。

The technological advance that has likely had the greatest impact on our understanding of the operation of the human brain is functional magnetic resonance imaging (fMRI). Its remarkable capacity to look inside the head and view not only the structure of the brain, but its function, serves both basic and clinical research programs. At its core, fMRI represents a readout of local changes in blood flow that is most often derived from local changes in neural activity. Since blood flow and neural activity operate by entirely different principles, pinpointing their specific connection has been elusive and seems to depend on a number of factors. This is not surprising, for how can one make a one-to-one mapping between a specific pattern of activity among millions of neurons in a voxel to a slow change of single scalar values measured as the local the hemodynamic signal? Frustrating as the problem is, the topic is of great importance, since any clues about the link to local neural activity or ascending neuromodulation can have wide-reaching consequences for interpreting results in humans, including in psychiatric patients. While our laboratory does not focus on the study of neurovascular coupling per se, we do undertake experiments that bring new insights into the interpretation of the hemodynamic fMRI signal. For example, we are studying the nature of local neural diversity of in the spiking responses to different types of signals, and how this bears on the hemodynamic responses from the same voxel or area. We are also investigating the relationship between activity in large-scale functional MRI networks across the brain to local neural activity measured at a single position. In the past year, we have made headway on understanding the link between a single neurons firing pattern and the blood-based activity measured throughout the brain. In one recently published study (Zaldivar et al, 2022, PNAS) we investigated the relationship of spontaneous single-unit activity to concurrently measured fMRI signals. In effect, we were able to use brain-wide fMRI to characterize the spontaneous firing of neurons within a local population. This required the simultaneous collection of single-unit and fMRI responses inside the MR scanner. As the technical hurdles associated with this type of work are immense, we needed to develop and acquire MR-compatible electrodes, microdrive, suitable RF coils, preamplifiers, cables, and filters to achieve such simultaneous recording. In the past year, we have discovered that, in contrast to the naturalistic visual responses described in the first study, the fMRI mapping of spontaneous activity is more homogeneous and more restricted across cortical and subcortical areas. In a related study more directly related to natural vision, we have also combined single unit and whole-brain fMRI to study the functional composition of local neural populations, in this case during the viewing of natural movies (Park et al, 2022, Sci Adv). In that study, we came to the surprising conclusion that functionally distinct regions of the visual cortex, which are generally believed to divide their labor in the analysis of a visual scene, have similar mixtures of neurons whose functional specialization varies greatly. Our study came to this conclusion after departing from the standard mode of showing isolated images onto a blank screen. Instead, we tested subjects during the free viewing of naturalistic movies. We found that neighboring neurons, all nominally selective for faces, respond to very different types of features under these conditions. Furthermore, by comparing their response profiles to that of voxels throughout the brain, we found neurons within a local population showed a diverse range of correpondences with networks across the brain. The implications of these findings is that, under natural modes of vision, the visual cortex is not composed of discrete, functionally homogeneous areas that process stimuli in a stepwise fashion. Instead, more broadly specialized regions are pervaded by parallel functional subnetworks, which contribute to the functioning of multiple brain areas. This finding contradicts the commonly held view that the fundamental organization of the visual brain is marked by strict segregation of function, which is a critical assumption for many experiments. Together, these two studies provide new insights into the nature of fMRI responses, including the local neurovascular relationship, as well as the network layout principles of high-level vision.