本研究在猕猴下颞叶皮层(IT)约 1 mm³ 的局部微组织柱内利用 64 通道高密度阵列同时记录约 80 个神经元的放电活动,在逐试次尺度上检验了神经元间的“噪声共变”(noise covariation,即无外加刺激变化时神经元放电在试次间的自发波动关联)对物体表征效率的影响。研究表明:神经元间的弱噪声共变确实降低了 IT 皮层的物体分类编码表现;然而,决定整体编码效率的核心因素是网络各向异性(network anisotropy,即信号相关在神经群体中的非均一空间分布)——位于皮层输出层且具有高度信号相关的神经元亚群构成了物体表征的核心载体。该工作为理解高级视皮层利用低维流形表征复杂客体信息的计算机制提供了电生理依据。
研究背景
神经元放电相关性如何影响大脑皮层群体的编码效率是计算神经科学的核心论题。早先基于初级视皮层(V1)或中颞区(MT)的研究指出,物理临近神经元的信息传递在很大程度上相互独立(噪声相关 rSC 均值较低;Gawne 与 Richmond 1993;Kohn 与 Smith 2005),从而支持了 Barlow 经典的“去相关/稀疏化表征以实现高效编码”假说(Barlow 1961;Olshausen 与 Field 1996)。然而理论推导表明(Abbott 与 Dayan 1999;Averbeck 等 2006),噪声共变对信息编码的影响取决于信号相关(rsignal,即感受野调谐相似度)与噪声相关之间的空间相对关系:当信号相关与噪声相关同向对齐时,神经噪声会沿信号轴扩散从而损害信息读取;若二者正交,噪声共变的影响则相对有限。
既往关于神经相关的实验多局限于低级感觉皮层的简单参数辨别任务(如朝向辨析),缺乏在支持高阶物体识别的高级视皮层中对密集神经元群体开展的系统测量。作者团队的前期工作(Hung 等 2005;Lin 等 2014)通过多通道微电极记录发现,在下颞叶皮层内部,调谐高度相似的神经元亚群在物体分类解码中表现出更高的信息载荷,这一现象与传统去相关假说存在分歧。由此引出的科学问题是:在逐试次的神经动态中,自发的噪声共变究竟如何影响这些高相关神经元群体的物体表征,以及下游线性读出机制是否依赖于局部相关结构。
研究思路
作者对密集微电极阵列采集的高通量神经信号进行逐试次重构,检验“噪声共变在规模更大、同质性更高的 IT 神经群体中会削弱物体编码与解码”的假说。分析采用三条相互印证的技术路线:第一,利用支持向量机(SVM)线性分类器对比“未打乱数据”(保留真实同时记录的试次间相关)与“试次打乱数据”(在每个物体条件下独立随机打乱试次时序以消除神经元间的瞬时共变),并区分编码评估(在相同相关条件下训练并测试)与解码评估(利用打乱数据训练分类器,再应用于未打乱数据测试,评估解码端掌握相关结构是否带来增益);第二,测定皮层内部的微构筑相关特性——解析信号相关与噪声相关的偶联程度,评估其随皮层深度(上颗粒层、颗粒层、下颗粒层)的分布规律,并依据神经元与局部同伴的平均信号相关强度,将群体划分为第 1 组(Group 1,高信号相关亚群)与第 2 组(Group 2,中等相关亚群);第三,向试次平均响应中注入受控的合成噪声相关以进行仿真,检验不同噪声水平对分类表现的定量影响。
方法
实验记录来自 3 只轻度神经安定麻醉与肌松状态下的台湾猕猴(1 雄 2 雌;静脉维持芬太尼 0.9 µg/kg/h,配合氧化亚氮、异氟醚与肌松剂),在右侧下颞叶皮层前部侧凸面(AP16 位点)植入 64 通道硅基微电极阵列(8 针柄 × 每柄 8 个触点,触点水平及垂直间距均为 0.2 mm,覆盖 1.4×1.4 mm 的微小柱状容积),共完成 5 次独立记录会话,单次会话最多同步分离出 306 个神经单元(图 1A)。信号经 400–5000 Hz 滤波与 24.4 kHz 数字化,采用 WaveClus 算法分离单单元活动(SUA),结合多单元活动(MUA)与阈值穿越散列信号(hash MUA)作为对照,剔除放电频率低于 4 Hz 或存在慢速状态漂移的单元;皮层深度在穿刺时直接标定并依据层级电生理特征划分为上颗粒层、颗粒层与下颗粒层。
视觉刺激包含 240 幅无色彩与表面纹理的灰度三维物体模型渲染图(涵盖动物、人脸、植物、工具、车辆、家具等类别),中央视场单眼呈现(视视角 10°),采用 5 Hz 快速连续呈现范式(94 ms 刺激呈现 / 106 ms 灰屏间隔),伪随机重复 10 次;另设姿态/光照泛化测试块,对 10 个代表性物体呈现 25 种变换组合(5 种三维偏转姿态 × 5 种照明光源方位)。分类解码采用软间隔线性 SVM 分类器,输入特征为刺激呈现后 100–300 ms 的标准化放电计数。测试任务包括四类:① 类别内物体泛化(训练同类 8 个物体,测试未见的另外 5 个同类物体,随机水平 12.5%);② 240 个单一物体的二分类识别(随机水平 0.42%);③ 姿态与光照条件泛化(训练 4 种姿态/光照,测试未见条件,随机水平 10%);④ 8 分类任务的时序动态分析(25 ms 滑动时间窗)。信号相关计算各神经元试次平均调谐向量间的皮尔逊相关,噪声相关计算消除刺激均值后逐试次放电波动间的相关,统计均在不同物理电极通道之间开展以排除动作电位波形重叠干扰。
主要结果
-
噪声共变降低物体编码表现,但掌握相关结构的解码器未见额外增益:在类别内物体泛化任务中,试次打乱(消除瞬时噪声相关)后的分类识别率显著高于真实未打乱数据,编码表现下降幅度约为 5%–10%,多类别达到统计显著(人脸与动物 p<0.005,其他类别处于 p<0.01 或 0.05 水平;图 1B);在 240 个物体的单一物体识别中,当积分时间窗大于 50 ms 时同样表现出一致的编码损失(图 2);在未见姿态与光照条件的泛化任务中亦观察到 5%–10% 的编码下降(图 3)。然而,在所有任务中,使用消除相关的打乱数据进行训练、并在真实未打乱数据上进行测试的解码分类器,其识别率与基线解码器相比未见统计学提升。该效应在类别间表现不均:在具有集群表征倾向的人脸与动物类别中,编码受损效应较强,而工具类别未见显著效应。
-
编码损失随群体规模扩大而增强,且在早期反应阶段即已确立:编码损失随神经元群体规模增加而显著加剧,在单阵列包含 32–64 个单元时,相对表现下降(ΔP/Pcorrected)达到约 10%–15%,而在少于 8 个神经元的小群体中效应相对较弱;相比之下,解码端效应在所有规模梯度下均未达显著性(图 4A–C)。时间进程分析显示,编码损失在刺激呈现后最早期的神经放电时程(约 100 ms)即已出现,其动态轨迹与物体分类本身的涌现时程高度重叠,表明该现象伴随初级前馈视觉信号产生(图 4D)。
-
信号相关与噪声相关呈现高度正向耦合:在五个独立阵列中,神经元对之间的信号相关 rsignal 与噪声相关 rSC 均呈现显著正相关(阵列内 Fisher 变换相关系数 R 介于 0.38 至 0.79 之间,SUA 均值为 R=0.63;图 5)。总体而言,下颞叶局部的平均相关系数处于较低水平(rsignal 均值为 0.11,rSC 均值为 0.05,共计 10,645 个神经元对)。理论上,调谐相似性与噪声波动的正向排列构成了信息量损失的几何结构,为结果第 1 条提供了生理学基础。
-
编码效率与皮层层次构筑及网络各向异性密切相关:噪声相关在输入颗粒层(皮层深度 1.0–1.2 mm)表现最弱(SUA 均值为 0.021),显著低于上颗粒层(0.066)与下颗粒层(0.065,p<0.001;图 6B)。在功能表现上,上、下输出层的物体分类性能优于颗粒层,且噪声共变引发的编码性能损失特异性集中在输出层,在颗粒层不显著(图 6C)。依据各向异性分组后,高信号相关亚群(Group 1,主要分布于输出层)在类别泛化、个体识别及姿态/光照泛化中的绝对解码表现全面优于中等相关亚群(Group 2);尽管 Group 1 承担了更明显的噪声共变损失,但在相同神经元数量下其分类表现更高,展现出更高的群体编码效率(图 7)。
-
数值仿真复现了实验观察并指示其亚群来源:通过向试次平均响应中注入已知强度的合成噪声,模拟表明负向 rSC 可微弱提升编码表现,而真实实验数据的表现对应于有效 rSC 约为 0.2 的模拟状态(图 8A)——显著高于全脑区实测均值 0.05,表明物体信息的主要编码载体是局部高相关性的 Group 1 输出层神经元。在全部实测配对中,检测到一小部分稳定的负向噪声相关对(约占 4.1%),提示局部微回路中存在抑制性神经元的侧向调控作用。
图注解读
图 1 · 噪声共变降低类别泛化的编码
原文图注:Figure 1. Noise covariation reduces object category generalization encoding in IT. A, Array locations 1–5, centered at A16 of right lateral inferior temporal cortex, recorded in separate sessions across 3 macaque monkeys under light neurolept anesthesia. Arrays had 64 contacts (8 shanks, 8 contacts/shank) spaced 0.2 mm apart, spanning 1.4 × 1.4 mm horizontally and in depth. STS, Superior temporal sulcus; AMTS, anterior medial temporal sulcus. B, Effect of noise correlation on classifier performance for within-category generalization. Linear classifiers were trained on 8 objects per category and tested on 5 other objects per category. Chance is 12.5% for 8 categories. For encoding, classifiers were trained and tested on unshuffled (actual) data versus trained and tested on trial-shuffled data, shown for individual categories (colors) and for the average across categories (black). For decoding, classifiers were trained and tested on unshuffled data, versus trained on shuffled data and tested on unshuffled data. Classifier input was spike count in the 100:300 ms period, pooled across 5 arrays and all trials. Data are mean ± SEM (50 permutations of objects and trials). p<0.05 (two-tailed t test, uncorrected). p<0.01 (two-tailed t test, uncorrected). **p<0.005 (two-tailed t test, uncorrected). C, Average category responses for the 5 arrays. Spike counts 100:300 ms after stimulus onset were averaged across 10 trials, z-normalized across all 240 objects for each contact, and then averaged within each category. z-scores are low because different objects in a category tend to activate different sets of contacts. White represents broken/inactive contacts.

读图说明:(A) 展示 5 次微电极阵列植入的解剖位置,位于下颞叶侧凸面 A16 坐标,介于颞上沟(STS)与前内侧颞沟(AMTS)之间;(B) 展示类别内泛化任务的分类性能对比,彩色柱代表 8 个具体类别,黑色柱代表平均表现;每组对比包含编码比对(真实未打乱 vs 打乱数据)与解码比对(真实解码 vs 打乱训练/真实测试)。平均柱在试次打乱后显著升高,而解码比较无差异,支撑主要结果第 1 条中关于编码受损与解码无补偿的结论;(C) 呈现各阵列在不同类别下的平均响应热图(z 分数,白色表示无响应触点),反映了类别反应在局部空间上的拓扑聚集。
图 2 · 噪声共变也损害单个物体的识别
原文图注:Figure 2. Noise covariation reduces object identification encoding. Effect of noise covariation on classifier performance for object identification, for encoding and decoding across different bin sizes within the 100:300 ms window. Classifier training was based on 5 trials, and classifier testing was based on 5 other trials. Performance is shown for the best bin of each bin size (e.g., the 150:175 ms bin for the 25 ms bin size). Chance is 0.42% for 240 objects. Data are mean ± SEM (50 permutations). p<0.05 (two-tailed t test, uncorrected). **p<0.005 (two-tailed t test, uncorrected).

读图说明:横坐标表示时间窗宽度(12.5–200 ms),纵坐标为 240 个单一物体的二分类识别准确率(随机概率水平为 0.42%),绘制最佳时间窗下的表现。当积分时间窗大于 50 ms 时,打乱试次后的编码识别表现显著高于未打乱状态,证实噪声共变损害单一客体的细粒度编码,而非单纯源于类别归纳时的概念合并。该图对应主要结果第 1 条。
图 3 · 姿态与光照泛化同样受损
原文图注:Figure 3. Noise covariation reduces pose and illumination generalization. A, Pose and illumination variations for one object. B, C, Classifier performance for generalization across pose and illumination (colors) for encoding and decoding. Based on arrays 2 and 4, recorded in separate sessions. For each array, we tested 25 variations of 10 preferred objects. Black represents average across poses/illuminations. Classifiers were trained on 4 poses or 4 illuminations (20 variations per object) and tested on object identification at the unseen pose or illumination. Data are mean ± SD (50 trial permutations). Chance is 10% for 10 objects. p<0.05. **p<0.005. n.s., Not significant.

读图说明:(A) 展示单个物体在 5 种偏转姿态与 5 种照明光源方位下的变异矩阵;(B) 与 (C) 分别展示跨姿态与跨光照泛化分类性能,彩色柱代表各个测试条件,黑色柱代表平均表现(随机水平为 10%)。未打乱数据在编码测试中显著落后于打乱数据(B 中 p<0.005,C 中 p<0.05),而解码端未见显著差异,证实视角与光照容忍性泛化编码同样受噪声共变牵制,支撑主要结果第 1 条。
图 4 · 效应随群体规模增大且出现极早
原文图注:Figure 4. Dependence of noise covariation effect on ensemble size and latency. A, Effect of noise covariation on category encoding as a function of number of units in each of 5 arrays (colors). Classifiers were trained and tested on unshuffled data (solid lines) or trained and tested on shuffled data (dashed lines). Classifiers were trained on 5 trials and then tested on the remaining 5 trials, based on spike count in the 100:300 ms window. Data are mean ± SEM (50 permutations of units and trials). Chance = 12.5% for 8 categories. B, Effect of noise correlation on category decoding. Classifiers were trained and tested on unshuffled data (colored solid lines) or trained on shuffled and tested on unshuffled data (gray dashed lines). C, Dependence of effect on population size for encoding (left) and decoding (right). Colors show results for individual arrays, based on SUA. ΔP/Pcorrected is calculated as follow: (PU − PS)/(PU − PC), where PU, PS, and PC are performance on unshuffled training trials, shuffled training trials, and chance (12.5% for 8 categories), respectively. Mean ΔP/Pcorrected is shown as thick black line for SUA, dashed black line for MUA (pooled from SUA), and dashed red line for hash MUA. Shaded region is SEM (50 permutations). D, Time course of object categorization performance for encoding (left) and decoding (right). Classifier input was the average spike rate for one 25 ms bin in the 100:300 ms window, pooled across 5 arrays. Data are mean ± SEM (50 permutations, SUA). p<0.05; p<0.01; **p<0.005 (two-tailed t test, uncorrected).

读图说明:(A) 与 (B) 描绘 5 个阵列中分类性能随神经元数量扩增的演化轨迹,实线为真实数据,虚线为打乱数据,编码端两线间距随规模增大而扩张,解码端两线重合;(C) 将间距转化为归一化相对变化量 ΔP/Pcorrected,显示编码端相对损失随规模上升至 10%–15%,解码端保持平直;(D) 为 25 ms 滑动窗的时间动力学曲线,编码端差异在刺激后约 100 ms 即达到统计学显著。该图支撑主要结果第 2 条。
图 5 · 信号相关与噪声相关同向而行
原文图注:Figure 5. Signal correlation and noise correlation are strongly linked. A, Signal correlation (rsignal) versus noise correlation (rSC) of neuronal pairs in array 1, based on 81 SUAs (3214 pairs, recorded from separate contacts). rsignal and rSC are measured as Pearson correlation of trial-averaged stimulus responses and trial-by-trial response deviations, respectively, based on spike count from 100 to 300 ms after stimulus onset. Stimuli were 240 objects shown for 10 repetitions in pseudorandom order. Gray line indicates model II regression of rsignal versus rSC. Red marginal represents rSC after trial shuffling. ***p<0.005. B, Correlation (R) between rsignal and rSC across 5 arrays. Horizontal lines are mean across 5 arrays for hash MUA (r=0.69) and for SUA (r=0.63). p<0.005 for each array.

读图说明:(A) 呈现阵列 1 中 3214 个单单元配对的信号相关(rsignal)与噪声相关(rSC)散点分布,灰色实线为模型 II 回归拟合线,右侧红色边际直方图证实打乱试次后噪声相关归零;(B) 汇总 5 个阵列的偶联相关系数 R,SUA 均值为 0.63,hash MUA 均值为 0.69,且各阵列独立显著(p<0.005)。证实下颞叶神经元中调谐相似度与瞬时噪声波动正向偶联,支撑主要结果第 3 条。
图 6 · 相关结构与编码的皮层层次分布
原文图注:Figure 6. Cortical depth dependency of signal correlation, noise correlation, and coding efficiency. A, Cortical depth versus proportion of Group 1 units, defined as the top 20th percentile of units in each array sorted by average signal with other units in the same array, based on SUA. B, Noise correlation (rSC) at supragranular (0.2–0.8 mm), granular (1.0–1.2 mm), and infragranular (1.4–1.6 mm) depths for arrays 1–3, based on hash MUA and SUA. Layers were estimated from cortical depth. Red lines indicate mean ± SD. Arrays 4 and 5 were excluded because they sampled mainly supragranular layers. p<0.05. p<0.01. **p<0.005. C, Effect of noise correlation on encoding (blue, red, and black) and decoding (gray) at supragranular, granular, and infragranular depths for within-category identification, categorization, and within-category generalization, for arrays 1–3. Based on 8 units with highest average rsignal per depth group per array, defined without training/test objects. S, Shuffled; U, unshuffled. Data are mean ± SEM (50 permutations of stimuli and trials). Chance is 12.5% for all tests, for 8 categories and 8 objects/category.

读图说明:(A) 展示 Group 1 高相关神经元在不同皮层深度的分布比例,其在输入颗粒层显著匮乏;(B) 描绘三个皮层深度的噪声相关 rSC(红线为均值 ± 标准差),颗粒层(约 0.02–0.035)显著低于上颗粒层与下颗粒层(0.065–0.096);(C) 展示三类任务在各深度的编码与解码表现,输出层的分类性能整体优于颗粒层,且编码噪声受损效应仅特异性出现在输出层。该图支撑主要结果第 4 条。
图 7 · 各向异性:谁受益、谁买单
原文图注:Figure 7. Noise covariation effect on coding efficiency depends on network anisotropy. A, Effect of noise covariation on category encoding (colors) and decoding (gray) for Group 1 versus Group 2 units in each array. Group 1 is units with the highest average signal with other units in the same array, and Group 2 is units with average signal closest to the median of each array, defined without training/test objects. Classifier training/testing is same as in Figure 4, cross-validated across trials. For both panels: p<0.05; **p<0.005. Data are mean ± SEM (50 permutations). n.s., Not significant. B, Effect of noise covariation on encoding (red, black) and decoding (gray) for categorization, within-category generalization, and pose and illumination generalization for Group 1 and Group 2 units pooled across arrays (compare with Figs. 1B, 3B, C). Solid lines are Group 1 and Group 2 units defined as the 30% highest (red) and 30% median (black, 35th-65th percentile) average rsignal of each array, defined without training/test objects. Dashed lines are Group 1 and Group 2 units defined as the 20% highest (red) and 20% median (black) average signal of each array. Data are mean ± SEM (50 permutations of stimuli and trials).

读图说明:(A) 分阵列对比 Group 1 与 Group 2 单元的编码与解码轨迹,Group 1 曲线整体位于上方,展现出更高的分类表现与更明显的编码噪声差距;(B) 跨阵列汇总展示在不同任务下的分组对比,实线为 30% 分界,虚线为 20% 分界。Group 1 承受了更大的噪声共变损失,但在相同神经元数量下其分类表现显著更高,证明其以较小群体规模实现了更高效的信息表征,对应主要结果第 4 条后半部分。
图 8 · 模拟注入人工噪声相关
原文图注:Figure 8. Effect of simulated noise correlation on category encoding and decoding. A, Effect of simulated noise correlation on category encoding. ΔIshuffled/I versus population size for simulated noise correlation rSC of -0.01 to 0.2 (black). I, Information; Ishuffled, trial-shuffled information; ΔIshuffled, I−Ishuffled. Noise correlation was simulated by injecting noise to trial-averaged data (blue). Results for unshuffled data (actual, red) approximate those of simulated rSC of 0.2. ΔIshuffled/I is positive for negative rSC. Compare with Averbeck et al. (2006; their Fig. 2). B, ΔIdiag/I versus population size. Idiag, Information that would be extracted by a decoder trained on trial-shuffled data but tested on actual correlated (unshuffled) data; ΔIdiag, I−Idiag. Compare with Averbeck et al. (2006; their Fig. 4).

读图说明:(A) 描绘信息量相对变动 ΔIshuffled/I 随神经元数量的变化,黑色曲线为不同模拟 rSC 水平(−0.01 至 0.2),红色曲线为真实实测数据。实测数据与 rSC = 0.2 的模拟轨迹高度接近,显著偏离全区均值 0.05,指示信息主要由局部强相关子群承载;(B) 描绘解码端相对信息差额 ΔIdiag/I,随群体规模由负值转为正值,提示非线性或高阶解码机制在理论上存在利用相关结构的可能性。该图支撑主要结果第 5 条。
讨论
作者在讨论中系统总结了研究的核心论断:尽管弱噪声相关(均值 rSC 仅 0.05)对下颞叶皮层的高阶物体分类编码造成了一定程度的性能损失,但在支持高维物体流形表征时,群体内部由网络各向异性所塑造的信号相关结构是维持高效表征的关键。作者提出了结构分工的定性模型:Group 1 高相关神经元主要分布于皮层输出层,通过时间上的同步放电构成低维流形的核心骨架;Group 2 神经元呈现更为广泛的去相关特性,为流形表征补充高维细节。这一结构与皮层微柱功能构筑、网络自适应重组速度取决于各向异性(Panas 等 2015)以及腹侧视流通路逐级提升特征容忍性(Rust 与 DiCarlo 2010)的理论相契合;效应在刺激呈现后 100 ms 内迅速涌现,排除了迟发反馈或注意调控的解释。
作者指出了本研究的边界条件与假设前提:首先,研究采用线性支持向量机作为下游读出模型,若下游脑区采用非线性核或高阶重合检测算法,解码端对相关结构的利用可能呈现不同模式;其次,实验在轻度神经安定麻醉条件下开展,尽管监测证实局部噪声相关高于 3.3 Hz 频段且肌松排除了扫视运动等主动行为干扰,麻醉状态是否在一定程度上弱化了大尺度网络状态的自发波动,仍需在清醒主动任务动物中开展对照验证;第三,群体水平的信号相关与噪声相关如何定量映射至动物真实的行为辨别阈值与反应时间,仍有待未来开展联机行为闭环电生理研究加以澄清。
一句话总结
本研究证实猕猴下颞叶皮层内部的弱噪声共变虽对逐试次物体编码表现产生抑制,但决定整体编码效率的核心在于网络各向异性:位于皮层输出层、具有高度信号相关的局部神经元亚群构成了客体流形表征的主要信息载体。这一发现揭示了高级感觉皮层利用低维关联结构平衡神经噪声与高维表征容量的回路计算机制。