NEW YORK – A novel cell segmentation method developed by scientists at the Fred Hutchinson Cancer Center may help researchers conduct more accurate spatial biology studies, advancing key efforts such as cancer biomarker and drug target discovery.

纽约——弗雷德·哈钦森癌症中心的科学家们开发了一种新的细胞分割方法，可能帮助研究人员进行更精确的空间生物学研究，推动癌症生物标志物和药物靶点发现等关键工作。

Cell segmentation is a fundamental first step in understanding spatial gene expression; essentially, which mRNA transcripts come from which cell.

细胞分割是理解空间基因表达的基本第一步；本质上，就是哪些mRNA转录本来自哪些细胞。

Evan Newell, an associate professor of vaccines and infectious diseases at the Fred Hutchinson Cancer Center and senior author of the study — published this morning in

埃文·纽厄尔，弗雷德·哈钦森癌症中心疫苗和传染病副教授，也是今天早上发表的这项研究的资深作者——

Nature

自然

—said that despite rapid advances in the field of spatial biology, there remains 'no great way to determine which transcripts to assign to which cells.'

—表示尽管空间生物学领域取得了快速进展，但“仍然没有很好的方法来确定将哪些转录本分配给哪些细胞”。

This ambiguity impacts differential gene expression analysis in a number of ways. A gene can appear upregulated in one cell population, for example, simply by proximity to another cell population expressing that same gene. Similarly, cells can differ in the number of transcripts they contain by several orders of magnitude, meaning that spillover between a cell with 1,000 transcripts and one with 50 transcripts can result in the former's signal completely dominating the latter's and even burying the identity of the low population cell..

这种模糊性在许多方面影响着差异基因表达分析。例如，一个基因在一个细胞群体中可能表现为上调，这仅仅是因为它靠近另一个表达相同基因的细胞群体。同样，细胞所含的转录本数量可能相差几个数量级，这意味着含有1000个转录本的细胞和含有50个转录本的细胞之间的溢出效应，可能导致前者的信号完全掩盖后者的信号，甚至淹没低数量细胞的身份特征。

'Say I want to know what's different about T cells in a tumor versus not in a tumor,' Newell said. 'The T cells that are in the tumor are going to be in close proximity to tumor cells and the T cells that are outside of the tumor are not.'

“比如说，我想知道肿瘤中的T细胞与不在肿瘤中的T细胞有何不同，”纽厄尔说。“肿瘤内的T细胞会与肿瘤细胞非常接近，而肿瘤外的T细胞则不会。”

Any misassignment that occurs between these two T cell populations will then confound one's ability to identify tumor versus non-tumor gene expression, which can be relevant for everything from understanding how tumors evolve to identifying potential diagnostic and therapeutic targets.

任何在这两种T细胞群体之间发生的错误分配将会混淆识别肿瘤与非肿瘤基因表达的能力，这可能与从理解肿瘤如何演变到识别潜在的诊断和治疗靶点等所有方面都相关。

To infer more morphologically plausible cell boundaries and reduce such misassignment, the Fred Hutchinson team developed and tested a cell segmentation method called Proseg, for probabilistic segmentation.

为了推断出更多形态上合理的细胞边界并减少此类错误分配，弗雷德·哈钦森团队开发并测试了一种名为Proseg的概率分割细胞分割方法。

Proseg initially estimates cell morphologies from nuclear stains, in order to prevent the algorithm from generating an improbable number of false cells with artificially low gene expression. It then expands and alters these in an unsupervised manner until it arrives at shapes that best explain the distribution of observed transcripts, repositioning those that appear in implausible locations..

Proseg 首先从核染色中初步估计细胞形态，以防止算法生成数量不合理的假细胞以及人为的低基因表达。然后，它以无监督的方式扩展和调整这些形态，直到找到最能解释观察到的转录本分布的形状，并重新定位那些出现在不合理位置的转录本。

The investigators compared Proseg to other segmentation methods including Baysor, Cellpose, SCS, GeneSegNet, and Bering, which are all widely used and well known in the cell segmentation field. They compared these methods in various combinations using four datasets from Vizgen's MERSCOPE, NanoString's CosMx, and 10x Genomics' Xenium commercial platforms..

研究人员将Proseg与其他分割方法进行了比较，包括Baysor、Cellpose、SCS、GeneSegNet和Bering，这些方法在细胞分割领域都被广泛使用且众所周知。他们通过来自Vizgen的MERSCOPE、NanoString的CosMx以及10x Genomics的Xenium商业平台的四个数据集，以各种组合对这些方法进行了比较。

In benchmarking tests, Proseg ran approximately an order of magnitude faster than Baysor and Bering, due in part to a more effective parallelization scheme. Although it ran slightly slower than Cellpose, Proseg generally produced more accurate segmentation than the others.

在基准测试中，由于部分原因归功于更有效的并行化方案，Proseg 的运行速度比 Baysor 和 Bering 快大约一个数量级。尽管它比 Cellpose 稍微慢一些，但 Proseg 通常比其他方法产生更准确的分割结果。

In comparisons using several lung cancer datasets, Proseg consistently showed better transcript assignments, more accurate cell type identification, spatial co-location patterns between neutrophils and tumor cells that more closely agreed with nuclear segmentation, and reduced spurious co-expression..

在使用多个肺癌数据集进行的比较中，Proseg 始终表现出更好的转录本分配、更准确的细胞类型识别、中性粒细胞与肿瘤细胞之间的空间共定位模式更符合核分割结果，并减少了虚假的共表达。

Proseg's improved cell segmentation may help researchers better understand things like tumor biology, which could in turn, inform cancer biomarker and drug target discovery.

Proseg改进的细胞分割可能有助于研究人员更好地理解肿瘤生物学等内容，从而为癌症生物标志物和药物靶点的发现提供信息。

In one experiment, the Fred Hutch investigators applied Proseg to measuring the extent and variation of T-cell infiltration found in four renal cell carcinoma (RCC) patient samples run on 10x's Xenium platform. A high degree of T-cell infiltration is one marker of successful immunotherapy but the frequencies of tumor-specific T cells to non-specific bystanders can vary dramatically between tumors, complicating any measurement of infiltration..

在一项实验中，Fred Hutch 的研究人员应用 Proseg 来测量在 10x 的 Xenium 平台上运行的四个肾细胞癌 (RCC) 患者样本中发现的 T 细胞浸润的程度和变化。高度的 T 细胞浸润是成功免疫治疗的一个标志，但肿瘤特异性 T 细胞与非特异性旁观者的频率在不同肿瘤间可能会有很大差异，这使得任何浸润测量都变得复杂。

T-cell populations identified by Proseg clustered into CCL4+ CD8+ T-cells, CXCL13+ CD8+ T-cells, other CD8+ T-cells, other CD4+ T-cells, and natural killer cells. These populations were observed in very different proportions across the four samples. Two tumors showed primarily CXCL13+ CD8+ T-cells, which co-localized with tumor cells.

通过Proseg识别的T细胞群可分为CCL4+ CD8+ T细胞、CXCL13+ CD8+ T细胞、其他CD8+ T细胞、其他CD4+ T细胞和自然杀伤细胞。这些群体在四个样本中的比例差异很大。两个肿瘤主要显示CXCL13+ CD8+ T细胞，它们与肿瘤细胞共定位。

Both of these samples and, to a lesser degree, a third sample, showed the diminished tumor density indicative of a higher degree of T-cell infiltration..

这两个样本，以及在较小程度上，第三个样本，均显示出肿瘤密度降低，这表明T细胞浸润程度更高。

'The significance of the RCC findings is that there are far more T cells infiltrating these particular tumors than the other segmentation methods would have us believe,' said Daniel Jones, a computer scientist in Newell's lab and lead author of the study. 'Having more T cells helped us find some interesting subsets based on CXCL13 and CCL4 expression.

“RCC研究结果的意义在于，这些特定肿瘤中浸润的T细胞数量远比其他分割方法让我们相信的要多，”纽厄尔实验室的计算机科学家、该研究的主要作者丹尼尔·琼斯说。“拥有更多的T细胞帮助我们基于CXCL13和CCL4表达发现了一些有趣的亚群。”

Our results suggest CXCL13 as reliable marker of exhaustion, and that these exhausted T cells are those that have most infiltrated the tumor.'.

我们的研究结果表明，CXCL13 是耗竭的可靠标志物，这些耗竭的 T 细胞是那些最浸润肿瘤的细胞。

Ali Shariati, assistant professor of biomolecular engineering at UC Santa Cruz and an expert in cell segmentation, said that the field of spatial transcriptomics has advanced rapidly in recent years, particularly with respect to increasing the number of transcripts that can be detected within individual cells.

阿里·沙里亚提是加州大学圣克鲁兹分校生物分子工程的助理教授，也是细胞分割领域的专家，他表示，近年来空间转录组学领域发展迅速，尤其是在提高可检测到的单个细胞内转录本数量方面。

However, he explained, accurately defining cell boundaries so that transcripts can be assigned to the correct cell with high precision has remained a key bottleneck..

然而，他解释说，准确定义细胞边界以高精度将转录本分配给正确的细胞仍然是一个关键瓶颈。

'Proseg is a major advancement toward solving this problem,' Shariati said in an email.

“Proseg 是解决这个问题的重大进步，”沙里亚蒂在一封电子邮件中说。

Shariati praised Proseg's generalizability as one of the model's more impressive aspects, saying that its ability to perform well across multiple spatial transcriptomics platforms and cell types suggests a 'strong potential' for broad application in the spatial transcriptomics field.

Shariati称赞Proseg的通用性是该模型更令人印象深刻的特点之一，并表示其在多个空间转录组学平台和细胞类型中表现出色的能力，表明其在空间转录组学领域具有“广泛的应用潜力”。

'I'm curious to see how easy it will be to adopt this method and what level of tuning is required to apply Proseg to new datasets, [such as] spatial transcriptomics data from organoid or embryo models,' he said.

“我很好奇采用这种方法会有多容易，以及将 Proseg 应用于新数据集需要什么程度的调整，[例如] 来自类器官或胚胎模型的空间转录组学数据，”他说道。

Proseg's code and documentation are currently available to the public via Github, although Shariati suggested that certain adjustments could make it even more broadly accessible.

Proseg 的代码和文档目前可以通过 Github 公开获取，尽管 Shariati 表示某些调整可能会使其更加广泛地被访问。

'Integrating Proseg as a plugin within commonly used image analysis platforms like Napari or ImageJ could further make it easier for the community to incorporate [it] into their workflows,' Jones said.

琼斯表示：“将Proseg作为插件集成到常用图像分析平台（如Napari或ImageJ）中，可以进一步方便社区将其纳入他们的工作流程。”

Jones said that while Napari and ImageJ are both currently focused on pure imaging data and lack 'first class' support for transcriptomics data, the Proseg team would definitely consider making plugins for those services in the future.

琼斯表示，虽然 Napari 和 ImageJ 目前都专注于纯影像数据，并且缺乏对转录组学数据的“一流”支持，但 Proseg 团队未来肯定会考虑为这些服务开发插件。

Newell said that Proseg is already being used regularly by investigators at Fred Hutch in studies on cancer, infectious disease, and autoimmunity. His own lab is currently using it to study graft-versus-host disease.

纽厄尔说，Proseg 正在被弗雷德·哈奇的研究人员在癌症、传染病和自身免疫性疾病的研究中常规使用。他自己的实验室目前正用它来研究移植物抗宿主病。

Although Fred Hutch has patented Proseg, Newell said that there are no commercial plans related to it and that it will remain freely available for the foreseeable future. The team also intends to update the algorithm regularly as new improvements are found and new information comes to light.

尽管弗雷德·哈钦森中心已经为Proseg申请了专利，但纽厄尔表示，目前没有任何与之相关的商业计划，并且在可预见的未来它将保持免费可用。团队还打算随着新改进的发现和新信息的出现定期更新该算法。

'[Proseg] doesn't solve cell segmentation,' Newell said, 'but it drastically improves the interpretability of datasets.'

“[Proseg] 不能解决细胞分割问题，”纽厄尔说，“但它大大提高了数据集的可解释性。”