选择了五百个SGRNA作为初始目标库
。在500个SGRNA中
，有50个是非靶向控制指南RNA。每个基因的三个SGRNA针对人类HNSCC和SSCC中最常见的150个最常见或CNV基因 。我们选择了HNSCC突变频率≥6％的基因，在SSCC中≥30％ ，两者的拷贝数变化的15％≥15％
。使用CBIOPORTAL的癌症基因组学51,52评估突变频率
，用于SSCC53,54和HNSCC5555,55,56,57,58,59。基因的确切选择及其突变频率可以在扩展数据中找到 。图1上图1B（Venn图）以图形方式表示HNSCC和SSCC基因之间的重叠和差异
。单个指南是使用Broad Institute SGRNA设计工具（SGRNA设计师：CRISPRKO（https://portals.broadinstitute.org/gpp/public/analisy-tool-tools/sgrna-design）设计的）。选择每个目标的三个顶级指南 。从Brie Library60（Addgene＃73633）随机挑选了五十个非目标控制SGRNA
。所有500个SGRNA序列均在补充表9中列出。 　　原始的作物隔离式旋转媒介（Addgene＃86708）中的紫霉素耐药盒被用MCHERRY序列取代，该序列从PAAVS1-NDI-CRISPRI（GEN1）（Addgene＃73497）放大 ，并通过PFL23II和MLUI限制点进行克隆 。如前所述5
。对于所有单细胞实验
，将三个不同的库批次克隆并单独测序以确认均匀的SGRNA表示。 　　通过磷酸钙转染Lenti-X 293 T细胞（Takara Clontech，632180） ，用作物 - 麦克雷和助手质质粒PMD2.g和pspax2（添加质子质质质体122259和12260）
，通过磷酸钙转染（Takara Clontech，632180）进行了杀虫剂的生产 。转染后十六小时后 ，将培养基改为病毒生产培养基（DMEM（GIBCO 11965092）
，1％青霉素 - 链霉素 - 链霉素 - 谷氨酰胺（Gibco 10378016），1％100 mm pyruvate（Gibco 11360070）（Gibco 11360070） ，1％sodium bicarate bicarate bicarate bicarate bicarate bicarate bicarate bicarate bicarate bicarate bicarate bicarate bicarate bicarate soveralate bicarate bicarate bicarate ,，丁酸钠（Sigma-Aldrich B5887））
，以37°C/7.5％CO2孵育。转染后46小时收集病毒上清液，并通过0.45μm滤波器过滤（Millipore stericup快速释放Durapore S2HVU02RE）
。对于体内慢病毒转导 ，使用100 kDa MW截止点米利班中心70（Merck Millipore; UFC710008）
，将病毒上清液浓缩约2,000倍。使用SW 55 Ti转子在45,000 rpm和4°C下，使用SW 55 Ti转子进一步浓缩病毒 。将最终病毒颗粒重悬于病毒重悬浮缓冲液中（20 mM Tris pH 8.0 ，250 mm NaCl
，10 mM MgCl2，5％山梨糖醇） ，并在-80°C下储存
，直到用于滴定或注射
。对于低映射病毒，与上述相似地进行慢病毒的产生 ，并使用0.45-μm注射器过滤器（SARSTEDT 83.1826）收集病毒上清液并过滤
，并在-80°C下存储直至使用。 　　所有动物实验均严格符合《瑞士动物保护法》和瑞士联邦食品安全与动物福利办公室（BLV）的要求 。苏黎世广州的动物福利委员会批准了本研究中执行的所有动物方案和实验（动物允许ZH074/2019，ZH196/2022）
。 　　菌株TG（B6J.129（CG）-GT（ROSA）26Sortm1.1（CAG-CAS9* ， - EGFP）FEZH/J（表示为B6.CAS9）的遗传改性小鼠（CAG-CAS9*， - EGFP）是从杰克逊实验室购买的（026179）。（作为杰克逊实验室003243的TNFRSF1A-KO）通过瑞士免疫学鼠标存储库（Swimmr）获取 。Cas9等位基因的杂合子
，适用于胚胎发育阶段E9.5的慢病毒注射
。 　　如前所述，进行了子宫注射中的超声引导。简而言之 ，妊娠E9.5的女性用异氟烷麻醉 。将每个胚胎注入0.5μl高映射病毒
，并每个垃圾注入多达8个胚胎
。手术程序限制为30分钟，以确保快速恢复。 　　所有小鼠均位于苏黎世大学实验室动物服务中心（LASC） ，在湿度和轻度控制的环境中（22°C
，45-50％，12 h轻/黑暗周期）中置于单独通风的笼子中 ，并可以使用食物和水量litbitum 。B6.CAS9雄性用于杂交和怀孕的CD1女性。所有其他小鼠都是小组的
。通过双重荧光蛋白手电筒（Nightsea
，DFP-1）激发通过眼睛控制小鼠的成功感染。然后通过斩首或用二氧化碳（P60）对阳性感染的小鼠进行安乐死，如有必要 ，剃须并用脱毛霜处理 。 　　如先前所述8
，对P4背皮进行处理，并进行了较小的修改
。手术切除并刮除脂肪后 ，将后皮肤洗在冷PBS中一次洗涤，然后在37°C下放入35分钟的配置中（康宁354235）
，真皮侧面朝下，在轨道振荡器上。接下来 ，将表皮与真皮分开
，用细镊子撕成较小的碎片，并放入4 ml 0.25％胰蛋白酶-EDTA（1× ，Gibco; 25200056）
，在37°C的情况下15分钟，并带有轨道摇动 。用冷PBS洗涤表皮 ，并剧烈移动以实现单细胞悬浮液
。然后
，将悬浮液通过70 µm和40 µm的过滤器（Corning; 431750，431751）连续地进行过滤 ，然后在400G下以400G离心10分钟
，然后重悬于FACS缓冲液中（PBS + 2％Chelexed FBS（FBS（FBS（FBS）） + DAPI）。 　　如前所述，对P60背皮进行处理 。62
。简而言之 ，用手术剪刀收集了背皮，并将面向真皮的针染色托盘放在带有销的泡沫聚苯乙烯托盘上。用手术刀刮掉脂肪和亚白水粉 。然后洗涤无脂肪的皮肤
，在1×PBS中面对真皮侧
。然后将皮肤放在0.5％胰蛋白酶-EDTA（10×，Gibco; 15400054）中的同一方向上 ，并在37°C的轨道振荡器上孵育25（雌性）或50分钟（男性）。使用玻璃显微镜载玻片
，将皮肤固定在盘子的底部，然后用手术刀再次刮擦直到开始破裂 。添加了2％FBS（ - ）的过量冷PBS中和胰蛋白酶。首先通过70 µm的滤网将细胞悬浮液在冰上过滤 ，并用另外15 ml的1×1×PBS用2％FBS（ - ）洗涤
，然后通过40 µm的滤网过滤
。细胞悬浮液以400克旋转10分钟，并重悬于FACS缓冲液中。 　　P4：在7个独立的单细胞RNA-seq运行中 ，总共收集了58只注射的P4小鼠并评估 。P60：在8个独立的单细胞RNA-seq运行中
，总共收集了10只注射小鼠并评估
。由于P4皮肤含有较少的细胞，因此必须将更多的小鼠汇总到足够数量的麦克利阳性细胞。 　　使用70 µM喷嘴在BD Facsaria III上对MCHERRY阳性细胞进行分选 ，并使用BD Rhapsody单细胞分析系统使用BD Rhapsody Cartridge套件（6333733）处理单细胞捕获 。对于P4
，将7个单独的墨盒加载为50,037个细胞
。对于P60，将8个平均64,625个细胞的墨盒加载。根据制造商的说明（BD Biosciences ，Doc ID：210967Rev 。1.0）进行了珠子和测序库生产的逆转录
。 　　在单独的嵌套PCR反应中，从BD狂想曲珠中扩增了含SGRNA的区域。In PCR1, we used 100 µl KAPA HiFi HotStart ReadyMix (Roche 07958935001), 6 µl forward primer (5′-ACACGACGCTCTTCCGATCT-3′, 10 μM), 6 µl reverse primer (5′-TCTTGTGGAAAGGACGA-3′, 10 μM), 12 μl Bead RT/PCR来自BD生物科学的增强剂试剂和72μl无核酸酶的水总体积为200μl 。将来自每个单独的墨盒制剂中的狂想珠重悬于PCR混合物中
，以4×50μl的反应等分，并迅速移入预热的PCR机中 ，而无需使珠子沉降
。使用以下条件：在95°C下进行初始变性5分钟
，然后在95°C的25个循环中进行30 s，将53°C退火30 s ，扩展72°C 20 s
，最后在72°C的最终延伸10分钟。 　　根据制造商的指示，在第二个PCR之前的指示下 ，根据制造商的指示
，将PCR1产品汇总，用磁性XP珠子（Beckman Coulter A63881）清理珠子 ，并用Agencourt Ampure XP珠（Beckman Coulter A63881）清理扩增子 。For PCR 2, 3 μl of cleaned up PCR1 was used as template along with 2 μl forward primer (5′-ACACGACGCTCTTCCGATCT-3′, 10 μM), 2 μl reverse primer (CAGACGTGTGCTCTTCCGATCTCTTGTGGAAAGGACGAAACA*C*C*G-3′, 10 μM), 18 μl无核酶的水
，以及25μl的kapa hifi hotstart readymix，总体积为50μl。在以下条件下进行PCR2：在95°C下进行初始变性3分钟 ，然后在10个变性95°C的10个循环中进行30 s，将60°C退火3分钟
，延伸72°C 60 s，最后在72°C下延伸5分钟
。在索引PCR之前 ，根据制造商的说明
，用Agencourt Ampure XP珠子清理PCR2产品。根据BD Biosciences“ MRNA针对库制备”协议（DOC ID：210968 REV 3.0）进行索引PCR 。 　　为了计算SGRNA的富集和耗竭，我们将SGRNA细胞计数用于P4和P60
。对于T0 ，使用了来自单独的批次质粒制剂的测序预注射库A
，B和C的读数。由于每个PX（P60或P4）样品收到了三个T0库批次之一，因此我们按以下方式分别对每个样本进行了归一化 。首先将PX和T0的计数转换为比总数的比例 ，无论是整个样本还是库批次
。然后
，通过将每个PX样品指南比例分配给其T0比例（即其在匹配的注射库中的初始表示）。这会产生折叠变化（比率） 。然后将靶向同一基因的指南的三胞胎变化在一起（并用靶向基因的象征指示），50个对照指南（总共分组17个伪 - 三个伪长的对照：Ctrl_1的CTRL_1用于控制SGRNA的CTRL_1 ，用于1至3
，CTRL_2，CTRL_2 ，CTRL_2，用于控制SGRNA
，来自4到6NN和CRTR，以及50 sna和CRLL的控制 ，以及50 snN和CRLL的控制
。这导致每个PX样本中的167倍变化值：目标基因为150
，伪三局对照为17。根据这些折叠量的减小顺序，为每个PX样本定义了等级 ，因此首先排名最高 。将纽带（通常为零值零）分配给相同的最低等级（请参见R package dplyr的min_rank函数）
。这些等级用于计算八个P60样品（生物学重复）中前20个和底部20个扰动（指导三重扰）的相关性
，如扩展数据所示。 　　最后，在同一时间点样品（P4或P60）上也将获得的倍数变化平均 。对于这个平均值 ，使用了加权均值
，重量是每个样品的引导阳性细胞的总数。这些结果以图2a和扩展数据图中的log2量表报告
。1i和3。对于P60与P4指南表示形式（扩展数据图3C），将扩展数据中显示的倍数变化图3a除以扩展数据中的数据图3B（通过匹配的扰动身份）并重新排列 。将表皮干细胞特异性的SGRNA表示（如扩展数据图3D所示）以相同的方法计算 ，在将P4或P60数据集取决于仅包括表皮干细胞（群集0
，图1E，H）
。 　　如前所述进行了两级皮肤化学癌变。63 。简而言之 ，每100 µL（溶解在丙酮中），用电剪剪子的6至8周雌性小鼠的背皮被剃光
，并用400 nmol DMBA（Sigma-Aldrich D3254）处理一次
。经过2周的休息时间，每周两次溶于100 µL 100％乙醇中40 nmol TPA（Sigma-Aldrich 79346）。定期通过视觉检查监测肿瘤的形成 ，并在治疗12周后对小鼠安乐死 。苏黎世广州动物福利委员会允许的最大累积肿瘤大小为2 cm（动物允许ZH074/2019
，ZH196/2022）
。 　　切除了DMBA/TPA处理的小鼠的肿瘤和1-2 mm的周围组织，并用手术刀细细地切碎。将肿瘤片浸入5 mL预热的DMEM中 ，在37°C下
，在37°C的轨道振荡器上以0.25％的胰蛋白酶/EDTA和3 u ml-1 DNase I浸没，每5分钟严格重悬于每5分钟 ，总计30分钟 。将反应用1 mL FBS淬灭
，并通过70 µm的网格拧紧，在400g的400g固定中 ，用过量的PBS沉淀10分钟
，然后在排序之前重悬于0.5 mL FACS缓冲液中
。如上所述，使用Rhapsody增强的墨盒试剂盒（BD Biosciences 664887）进行单细胞捕获。根据制造商的说明（BD Biosciences ，Doc ID：210967Rev 。1.0）进行了珠子和测序库生产的逆转录。 　　DMBA/TPA诱导的肿瘤未用于SCRNA-SEQ或空间分析
。使用Qiagen Dneasy血液和组织试剂盒提取基因组DNA。在嵌套的PCR中将SGRNA区域扩增 。In PCR1, we used 12.5 µl KAPA HiFi HotStart ReadyMix (Roche 07958935001), 0.75 µl forward primer (5′-CTTGTGGAAAGGACGAAACACCG-3′, 10 μM), 0.75 µl reverse primer (5′-GTGTCTCAAGATCTAGTTACGCCAAGC-3′,10μM），1 µL提取的基因组肿瘤DNA（浓度在50–100ngμl-1之间）和10μl无核酸酶的水
，总体积为每个肿瘤25μl
。使用以下条件进行PCR1：在98°C下初始变性2:30分钟，然后是22个变性循环98°C 20 s ，将62°C退火30 s
，延伸72°C 30 s。用Genelute PCR清洁试剂盒（Sigma-Aldrich NA1020-1KT）单独将每种肿瘤的PCR1产物分别纯化 。对于PCR2，将1μl的清除PCR1与0.75μl的前进条形码i5底漆一起用作模板（请参阅补充表1 ，10μm）
，0.75μl反向条形码i7引物（请参阅补充表1，10μm） ，10μl-fore-foree for-Fore-Free Water vole for 12.5.5 kap a的kapa kapa hifi h
。25μl。八个不同的i5和i7引物允许总共64种不同的组合 。在以下条件下进行PCR2：在95°C下初始变性3分钟
，然后在32个变性98°C的32个循环中进行30 s，将62°C退火30 s ，延伸72°C 30 s
。根据制造商的说明
，根据制造商的说明，将最多64个单独的条形肿瘤扩增的PCR2产品与Agencourt Ampure XP珠（Beckman Coulter A63881）一起清理。 　　除全转录组扩增分析7（发送给Novogene并在Illumina Novaseq的S4流动池上进行测序）外 ，所有单细胞或扩增子测序均作为现成的库制备，并在功能基因组中心的Illumina Novaseq或NextSeq 500 Instruments上进行了测序 。使用DNA高灵敏度芯片检查了测序文库的峰值大小和浓度
。使用量子荧光计进一步测量浓度。来自狂想曲弹药筒的全转录组扩增分析以1.8 nm的浓度为1.8 nm的全sp流细胞
，使用20％的phix和以下读取构型：读取1 60，i1 8 ，读取2 62个周期 。用10％PHIX
，双重指数i5，8 bp对肿瘤进行测序；i7 ，8 bp;并阅读1
，64个周期。 　　将MCHERRY阳性或阴性角质形成细胞直接在BD FACSARIA III上进行FAC，直接进入Trizol ls ，并用直接-Zol RNA Miniprep Kit（Zymo Research）进行柱状纯化
。用TRUSSEQ mRNA库试剂盒制备每个样品中1,000 ng RNA的测序文库
，并在Zurich功能基因组中心的Novaseq x 2×150 bp上测序。使用FASTQC程序对配对末端的RNA-seq数据进行了质量控制 。Casadapt进行了Illumina通用适配器（Agatcggaagag）的适配器修剪（Agatcggaagag）
。Salmon64使用Gencode小鼠M25参考转录组进行了转录水平定量。DESEQ2用于识别不同条件之间的差异表达基因65 。计数数据的初始质量控制是通过PCA和分层聚类进行的，使用方差稳定归一化后 ，使用前500个可变基因进行
。通过比率方法中位数将基因水平计数数据标准化
，并估计分散。接下来，将拟合一个广义线性模型来识别差异表达的基因 。P值是通过WALD检验获得的 ，并使用Benjamini -Hochberg程序校正了FDR
。FDR截止 <0.05 was used to detect significant differentially expressed genes for subsequent analysis. 　　Back skin from P4 and P60 mice was scraped to remove fat, placed on Whatman paper and cut into strips. Sections were embedded in OCT (Tissue-Tek; 4583), frozen by placing them in a liquid nitrogen-cooled isopentane filled metal beaker, stored at −80 °C and sectioned at 10–12 μm thickness on a Leica CM1900. Sections were immobilized on Superfrost glass slides, fixed for 10 min at room temperature with 4% paraformaldehyde (PFA), washed twice in 1× PBS before blocking in blocking solution (1% BSA, 1% gelatin, 2.5% normal goat serum, 2.5% normal donkey serum, 0.30% Triton X-100, 1× PBS). Primary antibodies (TNF [D2D4] XP[R] Rabbit anti-mouse monoclonal antibody, CST 11948; 1:300) and purified rat anti-mouse CD104 (also known as β4-integrin) (BD Pharmingen 553745, 1:300) were incubated overnight at 4 °C. After washing twice with 1× PBS, slides were incubated with secondary antibodies (Alexafluor488, Cy3, or AlexaFluor647, Jackson ImmunoResearch Laboratory; 1:500–1:1,000) and 0.5 µg ml−1 4′,6-diamidino-2-phenylindole (DAPI) at room temperature for 1 h. Sections were then washed again with 1× PBS, dried, covered with home-made mounting medium and sealed with nail polish before image acquisition. Pictures were acquired with a 20× objective on a ZEISS Axio Observer microscope controlled by the ZEN microscopy software (version 3.1). 　　Dissected back skin from P4 and P60 mice injected with CROP-mCherry library were fixed in 4% PFA for 1 h at room temperature. Following fixation, samples were permeabilized in 0.8% PBS-Triton overnight. All steps during staining were carried out at room temperature. Primary antibodies were diluted into blocking buffer (5% donkey serum, 2.5% fish gelatin, 1% BSA, 0.8% Triton in PBS) and were incubated for at least 16–20 h at room temperature. Samples were then washed for 3–4 h in 0.8% PBS-Triton, and incubated with secondary antibodies (in blocking buffer) together with DAPI (to label nuclei, 0.25 mg ml−1) for at least 16–20 h. After staining, samples were extensively washed with 0.8% PBS-Triton every hour for 4–6 h. Samples were then dehydrated in increasing concentrations of Ethanol: 30%, 50% and 70% in doubled-distilled water (with the pH adjusted to 9.0 with NaOH/HCL) for 1 h each, and finally in 100% ethanol for 1 h. For tissue clearing, samples were transferred to Eppendorf tubes with 500 μl ethyl cinnamate (ECi, Sigma 112372) and shaken overnight at room temperature under dark conditions. Fresh ECi was used for mounting for imaging. 　　For fixed sections, back skin was cut into strips, embedded and frozen in OCT (Leica), and sectioned with a Leica cryostat (10-μm sections). Sections were permeabilized in 0.3% PBS-Triton for 15 min at room temperature, blocked in blocking buffer (5% donkey serum, 2.5% fish gelatin, 1% BSA, 0.3% Triton in PBS) for 15 min at room temperature. Primary antibodies in blocking buffer were incubated on slides overnight at 4 °C, washed off with 0.3% PBS-Trition before adding secondary antibodies, with DAPI for 1 h at room temperature. After washing off the secondary, samples were mounted for imaging using ProLong Diamond Antifade Mountant (P36961, Invitrogen). 　　Antibodies used were as follows: rat anti-RFP (Chromotek, 5F8; 1:200), rabbit anti-RFP (MBL, PM005; 1:200), chicken anti-GFP (Abcam, ab13970; 1:200), rat anti-CD45-biotin (Biolegend, 103104; 1:200), goat anti-TNFR1 (R&D, AF-425-PB; 1:200). All secondary antibodies used were raised in a donkey host and were conjugated to Alexa Fluor 488, Cy3 or Alexa Fluor 647 (Jackson ImmunoResearch Laboratory; 1:500- 1:1000). 4′,6-diamidino-2-phenylindole (DAPI) was used to label nuclei (0.25 mg ml−1). 　　Whole-mount images were acquired using an LSM980/Airyscan module combined with Tiling (where indicated) 40× multi-immersion objective (LD LCI Plan-Apochromat 40×/1.2 Imm Corr DIC (silicone oil, glycerol or water immersion), WD 0.41 mm) with glycerine as an immersion medium for cleared samples, or a 40× oil immersion objective (Plan-Apochromat 40×/1.4 Oil DIC, WD 0.13 mm) for sections. 　　All processing was done in Zen Blue, with a custom batch processing macro (Thomas Peterbauer). All images depicted are maximum intensity projections. 　　One hour before they were euthanized, mice were injected intraperitoneally (50 µg per g body weight) with 5-ethynyl-2′-deoxyuridine (EdU, Targetmol, TMO-T17341-500mg) dissolved in PBS at a concentration of 10 mg ml−1. After cryo-sectioning, click-chemistry was performed for staining with the In Vivo EdU Click Kit 488 (Merck, BCK488-IV-IM-S) according to manufacturer’s instructions. In brief, 10–12 µm thin frozen tissue sections were fixed with 4% PFA in PBS for 15 min at room temperature. Cells were washed twice with 1× PBS before permeabilization with 0.3% Triton X-100 in PBS for 15 min at room temperature. Five hundred microlitres of EdU reaction cocktail was added to each section and incubated for 30 min at room temperature, protected from light. Afterwards, sections were washed in PBS and antibody staining was carried out as described above. 　　Newborn, primary mouse epidermal keratinocytes derived from B6-LSL-Cas9-eGFP mice were isolated as previously described66. In brief, isolated epidermal keratinocytes were cultured on 3T3-S2 feeder layer previously treated with Mitomycin-C in 0.05 mM Ca2+ E-media supplemented with 15% serum. After 3 passages on 3T3-S2 feeder layer, cells were cultured in 0.05 mM Ca2+ E-media, made in house as previously described14. The B6-LSL-Cas9-eGFP keratinocytes were transiently transfected with 1 µg of a Cre-expressing plasmid in 6 wells using Lipofectamine 2000 (Invitrogen 11668-027) to activate Cas9–GFP. Cells were then sorted for GFP-positive signal on a BD Aria III and put back in culture using 0.05 mM Ca2+ E-media. For this project, all further modified cell lines derived from the Cas9–GFP activated cell line and were grown at standard conditions, 37 °C and 5% CO2. The SCC cell line was previously isolated from DMBA/TPA chemically induced skin tumours and brought into culture. Cell lines were tested for mycoplasma using the Mycoplasma PCR detection kit (Sigma; D9307). 　　For lentiviral infections in culture, B6.Cas9 keratinocytes (see above) were plated in 6-well plate (Thermo Scientific Nunclon TM Delta Surface 140675) at 1.5 × 105 cells per well and infected with 100–300 µl of low-titre virus in the presence of infection mix (1 in 10 dilution of polybrene (10 mg ml−1 Sigma 107689-100MG in PBS) in FBS[-]), by centrifuging plates at 1,100g for 30 min at 37 °C in a Thermo Heraeus Megafuge 40 R centrifuge. Infected cells were sorted by FACS for mCherry on a BD FACSAria III. Viruses used for generation of knockout keratinocytes carried the following sgRNAs: Trp53_sgRNA.3, Fat1_sgRNA.3, Notch1_sgRNA.2, ctrl_44. Additionally, we used constructs carrying sgRNAs and Tnf cDNA: Trp53_sgRNA.3 + Tnf, Notch1_sgRNA.2 + Tnf, ctrl44-Tnf. 　　Mouse TNF uncoated ELISA Kit from Invitrogen (88-7324-22) was used according to the manufacturer’s instructions to quantify TNF concentration in supernatants from confluent 6 wells of keratinocyte cultures after 24 h of incubation. Measurements of 96-well plates in technical and biological triplicates from separate wells were taken on a Tecan Infinite M1000Pro plate reader. Concentrations in supernatant were calculated according to included recombinant TNF standard curve. 　　In vitro-infected keratinocytes and SCC cells were lysed in 6 wells with 1 ml TRIzol Reagent and the RNA was purified using the Direct-zol RNA Miniprep kit (Zymo Research). The procedure was performed according to the manufacturer protocol except for an additional 1 min centrifugation after the last washing step to completely remove residual ethanol. cDNA was synthesized using the Promega GoScript Reverse Transcription Mix, Oligo Protocol. In this procedure, 500–1,000 ng of RNA were converted into oligo(dT)-primed first-strand cDNA. iTaq Universal SYBR Green Supermix was used according to the manufacturer’s protocol for RT–qPCR reaction in a Quant Studio 7 Flex (Applied Biosystem by Life Technologies). RT–qPCR primers can be found in Supplementary Table 9. The delta-Ct method in Quant Studio Real-Time PCR software (v1.3) was used for analysis and to calculate fold changes based on Ct values. 　　Keratinocytes were cultured as described above, washed with PBS and lysed in protein sample buffer (100 mM Tris-HCl pH 6.8, 4% SDS, 20% glycerol, and 0.2 M DTT). The lysate was heated for 5–10 min to 95 °C and vortexed briefly to shear genomic DNA. Isolated proteins were stored at −20 °C for later use or loaded directly (10–20 µg). Proteins were separated by 4–12% Bis-Tris SDS–PAGE electrophoresis (at 180 V, 300 W, 55 min) and transferred onto a nitrocellulose membrane (GE Healthcare). Membranes were blocked in 5% BSA (Sigma A3059-100G) in TBS-Tween (0.1%) for 1 h at room temperature, incubated with primary antibodies in 3% BSA at 4 °C overnight and washed with TBS-Tween. The secondary antibodies were added at room temperature in 3% BSA for 2 h. Western blots were developed with freshly mixed ECL solutions (GE Healthcare). The following antibodies were used: anti-TNF (Cell Signaling Technology 11948), anti-vinculin (Abcam, ab129002). Secondary Goat Anti-Rabbit IgG HRP Linked Antibody (1:1,000 dilution, Cell Signaling Technology 7074S). 　　Cells were seeded in an ibidi Culture-Insert 2-well (81176) according to the manufacturer’s recommendations. This insert creates a clean cell-free gap in a 6- or 12-well plate (Thermo Scientific, 150239). The assay was performed according to the manufacturer’s protocol (ibidi). In brief, once confluency was reached, the insert was removed, and detached cells were washed away with PBS. New medium (with or without 10 ng ml−1 recombinant TNF (R&D Systems 410-MT) was added, and the wells were imaged every hour for 22 h in an incubation chamber at 37 °C and 5% CO2. The images were taken with a Zeiss Axio Observer controlled via the ZEN software. The data analysis was performed with an Image J plugin67 (version 1.54 h) and default settings. 　　Full-length mouse Tnf was amplified by PCR from GFP-TNF-alpha plasmid (Addgene #28089) via Phusion PCR according to the manufacturer’s protocol with the following primer: forward: 5′-agcctgctgaagcaggccggcgacgtggaggagaaccccggccccatgagcacagaaagcatgatccgcgacgt-3′, reverse: 5′-ATATaacgcgtCTAcagagcaatgactccaaag’. Primers created a 5′-P2A-site overhang with the opposing site added to the 3′ end of mCherry. mCherry-P2A-TNF was then ligated into the CROP-seq vector via Pfl23 and MluI restriction sites. Expression was tested with quantitative PCR, western blot and immunofluorescence (see Fig. 5). 　　Full-length cDNA for Jun, Ccn1 and Fos was Phusion PCR amplified from Addgene plasmids #47443, #12519 and #193089 with the following primers: Fos fw (5′-ATGGATCCGGGgccaccaacttcagcctgctgaagcaggccggcgacgtggaggagaaccccggccccATGATGTTCTCGGGTT-3′), Fos rv (5′-taacgcgtTCACAGGGCCAGCAGCGTG-3′); Jun fw (GGATCCGGGgccaccaacttcagcctgctgaagcaggccggcgacgtggaggagaaccccggccccatgactgcaaagatggaaac-3′), Jun rv (5′-ataacgcgttcaaaacgtttgcaact-3′); Ccn1 fw (5′-ataGGATCCGGGgccaccaacttcagcctgctgaagcaggccggcgacgtggaggagaaccccggccccATGAGCTCCAGCACCTTC -3′), Ccn1 rv (5′-taaacgcgtTTAGTCCCTGAACTTGTGGATGTC-3′). 　　PCR amplicons were ligated via BamHI and MluI restriction site overhangs into the same EF1-mCherry-2A expression vector as TNF. 　　Proliferation rates for different treatments and overexpression constructs were tested using the MTT Cell Growth Assay Kit (Merck CT01). In short, 10,000 keratinocytes were seeded per 96-well of a flat-bottom tissue culture plate and incubated in 100 μl regular E-low medium for 24 h at 37 °C, 5% CO2 before adding 10 μl of freshly prepared MTT solution for four hours. The resulting formazan crystals were dissolved by adding 100 μl isopropanol with 0.04 N HCl to each well with vigorous pipetting. Absorbance at 570 nm with a reference wavelength of 630 nm was measured directly on a Tecan Infinite M1000Pro plate reader. 　　Eleven sgRNAs representing top, middle and bottom enrichment cohorts were selected and sgRNA efficiency was assessed by infecting B6.Cas9 keratinocytes. Lentivirus was derived previously with low-titre lentiviral preparations of the CROP-Seq-Puro vector and single guides. Cells were harvested after 10 days of puromycin selection (1 µg ml−1, Gibco A11138-03). Genomic DNA was extracted using the DNeasy Blood and Tissue Kit (QIAGEN 69504), followed by Taq (NEB M0273) PCR following the manufacturer’s instructions. Amplicons were 800–1,000-bp-long and amplified by primers, designed with primer 3 listed in Supplementary Table 1. T7 endonuclease assay was performed as described in the ALT-R genomic editing detection kit (IDT). Samples were quantified on a Bioanalyzer 2100 with a DNA high-sensitivity chip. Primers for amplification of genomic target regions of guides can be found in Supplementary Table 1. 　　The workflow of Mirzazadeh et al.68 for RNA-rescue spatial transcriptomics was followed to overcome low RNA integrity number score. A microarray of fresh-frozen DMBA/TPA-induced squamous cell carcinoma samples was cryo-sectioned with 10 µm thickness (Leica, CM3050S) and placed on the capture area of a spatial gene-expression slide (10X Genomics, 1000188). Samples were stored in −80 °C before processing. The samples were fixed with 4% PFA and H&E staining was performed. The spatial libraries were then generated from the probe hybridization step (10X Genomics, 1000365) according to Visium Spatial Gene Expression Reagent Kits for FFPE (10X Genomics, User Guide CG000407 rev D, 1000361). The resulting libraries were sequenced by the Genomics Facility Basel. The sequencing was performed in a paired-end manner with dual indexing (10X Genomics, 1000251) on a NovaSeq 6000 (Illumina) with a S4 or SP (PE 51) flow cell. S4 libraries were then sequenced with the following cycle settings: Read 1 101 cycles, i7 10 cycles, i5 10 cycles, Read 2 101 cycles. 　　Sequenced libraries were processed using Space Ranger software (version 2.0.1, 10X Genomics). Reads were aligned with the ‘spaceranger count’ pipeline to the pre-built mouse reference genome provided by 10X Genomics version 2020-A (comprising the STAR-indexed mm10 mouse genome and the GTF gene annotation from GENCODE version M23 (Ensembl 98)), and the probe set reference CSV file (Visium Mouse Transcriptome Probe Set v1.0) was provided for filtering of valid genes. The resulting filtered count matrix of features (that is, genes) per spot barcodes were employed for the downstream analysis. Proportions of cell-type mixtures in each Visium spot were deconvoluted with the robust cell-type decomposition (RCTD)69 method from the spacexr package (version 2.2.1; https://github.com/dmcable/spacexr), leveraging as single-cell reference the union of mCherry-positive (Fig. 4c) and mCherry-negative (Extended Data Fig. 6c) tumour cells (merged dataset in Extended Data Fig. 7e). Seurat R package (version 4.9.9.9041) was used to plot these cell-type deconvolution results. Spatial co-occurrence of cell types was analysed with the ISCHIA package40 (version 1.0.0.0), and resulting probabilities for positive (Pgt) or negative (Plt) co-occurrence were plotted in −log10 scale. 　　The raw sequencing data comprising the BCL files were demultiplexed using the Illumina bcl2fastq program v2.20 with default options, allowing one mismatch of the sample barcode sequences. The resulting fastq files were processed by the BD Rhapsody Pipeline version 1.9.1 hosted at the Seven Bridge’s cloud platform (https://www.sevenbridges.com/). A custom mouse reference genome based on the Gencode version 25 was utilized and combined with our 500 sgRNA sequences. Since the BD pipeline uses STAR aligner70 in the backend, the custom STAR reference genome was generated by the genomeGenerate command in STAR. The Refined Putative Cell Calling option was disabled while running the pipeline. The output unique molecular identifier (UMI) counts data corrected by the BD Genomics RSEC (Recursive Substitution Error Correction) method were used for further downstream analysis. 　　The PCR dial-out data was processed by a custom in-house Python script. In brief, the script extracted three 9-nucleotide (nt) long cell barcodes out of possible 97 barcodes at specific locations from the Read 1, allowing one mismatch per barcode (Hamming distance of 1). It also obtained 8-nt UMI sequences from the Read 1, as provided in the BD manual. For valid barcodes, it counted the presence of sgRNAs in Read 2 by an exact match of the 20-nt sgRNA sequence. Next, the UMI counts per cell were deduplicated and sgRNA UMI counts per cell were obtained. If multiple sgRNAs were detected in one cell, the cell was only assigned to a specific sgRNA if the sgRNA UMI count was higher than the quantile 0.99. 　　To remove the doublets, the Scrublet71 Python package was utilized. Doublets were removed from each dataset separately with initial filtering of genes expressed in minimum 5 cells and expected doublet rate of 20% as detected by the BD Rhapsody. Next, 7 and 8 scRNA-seq datasets for P4 and P60, respectively, were merged in Seurat 472. Cells with detected sgRNAs and that were not doublets were selected for further analysis. Cells with UMI counts >500，UMI计数低于总UMI计数（去除异常值）的分数0.99
，并且细胞<20％的线粒体基因被过滤。之后，通过标准的Seurat管道处理数据 ，该管道将数据归一化的数据归一化
，该数据将计数扩展到每个单元格，然后使用Log1p进行自然对数变换 。通过FindVariableGenes函数检测到两千个可变基因
。通过鳞片函数缩放归一化数据 ，然后使用RUNPCA函数进行PCA分析。每个主要成分用肘部功能解释数据方差的贡献
，并为每个P4，P60和肿瘤的每个数据集选择了最高主要成分 。选定的主要组件用于查找具有Findneighbors功能的最近邻居。Louvain算法与FindClusters一起使用模块化优化的数据聚类
。UMAP尺寸还原技术用于在两个维度上可视化数据。在UMAP预测中未观察到P4和P60数据集中的批处理效应 。因此未进行批处理效应校正
。通过Findallmarkers函数获得每个簇的标记基因 ，并根据已知标记基因的表达来用细胞类型注释簇。使用MAST Package73和Wilcoxon Rank-sum检验进行了每种扰动的SCRNA-SEQ数据的差分表达分析
，从而得出了可比的结果 。基因表达的核密度估计值是通过Nebulosa包推断的
。 　　为了获得每种扰动的转录表型的概述，计算了基础细胞和所有p60细胞中每个扰动的平均细胞基因表达。接下来 ，选择了这些扰动中的前2,000个最可变的基因
，并使用带有Ward.d2方法进行聚类的Pheatmap软件包计算了相关矩阵的热图 。此外，我们从Pymde软件包（pymde.org）利用了Preserve_neighbors函数 ，以在两个维度上嵌入数据的最小渗透率
。 　　测序一百六十8个单独的肿瘤以量化每个肿瘤中的SGRNA表示。用相应的i5和i7索引条形码将原始测序数据用BCL2FASTQ函数反复插图 。示例中的SGRNA数量是通过Mageck74软件包的Count Command计数的
。肿瘤中所得的SGRNA表示用于肿瘤中的总代表和选择评分。 　　CD4+或CD8+ T细胞通过腹膜内注射单克隆大鼠抗小鼠CD4（YTS 191.1，Hoelzel Diagnostika Lein-C3210）和大鼠抗小鼠CD8（YTS169.4 Hoelzel Diagnostika Leinostika Lein-C2850）中性分化10 mgibibibibodies artibibodies artibibodies artibibodies 。从P4到P60每两周注射抗体。通过腹膜内注射大鼠抗小鼠CD115（CSF-1R
，Hoelzel Diagnostika Lein-C2268）的巨噬细胞以0.5 mg的剂量为0.5 mg，然后每周两次每周两次 ，从P4到P60
，以25 mg的每公斤体重进行了25 mg
。对照小鼠接受了大鼠IgG同种型对照（Hoelzel Diagnostika，Lein-I-1177）。我们使用6只小鼠用于T细胞耗竭 ，7只小鼠用于巨噬细胞耗竭和7个对照IgG小鼠 。在P60时
，如上所述制备表皮，并分成两个部分（前后皮肤和后背部皮肤）进行单独的扩增子测序
。基因组DNA分离后 ，使用与肿瘤的SGRNA扩增方案相同的方案进行了SGRNA盒的PCR扩增。在具有10％phix的Novaseq 6000 SP流动池上测序扩增子
，双索引i5，8 bp；i7 ，8 bp;R1进行64个周期 。Mageck计数FASTQ文件中的SGRNA数量
。DESEQ2用于基因级别的SGRNA的差异分析。使用50个对照SGRNA使用比率方法中值将样品的计数数据归一化
，并估计样品之间的分散体 。为每个指南拟合了负二项式通用线性模型，以检测差异富集或耗尽的指南
。通过Benjamini -Hochberg多重测试程序校正了通过WALD检验获得的P值。 　　对于WGCNA24分析 ，使用子集= 500选项的子集函数将数据集降低至每个扰动最多500个单元 。接下来，将数据归一化
，并使用标准化和FindVariable Features函数选择了2,000个大多数可变基因
。功率计算是通过从WGCNA软件包1到30的拾音器函数进行的。接下来，我们使用签名网络进行WGCNA分析 ，使用签名网络进行wgcna分析
，最小模块大小为10，并使用稳健函数bicor bicor bicor bicor bicor bicor bicor bicor for for the power vrop for the power velimate frop the power vaste 。除非另有规定 ，否则不包括具有对照SGRNA的细胞进行WGCNA分析。分别为每个群集进行WGCNA分析
。为了计算扰动得分
，运行模量函数以获得给定模块的模块特征元（第一主要组件）。接下来，使用R中的比例函数对第一个主成分的值进行了归一化 。接下来 ，在R中执行线性建模
，以获得每个扰动对使用公式的扰动得分：归一化基因得分〜扰动
。从线性模型中提取P值，标准误差和效应大小以进行绘图。这种方法类似于Jin等人25中描述的方法 。通过METAP R软件包将差异表达的模块基因的P值组合在一起 ，并结合了对数折叠的平均变化
，以计算平均扰动效果
。 　　为了在每个群集之间解密细胞到细胞信号，使用了具有截短的平均接近截面的CellChat33封装 ，最少有十个单元进行推断信号传导。 　　使用seurat IntegratedAta函数进行了p60群集0和肿瘤麦克伦阳性簇0和7的整合 。首先，将每个数据集归一化
，并获得2,000个可变基因
。下一个SelectIntegrationFeatures用于选择跨数据集重复变量以进行集成的功能。接下来，使用公共功能对数据集进行缩放 ，并执行PCA 。之后
，使用30个主组件，然后使用IntegratedAta进行集成数据集 ，使用RPCA算法运行Seurat Find Integrationanchors函数
。将结果与Harmony75算法进行了比较
，该算法产生了相似的整合。 　　为了识别SCRNA-Seq数据中的元细胞态，采用了SeaCells方法 ，该方法确定了一组定义为元素的细胞以发现细胞态 。首先
，将p60 EPSC数据集降采样至每个SGRNA的最多200个单元，以获得每个扰动数量的元素。选择约50个细胞来定义一个元细胞（SEACELL） ，并使用PCA构建核
。SEACELLS函数以选项N_Waypoint_eigs = no seacell +1和contgence_epsilon = 1E-5的选项运行
，每种扰动分别为1e-5，每次扰动1-4个metacells。SEACELLS模型最多拟合100次迭代 。此外 ，通过Seurat中的平均表达函数计算了每个元细胞的平均归一化表达，以绘制基因表达热图
。 　　使用内部Python 3.9和R 4.1脚本进行了分析。数据争吵是使用Python中的Pandas库进行的
，并在R.使用ComplexHeatMap和PheatMap软件包中创建了热图 。使用R的GGPLOT2库和Python的Seaborn Library制造了其他图
。 　　富集富集的过度占代表分析是通过Fisher的精确测试以及使用MSIGDB Hallmark 2020和其他Gene Sets的自定义R脚本进行的，它是基于Fisher的精确测试的Errichr31（https://maayanlab.cloud/enrichr）进行的。将FDR <0.05的切断设置为选择差异表达的基因 。使用GSEA31预先排名的方法使用GSEAPY76（https://github.com/zqfang/gseapy）进行了途径富集分析 ，从而可以同时分析上调基因和下调基因
。在FDR截止值为0.05的情况下
，进行了10,000个排列，以确定富集基因集。 　　TCGA数据是从UCSC Xena Soin77（http://xena.ucsc.edu）获得的 。在利用生存和生存库中 ，对HNSCC和其他TCGA肿瘤类型进行了生存分析
。COX比例危害回归模型用于拟合基因表达数据以生存以获得危险比。对基因表达的下层和上层和上部较低和底部1/3 mRNA的样品进行了Kaplan – Meier分析（在百万（TPM）中（TPM）中
，P值是通过对数秩检验计算的 。为了获得生存与TSK签名基因的关联，计算了MMP9 ，MMP10
，PTHLH，FEZ1 ，IL24
，KCNMA1，INHBA ，INHBA，MAGEA4
，NT5E，LAMC2和SLITRK6的平均表达水平。Kaplan – Meier分析是用这些基因平均表达的较低和上部三位数进行的
。 　　全部染色图中所示。1C和3E是来自三个小鼠的三个视野的代表性图像 。如图4所示 ，免疫荧光是来自四个不同肿瘤的六个视野的代表性图像
。图5J – N是来自两只小鼠的五个视野的代表性图像。 　　有关研究设计的更多信息可在与本文有关的自然投资组合报告摘要中获得 。