ST-FFPE-mIF: integrating spatial transcriptomics and multiplex immunofluorescence in formalin-fixed paraffin-embedded tissues using Stereo-seq Authors: Xue Zhang, Meng Zhang, Yuan Xu, Yangyang Song, Xingfeng Yang, Yanan Wu, Xuelin Zhao, Dongsheng Ran, Xin Liu, Huaqiang Huang, Wenxiao Lei, Hongyan Li, Yongfen Zhang, Feng Xi, Guibo Li, Xing Liu, Luohao Xu, Ao Chen, Sha Liao, Jiajun Zhang Published: 2025 ABSTRACT While Stereo-seq-FFPE captures spatial gene expression in formalin-fixed paraffin-embedded (FFPE) samples, it is limited to transcriptomic data. Here, we introduce ST-FFPE-mIF, a protocol that integrates eight-plex immunofluorescence staining with spatial transcriptomics in the same FFPE tissue section. Our approach extends Stereo-seq-FFPE to phenotypic analysis by enabling single-cell resolution through nuclear or membrane-based cell segmentation. We demonstrate how ST-FFPE-mIF supports robust spatial multi-omic profiling in cancer samples, offering a comprehensive view of molecular hierarchies from transcriptome to phenome.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13059-025-03900-3. Background Spatial transcriptomics (ST) technology is a cutting-edge biological technology enabling transcriptomic profiling within a spatial context to revolutionize biological research [1–3]. The commercial ST technologies can be classified into imaging-based such as MERFISH, seqFISH, osmFISH and STARmap [4], and sequencing-based methods including 10X Visium, Slide-seq V2, and Stereo-seq [2, 5, 6]. Despite the high resolution, imaging-based ST methods present several notable disadvantages: low throughput, complexity, high costs, time consumption, and comparability of transcripts [7]. Although addressing the challenges of low throughput [8], most sequencing-based ST methods are limited by resolution constraints. Emerging of advanced ST technology, like Stereo-seq, overcomes the mentioned limitation. Stereo-seq utilizes a DNA nanoball (DNB) patterned chip with a center-to-center distance of 500 nm, enabling the detection of spatially resolved and unbiased transcriptomic profiling at subcellular resolution [9]. A newly released version of Stereo-seq upgraded the probes with randomers (Stereo-seq-FFPE) [10], which made the technique compatible with the FFPE sample. ST technology allows for the visualization and analysis of gene expression patterns within the spatial context of tissue sections. However, unimodal (transcriptome) measurement also limits the application to link the mRNA information to functional activities. Given that protein markers serve as the gold standard for indicating biological functions and identifying specific cell types, it is necessary to develop a technique to add protein expression on top of ST data for the multi-modal analysis. The currently available spatial co-detection technologies exhibit several limitations, including a lack of sub-cellular resolution in Visium, DBiT-seq and SM-Omics [7], unable to achieve unbiased transcriptomic profiling in MERFISH [11], and the necessity for costly equipment in GeoMx DSP [12] and Visium HD [13, 14]. These challenges hinder the potential of these spatial techniques to adequately fulfill the requirements of clinical applications and translational research [15, 16]. The microscope-based multiplexing technique represents a pathology-related and relatively time-efficient method for assessing protein expression [17]. In light of the aforementioned, there is a clear and pressing need to develop a multi-omics co-detection technology with subcellular resolution capable of performing spatial whole-transcriptome capturing and multiplex immunofluorescence staining. This study presents an innovative solution termed spatial transcriptome and multiplex immunofluorescence for formalin-fixed paraffin-embedded samples (ST-FFPE-mIF). This methodology facilitates the simultaneous detection of eight-plex immunofluorescence staining together with whole transcriptome by Stereo-seq on one section of FFPE tissue. We provide two mIF staining protocols, named ST-FFPE-mIF-direct and ST-FFPE-mIF-indirect, respectively (Additional file 1 and Additional file 2). Given the shorter operation time (less than 4 h), more numbers of protein targets (up to eight targets) and wider application scenario, present article mainly discuss the performance of ST-FFPE-mIF-direct. Meanwhile, the user guide and fundamental performance of ST-FFPE-mIF-indirect were provided accordingly. Furthermore, the solution includes a comprehensive analytical workflow for image processing, data visualization, and multi-omic data integration. ST-FFPE-mIF also provides a recommended guideline for cell segmentation based on either nuclear or membrane staining. Additionally, we demonstrate the application of ST-FFPE-mIF in investigating cellular and molecular landscape in colorectal cancer (CRC) by integrated analysis of spatial transcriptomic data and mIF images at single-cell resolution. This innovative solution has the potential to enhance the understanding of physiological and pathological processes through the integration of multi-omic and architectural data. Results Workflow of ST-FFPE-mIF on formalin-fixed paraffin-embedded sample The Stereo-seq-FFPE technology is a powerful method that enables precise and accurate visualization of spatial mRNA information at single-cell resolution in FFPE tissue. We have enhanced the capabilities of Stereo-seq-FFPE by providing a protocol (Additional file 1) to enable the mIF staining before the standard ST protocol, referred to as ST-FFPE-mIF (Fig. 1, step 1). To overcome the challenges associated with significant crosslinking in FFPE samples, we implemented a refined step for antigen retrieval (Additional file 1) to facilitate the sufficient unmasking of antigenic epitopes while not affecting the performance of ST (Fig. 1, step 2). Given the limited antibody sources, fluorescence-labeled conjugated (FL-conjugated) primary antibodies were utilized for mIF staining to ensure compatibility with various different microscope systems [18].Fig. 1Schematic workflow and bioinformatics analysis of the ST-FFPE-mIF whole solution. (1) FFPE tissue section was mounted on the chip and performed de-paraffinization (2) The section on the chip was subjected to antigen repair, blocking, antibody incubation, and mIF imaging. Before the 2nd round of mIF staining, antibodies from the 1 st round should be removal, and the epitope would be re-exposed. Then, the above mIF protocol should be repeated to generate another mIF image. Nine-colored mIF images were obtained accordingly (3) Permeabilization, in situ cDNA synthesis and collection were conducted after De-crosslink (4) cDNA amplification, library construction, and sequencing were performed according to manufacturer’s protocol to obtain Stereo-seq sequencing result (5) mIF images and spatially resolved gene expression was subjected to downstream spatial visualization and bioinformatics analyses including image processing and spatial multi-omics analysis to address scientific questions Schematic workflow and bioinformatics analysis of the ST-FFPE-mIF whole solution. (1) FFPE tissue section was mounted on the chip and performed de-paraffinization (2) The section on the chip was subjected to antigen repair, blocking, antibody incubation, and mIF imaging. Before the 2nd round of mIF staining, antibodies from the 1 st round should be removal, and the epitope would be re-exposed. Then, the above mIF protocol should be repeated to generate another mIF image. Nine-colored mIF images were obtained accordingly (3) Permeabilization, in situ cDNA synthesis and collection were conducted after De-crosslink (4) cDNA amplification, library construction, and sequencing were performed according to manufacturer’s protocol to obtain Stereo-seq sequencing result (5) mIF images and spatially resolved gene expression was subjected to downstream spatial visualization and bioinformatics analyses including image processing and spatial multi-omics analysis to address scientific questions To achieve multi-round mIF staining, a protocol was developed for the elution of antibodies from the tissue section after 1 st round imaging. We optimized incubation time and elution duration to interfere with the non-covalent interaction between the antibody and its epitope. Additionally, we used dimethyl sulfoxide (DMSO) as the washing buffer after elution to remove residual antibodies and to break the residual non-covalent interaction [19]. Afterwards, the users could perform a 2nd round of mIF staining and imaging (Fig. 1, step 2). Multi-target co-staining, enabled by a multi-channel microscope, significantly simplifies the entire staining procedure. Meanwhile, using of fluorescence-conjugated antibodies could save time by eliminating the secondary antibody incubation step. After the 2nd round of mIF imaging, the protocol of Stereo-seq-FFPE should be followed, including permeabilization, probe hybridization, in situ reverse transcription (Fig. 1, step 3), library construction and sequencing (Fig. 1, step 4). Following rigorous data processing, a spatially resolved gene expression matrix is generated that can be aligned with mIF images. The aligned images are subjected to cell segmentation, yielding a single-cell resolved expression matrix that enables advanced spatial multi-omics analyses (Fig. 1, step 5). A detailed workflow of the ST-FFPE-mIF and a step-by-step protocol can be consulted in the supplementary section. Evaluation of spatially resolved transcriptomic profiles and mIF images from ST-FFPE-mIF in the human colorectal cancer FFPE sample To evaluate the performance of ST-FFPE-mlF, we applied it to FFPE sections of human CRC. The tested mIF panel was designed to identify cancerous regions (HER2, Villin, P53, CK19), T cells (CD3), macrophages (CD68), and fibroblasts (SMA, Vimentin). ST-FFPE-mIF detected an average of 798 genes and 1235 unique molecular identifiers (UMIs) per bin50 (25 μm × 25 μm). Both the mIF staining image and spatial distribution heat maps of detected gene types and UMI counts displayed a strong concordance with histomorphology (H&E Staining) (Fig. 2A). The detailed mIF images accurately reflected the histological structure, and the IF images for each marker exhibited unique spatial distributions and specificities (Fig. 2B).Fig. 2Evaluation of performance of ST-FFPE-mIF using human colorectal cancer section. A ST-FFPE-mIF profiling of a human colorectal cancer FFPE section. Left: H&E staining of an adjacent section. Middle left: Co-detection of mIF image. Middle right: Spatial gene expression count map at bin50. Right: Spatial UMI count map at bin50. B Representative images for mIF staining of eight-plex antibody panel in CRC tissue. Cells were colored by CK19 (green), CD3 (gray), CD68 (red), P53 (orange), Villin (blue green), SMA (turquoise), HER2 (yellow), Vimentin (violet), DAPI (blue). Scale bars = 100 μm. C Bar plots show the distribution of unique genes and UMI counts for Stereo-seq-FFPE (n = 3) and ST-FFPE-mIF (n = 3) data generated from consecutive tissue sections from the same CRC tissue specimen. ns : not significant (P > 0.05). D The Venn diagram shows the confluence and variance in gene capture between Stereo-seq-FFPE and ST-FFPE-mIF. E Scatter plot shows the gene expression significantly correlated between Stereo-seq-FFPE and ST-FFPE-mIF. Correlation coefficients were determined by the Pearson correlation coefficient. F Cell annotation is performed based on transcriptomic data at bin50 resolution. Representative figures show spatial visualization of cell identification and heatmap depicting the expression of classic markers of 10 annotated cell types. G Bar chart displays the percentage of each cell type within the analyzed sample Evaluation of performance of ST-FFPE-mIF using human colorectal cancer section. A ST-FFPE-mIF profiling of a human colorectal cancer FFPE section. Left: H&E staining of an adjacent section. Middle left: Co-detection of mIF image. Middle right: Spatial gene expression count map at bin50. Right: Spatial UMI count map at bin50. B Representative images for mIF staining of eight-plex antibody panel in CRC tissue. Cells were colored by CK19 (green), CD3 (gray), CD68 (red), P53 (orange), Villin (blue green), SMA (turquoise), HER2 (yellow), Vimentin (violet), DAPI (blue). Scale bars = 100 μm. C Bar plots show the distribution of unique genes and UMI counts for Stereo-seq-FFPE (n = 3) and ST-FFPE-mIF (n = 3) data generated from consecutive tissue sections from the same CRC tissue specimen. ns : not significant (P > 0.05). D The Venn diagram shows the confluence and variance in gene capture between Stereo-seq-FFPE and ST-FFPE-mIF. E Scatter plot shows the gene expression significantly correlated between Stereo-seq-FFPE and ST-FFPE-mIF. Correlation coefficients were determined by the Pearson correlation coefficient. F Cell annotation is performed based on transcriptomic data at bin50 resolution. Representative figures show spatial visualization of cell identification and heatmap depicting the expression of classic markers of 10 annotated cell types. G Bar chart displays the percentage of each cell type within the analyzed sample Comparative assessment between results from ST-FFPE-mIF and standard Stereo-seq-FFPE was performed to estimate the capture efficiency. The sequencing depth was meticulously controlled, employing technical triplicates of continuous sections for each methodology. Figure 2C indicated that ST-FFPE-mIF could profile the spatial transcriptome with a comparable number of detected genes and UMIs per spot when compared to standard Stereo-seq-FFPE. Moreover, there was a significant overlap of detected gene types (Fig. 2D) and a strong concordance in gene expression (Fig. 2E) between the two techniques. In addition to evaluating general capture efficiency, we further compare the annotation results. We annotated the ST data at a resolution of bin50 using cell2location [20, 21], with a corresponding single-cell RNA sequence (scRNA-seq) reference [22]. The distribution of annotated cell types was consistent with the H&E staining of the adjacent section. Marker genes of the corresponding cell types expressed specially, demonstrating the accuracy of annotation (Fig. 2F, Additional file 3: Fig. S1, and Fig. S2A). We also observed the proportions of annotated cell types were consistent between the two techniques (Fig. 2G). This suggests that the spatial transcriptomic data acquired through the ST-FFPE-mIF maintains a high consistency with the data generated by the commercially available Stereo-seq-FFPE technology. Collectively, these results indicate that ST-FFPE-mIF can simultaneously obtain multi-omics data without compromising the capture efficiency. ST-FFPE-mIF guarantees the production of authentic, high-quality mIF images Chromogenic immunohistochemistry (IHC) is widely employed to assess protein biomarker expression in FFPE tissue, which is widely used in pathology [23]. To verify the precision and reliability of mIF staining in ST-FFPE-mIF, we utilized adjacent sections for IHC staining of each target. This approach facilitated a direct comparison of signal consistency within identical field of view (FOV), thereby enabling an unequivocal assessment of individual IF results in relation to IHC outcomes from corresponding regions. Despite the unavoidable effects of intrinsic autofluorescence and signal channel crosstalk [24, 25], the tissue-specific structural features and positive signals observed between IHC and individual IF demonstrated a high degree of concordance (Fig. 3A). Furthermore, we employed ImageJ software to calculate the positive area across 20 randomly selected FOVs in each image, which permitted a comparative analysis of IHC and mIF results based on specific targets. The comparison revealed significant correlation in positive areas between mIF and IHC for each target (Fig. 3B). The observed discrepancies in positive signal localization can be attributed to the fundamentally distinct imaging principles and detection sensitivities of IHC versus mIF. In summary, our mIF images exhibited authentic and reliable fluorescence signals that could support precise identification of cells types within tumor microenvironment.Fig. 3Assess the accuracy of mIF staining from ST-FFPE-mIF staining by using chromogenic IHC as gold standard. A Representative FOVs in photomicrograph of mIF and chromogenic IHC show comparable results. B These scatter plots illustrate the significant correlation between the positive areas of IHC and mIF across 20 fields of view for each marker Assess the accuracy of mIF staining from ST-FFPE-mIF staining by using chromogenic IHC as gold standard. A Representative FOVs in photomicrograph of mIF and chromogenic IHC show comparable results. B These scatter plots illustrate the significant correlation between the positive areas of IHC and mIF across 20 fields of view for each marker To assess the effectiveness of the antibody elution, slides were stained with a four-plex antibody panel of 1 st round and imaged, then subjected to elution followed by re-imaging of the same regions. After elution, no specific antibody-derived signals were detected. Quantitative analysis further confirmed a significant reduction in positive area from primary staining compared to post-elution. It also showed that post-elution signals were relatively comparable to the negative control (Additional file 3: Fig. S3), demonstrating that the elution procedure effectively removed bound antibodies, reducing residual signals similar to background levels. Evaluation of technical reproducibility of ST-FFPE-mIF Three adjacent sections from the same human CRC sample were utilized to assess technical reproducibility. The ST-FFPE-mIF datasets generated from these samples revealed an outstanding consistency across several fundamental metrics at bin50 resolution, which includes similar distribution of number of genes and UMIs in violin plots, a high overlap of captured gene types, and correlations of captured gene profiles exceeding 0.99 (Fig. 4A-C and Additional file 3: Fig. S4). Then, we calculated Moran’s I score, which measured global spatial autocorrelation for genes captured in three sections. The correlation analysis of Moran’s I score showed that there were high correlations among the adjacent three sections. Simultaneously, we illustrated the spatial distribution of specific genes such as COL3A1 across the three sections. Notably, spatial pattern of selected gene expression in all three sections was similar (Fig. 4D and Additional file 3: Fig. S2B). To assess the performance at cellular level, annotations were performed at single-cell resolution. Subsequently, correlations between identical cell types in the three samples were computed. The results show high consistency in spatial distributions, compositional proportions, expression of classic marker genes and general gene profiles among annotated cells across the three replicates (Fig. 4E - F and Additional file 3: Fig. S2A). Furthermore, we conducted an assessment of capture efficiency and RNA quality tolerance of ST-FFPE-mIF across three tissue types, including breast cancer, gastric cancer, and renal cancer. Despite being applied to samples with a range of DV200 values and low RNA quality numbers (RQN) (Additional file 4), ST-FFPE-mIF method demonstrated robust performance, confirming its compatibility with challenging sample conditions. (Additional file 3: Fig. S5).Fig. 4Estimation of reproducibility and stability of ST-FFPE-mIF protocol. A Violin plots show the number of captured genes and UMIs at bin50 (50 × 50 DNB) resolution among three consecutive sections. B Upset plot shows commonality of captured gene type among three consecutive sections. C Correlations of normalized gene expression among three consecutive sections show stability of the technique. D Scatter plots show a high correlation of the Moran’s I score of the corresponding genes among three sections. The heatmaps show the spatial pattern of COL3A1 was similar in the three sections. E Cell annotation and the proportion of each cell type show similarities among three consecutive sections. F The stability of the technique is demonstrated by the correlations of average gene expression in identical cell types (B, T, Myeloid, Epithelial) across three sections at single cell resolution Estimation of reproducibility and stability of ST-FFPE-mIF protocol. A Violin plots show the number of captured genes and UMIs at bin50 (50 × 50 DNB) resolution among three consecutive sections. B Upset plot shows commonality of captured gene type among three consecutive sections. C Correlations of normalized gene expression among three consecutive sections show stability of the technique. D Scatter plots show a high correlation of the Moran’s I score of the corresponding genes among three sections. The heatmaps show the spatial pattern of COL3A1 was similar in the three sections. E Cell annotation and the proportion of each cell type show similarities among three consecutive sections. F The stability of the technique is demonstrated by the correlations of average gene expression in identical cell types (B, T, Myeloid, Epithelial) across three sections at single cell resolution ST-FFPE-mIF enables reliable analysis at a single-cell level Cell segmentation was performed on a DAPI-stained image that had been registered with the corresponding ST data. Fig. 5A showed a representative FOV. To evaluate the accuracy of single-cell segmentation, Intersection over Union (IoU), average precision (AP) and recall rate were calculated by comparing DAPI-based segmentation results with manually corrected segmentation as the ground truth. The analysis revealed that Cellpose segmentation achieved a high average IoU of 0.914. Furthermore, the 20 selected fields of view yielded an average AP of 0.856 and an average recall of 0.901 at an IoU threshold of 0.5. The performance of Cellpose on ST-FFPE-mIF images is comparable to that of the validated Cellpose built-in model [26, 27], confirming its high accuracy (Additional file 3: Fig. S6). Based on the single-cell segmentation results, we derived cell locations and compiled a gene expression matrix at single-cell resolution by integrating transcriptome data using an in-house script (source code in Zenodo).Fig. 5ST-FFPE-mIF enables reliable analyses at single-cell level. A Representative FOV illustrates cell segmentation in the DAPI-stained image. The red circles represent the cell boundaries inferred by Cellpose. Scale bar: 50 μm. B The spatial visualization displays each cell colored according to its cell type. C The heatmap shows specific expression of the classic markers in each cell type. D The histogram plot shows the proportion of each cell type. E The FOV represents the expanded cell boundaries based on DAPI segmentation results. The red circles represent cell boundaries inferred by Cellpose. The boundary dilation of 3 pixels is rendered in orange. Scale bar: 50 μm. F A dotted heatmap shows the scaled expression of representative biomarkers for each cell type. G Semi-qualitative analysis is performed based on HER2 signal intensity in each cell. The scatter plot shows cell rank based on HER2 signal intensity. The box plot shows there are significant differences among the three groups and the spatial visualization shows spatial distribution of these cell types (The “high,” “medium,” and “low” groups consist of 51361, 102750, and 51337 cells, respectively. ***P < 0.001) (H) GSEA analysis is performed between “high” and “low” groups. Representative figures demonstrate activation of three pathways associated with tumor development ST-FFPE-mIF enables reliable analyses at single-cell level. A Representative FOV illustrates cell segmentation in the DAPI-stained image. The red circles represent the cell boundaries inferred by Cellpose. Scale bar: 50 μm. B The spatial visualization displays each cell colored according to its cell type. C The heatmap shows specific expression of the classic markers in each cell type. D The histogram plot shows the proportion of each cell type. E The FOV represents the expanded cell boundaries based on DAPI segmentation results. The red circles represent cell boundaries inferred by Cellpose. The boundary dilation of 3 pixels is rendered in orange. Scale bar: 50 μm. F A dotted heatmap shows the scaled expression of representative biomarkers for each cell type. G Semi-qualitative analysis is performed based on HER2 signal intensity in each cell. The scatter plot shows cell rank based on HER2 signal intensity. The box plot shows there are significant differences among the three groups and the spatial visualization shows spatial distribution of these cell types (The “high,” “medium,” and “low” groups consist of 51361, 102750, and 51337 cells, respectively. ***P < 0.001) (H) GSEA analysis is performed between “high” and “low” groups. Representative figures demonstrate activation of three pathways associated with tumor development Expression of transcriptional markers was relatively specific in corresponding cell types. Proportion and spatial location of each cell type were reasonable (Fig. 5B-D). To validate results from ST-based cell annotation, we calculated fluorescence signals for each biomarker within individual cells. The results demonstrated that fluorescent signals were correctly expressed in their respective cell types (Fig. 5E-F). Additionally, we inferred spatial intercellular communications between different cell types utilizing StereoSiTE [28] (Additional file 3: Fig. S7A). Specific interactions mediated by ligand-receptor genes involved in the EGF pathway were primarily observed between Mast/Myeloid/Endothelial-Epithelial (Additional file 3: Fig. S7B). The pattern recognition results exhibited EGF pathway in pattern 3 as an incoming signal of Epithelial (Additional file 3: Fig. S7C). ST-FFPE-mIF could simultaneously acquire both nuclear and membrane fluorescence signals, which could be potentially applied for accurate cell segmentation. For example, we compared CD20 staining-based segmentation with DAPI staining-based segmentation in diffuse large B cell lymphoma. The results showed that the IoU of CD20-based segmentation reached 0.8386, which was higher than that of DAPI-based segmentation (0.4573). This indicated that membrane-based cell segmentation has the potential to obtain a relatively reliable cell boundary and high fidelity single-cell resolved data matrix (Additional file 3: Fig. S8). In addition to transcriptome profiling, the mIF images were utilized for semi-quantitative analysis to facilitate a combined multi-omics investigation. As a member of the EGFR family of receptor tyrosine kinases, the human epidermal growth factor receptor 2 (HER2) is considered to be a potential therapy target and a prognostic marker in CRC [29]. We calculated HER2 fluorescence signal intensity of all annotated epithelial cells and divided them into three groups according to percentiles, which were named “high” (75th percentile), “middle”, and “low” (25th percentile), respectively (Fig. 5G). We calculated the differentially expressed genes in groups of “high” and “low” for enrichment analysis. GSEA analysis indicated that three pathways “HALLMARK_TNFA_SIGNALING_VIA_NFKB”, “HALLMARK_IL2_STAT5_SIGNALING” and “HALLMARK_KRAS_SIGNALING_UP”, were activated in cells with high HER2 signals (Fig. 5H). These analyses showed that the fluorescence signals of biomarkers can be combined with transcriptomic data to provide clues for exploring new insights for mechanisms. Discussion Spatial transcriptomics is a start-of-the-art method implemented in many research fields, including cancer biology and developmental biology [30]. Although all proteins are translated from messenger RNAs, post-transcriptional regulation and modifications often lead to discrepancies in their expression patterns [31]. Multi-omics methodologies at spatial resolution are significant for the investigation of the intermolecular/intercellular dynamics on the transcriptome and proteome levels in the same single cells across tumor development and progress. In this paper, we present a co-detection technology that offered subcellular resolution, enabling the simultaneous spatial profiling of the complete transcriptome alongside multiplex immunofluorescence staining. Herein, we provided a detailed protocol of ST-FFPE-mIF. Taking fluorescent channels, fluorophores, the isotypes, and species reactivity of secondary antibodies into consideration, we proposed an image-based strategy utilizing FL-conjugated primary antibodies together with antibody elution to increase the detection targets and to reduce time-consuming (Fig. 1 and Additional file 1). Given the accessibility to FL-conjugated primary antibodies, we additionally evaluated the secondary antibody-based mIF technique in the context of lymphoma and have included the suggested protocol in Additional file 2. The performance of the two protocols demonstrated consistency and comparability (Additional file 3: Fig. S9). Notably, FL-conjugated primary antibody is frequently regarded as having lower sensitivity for targets with low abundance [32–34]. It is advisable to carefully match fluorescence intensity with antibody signals; for instance, using stronger fluorophores for antibodies that yield weaker signals can effectively enhance the signal-to-noise ratio [35, 36]. Autofluorescence poses a significant challenge to the accurate interpretation of signals, potentially leading to false positive results. To address this issue, we provided an optional step to optimize image quality. Taking breast cancer samples as an example, specificity of marker expression proved the controllable impact of autofluorescence on overall signal-to-noise ratio of the images (Additional file 3: Fig. S5A). Furthermore, it has been reported that multiple rounds of antibody elution have a limited effect on marker intensity or tissue morphology [32–34]. This indicates that our technology has the potential to increase the target number by raising the number of cycles. Figures 2, 3 and 4 proved the authenticity, specificity, sensitivity, stability, and reproducibility of ST-FFPE-mIF. To further explore how ST-FFPE-mIF can be used in scientific research, we described the landscape of CRC with ST-FFPE-mIF by inferring spatial intercellular communication at single-cell resolution (Additional file 3: Fig. S7A-C). Variations in cell morphology and size introduce uncertainty in defining cellular boundaries and pose significant challenges for accurate segmentation [37, 38]. Nonetheless, leveraging fluorescence signals from the cell membrane, as demonstrated in Fig S8. In addition to cell segmentation, the current analytic pipeline also demonstrated two strategies for using mIF images: identifying of cell types and semi-quantitatively analyzing functional biomarker expression (Fig. 5F-H). The current protocols still have several limitations that need to be addressed in the future. Firstly, we need to improve the iterative cycle number for tagging. It is anticipated that advancements in imaging techniques and an increase in cyclic mIF staining will significantly expand the detection of protein targets. Secondly, it is essential to develop a standardized and fully automated mIF image processing and analysis software, including multi-round image registration, background noise elimination, and protein expression quantification. In summary, the integration of high-precision, unbiased whole transcriptome capture through Stereo-seq-FFPE with multiplex immunofluorescence imaging within a single tissue section establishes a robust framework for acquiring spatial multi-omics information at single-cell resolution. This innovative methodology facilitates comprehensive spatial omics analyses, allowing for precise elucidation of structural and functional dynamics within complex multicellular interactions in the tumor microenvironment. ST-FFPE-mIF shows its significant potential across diverse applications including fundamental biology, diagnostics, and therapeutic development. This technology enables the acquisition of aligned multimodal data, facilitating research on co-localization of cells, discovery of novel cell types, identification of function and integrated multi-modal analysis [39–41]. Conclusions Overall, co-detection of multi-omic technology opens up new possibilities for biological research. Future advancements are expected to provide valuable insights into biological processes, promoting precision medicine and enhancing the ability to address complex health challenges. Methods Sample collection Human FFPE tissue samples were obtained from OUTDO BIOTECH CO. LTD (Shanghai, China), an accredited and legally authorized human tissue bank. All samples were strictly anonymized and coded prior to delivery. ST-FFPE-mIF solution methodology Tissue de-paraffinization and antigen retrieval For the human colorectal cancer FFPE sample included in the ST analysis, we performed ST-FFPE-mIF assays using Stereo-FFPE-seq technology (BGI Shenzhen, China) following the manufacturer’s instructions. Tissue sections were cut to a thickness of 5 μm and mounted on Stereo-seq N transcriptomics chips (Cat#211SN114, STOmics). Next, the Stereo-seq N chip with the tissue section was dried and baked overnight at 38 °C. Following this, the chip underwent a one-hour baking process at 60 °C to facilitate softening and melting of the paraffin. De-paraffinized in Histo-clear substitute and graded ethanol according to Stereo-seq Transcriptomics N Kit (Cat#211SN114, STOmics). Subsequently, the chip with the tissue section was submerged in Tris-EDTA antigen retrieval solution (Cat#C1037, Solarbio) and incubated at 95 °C for 25 min to facilitate antigen retrieval. Multiple immunofluorescence staining After de-paraffinization and antigen retrieval, blocking was performed by adding 100 µL of blocking buffer containing 1x SSC Buffer (Cat#AM9770, Thermo), 0.05 U/µL RNase inhibitor (Cat#M0314L, NEB), 10% 1:1 diluted Gibco™ Horse Serum (Cat#26050070, Thermo) and Gibco™ Goat Serum (Cat#16210064, Thermo), followed by a 20 min incubation at room temperature. After removing the blocking buffer, primary antibodies labeled fluorophores were added at the appropriate concentrations, diluted in the blocking buffer, and incubated for 45 min at room temperature. This was followed by washing the chip with 0.1x SSC Buffer with 0.05 U/µL RNase inhibitor for 2 min, repeated twice. Subsequently, the chip surface was treated with 5 µL of glycerol (Cat#4100854050, Sangon Biotech) to ensure that the tissue was completely covered by glycerol and the coverslip. Imaging was conducted using a ZEISS Axioscan 7 microscope. After imaging, the chip was washed with 0.1 x SSC buffer supplemented with 0.05 U/µL RNase inhibitor. AbEraser (Cat#AXT9710000, AlphaX (Beijing) Biotech Co) was used to elute the antibody from the section. The elution buffer was then applied to the chip surface and incubated at 50 °C for 5 min. After removing the solution, the chip was sequentially washed with DMSO (Cat#A610163, Sangon Biotech) and 0.1x SSC Buffer supplemented with 0.05 U/µL RI. For the 2nd round of immunofluorescence staining, the primary antibodies incubation was repeated as previously mentioned. Nuclei were stained with DAPI and performed according to standard procedure. Multispectral images of stained chip were captured using a ZEISS Axioscan 7 microscope. Primary antibodies included Alpha-Smooth Muscle Actin (Cat#S0B1508, Starter, 1:100), Vimentin (Cat#S0B1507, Starter, 1:200), Cytokeratin 19 fragment (Cat#S0B1566, Starter, 1:50), Villin (Cat#S0B1572, Starter, 1:100), p53 (Cat#S0B1569, Starter, 1:200), CD3 epsilon (Cat#S0B0217, Starter, 1:100), CD68 (Cat#S0B1501, Starter, 1:100), HER2 (Cat#S0B1567, Starter, 1:50), CK-Pan (Cat#S-P001, Starter, 1:200) CD163 (Cat#S-222-171, Starter, 1:100), CD45 (Cat#S-R035, Starter, 1:200), CD20 (Cat#S-R133, Starter, 1:200), CD79a (Cat#S-030-18, Starter, 1:200), CD117 (Cat#S-125-51, Starter, 1:200). For samples characterized by elevated autofluorescence, a sequential addition of 100 µL of 1 M copper (II) sulfate (Cat#A603008-0250, Sangon Biotech) and 10% ammonium chloride (Cat#A631002, Sangon Biotech) was conducted, followed by a 3 min incubation at room temperature. The chip was subsequently rinsed with a wash buffer prior to the blocking step. In accordance with the previously described immunofluorescence staining protocol, the surface of the chip was treated with a quench-resistant sealant (Cat#P0128S, Beyotime or Cat#P36971, Thermo Fisher) in lieu of glycerol, ensuring complete coverage of the tissue with a coverslip. Imaging was then carried out utilizing a ZEISS Axioscan 7 microscope. Library preparation and sequencing Sections were processed according to manufacturer’s instructions for the Stereo-seq Transcriptomics N Kit (Cat#211SN114, STOmics). In short, the chip with tissue section was submerged in the FFPE Decrosslinking Reagent for decrosslinking (Cat#211SN114, STOmics). Fixation in pre-cooled methanol (Cat# 34860, Sigma) for 30 min at − 20 °C ensued. Post-fixation, the Stereo-seq chip was air-dried, and the tissue section was incubated in permeabilization buffer (Cat# 211SN114, STOmics) supplemented with 3 µL of Tetramethylene sulfone (Cat#A610163, Sangon Biotech) for 25 min at 37 °C. After permeabilization, the FFPE Dimer mix (Cat#211SN114, STOmics) was added and incubated at room temperature for 40 min. Subsequently, the reverse transcription mix (Cat#211SN114, STOmics) was pipetted onto the chip surface and incubated at 42 °C overnight. After reverse transcription, cDNA products were released from the chip by Release Enzyme Mix and collected. Library construction was performed according to the Library preparation Kit (Cat# 111KL114, STOmics) and subsequent DNB generation. Finally, the DNBs were sequenced with the MGI DNBSEQ-T7 or DNBSEQ-2000RS Sequencer. H&E stain and immunohistochemistry (IHC) H&E staining and IHC were performed on adjacent FFPE sections using a standard procedure. Samples were sectioned at a thickness of 5 μm, and tissue H&E staining was carried out using Hematoxylin (Cat#S330930-2, Dako) and Eosin (Cat#HT110216, Sigma-Aldrich). The staining times varied depending on tissue type. For the biochemical process of IHC, the 4 μm thick section underwent a series of standard procedures including de-paraffinization, hydration, and antigen retrieval. The sections were incubated in 3% hydrogen peroxide to prevent non-specific background staining. The appropriate blocking, antibody incubation, Diaminobenzidine (Cat#DAB-1031, Fuzhou Maixin Biotech) and Hematoxylin (Cat#H8070, Solarbio) stain time were selected according to different sections. The H&E and IHC images were captured with a 20X objective using a ZEISS Axioscan 7 microscope. ST-FFPE-mIF Raw data processing SAW [42] was used to generate the gene expression matrix from raw FASTQ data of the ST-FFPE-mIF assay. We aligned the coordination identity (CID) from read 1 to a whitelist of barcodes with known coordinates, allowing for one base mismatch. UMI sequences from read 2 were filtered according to the following criteria: (1) the presence of N bases and (2) having more than two bases with a Phred score lower than 10. Only reads with valid CID and high quality UMI were used for subsequent analyses. The associated read 2 was aligned to the reference genome (GRCh38) using STAR [43], with only uniquely mapped reads having mapping scores >10 were kept for gene annotation and read counting. Finally, we counted the deduplicated UMI numbers for each CID and each gene as gene expression counts, and aggregated them to a matrix file with spatial coordinates. Staining image registration Two round mIF registration To register two round mIF images, TrakEM2 within Fiji (v2.16.0) [44] was utilized. The two images were loaded into a new TrarkEM2 project, then one image was locked and the other was adjusted. During the process, we adjusted the position and size so that they matched accurately. Finally, we export two registered flat images. mIF images registration with transcriptome To register mIF images and transcriptome, Cellbin2 was utilized (https://github.com/STOmics/cellbin2). Initially, the quality of the DAPI-stained image was evaluated. Then, the DAPI-stained image was used for registration with the transcriptome to obtain image transform parameters, which were used to adjust other fluorescence images. Quantification and comparison of mIF and IHC images To assess the accuracy of signals, we used a pixel-based approach to quantify signals within mIF and IHC images. The approach was a measure of positive pixel area within a given region. Registered image was divided into 20 equal regions, and a field of view was randomly selected from each region to represent the entire images as much as possible. If a FOV has no tissue, we choose a field of view from neighboring regions, distinguished positive and negative pixels based on thresholds, and measured the area using ImageJ. Then, pearson correlation coefficients were calculated by R software. Evaluation of antibody elution efficiency To evaluate the effectiveness of antibody elution, the primary-staining images, post-elution images and negative control of the 1 st round markers were used to assess the effectiveness. 20 FOVs were captured from each image for further analysis, following the same methodology as the IHC and mIF assessments described above. Cell cluster and annotation We used Scanpy (v1.9.3) [45] to construct the data structure and perform cell cluster. The workflow comprised the following operations: (1) Data normalization and scaling; (2) Data dimensionality reduction; (3) Cell clustering using the Leiden algorithm. A single-cell dataset of human CRC [22] was used as a reference to deconvolute a mixture of cells in ST-FFPE-mIF data by Cell2location with N_cell_per_location = 6, detection_alpha = 20. Each cell was assigned a cellular identity. Finally, the proportions of cell types were calculated and visualized in R software. Identification of spatially auto-correlated genes To assess spatial similarity between consecutive sections, we calculated Moran’s I score for captured gene, using Scanpy and Squidpy (https://github.com/scverse/squidpy). Initially, Scanpy was used to filter out cells and genes according to two criteria: (1) Cells with UMI less than 50 and the proportion of mitochondrial genes more than 10%; (2) Genes expressed in less than 10 cells. Data normalization and logarithmic transformation were performed after filtering. Finally, Squidpy was used to calculate Moran’s I score using the spatial_neighbors and spatial_autocorr functions. Cell segmentation DAPI-based cell segmentation The deep-leaning toolkit Cellpose [27] was used to obtain single-cell spatial coordinates in DAPI-stained images. The built-in ‘‘nuclei model’’, which is pre-trained to handle segmentation of nuclear fluorescence staining, was used to perform cell segmentation. We set the parameter channel = [3, 0], flow_threshold = 0.8 and diameter = 0. Fluorescence-based cell segmentation Cellpose was also used to perform cell segmentation in CD20-stained fluorescence images using a built-in “cyto3” model. The parameters were set to channel = [2,0], flow_threshold = 0.8 and diameter = 0. Evaluation of single-cell segmentation To assess accuracy of the cell segmentation based on DAPI, 20 regions were extracted from the results of segmentation by Cellpose and manually corrected them using Qupath [46] software to establish the ground truth. Subsequently, the IoU, average precision and recall rate of corresponding region using the built-in functions in Cellpose. Comparing the cell segmentation based on CD20 and DAPI using the identical method for calculating IoU. Analysis of intercellular communications StereoSiTE (https://github.com/STOmics/StereoSiTE) [28] was downloaded and used to perform spatial cell interaction analysis, The parameters were set to Secreted Signaling = 200, ECM-Receptor = 200, Cell-Cell Contact = 30. The built-in database named “CellChatDB-human.csv” was selected. Cell boundary expansion and fluorescence signals calculation The function “skimage.segmentation.expand_labels” within scikit-image [47] was used to expand the cell boundary. Setting parameter distance = 3. Then, the function “cv2.cvtColor” within opencv (https://github.com/opencv/opencv) was used to transform the RGB image into gray-scale image, and add the gray-scale values of each biomarker within the cell separately to generate matrix. Epithelial cell subtyping and GSEA analysis Epithelial cells were categorized according to the signal intensity of HER2, followed by a t-test analysis in R software. We used Scanpy to extract all epithelial cells and used Numpy (https://github.com/numpy/numpy) to calculate the 25th and 75th percentiles of HER2 intensity, which reached 983 and 8414, respectively. The function “scanpy.tl.rank_genes_groups” within Scanpy was used to calculate the difference of gene expression. ClusterProfiler (https://github.com/YuLab-SMU/clusterProfiler) was used to perform GSEA analysis, and filter pathways with p-value < 0.05. Supplementary Information Additional file 1. ST-FFPE-mIF-direct multiplex immunofluorescence (mIF) staining procedure. Additional file 1. ST-FFPE-mIF-direct multiplex immunofluorescence (mIF) staining procedure. Additional file 2. ST-FFPE-mIF-indirect multiplex immunofluorescence (mIF) staining procedure. Additional file 2. ST-FFPE-mIF-indirect multiplex immunofluorescence (mIF) staining procedure. Additional file 3. Fig S1-S9. Contains all supplementary figures referenced in the manuscript. Additional file 3. Fig S1-S9. Contains all supplementary figures referenced in the manuscript. Additional file 4. DV200 values of RNA-degraded FFPE samples (breast cancer, gastric cancer, lymphoma, renal carcinoma). Additional file 4. DV200 values of RNA-degraded FFPE samples (breast cancer, gastric cancer, lymphoma, renal carcinoma).