Reversible fixation of human cells in suspension

Extending previous studies using DSP to preserve cell lines for RNA sequencing [16], we initially confirmed its applicability for single-cell analysis of cells in suspension (human peripheral blood mononuclear cells; PBMCs) and microfluidics systems (10x Genomics Chromium Controller), before combining fixation and dissociation of complex solid tissues. To this end, we compared cell morphology, RNA integrity, and reverse transcription efficiency of fresh and DSP-fixed PBMCs. Fixed cells showed highly similar morphology compared to fresh PBMCs in bright-field microscopy, with no evident changes in cell phenotypes or sample clumping after fixation (Fig. S1a). Next, we captured and barcoded single cells from both fresh and DSP-fixed samples using the Next GEM Single Cell 3′ Reagent Kits v3.1 from 10x Genomics. Bioanalyzer profiles of the amplified cDNA from both samples were virtually identical, demonstrating DSP fixation not to affect RNA integrity or reverse transcription performance (Fig. S1b,c). After sequencing, we confidently mapped over 80% of the reads from both sequencing libraries to the human reference genome, with over 50% of exonic reads usable for quantifying gene expression levels (Fig. 1a). We observed a comparable correlation between the number of detected genes or the number of unique molecular identifiers (UMIs) and sequencing depth for fresh and fixed samples (95% CI, 3.65e−05 ± 1.11e−05 and 0.44 ± 0.06 vs. 3.56e−05 ± 1.14e−05 and 0.35 ± 0.06, respectively), indicating DSP fixation to conserve library complexity (Fig. 1b). Briefly, we captured a total of 22,481 genes in both conditions, together with 1667 and 1482 specific for fresh and fixed samples, respectively. We confirmed this observation at the single-cell level, where we found a similar relationship between sequencing depth and the number of detected UMIs or genes per cell (Fig. S1d). In line, we observed similar gene counts in single blood cells in fresh and fixed samples (Fig. 1c). After filtering out low-quality cells, we found a similar distribution of the main quality control (QC) metrics between both protocols (Fig. S1e), except for a few specific cell subpopulations (Fig. S1f). These results suggest DSP fixation to conserve the ability to detect gene transcripts in single cells compared to fresh samples in scRNA-seq experiments.

Fig. 1

FixNCut protocol in human peripheral blood mononuclear cells (PBMCs). a Mapping analysis of sequencing reads within a genomic region. b Comparative analysis of the number of detected genes (top) and UMIs (bottom) across various sequencing depths. c Cumulative gene counts analyzed using randomly sampled cells. d Principal component analysis (PCA) and uniform manifold approximation and projection (UMAP) representation of gene expression profile variances of fresh and fixed samples. e, f Linear regression model comparing average gene expression levels of expressed genes (e) and main biological hallmarks, including apoptosis, hypoxia, reactive oxygen species (ROS), cell cycle G2/M checkpoint, unfolded protein response (UPR), and inflammatory response genes (f). The coefficient of determination (R2) computed with Pearson correlation and the corresponding p-value are indicated. g UMAP visualization of 17,483 fresh and fixed PBMCs, colored by 19 cell populations. h Comparison of cell population proportions between fresh (n = 9754) and fixed (n = 7729) PBMCs with the Bayesian model scCODA. Asterisks (*) indicate credible changes. i Differential gene expression analysis between fresh and fixed samples. The top differentially expressed genes (DEGs) with significant adjusted p-values (FDR) < 0.05, upregulated (red), and downregulated (blue) with Log2FC > |1| are indicated. j Violin plot of ex-vivo blood handling gene signature score [19] for fresh and fixed human PBMCs. Statistical analysis between fixed and fresh cells was performed using the Wilcoxon signed-rank test. k Dotplot showing the average expression of sampling-time DEGs for Fresh (y-axis) for all 19 cell types (x-axis) split by protocol. The dot size reflects the percentage of cells in a cluster expressing each gene, and the color represents the average expression level

To further assess potential technical variation between protocols, we identified highly variable genes (HVGs) independently in fixed and fresh PBMCs. We found that 70% of HVGs were shared between the two protocols, indicating a conserved representation of the transcriptome and suitability for joint downstream processing (Fig. S1g). Additionally, when we examined the variation captured in the main principal components (PCs) and displayed single-cell transcriptomes in two dimensions (uniform manifold approximation and projection; UMAP), we did not observe any notable outliers due to the sampling protocol (Fig. 1d). Cells clustered together based on biological differences rather than preparation protocol, suggesting fixed and fresh cells to have similar capacity for cellular phenotyping. The pseudo-bulk gene expression profiles between fixed and fresh samples were highly correlated (R2 = 0.99, p < 2.2e−16) (Fig. 1e), indicating DSP fixation not to alter the expression of specific genes. This was further confirmed at the cell population level (Fig. S1h). Moreover, the biological processes such as apoptosis, hypoxia, reactive oxygen species (ROS), cell-cycle (G2/M checkpoint), unfolded protein response (UPR), and inflammation hallmarks remained unchanged across libraries (Fig. 1f).

Next, we performed a joint analysis of 17,483 fresh and fixed cells, which were clustered to define 19 distinct cell populations (Fig. 1g; Fig. S1i). All cell types and states were found across both protocols at similar proportions, except for classical monocytes and NK cells, which showed small but significant differences, being slightly increased in fresh and fixed, respectively (Fig. 1h). Fixation did not affect differential expression analysis (DEA), with only four upregulated genes representing hemoglobin subunits (HBA1, HBA2, and HBB) and a mitochondrial gene (MT-NDL4) (Fig. 1i; Additional file 2: Table 1). These genes were consistently found across all cell populations (Additional file 3: Table 2), a phenomenon also observed when performing digestion protocols at low temperatures [3]. The FixNCut protocol may prevent erythrocyte lysis, leading to their co-encapsulation with nucleated blood cells and the detection of specific transcripts.

Importantly, we observed a reduction in technical artifacts introduced during sample processing prior to single-cell experiments upon fixation. Specifically, gene expression alterations previously defined to correlate with ex vivo PBMC handling [19] showed a significant reduction upon fixation (p < 2.2e−16) (Fig. 1j). Moreover, a sampling-time gene signature obtained from single-cell benchmarking studies [4] also showed a significant reduction in the fixed PBMCs (p < 2.2e−16) (Fig. 1k). Interestingly, T lymphocytes appeared to be particularly affected (p < 0.0001; except for gdT cells, p < 0.05), showing the strongest protection from sampling artifacts in fixed cells (Fig. S1j). DSP also protected against the general reduction of gene expression activity, previously reported during PBMC sample processing [4]. Notably, more than 30% of genes from the sampling-time signature were also detected as enriched in the fresh PBMCs (Fig. S1k). The results suggest that fixed PBMCs have comparable cellular composition and gene expression profiles to freshly prepared samples, while reducing gene expression artifacts introduced during sample preparation.

FixNCut protocol applied on mouse solid tissues

Beyond the benefits of cell fixation in standardizing sample processing and preserving gene expression profiles of cells in suspension, the FixNCut protocol was specifically designed for solid tissues. Specifically, it allows for fixation and subsequent digestion, which is particularly advantageous for complex and logistically challenging study designs, such as clinical trials. Here, sampling artifacts, including biases in gene expression and cell type composition, are frequently observed in fragile solid tissue types. For example, differentiated colonic epithelial cells (e.g., secretory or absorptive cells), tightly connected adult neurons, or processing-sensitive adipocytes are more susceptible to cell damage and death as a result of common tissue dissociation protocols [20, 21]. Fixation prior to digestion using the FixNCut protocol can reduce these artifacts. Thus, we next evaluated the effectiveness of the FixNCut protocol with subsequent scRNA-seq readout in different solid mouse tissues before extending its application to challenging human patient samples, such as tissue biopsies.

Fresh mouse lung samples were minced, mixed, and split into two aliquots, one processed fresh and the other fixed using the FixNCut protocol with subsequent 30-min tissue digestion using Liberase TL. The fixed sample showed a slight decrease in cell size and an increase in DAPI+ cells, but overall, cell morphology was similar to the fresh sample (Fig. S2a). Single-cell encapsulation and scRNA-seq (10x Genomics, 3′ RNA v3.1) showed comparable proportions of reads mapped to the mouse reference genome and exonic genomic regions for both fresh and fixed samples (Fig. 2a). We further observed a similar correlation between the number of detected genes or UMIs and sequencing depth in fresh and fixed samples (95% CI, 2.31e−05 ± 8.03e−06 and 0.40 ± 0.06 vs. 2.64e−05 ± 7.86e−06 and 0.40 ± 0.06, respectively) (Fig. 2b). At the cell level, we confirmed the similar complexity of fixed libraries, as reflected by the number of detected UMIs and genes (Fig. S2b). Genes identified in both libraries (n = 20,684) were mostly protein-coding genes (76%). Conversely, genes exclusively captured in either fixed (n = 1383) or fresh (n = 1157) samples were largely non-coding RNA genes, specifically lncRNAs (52% vs 47%) (Fig. S2c). We further observed that more genes were captured for the fixed samples after accumulating information from a few number of individual lung cells (Fig. 2c). Importantly, after filtering out low-quality cells, the main QC metrics in fixed samples showed consistent distributions across all characterized cell types (Fig. S2d,e), suggesting that FixNCut protocol preserves the capacity for scRNA-seq profiling after fixation and digestion.

Fig. 2

FixNCut protocol tested in mouse lung samples. a Mapping analysis of sequencing reads within a genomic region. b Comparative analysis of the number of detected genes (top) and UMIs (bottom) across various sequencing depths. c Cumulative gene counts analyzed using randomly sampled cells. d Principal component analysis (PCA) and uniform manifold approximation and projection (UMAP) representation of gene expression profile variances of fresh and fixed samples. e Linear regression model comparing average gene expression levels of expressed genes between protocols. The coefficient of determination (R2) computed with Pearson correlation is indicated. f Hierarchical clustering of coefficient of determination (R2) obtained for all pair comparisons between protocols for biological hallmarks, including apoptosis, hypoxia, reactive oxygen species (ROS), cell cycle G2/M checkpoint, unfolded protein response (UPR), and inflammatory response genes. g UMAP visualization of 19,606 fresh and fixed mouse lung cells, colored by 20 cell populations. h Comparison of cell population proportions between fresh (n = 10,289) and fixed cells (n = 9317). The top figure shows the difference in cell population proportions between fresh and fixed samples, and the bottom figure shows the results of the compositional cell analysis using the Bayesian model scCODA. Asterisks (*) indicate credible changes, upregulated for the fresh sample. i Differential gene expression analysis between fresh and fixed samples. The top differentially expressed genes (DEGs) with significant adjusted p-values (FDR) < 0.05, upregulated (red), and downregulated (blue) with Log2FC > |1| are indicated

An overlap of almost 80% of sample-specific HVGs was found when comparing the fresh and FixNCut protocols (Fig. S2f). The absence of batch effects linked to protocols was demonstrated by the PCA and UMAP representations (Fig. 2d), indicating bias-free transcriptome profiles after cell fixation and digestion. Highly comparable profiles of mean gene expression values were observed between fresh and fixed mouse lung samples (R2 > 0.99, p < 2.2e−16) (Fig. 2e), a finding also confirmed at the population level (Fig. S2g). Moreover, the high correlation across gene programs supported the absence of alterations in major biological processes (Fig. 2f).

We then performed a joint analysis of all 19,606 mouse lung cells, which were segregated into 20 distinct cell populations, encompassing both lung and tissue-resident immune cells (Fig. 2g; Fig. S2h). All characterized cell types were detected in both fresh and fixed samples, with slight variability in cell type proportions between both protocols. The fixed protocol showed an improved representation of tightly connected epithelial and endothelial cell types, while immune cells (B and T cells, monocytes, monocyte-derived DCs, and neutrophils) were proportionally increased in the fresh sample (Fig. 2h). To validate preserved gene expression profiles in fixed tissues, we performed differential expression analysis (DEA) between the two protocols. We observed upregulation of genes related to pneumocytes (Sftpc), myeloid enhancer binding protein (Cebpb), and endothelial cells promoting cell migration (Cxcl2) for the fixed protocol, which could be largely explained by the enrichment of this population upon fixation. In contrast, fresh samples were enriched in genes related to inflammatory and immune processes (Ms4a4b and Trbc2), in accordance with the increased proportion of recovered immune cells (Fig. 2i; Additional file 4: Table 3). At the cellular level, we found a uniform enrichment in stress-related genes in non-immune populations from the fresh sample, while the fixed sample showed this enrichment in immune populations (Additional file 5: Table 4). Gene set enrichment analysis (GSEA) stratified by cell type revealed that freshly prepared endothelial cells were enriched in ROS, apoptosis, and cellular response to external stimuli, whereas the opposite patterns were observed upon fixation (Additional file 6: Table 5). Overall, these results suggest the global conservation of library complexity and quality, along with the inclusion of tightly connected, challenging-to-isolate cell types in fixed mouse lung samples.

We further evaluated the performance of the FixNCut protocol in a different challenging solid tissue context. To do so, we minced and mixed mouse colon samples that were split and subjected to scRNA-seq after digestion of either fresh or fixed tissues. Our results indicate that FixNCut provides several benefits, including improved transcriptome capture accuracy, as evidenced by a higher number of total reads mapped to the reference and a higher exonic fraction (Fig. 3a). Additionally, the fixed sample exhibited a higher non-significant library complexity based on the total number of detected genes (95% CI, 8.81e−05 ± 2.83e−05 vs. 9.03e−05 ± 3.16e−05) coupled with an increased number of total UMIs at deeper sequencing (95% CI, 0.71 ± 0.05 vs. 0.73 ± 0.07) (Fig. 3b), with fixed cells showing increased numbers of detected UMIs and genes (Fig. S3a). Genes identified in both libraries (n = 18,314) were mostly protein-coding genes (81%). Conversely, genes exclusively captured in fixed samples (n = 2225) compared to those from fresh (n = 1011) showed a larger percentage of protein-coding (45% vs 35%) coupled with a smaller fraction of lncRNA (44% vs 53%) (Fig. S3b), indicating that FixNCut enhances the gene capture efficiency, potentially allowing for a more fine-grained cell phenotyping after sample fixation. Notably, the cumulative gene count was greater for the fixed colon, particularly when considering a larger number of sampled cells (Fig. 3c), and we observed improved QC metrics for the FixNCut sample after filtering out low-quality cells, which held true across all cell populations (Fig. S3c,d).

Fig. 3

FixNCut protocol tested in mouse colon samples. a Mapping analysis of sequencing reads within a genomic region. b Comparative analysis of the number of detected genes (top) and UMIs (bottom) across various sequencing depths. c Cumulative gene counts analyzed using randomly sampled cells. d Principal component analysis (PCA), uniform manifold approximation and Projection (UMAP) prior data integration, and harmony integrated UMAP representation of gene expression profile variances of fresh and fixed samples. e, f Linear regression model comparing average gene expression levels of expressed genes (e) and biological hallmarks, including apoptosis, hypoxia, reactive oxygen species (ROS), cell cycle G2/M checkpoint, unfolded protein response (UPR), and inflammatory response genes (f). The coefficient of determination (R2) computed with Pearson correlation and the corresponding p-values are indicated. g UMAP visualization of 14,387 fresh and fixed mouse colon cells, colored by 16 cell populations. h Comparison of cell population proportions between fresh (n = 6009) and fixed (n = 8378) mouse colon samples with the Bayesian model scCODA. Asterisks (*) indicate credible changes, upregulated for the fixed sample. i Differential gene expression analysis between fresh and fixed samples. The top differentially expressed genes (DEGs) with significant adjusted p-values (FDR) < 0.05, upregulated (red), and downregulated (blue) with Log2FC > |1| are indicated

The overlap of HVGs between the fresh and fixed colon samples was slightly lower than that observed in lung tissues (> 60%) (Fig. S3e). Further, we identified noticeable differences in the transcriptomic profile, as demonstrated in both PCA and UMAP representations (Fig. 3d), which were attributed to the aforementioned improvements in library complexity after DSP fixation. Given that the overall cellular transcriptomic profile remains intact, confirmed by a high correlation in mean gene expression values between the two protocols (R2 = 0.96, p < 2.2e−16) (Fig. 3e, f), we applied sample integration to collectively annotate cells and to address technical differences at cell type level (Fig. 3d). Notably, cell populations that exhibited a diminished correlation between the fresh and fixed samples coincided with cell types that were specifically enriched in the fixed sample (Fig. S3f).

We captured a total of 14,387 cells that were clustered into 16 cell populations, representing both immune and colon-epithelium cells (Fig. 3g; Fig. S3g). All cell types were detected in both conditions, but we observed a clear shift in cell type composition with an enrichment of sensitive epithelial and stromal cells in the fixed sample (Fig. 3h). Differential expression analysis revealed a higher representation of ribosomal protein and mitochondrial genes in the fixed sample, mostly explained by the larger capture of actively cycling epithelial cell population known as transit-amplifying (TA) cells (Fig. 3i; Additional file 4: Table 3). In line, both epithelial and stromal populations were also enriched in mitochondrial and ribosomal protein genes (Additional file 7: Table 6). Additionally, GSEA by cell population showed enrichment of ribosomal-dependent and metabolic processes pathways in fixed cells, especially in the sensitive populations (Additional file 8: Table 7). Together, our results demonstrate the FixNCut protocol to enhance library complexity and quality metrics, while also capturing fragile epithelial and stromal cell populations from delicate tissues, such as the colon. Thus, DSP-based fixation preserves the integrity of tightly connected cell types that are otherwise difficult to isolate for single-cell experiments, resulting in an improved representation of cell types and states in these solid tissues.

Long-term storage of fixed tissues

Conducting multi-center clinical studies can be challenging due to centralized data production and the need for storage and shipment. To address this challenge, we evaluated the combination of the FixNCut protocol with cryopreservation (cryo; 90% FBS and 10% DMSO; see the “Methods” section) to allow for the separation of sampling time and location from downstream data generation, while preserving sample composition and gene expression profiles. To test this, fresh mouse lung samples were minced, mixed, and split into three pools for fixation-only, cryo-only, and fixation+cryo sample processing. After single-cell capture and sequencing (10x Genomics, 3′ RNA v3.1), all three libraries showed comparable statistics of mapped and exonic reads across conditions, indicating successful preservation of the transcriptome (Fig. 4a). We also observed a similar relationship between the number of detected genes and the sequencing depth for all three protocols (95% CI, 2.88e−05 ± 8.64e−06 cryo vs. 2.37e−05 ± 8.07e−06 fixed/cryo), although the number of detected UMIs was statistically significantly reduced in the fixed/cryo sample compared to cryo-only (95% CI, 0.46 ± 0.06 cryo vs. 0.24 ± 0.06 fixed/cryo) (Fig. 4b), which was consistent considering the detected UMIs and genes across individual cells (Fig. S4a), but hardly noticeable when accumulating gene counts across multiple cells (Fig. 4c). Genes identified across all libraries (n = 19,509) were mostly protein-coding genes (78%). Conversely, genes exclusively captured in fixed/cryo samples compared to those from cryo-only showed a similar percentage of protein-coding genes (34% vs 36%) and comparable fractions of lncRNA (52% vs 50%) (Fig. S4b). After removing low-quality cells, we found highly comparable distributions for the main QC metrics across all samples. However, we noticed a small increase in the percentage of mitochondrial gene expression detected in the fixed/cryo sample (Fig. S4c). Similarly, the different cell populations showed consistent QC across conditions (Fig. S4d).

Fig. 4

Long-term storage of fixed mouse lung samples. a Mapping analysis of sequencing reads within a genomic region. b Comparative analysis of the number of detected genes (top) and UMIs (bottom) across various sequencing depths. c Cumulative gene counts analyzed using randomly sampled cells. d Principal component analysis (PCA) and uniform manifold approximation and projection (UMAP) representation of gene expression profile variances of fixed, cryopreserved, and fixed/cryopreserved samples. e Linear regression model comparing average gene expression levels of expressed genes across protocols used. The coefficient of determination (R2) computed with Pearson correlation is indicated. f Hierarchical clustering of coefficient of determination (R2) obtained for all pair comparisons across protocol for biological hallmarks, including apoptosis, hypoxia, reactive oxygen species (ROS), cell cycle G2/M checkpoint, unfolded protein response (UPR), and inflammatory response genes. g UMAP visualization of 24,291 fixed, cryo, and fixed/cryo mouse lung cells, colored by 20 cell populations. h Comparison of cell population proportions between fixed (n = 10,256), cryopreserved (n = 8609), and fixed/cryopreserved cells (n = 5426). The top figure shows the difference in cell population proportions between fixed, cryo, and fixed/cryo samples, and the bottom figure shows the results of the compositional cell analysis using the Bayesian model scCODA. Credible changes and Log2FC are indicated. i Differential gene expression analysis across conditions: fixed vs cryo (top-left), fixed/cryo vs cryo (top-right), and fixed/cryo vs fixed (bottom). The top differentially expressed genes (DEGs) with significantly adjusted p-values (FDR) < 0.05, upregulated (red), and downregulated (blue) with Log2FC > |1| are indicated

We confirmed the absence of DSP-fixation biases after cryopreserving fixed samples, as indicated by a high overlap (> 70%) of HVGs across all three protocols (Fig. S4e). In addition, both PCA and UMAP dimensionality reduction plots showed no discernible biases between preservation protocols (Fig. 4d). We also observed highly comparable expression profiles and gene programs when correlating the mean gene expression values for all protocol comparisons (R2 > 0.99, p < 2.2e−16) (Fig. 4e). Moreover, there was no appreciable alteration in biological processes at the gene program or population level when comparing across protocols (Fig. 4f; Fig. S4f).

We next analyzed 24,291 mouse lung cells processed with the three different protocols and annotated 20 lung and tissue-resident immune cell populations (Fig. 4g; Fig. S4g), and all cell types and states were found across the three conditions at similar proportions. However, we observed slight changes in composition, with fixed/cryo samples showing a decrease of B cells coupled with an increase of gCap compared to the cryo sample, and an increase of Monocytes compared to fixed-only. Comparing fixed-only with cryo, we found an increase in arterial and pneumocyte type I cells compared to the cryo sample (Fig. 4h). Additionally, we observed downregulation of genes associated with immune function (e.g., Igkc, Ccl4, Scgb1a1) in the fixed/cryo, explained by the aforementioned shift in cell type composition. Importantly, the cryo-only sample showed upregulated genes related to stress response, such as Fosb (Fig. 4i; Additional file 4: Table 3). A closer inspection of the different cell populations validated the expression of stress-related genes across all cells in the cryo-only compared fixed/cryo samples, specifically in non-immune cells (Additional file 9: Table 8). Accordingly, GSEA detected an enrichment of regulatory or response pathways for almost all cryopreserved cell types compared to fixed/cryo samples (Additional file 10: Table 9). These results support the feasibility of cryopreservation after fixation to combine the robustness and logistical advantages of the respective methods for scRNA-seq experiments.

Minimization of technical artifacts in FixNCut tissue samples

Fixing tissues after sample collection preserves the natural state of a cell and avoids technical biases, previously shown to affect bulk and single-cell transcriptomics analysis [2,3,4, 22]. In addition to the abovementioned differences in stress-response genes, we further aimed to demonstrate the ability of FixNCut to preserve gene expression profiles by examining previously identified artifact signatures. Specifically, we investigated condition-specific gene signatures from published studies using our mouse lung and colon data (see the “Methods” section).

After analyzing mouse lung samples, we found that fixed samples had comparable dissociation/temperature-signature scores, except for the warm collagenase which was significantly lower (p < 4.3e−07) compared to fresh samples (Fig. 5a). We observed that external tissue stressors had a greater impact on fresh lung resident cells compared to the infiltrating immune cell fraction (Fig. 5b). Interestingly, the signature scores for these populations displayed a bimodal-like distribution, indicating an uneven effect within cell populations (Fig. 5b). Additionally, stress-signature genes were not only found to be differentially expressed in the fresh lung but also in the fixed samples, regardless of their level of expression (Fig. 5c).

Fig. 5

Minimization of technical artifacts using FixNCut protocol on mouse tissues. This figure shows the impact of various dissociation-induced gene signature scores, including dissociation on mouse muscle stem cells [22], warm dissociation on mouse kidney samples [3], and warm collagenase dissociation on mouse primary tumors and patient-derived mouse xenografts [2], across mouse tissues and processing protocols used. All statistical analyses between protocols were performed using the Wilcoxon signed-rank test; significance results are indicated with the adjusted p-value, either with real value or approximate result (ns, p > 0.05, *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001). a Violin plots of dissociation-induced gene signatures scores for fresh and fixed mouse lung. b Score of warm collagenase gene signature for fresh and fixed mouse lung samples across cell populations. c Overlap of differentially expressed genes in the fresh and fixed mouse lung sample with genes from the three dissociation-induced signatures. d Violin plots of dissociation-induced gene signatures scores for fresh and fixed mouse colon. e Score of warm collagenase gene signature for fresh and fixed mouse colon samples across cell populations. f Overlap of differentially expressed genes in the fresh and fixed mouse colon sample with genes from the three dissociation-induced signatures. g Violin plots of dissociation-induced gene signatures scores for cryo and fixed/cryo mouse lung. h Score of warm collagenase gene signature for cryo and fixed/cryo mouse lung samples across cell populations. i Overlap of differentially expressed genes in the cryopreserved and Fixed/Cryo mouse lung sample with genes from the three dissociation-induced signatures

Similarly, fixed colon samples showed a significantly larger decrease in dissociation/temperature-stress signature scores compared to fresh samples (Fig. 5d). Here, we also observed a lineage-dependent impact of cell stress; colonocytes were greatly affected with differences in subtypes, whereas immune cells largely escaped stress biases (Fig. 5e). Endothelial and stromal cells suffered the largest dissociation-related stress in the fresh samples, which was drastically reduced upon fixation (Fig. 5e). Moreover, stress-signatures genes were also differentially expressed in the fresh colon sample, while largely absent in the fixed sample (Fig. 5f).

Furthermore, we demonstrated the effectiveness of FixNCut for long-term sample storage by examining the dissociation/temperature-stress signature scores in cryo-only and fixed/cryo mouse lung samples. Our results showed that cryopreserved samples had significantly higher stress-related signature scores compared to fixed/cryo (p < 2.2e−16) (Fig. 5g). Interestingly, the stress signature score for endothelial and stromal cells exhibited a bimodal distribution exclusively in the cryo-only sample, with cells showing larger dissociation-related effects in the same population (Fig. 5h), consistent with previous observations. Over 70% of signature-specific genes were significantly differentially expressed in the cryo-only sample, an even higher proportion compared to fresh lung and colon samples, whereas the fixed/cryo samples had almost no overlapping DEGs (Fig. 5i).

Moving towards the use of FixNCut on clinical samples

As a proof-of-concept for a multi-center clinical research study focused on autoimmune diseases, we evaluated the performance of FixNCut on human patient biopsies. To this end, we obtained fresh colon biopsies from two IBD patients in remission. The biopsies were mixed and split into four aliquots, which were processed as follows: fresh, fixed-only, cryo-only, or fixed/cryo. The fixed human colon samples exhibited a similar proportion of reads mapped to the reference genome and exonic regions as mouse colon tissues (Fig. 6a) and displayed comparable library complexity for short-term (fixed vs. fresh) and long-term (fixed/cryo vs. cryo) conditions, considering the total number of detected genes (95% CI 6.29e−05 ± 2.03e−05 fresh vs. 5.53e−05 ± 2.13e−05 fixed; 6.65e−05 ± 2.23e−06 cryo vs. 6.10e−05 ± 2.11e−05 fixed/cryo) and captured UMIs (95% CI 0.48 ± 0.07 fresh vs. 0.37 ± 0.08 fixed; 0.47 ± 0.07 cryo vs. 0.39 ± 0.08 fixed/cryo) (Fig. 6b). A similar pattern was observed comparing the number of captured genes and UMIs at the cell level (Fig. S5a). The cumulative detected gene count was highest for the fixed biopsy, with fresh being the worst condition (Fig. 6c). Genes identified across all libraries (n = 21,637) were mostly protein-coding genes (70%), followed by non-annotated genes (21%). Genes exclusively captured in one condition (fresh, fixed, cryo-only, or fixed/cryo) showed a highly similar distribution of gene features (Fig. S5b). After removing low-quality cells, the main quality control metrics had similar distributions, with slightly improved median UMI, gene counts, and reduced mitochondrial gene percentage for the fixed sample (Fig. S5c), which were consistently observed across almost all populations (Fig. S