A high throughput PIXUL-Matrix-based toolbox to profile frozen and FFPE tissues multiomes

Abstract

Large-scale high-dimensional multi-omics studies are essential to unravel molecular complexity in health and disease. We developed an integrated system for tissue sampling (CryoGrid), analytes preparation (PIXUL), and downstream multiomic analysis in a 96-well plate format (Matrix), MultiomicsTracks96, which we used to interrogate matched frozen and FFPE (formalin fixed paraffin embedded) mouse organs. Using this system, we generated 8-dimensional omics datasets encompassing 4 molecular layers of intracellular organization: epigenome (H3K27Ac, H3K4m3, RNA polymerase II, and 5mC levels), transcriptome (mRNA levels), epitranscriptome (m6A levels), and proteome (protein levels) in brain, heart, kidney, and liver. There was high correlation between data from matched frozen and FFPE organs. The Segway genome segmentation algorithm applied to epigenomic profiles confirmed known organ-specific super-enhancers in both FFPE and frozen samples. Linear regression analysis showed that proteomic profiles, known to be poorly correlated with transcriptomic data, can be more accurately predicted by the full suite of multi-omics data, compared to using epigenomic, transcriptomic, or epitranscriptomic measurements individually.

BACKGROUND

Multi-omics is a relatively new discipline that aims to reverse engineer biological systems by a) acquiring large molecular datasets for different ome layers of intracellular organization ¹ and b) computationally integrating these heterogeneous datasets to gain a deeper understanding of phenotypes ². Multi-omic analysis include genomics (DNA sequence and structure)³, epigenomics (chromatin modifications and structure)⁴, transcriptomics (RNA sequence and structure)⁵, epitranscriptomics (posttranscriptional RNA modifications)⁶, proteomics (protein interaction, composition, structure)⁷, and metabolomics (interactions and composition of small molecules)⁸ datasets. The exponential growth of the multi-omics field has been fueled, in part, by advances in high throughout analytical technologies such as next generation sequencing, mass spectrometry, and others ^1,9,10.

Freezing is the preferred way to store and transport biospecimens for research ^11–13. In clinical settings, formalin fixation and paraffin embedding (FFPE) is the most commonly used method for longitudinal tissue specimen storage. FFPE blocks are also commonly used to transport clinical samples for molecular diagnosis ¹⁴. All clinical laboratories in the US are required to hold onto diagnostic tissue blocks (mostly FFPEs) for several years, greatly multiplying the number of human tissue specimens in archives. The estimated hundreds of millions of FFPE tissue samples acquired over the years provide a vast resource for the discovery of disease molecular pathways linked to histopathology. Still, these biospecimens remain highly underutilized in the multi-omics field ^15,16 because sample and analyte preparation methods are inefficient, slow, labor intensive, and low throughput ¹⁶, let alone the challenge of using a single biospecimen to generate multidimensional profiles. Thus, better tools and methods are needed to retrieve chromatin, DNA, RNA, and protein from biospecimens such as frozen and FFPE tissues to advance omics research in health and disease.

We have previously developed a 96-well microplate sonicator, PIXUL, that offers unmatched sample preparation throughput capabilities for a broad range of high throughput analytical workflow applications ^17–19. Here, we took advantage of mouse organs’ epigenomic (several chromatin modifications), transcriptomic (RNA levels), epitranscriptomic (RNA modification), and proteomic (protein levels) differences as substrates to develop systems that integrate tissue sampling (CryoGrid²⁰) with 96-well plate format sample preparation, PIXUL¹⁷, and analytical, Matrix ^21–23, platforms for multiome profiling of FFPE and frozen tissues.

MATERIALS, DEVICES, AND METHODS SUMMARY

Details of each section listed below are provided in the Supplement.

Materials.

Buffer recipes are listed in the Supplement. Hardware/labware (Table S1) and kit/enzyme (Table S2) catalog numbers and commercial suppliers are listed in the supplementary tables.

Devices.

The CryoGrid system ²⁰,was used to cryostore and sample tissue and PIXUL¹⁷ multi-sonicator was used to extract and prepare chromatin, DNA, RNA and protein.

Methods.

Frozen and FFPE mouse organ tissues.

Post-mortem brains, hearts, kidneys, livers, lungs, and muscle were used from 12-week-old C57bl/6 mice. Tissues were stored in CryoTrays at −80°C (Frozen)²⁰. FFPE blocks were prepared from matched frozen tissues as described in ²⁴ and sectioned using a microtome. Prior to analytes extractions, FFPE curls were deparaffinized and rehydrated²⁵.

Analysis of chromatin, methylated DNA, RNA, methylated RNA, and protein.

Chromatin (histone modifications and RNA polymerase II (Pol II)):

Chromatin immunoprecipitations in 96-well microplates (Matrix-ChIP) was followed by either qPCR/PCRCrunch analysis ²³ or library production, sequencing, generation BAM and bigwig files, and analysis ²⁶.

Methylated DNA:

Methylated DNA immunoprecipitation in 96-well microplates (Matrix-MeDIP²²) was followed by library production, sequencing, generation BAM and bigwig files analysis.

RNA:

Reverse transcription was done in 96-well microplates (Matrix-RT) ²⁰. cDNAs were analyzed either by qPCR/PCRCrunch or library sequencing²⁰.

Methylated RNA:

Methylated RNA immunoprecipitation in 96-well microplates (Matrix MeRIP) was followed by reverse transcription (²⁰. cDNAs were analyzed by library sequencing ²⁰.

Proteins.

Proteins isolated from tissue fragments were digested using the SP3 protocol ²⁷, labeled with 16-plex TMTpro and quantified by LC-MS/MS on Orbitrap Exploris480 ²⁸.

Machine learning algorithm.

Segway genome segmentation ^29,30 automatically segments and annotates the genome based on multiple tracks of functional data. Segway was used on FFPE and frozen multi-omics profiles to generate segmentations and annotations for mouse brain, heart, kidney, and liver tissues.

RESULTS

A high-throughput system for storing and sampling of frozen tissues (CryoGrid²⁰) integrated with PIXUL-based RNA and chromatin analyses in 96-well plate-based assays (Matrix) (Fig.1).

Fig.1. CryoGrid system integrated with PIXUL: A high-throughput system for rapid sampling of frozen tissue for histology, transcript, and epigenetic analysis.

A. CryoTray with multiple frozen mouse organs in CryoBlock (<−70°C). Each row contains organs from the same mouse; from left to right, brain, heart, kidney, liver, lung, and muscle. A single tray holds all six organs from four mice. CryoCore is a battery-operated, easy-to-use tool to reproducibly extract tissue cores (1–2mm long and 1mm in diameter) from frozen tissues. Left, cores were jetted into wells of 96-well PIXUL plates, crosslinked, and sonicated in PIXUL. PlateHandle (black), next to PIXUL, is a hand-held tool for picking up plates, thus avoiding contamination and the need to handle plates directly with hands. Sheared chromatin was used in Matrix ChIP. Center, cores were used to prepare FFPE blocks for H&E staining. Right, protocol for RNA extraction. B. CryoCore cores from CryoTray – mouse frozen brain, heart, kidney, liver, lung, and muscle stained with H&E (Haemotoxylin and Eosin). C. Cartoons of the genes. Blue arrows show positions of qPCR primers used in ChIP and green arrows show the positions of qPCR primers used in RT-qPCR. D. CryoCore-PIXUL extracted RNA was assayed in Matrix RT-qPCR using primers to 3’ ends of the genes. E. Cores from the frozen organs were cross-linked and sonicated. PIXUL-sheared chromatin samples were analyzed for RNA Polymerase II CTD (Pol II CTD) (proxy readout for transcription), H3K4m3, H3K27Ac, and histone H3 levels at indicated organ-specific genes using Matrix ChIP-qPCR. ChIP DNA was analyzed by qPCR, expressed as a fraction of input. In-house PCRCrunch software tool was used to download, analyze, and plot results. Data represent mean+SEM (n=4 for each frozen organ) expressed as a fraction of input. Syn1, brain-specific gene; Tnnt2, heart-specific gene; Fxyd2, kidney-specific gene; Alb, liver-specific gene; Sftpa1, lung-specific gene; and Tnnc2, muscle-specific gene.

We used mouse brain, heart, kidney, liver, lung, and muscle fragments to integrate and optimize CryoGrid and CryoCore with PIXUL protocol to extract RNA and chromatin from frozen tissue samples (Fig.1A). One core (1–2mm) was used for each of the following: H&E histology (Fig.1B), RNA isolation followed by RT-qPCR, and ChIP (Matrix ChIP-qPCR). In Matrix-RT and Matrix-ChIP assays, we used primers to genes specific to brain (Syn1), heart (Tnnt2), kidney (Fxyd2), liver (Alb), lung (Sftpa1), and muscle (Tnnc2) to assess the quality of the methods (Fig.1C). Matrix ChIP was done using antibodies to RNA polymerase II (Pol II), histone H3, H3K27Ac, and H3K4m3. In our previous work we have found that Pol II ChIP signal intensity at 5’ ends of genes recapitulates mRNA levels ^17,21,23. Consistently, results of CryoGrid/CryoCore-based assays demonstrated excellent agreement between the levels of mRNAs, Pol II, and permissive H3K27Ac and H3K4m3 histone modifications assayed in these organs (Fig.1D–E). We used CryoGrid-PIXUL frozen organs data as a benchmark for PIXUL-FFPE method development (below).

A high-throughput system for extracting soluble chromatin, DNA, and RNA from FFPE tissue blocks (Fig.2).

Fig.2. PIXUL-FFPE-Matrix-ChIP and mRNA analysis at mouse organ-specific genes.

A. PIXUL-FFPE-Matrix-ChIP-qPCR and PIXUL-FFPE-mRNA-qPCR protocol. FFPE slices (5μm) were generated from FFPE blocks using a microtome, total 8 series for each organ. After deparaffinization, EtOH rehydration, and heat retrieval, samples were treated in PIXUL, and chromatin (blue) and RNA (green) were isolated. B. FFPE blocks of mouse brain, heart, kidney, liver, lung, and muscle. C. Cartoons of the genes. Blue arrows show positions of qPCR primers used in ChIP-qPCR and green arrows show positions of qPCR primers used in RT-qPCR. D. Extracted RNA was assayed in Matrix RT-qPCR using primers to 3’ ends of the genes. E. Extracted chromatin was analyzed in Matrix ChIP-qPCR using antibodies to Pol II CTD, H3K4m3, H3K27Ac, and histone H3. Mouse IgG was used for background subtraction. Inputs were diluted 20X to overcome PCR interference. Results (expressed as a fraction of input) represent mean±SEM (n=8 different FFPE extractions).

We used mouse brain, heart, kidney, liver, lung, and muscle FFPE blocks to develop and optimize the procedure for extracting analytes from FFPEs (Fig.2A). Standard formalin fixation (10% formalin for 24hrs) and paraffin embedding procedure was used to generate mouse organ FFPE blocks (Fig.2B). A microtome was used to cut FFPE slices. We began optimizing the methods with a standard protocol to deparaffinize FFPE curls with xylene, ethanol-H₂O rehydration, and heat retrieval ¹⁶ followed by ultrasound shearing of soluble chromatin in PIXUL. The quality of soluble chromatin was assessed in Matrix ChIP using antibody to RNA polymerase II (Pol II) with IgG used as a background control, with the expectation that Pol II levels will be highest at genes expressed in a given organ. Primers were designed to genes specifically expressed in brain (Syn1), heart (Tnnt2), kidney (Fxyd2), liver (Alb), lung (Sftpa1), and muscle (Tnnc2) (Fig.2C). We found that input DNA generated from FFPE blocks contains PCR inhibitor(s). It has been reported that extracts from FFPEs can inhibit PCR due to small DNA degradation artifacts ³¹. To diminish this inhibition, we found that dilution of input samples by 20-fold minimized the qPCR interference. We tested a range of specimen slice thicknesses from 5 to 20μm and found that, for all the organs, 5μm worked well in PIXUL-ChIP assays while preserving more FFPE material compared to 10μm (Fig.S1A–B). We found that for some organs the Pol II ChIP signal could be noisy but subtracting IgG ChIP background signal greatly improved the consistency of PIXUL-FFPE results, a correction that has been included in the protocol.

Xylene solvent has been traditionally used to deparaffinize FFPE blocks, including in epigenetics studies ^16,32,33. However, it is toxic and, as such, hard to work with and less suitable for future PIXUL-FFPE automation. SafeClear is an organic, nonhazardous, nonflammable clearing and deparaffinization Xylene substitute (ThermoFisher). Fig.S2A–B show results of PIXUL-ChIP analysis using SafeClear compared to Xylene. Both solvents yield comparable tissue-specific patterns of permissive histone marks, H3K27Ac and H3K4m3, at 5’ ends of organ-specific genes, similar to Pol II profiles (Fig.S2B). As expected, histone H3 profiles were the same for each organ. These results demonstrate that both solvents perform similarly; therefore, SafeClear is used to deparaffinize blocks in the PIXUL-FFPE workflow.

Heating, typically at 95°C, is considered a critical step for retrieval of analytes from FFPE samples ^34,35. And yet, heating can accelerate analyte degradation. Therefore, we tested temperatures lower than the standard 95°C for soluble chromatin retrieval. We compared chromatin extraction from rehydrated FFPEs at 95°C, 85°C, and 75°C. Samples were transferred to 96-well plates, treated with ultrasound in PIXUL, and analyzed in Matrix ChIP (Fig.S3A). These experiments showed that incubation at 95°C was more efficient than 85°C (Fig.S3B) and 75°C (not shown). We chose to use 95°C for heat retrieval of chromatin. Duration of incubation at 95°C was also important. We found that for ChIP analysis of extracted chromatin (Pol II, H3K4m3, H3K27Ac, and H3) 20min was better than 10min (Fig.S4A–B) while 5min yielded poor results.

We wondered if deparaffinization using SafeClear at higher temperature would improve the efficiency of chromatin extraction. We compared standard deparaffinization at room temperature (10min) which includes manual homogenization to 50°C (3min) without homogenization and found similar results in Matrix ChIP (Pol II, H3K4m3, H3K27Ac and H3) (Fig.S5A–B). We chose to carry out deparaffinization at 50°C to save time and eliminate the need for initial homogenization via pipette tip.

Some protocols include RNase treatment to increase chromatin recovery from rehydrated FFPEs ^16,36,37. To test this, we incubated samples with or without RNase A (10μg/mL) for 30min at room temperature before the 95°C heating step. We found that application of RNase to rehydrated FFPEs did not make a difference in Matrix ChIP results (Pol II, H3K4m3, H3K27Ac and H3) (Fig.S6A–B). We chose not to include RNase treatment.

PIXUL uses polystyrene (PS) plates ^17,18. For protocols that use high temperature incubation, 96-well polypropylene (PP) plates would be more suitable (PP is more heat resistant and has lower unspecific binding background). Fig.S7 demonstrates that using 96-well PP plates in the PIXUL-FFPE workflow yields ChIP-qPCR results similar to those using 96-well PS plates (compare to Fig.S6). Fig.S8 shows that substituting test tubes with 96-well PP plates in the heating step and then using the same plate for sonication in PIXUL yields the same ChIP-qPCR outcomes. These results demonstrate that using 96-well PP plates increases throughput of PIXUL procedure and simplifies workflow.

FFPE blocks can be stored at room temperature for several years ^16,38. We found that soluble chromatin retrieved from 1.5-year-old FFPE blocks of mouse organs yielded Matrix ChIP results indistinguishable to those results obtained earlier from the same freshly prepared FFPE blocks (Fig.S9A–B). The blocks used in all the NGS studies were 1.5 years old (below).

We modified the PIXUL chromatin protocol for extraction of RNA from FFPEs. After incubation in extraction buffer at 95°C for 20 min, samples were transferred to 96-well plates and treated with ultrasound in PIXUL. Samples were transferred to 1.5mL tubes for standard TRizol RNA isolation protocol ³⁹ followed by DNase treatment to eliminate residual genomic DNA. Finally, RNA was cleaned up either by EtOH precipitation or by Zymo clean up column (Table S2).

Fig.2A outlines final protocols of chromatin and RNA extraction from mouse brain, heart, kidney, liver, lung, and muscle FFPEs (Fig.2B) that were analyzed for transcript levels and epigenetic marks (Fig.2C). As in case of frozen organs (Fig.1), Matrix-RT-qPCR (Fig.2D) and Matrix-ChIP-qPCR (Fig.2E) data for organ-specific genes show that mRNA expression matches the pattern of Pol II and permissive histone marks (H3K4m3 and H3K27Ac) (Fig.2D–E). These comparisons serve to validate PIXUL-FFPE protocols.

The detailed workflows for extraction of RNA, DNA, and chromatin are shown in Fig.S10 for frozen tissues and in Fig.S11 for FFPEs. These protocols were next adapted for next generation sequencing.

PIXUL-Matrix-RNA-seq analysis of FFPE and frozen mouse tissues.

RNA extracted with PIXUL from mouse FFPE (1.5-year-old) and frozen organs were used to generate sequencing libraries (Fig.3A). Correlation scatterplots ¹⁷ (deepTools) of FFPE vs frozen tissue BAM data showed Spearman correlation coefficients as follows: brain 0.91; heart 0.83; kidney 0.87; and liver 0.85 (Fig.S13A). IGV transcript snapshots shown in Fig.3B recapitulates results of RT-qPCR analysis of organ-specific transcripts in frozen (Fig.1D) and FFPE (Fig.2D) tissues. Housekeeping genes Actb and Rpl32 are expressed in all four organs. The third intron of Rpl32 gene encodes small nucleolar RNA, Snord68, that is also expressed in all four organs. Results of RNA-seq analysis further validate the PIXUL-FFPE RNA extraction method, as high-quality RNA can be extracted from FFPE specimens that were stored at room temperature for a long time.

Fig.3. RNA-seq and methylated RNA (m6A)-seq of FFPE vs. frozen mouse organ tissues analyzed by PIXUL-Matrix-RNA-seq and PIXUL-Matrix-methylated RNA immunoprecipitation (MeRIP), respectively.

A. RNA was isolated from frozen (Fig.1) and FFPE (Fig.2) mouse tissues. For MeRIP RNA was fragmented using PIXUL and immunoprecipitated with m6A antibody. Libraries were prepared from precipitated RNA and sequenced. B. Transcript sequencing reads profile snapshots were generated from BAM files using Integrative Genomics Viewer (IGV) of organ specific genes, brain: Syn1; heart: Tnnt2; kidney: Fxyd2; liver: Alb; and housekeeping genes: Actb and Rpl32. C. MeRIP-seq analysis of RNA isolated from mouse liver, kidney, heart, and brain. Anti-m6A antibody was used in 96-well plate Matrix-MeRIP to immunoprecipitate methylated RNA. Same genes as in B are shown. D. IGV snapshot demonstrating that snoRNAs are not m6A methylated. Snord68 and Mir7079 are small nucleolar and micro RNAs respectively.

Methylated RNA immunoprecipitation (PIXUL-Matrix-MeRIP)-seq analysis of FFPE and frozen mouse tissues.

RNA modifications have been known for decades ^40,41. With the introduction of MeRIP the number of studies to examine mRNA base modifications have increased, as they are thought to control transcripts’ translation and processing, a regulation called epitranscriptomics. m6A is one of the most common, reversible epitranscriptomic modifications ^42,43. We have previously developed methylated DNA immunoprecipitation assay in 96-well plates (Matrix-MeDIP) ²² which we adapted here for MeRIP. RNA isolated from fresh frozen and FFPE tissues was further fragmented in PIXUL and immunoprecipitated with anti-m6A antibody bound to walls of 96-well plate via Protein A. After washes, MeRIP RNA was eluted and used to generate sequencing libraries (Fig.3A). Correlation scatterplots of FFPE vs frozen tissue BAM data showed Spearman correlation coefficient as follows: brain 0.85; heart 0.53; kidney 0.65; and liver 0.56 (Fig.S13B). Fig.3C shows the IGV m6A MeRIP-seq snapshot for the mouse brain, heart, kidney, and liver for organ-specific and housekeeping genes. MeRIP-seq looks very similar to RNA-seq (compare Fig.3B with 3C) except that there is no signal for the small nuclear RNA Snord68, suggesting that this RNA is not m6A modified. We found several other small nucleolar RNAs (snoRNAs) that are likewise not m6A modified, some of which are shown in Fig.3D. To our knowledge, the lack of m6A base modification in small nucleolar RNAs has not been previously reported.

FFPE and frozen mouse tissues PIXUL-Matrix-ChIP-seq (Fig.4).

Fig.4. Results of H3K27Ac, H3K4m3, and Pol II Matrix-ChIP-seq and Matrix-MeDIP-seq analyses of frozen and FFPE mouse organs.

A. Chromatin and DNA was prepared from frozen and FFPE mouse tissues and sheared using PIXUL. B-D. Matrix-ChIP assays were carried out using chromatin preparations and anti-H3K27Ac, anti-H3K4m3 and anti-Pol II CTD antibodies. ChIP DNA was isolated and libraries were prepared and sequenced. Genomic snapshots were generated from bigwig files using IGV for H3K27Ac, H3K4m3 anti-Pol II CTD profiles at organ specific genes, brain: Syn1; heart: Tnnt2; kidney: Fxyd2; liver: Alb; and housekeeping genes Actb and Rpl32. E. DNA from organs was sheared and denatured to yield ssDNA. Matrix-MeDIP in 96-well plate was carried out using anti-5mC antibody. Libraries were prepared and sequenced. Fastq files were processed as in B-D. 5mC snapshots for the tandem 4.5S RNA loci were generated from bigwig files using IGV.

We used PIXUL-isolated chromatin from mouse FFPEs (1.5years old) and frozen organs in Matrix ChIP using antibodies to H3K27Ac (Fig.4B), H3K4m3 (Fig.4C), and RNA polymerase II (Pol II) (Fig.4D). Scatterplots of FFPE vs frozen tissue BAM data showed good agreement with Spearman correlation coefficient as follows: H3K27Ac – brain 0.75; heart 0.82; kidney 0.80; and liver 0.79; (Fig. S13C); H3K4m3 – brain 0.76; heart 0.76; kidney 0.73; and liver 0.64 (Fig.S13D); Pol II CTD– brain 0.77; heart 0.74; kidney 0.76; and liver 0.69 (Fig.S13E). IGV snapshots shown in Fig.4B (H3K27Ac), Fig.4C (H3K4m3 and Fig.4D (Pol II) recapitulate frozen (Fig.1D) and FFPE (Fig.2D) ChIP-qPCR results at loci that encode organ-specific transcripts. Results of ChIP-seq analysis serve to validate the PIXUL-FFPE chromatin extraction method and demonstrate the usefulness of FFPE specimens in epigenetic analysis even after long-term storage at room temperature.

PIXUL-Matrix-MeDIP-seq analysis of FFPE and frozen mouse tissues.

Methylation of the 5th position of cytosine (5mC) is the most common DNA modification ⁴⁴. We adapted PIXUL protocols to extract DNA from FFPEs as follows (Fig.4A). Scatterplots of FFPE vs frozen tissue BAM data showed excellent agreement with Spearman correlation coefficient as follows: brain 0.89; heart 0.92; kidney 0.87; and liver 0.90 (Fig.S13F). Sequencing analysis reveals alternating DNA methylation pattern at 4.5S RNA loci corresponding to CpG islands (Fig.4E). Analysis of the RNA-seq, H3K4m3-seq, H3K27Ac-seq, and Pol II-seq data corresponding to 4.5S loci uncovered peaks of these marks in regions that are not methylated (Fig.S14), suggesting that these permissive marks and 5mC are mutually exclusive ⁴⁵. This data also show that like snoRNAs (Fig.3C–D) 4.5S RNA appears not to be m6A modified. These Matrix-MeDIP-seq data validate the PIXUL-based DNA isolation method from FFPEs and demonstrate high quality of isolated DNA.

PIXUL-LC-MS/MS in FFPE and frozen mouse tissues.

To the best of our knowledge, there are no studies that combine proteomic measurements with epigenomic, transcriptomic, and epitranscriptomic assays in the same biospecimens, let alone FFPEs. To mitigate sample bias and batch effects ⁴⁶, our goal was to develop a 96-well format proteomic workflow aligned with all the other omics. (Figs.5A and S13). Comparable numbers of proteins were identified and quantified in FFPE vs. frozen organs: brain 5267 vs. 5182; heart 5321 vs. 4978; kidney 5292 vs. 5162 and liver 5299 vs. 5076 (Fig.5B). There was a high correlation of protein abundances between FFPE and matched frozen organs: brain 0.969; heart 0.947; kidney 0.957; and liver 0.953 (Pearson R²) (Fig.5C). In agreement with large-scale proteomic studies ⁴⁷ Venn diagrams illustrate that the number of organ-specific proteins ranged between 24–90 or 0.5–1.9% of the total number of a given organ (Fig.5D). These results demonstrate that proteomic analysis can be combined with epigenomic, trancriptomics and epitransciptomic assays from the same samples using the PIXUL.

Fig.5. Protein extracted from FFPE and frozen tissue analyzed by PIXUL-LC-MS/MS.

A. PIXUL-based protocol was used to isolate proteins from matched FFPE and frozen mouse organs. CryoCore was used to sample frozen tissues and microtome to sample FFPEs. Proteins were extracted (Methods) using PIXUL and trypsin-digested using SP3 protocol, and TMT-labeled peptides were fractionated and analyzed by LC-MS/MS. B. Total number of proteins identified in organ from frozen and FFPE blocks. C. Scatter plots of comparative analysis of PIXUL-LC-MS/MS generated protein datasets from mouse organs frozen and FFPE blocks. D. Venn diagram comparing proteins identified across the four organs.

The Segway algorithm produces annotations of the genome based on ChIP-seq and MeDIP-seq datasets.

Segway ³⁰ is a dynamic Bayesian network model that takes as input heterogeneous genome-wide measurements and coalesces them into a segmentation of the genome and an associated set of labels. Here we ran Segway across multi-tissue multi-omics profiles (ChIP-seq: H3K27Ac/H3K4m3/Pol II and MeDIP-seq: 5mC datasets) on FFPE and frozen tissues separately, and we generated genome segmentations and annotations for mouse brain, heart, kidney, and liver tissues (Fig.6A). Genomes in each tissue were annotated with 15 labels, each representing a specific type of genomic function (e.g., promoters and enhancers). Specifically, we identified one segmentation label for FFPE and frozen samples each that is enriched for previously annotated super-enhancers (dbSUPER, ⁴⁸) (Fig.6B). Interestingly, Segway’s organ-specific annotation of super-enhancers agrees with previous analyses ⁴⁸, suggesting that multi-omics data generated by PIXUL-Matrix-ChIP/MeDIP-seq can reveal organ-specific epigenomic regulation. This result also demonstrates that FFPE measurements alone can be used to capture organ-specific regulatory elements. We further derived a set of super-enhancers that are predicted to only occur in one organ and are identified in both FFPE and frozen samples (brain 3342; heart 1059; kidney 1642; and liver 3029, Fig.6C). Thus, in addition to re-identifying enhancers that are already annotated in dbSUPER, Segway is able to generate new hypotheses by predicting a set of new organ-specific super-enhancers based on multi-omics profiles.

Fig.6. Tissue-specific super-enhancer prediction by Segway and comparison to dbSUPER’s tissue-specific super-enhancer annotations.

A. Segway protocol. ChIP-seq: H3K27Ac/H3K4m3/Pol II and MeDIP-seq: 5mC datasets were used. A Segway model is trained on frozen and FFPE samples separately and annotates the four genomes into 15 types of genome segments, one of which are enriched for super-enhancers (pink region). B. For each organ, an F1 score based on precision and recall with respect to dbSUPER’s super-enhancer annotations was calculated. Dotted lines show F1 scores from a random baseline. dbSUPER’s tissue-specific super-enhancers were best captured (with the highest F1 score) by Segway’s annotation in the corresponding tissue compared to unrelated tissues. C. Numbers of predicted organ-specific super-enhancer sites shared in both frozen and FFPE organs and numbers of these sites also found in dbSUPER database.

Predicting protein abundance using multi-omics profiles.

Tissue protein and mRNA levels for a given gene are not well correlated ^47,49,50. For instance, in the ageing kidney changes in protein levels can occur without changes in cognate mRNAs ⁵¹. Because there is a large discrepancy between transcriptomic and proteomics expression measurements (Fig.7A), we further investigated whether multi-omics profiles can provide a more accurate predictor of protein expression. To do that, we trained linear regressors based on different combinations of multi-omics profiles to predict protein expression levels across organs and samples. We found that protein expression is best predicted when using both DNA- and RNA- based predictors (Fig.7B–C, Fig.S15). In addition, compared to RNA expression, m6A RNA methylation level is a better indicator of protein expression (Fig.7D, Fig.S16). Furthermore, combining epitranscriptomic measurements with transcriptomics markedly improves protein expression prediction (Fig.7E). These findings suggest that epigenomic and epitranscriptomic assays provide complementary information to transcriptomics measurements in relation to protein quantities, demonstrating the potential of integrating multi-omics measurements to model proteomics profiles.

Fig.7. Protein expression prediction using multi-omics profiles.

A. Pairwise Pearson correlation between observed multi-omics datasets profiles. B. Pearson correlation between predicted and observed protein expression, separated by organ and sample prep. Protein expression is predicted by several sets of multi-omics profiles: “All”: all other seven omics types in this study; “RNA + m5C + m6A”: transcriptomic and epitranscriptomic profiles; “RNA”: transcriptomic profiles; “All – RNA – m5C – m6A”: epigenomic profiles. C-E. Violin plots comparing the performance of protein expression prediction across different sets of predictors. Statistical significance is calculated through one-sided paired Wilcoxon signed-rank test on the Pearson correlation coefficients across cross-validation folds, organ, and sample prep. P-values were subjected to Benjamini-Hochberg correction.

DISCUSSION

Progress in high throughput and efficient extraction of analytes from biospecimens has been slow, and multi-omics studies have been limited to selected cultures or tissues. To facilitate multiome analysis, we used FFPE and frozen blocks from mouse organs to develop PIXUL ultrasound-based 96-well-format methods to extract chromatin, DNA, RNA, and proteins from these biospecimens. We integrated these sample preparation protocols with downstream multi-omic analysis platforms, and in so doing, made these workflows, which we call MultiomicsTracks96, useful for high throughput multiome analysis of FFPE and frozen biospecimens.

FFPE tissue blocks for nucleic acids analysis.

For decades, FFPE blocks were used primarily for histopathology until the late 1980s, when DNA ⁵² and RNA ⁵³ were first isolated from FFPEs and used in PCR ^53–55. Given the enormous potential of FFPEs in basic research and clinical applications ¹⁵, many companies have developed kits to extract DNA, RNA, or both from these biospecimens. These commercially available tools use Proteinase K, heating, sonication, and silica adsorption techniques developed by the early pioneers of the field ⁵⁶. Comparative analysis revealed that no single kit was superior to the others, and all performed well in downstream next generation sequencing applications (NGS) ^57–60 that showed high concordance with paired fresh frozen tissues ^60–62. FFPE as old as 12 years can be efficiently used for NGS analysis ¹⁵. The 96-well format PIXUL-Matrix method also demonstrated high RNA-seq concurrence between FFPE and frozen mouse organs tissues (Fig.S13A).

FFPE blocks for proteomic analysis.

In 2005, Hood et al. tested whether archived FFPE could be used for proteomic analysis and isolated peptides representing several hundred proteins from prostate cancer and benign prostate hyperplasia ⁶³. More recently, FFPEs were tested in large-scale proteomic analysis, taking advantage of either ultrasound ^64,65 or pressure cycling technology ⁶⁶. These latest studies demonstrated that FFPEs as old as 15–20 years can be used effectively for proteomic analysis even though there is some decrease in peptide identification compared to newer samples ⁶⁵. Correlation of paired fresh frozen and FFPE tissues in these studies was high with R²>0.9 ^64,66. PIXUL-LC-MS/MS analysis yielded R² ≥0.95 (Fig.5C).

FFPE blocks for epigenetic analysis.

In 2010, Fanelli et al. reported an improved method for epigenomic analysis from FFPEs, which they termed PAT-ChIP ³⁷. More recently there has been significant progress made using these biospecimens for epigenetic analysis (reviewed ¹⁶) but these methods remain slow. In contrast, the PIXUL-Matrix-ChIP FFPE method is simple and takes less time: starting from a tissue fragment it takes 1 day for qPCR-ready DNA and 1.5 days for sequencing library ready DNA (Fig.S11, Table S5).

FFPE blocks for epitranscriptomic analysis.

Until now, to the best of our knowledge, FFPE tissue blocks have not been used in epitranscriptomics studies. The regulatory role of m6A has been implicated in virtually every step of RNA-mediated gene expression, including splicing, nuclear export, and translation, to name just a few ^42,67. The fundamental role of m6A in gene expression does not appear to be unique. Rather, its role seems to be shared with other RNA base modifications, even though their levels in mRNAs are 10-fold lower (e.g., m1A, m5C) ⁶⁸. Thus, it is conceivable that the combination of mRNA base modifications constitutes an epitranscriptomic code. In this regard, the PIXUL-Matrix-MeRIP platform should greatly facilitate epitranscriptome profiling of a wide range of biospecimens, including FFPEs.

High dimensional multi-omics.

Multi-omics is a nascent discipline in which most studies have included only two omics datasets ¹. For instance, in a 2022 review of the NIH multi-omics grant portfolio, which included metabolomics, only 7% and 2% of projects consisted of four and five omics approaches, respectively ⁶⁹. This study measures 8 different omic datasets, including, for the first time, epitranscriptome profiles spanning 4 layers of intracellular organization. Further, the multidimensional survey covered several organs while comparing frozen and FFPE tissue blocks.

Despite great advances in analytical tools, integrating heterogeneous multi-omics datasets remains computationally challenging ^1,2,70. Technical variation and sample batch effects render the integration of multi-dimensional datasets more difficult ^46,70,71. To mitigate experimental biases, we developed protocols that use a single sample for all the bulk multiome assays. Additionally, we aimed to develop workflows to maximize the number of sample preparation steps shared by the different multi-omics assays.

Computational integration of multi-omics datasets.

In the current study, using the Segway algorithm we demonstrated a use case of predicting tissue-specific super-enhancers based on multi-omics epigenomics measurements. The Segway genome segmentation method can also be applied to identify other types of regulatory elements if there is some prior knowledge to help assign semantics to the automatically inferred Segway labels. The multi-omics profiles, in conjunction with Segway, also have great potential to reveal disease-specific regulatory elements by identifying genomic labels that specifically occur in a group of patients vs. control samples, including drug toxicities.

Here, as a proof of concept, we also demonstrated that epigenomics and epitranscriptomics can be used as predictors for protein quantification. However, the Pearson correlation between predicted and observed protein level is still quite low. This is not entirely surprising, because many translational and post-translational effects are not yet captured, even with this rich, multi-omic analysis.

Challenges.

MultiomicsTracks96 is capable of generating high-dimensional multiome datasets. Still, one of the biggest challenges of multi-omics is the development of computational tools that can integrate such large-scale heterogenous datasets to define the flow of information and characterize multi-omic organization in health and disease, include the effects of drug treatments and environmental perturbations.

There are significant advantages to preserving tissues as FFPE versus freezing, but for molecular analysis FFPE samples are more challenging. We have found excellent agreement between FFPE and frozen tissue for proteomic and DNA methylation measurements (R² >0.9). Further improvements in efficiency of retrieval of soluble chromatin and RNA from FFPE samples are needed to facilitate integration of these heterogeneous datasets.

Availability of data and materials:

Sequence data was deposited in Gene Expression Omibus database under entry GSE230109. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD041462.