Intracellular polarization of RNAs and proteins in the human small intestinal epithelium

Reconstructing the apicome of the human and mouse intestinal epithelium

To map the mRNA and protein apicome in the human and mouse intestinal epithelium, we used LCM, followed by bulk RNA sequencing (RNA-seq) and mass spectrometry proteomics (Fig 1A). We collected the proximal jejunums from 8 patients who have undergone pancreaticoduodenectomy (Whipple surgery) due to pancreatic pathologies (Fig 1B and S1 Table). In these procedures, the proximal small intestine is removed together with the head of the pancreas that harbors a tumor. In these patients, the intestine is considered disease-free [4]. Additionally, we collected the jejunum from 5 healthy adult mice. To identify potential differences in the apicome along the villus axis, we collected samples from the apical and basal compartments, in both the bases and tips of villi (Fig 1A and 1C). Epithelial segments were carefully dissected to avoid inclusion of stroma. Since the basal compartment in human is significantly smaller compared to the apical compartment (width of 15.4 μm versus 6.6 μm, S1A Fig and S2 Table), this resulted in generally lower mRNA yields in the basal compartments (S1B Fig). We therefore normalized the data in each compartment to enable comparison of mRNA and protein concentrations (Methods).

Fig 1. Reconstructing the apicome of the human and mouse intestinal epithelium.

(A) Diagram of experimental setup. LCM was used on human and mouse small intestine tissues. Apical and basal sides of the epithelial cells were collected from the bases and tips of villi, followed by high-throughput RNA-seq and mass spectrometry proteomics. Created with BioRender.com. (B) Hematoxylin and eosin staining of human jejunum, obtained during Whipple surgery. Scale bar, 1 mm. (C) Human villus colored with HistoGene staining solution before (left) and after (right) LCM. Colors indicate collection samples—apical tip (red), apical base (yellow), basal tip (green), and basal base (orange). Scale bar, 150 μm. LCM, laser capture microdissection.

https://doi.org/10.1371/journal.pbio.3002942.g001

Human intestinal epithelium exhibits mRNA polarization

To map the mRNA polarization in the apical and basal compartments, we performed RNA-seq on the LCM samples (S3 Table). The samples clustered by the apical/basal origins and villi zones rather than by patients (S1C Fig). For each gene, we calculated the ratio between apical and basal transcript normalized abundances (Methods). The mouse log2 (apical/basal) significantly correlated with those shown by Moor and colleagues [3] (S1D Fig).

We measured 8,103 genes above mean normalized expression of 10⁻⁵, 632 of which exhibited significant mRNA apical-basal polarization (sign-rank test q-value Fig 2A). These included the basally polarized genes MYH14, DSP, and CPS1 and the apically polarized genes CDH1, SLC5A1, and APOA4, which we validated using single-molecule fluorescence in situ hybridization (smFISH, Figs 2B, S1E, and S1F and S2 Table).

The apical bias of mRNAs had a negative correlation with protein lengths, and positive correlation with mRNA expression levels and stabilities, as well as with translation rates, protein abundances, and stabilities (S2A Fig). Therefore, apical mRNAs seem to be shorter, more stable, and more efficiently translated. Additionally, the apical genes were enriched in metabolic functions such as oxidative phosphorylation, glycolysis, and fatty acid metabolism, while the basal genes were enriched in mitotic spindle pathway genes (S2B Fig and S3 Table). None of the highly expressed genes changed their polarization between the base and tip of the villi in either human or mouse (S2C Fig, no genes had q-value lower than 0.5). Our data therefore exposes mRNA polarization in the human intestinal epithelium that seems to be invariable in distinct villi zones.

Next, we compared the inter-species changes of the mRNA polarization between humans and mice. Apical bias was significantly correlated between the 2 species (R = 0.36, p = 2.2 × 10⁻¹⁶, Fig 2C). Only 5 genes changed their polarization between the 2 species and were also significantly polarized in at least one of them—PIGR, SLC5A1, RACK1, MGAM, CDHR5 (Fig 2C). We validated 2 of these genes with smFISH–PIGR was significantly apical in humans yet significantly basal in mice, whereas SLC5A1 was significantly apical in humans yet non-polarized in mice. We also validated APOB, which shows apical polarization in both species (Fig 2D). As inflammation has been shown to up-regulate Pigr expression via cytokines [5], we checked whether the polarization of Pigr mRNA changes in germ-free mice, antibiotic-treated mice, and lipopolysaccharides (LPS)-injected mice. We found that the polarization remained basal in all conditions (S2D Fig).

Fig 2. Human intestinal epithelium exhibits mRNA polarization.

(A) MA-plot of RNA-seq of human samples collected by LCM from jejunums of 8 patients. Significantly polarized genes are in blue (signed rank test, q-value (B) smFISH staining of selected basal (left) and apical (right) genes in human jejunum. Red staining is anti E-cadherin antibody which stains epithelial basolateral membranes, blue staining is DAPI which stains DNA. Scale bar is 20 μm. (C) Spearman correlation between the mRNA log2(apical/basal) of human and mouse genes. Genes that significantly change their apicome between mouse and human (Methods) are annotated in red. Other selected genes are annotated in black. Data is available in S3 Table and S1 Data. (D) smFISH staining of selected genes in mouse and human jejunums. Red staining is anti E-cadherin antibody which stains epithelial basolateral membranes, blue staining is DAPI which stains DNA. Scale bar is 20 μm. Data is available in S3 Table and in S1 Data. LCM, laser capture microdissection; smFISH, single-molecule fluorescence in situ hybridization.

https://doi.org/10.1371/journal.pbio.3002942.g002

Polarization of mRNA and proteins largely overlap in human

To explore the human protein apicome, we performed LCM followed by mass spectrometry proteomics, identifying 1,582 proteins (S4 Table). We found a significant correlation between the zonated changes in mRNAs and proteins along the villus axis (S3A Fig, Methods). This positive correlation is in line with recent analysis of whole epithelial segments [4]. Among the measured proteins, 288 (approximately 18%) demonstrated significant apical/basal bias (DESeq2, q-value Fig 3A). These included the apical proteins encoded by the genes ANPEP, ACE2, and SLC5A1 and the basal proteins encoded by CASK, ITGA6, and GPA33 (Fig 3A and 3B). Proteins previously shown to be enriched in the mouse epithelial brush border [6] were strongly apical in our measurements (S3B Fig).

We next investigated if an mRNA and its respective protein would localize to the same intracellular compartment. We found a weak, yet significant positive correlation between the apical bias of mRNAs and their matching proteins (Spearman’s R = 0.13, p = 7.8 × 10⁻⁶, Fig 3C), with the majority of genes exhibiting co-localization of proteins and mRNAs (61%, S3C Fig). An example of a concordant gene was SLC5A1, which showed both apical mRNA and protein polarization (Fig 3C and 3D). Notably, some genes also exhibited discordant mRNA-protein polarization, such as CDH1, with apically polarized mRNAs and basolateral polarized E-Cadherin proteins, and Myosin heavy chain 14 (MYH14), which exhibits the opposite mRNA/protein polarization pattern (Fig 3C and 3D). Although our LCM was carefully performed to focus on the epithelial layer (Fig 1C), the basal compartment included some stromal and immune mRNAs and proteins, which could affect protein-mRNA correlations. We also computed the mRNA-protein correlation using enterocyte-specific genes (Methods, S3 Table) and found a slightly higher positive correlation between the apical bias of mRNAs and their matching proteins (Spearman’s R = 0.32, p = 3.7 × 10⁻³, S3D Fig). This suggests that some proteins and mRNAs in human enterocytes co-localize in their apical/basal cellular compartments.

Fig 3. Human intestinal epithelium exhibits protein polarization.

(A) MA-plot of proteomics of human samples collected by LCM, from jejunums of 4 patients, followed by mass spectrometry. Significantly polarized genes are in blue (q-value (B) IHC staining of selected basal (top) and apical (bottom) proteins from the human protein atlas [7] of healthy duodenum or small intestines. Scale bar is 50 μm. Direct URLs: ANPEP, ACE2, SLC5A1, CASK, ITGA6, GPA33. (C) Log2(apical/basal) of mRNAs and their corresponding proteins. R is Spearman correlation, selected genes are annotated in black, and red genes are shown in D. RNA-seq and proteomics data are available in S3 and S4 Tables, respectively. (D) (Left) smFISH images of human jejunum, of selected genes, with DAPI in blue and CDH1 protein in red. (Right) IHC images from the human protein atlas [7] of healthy duodenum or small intestines, direct URLs: CDH1, SLC5A1, MYH14. Scale bar is 20 μm. Data is available in S1 Data. IHC, immunohistochemistry; LCM, laser capture microdissection; smFISH, single-molecule fluorescence in situ hybridization.

https://doi.org/10.1371/journal.pbio.3002942.g003

Streamlined polarization of protein groups in the human intestinal epithelium

We next focused the analysis on polarization of selected protein groups (Figs 4A, 4B, and S4). As expected, histones showed higher abundances in the basal side, since the nuclei of enterocytes are basally shifted. Extra cellular matrix (ECM) proteins, as drawn from the Matrisome [8] and integrins, were also basal, whereas mucus components, including membrane-associated and secretory mucins [9] were highly apical (Figs 4A, 4B, and S4). Various digestive enzymes, xenobiotic metabolism proteins, together with sulfotransferases were apically polarized. Solute carrier proteins (SLC), which transport various nutrients from the lumen, and particularly amino-acid transporters, showed apical polarization (Figs 4A, 4B, and S4). As reported previously [10], the proteins SGLT1 (SLC5A1) and GLUT5 (SLC2A5) which transport glucose and fructose, respectively, from the lumen into the enterocytes, were apically localized, whereas GLUT2 (SLC2A2), which transports the monosaccharides from the enterocytes to the stroma was slightly basally polarized (Figs 4A, S4, and S5).

Intestinal lipid absorption is a complex process that begins outside of the enterocytes with the breakdown of dietary triacylglycerols into free fatty acids. These are then imported into enterocytes, where they are re-assembled into triacylglycerols and incorporated into chylomicrons, which are secreted into the lymphatics. We found that the polarization of lipid absorption proteins followed this streamlined process (Figs 4A, S4, and S6). The pancreatic lipase (PNLIP), which initiates the extracellular breakdown of triglycerides, was strongly localized in the apical side. The fatty-acid transporters CD36 and FATP4 (SLC27A4), as well as the intestinal fatty acid-binding protein (I-FABP, FABP2) and the triacylglycerol assembly enzymes MGAT and DGAT were also apically biased. Chylomicron apolipoproteins exhibited diverse polarization patterns. APOA4 was apical, APOB-48 was balanced, and APOA1 was basal. These localization patterns could be explained by the role of APOA1 as both a late-incorporated chylomicron protein and a component of high density lipoprotein (HDL), which is exported basolaterally with ingested cholesterol [11,12]. Our polarization measurements therefore capture the streamlined process of carbohydrate and lipid absorption. Notably, these polarization trends remain unchanged after filtering out nuclear and matrix proteins, which are more abundant in the basal samples in our data (S7A Fig).

Our protein measurements revealed differences in the polarization patterns of ribosomal and mitochondrial proteins between humans and mice (Fig 4B). Whereas in mice, ribosomes were shown to localize apically and mitochondria basally [3], in humans, ribosomal proteins were slightly basal and mitochondria were slightly apical. We confirmed this finding by smFISH for 18S rRNA and MT-RNR1/mt-Nd6 mitochondrial RNA (Figs 4C, 4D, 4F, 4G, and S7B and S2 Table), and with RNA-seq (Fig 4E).

Fig 4. Intracellular polarization of protein groups.

(A) Max-normalized abundance of proteins involved in nutrient processing and absorption on the apical and basal sides from LCM-proteomics data. (B) Log2(apical/basal) of protein groups based on LCM-proteomics. Q-values are based on Wilcoxon rank-sum test with Benjamini–Hochberg correction. ECM, extracellular matrix, SLC, solute carrier proteins. Horizontal bars are medians, boxes delineate the 25–75 percentiles. (C, D) smFISH of mitochondrial-RNA (C) and ribosomal-RNA (D). Red staining is anti E-cadherin antibody which stains basolateral epithelial membranes, blue staining is DAPI which stains DNA. Scale bar is 20 μm. Quantification shown in (F and G). (E) The log2(apical/basal) of mitochondrially encoded genes with normalized expression above 10⁻⁴ in both human and mouse, based on RNA-seq. P-values are based on the paired Wilcoxon signed rank test. Horizontal bars are medians, boxes delineate the 25–75 percentiles. (F, G) Fluorescence intensity of smFISH stainings in the apical and basal sides of ribosomal (F) and mitochondrial (G) RNA. Each measurement consists of the median intensity across 3–5 epithelial cells from 8 patients and 3 mice. P-values are based on the paired Wilcoxon signed rank test for each patient/mouse, combined by the Fisher method for multiple p-values (Methods). Horizontal bars are medians, boxes delineate the 25–75 percentiles. Data is available in S4 Table and S1 Data.

https://doi.org/10.1371/journal.pbio.3002942.g004

Experimental methods

Animal experiments and tissue preparations

Mouse experiments were approved by the Institutional Animal Care and Use Committee of the Weizmann Institute of Science (approval number 13000419–2) and performed in accordance with institutional guidelines. C57BL/6 Male mice aged 2 to 5 months, housed under regular 12 h light-dark cycle and fed ad libitum were used in the experiments.

For the antibiotic treatment, mice were given a combination of the following antibiotics for 4 weeks, vancomycin (1 g/l), ampicillin (1 g/l), kanamycin (1 g/l), and metronidazole (1 g/l) in their drinking water [29–31], all antibiotics were obtained from Sigma Aldrich. For LPS treatment—LPS was dissolved in sterile PBS. Mice were injected 20 μg/kg body weight LPS intraperitoneally, mice were sacrificed 12 h after the injection. Overall, 2 mice from each condition and 5 controls were used in the study.

Mice sacrificed by cervical dislocation and the jejunums were harvested and washed with cold PBS. The proximal half of each jejunum was lateralized and spread onto a Whatman paper, embedded in OCT, and kept at −80°C. The other half of each jejunum was fixed in cold 4% FA in PBS for 3 h at 4°C followed by overnight fixation in 4% FA + 30% sucrose at 4°C while revolving. The jejunums were rolled to a “Swiss-roll,” embedded in OCT, and kept at −80°C.

Single molecule fluorescence in situ hybridization

Probe libraries were designed using the Stellaris FISH Probe Designer Software (Biosearch Technologies, Inc., Petaluma, California, United States of America). Probe libraries (S5 Table) were coupled to Cy5, TMR, or Alexa594. Cryosections (6 μm) of fixed mouse or human intestines were mounted on poly-L-lysine pre-coated coverslips and used for probe hybridization, as previously described [32]. Briefly, sections were fixed for 15 min in 4% FA in PBS, washed with PBS, and incubated for 2 h in 70% ethanol in 4°C. Sections were washed with 2× SSC (Ambion AM9765) for 5 min, then permeabilized for 10 min with proteinase K (10 mg/ml, Ambion AM2546) followed by 2 washes with 2× SSC (Ambion AM9765) for 5 min. Tissues were incubated in wash buffer (20% Formamide Ambion AM9342, in 2× SSC) for 10 min in a 30°C incubator. Tissues were incubated with hybridization mix (10% Dextran sulfate Sigma D8906, 20% formamide, 1 mg/ml E. coli tRNA Sigma R1753, 10% 20× SSC, 0.02% BSA Ambion AM2616, 2 mM Vanadylribonucleoside complex NEB S1402S, in Rnase-free water) mixed with probes and FITC-conjugated anti-CDH1 antibody (1:100, BD Biosciences cat. 612131) overnight in 30°C. For negative control staining—hybridization mix was used without addition of probes. After the hybridization, tissues were washed with wash buffer for 30 min in 30°C, and with wash buffer with DAPI (50 ng/ml, Sigma-Aldrich, D9542) for 5 min. Additional wash with 2× SSC was performed, and coverslips were mounted with ProLong Gold antifade reagent (Invitrogen P36934). All images were performed on a Nikon-Ti-E inverted fluorescence microscope using the NIS element software AR 5.02.03. Images were acquired using 100× magnification. All smFISH experiment were performed on tissues from at least 3 patients from multiple fields of view.

Laser capture microdissection

Cryosections (10 μm) of fresh-frozen mouse or human intestines were mounted on polyethylene-naphthalate membrane-coated glass slides (Zeiss, A4151909081000), air-dried for 20 s at room temperature, washed in 70% ethanol for 30 s, incubated in Rnase-free water for 30 s, followed by staining with HistoGene Staining Solution for 20 s (Thermo Fisher Scientific, KIT0401). The stained sections were dehydrated with subsequent 30 s incubations in water, and then in 70%, 95%, and 100% ethanol and air dried for 4 min before microdissection. Tissue sections were microdissected on a UV laser-based PALM Microbeam (Zeiss). The system makes use of a pulsed UV laser that cuts the tissue at indicated marks with minimal damage to surrounding cells; the cutting was performed with the following parameters: PALM 40X lens, cut energy 43 (1–100), cut focus 52 (1–100). Tissue fragments were catapulted and collected in 0.2 ml adhesive cap tubes (Zeiss, A4151909191000) with these settings: LPC energy 68 (1–100), LPC focus 50 (1–100). For RNA-seq, the cutting time of each slide was limited up to 80 min, and buffer RLT from QIAGEN with 0.25 M Dithiothreitol was added and immediately frozen on dry ice. For MS, 5% SDS, 50 mM Tris (pH 8) was added before freezing.

RNA-seq

RNA-seq Library preparation was performed following mcSCRBseq protocol ([33]). The cDNA was pre-amplified with 17 cycles. Up to 0.6 ng of the amplified cDNA was converted into sequencing library with the Nextera XT DNA Library kit (Illumina, FC131-1024), according to supplier protocol. Quality control of the resulting libraries was performed with an Agilent High Sensitivity D1000 ScreenTape System (Agilent, 5067–5584). Libraries were loaded with a concentration of 1.8 nM and sequenced on a Novaseq 6000 (Illumina) with the following cycle distribution: 8 bp index1, 26 bp read1, 8 bp index2, and 66 bp read2. Illumina output files were demultiplexed and aligned to the genome with UTAP [34], with CUTADAPT with default parameters. A total of 32 human samples from 8 patients were aligned to the HG38 genome (Gencode_V34_24.3.2020), and 18 mouse samples from 5 mice were aligned to the mm10 genome (Gencode_vM25_24.3.2020).

Proteomics of LCM samples

LCM samples from 4 patients were lysed and digested with trypsin using the S-trap method [35]. Lysates in 5% SDS in 50 mM Tris-HCl were incubated at 96°C for 5 min, followed by 6 cycles of 30 s of sonication (Bioruptor Pico, Diagenode, USA). Proteins were reduced with 5 mM dithiothreitol and alkylated with 10 mM iodoacetamide in the dark. Phosphoric acid was added to the lysates to a final concentration of 1.2%, and 90:10% methanol/50 mM ammonium bicarbonate and then added to the samples. Each sample was then loaded onto S-Trap microcolumns (Protifi, USA). Samples were then digested with 250 ng trypsin for 4 h at 37°C. The digested peptides were eluted using 50 mM ammonium bicarbonate; trypsin was added to this fraction and incubated overnight at 37°C. Two more elutions were made using 0.2% formic acid and 0.2% formic acid in 50% acetonitrile. The 3 elutions were pooled together and vacuum-centrifuged to dry. Samples were kept at −80°C until analysis.

Liquid chromatography

ULC/MS grade solvents were used for all chromatographic steps. Each sample was loaded using split-less nano-Ultra Performance Liquid Chromatography (10 kpsi nanoAcquity; Waters, Milford, Massachusetts, USA). The mobile phase was: (A) H2O + 0.1% formic acid; and (B) acetonitrile + 0.1% formic acid. Desalting of the samples was performed online using a reversed-phase Symmetry C18 trapping column (180 μm internal diameter, 20 mm length, 5 μm particle size; Waters). The peptides were then separated using a T3 HSS nano-column (75 μm internal diameter, 250 mm length, 1.8 μm particle size; Waters) at 0.35 μl/min. Peptides were eluted from the column into the mass spectrometer using the following gradient: 4% to 30%B in 105 min, 30% to 90%B in 10 min, maintained at 90% for 7 min and then back to initial conditions.

Mass spectrometry

The nanoUPLC was coupled online through a nanoESI emitter (10 μm tip; Fossil, Spain) to a quadrupole orbitrap mass spectrometer (Exploris 480, Thermo Scientific) using a FlexIon nanospray apparatus (Thermo Scientific). Data was acquired in data-dependent acquisition (DDA) mode, with a cycle time limit of 2 s. MS1 resolution was set to 120,000 (at 200 m/z), mass range of 380–1,500 m/z, AGC of 200% and maximum injection time was set to 50 msec. MS2 resolution was set to 15,000, quadrupole isolation 1.4 m/z, AGC of 150%, dynamic exclusion of 40 sec and maximum injection time of 150 msec. Raw data was processed with MaxQuant (v2.0.1) [36]. The data was searched with the Andromeda search engine against the human proteome database appended with common lab protein contaminants and the default modifications. Quantification was based on the intensity-based absolute quantification (iBAQ) method, based on unique peptides, missing values were imputed from a low, random distribution.

Computational methods

RNA-seq analysis pipeline

RNA-seq results yielded alignment of 17,947 and 17,019 genes for human and mouse samples, respectively. In mouse data, one low-quality sample was removed due to excessively low unique molecular identifier (UMI) counts (below 10,000). Non-protein-coding genes were removed, based on BioMart [40] gene_biotype values.

Data were normalized by dividing each value in a sample by the sum of UMIs in the sample. For principal component analysis (PCA), samples with more than 20,000 UMIs were used, and genes with mean expression >10⁻⁴ were scaled, and PCA was performed.

Pooled apical bias for each gene was calculated by geometric mean of the apical/basal ratios of all patients/mice, as follows:

Where is number of patients/mice (n = 8 for human and n = 5 for mouse), A and B are the normalized apical and basal expression, respectively, and pn is pseudo-number which is the minimal normalized expression across all samples and all genes, which is not 0. This value is added to avoid division by zero and consequently infinite or non-defined values when expression levels are low, without skewing the ratios. P-values for apical/basal bias and for changes between top/bottom of villi were calculated using exact paired sign-rank test with two-sided alternative, on highly expressed genes (mean normalized expression >10⁻⁵). Q-values were calculated using the Benjamini–Hochberg method.

Gene set enrichment analysis (GSEA) was performed on the log2(apical/basal) values of all genes, using Fgsea 1.28.0 [41] package, fgseaSimple function, with the “hallmark” database.

Correlation of RNA-seq with previously published data

Spearman correlation was performed to assess the relationship between log2(apical/basal) and various biological parameters. The parameters included data from Schwanhäusser and colleagues [26] and Harnik and colleagues [25]. mRNA expression and protein abundance are from the current study, calculated as the mean expression of the normalized LCM RNA-seq and proteomics.

Spearman’s correlation was also calculated on the log2(apical/basal) from Moor and colleagues [3] in comparison to the log2(apical/basal) from the current study, on significant and highly expressed genes (q-value in both data sets 10⁻⁵) using the ggpubr 0.6.0 [39] and ggplot 2 3.5.0 [38] libraries (S1B Fig).

Comparison of mouse and human RNA-seq

RNA-seq analysis of both species was merged. Genes were renamed based on orthology table from Ensembl (V.109), obtained with BioMart [40]. Spearman correlation analysis was performed based on log2(apical/basal) as described in the RNA-seq analysis section. For each gene, Wilcoxon rank-sum test was performed for the human samples and the mouse samples. P-values were corrected for multiple hypotheses using the Benjamini–Hochberg method. Genes were determined as significant based on the q-value of the inter-species q-value 10⁻⁴.

smFISH quantification of gene expression

smFISH images were analyzed by manual segmentation of apical and basal compartments of few cells and identification of dots using imageM [32]. For each gene, multiple FOVs from distinct 3 villi per patient, across 2 to 3 patients were quantified. Spot intensity was normalized by the segmented area and the log2(apical/basal) ratio was calculated for each paired segmentation. The mean and standard error of the mean were calculated for the log2(apical/basal) ratios.

Proteomics data analysis

Suspected contaminants, as well as proteins with number of unique peptides lower than 2 were removed. The minimal iBAQ value was subtracted from all values, and each iBAQ value was divided by the sum of iBAQ of its sample. Low-quality samples, with number of identified proteins 42] FindMarkers function with DESeq2 method [43]. Log2(apical/basal) was defined as where pn = 10⁻⁶. ECM proteins were assigned based on Matrisome database [8], mitochondrial proteins were assigned by gene description from Ensembl (V.109), obtained with BioMart [40] and metabolic proteins were obtained from Burclaff and colleagues [44]. For each group, p-value was calculated with two-sided Wilcoxon rank sum test (proteins in group versus all the rest of the proteins), and Benjamini–Hochberg correction was performed.

For heatmaps visualization—normalized abundance of each protein was defined as follows:

And heatmaps were visualized using ggplot2 3.5.0 [38]. KEGG pathways were visualized with KEGGREST 1.42.0 [45] and pathview libraries [46].

The analysis was repeated, focusing exclusively on cytoplasmic proteins. Subcellular localization data was obtained from the Human Protein Atlas [47] and alternative protein names were retrieved from UniProt [48]. Proteins identified solely as nuclear based on subcellular location gene annotation, as well as ECM proteins according to the Matrisome [8], were excluded. Following this filtration, data was renormalized by dividing each value by the sum of values in the sample.

mRNA and protein correlation in cellular polarization

The log2(tip/base) ratios from the average of RNA-seq apical and basal samples were merged with the corresponding proteomics apical samples. Spearman’s correlation was then calculated and visualized for relatively highly expressed genes (normalized expression >10⁻⁴) using the ggpubr 0.6.0 [39] and ggplot2 3.5.0 [38] libraries. Additionally, Spearman’s correlation was calculated on the log2(apical/basal) ratios of RNA-seq and proteomics on relatively highly expressed genes (normalized expression >10⁻⁵).

Epithelial-specific genes were classified based on integration of Elmentaite and colleagues [49] and Burclaff and colleagues [44] single-cell RNA-seq data and cell type classifications. The mean of cells from each cell type was calculated and then the signature-matrix was normalized where each value was divided by the sum of values in the cell type. Enterocyte-specific genes were classified based on the following:

Where E is the expression in mature enterocytes, N is the maximal expression across all non-enterocyte cell types, and pn is the minimal non-zero value present in the signature-matrix. This value is added to avoid division by zero and consequently infinite or non-defined values when expression levels are low, without skewing the ratios.

Mitochondrial/ribosomal intensity quantification of smFISH images

Segments from the apical and basal sides of 3 to 5 epithelial cells from spatially distant villi were manually segmented on nuclear and membrane staining using ImageJ [37] for each of the 8 patients and of 3 mice (S2C Fig). The median intensity for mitochondrial and ribosomal stainings were quantified in both sides, and the corresponding median intensity of the background was subtracted. Ratio of apical/basal background-subtracted intensities was calculated, and 2 signed-rank tests were performed on each patient, one test with left-tail and one with right-tail hypothesis. The minimal p-value of both tests was combined across the 8 patients and across the 3 mice using Fisher’s method.