*Review* **Dietary Modulation of Bacteriophages as an Additional Player in Inflammation and Cancer**

**Luigi Marongiu 1, Markus Burkard 2, Sascha Venturelli 2,3,\* and Heike Allgayer 1,\***


**Simple Summary:** The role and function of bacteriophages (phages) in the intestine, its health and microbial homeostasis has been underestimated so far. This interdisciplinary review highlights the effect of dietary compounds on phages and puts this into perspective with putative contributions of phages to gastrointestinal diseases, specifically inflammation, infection, and cancer. The review discusses novel fields of opportunities in this context. These include, but are not limited to, perspectives how a better understanding of modulating the activity of specific phages by particular nutritional components may contribute to reorganizing the microbial network, thus supporting in the combat, or even prevention, of inflammation or even cancer in the gut.

**Abstract:** Natural compounds such as essential oils and tea have been used successfully in naturopathy and folk medicine for hundreds of years. Current research is unveiling the molecular role of their antibacterial, anti-inflammatory, and anticancer properties. Nevertheless, the effect of these compounds on bacteriophages is still poorly understood. The application of bacteriophages against bacteria has gained a particular interest in recent years due to, e.g., the constant rise of antimicrobial resistance to antibiotics, or an increasing awareness of different types of microbiota and their potential contribution to gastrointestinal diseases, including inflammatory and malignant conditions. Thus, a better knowledge of how dietary products can affect bacteriophages and, in turn, the whole gut microbiome can help maintain healthy homeostasis, reducing the risk of developing diseases such as diverse types of gastroenteritis, inflammatory bowel disease, or even cancer. The present review summarizes the effect of dietary compounds on the physiology of bacteriophages. In a majority of works, the substance class of polyphenols showed a particular activity against bacteriophages, and the primary mechanism of action involved structural damage of the capsid, inhibiting bacteriophage activity and infectivity. Some further dietary compounds such as caffeine, salt or oregano have been shown to induce or suppress prophages, whereas others, such as the natural sweeter stevia, promoted species-specific phage responses. A better understanding of how dietary compounds could selectively, and specifically, modulate the activity of individual phages opens the possibility to reorganize the microbial network as an additional strategy to support in the combat, or in prevention, of gastrointestinal diseases, including inflammation and cancer.

**Keywords:** Phage; bacteriophages; diet; infection; colorectal; cancer; nutrition

### **1. Introduction**

The impact of the gut bacteriome on the human physiology is currently being investigated and seems to have a significant influence on the development and treatment

**Citation:** Marongiu, L.; Burkard, M.; Venturelli, S.; Allgayer, H. Dietary Modulation of Bacteriophages as an Additional Player in Inflammation and Cancer. *Cancers* **2021**, *13*, 2036. https://doi.org/10.3390/ cancers13092036

Academic Editor: Damián García-Olmo

Received: 9 April 2021 Accepted: 21 April 2021 Published: 23 April 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

of various diseases. Collectively, the over one thousand bacterial species residing in the human gut encode 3.3 million genes, expanding the human genome 150 times over [1]. Several studies have demonstrated that microorganisms present in the human gut (the gut microbiome) modulate human physiology at different levels. Intestinal bacteria not only metabolize polysaccharides that would be otherwise indigestible [2], but also regulate peristalsis [3], help to keep a proper intestinal morphology as it has been shown in a gnotobiotic pig model [4], maintain the integrity of the intestinal barrier [5–7], attenuate inflammation [8,9], reduce the virulence of pathogenic species [10], and even influence the action of anticancer drugs [11]. Although it has been proposed to consider the intestinal microorganisms as symbionts rather than simple commensal species [12], our understanding of the dynamics underlying the interactions between host and gut microbiome is still limited [13,14].

Bacteriophages (or phages for short) represent a significant modulator of the gut microbiome [15]. By definition, phages infect bacteria, but more and more data highlight the interrelation between eukaryotic cells and bacterial viruses. Phages can interact directly with the human body since they can translocate inside eukaryotic cells [16] and activate the immune system, exacerbating ongoing colitis symptoms and boosting the antibacterial response [17]. It has recently been proposed to consider phages as human pathogens [18].

In the last few years, phages have become a crucial topic in the medical and microbiological fields because these viruses can be used as a treatment of bacterial infections in the context of the rising problem of antibiotic resistance [19–21]. As our understanding of phage biology increases, the applications of phage therapy also expand. Phages have been applied to treat bacterial infections ever since their discovery, and phage therapy is becoming more and more popular in fields ranging from dentistry to medical microbiology [22–25]. For example, phages are currently being evaluated to fight infections in poultry that are still an economic and health issue [24]. Recent studies suggest that phages can also be applied in antiviral and anticancer therapies. For instance, it has been proposed that phage T4 might be used as a co-treatment for COVID-19 because this phage reduces the immune response, which is an important contributor to the fatality associated with this disease [26]. Furthermore, it has been shown that phages bind to cancerous cells and reduce the size of the tumor mass in different mouse models [17,27,28], opening the possibility of phage-mediated oncolytic virotherapy.

Diet can influence the gut microbiome, and it is actively used as an intervention to reduce the risk of developing diseases [29]. Particular components have been shown to be of benefit in the treatment of even severe disease conditions up to cancer. For example, in own previous studies, it was demonstrated that the plant-derivatives curcumin and artesunate inhibit tumor cell invasion and metastasis, at least in part via regulating the expression of proteolytic enzymes, the molecular cascades involving transcriptional factors and microRNAs, respectively [30–33]. However, there is a lack of studies describing how dietary compounds impact microorganisms in general and phages in particular. Seminal studies in the 1950 s demonstrated the antiviral activity of tannins, which are contained in popular beverages such as tea and coffee, and of acerin, the active component of maple fruit [34,35]. Especially, tea showed broad antimicrobial activity, including inactivation of phages [36]. It is also known that essential oils have antibacterial and antiviral properties as well as anti-inflammatory and regenerative activities [37]. Nevertheless, gaining experimental knowledge on the influence of dietary compounds on phages as modulators of microbiota has not yet been in focus of attention in the research community.

Most of the studies related to the effect of dietary compounds on phages have been focusing on human viruses associated with gastroenteritis. Phages have been used as surrogates for viruses that cannot be easily cultivated, such as norovirus, rather than for studying bacterial virus biology as such. Also, most of the bacteriophage studies so far have been limited to phages infecting *Escherichia coli* (coliphages). Nonetheless, *E. coli* plays an essential role in human health since certain strains of this species, known as Shiga toxin-producing *E. coli* (STEC), are widespread food-borne pathogens. The most prevalent STECs are O157, O26, O45, O103, O111, O121, and O145. These seven serotypes induce diseases ranging from acute diarrhea to hemorrhagic colitis and fatal hemolytic syndrome [38–40].

The main STEC derived virulence factor is shiga-like toxin (Stx), which is encoded by the prophages 933 J (*Stx1*) and 933 W (*Stx2*) [41,42]. Upon activation, these prophages express *Stx*, and they can horizontally spread this gene by transduction [43,44]. Genotoxins, such as cytolethal distending toxins and colibactin, are considered cancer risk factors and can be found in pathogenic *E. coli* strains [45]. Interestingly, many natural compounds have been shown to be bactericidal against pathogens [46], and to suppress the biological activity of toxins, including the cholera and ricin toxins [46–49]. Peas showed to bind with high efficiency Stx, acting as toxic-scavengers, whereas beans can reduce the intake of Stx [50].

We speculate that a better understanding of how phages are activated or inhibited in the human gut might be pivotal in modulating the intestinal microbiome, e.g., to counteract bacterial infections, inflammatory conditions, and even carcinogenesis and cancer progression. Such indirect antibacterial activity is a particularly relevant feature in light of the urgent need to identify alternatives and additional strategies to antibiotics to defeat (therapy-resistant) bacterial infections. The present review will summarize the current knowledge on the effect of dietary compounds on phages, their activity, and infectivity.

### **2. Interactions between Phages and Bacteria in the Gut Microbiome**

Phages were first described by the French-Canadian Félix d'Hérelle, of Institute Pasteur in 1917, who also defined the term 'bacteriophage' ("eater of bacteria"). As a first, pioneering phage-based therapy, he applied bacteriophages to treat *Shigella* infections in soldiers, establishing what became known as phage therapy [51–53]. Phages can be subdivided into two groups: virulent (lytic) and temperate (lysogenic) (Figure 1) [54]. Lytic viruses start the replication process soon after the infection of the bacterial host. Once the progeny virions have assembled in a sufficient number (the burst size), the cell bursts open, releasing the new phages in the surrounding environment. Lysogenic phages have an additional phase: they can integrate as prophages in the bacterial chromosome and undergo a latency period where only a viral transcription suppressor is produced actively. In particular contexts, such as bacterial starvation or DNA damage, the suppression control is relieved, and the prophage enters the lytic phase. Conversely, in the presence of a high number of infected bacteria, phages exit the lytic phase and initiate lysogeny [55]. Both virulent and temperate phages modulate the bacterial population through lysis.

Phages can also modulate the bacterial population, indirectly. It is well known that bacteria must undergo a fierce competition within each ecological niche, and, therefore, some species have developed virulence factors to improve their chances of survival [56]. Moreover, the microbial competition is complex and difficult to predict. For instance, *Lactobacillus delbrueckii* and *L. rhamnosus* inhibit *E. coli* O157, but *L. plantarum* suppresses the commensal strains of *E. coli* but not O157, and *L. paracasei* does not constrain *E. coli* at all [57]. In addition, the suppression of one species might cause the unexpected expansion of a species not apparently associated with the suppressed one. For instance, *E. coli* fosters the growth of *B. fragilis* but represses *B. vulgatus*. Knocking down *E. coli* by phage T4 is, therefore, followed by a contraction of the prevalence of *B. fragilis* and an increased growth of *B. vulgatus,* but also of *Proteus mirabilis* and *Akkermansia muciniphila* [58]. It is also known that commensal species can neutralize toxins, reducing the fitness of the pathogens. For instance, surface proteins of *L. plantarum* can neutralize Stx, reducing the cytotoxicity (and, thus, the fitness) of *E. coli* O157 [59]. Therefore, the alteration of even one species due to phagial predation can have drastic consequences for the microbiome.

Mounting evidence suggests that phages have access to eukaryotic (and human) cells [60]. Even though tissues are expected to be sterile, it has been known for decades that an ingestion of phage preparations during phage therapy is followed by a recovery of phages in human urine and blood within a few minutes from the administration [61,62]. This recovery implies that the viruses had somehow crossed the gastrointestinal barrier. Recent virome studies have identified genes belonging to phages in both blood and brain [63,64]. The circulation of phages in the peripheral blood has been named 'phagemia', but there is a lack of hard evidence for its actual existence in physiological conditions [65]. Furthermore, phages can be actively transported from one side to another of intestinal cells (transcytosis) via the Golgi network [16].

**Figure 1.** Outcomes of phagial infection of bacteria. A virulent phage (yellow particle on the left) can infect a bacterium (in blue). The replication of the phage leads to lysis of the host cell, releasing the viral progeny (yellow particles on the right). Alternatively, some viral species known as temperate can establish an additional step known as latency. The phagial genome can remain independent from that of the bacterium (pseudo-lysogeny) or become integrated into the host's genome (lysogeny). In both cases, the viral expression is kept to a minimum and there is no virion production until several cellular conditions are met. Upon induction, temperate phages enter the lytic pathway and determine the lysis of the host.

### **3. Effect of Dietary Compounds on Phages**

Several dietary compounds can alter the physiology of phages, as summarized in Figure 2. Although many studies showed a connection between nutrition and intestinal microbiome, there are only a few studies that deal with the effects of nutrition on the activity of phages. Seminal work in the 1960 s indicated that amino acids and vitamins had a different impact on the induction of prophage λ in *E. coli* [66]. For instance, the amino acid cysteine was an inducer, but its oxidized derivative cystine was not. About four decades later, it was shown that essential oils extracted from chamomile, lemongrass, cinnamon, and geranium could greatly reduce the infectivity of *E. coli* T7 and *S. aureus* SA, whereas others (such as angelica, cardamom, lime, and rosemary) affected only the former phage [67]. A recent study reported how different compounds could selectively activate some viruses but not others in bacterial growth and prophage-induction assays [68]. This study demonstrated how stevia, a natural sweetener obtained from the Brazilian shrub *Stevia rebaudiana* [69], could strongly induce prophages present in *Bacteroides thetaiotamicron* and *Staphylococcus aureus* but not in *Enterococcus faecalis*, whereas uva ursi (derived from the shrub plant *Arctostaphylos uva-ursi*), aspartame (a peptide), and propolis (a flavonoid) resulted in the opposite. These data indicate that dietary compounds can modulate the gut virome and, consequently, alter the gut bacteriome.

**Figure 2.** Summary of actions on phages on dietary compounds. There are three main mechanisms of action of dietary compounds upon phages. A dietary compound can modify the capsid, blocking the infectivity of the targeted phage (capsid alteration). Alternatively, dietary compounds can lead to the degradation of nucleic acids (genome damage). In this case, a phage can infect the host, but there will be neither lysis nor viral progeny. However, DNA damage to the host cell's genome triggers the induction of prophages (dotted arrows). A final mechanism of action (repression of replication) involves an interference with the replication of the viral genome. Even in this case, there is infection, but no viral progeny is produced.

Experiments measuring the effect of dietary compounds on phage activity have been based on few classes of compounds, mainly polyphenols. These are molecules that contain one or more phenolic aromatic rings (benzenes with hydroxide moieties). Polyphenols can be subdivided into phenolic acid derivatives and flavonoids [70]. The former can, in turn, be subdivided into derivatives of either hydroxybenzoic acid (for instance, gallic acid) or cinnamic acid (for example, caffeic acid) [71]. Tea, the second most frequently consumed beverage after water, is a primary source for gallic acid [72]. Coffee, whose consumption is increasing worldwide [73], contains chlorogenic acid (a combination of caffeic acid and quinic acid) [71]. Tannic acid, which contains several hydroxybenzoic acid moieties, is particularly abundant in berries; soy is rich in isoflavonoids, such as genistein and daidzein [74]. The exact mechanism of action of these phenol-compounds is not entirely understood. Still, it is known that they can be beneficial for human physiology and have been used in folk medicine since millennia [75]. They are currently being investigated for their anticancer activity [76–78].

The chemical structure of the compounds discussed herein is shown in Figures 3–5. A summary of the activities identified is given in Table 1. The most common outcome of exposure to a given nutrient is a loss of infectivity; this is measured by comparing the plaque-forming units (PFU) of a control and an exposed suspension (measured in mL) of phages. If the control and the exposed suspensions showed, for instance, 1010 and 10<sup>9</sup> PFU/mL, then the reduction is said to be one log10. Herein, we will report the results using this notation.

**Figure 3.** Chemical structures of the phenolic acids reported in the present review.

**Figure 4.** Chemical structures of the flavonoids reported in the present review.

**Figure 5.** Chemical structures of the other active dietary compounds reported in the present review.

**Table 1.** Effect of the dietary compounds reported in the present review, stratified by chemical class and viral target. The principle of action is also reported, as far as known so far.


### *3.1. Phenolic Acids*

Roasted coffee, but not freshly brewed coffee, has been shown to induce the prophage λ in *E. coli* [79]. However, the λ progeny suffered from aberrant replication, and most of the resulting virions were not infective [110]. Therefore, one hypothesis to explain this is that the compounds produced during the roasting process of coffee beans, such as aliphatic carbonyls or volatile substances [73], can cause DNA damage that, in turn, initiates a stress response and the consequent induction of λ prophages. The DNA damage also would explain why the viral progeny, whose genome is a linear double-stranded DNA (dsDNA) molecule, displays a reduced level of infectivity.

Potatoes are commonly used as food worldwide, particularly in the Western diet [111,112]. Potato peel extracts (PPE) contain a mixture of polyphenols (e.g., gallic acid, chlorogenic acid, caffeic acid, and ferulic acid) and flavonoids (such as quercetin and rutin). Exposure of the *E. coli* O157 phages MS2 and Av-05 to 5 mg/mL of PPE for three hours in vitro resulted in a 2.8 and 3.9 log10 reduction, respectively [113]. Hence, Av-05 was more susceptible than MS2 to PPE exposure. The inhibitory mechanism was probably due to interference with the replication stage. Also, tomatoes contain different polyphenols, mainly in leaves and stems. Although the exact composition of these polyphenols varies, gallic acid and chlorogenic acid belong to the most prevalent. Exposure to 5 mg/mL of tomato leaf extract (TLE) for 12 h reduced the infectivity of both MS2 and Av-05; the magnitude of the reduction depended on the tomato subspecies: the *Pitenza* cultivar reduced the infectivity of MS2 and Av-05 by 3.8 and 5 log10, respectively, compared to 0.57 and 1.6 log10 obtained with the *Floradade* cultivar [80]. Even in this case, the inhibitory mechanism was supposed to be linked to viral replication.

Caffeic acid could also inhibit the cytotoxicity of Stx in a Vero-d2EGFP cell-based assay, in a process independent from the alteration of the induction of 933J and 933W [114]. Gallic and caffeic acids at low concentration (around 10–6 mg/mL) and tannins (0.5 mg/mL) reduced the infectivity of PL-1 (infecting *L. casei*) by 80–90% [81]. Others reported that both tannic (0.01–0.1 mg/mL) and gallic (0.1–0.4 mg/mL) acids had negligible action upon the infectivity of MS2, with a reduction that reached a maximum of 0.06 log10 [82].

*Zataria multiflora* is an aromatic plant native from Iran and Afghanistan that is rich in the monoterpenoids carvacrol (or cymophenol) and thymol [115]. A 0.03% *v*/*v* of *Z. multiflora* extracts were bacteriostatic for *E. coli* O157, but sub-inhibitory concentrations reduced the induction of 933 W, measured by quantifying the expression of *Stx2* [83]. Several other compounds, including several derivatives of gallic acid, showed antiviral activity in vitro measured with the MTT method and estimated by the inhibition of viral cytopathic effects [116].

### *3.2. Flavonoids*

Flavonoids also belong to the class of polyphenols. In natural sources, they are usually mixed with other phenolic acids; thus, it is difficult to separate the former's activity from that of the latter. Nevertheless, the active compound of cranberry juice is believed to be proanthocyanidin, a flavonoid [90]. In contrast, the active compounds of pomegranate juice extracts (PJE) were identified in punicalagin, a phenolic acid with antioxidant properties that could also inhibit the influenza virus [117,118]. Flavonoids are classified as antioxidants because they can react with, and remove from the cellular environment, the highly reactive superoxide anions (O2 −) in a process known as scavenging [119]. Flavonoids include two products, catechin and genistein, with peculiar characteristics. Catechin is the basic block of tannins, found in fruit, tea, and wine; genistein is present in many medicinal plants.

Tea extracts were able to inactivate the *Salmonella* phages Felix 01 and P22, without affecting the growth of the bacterial host [84]. Exposure to 35 mg/mL of catechin for 24 h reduced the infectivity of the coliphage T4 by over two log10 in vitro, whereas the host did not show any reduction in population [86]. In addition, derivatives of catechins extracted from green tea could inhibit prophage induction. Epigallocatechin-3-gallate (EGCG) decreased the expression of *Stx1* but increased that of *Stx2* in *E. coli* O157 [87]. Since the expression of these two toxins is associated with the induction of 933 W in a germ-free mouse model [120], it needs to be assumed that, in this situation, EGCG is able to act as a virus inhibitor. The mechanism of action of EGCG involves the repression of the bacterial gene *recA* [87], an effector of the stress response that is central in the induction of 933 J, whereas the induction of 933 W relies on additional pathways not related to the

expression of *recA*. This difference explains why only *Stx1* was reduced upon exposure to EGCG. This nutrient is believed to cause membrane damage that affects the growth of *E. coli* O157 and triggers stress response [87]. Other studies suggested that *Stx1* was still produced upon stimulation with EGCG and gallocatechin gallate (GCG), but the toxin's extracellular release from *E. coli* O157 cultured at 37 ◦C for 24 h was hampered, probably due to both the galloyl and the hydroxyl moieties of these compounds [121].

Tannic acid is known to have antioxidant properties, since it can bind and remove singlet oxygen (1O2) from the cellular environment [122]. A 0.3% *w*/*v* solution of persimmon, a tannin, could induce a 3.13 log10 reduction in the infectivity of MS2. Electron microscopy confirmed that such exposure caused capsid denaturation [123].

Genistein and daidzein extracted from soybeans could protect the genome of phage ϕX174 from degradation induced by nitric oxide (NO) or peroxynitrite (ONOO–). Genistein was more effective than daidzein since a 25 μM solution of these dietary compounds protected about 75% and 45% of the viral ϕX174 DNA molecule confirmed by agarose gel electrophoresis, respectively [89]. This protection might be due to the scavenging properties of the flavonoids [124]. Genistein was also used to protect modified phages containing thymidine kinase derived from Herpes simplex virus during the delivery of this cytotoxin enzyme to tumor cells, thus increasing the targeted elimination of cancer cells [125].

Cranberries are fruits imported from North America and traditionally used by native Americans to treat bacterial infections. Investigations showed that cranberry juice could drastically reduce the growth of *E. coli* O157 in vitro [126]. Exposure for one hour to cranberry juice reduced the infectivity of the coliphages MS2 and ϕX174 by 1.67 and 1.22 log10, respectively, compared to the 0.05 and 0.29 log10 obtained by orange juice, 0.97 and 1.01 log10 obtained by grape juice, and 1.00 and 2.63 log10 obtained by purified proanthocyanidin [90,91]. Experiments with the coliphages T2 and T4 confirmed a complete and immediate loss of infectivity for these viruses when exposed to cranberry juice purchased from food shops [90,95]. Proanthocyanidin is also contained in blueberries; accordingly, exposure of MS2 to blueberry juice for 21 days induced a 6.32 log10 reduction in its infectivity when compared to incubation in phosphate buffered saline (PBS) [127].

In some studies, pomegranate and grape seed juices, which are rich in both flavonoids and phenolic acids, showed an antiviral activity. PJE at a 4 mg/mL concentration displayed a 0.12–0.32 log10 reduction upon MS2 infectivity in vitro [85,92]. This was in the same order of magnitude of other experiments carried out with pomegranate juice applied for 21 days, which showed a 0.14 log10 reduction in MS2 infectivity. Moreover, pomegranate juice diluted in PBS increased the inactivation to 1.84 log10 [127]. MS2 incubated in 1 mg/mL of grape seed extract (GSE) for two hours showed a 1.66 log10 reduction evaluated by plaque assay [93]. GSE also inhibited the growth of non-O157 *E. coli* serotypes, and GSE at a concentration of 4 mg/mL reduced the production of *Stx2* [83,94,128]. By comparison, pomegranate, grape, and orange juices showed lower, albeit still significant, reduction in phage infectivity in vitro [85,127]. In addition, grape seeds, which contain epicatechin, gallocatechin, GCG, and EGCG, could inactivate the cytotoxicity of Stx [114].

Su and colleagues suggested that cranberry juice in general, and proanthocyanidin in particular, inhibits the attachment phase of infecting phages in vitro, possibly via alterating the capsid [91]. This suggestion has been confirmed by electron microscopy analysis, which revealed that T4 treated with cranberry juice did not attach to their host [95]. Moreover, the feline calicivirus 9 showed structural modification of the capsid upon exposure to cranberry juice [91]. Likewise, apple juice, which is rich in procyanidins, increased the resistance of Vero cells against Stx [129].

Propolis ("bee glue", a mixture of the saliva of honey bees with beeswax and plant exudates) contains flavonoids [130]. As mentioned above, it has been shown to specifically induce prophages in *E. faecalis* but not *B. thetaiotamicron* and *S. aureus* [68]. Brazil is the major producer of propolis and this natural substance can be classified according to its color. Green propolis induced a 3.0 log10 reduction in the infectivity of MS2 and 3.5 log10 in Av-08; red propolis was even more effective in reducing PFU, with a 4.2 and 4.0 log10 reduction

for MS2 and Av-08 [96]. The main active molecule of red propolis is formononetin and the suggested mechanism of inhibition was alteration of the structure of the capsid [131].

### *3.3. Saccharides*

Chitosan is a family of polysaccharides present in the exoskeleton of crustaceans and insects as well as in the cell wall of fungi. The members of this family are classified according to their molecular weight [132]. A 0.7% *w*/*v* solution of chitosan applied for three hours could decrease the infectivity of MS2 by up to 2.80 log10 (when using a molecule with a molecular mass of 53 kDa) and 5.16 log10 (when using 222 kDa). By increasing the concentration to 1%, only the 222 kDa form could completely inhibit MS2 [98,133,134]. Higher concentrations of both forms (1.5% *w*/*v*) were needed to achieve the inactivation of ϕX174, albeit the magnitude was much smaller than that of MS2 (0.94 log10). Chitosan was also active against *Bacillus thuringiensis* phage 1–97 A [99] and *Lactococcus lactis* phage c2 in vitro [100]. Furthermore, in vivo experiments with mice showed that chitosan was able to reduce Stx expression and the diffusion of induced 933 W progeny into the tissues, and to improve the lifespan of mice infected with enterohemorrhagic *E. coli* [101]. The mechanism of action was hypothesized to be a structural modification of the capsid [134–136]. Moreover, mutagenic effects of a sucrose-rich diet were reported by Dragsted et al. when investigating the colon of Big Blue rats, a specific strain of Fischer rats that carries 40 copies of the lambda phage on chromosome 4. In this study, a sucrose-rich diet resulted in an increase of mutational frequency in the DNA of these colons [137]. Lysozyme, which is widely distributed among prokaryotes and eukaryotes, is expressed by the R gene of phage lambda. Accordingly, the latter is called bacteriophage lambda lysozyme (LaL), and it has been shown to have bacteriolytic capabilities [138]. In contrast to other lysozymes, however, LaL differs regarding the cleavage of the glycosidic bond between N-acetylmuramic acid and N-acetylglucosamine of bacterial peptidoglycan. Duewel and colleagues showed that high concentrations of β(1→4) N-acetyl-D-glucosamine oligomers inhibit LaL but are not cleaved by the enzyme [138]. A similar observation of degrading peptidoglycans into fragments has also been reported for lysates of phage Vi II [139].

### *3.4. Essential Oils and Vitamins*

Several essential oils show antibacterial and antioxidant activity, together with antiviral function [140]. For instance, oregano, thyme, cinnamon, and allspice (a berry from *Pimenta dioica* used commonly in the food industry) extracts, amongst others, can reduce the growth of *E. coli* O157 [141,142]. A 4% *v*/*v* solution of cinnamon oil, whose main component is cinnamaldehyde, inhibited the growth of *E. coli* O157 in vitro, but sub-inhibitory doses reduced the expression of *Stx2* and the release of viral progeny [72,94]. As in the case of EGCG, the interference over phage induction was accompanied by down-regulation of the effector of the stress pathway *recA*, but also of the quorum sensing (QS) (*qseB*, *qseC,* and *luxS*) and oxydative stress (*oxyR*, *soxR*, and *rpoS*) pathways, as well as the polynucleotide phosphorylase PAP I [94], which is also an inducer of 933 W [143]. These results suggest that cinnamon oil could interfere with 933 W induction as several overlapping levels. Furthermore, cinnamon oil disrupted *E. coli* O156 and *Pseudomonas aeruginosa* biofilms by interfering with the formation of the fimbriae, which are required to make inter-bacterial connections [72,94]. Oregano had a general suppressive action upon prophages, but the effect was stronger in *S. aureus* than in *E. faecalis* or *B. thetaiotamicron* [68]. Eugenol, which is rich in allspice and clove, reduced the induction of both Stx1 and Stx2, and inhibit the growth of *E. coli* O157 in vitro [144].

After a lag phase of few minutes, ascorbic acid (also known as vitamin C), reduced the infectivity of several phages: δA and ϕX174 (with a genome of ssDNA); T7, P22, D29, and PM2 (dsDNA); and MS2 (ssRNA) in vitro [88,102–106]. Supplementation of ascorbic acid with oxidants such as oxygen and hydrogen peroxide enhanced this effect, whereas antioxidants (for instance thiol compounds), nitrogen gas bubbling, or chelating agents suppressed it [102]. It was postulated that the autoxidation of ascorbic acid produced

hydrogen peroxide that damaged the genome of the phages, even though Murata and colleagues found that hydrogen peroxide produced by autoxidation of ascorbic acid did not exert effects on activity of phage δA, in contrast to free radical intermediates [145]. Thus, the scavenging activity provided by thiols and chelating agents was hypothesized to reduce the damage on the viral genome, and the initial delay in the activity of ascorbic acid was interpreted as the time required to internalize this hydrogen peroxide inside the capsid [102]. Subsequent in vitro studies confirmed this hypothesis and showed that ascorbic acid caused the accumulation of nicks in both DNA and RNA genomes, with double-stranded genomes being less affected than single-stranded ones [88,103]. In these studies, the overlap of the nicks determined the formation of double strands breaks, which in fact sometimes appeared after the nicks as a result of the stochastic overlapping of the single-stranded lesions. Furthermore, these damages could be restored by the host's cellular DNA repair system [88].

The oxidized form of vitamin C, dehydroascorbic acid, showed only very limited effects on phage activity and the amount of strand cleavages in ssDNA from phage δA was proportional to ascorbic acid concentration and incubation time. It was significantly increased by Cu2+ or hydrogen peroxide [102,146]. These DNA-damaging properties of the strong reducing combination of ascorbic acid with metal ions (especially Cu2+) [103] can have an impact on the phage population in the intestinal microbiome, but could also have implications in other fields such as the application of high-dose ascorbate in tumor patients. Towards this end, tumor entities like non-small-cell lung cancer and glioblastoma seem to be vulnerable towards the disruption of their intracellular iron metabolism and oxidative damage caused by the formation of hydrogen peroxide and hydroxyl radicals [147].

### *3.5. Other Compounds*

There are very few studies investigating the impact on phages of molecules other than those listed so far. Pioneering work in the late 1950 s demonstrated how hydroquinone and pyrogallol (both derivatives of phenol but not polyphenols) reduced the infectivity of T coliphages [148]. Psoralens belong to the family of furocoumarins, photoactive polyphenols that can induce DNA damage. They are particularly abundant in the peel of limes [149]. Accordingly, a six-hour exposure with lime juice in vitro reduced the infectivity of MS2 by 1.3 log10 even in the absence of photoactivation [107].

Coffee contains not only caffeic acid but also caffeine, an alkaloid; coffee is a beverage on its own and the base for a plethora of soft-drinks [150]. High consumption of coffee and its derivatives has been suggested to confer an increased risk of colorectal cancer, due to its antimicrobial activity that disrupts the intestinal homeostasis [151]. Caffeine is able to induce *E. coli* phage ϕX174 in mitomycin treated *E. coli* cells [108]. Since caffeine is known to distort DNA and cause mutations [152], its activity is supposedly similar to caffeic acid in terms of inducing a stress signal that starts the lytic process.

Finally, even common salt used for meat preservation has been reported to exert effects on phage biology. Towards this end, a 2% *w*/*v* concentration of sodium chloride increased the expression of *Stx2*, as measured by immunoblotting, and the activation of the 933 W, as measured by plaque assay, in *E. coli* O157 [109].

### *3.6. In Vivo Studies*

In contrast to a steadily increasing body of in vitro data that evaluates the interplay of diet or certain nutrients with bacteriophages as discussed in the chapters above, there are still only a few in vivo studies available. However, the possibility to modulate the microbiome by phage application is currently starting to attract more and more attention, especially in the field of inflammatory and malignant diseases. For instance, Zheng and colleagues covalently linked irinotecan-loaded dextran nanoparticles to azide-modified phages that were able to inhibit the growth of *Fusobacterium nucleatum* [153]. After i.v. or oral administration, these phage-guided irinotecan-loaded nanoparticles increased the chemotherapeutic efficacy in mice with colorectal tumors. In another study, a single in-

jection of a lytic bacteriophage cocktail was effective as a rescue treatment for murine severe septic peritonitis, resulting in a significant improvement of the disease state without harming the microbiome [154]. Wild-type phage T4 and the according substrain HAP1, which is characterized by enhanced affinity to melanoma cells, were able to reduce lung metastasis of murine B16 melanoma cells by 47% and 80%, respectively [28]. Moreover, the modulation of the intestinal microbiome and metabolome was investigated using cognate lytic phages in gnotobiotic mice that were colonized with defined human gut commensal bacteria. This approach directly impacted susceptible bacteria, but phage predation also regulated additional bacteria via interbacterial interactions, yielding strong cascading net effects on the gut metabolome [58]. In a gnotobiotic pig model, it was shown that bacteria species are able to affect intestinal morphology as well as the expression of proinflammatory cytokines such as IL-1β and IL-6. Therefore, it can be hypothesized that a modulation of, e.g., neonatal bacterial colonization would have strong implications for a healthy development of the intestine [4]. On the other hand, intestinal inflammation and ulcerative colitis can be aggravated by high levels of certain bacteriophages that induce interferon-γ release [17]. Different phage cocktails (*ShigActive*™ [155] and *ListShield* [156]) have been shown to reduce shigella colonization of the murine gut and to decrease *Listeria monocytogenes* in the gastrointestinal tract, respectively. *ShigActive*™ was found to have comparable therapeutic effects to ampicillin but without the harmful effects on the gut microbiota exerted by the antibiotic [155]. *ListShield* was applied via oral gavage before mice were orally infected with *Listeria monocytogenes*. Consequently, *Listeria monocytogenes* concentrations were found to be reduced in the liver, spleen, and intestines when compared to controls. Even though, this phage therapy was as effective as the treatment with an antibiotic, it did not result in weight loss of the animals in contrast to infected controls and antibiotic-treated mice [156]. In another study, mice with antibiotic-induced perturbed microbiomes were treated with autochthonous virome transfer and viable phages were effective in reshaping the murine gut microbiota in a way that closely resembled the pre-antibiotic situation [157]. In vivo targeting of specific bacterial pathogens with recombinant or wildtype phages was also investigated for *Clostridium difficile* infections [158], Vancomycin-resistant *Enterococcus faecalis* infections [159], Crohn's disease [160] and even for the attenuation of alcoholic liver disease [161]. The human Bacteriophage for Gastrointestinal Health (PHAGE) study and PHAGE-2 study demonstrated that an application of therapeutic doses of bacteriophages was both safe and tolerable [162–164]. The double-blinded, placebo-controlled crossover PHAGE trial with adults consuming bacteriophages for 28 days (32/43 participants finished the study) also demonstrated that bacteriophages are able to selectively decrease the amount of target organisms, without disrupting the gut microbiome globally [162]. In the randomized, parallel-arm, double-blinded, placebo-controlled PHAGE-2 study (68 participants, four weeks), it could be shown that adding supplemental bacteriophages (PreforPro) to the probiotic *Bifidobacterium animalis* subsp. *lactis* enhanced positive effects on gastrointestinal health [164]. Taken together, there is increasing evidence, in initial in vivo studies, for the high potential of treating different diseases with bacteriophages and for the ability to reshape the gut microbiome via tailored phage cocktails. Still, however, more in vivo studies are needed that investigate the complex interplay between diet and bacteriophages, especially in the context of the prevention and treatment of inflammatory diseases and cancer.

### **4. Conclusions**

In conclusion, the present review shows that many dietary compounds and food ingredients display significant bioactivity with documented effects on phages. The dietary compounds discussed in this review can be consumed directly by diet (as in tea or coffee) or indirectly as food supplements. Still, most of the data reviewed and discussed herein pertain to *E. coli* as the, so far, best studied phage target in humans. Although being a common gut commensal, certain serotypes of this species pose a threat to public health regarding severe infectious and (in part systemic) inflammatory conditions as discussed

in the Introduction. A few reports up to now even hypothesized particular *E. coli* strains such as those producing the genotoxin colibactin as potential tumor promoters [165], although their data were restricted to experimental models so far. In own genome data, we have preliminary evidence of sequences of particular *E. coli* strains in human colorectal carcinomas and even metastases (unpublished data based on genomes published in [166]).

The main activity of the dietary compounds discussed in the present review includes inhibitory effects on phages due to the alteration of the capsid, with subsequent reduction of infectivity. In other cases, the viral genome is being damaged, again inhibiting the infectivity of phages. However, some dietary compounds are able to induce (as with common salt or gallocatechin gallate) or repress (as with carvacrol) prophages. More importantly, a few dietary compounds display species-specific activities. For instance, stevia apparently acts as an inducer for *S. aureus* prophages, but not of those present in *E. faecalis,* whereas propolis displays the opposite actions.

Overall, most of the dietary compounds reviewed here, with documented actions towards phages, showed a beneficial effect for the host by interfering with the activity of the pathogens at several levels. Thus, a number of concluding scenaria can be summarized for the putative benefit of nutrients (including the modulation of phages) to human patients and their microbiota (Figure 6A): Several nutritional compounds can directly affect the growth of microbial pathogens, but not that of commensals. Also, dietary compounds are able to inactivate particular toxins produced by pathogens, thus reducing fitness of the latter. More importantly, dietary compounds can inactivate virulent phages, modifying the overall equilibrium of the intestinal microbiome. As a result, phages targeting a commensal species that is a competitor to a pathogen can be removed from the niche. The commensal species will then expand and compete with the pathogen, again reducing the latter's fitness. Finally, dietary compounds might induce prophages present in the pathogen, determining the hosts' lysis and a wave of active virulent phages, which in turn reduce the pathogen's population. Combining all of these inhibitory outcomes will reduce the pathogenicity of invading species and for example, help resolve infections or (chronic) inflammatory conditions.

To better understand how dietary compounds could selectively modulate bacterial infections, we carried out a simulation model (Figure 6b). This model shows that a pathogenic bacterium can wipe out a commensal species, but the selective induction of a prophage can then control the growth of the pathogen, reducing the virulence of the infecting species. The model suggests that it could be possible, in principle, to reorganize the microbial network to fight infections and further disease. Experimental data is required to assess the specificity of particular dietary compounds' action to, effectively and safely, direct such attempts of specific reorganization.

Similar considerations as for phage-directed attempts to counteract infections and inflammatory conditions could be speculated for the field of carcinogenesis and cancer. Interfering with particular commensals within the intestinal microbiome by phages of different activities and properties, with the result of changing the intestinal microenvironment towards a more pro- or anti-carcinogenic condition, could be an exciting novel field of colorectal and other intestinal cancer research, and of treatment development. In parallel, more specific research on particular dietary compounds, chemical components, and associated modulation of phages that exert controllable, specific effects on the microbiome could open exciting new possibilities to interfere with intra-intestinal conditions in ways to foster anti-carcinogenic, more cancer-preventive environments.

**Figure 6.** Overall impact of dietary compounds on infections. Dietary compounds can modify the bacterial population indirectly, based on the interaction between phages and bacteria, and by bacterial competition. (**a**) A dietary compound (nutrient) can interfere with the activity of a pathogenic bacterium at different levels by: inactivating a bacterial toxin (interposed inhibition); inducing a prophage already present in the invading bacterium, which then lyses the pathogen (indirect inhibition); acting bactericidal on invading species (direct inhibition); inactivating a lytic phage of an antagonist commensal species that, freed from the phagial burden, can compete with the pathogen (mediated inhibition). (**b**) Simulation of the interaction between bacteria and phages. The model considers the presence of a commensal bacterium (resident) and its phage. These reach an equilibrium where the number of cells or phages remains constant. A pathogenic bacterium (invader) will have virulence factors that favor its replication. As a result, it overgrows the commensal species. The activation of phages, namely through dietary-mediated induction of prophages, reduces the replication rate of the invader and re-establish, as a result, the commensal population. For the simulation, the parameters used were as follows. Carrying capacity: 2.2 <sup>×</sup> 107. Maximum growth rate, 0.47 (commensal), and 0.72 (invader). Phage adsorption rate: 10–9, Phage lyse rate: 1.0. Phage burst size: 50. Particle loss rate: 0.05. Initial population of commensal bacteria: 50 000 cells. Amount of invader bacteria inoculated: 500 cells. Amount of phages: 1000 particles each. The model was implemented in *Julia* language using the *DifferentialEquations* package.

**Author Contributions:** L.M., S.V., and H.A. conceptualized this interdisciplinary review. H.A. conceptualized the idea of linking phages, and their contributions to the microbiome and microenvironment, to cancer, carcinogenesis, progression, and metastasis. L.M. did the statistical analyses and carried out the literature research. L.M. and M.B. designed the figures. All of the authors (L.M., M.B., S.V., and H.A.) wrote and carefully revised the manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** H.A. was supported by the Alfried Krupp von Bohlen und Halbach Foundation, Essen; the Deutsche Krebshilfe, Bonn (70112168); the Deutsche Forschungsgemeinschaft (DFG, grant number AL 465/9-1); the HEiKA Initiative (Karlsruhe Institute of Technology/University of Heidelberg collaborative effort); the DKFZ-MOST Cooperation, Heidelberg (grant number CA149); the HIPO/POP-Initiative for Personalized Oncology, Heidelberg (H032 and H027). S.V. and M.B. were supported by a grant from the Else-Uebelmesser-Stiftung (grant no. D.30.21947) and PASCOE Pharmazeutische Praeparate GmbH. We further acknowledge support by Open Access Publishing Fund of University of Tuebingen.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Acknowledgments:** We would like to thank Maurizio Grilli, Library of the Medical Faculty Mannheim, Ruprecht Karls University of Heidelberg, Germany, for his support in the literature search of this review. A special thanks goes to Ingo Plag, Institute for English Language and Linguistics of the University of Düsseldorf, for his advice on the adjective 'phagial', which is still not officially recognized.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


## *Article* **The Immune Landscape of Colorectal Cancer**

**Artur Mezheyeuski 1,\*, Patrick Micke 1, Alfonso Martín-Bernabé 2, Max Backman 1, Ina Hrynchyk 3, Klara Hammarström 1, Simon Ström 1, Joakim Ekström 1, Per-Henrik Edqvist 1, Magnus Sundström 1, Fredrik Ponten 1, Karin Leandersson 4, Bengt Glimelius <sup>1</sup> and Tobias Sjöblom <sup>1</sup>**


**Simple Summary:** We sought to provide a detailed overview of the immune landscape of colorectal cancer in the largest study to date in terms of patient numbers and analyzed immune cell types. We applied a multiplex in situ staining method in combination with an advanced scanning and image analysis pipeline akin to flow cytometry, and analyzed 5968 individual multi-layer images of tissue defining in a total of 39,078,450 cells. We considered the location of immune cells with respect to the stroma, and tumor cell compartment and tumor regions in the central part or the invasive margin. To the best of our knowledge, this study is the first comprehensive spatial description of the immune landscape in colorectal cancer using a large population-based cohort and a multiplex immune cell identification.

**Abstract:** While the clinical importance of CD8+ and CD3+ cells in colorectal cancer (CRC) is well established, the impact of other immune cell subsets is less well described. We sought to provide a detailed overview of the immune landscape of CRC in the largest study to date in terms of patient numbers and in situ analyzed immune cell types. Tissue microarrays from 536 patients were stained using multiplexed immunofluorescence panels, and fifteen immune cell subclasses, representing adaptive and innate immunity, were analyzed. Overall, therapy-naïve CRC patients clustered into an 'inflamed' and a 'desert' group. Most T cell subsets and M2 macrophages were enriched in the right colon (*p*-values 0.046–0.004), while pDC cells were in the rectum (*p* = 0.008). Elderly patients had higher infiltration of M2 macrophages (*p* = 0.024). CD8+ cells were linked to improved survival in colon cancer stages I-III (q = 0.014), while CD4+ cells had the strongest impact on overall survival in metastatic CRC (q = 0.031). Finally, we demonstrated repopulation of the immune infiltrate in rectal tumors post radiation, following an initial radiation-induced depletion. This study provides a detailed analysis of the in situ immune landscape of CRC paving the way for better diagnostics and providing hints to better target the immune microenvironment.

**Keywords:** colorectal cancer; multiplex; tumor immunology; immune landscape

### **1. Introduction**

Cancer remains one of the leading causes of death worldwide and CRC is the third most common cancer type and the second most common cancer killer [1]. In addition to the traditional TNM classification system, molecular subgroups based on mutations and gene expression profiles are used to identify more homogeneous subgroups as CRC is intrinsically heterogeneous [2]. In particular, somatic mutations in driver genes, such as those of

**Citation:** Mezheyeuski, A.; Micke, P.; Martín-Bernabé, A.; Backman, M.; Hrynchyk, I.; Hammarström, K.; Ström, S.; Ekström, J.; Edqvist, P.-H.; Sundström, M.; et al. The Immune Landscape of Colorectal Cancer. *Cancers* **2021**, *13*, 5545. https:// doi.org/10.3390/cancers13215545

Academic Editor: Heike Allgayer

Received: 17 October 2021 Accepted: 2 November 2021 Published: 4 November 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

the RAS pathway, have major clinical implications for the response to specific therapies and molecular testing for such mutations is now a clinical standard in metastatic CRC (mCRC). During the past decade, a classification system based on the tumor immune environment has attracted attention. Galon et al. introduced an immune score that grouped CRCs with regard to the infiltration of T cells (CD3+ and CD8+ lymphocytes) in the tumor and the invasive margin [3]. The Immunoscore® provides independent prognostic information in addition to other clinical parameters including the TNM classification in CRC stage I–III [4,5]. Furthermore, not only the T cell lineage, but also the presence of other immune cell types including B cells and NK cells have been associated with better outcomes [6,7]. On the other hand, certain immune contexts of the primary tumor dominated by immune suppressive cells, like T regulatory cells or M2 type macrophages, were connected to tumor progression and poor prognosis [8]. These observations indicate an active involvement of the tumor immune environment in tumorigenesis and suggest a diagnostic, prognostic, and, potentially, also predictive value of a deeper immune classification of CRC.

The introduction of immune checkpoint inhibitors has demonstrated that cancer immunity can be modified, leading to immune-mediated long-lasting tumor regression in subsets of patients with several different solid tumor types [9]. Further, the pre-existing microenvironment seems of major relevance and high infiltration with immune cells is associated with better tumor response and long-term survival in patients treated with checkpoint inhibitors [10]. Transcriptomic analyses have revealed that these tumors also express inflammatory and effector cytokines, indicating a basic anti-tumor immune response, though not efficient enough to control tumor growth. This immune phenotype has been designated 'inflamed' or 'hot'. In contrast, tumors with less immune cell infiltrate were designated as 'desert' or 'cold' tumors [11,12]. In CRC, the 'inflamed' immune phenotype is often found in tumors with high microsatellite instability (MSI-H), most probably due to high tumor mutational load and the presentation of neoantigens leading to anti-cancer immunity [13,14]. The analysis of tumor exomes allows identification of such neoantigens. The number of mutations per exome ranges from ~100 in microsatellite stable (MSS) to ~1000 in MSI CRC [15–17]. Checkpoint inhibitor therapy is effective in these tumors [18,19] and is now approved for mCRC with MSI-H [18–21]. Taken together, there is evidence that the tumor immune microenvironment plays a major role in terms of CRC prognosis and, at the same time, indicates whether immune modulating treatment is beneficial.

Despite its obvious clinical relevance, knowledge of the immune microenvironment in CRC is fragmentary as most studies have focused on only a single cell type or a few subsets of immune cells. The most applied strategy is based on immunohistochemical analysis with semi-quantitative measurements, carrying a substantial risk of observer bias. Multiple markers may be analyzed in consecutive sections, but this has limited relevance when evaluating cell interactions [22]. Therefore, the focus of prior immunohistochemical studies was on the T cell lineage. More comprehensive studies rely on deconvolution of gene expression data, without spatial context of immune cells. This approach has disadvantages, as low abundance cell types are challenging to quantitate accurately in bulk mRNA profiles. Given these methodological difficulties, there are few comprehensive efforts towards in situ mapping of the tumor microenvironment [23]. However, novel immunofluorescence multiplex techniques in combination with advanced scanning and image analysis systems can tackle these obstacles to describe the immune response in cancer in a holistic and standardized manner [24].

The aim is to apply immunofluorescence multiplexing techniques to provide the first comprehensive overview of the immune landscape across a large population-based cohort of CRC patients. Relevant molecular and clinical subgroups are analyzed using antibody panels allowing in situ identification of 15 distinct subclasses of immune cells in association with clinical parameters and outcome. Finally, we compare the immune status in rectal tumors treated with different therapies and intervals prior to surgery to identify therapy-induced modulation of CRC immunity.

### **2. Materials and Methods**

### *2.1. Study Cohort*

The study cohort consists of prospectively collected CRC patients living in the Uppsala region of Sweden, most of which have been included in the Uppsala-Umeå Comprehensive Cancer Consortium (U-CAN) [25]. In total, 937 patients were diagnosed with CRC between 2010 and 2014 in the Uppsala region. Of them, 746 (80%) were included in tissue microarray (TMA). For the present study, only patients with TMA material from primary tumors were selected. After the staining procedures and quality control, 536 patients were available for analysis. The clinicopathological characteristics of the included patients and their tumors are presented in Table S1.

All patients received stage-stratified standard of care according to the Swedish national guidelines from 2008. According to the guidelines, colon tumors were recommended primary surgery and adjuvant chemotherapy if risk-factors for recurrence were present. If the colon tumor was considered inoperable, neoadjuvant chemotherapy was administered to shrink the tumor before surgery. Rectal tumors were grouped into three prognostic categories: early (low recurrence risk), intermediate (intermediate recurrence risk), and locally advanced (high recurrence risk) with recommendations of primary surgery or pre-operative radiotherapy or chemoradiotherapy with different time-intervals to surgery, dependent on group belonging. Formalin-fixed paraffin-embedded tissue blocks of primary tumors and distant metastases were used to construct TMAs. Each case was represented on the TMA with cores derived from the central part of the tumor and from the invasive margin. The study was approved by the regional ethical committee in Uppsala, Sweden (Dnr 2010/198 and Dnr 2015/419).

MSI status was evaluated in available cases by IHC analysis with antibodies against the two MMR proteins, PMS2 and MSH6. The tumor was denoted as MSI-H if at least one of these proteins was absent.

### *2.2. Multiplex Immunofluorescence Staining*

For the multiplexed immunofluorescence staining, 4 μm thick sections were deparaffinized, rehydrated, and rinsed in distilled H2O. Three staining protocols were established with three panels of antibodies: a lymphocyte panel, with CD4, CD8, CD20, FoxP3, CD45RO, and pan-cytokeratin (pan-CK) (as described in [26]); a NK/macrophage panel encompassing CD56, NKp46, CD3, CD68, CD163, and pan-CK; and a dendritic cell panel with CD3, CD1a, CD208, CD123, CD68, CD15, and pan-CK. The staining procedure was performed as described [27,28]. In total, 520 cases were evaluable for the lymphocyte panel, 508 cases for the NK/macrophage cell panel, and 498 cases for the dendritic cell panel (Table S1). Using a combination of immune markers, we quantified 15 immune cell subclasses (Figure 1a,b).

### *2.3. Imaging, Image Analysis, Thresholding and Immune Scores*

The stained TMAs were imaged using the Vectra Polaris system (Akoya Biosciences, Marlborough, MA, USA) in a multispectral mode at a resolution of 2 pixels per μm. This resulted in 5968 individual multi-layer images, each representing a TMA core. Spectral deconvolution and initial image analysis were conducted in the inForm (2.4.6) software (Akoya Biosciences) (Figure S1). Each of the images was reviewed and manually curated by a pathologist to exclude artefacts, staining defects, and accumulation of immune cells in necrotic areas and intraglandular structures. The vendor-provided machine learning algorithm was trained and applied to split tissue into three categories: tumor compartment, stromal compartment, or blank areas as described [29]. Cell segmentation was performed using DAPI nuclear staining as described [27,28]. The perinuclear region at 3 μm (6 pixels) from the nuclear border was considered the cytoplasm area. The nuclear or cytoplasmic area was evaluated for the expression of nuclear or cytoplasmic/membrane markers, respectively. The cell phenotyping function of the inForm software was used to manually define cells positive to each of the markers. The intensity of the marker expression in

selected cells was used to set the thresholds for marker positivity. The defined thresholds were then applied to the raw output data of the complete cohort outside the inForm pipeline. Every cell was characterized as positive or negative for each marker in the panel, and marker co-expression was used to define immune cell subtypes (Figure 1a,b). Immune cell infiltration was evaluated as the number of cells per analyzed tissue area, in the stromal compartment and tumor compartment. This algorithm was applied to quantify the 15 different immune cell subclasses in the stroma and tumor compartment in the center of the tumor and in the invasive margin, i.e., obtaining a cell quantification in four tissue regions. Immune scores were generated for each immune cell subclass. First, immune cell infiltration in each of the four localizations was dichotomized into 0 (low) and 1 (high), using the median as threshold. The sum of the values gives a score between 0 and 4 (Figure S1e).

**Figure 1.** Characterization of immune cell subsets in the tumor and stroma compartments at the invasive margin and core of the tumor of primary CRC. (**a**) Representative images of the multiplex staining with three immune panels; (**b**) scheme of the immune marker combinations used to define the subgroups of immune cells; (**c**) immune cell densities in Tumor and Stroma compartments in Central Tumor (CT) and Invasive Margin (IM) (boxes show median values and interquartile range, and numbers represent cell counts per mm2, cube root transformed); and (**d**) illustration of the mean immune cell infiltration in tumor center in tumor and stromal compartment.

### *2.4. Statistics*

Statistical analyses were performed using R (version 3.5.1) and SPSS V20 (SPSS Inc., Chicago, IL, USA). In radically operated stage I–III patients, recurrence-free survival (RFS) was computed as the time from surgery to the first documented disease progression including local recurrence or distant metastases or death due to any reason, whichever

occurred first [30]. Overall survival (OS) was the time from surgery to death due to any reason. To estimate relative hazards in both univariate and multivariable models, a Cox proportional hazards model was used. Hierarchical clustering analyses were conducted with the heatmap.plus package (version 1.3) in R. For the analyses of associations between MSS/MSI status and metastases type, the Chi-square test was used. The Ward algorithm was used for hierarchical clustering and *p* < 0.05 was considered statistically significant. The Benjamini–Hochberg procedure was used to adjust for multiple hypothesis testing, and adjusted q-values were reported.

### **3. Results**

### *3.1. Identification and Quantification of Immune Cell Subclasses in CRC by Multiplex Staining*

The successfully stained tissue microarray cores comprised of 536 surgically removed primary CRC cases with two cores from each tumor, representing the invasion margin and the central tumor area. Two thirds of the patients had colon cancer while one third rectal cancer, 54% of the patients were male and 35% were older than 70 years. In total, 59% of patients had stage III disease and 15% had metastatic disease at diagnosis. Most colon cancer patients (99%) were therapy-naïve at the time of surgery, while many rectal cancer patients had received pre-operative treatment (61%) (Table S1). The TMAs were stained with three different panels of immune markers along with pan-cytokeratin and DAPI as nuclear stain. Examples of multiplex immunofluorescent images are shown in Figure 1a and the analysis pipeline is illustrated in Figure S1. The expression of different immune markers was combined to assign each cell to one of 15 immune cell subtypes (Figure 1b), including different lymphocytes, macrophages, natural killer (NK) cells, dendritic cells (DCs), and myeloid cells (Figure 1b). The density of immune cell subtypes was annotated in the stroma and tumor compartments of the tumor center and at the invasive margin, resulting in four different metrics for each immune cell class. Overall, the most abundant cell types were CD8 single positive cells and M1 macrophages with median (mean) values of 314 (832) and 431 (685) cells per mm2, respectively. NK cells and NKT cells demonstrated very low overall density with 77 and 81% of the cases being negative, respectively (Figure 1c,d). Taken together, the infiltration of immune cells was highly variable between tumors and immune cell subclasses, spanning from 0 to 11,994 cells/mm2.

### *3.2. Spatial Distribution of Immune Cells in CRC*

Next, we performed case-wise comparisons between the four tissue regions (stroma and tumor in the tumor center and stroma and tumor in the invasion margin). When comparing infiltration in stroma against tumor compartments (Figures 1d, S2 and S3), most immune cell subsets were more abundant in the stroma. Only CD8 single positive cells (see Figure 1B for immune cell sub-classification) and myeloid cells were more numerous in the tumor compartment (q < 0.001, Mann–Whitney U test with Benjamini–Hochberg correction). The distribution of immune cells between the center of the tumor and the invasive margin were similar, with a few notable exceptions. The most striking difference was observed for T cells, which were more abundant in the tumor center (Figure S4). In conclusion, there was greater enrichment of immune infiltrate in the stroma compared to the tumor cell compartment, but no significant differences between tumor center and invasive margin.

### *3.3. Interrelationship of Immune Cells and Immune Scores*

We hypothesized that immune cells of the same lineage infiltrate tumor tissue in a coordinated fashion. Therefore, we correlated the abundance of all immune cell subtypes to each other in the four analyzed tumor regions (Figure S5). Indeed, the correlations for each specific immune subclass between the four tissue regions of the same tumor sample were in generally high. Due to this observation, we summarized the immune cell values in a single score for each immune cell subclass. These scores were generated in analogy to the original Immunoscore® [3] by summarizing the cell densities in all four regions into one score ranging from 0 to 4 (Figure S1e). Analysis using the immune scores revealed interrelations between different immune cells, with the highest correlations between lymphocyte subtypes, and between M1 macrophages and CD8+ lymphocytes. Interestingly, NK and NKT cells correlated negatively to mature dendritic cells and plasmacytoid dendritic cells (mDCs and pDCs). There was also a negative correlation of T cells to myeloid cells and M2 macrophages (Figure 2a). In conclusion, we identified distinct dominating immune infiltration patterns when a set of immune cell subclasses infiltrate tumor tissue coordinately.

**Figure 2.** The immune scores interrelations, distribution across different clinical and pathological groups and unsupervised hierarchical clustering. (**a**) Graphical representation of Spearman's correlation matrix between immune scores. Pie charts and the intensity of shading represent the strength of correlation (Spearman correlation coefficient), blue color indicates direct while red color indicates inverse correlation. Asterisks indicate statistical significance (*p* < 0.05). (**b**) Immune scores mean levels (black line) and 95% confidence intervals (pink areas limited by gray lines) at specific primary tumor locations. For additional data, see Table S2. (**c**) Unsupervised hierarchical cluster analysis of immune scores. Cases were clustered based on the levels of immune scores. A total of 373 cases with complete immune score data from therapy-naïve patients were available. Clusters with enriched CD4 or CD8 cells are marked by dashed black line, while the cluster with low lymphocyte level is marked by dashed red line. For additional data, see Table S3.

### *3.4. Clustering of CRC Cases by Immune Cell Scores and Relation to Clinical Parameters*

We next evaluated whether the immune scores were related to clinicopathological parameters. The findings largely replicated associations observed in region-restricted immune cell densities (Table S4). In line with published data [31,32], tumors of the right colon were characterized by higher immune scores for most T cell subclasses, M2 macrophages, and myeloid cells in comparison to the left colon and rectum, while pDC cells were enriched in the rectum. The most abundant immune infiltrates were seen in the tumors from flexura hepatica and colon transversum (Figure 2b and Table S2). Higher immune scores of CD8 single positive cells were observed in tumors with lower N stage. Most T cell subclasses,

macrophages, and myeloid cells were enriched in MSI-H tumors. The M2 macrophages were associated with higher patient age.

To capture the dominating immune landscapes, we performed hierarchical clustering based on immune scores across all 373 therapy-naïve cases. The cluster analysis revealed two distinct groups (Figure 2c). The smaller cluster (*n* = 145) included tumors with high immune scores for T cells, reflecting an 'inflamed' phenotype. Interestingly, this cluster consisted of two distinct subgroups with high CD8 or CD4 scores. The second, larger cluster (*n* = 228), demonstrated low immune scores for T cells, representing the immune 'desert' phenotype. Within this cluster, several smaller subgroups were observed with either increased M2/myeloid cell scores, NK/NKT cell scores, immature dendritic cell (iDC) scores, or mDC/pDC scores. The 'inflamed' cluster was enriched with tumors (i) from the right colon, (ii) with high differentiation grade, (iii) without neural invasion, and (iv) with MSI. Other parameters, such as stage, sex, vascular engagement, local lymph node involvement, presence of distant metastases, or BRAF mutation status did not affect the distribution across the main clusters (Table S3). Interestingly, when we analyzed the impact on OS, the 'inflamed' and 'desert' immune clusters did not demonstrate significant differences. Taken together, tumors with an 'inflamed' or a 'desert' immune phenotype were clearly distinguishable, although, unexpectedly, not associated with improved or reduced survival.

### *3.5. Immune Scores and Survival*

Clearly defined 'hot' and 'desert' tumors did not have significant survival differences. Therefore, we hypothesized that individual variations in different immune cell subsets may play more important roles in predicting patient survival and focused on the analyses of single immune scores. In a first set of survival analyses, we evaluated OS in all therapy naive patients (Figure 3a and Table S5); since preoperative treatment may influence the immune scores, these analyses were restricted to untreated patients. In line with previously published data [32], T cell immune scores had positive associations with improved survival, but only the immune score for CD8 single positive cells reached statistical significance (HR = 0.64, 95%CI [0.49–0.84], q = 0.014). In contrast, higher M2 macrophage scores were associated with shorter survival (HR = 1.50, 95%CI [1.20–2.00], q = 0.014). Due to the heterogeneity of CRC, both in terms of natural course of the disease and treatments, we investigated survival in specific patient subgroups with relevant endpoints. The same survival impact of CD8 cells and M2 macrophages was seen for radically operated stage I–III colon cancer patients, when disease-free survival (RFS) was analyzed. However, in the multivariable analysis, adjusted to clinicopathological factors, only single positive CD8 cells had a significant impact on prolonged RFS (HR = 0.64, 95%CI [0.41–0.98], *p* = 0.039) (Figure 3b and Tables S6 and S7). Subsequently, we evaluated stage IV patients separately. Immune scores for single positive CD4, single positive CD8 cells, and mDCs were associated with longer survival in the univariable analyses, although only the first retained statistical significance after adjustment for multiple testing (Figure 3c and Table S8). Thus, survival analysis in a therapy-naïve cohort and in colon cancer stage I–III confirmed previous findings indicating a major impact of CD8+ cells. In stage IV, single positive CD8+ were accompanied by mDCs and even stronger survival-predictive impact of single positive CD4+ cells.

**Figure 3.** Immune scores predict patient survival. Forest plot of hazard ratios (HR) for immune scores in the univariable and multivariable Cox regression models. Filled squares indicate HR and whiskers represent 95% CI. Blue-colored squares indicate statistically significant (*p* < 0.05 and, where applicable, FDR q < 0.05) associations of the respective immune score with improved survival, while red squares represent association with reduced survival. Blue-colored squares with black contour indicate that the association was statistically significant in an individual test (*p* < 0.05) but lost statistical significance after adjustment for multiple testing (FDR q ≥ 0.05). (**a**) Univariable associations of immune scores with OS in a complete cohort of therapy-naïve patients. For detailed information see Table S5. (**b**) Association of immune scores with RFS in stage I–III colon cancer. Left panel illustrates the result of the univariable Cox regression models. Right panel illustrates the result of the multivariable Cox regression model, adjusted to clinicopathological parameters: pT, pN stages, tumor differentiation, patient age, surgery type (elective or acute), and adjuvant treatment. For detailed information see Supplementary Tables S6 and S7. (**c**) Univariable associations of immune scores with OS in stage IV therapy-naïve colorectal cancer patients. For detailed information, see Table S8.

### *3.6. Rectal Cancer*

Rectal and colon cancer are often considered separate diseases [33]. This is also reflected by the different immune phenotype observed with lower numbers of CD4 single positive cells, CD4 Treg cells, and higher mDC, pDC, NKT cells in therapy-naïve rectal cancer compared to colon cancer (Table S2, Figure 4a), suggesting a lower level of natural immune activation in rectal cancer patients. Since most rectal cancer patients receive neoadjuvant radiotherapy or chemoradiotherapy (RT/CRT), we analyzed samples from 78 patients treated with RT/CRT. These patients were dichotomized regarding neoadjuvant treatment type, that was either (i) short-course RT (5 × 5 Gy in one week) followed by immediate surgery or (ii) short-course RT with delayed surgery (later than three weeks), CRT with delayed surgery, or short-course RT and chemotherapy in the interval to surgery. The analyses revealed that many immune cell counts decreased in the group that received RT/CRT therapy and was operated soon after the treatment and increased again in the delayed surgery group (Figure 4b), with the most characteristic profile seen for CD4+CD45RO+, CD8+CD45RO+, CD8 regulatory cells, M2 macrophages, iDCs, and pDCs. Interesting, CD4 and CD8 single cells, as well as B cells, showed quite a stable level of infiltration independent from neoadjuvant treatment type. None of the immune cell subclasses showed statistically significant differences when comparing tumors

from primary surgery and those from the delayed surgery group, with the exception of M1 macrophages which demonstrated lower densities in the pretreated delayed surgery group. Taken together, the immune profiles differ between rectal and colon cancers in several aspects. In rectal cancer, the pattern reflects repopulation of the immune infiltrate in tumor tissue post radiation, following an initial radiation-induced depletion.

**Figure 4.** Immune infiltration in rectal cancers is restored after RT/CRT pre-treatment and delayed surgery, while vasculature is changed. (**a**) Radar plots of immune scores in therapy-naïve colon cancer patients (green) and rectal cancer patients (brown). (**b**) Immune infiltrate levels for patients who had primary surgery (white), radiation therapy followed by immediate surgery (<21 days), or delayed surgery after (chemo)radiotherapy. Numbers represent cell counts per mm2, cube root transformed. Boxes show median values and interquartile range of the ratios, whiskers represent 1.5 IQR. Wilcoxon signed-rank test with Pratt method assuming asymptotic distribution was used for statistical analysis. Statistically significant differences: \* *p* < 0.05, \*\* *p* < 0.01 and \*\*\* *p* < 0.001; not statistically significant differences: n.s.

### **4. Discussion**

Tumors are composed of malignant cells and host elements of the tumor microenvironment which can support or suppress tumor progression and influence anti-cancer treatment. Although T cells have been considered as the most important anti-tumoral immune cells, detailed analysis of T cell subtypes and of other immune cells classes has been limited due to methodological difficulties. The functions of different immune cells can vary dramatically, depending on their activation and differentiation status. These differences are reflected in unique protein expression profiles, requiring techniques for multiplex in situ analysis to enable quantification of immune cell classes in clinical samples [34]. This study describes the immune cell microenvironment of CRC with 15 subgroups of immune cells at a hitherto unrivaled resolution. We applied a multiplex in situ staining method in combination with an advanced scanning and image analysis pipeline, akin to flow cytometry in situ, and analyzed 5968 individual multi-layer images of tissue defining a total of 39,078,450 cells. Each image was reviewed and thoroughly curated by a pathologist to exclude artefacts, staining defects, and necrotic areas. Furthermore, we considered the location of immune cells with respect to the stroma and tumor cell compartment as well as tumor regions in the central part or the invasion margin. To the best of our knowledge, this study is the first comprehensive spatial description of the immune landscape in CRC using a large population-based cohort and a multiplex immune cell identification.

In addition to commonly analyzed immune cells, like CD4 and CD8 cells, or FoxP3+ cells, we could accurately discriminate additional subsets of T lymphocytes. This increased the depth of cell sub-classification and, at the same time, improved the purity of each cell class. For instance, in conventional immunohistochemical analysis, FoxP3 positive cells have usually been considered regulatory T cells. Our approach refined cell counting by excluding FoxP3+ cells of non-lymphocyte or unknown origin, e.g., cancer cells or immune cells negative for other markers [28,35]. Furthermore, we found that a large

proportion of the FoxP3+ cells are of the CD8+ lineage. CD8+FoxP3+ T cells have previously been suggested to be immunosuppressive CD8+ Tregs [36], although conflicting data exist showing that FoxP3 may be induced upon CD8+ T cell activation [37]. The unexpected high abundance of the specific cell type observed here should be the subject of further investigations.

We could clearly identify two distinct immune phenotypes: immune 'inflamed', characterized by high infiltration of lymphocytes, and immune 'desert' tumors. Interestingly, the 'inflamed' cluster in our analysis consisted of two subgroups with tumors with either CD4+ or CD8+ infiltration, and with only a small group of cases with concurrent high CD4 and CD8 levels. One may speculate that this finding might explain the general resistance of CRC to immune checkpoint inhibitors, considering that the presence of both immune cell linages is necessary for effective cancer cell elimination. Finally, the presence of dendritic cells, cells of the myeloid lineage, and NK cells define further subgroups within the immune desert background. Taken together our analyses refine the immune classification of CRC.

Despite being expectedly associated with dMMR/MSI-H cases, immune 'inflamed' tumors did not demonstrate a statistically significant association with improved patient survival. We next extended our analysis of individual immune cells in the context of clinical outcome. In an objective and unbiased analysis, previously reported relations were confirmed, but we could also uncover new information about the prognostic impact of further immune cell subclasses. Thus, Immunoscore®, which considers amount and localization of CD3 and CD8 cells showed an independent prognostic impact in a large multicenter prospective study [5]. Our CD8 immune score, although generated slightly differently from the Immunoscore®, had prognostic value in this cohort. Another immune cell subclass which emerged as a potent prognostic biomarker was the M2 macrophages. These cells have a broad and not yet fully understood role, but can be considered as protumoral elements and hallmarks of an immunosuppressive microenvironment [38]. Our findings suggest that the adverse effect of M2 macrophages should be considered in efforts to improve the prognostic accuracy of immune scoring systems.

Finally, we evaluated changes in the immune microenvironment of primary rectal cancer tissue subjected to preoperative RT/CRT. Our results demonstrated immune deprivation in tumor tissue undergoing resection directly after irradiation. The local immunosuppressive effect of irradiation is well established, and diverse radio-sensitivity of different immune cell types has been reported (reviewed in [39]). In agreement with these reports, we observed lower cell counts for all evaluated cell types. While there are several studies describing the immediate effect of RT/CRT on the tumor microenvironment, data about delayed effects, after weeks or months, are largely missing. Here, tumors resected after a delay following RT or CRT were characterized by an immune microenvironment largely similar to non-irradiated tumors. Accordingly, the immune suppressive effect of therapy should be considered when combinations of immune and conventional therapy are planned and may give a rationale for the sequencing of different therapy modalities in clinical trials. However, this simplistic explanation is complicated by the fact that the tumors that received or did not receive neoadjuvant treatment were not randomized, but rather selected according to stage and other characteristics on magnetic resonance imaging and type and intensity of therapy varied. The patients of the three analyzed groups, i.e., primary surgery, preoperative RT followed by immediate surgery, and preoperative RT/CRT followed by delayed surgery, are not comparable with regards to clinicopathological characteristics. With this caveat, the causes for the reported observations may be more elaborate than a direct link between RT/CRT and immune cell count. Overall, early (or so called 'good') rectal tumors [40] were operated immediately and had lower stages and few other risk factors (like extramural vascular invasion) than intermediate or 'bad' tumors subjected to RT and immediate surgery. Further, the group of locally advanced tumors receiving preoperative RT/CRT and delayed surgery (so called 'ugly' tumors) usually represent even more advanced tumor forms: stage cT4a/b or cT3 tumors with threatened/involved mesorectal fascia. Taking this together, the immune characteristics can not only be compared between neoadjuvant treatment groups, but also need to be normalized to their non-pretreated counterparts, with respective T and N-stages. With this background, using data presented in Table S4 as reference, one could expect 'ugly' tumors to have lower levels of immune infiltration. However, our data demonstrate that 'ugly' tumors after RT/CRT were immunologically comparable with non-pretreated 'good' tumors for most of the immune cells (except M1 macrophages). Therefore, we hypothesize that RT/CRT may convert 'ugly' tumors into immunologically 'good' ones. Although intriguing, this interpretation should be considered with caution because the number of cases is relatively small and due to the absence of proper non-treated reference tissue for RT/CRT cases. Further studies, involving patients randomized with regards to pre-operative treatment (in terms of the type of treatment and of the timing prior surgery) are therefore warrantied. Ideally, such studies should include sampling before the neoadjuvant treatment.

### **5. Conclusions**

In conclusion, to the best of our knowledge this study is the largest in terms of patient numbers and analyzed immune cell subclasses in CRC. We provide a detailed un-biased overview of the in situ immune landscape of CRC and were able to confirm but also extend the concept of cancer immunity. Many of the observations may have clinical relevance for CRC patients by paving the way for better cancer diagnostics or by providing hints to better target the immune microenvironment therapeutically. The applied multiplex technique and the analysis pipeline are applicable on common diagnostic tissue samples; therefore, it is possible that a comprehensive analysis of the immune microenvironment will become a part of the future clinical routine in the era of immunotherapy.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/ 10.3390/cancers13215545/s1. Figure S1: flow chart illustrating the analytical pipeline: (a) original multispectral image 'mixed' channels; (b) multispectral multilayer image with separated channels; (c) schematic illustration of the image processing related to compartment segmentation. The Tumor (brown), Stroma (green) and Non-tissue (yellow) compartments are segmented, based on machinelearning image analysis approach. Areas of necrosis and artefacts were marked for exclusion from further analysis (gray). Immune cell infiltrates were analyzed or in Tumor and Stroma compartment separately. (d) Individual cell segmentation. (e) Generation of immune scores for each of 15 immune cell subclasses. Figure S2: pairwise comparison of immune cell densities in Tumor and Stroma compartments. Figure S3: illustration of the mean immune cell infiltration in invasive margin in tumor and stroma compartment. Numbers represent cell counts per mm2, cube root transformed. Figure S4: pairwise comparison of immune cell densities in CT and IM. Figure S5: cross-correlation between the abundance of immune cells. (a) Graphical representation of the Spearman's correlation matrix of the abundance of immune cell subclasses. Circle size represents the strength of correlation, blue color indicates direct while red color indicates inversed correlation. (b, c, d, e) Graphical representation of the Spearman's correlation matrix of the abundance of immune cell subclasses only in: (b) stroma compartment in invasive margin; (c) tumor compartment in invasive margin; (d) stroma compartment in tumor center; (e) tumor compartment in tumor center, Table S1: baseline clinicopathological characteristics. Values are the number (percentage) unless indicated otherwise. Percentages may not add to 100% due to rounding. Table S2: immune scores in tumors with different clinical and pathological characteristics. Chi-square test was used for statistical analysis. Table S3: distribution of clinicopathological parameters in tumors within 'inflamed' and 'Immune desert' clusters. See also Figure 2c. Table S4: differences in immune cell distribution across cancer samples from patients with different clinicopathological characteristics. The direction of the association is illustrated by the location of the asterisks. \*-*p* < 0.05; \*\*-*p* < 0.01; \*\*\*-*p* < 0.001. Table S5: univariable associations of immune scores with OS in a complete cohort of therapy-naïve patients. See also Figure 3a. Table S6: association of immune scores with RFS in stage I–III colon cancer, univariable Cox regression models. See also Figure 3b, left panel. Table S7: association of immune scores with RFS in stage I–III colon cancer, multivariable Cox regression model, adjusted to clinicopathological parameters. See also Figure 3b, right panel. Table S8: univariable associations of immune scores with OS in stage VI therapy-naïve colon cancer patients. See also Figure 3c.

**Author Contributions:** Conceptualization, A.M., P.M., B.G. and T.S.; methodology, A.M.; software, A.M.; formal analysis, AM, A.M.-B., M.S., K.H., S.S. and P.-H.E.; resources, A.M., P.M., F.P., B.G. and T.S.; data curation, A.M., M.B. and I.H.; writing—original draft preparation, A.M. and P.M.; writing review and editing, A.M., P.M., F.P., K.L., B.G. and T.S.; visualization, A.M. and J.E.; supervision, P.M, K.L., B.G. and T.S.; project administration, A.M. and P.-H.E.; funding acquisition, A.M., P.M., B.G. and T.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This study was supported by a postdoctoral grant from the Swedish Cancer Society to A.M. (CAN 2017/1066) and project grants from the Swedish Cancer Society to T.S. (CAN 2018/772), B.G. (CAN 2019/0382), and P.M. (CAN 2018/816); the Lions Cancer Foundation Uppsala to P.M.; and the Selanders foundation and P.O. Zetterling Foundation to A.M. U-CAN is supported by the Swedish Government (SRA CancerUU) and locally by Uppsala University and Region Uppsala.

**Institutional Review Board Statement:** The study was approved by the regional ethical committee in Uppsala, Sweden (Dnr 2010/198 and Dnr 2015/419).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** Data regarding methodology, image analysis, curation and data processing, and raw data of stroma fraction are available from the corresponding author.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**

