Interaction of kidney with thyroid

[dariushghasemi]

Here we tested the potential for the interaction between the replicated Kidney-associated variants and Thyroid function. To do so, we investigated by including the interaction term between the variants in the linear regression in two strategies:

1. A linear interaction between the variants and quantitative Thyroid Stimulating Hormone (TSH) level. lm(eGFRw.log.Res ~ SNP * TSH + PC1 + PC2 + PC3 + PC4 + PC5 + PC6 + PC7 + PC8 + PC9 + PC10)
2. A non-linear interaction between the variants and the categorized TSH level (Hyper- and Hypo-thyroidism). lm(eGFRw.log.Res ~ SNP * TSH_cat + PC1 + PC2 + PC3 + PC4 + PC5 + PC6 + PC7 + PC8 + PC9 + PC10)

Note: To account for the samples relatedness and population stratification we added the 10 first principal components to the model.

[dariushghasemi]

• As we observed the significant interaction for two SNPs out of 3 in one replicated locus, we tried to elucidate this findings. By centralizing TSH (0-mean-centered) we realized adjustment in SNP regression coefficients but still interaction effects and the corresponding P-values remained stable.
• Another important thing we found was that the significant interaction only occurs when we exclude the cases with cancers/renal carcinoma/alternation of thyroid function due to pregnancy which were ~9,700 individuals. If we apply the previous model on the entire CHRIS 10K consisting of ~10,140 individuals the significance of interaction terms goes away and there would not be any significant interaction anymore.

# Regular TSH levels
lm(eGFRw.log.Res ~ SNP * TSH + Sex + Age + PC1 + PC2 + PC3 + PC4 + PC5 + PC6 + PC7 + PC8 + PC9 + PC10)

• We replaced log(eGFR) with eGFRw.log.Res in the above model and ran it again. The results stay consistent in a way that those two variants with significant interactions with TSH remain significant. In this way we can avoid over-adjustment of the covariates: age and sex (03-Feb-23).

[dariushghasemi]

• We modified our exclusion criteria in the way that we omitted the observations with:

  1. missing values on both TSH and thyroid drugs (n = 4)
  2. thyroid cancer (n = 16)
  3. kidney cancer (n = 1)
  4. goitre (n = 277)
  5. operation on thyroid gland (n = 312)

• After excluding these individuals, exactly 9,730 people remined out of 10,146 for SNP-TSH interaction analysis. Here is the script to do the manipulations:

  vcfReg_TSHmod <-
  vcfReg %>%
  mutate(Thyroid_DrugName = na_if(Thyroid_DrugName, "")) %>%
  filter(
  !if_all(c(Thyroid_DrugName, TSH), is.na),                       # Removed are:   4 with NAs
  kidneyCancer  != 1 | is.na(kidneyCancer),                       # Removed are:   1 with Kidney cancer
  Goiter != 1 | is.na(Goiter),                                    # Removed are: 277 with Goitre
  Operation_thyroid_gland != 1 | is.na(Operation_thyroid_gland),  # Removed are: 312 with operation on thyroid gland
  ) %>%
  mutate(TSH_cat = replace(TSH_cat, Thyroid_DrugName == "Iodine therapy",       "HyperT"),
       TSH_cat = replace(TSH_cat, Thyroid_DrugName == "Levothyroxine sodium", "HypoT"),
       TSH_cat = replace(TSH_cat, Thyroid_DrugName == "Propylthiouracil",     "HyperT"),
       TSH_cat = replace(TSH_cat, Thyroid_DrugName == "Thiamazole",           "HyperT"),
       #Changing the reference level of TSH_cat to TSH_cat = "2" or "NormT"
       TSH_cat = relevel(as.factor(TSH_cat), ref = 2))

• Thus, the interaction models were executed as follows:

  # Interaction of SNP-TSH
  lm(log(eGFR) ~ SNP * TSH + PC1 + ... + PC10)
  # Interaction of SNP-thyroid_disease
  lm(log(eGFR) ~ SNP * TSH_cat + PC1 + ... + PC10)

• At the end, we found there was no significant interaction between kidney variants neither TSH nor thyroid disease at 0.05/11 level (10-Feb-2023).

[dariushghasemi]

• We decided to use the natural log-transformed winsorized eGFR (so called eGFRw.log) for interaction analysis rather than log(eGFR). Previously we had found a significant interaction with TSH in one single locus.

  # Interaction of SNP-TSH
  lm(eGFRw.log ~ SNP * TSH + PC1 + ... + PC10)
  # Interaction of SNP-thyroid_disease
  lm(eGFRw.log ~ SNP * TSH_cat + PC1 + ... + PC10)

• After changing the outcome, we realized if we use as eGFRw.log the outcome, we will have two significant n=interaction terms for the SNPs with TSH in one locus (13-Feb-2023).